diff options
Diffstat (limited to 'llvm/test')
32 files changed, 4834 insertions, 2996 deletions
diff --git a/llvm/test/CodeGen/AArch64/isinf.ll b/llvm/test/CodeGen/AArch64/isinf.ll index e68539b..e8bbaf9 100644 --- a/llvm/test/CodeGen/AArch64/isinf.ll +++ b/llvm/test/CodeGen/AArch64/isinf.ll @@ -27,9 +27,8 @@ define i32 @replace_isinf_call_f32(float %x) { ; CHECK-LABEL: replace_isinf_call_f32: ; CHECK: // %bb.0: ; CHECK-NEXT: fmov w9, s0 -; CHECK-NEXT: mov w8, #2139095040 // =0x7f800000 -; CHECK-NEXT: and w9, w9, #0x7fffffff -; CHECK-NEXT: cmp w9, w8 +; CHECK-NEXT: mov w8, #-16777216 // =0xff000000 +; CHECK-NEXT: cmp w8, w9, lsl #1 ; CHECK-NEXT: cset w0, eq ; CHECK-NEXT: ret %abs = tail call float @llvm.fabs.f32(float %x) @@ -43,9 +42,8 @@ define i32 @replace_isinf_call_f64(double %x) { ; CHECK-LABEL: replace_isinf_call_f64: ; CHECK: // %bb.0: ; CHECK-NEXT: fmov x9, d0 -; CHECK-NEXT: mov x8, #9218868437227405312 // =0x7ff0000000000000 -; CHECK-NEXT: and x9, x9, #0x7fffffffffffffff -; CHECK-NEXT: cmp x9, x8 +; CHECK-NEXT: mov x8, #-9007199254740992 // =0xffe0000000000000 +; CHECK-NEXT: cmp x8, x9, lsl #1 ; CHECK-NEXT: cset w0, eq ; CHECK-NEXT: ret %abs = tail call double @llvm.fabs.f64(double %x) diff --git a/llvm/test/CodeGen/AArch64/masked-integer-compare.ll b/llvm/test/CodeGen/AArch64/masked-integer-compare.ll new file mode 100644 index 0000000..363cd10 --- /dev/null +++ b/llvm/test/CodeGen/AArch64/masked-integer-compare.ll @@ -0,0 +1,178 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5 +; RUN: llc -mtriple=aarch64-none-linux-gnu < %s -o -| FileCheck %s + +; Test code generation support for SUBS (shifted register) from masked integer +; compare sequences. These sequences appear in isinf tests, for example. + +define i1 @combine_masked_i32(i32 %x) { +; CHECK-LABEL: combine_masked_i32: +; CHECK: // %bb.0: +; CHECK-NEXT: mov w8, #-16777216 // =0xff000000 +; CHECK-NEXT: cmp w8, w0, lsl #1 +; CHECK-NEXT: cset w0, eq +; CHECK-NEXT: ret + %and = and i32 %x, u0x7fffffff + %sub = sub i32 %and, u0x7f800000 + %cmp = icmp eq i32 %sub, 0 + ret i1 %cmp +} + +define i1 @combine_masked_i64(i64 %x) { +; CHECK-LABEL: combine_masked_i64: +; CHECK: // %bb.0: +; CHECK-NEXT: mov x8, #-9007199254740992 // =0xffe0000000000000 +; CHECK-NEXT: cmp x8, x0, lsl #1 +; CHECK-NEXT: cset w0, eq +; CHECK-NEXT: ret + %and = and i64 %x, u0x7fffffffffffffff + %sub = sub i64 %and, u0x7ff0000000000000 + %cmp = icmp eq i64 %sub, 0 + ret i1 %cmp +} + +define i1 @combine_masked_ne(i32 %x) { +; CHECK-LABEL: combine_masked_ne: +; CHECK: // %bb.0: +; CHECK-NEXT: mov w8, #-16777216 // =0xff000000 +; CHECK-NEXT: cmp w8, w0, lsl #1 +; CHECK-NEXT: cset w0, ne +; CHECK-NEXT: ret + %and = and i32 %x, u0x7fffffff + %cmp = icmp ne i32 %and, u0x7f800000 + ret i1 %cmp +} + +define i1 @combine_masked_lsl4(i32 %x) { +; CHECK-LABEL: combine_masked_lsl4: +; CHECK: // %bb.0: +; CHECK-NEXT: mov w8, #-134217728 // =0xf8000000 +; CHECK-NEXT: cmp w8, w0, lsl #4 +; CHECK-NEXT: cset w0, eq +; CHECK-NEXT: ret + %and = and i32 %x, u0x0fffffff + %cmp = icmp eq i32 %and, u0x0f800000 + ret i1 %cmp +} + +define i1 @dont_combine_not_mask(i32 %x) { +; CHECK-LABEL: dont_combine_not_mask: +; CHECK: // %bb.0: +; CHECK-NEXT: mov w8, #2139095040 // =0x7f800000 +; CHECK-NEXT: and w9, w0, #0x7ffffffe +; CHECK-NEXT: cmp w9, w8 +; CHECK-NEXT: cset w0, eq +; CHECK-NEXT: ret + %and = and i32 %x, u0x7ffffffe + %cmp = icmp eq i32 %and, u0x7f800000 + ret i1 %cmp +} + +define i1 @dont_combine_cmp_not_masked(i32 %x) { +; CHECK-LABEL: dont_combine_cmp_not_masked: +; CHECK: // %bb.0: +; CHECK-NEXT: mov w8, #2139095040 // =0x7f800000 +; CHECK-NEXT: and w9, w0, #0x3fffffff +; CHECK-NEXT: cmp w9, w8 +; CHECK-NEXT: cset w0, eq +; CHECK-NEXT: ret + %and = and i32 %x, u0x3fffffff + %cmp = icmp eq i32 %and, u0x7f800000 + ret i1 %cmp +} + +define i1 @dont_combine_not_constant_mask(i32 %x, i32 %m) { +; CHECK-LABEL: dont_combine_not_constant_mask: +; CHECK: // %bb.0: +; CHECK-NEXT: mov w8, #2139095040 // =0x7f800000 +; CHECK-NEXT: and w9, w0, w1 +; CHECK-NEXT: cmp w9, w8 +; CHECK-NEXT: cset w0, eq +; CHECK-NEXT: ret + %and = and i32 %x, %m + %cmp = icmp eq i32 %and, u0x7f800000 + ret i1 %cmp +} + +define i1 @dont_combine_not_constant_cmp(i32 %x, i32 %c) { +; CHECK-LABEL: dont_combine_not_constant_cmp: +; CHECK: // %bb.0: +; CHECK-NEXT: and w8, w0, #0xfffffff +; CHECK-NEXT: cmp w8, w1 +; CHECK-NEXT: cset w0, eq +; CHECK-NEXT: ret + %and = and i32 %x, u0x0fffffff + %cmp = icmp eq i32 %and, %c + ret i1 %cmp +} + +define i1 @dont_combine_subs_imm(i32 %x) { +; CHECK-LABEL: dont_combine_subs_imm: +; CHECK: // %bb.0: +; CHECK-NEXT: and w8, w0, #0x7fffffff +; CHECK-NEXT: cmp w8, #291 +; CHECK-NEXT: cset w0, eq +; CHECK-NEXT: ret + %and = and i32 %x, u0x7fffffff + %cmp = icmp eq i32 %and, u0x123 + ret i1 %cmp +} + +define i1 @dont_combine_subs_imm_lsl12(i32 %x) { +; CHECK-LABEL: dont_combine_subs_imm_lsl12: +; CHECK: // %bb.0: +; CHECK-NEXT: and w8, w0, #0x7fffffff +; CHECK-NEXT: cmp w8, #291, lsl #12 // =1191936 +; CHECK-NEXT: cset w0, eq +; CHECK-NEXT: ret + %and = and i32 %x, u0x7fffffff + %cmp = icmp eq i32 %and, u0x123000 + ret i1 %cmp +} + +define { i1, i1 } @dont_combine_multi_use_cmp(i32 %x) { +; CHECK-LABEL: dont_combine_multi_use_cmp: +; CHECK: // %bb.0: +; CHECK-NEXT: mov w8, #2139095040 // =0x7f800000 +; CHECK-NEXT: and w9, w0, #0x7fffffff +; CHECK-NEXT: cmp w9, w8 +; CHECK-NEXT: cset w0, eq +; CHECK-NEXT: cset w1, lt +; CHECK-NEXT: ret + %and = and i32 %x, u0x7fffffff + %eq = icmp eq i32 %and, u0x7f800000 + %lt = icmp slt i32 %and, u0x7f800000 + %r1 = insertvalue { i1, i1 } poison, i1 %eq, 0 + %r2 = insertvalue { i1, i1 } %r1, i1 %lt, 1 + ret { i1, i1 } %r2 +} + +define { i32, i1 } @dont_combine_multi_use_sub(i32 %x) { +; CHECK-LABEL: dont_combine_multi_use_sub: +; CHECK: // %bb.0: +; CHECK-NEXT: mov w8, #-2139095040 // =0x80800000 +; CHECK-NEXT: and w9, w0, #0x7fffffff +; CHECK-NEXT: adds w0, w9, w8 +; CHECK-NEXT: cset w1, eq +; CHECK-NEXT: ret + %and = and i32 %x, u0x7fffffff + %sub = sub i32 %and, u0x7f800000 + %cmp = icmp eq i32 %sub, 0 + %r1 = insertvalue { i32, i1 } poison, i32 %sub, 0 + %r2 = insertvalue { i32, i1 } %r1, i1 %cmp, 1 + ret { i32, i1 } %r2 +} + +define { i32, i1 } @dont_combine_multi_use_and(i32 %x) { +; CHECK-LABEL: dont_combine_multi_use_and: +; CHECK: // %bb.0: +; CHECK-NEXT: mov w8, #2139095040 // =0x7f800000 +; CHECK-NEXT: and w0, w0, #0x7fffffff +; CHECK-NEXT: cmp w0, w8 +; CHECK-NEXT: cset w1, eq +; CHECK-NEXT: ret + %and = and i32 %x, u0x7fffffff + %cmp = icmp eq i32 %and, u0x7f800000 + %r1 = insertvalue { i32, i1 } poison, i32 %and, 0 + %r2 = insertvalue { i32, i1 } %r1, i1 %cmp, 1 + ret { i32, i1 } %r2 +} diff --git a/llvm/test/CodeGen/AMDGPU/fmax_legacy.f16.ll b/llvm/test/CodeGen/AMDGPU/fmax_legacy.f16.ll index ed48999..bd28f72 100644 --- a/llvm/test/CodeGen/AMDGPU/fmax_legacy.f16.ll +++ b/llvm/test/CodeGen/AMDGPU/fmax_legacy.f16.ll @@ -1,734 +1,759 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py -; RUN: llc -mtriple=amdgcn-- -mcpu=gfx900 < %s | FileCheck -enable-var-scope -check-prefixes=GFX9-SAFE %s -; RUN: llc -enable-no-nans-fp-math -enable-no-signed-zeros-fp-math -mtriple=amdgcn-- -mcpu=gfx900 < %s | FileCheck -enable-var-scope -check-prefixes=GFX9-NNAN %s +; RUN: llc -mtriple=amdgcn-- -mcpu=gfx900 < %s | FileCheck -enable-var-scope -check-prefixes=GFX9 %s -; RUN: llc -mtriple=amdgcn-- -mcpu=fiji < %s | FileCheck -enable-var-scope -check-prefix=VI-SAFE %s -; RUN: llc -enable-no-nans-fp-math -enable-no-signed-zeros-fp-math -mtriple=amdgcn-- -mcpu=fiji < %s | FileCheck -enable-var-scope -check-prefixes=VI-NNAN %s +; RUN: llc -mtriple=amdgcn-- -mcpu=fiji < %s | FileCheck -enable-var-scope -check-prefix=VI %s -; RUN: llc -mtriple=amdgcn-- < %s | FileCheck -enable-var-scope -check-prefix=SI-SAFE %s -; RUN: llc -enable-no-nans-fp-math -enable-no-signed-zeros-fp-math -mtriple=amdgcn-- < %s | FileCheck -enable-var-scope -check-prefixes=SI-NNAN %s +; RUN: llc -mtriple=amdgcn-- < %s | FileCheck -enable-var-scope -check-prefix=SI %s -; RUN: llc -mtriple=amdgcn-- -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX11-SAFE-TRUE16 %s -; RUN: llc -mtriple=amdgcn-- -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX11-SAFE-FAKE16 %s -; RUN: llc -enable-no-nans-fp-math -enable-no-signed-zeros-fp-math -mtriple=amdgcn-- -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX11-NNAN,GFX11-NNAN-TRUE16 %s -; RUN: llc -enable-no-nans-fp-math -enable-no-signed-zeros-fp-math -mtriple=amdgcn-- -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX11-NNAN,GFX11-NNAN-FAKE16 %s +; RUN: llc -mtriple=amdgcn-- -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX11,GFX11-TRUE16 %s +; RUN: llc -mtriple=amdgcn-- -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX11,GFX11-FAKE16 %s define half @test_fmax_legacy_ugt_f16(half %a, half %b) #0 { -; GFX9-SAFE-LABEL: test_fmax_legacy_ugt_f16: -; GFX9-SAFE: ; %bb.0: -; GFX9-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v0, v1 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc -; GFX9-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; GFX9-NNAN-LABEL: test_fmax_legacy_ugt_f16: -; GFX9-NNAN: ; %bb.0: -; GFX9-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-NNAN-NEXT: v_max_f16_e32 v0, v0, v1 -; GFX9-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; VI-SAFE-LABEL: test_fmax_legacy_ugt_f16: -; VI-SAFE: ; %bb.0: -; VI-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v0, v1 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc -; VI-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; VI-NNAN-LABEL: test_fmax_legacy_ugt_f16: -; VI-NNAN: ; %bb.0: -; VI-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-NNAN-NEXT: v_max_f16_e32 v0, v0, v1 -; VI-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; SI-SAFE-LABEL: test_fmax_legacy_ugt_f16: -; SI-SAFE: ; %bb.0: -; SI-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v0, v0 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v1, v1 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v1, v1 -; SI-SAFE-NEXT: v_max_legacy_f32_e32 v0, v1, v0 -; SI-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; SI-NNAN-LABEL: test_fmax_legacy_ugt_f16: -; SI-NNAN: ; %bb.0: -; SI-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v1, v1 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v0, v0 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v1, v1 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-NNAN-NEXT: v_max_f32_e32 v0, v0, v1 -; SI-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-TRUE16-LABEL: test_fmax_legacy_ugt_f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v0.l, v1.l -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v1.l, v0.l, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: test_fmax_legacy_ugt_f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v0, v1 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NNAN-TRUE16-LABEL: test_fmax_legacy_ugt_f16: -; GFX11-NNAN-TRUE16: ; %bb.0: -; GFX11-NNAN-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NNAN-TRUE16-NEXT: v_max_f16_e32 v0.l, v0.l, v1.l -; GFX11-NNAN-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NNAN-FAKE16-LABEL: test_fmax_legacy_ugt_f16: -; GFX11-NNAN-FAKE16: ; %bb.0: -; GFX11-NNAN-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NNAN-FAKE16-NEXT: v_max_f16_e32 v0, v0, v1 -; GFX11-NNAN-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX9-LABEL: test_fmax_legacy_ugt_f16: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_cmp_nle_f16_e32 vcc, v0, v1 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; VI-LABEL: test_fmax_legacy_ugt_f16: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_cmp_nle_f16_e32 vcc, v0, v1 +; VI-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc +; VI-NEXT: s_setpc_b64 s[30:31] +; +; SI-LABEL: test_fmax_legacy_ugt_f16: +; SI: ; %bb.0: +; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; SI-NEXT: v_cvt_f16_f32_e32 v0, v0 +; SI-NEXT: v_cvt_f16_f32_e32 v1, v1 +; SI-NEXT: v_cvt_f32_f16_e32 v0, v0 +; SI-NEXT: v_cvt_f32_f16_e32 v1, v1 +; SI-NEXT: v_max_legacy_f32_e32 v0, v1, v0 +; SI-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-TRUE16-LABEL: test_fmax_legacy_ugt_f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v0.l, v1.l +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v1.l, v0.l, vcc_lo +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: test_fmax_legacy_ugt_f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v0, v1 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc_lo +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = fcmp ugt half %a, %b %val = select i1 %cmp, half %a, half %b ret half %val } +define half @test_fmax_legacy_ugt_f16_fast(half %a, half %b) #0 { +; GFX9-LABEL: test_fmax_legacy_ugt_f16_fast: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_max_f16_e32 v0, v0, v1 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; VI-LABEL: test_fmax_legacy_ugt_f16_fast: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_max_f16_e32 v0, v0, v1 +; VI-NEXT: s_setpc_b64 s[30:31] +; +; SI-LABEL: test_fmax_legacy_ugt_f16_fast: +; SI: ; %bb.0: +; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; SI-NEXT: v_cvt_f16_f32_e32 v1, v1 +; SI-NEXT: v_cvt_f16_f32_e32 v0, v0 +; SI-NEXT: v_cvt_f32_f16_e32 v1, v1 +; SI-NEXT: v_cvt_f32_f16_e32 v0, v0 +; SI-NEXT: v_max_f32_e32 v0, v0, v1 +; SI-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-TRUE16-LABEL: test_fmax_legacy_ugt_f16_fast: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_max_f16_e32 v0.l, v0.l, v1.l +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: test_fmax_legacy_ugt_f16_fast: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_max_f16_e32 v0, v0, v1 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] + %cmp = fcmp ugt half %a, %b + %val = select nnan nsz i1 %cmp, half %a, half %b + ret half %val +} + define <2 x half> @test_fmax_legacy_ugt_v2f16(<2 x half> %a, <2 x half> %b) #0 { -; GFX9-SAFE-LABEL: test_fmax_legacy_ugt_v2f16: -; GFX9-SAFE: ; %bb.0: -; GFX9-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v2, 16, v1 -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v3, 16, v0 -; GFX9-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v3, v2 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v2, v2, v3, vcc -; GFX9-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v0, v1 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc -; GFX9-SAFE-NEXT: s_mov_b32 s4, 0x5040100 -; GFX9-SAFE-NEXT: v_perm_b32 v0, v2, v0, s4 -; GFX9-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; GFX9-NNAN-LABEL: test_fmax_legacy_ugt_v2f16: -; GFX9-NNAN: ; %bb.0: -; GFX9-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-NNAN-NEXT: v_pk_max_f16 v0, v0, v1 -; GFX9-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; VI-SAFE-LABEL: test_fmax_legacy_ugt_v2f16: -; VI-SAFE: ; %bb.0: -; VI-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v2, 16, v1 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v3, 16, v0 -; VI-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v3, v2 -; VI-SAFE-NEXT: v_cndmask_b32_sdwa v2, v2, v3, vcc dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; VI-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v0, v1 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc -; VI-SAFE-NEXT: v_or_b32_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD -; VI-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; VI-NNAN-LABEL: test_fmax_legacy_ugt_v2f16: -; VI-NNAN: ; %bb.0: -; VI-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-NNAN-NEXT: v_max_f16_sdwa v2, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; VI-NNAN-NEXT: v_max_f16_e32 v0, v0, v1 -; VI-NNAN-NEXT: v_or_b32_e32 v0, v0, v2 -; VI-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; SI-SAFE-LABEL: test_fmax_legacy_ugt_v2f16: -; SI-SAFE: ; %bb.0: -; SI-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v1, v1 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v3, v3 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v0, v0 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v2, v2 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v1, v1 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v3, v3 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v2, v2 -; SI-SAFE-NEXT: v_max_legacy_f32_e32 v0, v2, v0 -; SI-SAFE-NEXT: v_max_legacy_f32_e32 v1, v3, v1 -; SI-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; SI-NNAN-LABEL: test_fmax_legacy_ugt_v2f16: -; SI-NNAN: ; %bb.0: -; SI-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v3, v3 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v1, v1 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v2, v2 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v0, v0 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v3, v3 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v1, v1 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v2, v2 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-NNAN-NEXT: v_max_f32_e32 v0, v0, v2 -; SI-NNAN-NEXT: v_max_f32_e32 v1, v1, v3 -; SI-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-TRUE16-LABEL: test_fmax_legacy_ugt_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v0.h, v1.h -; GFX11-SAFE-TRUE16-NEXT: v_cmp_nle_f16_e64 s0, v0.l, v1.l -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v1.h, v0.h, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v1.l, v0.l, s0 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: test_fmax_legacy_ugt_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v2, 16, v1 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v3, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v2, v2, v3, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v0, v1 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v2, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NNAN-LABEL: test_fmax_legacy_ugt_v2f16: -; GFX11-NNAN: ; %bb.0: -; GFX11-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NNAN-NEXT: v_pk_max_f16 v0, v0, v1 -; GFX11-NNAN-NEXT: s_setpc_b64 s[30:31] +; GFX9-LABEL: test_fmax_legacy_ugt_v2f16: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_lshrrev_b32_e32 v2, 16, v1 +; GFX9-NEXT: v_lshrrev_b32_e32 v3, 16, v0 +; GFX9-NEXT: v_cmp_nle_f16_e32 vcc, v3, v2 +; GFX9-NEXT: v_cndmask_b32_e32 v2, v2, v3, vcc +; GFX9-NEXT: v_cmp_nle_f16_e32 vcc, v0, v1 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc +; GFX9-NEXT: s_mov_b32 s4, 0x5040100 +; GFX9-NEXT: v_perm_b32 v0, v2, v0, s4 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; VI-LABEL: test_fmax_legacy_ugt_v2f16: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_lshrrev_b32_e32 v2, 16, v1 +; VI-NEXT: v_lshrrev_b32_e32 v3, 16, v0 +; VI-NEXT: v_cmp_nle_f16_e32 vcc, v3, v2 +; VI-NEXT: v_cndmask_b32_sdwa v2, v2, v3, vcc dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; VI-NEXT: v_cmp_nle_f16_e32 vcc, v0, v1 +; VI-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc +; VI-NEXT: v_or_b32_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD +; VI-NEXT: s_setpc_b64 s[30:31] +; +; SI-LABEL: test_fmax_legacy_ugt_v2f16: +; SI: ; %bb.0: +; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; SI-NEXT: v_cvt_f16_f32_e32 v1, v1 +; SI-NEXT: v_cvt_f16_f32_e32 v3, v3 +; SI-NEXT: v_cvt_f16_f32_e32 v0, v0 +; SI-NEXT: v_cvt_f16_f32_e32 v2, v2 +; SI-NEXT: v_cvt_f32_f16_e32 v1, v1 +; SI-NEXT: v_cvt_f32_f16_e32 v3, v3 +; SI-NEXT: v_cvt_f32_f16_e32 v0, v0 +; SI-NEXT: v_cvt_f32_f16_e32 v2, v2 +; SI-NEXT: v_max_legacy_f32_e32 v0, v2, v0 +; SI-NEXT: v_max_legacy_f32_e32 v1, v3, v1 +; SI-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-TRUE16-LABEL: test_fmax_legacy_ugt_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v0.h, v1.h +; GFX11-TRUE16-NEXT: v_cmp_nle_f16_e64 s0, v0.l, v1.l +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v1.h, v0.h, vcc_lo +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v1.l, v0.l, s0 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: test_fmax_legacy_ugt_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v2, 16, v1 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v3, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v2, v2, v3, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v0, v1 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v2, v0, 0x5040100 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = fcmp ugt <2 x half> %a, %b %val = select <2 x i1> %cmp, <2 x half> %a, <2 x half> %b ret <2 x half> %val } +define <2 x half> @test_fmax_legacy_ugt_v2f16_fast(<2 x half> %a, <2 x half> %b) #0 { +; GFX9-LABEL: test_fmax_legacy_ugt_v2f16_fast: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_pk_max_f16 v0, v0, v1 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; VI-LABEL: test_fmax_legacy_ugt_v2f16_fast: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_max_f16_sdwa v2, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; VI-NEXT: v_max_f16_e32 v0, v0, v1 +; VI-NEXT: v_or_b32_e32 v0, v0, v2 +; VI-NEXT: s_setpc_b64 s[30:31] +; +; SI-LABEL: test_fmax_legacy_ugt_v2f16_fast: +; SI: ; %bb.0: +; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; SI-NEXT: v_cvt_f16_f32_e32 v3, v3 +; SI-NEXT: v_cvt_f16_f32_e32 v1, v1 +; SI-NEXT: v_cvt_f16_f32_e32 v2, v2 +; SI-NEXT: v_cvt_f16_f32_e32 v0, v0 +; SI-NEXT: v_cvt_f32_f16_e32 v3, v3 +; SI-NEXT: v_cvt_f32_f16_e32 v1, v1 +; SI-NEXT: v_cvt_f32_f16_e32 v2, v2 +; SI-NEXT: v_cvt_f32_f16_e32 v0, v0 +; SI-NEXT: v_max_f32_e32 v0, v0, v2 +; SI-NEXT: v_max_f32_e32 v1, v1, v3 +; SI-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-LABEL: test_fmax_legacy_ugt_v2f16_fast: +; GFX11: ; %bb.0: +; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-NEXT: v_pk_max_f16 v0, v0, v1 +; GFX11-NEXT: s_setpc_b64 s[30:31] + %cmp = fcmp ugt <2 x half> %a, %b + %val = select nnan nsz <2 x i1> %cmp, <2 x half> %a, <2 x half> %b + ret <2 x half> %val +} + define <3 x half> @test_fmax_legacy_ugt_v3f16(<3 x half> %a, <3 x half> %b) #0 { -; GFX9-SAFE-LABEL: test_fmax_legacy_ugt_v3f16: -; GFX9-SAFE: ; %bb.0: -; GFX9-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v5, 16, v0 -; GFX9-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v5, v4 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v4, v4, v5, vcc -; GFX9-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v1, v3 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc -; GFX9-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v0, v2 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc -; GFX9-SAFE-NEXT: s_mov_b32 s4, 0x5040100 -; GFX9-SAFE-NEXT: v_perm_b32 v0, v4, v0, s4 -; GFX9-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; GFX9-NNAN-LABEL: test_fmax_legacy_ugt_v3f16: -; GFX9-NNAN: ; %bb.0: -; GFX9-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-NNAN-NEXT: v_pk_max_f16 v1, v1, v3 -; GFX9-NNAN-NEXT: v_pk_max_f16 v0, v0, v2 -; GFX9-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; VI-SAFE-LABEL: test_fmax_legacy_ugt_v3f16: -; VI-SAFE: ; %bb.0: -; VI-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v5, 16, v0 -; VI-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v5, v4 -; VI-SAFE-NEXT: v_cndmask_b32_sdwa v4, v4, v5, vcc dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; VI-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v1, v3 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc -; VI-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v0, v2 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc -; VI-SAFE-NEXT: v_or_b32_sdwa v0, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD -; VI-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; VI-NNAN-LABEL: test_fmax_legacy_ugt_v3f16: -; VI-NNAN: ; %bb.0: -; VI-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-NNAN-NEXT: v_max_f16_sdwa v4, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; VI-NNAN-NEXT: v_max_f16_e32 v0, v0, v2 -; VI-NNAN-NEXT: v_max_f16_e32 v1, v1, v3 -; VI-NNAN-NEXT: v_or_b32_e32 v0, v0, v4 -; VI-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; SI-SAFE-LABEL: test_fmax_legacy_ugt_v3f16: -; SI-SAFE: ; %bb.0: -; SI-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v2, v2 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v5, v5 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v1, v1 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v4, v4 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v0, v0 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v3, v3 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v2, v2 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v5, v5 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v1, v1 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v4, v4 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v3, v3 -; SI-SAFE-NEXT: v_max_legacy_f32_e32 v0, v3, v0 -; SI-SAFE-NEXT: v_max_legacy_f32_e32 v1, v4, v1 -; SI-SAFE-NEXT: v_max_legacy_f32_e32 v2, v5, v2 -; SI-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; SI-NNAN-LABEL: test_fmax_legacy_ugt_v3f16: -; SI-NNAN: ; %bb.0: -; SI-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v5, v5 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v2, v2 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v4, v4 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v1, v1 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v3, v3 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v0, v0 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v5, v5 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v2, v2 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v4, v4 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v1, v1 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v3, v3 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-NNAN-NEXT: v_max_f32_e32 v0, v0, v3 -; SI-NNAN-NEXT: v_max_f32_e32 v1, v1, v4 -; SI-NNAN-NEXT: v_max_f32_e32 v2, v2, v5 -; SI-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-TRUE16-LABEL: test_fmax_legacy_ugt_v3f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v0.h, v2.h -; GFX11-SAFE-TRUE16-NEXT: v_cmp_nle_f16_e64 s0, v0.l, v2.l -; GFX11-SAFE-TRUE16-NEXT: v_cmp_nle_f16_e64 s1, v1.l, v3.l -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v0.h, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, s0 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v1.l, v3.l, v1.l, s1 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: test_fmax_legacy_ugt_v3f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v0 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v0, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_3) | instid1(VALU_DEP_3) -; GFX11-SAFE-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v5, v4 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v2, v4, v5, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v1, v3 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v2, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NNAN-LABEL: test_fmax_legacy_ugt_v3f16: -; GFX11-NNAN: ; %bb.0: -; GFX11-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NNAN-NEXT: v_pk_max_f16 v0, v0, v2 -; GFX11-NNAN-NEXT: v_pk_max_f16 v1, v1, v3 -; GFX11-NNAN-NEXT: s_setpc_b64 s[30:31] +; GFX9-LABEL: test_fmax_legacy_ugt_v3f16: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; GFX9-NEXT: v_lshrrev_b32_e32 v5, 16, v0 +; GFX9-NEXT: v_cmp_nle_f16_e32 vcc, v5, v4 +; GFX9-NEXT: v_cndmask_b32_e32 v4, v4, v5, vcc +; GFX9-NEXT: v_cmp_nle_f16_e32 vcc, v1, v3 +; GFX9-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc +; GFX9-NEXT: v_cmp_nle_f16_e32 vcc, v0, v2 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc +; GFX9-NEXT: s_mov_b32 s4, 0x5040100 +; GFX9-NEXT: v_perm_b32 v0, v4, v0, s4 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; VI-LABEL: test_fmax_legacy_ugt_v3f16: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; VI-NEXT: v_lshrrev_b32_e32 v5, 16, v0 +; VI-NEXT: v_cmp_nle_f16_e32 vcc, v5, v4 +; VI-NEXT: v_cndmask_b32_sdwa v4, v4, v5, vcc dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; VI-NEXT: v_cmp_nle_f16_e32 vcc, v1, v3 +; VI-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc +; VI-NEXT: v_cmp_nle_f16_e32 vcc, v0, v2 +; VI-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc +; VI-NEXT: v_or_b32_sdwa v0, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD +; VI-NEXT: s_setpc_b64 s[30:31] +; +; SI-LABEL: test_fmax_legacy_ugt_v3f16: +; SI: ; %bb.0: +; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; SI-NEXT: v_cvt_f16_f32_e32 v2, v2 +; SI-NEXT: v_cvt_f16_f32_e32 v5, v5 +; SI-NEXT: v_cvt_f16_f32_e32 v1, v1 +; SI-NEXT: v_cvt_f16_f32_e32 v4, v4 +; SI-NEXT: v_cvt_f16_f32_e32 v0, v0 +; SI-NEXT: v_cvt_f16_f32_e32 v3, v3 +; SI-NEXT: v_cvt_f32_f16_e32 v2, v2 +; SI-NEXT: v_cvt_f32_f16_e32 v5, v5 +; SI-NEXT: v_cvt_f32_f16_e32 v1, v1 +; SI-NEXT: v_cvt_f32_f16_e32 v4, v4 +; SI-NEXT: v_cvt_f32_f16_e32 v0, v0 +; SI-NEXT: v_cvt_f32_f16_e32 v3, v3 +; SI-NEXT: v_max_legacy_f32_e32 v0, v3, v0 +; SI-NEXT: v_max_legacy_f32_e32 v1, v4, v1 +; SI-NEXT: v_max_legacy_f32_e32 v2, v5, v2 +; SI-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-TRUE16-LABEL: test_fmax_legacy_ugt_v3f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v0.h, v2.h +; GFX11-TRUE16-NEXT: v_cmp_nle_f16_e64 s0, v0.l, v2.l +; GFX11-TRUE16-NEXT: v_cmp_nle_f16_e64 s1, v1.l, v3.l +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v0.h, vcc_lo +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, s0 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v1.l, v3.l, v1.l, s1 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: test_fmax_legacy_ugt_v3f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v0 +; GFX11-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v0, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc_lo +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_3) | instid1(VALU_DEP_3) +; GFX11-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v5, v4 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v2, v4, v5, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v1, v3 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v2, v0, 0x5040100 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = fcmp ugt <3 x half> %a, %b %val = select <3 x i1> %cmp, <3 x half> %a, <3 x half> %b ret <3 x half> %val } +define <3 x half> @test_fmax_legacy_ugt_v3f16_fast(<3 x half> %a, <3 x half> %b) #0 { +; GFX9-LABEL: test_fmax_legacy_ugt_v3f16_fast: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_pk_max_f16 v3, v3, v3 +; GFX9-NEXT: v_pk_max_f16 v1, v1, v1 +; GFX9-NEXT: v_pk_max_f16 v1, v1, v3 +; GFX9-NEXT: v_pk_max_f16 v0, v0, v2 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; VI-LABEL: test_fmax_legacy_ugt_v3f16_fast: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_max_f16_sdwa v4, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; VI-NEXT: v_max_f16_e32 v0, v0, v2 +; VI-NEXT: v_max_f16_e32 v1, v1, v3 +; VI-NEXT: v_or_b32_e32 v0, v0, v4 +; VI-NEXT: s_setpc_b64 s[30:31] +; +; SI-LABEL: test_fmax_legacy_ugt_v3f16_fast: +; SI: ; %bb.0: +; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; SI-NEXT: v_cvt_f16_f32_e32 v5, v5 +; SI-NEXT: v_cvt_f16_f32_e32 v2, v2 +; SI-NEXT: v_cvt_f16_f32_e32 v4, v4 +; SI-NEXT: v_cvt_f16_f32_e32 v1, v1 +; SI-NEXT: v_cvt_f16_f32_e32 v3, v3 +; SI-NEXT: v_cvt_f16_f32_e32 v0, v0 +; SI-NEXT: v_cvt_f32_f16_e32 v5, v5 +; SI-NEXT: v_cvt_f32_f16_e32 v2, v2 +; SI-NEXT: v_cvt_f32_f16_e32 v4, v4 +; SI-NEXT: v_cvt_f32_f16_e32 v1, v1 +; SI-NEXT: v_cvt_f32_f16_e32 v3, v3 +; SI-NEXT: v_cvt_f32_f16_e32 v0, v0 +; SI-NEXT: v_max_f32_e32 v0, v0, v3 +; SI-NEXT: v_max_f32_e32 v1, v1, v4 +; SI-NEXT: v_max_f32_e32 v2, v2, v5 +; SI-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-LABEL: test_fmax_legacy_ugt_v3f16_fast: +; GFX11: ; %bb.0: +; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-NEXT: v_pk_max_f16 v3, v3, v3 +; GFX11-NEXT: v_pk_max_f16 v1, v1, v1 +; GFX11-NEXT: v_pk_max_f16 v0, v0, v2 +; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX11-NEXT: v_pk_max_f16 v1, v1, v3 +; GFX11-NEXT: s_setpc_b64 s[30:31] + %cmp = fcmp ugt <3 x half> %a, %b + %val = select nnan nsz <3 x i1> %cmp, <3 x half> %a, <3 x half> %b + ret <3 x half> %val +} + define <4 x half> @test_fmax_legacy_ugt_v4f16(<4 x half> %a, <4 x half> %b) #0 { -; GFX9-SAFE-LABEL: test_fmax_legacy_ugt_v4f16: -; GFX9-SAFE: ; %bb.0: -; GFX9-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v6, 16, v3 -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v7, 16, v1 -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v5, 16, v0 -; GFX9-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v7, v6 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v6, v6, v7, vcc -; GFX9-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v5, v4 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v4, v4, v5, vcc -; GFX9-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v1, v3 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc -; GFX9-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v0, v2 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc -; GFX9-SAFE-NEXT: s_mov_b32 s4, 0x5040100 -; GFX9-SAFE-NEXT: v_perm_b32 v0, v4, v0, s4 -; GFX9-SAFE-NEXT: v_perm_b32 v1, v6, v1, s4 -; GFX9-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; GFX9-NNAN-LABEL: test_fmax_legacy_ugt_v4f16: -; GFX9-NNAN: ; %bb.0: -; GFX9-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-NNAN-NEXT: v_pk_max_f16 v0, v0, v2 -; GFX9-NNAN-NEXT: v_pk_max_f16 v1, v1, v3 -; GFX9-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; VI-SAFE-LABEL: test_fmax_legacy_ugt_v4f16: -; VI-SAFE: ; %bb.0: -; VI-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v6, 16, v3 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v7, 16, v1 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v5, 16, v0 -; VI-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v7, v6 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v6, v6, v7, vcc -; VI-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v5, v4 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v4, v4, v5, vcc -; VI-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v1, v3 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc -; VI-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v0, v2 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc -; VI-SAFE-NEXT: v_lshlrev_b32_e32 v2, 16, v4 -; VI-SAFE-NEXT: v_or_b32_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD -; VI-SAFE-NEXT: v_lshlrev_b32_e32 v2, 16, v6 -; VI-SAFE-NEXT: v_or_b32_sdwa v1, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD -; VI-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; VI-NNAN-LABEL: test_fmax_legacy_ugt_v4f16: -; VI-NNAN: ; %bb.0: -; VI-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-NNAN-NEXT: v_max_f16_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; VI-NNAN-NEXT: v_max_f16_sdwa v5, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; VI-NNAN-NEXT: v_max_f16_e32 v1, v1, v3 -; VI-NNAN-NEXT: v_max_f16_e32 v0, v0, v2 -; VI-NNAN-NEXT: v_or_b32_e32 v0, v0, v5 -; VI-NNAN-NEXT: v_or_b32_e32 v1, v1, v4 -; VI-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; SI-SAFE-LABEL: test_fmax_legacy_ugt_v4f16: -; SI-SAFE: ; %bb.0: -; SI-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v3, v3 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v7, v7 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v2, v2 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v6, v6 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v1, v1 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v5, v5 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v0, v0 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v4, v4 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v3, v3 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v7, v7 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v2, v2 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v6, v6 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v1, v1 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v5, v5 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v4, v4 -; SI-SAFE-NEXT: v_max_legacy_f32_e32 v0, v4, v0 -; SI-SAFE-NEXT: v_max_legacy_f32_e32 v1, v5, v1 -; SI-SAFE-NEXT: v_max_legacy_f32_e32 v2, v6, v2 -; SI-SAFE-NEXT: v_max_legacy_f32_e32 v3, v7, v3 -; SI-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; SI-NNAN-LABEL: test_fmax_legacy_ugt_v4f16: -; SI-NNAN: ; %bb.0: -; SI-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v7, v7 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v3, v3 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v6, v6 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v2, v2 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v5, v5 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v1, v1 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v4, v4 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v0, v0 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v7, v7 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v3, v3 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v6, v6 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v2, v2 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v5, v5 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v1, v1 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v4, v4 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-NNAN-NEXT: v_max_f32_e32 v0, v0, v4 -; SI-NNAN-NEXT: v_max_f32_e32 v1, v1, v5 -; SI-NNAN-NEXT: v_max_f32_e32 v2, v2, v6 -; SI-NNAN-NEXT: v_max_f32_e32 v3, v3, v7 -; SI-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-TRUE16-LABEL: test_fmax_legacy_ugt_v4f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v1.h, v3.h -; GFX11-SAFE-TRUE16-NEXT: v_cmp_nle_f16_e64 s0, v0.h, v2.h -; GFX11-SAFE-TRUE16-NEXT: v_cmp_nle_f16_e64 s1, v0.l, v2.l -; GFX11-SAFE-TRUE16-NEXT: v_cmp_nle_f16_e64 s2, v1.l, v3.l -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v1.h, v3.h, v1.h, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v0.h, s0 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, s1 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v1.l, v3.l, v1.l, s2 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: test_fmax_legacy_ugt_v4f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v3 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v1 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v7, 16, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_3) -; GFX11-SAFE-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v5, v4 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v4, v4, v5, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v7, v6 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v5, v6, v7, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v0, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v1, v3 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v5, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v1, v4, v1, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NNAN-LABEL: test_fmax_legacy_ugt_v4f16: -; GFX11-NNAN: ; %bb.0: -; GFX11-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NNAN-NEXT: v_pk_max_f16 v0, v0, v2 -; GFX11-NNAN-NEXT: v_pk_max_f16 v1, v1, v3 -; GFX11-NNAN-NEXT: s_setpc_b64 s[30:31] +; GFX9-LABEL: test_fmax_legacy_ugt_v4f16: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_lshrrev_b32_e32 v6, 16, v3 +; GFX9-NEXT: v_lshrrev_b32_e32 v7, 16, v1 +; GFX9-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; GFX9-NEXT: v_lshrrev_b32_e32 v5, 16, v0 +; GFX9-NEXT: v_cmp_nle_f16_e32 vcc, v7, v6 +; GFX9-NEXT: v_cndmask_b32_e32 v6, v6, v7, vcc +; GFX9-NEXT: v_cmp_nle_f16_e32 vcc, v5, v4 +; GFX9-NEXT: v_cndmask_b32_e32 v4, v4, v5, vcc +; GFX9-NEXT: v_cmp_nle_f16_e32 vcc, v1, v3 +; GFX9-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc +; GFX9-NEXT: v_cmp_nle_f16_e32 vcc, v0, v2 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc +; GFX9-NEXT: s_mov_b32 s4, 0x5040100 +; GFX9-NEXT: v_perm_b32 v0, v4, v0, s4 +; GFX9-NEXT: v_perm_b32 v1, v6, v1, s4 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; VI-LABEL: test_fmax_legacy_ugt_v4f16: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_lshrrev_b32_e32 v6, 16, v3 +; VI-NEXT: v_lshrrev_b32_e32 v7, 16, v1 +; VI-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; VI-NEXT: v_lshrrev_b32_e32 v5, 16, v0 +; VI-NEXT: v_cmp_nle_f16_e32 vcc, v7, v6 +; VI-NEXT: v_cndmask_b32_e32 v6, v6, v7, vcc +; VI-NEXT: v_cmp_nle_f16_e32 vcc, v5, v4 +; VI-NEXT: v_cndmask_b32_e32 v4, v4, v5, vcc +; VI-NEXT: v_cmp_nle_f16_e32 vcc, v1, v3 +; VI-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc +; VI-NEXT: v_cmp_nle_f16_e32 vcc, v0, v2 +; VI-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc +; VI-NEXT: v_lshlrev_b32_e32 v2, 16, v4 +; VI-NEXT: v_or_b32_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD +; VI-NEXT: v_lshlrev_b32_e32 v2, 16, v6 +; VI-NEXT: v_or_b32_sdwa v1, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD +; VI-NEXT: s_setpc_b64 s[30:31] +; +; SI-LABEL: test_fmax_legacy_ugt_v4f16: +; SI: ; %bb.0: +; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; SI-NEXT: v_cvt_f16_f32_e32 v3, v3 +; SI-NEXT: v_cvt_f16_f32_e32 v7, v7 +; SI-NEXT: v_cvt_f16_f32_e32 v2, v2 +; SI-NEXT: v_cvt_f16_f32_e32 v6, v6 +; SI-NEXT: v_cvt_f16_f32_e32 v1, v1 +; SI-NEXT: v_cvt_f16_f32_e32 v5, v5 +; SI-NEXT: v_cvt_f16_f32_e32 v0, v0 +; SI-NEXT: v_cvt_f16_f32_e32 v4, v4 +; SI-NEXT: v_cvt_f32_f16_e32 v3, v3 +; SI-NEXT: v_cvt_f32_f16_e32 v7, v7 +; SI-NEXT: v_cvt_f32_f16_e32 v2, v2 +; SI-NEXT: v_cvt_f32_f16_e32 v6, v6 +; SI-NEXT: v_cvt_f32_f16_e32 v1, v1 +; SI-NEXT: v_cvt_f32_f16_e32 v5, v5 +; SI-NEXT: v_cvt_f32_f16_e32 v0, v0 +; SI-NEXT: v_cvt_f32_f16_e32 v4, v4 +; SI-NEXT: v_max_legacy_f32_e32 v0, v4, v0 +; SI-NEXT: v_max_legacy_f32_e32 v1, v5, v1 +; SI-NEXT: v_max_legacy_f32_e32 v2, v6, v2 +; SI-NEXT: v_max_legacy_f32_e32 v3, v7, v3 +; SI-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-TRUE16-LABEL: test_fmax_legacy_ugt_v4f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v1.h, v3.h +; GFX11-TRUE16-NEXT: v_cmp_nle_f16_e64 s0, v0.h, v2.h +; GFX11-TRUE16-NEXT: v_cmp_nle_f16_e64 s1, v0.l, v2.l +; GFX11-TRUE16-NEXT: v_cmp_nle_f16_e64 s2, v1.l, v3.l +; GFX11-TRUE16-NEXT: v_cndmask_b16 v1.h, v3.h, v1.h, vcc_lo +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v0.h, s0 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, s1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v1.l, v3.l, v1.l, s2 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: test_fmax_legacy_ugt_v4f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v3 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v1 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v7, 16, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_3) +; GFX11-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v5, v4 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v4, v4, v5, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v7, v6 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v5, v6, v7, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v0, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v1, v3 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc_lo +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v5, v0, 0x5040100 +; GFX11-FAKE16-NEXT: v_perm_b32 v1, v4, v1, 0x5040100 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = fcmp ugt <4 x half> %a, %b %val = select <4 x i1> %cmp, <4 x half> %a, <4 x half> %b ret <4 x half> %val } +define <4 x half> @test_fmax_legacy_ugt_v4f16_fast(<4 x half> %a, <4 x half> %b) #0 { +; GFX9-LABEL: test_fmax_legacy_ugt_v4f16_fast: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_pk_max_f16 v0, v0, v2 +; GFX9-NEXT: v_pk_max_f16 v1, v1, v3 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; VI-LABEL: test_fmax_legacy_ugt_v4f16_fast: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_max_f16_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; VI-NEXT: v_max_f16_sdwa v5, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; VI-NEXT: v_max_f16_e32 v1, v1, v3 +; VI-NEXT: v_max_f16_e32 v0, v0, v2 +; VI-NEXT: v_or_b32_e32 v0, v0, v5 +; VI-NEXT: v_or_b32_e32 v1, v1, v4 +; VI-NEXT: s_setpc_b64 s[30:31] +; +; SI-LABEL: test_fmax_legacy_ugt_v4f16_fast: +; SI: ; %bb.0: +; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; SI-NEXT: v_cvt_f16_f32_e32 v7, v7 +; SI-NEXT: v_cvt_f16_f32_e32 v3, v3 +; SI-NEXT: v_cvt_f16_f32_e32 v6, v6 +; SI-NEXT: v_cvt_f16_f32_e32 v2, v2 +; SI-NEXT: v_cvt_f16_f32_e32 v5, v5 +; SI-NEXT: v_cvt_f16_f32_e32 v1, v1 +; SI-NEXT: v_cvt_f16_f32_e32 v4, v4 +; SI-NEXT: v_cvt_f16_f32_e32 v0, v0 +; SI-NEXT: v_cvt_f32_f16_e32 v7, v7 +; SI-NEXT: v_cvt_f32_f16_e32 v3, v3 +; SI-NEXT: v_cvt_f32_f16_e32 v6, v6 +; SI-NEXT: v_cvt_f32_f16_e32 v2, v2 +; SI-NEXT: v_cvt_f32_f16_e32 v5, v5 +; SI-NEXT: v_cvt_f32_f16_e32 v1, v1 +; SI-NEXT: v_cvt_f32_f16_e32 v4, v4 +; SI-NEXT: v_cvt_f32_f16_e32 v0, v0 +; SI-NEXT: v_max_f32_e32 v0, v0, v4 +; SI-NEXT: v_max_f32_e32 v1, v1, v5 +; SI-NEXT: v_max_f32_e32 v2, v2, v6 +; SI-NEXT: v_max_f32_e32 v3, v3, v7 +; SI-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-LABEL: test_fmax_legacy_ugt_v4f16_fast: +; GFX11: ; %bb.0: +; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-NEXT: v_pk_max_f16 v0, v0, v2 +; GFX11-NEXT: v_pk_max_f16 v1, v1, v3 +; GFX11-NEXT: s_setpc_b64 s[30:31] + %cmp = fcmp ugt <4 x half> %a, %b + %val = select nnan nsz <4 x i1> %cmp, <4 x half> %a, <4 x half> %b + ret <4 x half> %val +} + define <8 x half> @test_fmax_legacy_ugt_v8f16(<8 x half> %a, <8 x half> %b) #0 { -; GFX9-SAFE-LABEL: test_fmax_legacy_ugt_v8f16: -; GFX9-SAFE: ; %bb.0: -; GFX9-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v14, 16, v7 -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v15, 16, v3 -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v12, 16, v6 -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v13, 16, v2 -; GFX9-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v15, v14 -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v10, 16, v5 -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v11, 16, v1 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v14, v14, v15, vcc -; GFX9-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v13, v12 -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v8, 16, v4 -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v9, 16, v0 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v12, v12, v13, vcc -; GFX9-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v11, v10 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v10, v10, v11, vcc -; GFX9-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v9, v8 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v8, v8, v9, vcc -; GFX9-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v3, v7 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v3, v7, v3, vcc -; GFX9-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v2, v6 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v2, v6, v2, vcc -; GFX9-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v1, v5 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v1, v5, v1, vcc -; GFX9-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v0, v4 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v0, v4, v0, vcc -; GFX9-SAFE-NEXT: s_mov_b32 s4, 0x5040100 -; GFX9-SAFE-NEXT: v_perm_b32 v0, v8, v0, s4 -; GFX9-SAFE-NEXT: v_perm_b32 v1, v10, v1, s4 -; GFX9-SAFE-NEXT: v_perm_b32 v2, v12, v2, s4 -; GFX9-SAFE-NEXT: v_perm_b32 v3, v14, v3, s4 -; GFX9-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; GFX9-NNAN-LABEL: test_fmax_legacy_ugt_v8f16: -; GFX9-NNAN: ; %bb.0: -; GFX9-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-NNAN-NEXT: v_pk_max_f16 v0, v0, v4 -; GFX9-NNAN-NEXT: v_pk_max_f16 v1, v1, v5 -; GFX9-NNAN-NEXT: v_pk_max_f16 v2, v2, v6 -; GFX9-NNAN-NEXT: v_pk_max_f16 v3, v3, v7 -; GFX9-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; VI-SAFE-LABEL: test_fmax_legacy_ugt_v8f16: -; VI-SAFE: ; %bb.0: -; VI-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v14, 16, v7 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v15, 16, v3 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v12, 16, v6 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v13, 16, v2 -; VI-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v15, v14 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v10, 16, v5 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v11, 16, v1 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v14, v14, v15, vcc -; VI-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v13, v12 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v8, 16, v4 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v9, 16, v0 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v12, v12, v13, vcc -; VI-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v11, v10 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v10, v10, v11, vcc -; VI-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v9, v8 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v8, v8, v9, vcc -; VI-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v3, v7 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v3, v7, v3, vcc -; VI-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v2, v6 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v2, v6, v2, vcc -; VI-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v1, v5 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v1, v5, v1, vcc -; VI-SAFE-NEXT: v_cmp_nle_f16_e32 vcc, v0, v4 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v0, v4, v0, vcc -; VI-SAFE-NEXT: v_lshlrev_b32_e32 v4, 16, v8 -; VI-SAFE-NEXT: v_or_b32_sdwa v0, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD -; VI-SAFE-NEXT: v_lshlrev_b32_e32 v4, 16, v10 -; VI-SAFE-NEXT: v_or_b32_sdwa v1, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD -; VI-SAFE-NEXT: v_lshlrev_b32_e32 v4, 16, v12 -; VI-SAFE-NEXT: v_or_b32_sdwa v2, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD -; VI-SAFE-NEXT: v_lshlrev_b32_e32 v4, 16, v14 -; VI-SAFE-NEXT: v_or_b32_sdwa v3, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD -; VI-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; VI-NNAN-LABEL: test_fmax_legacy_ugt_v8f16: -; VI-NNAN: ; %bb.0: -; VI-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-NNAN-NEXT: v_max_f16_sdwa v8, v3, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; VI-NNAN-NEXT: v_max_f16_sdwa v9, v2, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; VI-NNAN-NEXT: v_max_f16_sdwa v10, v1, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; VI-NNAN-NEXT: v_max_f16_sdwa v11, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; VI-NNAN-NEXT: v_max_f16_e32 v3, v3, v7 -; VI-NNAN-NEXT: v_max_f16_e32 v2, v2, v6 -; VI-NNAN-NEXT: v_max_f16_e32 v1, v1, v5 -; VI-NNAN-NEXT: v_max_f16_e32 v0, v0, v4 -; VI-NNAN-NEXT: v_or_b32_e32 v0, v0, v11 -; VI-NNAN-NEXT: v_or_b32_e32 v1, v1, v10 -; VI-NNAN-NEXT: v_or_b32_e32 v2, v2, v9 -; VI-NNAN-NEXT: v_or_b32_e32 v3, v3, v8 -; VI-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; SI-SAFE-LABEL: test_fmax_legacy_ugt_v8f16: -; SI-SAFE: ; %bb.0: -; SI-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v7, v7 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v15, v15 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v6, v6 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v14, v14 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v5, v5 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v13, v13 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v4, v4 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v12, v12 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v3, v3 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v11, v11 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v2, v2 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v10, v10 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v1, v1 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v9, v9 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v0, v0 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v8, v8 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v7, v7 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v15, v15 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v6, v6 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v14, v14 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v5, v5 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v13, v13 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v4, v4 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v12, v12 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v3, v3 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v11, v11 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v2, v2 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v10, v10 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v1, v1 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v9, v9 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v8, v8 -; SI-SAFE-NEXT: v_max_legacy_f32_e32 v0, v8, v0 -; SI-SAFE-NEXT: v_max_legacy_f32_e32 v1, v9, v1 -; SI-SAFE-NEXT: v_max_legacy_f32_e32 v2, v10, v2 -; SI-SAFE-NEXT: v_max_legacy_f32_e32 v3, v11, v3 -; SI-SAFE-NEXT: v_max_legacy_f32_e32 v4, v12, v4 -; SI-SAFE-NEXT: v_max_legacy_f32_e32 v5, v13, v5 -; SI-SAFE-NEXT: v_max_legacy_f32_e32 v6, v14, v6 -; SI-SAFE-NEXT: v_max_legacy_f32_e32 v7, v15, v7 -; SI-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; SI-NNAN-LABEL: test_fmax_legacy_ugt_v8f16: -; SI-NNAN: ; %bb.0: -; SI-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v15, v15 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v7, v7 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v14, v14 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v6, v6 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v13, v13 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v5, v5 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v12, v12 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v4, v4 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v11, v11 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v3, v3 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v10, v10 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v2, v2 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v9, v9 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v1, v1 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v8, v8 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v0, v0 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v15, v15 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v7, v7 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v14, v14 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v6, v6 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v13, v13 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v5, v5 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v12, v12 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v4, v4 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v11, v11 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v3, v3 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v10, v10 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v2, v2 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v9, v9 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v1, v1 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v8, v8 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-NNAN-NEXT: v_max_f32_e32 v0, v0, v8 -; SI-NNAN-NEXT: v_max_f32_e32 v1, v1, v9 -; SI-NNAN-NEXT: v_max_f32_e32 v2, v2, v10 -; SI-NNAN-NEXT: v_max_f32_e32 v3, v3, v11 -; SI-NNAN-NEXT: v_max_f32_e32 v4, v4, v12 -; SI-NNAN-NEXT: v_max_f32_e32 v5, v5, v13 -; SI-NNAN-NEXT: v_max_f32_e32 v6, v6, v14 -; SI-NNAN-NEXT: v_max_f32_e32 v7, v7, v15 -; SI-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-TRUE16-LABEL: test_fmax_legacy_ugt_v8f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v0.h, v4.h -; GFX11-SAFE-TRUE16-NEXT: v_cmp_nle_f16_e64 s0, v1.h, v5.h -; GFX11-SAFE-TRUE16-NEXT: v_cmp_nle_f16_e64 s1, v2.h, v6.h -; GFX11-SAFE-TRUE16-NEXT: v_cmp_nle_f16_e64 s2, v3.h, v7.h -; GFX11-SAFE-TRUE16-NEXT: v_cmp_nle_f16_e64 s3, v0.l, v4.l -; GFX11-SAFE-TRUE16-NEXT: v_cmp_nle_f16_e64 s4, v1.l, v5.l -; GFX11-SAFE-TRUE16-NEXT: v_cmp_nle_f16_e64 s5, v2.l, v6.l -; GFX11-SAFE-TRUE16-NEXT: v_cmp_nle_f16_e64 s6, v3.l, v7.l -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v3.h, v7.h, v3.h, s2 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v2.h, v6.h, v2.h, s1 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v1.h, v5.h, v1.h, s0 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v4.h, v0.h, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v4.l, v0.l, s3 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v1.l, v5.l, v1.l, s4 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v2.l, v6.l, v2.l, s5 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v3.l, v7.l, v3.l, s6 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: test_fmax_legacy_ugt_v8f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v10, 16, v7 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v11, 16, v3 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v12, 16, v6 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v13, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v14, 16, v5 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v15, 16, v1 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v11, v10 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v8, 16, v4 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v9, 16, v0 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v10, v10, v11, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v13, v12 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v11, v12, v13, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v15, v14 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v12, v14, v15, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v9, v8 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v8, v8, v9, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v2, v6 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v2, v6, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v0, v4 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(VALU_DEP_2) -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v2, v11, v2, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, v4, v0, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v1, v5 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, v5, v1, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v3, v7 -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v1, v12, v1, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v3, v7, v3, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v8, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v3, v10, v3, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NNAN-LABEL: test_fmax_legacy_ugt_v8f16: -; GFX11-NNAN: ; %bb.0: -; GFX11-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NNAN-NEXT: v_pk_max_f16 v0, v0, v4 -; GFX11-NNAN-NEXT: v_pk_max_f16 v1, v1, v5 -; GFX11-NNAN-NEXT: v_pk_max_f16 v2, v2, v6 -; GFX11-NNAN-NEXT: v_pk_max_f16 v3, v3, v7 -; GFX11-NNAN-NEXT: s_setpc_b64 s[30:31] +; GFX9-LABEL: test_fmax_legacy_ugt_v8f16: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_lshrrev_b32_e32 v14, 16, v7 +; GFX9-NEXT: v_lshrrev_b32_e32 v15, 16, v3 +; GFX9-NEXT: v_lshrrev_b32_e32 v12, 16, v6 +; GFX9-NEXT: v_lshrrev_b32_e32 v13, 16, v2 +; GFX9-NEXT: v_cmp_nle_f16_e32 vcc, v15, v14 +; GFX9-NEXT: v_lshrrev_b32_e32 v10, 16, v5 +; GFX9-NEXT: v_lshrrev_b32_e32 v11, 16, v1 +; GFX9-NEXT: v_cndmask_b32_e32 v14, v14, v15, vcc +; GFX9-NEXT: v_cmp_nle_f16_e32 vcc, v13, v12 +; GFX9-NEXT: v_lshrrev_b32_e32 v8, 16, v4 +; GFX9-NEXT: v_lshrrev_b32_e32 v9, 16, v0 +; GFX9-NEXT: v_cndmask_b32_e32 v12, v12, v13, vcc +; GFX9-NEXT: v_cmp_nle_f16_e32 vcc, v11, v10 +; GFX9-NEXT: v_cndmask_b32_e32 v10, v10, v11, vcc +; GFX9-NEXT: v_cmp_nle_f16_e32 vcc, v9, v8 +; GFX9-NEXT: v_cndmask_b32_e32 v8, v8, v9, vcc +; GFX9-NEXT: v_cmp_nle_f16_e32 vcc, v3, v7 +; GFX9-NEXT: v_cndmask_b32_e32 v3, v7, v3, vcc +; GFX9-NEXT: v_cmp_nle_f16_e32 vcc, v2, v6 +; GFX9-NEXT: v_cndmask_b32_e32 v2, v6, v2, vcc +; GFX9-NEXT: v_cmp_nle_f16_e32 vcc, v1, v5 +; GFX9-NEXT: v_cndmask_b32_e32 v1, v5, v1, vcc +; GFX9-NEXT: v_cmp_nle_f16_e32 vcc, v0, v4 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v4, v0, vcc +; GFX9-NEXT: s_mov_b32 s4, 0x5040100 +; GFX9-NEXT: v_perm_b32 v0, v8, v0, s4 +; GFX9-NEXT: v_perm_b32 v1, v10, v1, s4 +; GFX9-NEXT: v_perm_b32 v2, v12, v2, s4 +; GFX9-NEXT: v_perm_b32 v3, v14, v3, s4 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; VI-LABEL: test_fmax_legacy_ugt_v8f16: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_lshrrev_b32_e32 v14, 16, v7 +; VI-NEXT: v_lshrrev_b32_e32 v15, 16, v3 +; VI-NEXT: v_lshrrev_b32_e32 v12, 16, v6 +; VI-NEXT: v_lshrrev_b32_e32 v13, 16, v2 +; VI-NEXT: v_cmp_nle_f16_e32 vcc, v15, v14 +; VI-NEXT: v_lshrrev_b32_e32 v10, 16, v5 +; VI-NEXT: v_lshrrev_b32_e32 v11, 16, v1 +; VI-NEXT: v_cndmask_b32_e32 v14, v14, v15, vcc +; VI-NEXT: v_cmp_nle_f16_e32 vcc, v13, v12 +; VI-NEXT: v_lshrrev_b32_e32 v8, 16, v4 +; VI-NEXT: v_lshrrev_b32_e32 v9, 16, v0 +; VI-NEXT: v_cndmask_b32_e32 v12, v12, v13, vcc +; VI-NEXT: v_cmp_nle_f16_e32 vcc, v11, v10 +; VI-NEXT: v_cndmask_b32_e32 v10, v10, v11, vcc +; VI-NEXT: v_cmp_nle_f16_e32 vcc, v9, v8 +; VI-NEXT: v_cndmask_b32_e32 v8, v8, v9, vcc +; VI-NEXT: v_cmp_nle_f16_e32 vcc, v3, v7 +; VI-NEXT: v_cndmask_b32_e32 v3, v7, v3, vcc +; VI-NEXT: v_cmp_nle_f16_e32 vcc, v2, v6 +; VI-NEXT: v_cndmask_b32_e32 v2, v6, v2, vcc +; VI-NEXT: v_cmp_nle_f16_e32 vcc, v1, v5 +; VI-NEXT: v_cndmask_b32_e32 v1, v5, v1, vcc +; VI-NEXT: v_cmp_nle_f16_e32 vcc, v0, v4 +; VI-NEXT: v_cndmask_b32_e32 v0, v4, v0, vcc +; VI-NEXT: v_lshlrev_b32_e32 v4, 16, v8 +; VI-NEXT: v_or_b32_sdwa v0, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD +; VI-NEXT: v_lshlrev_b32_e32 v4, 16, v10 +; VI-NEXT: v_or_b32_sdwa v1, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD +; VI-NEXT: v_lshlrev_b32_e32 v4, 16, v12 +; VI-NEXT: v_or_b32_sdwa v2, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD +; VI-NEXT: v_lshlrev_b32_e32 v4, 16, v14 +; VI-NEXT: v_or_b32_sdwa v3, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD +; VI-NEXT: s_setpc_b64 s[30:31] +; +; SI-LABEL: test_fmax_legacy_ugt_v8f16: +; SI: ; %bb.0: +; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; SI-NEXT: v_cvt_f16_f32_e32 v7, v7 +; SI-NEXT: v_cvt_f16_f32_e32 v15, v15 +; SI-NEXT: v_cvt_f16_f32_e32 v6, v6 +; SI-NEXT: v_cvt_f16_f32_e32 v14, v14 +; SI-NEXT: v_cvt_f16_f32_e32 v5, v5 +; SI-NEXT: v_cvt_f16_f32_e32 v13, v13 +; SI-NEXT: v_cvt_f16_f32_e32 v4, v4 +; SI-NEXT: v_cvt_f16_f32_e32 v12, v12 +; SI-NEXT: v_cvt_f16_f32_e32 v3, v3 +; SI-NEXT: v_cvt_f16_f32_e32 v11, v11 +; SI-NEXT: v_cvt_f16_f32_e32 v2, v2 +; SI-NEXT: v_cvt_f16_f32_e32 v10, v10 +; SI-NEXT: v_cvt_f16_f32_e32 v1, v1 +; SI-NEXT: v_cvt_f16_f32_e32 v9, v9 +; SI-NEXT: v_cvt_f16_f32_e32 v0, v0 +; SI-NEXT: v_cvt_f16_f32_e32 v8, v8 +; SI-NEXT: v_cvt_f32_f16_e32 v7, v7 +; SI-NEXT: v_cvt_f32_f16_e32 v15, v15 +; SI-NEXT: v_cvt_f32_f16_e32 v6, v6 +; SI-NEXT: v_cvt_f32_f16_e32 v14, v14 +; SI-NEXT: v_cvt_f32_f16_e32 v5, v5 +; SI-NEXT: v_cvt_f32_f16_e32 v13, v13 +; SI-NEXT: v_cvt_f32_f16_e32 v4, v4 +; SI-NEXT: v_cvt_f32_f16_e32 v12, v12 +; SI-NEXT: v_cvt_f32_f16_e32 v3, v3 +; SI-NEXT: v_cvt_f32_f16_e32 v11, v11 +; SI-NEXT: v_cvt_f32_f16_e32 v2, v2 +; SI-NEXT: v_cvt_f32_f16_e32 v10, v10 +; SI-NEXT: v_cvt_f32_f16_e32 v1, v1 +; SI-NEXT: v_cvt_f32_f16_e32 v9, v9 +; SI-NEXT: v_cvt_f32_f16_e32 v0, v0 +; SI-NEXT: v_cvt_f32_f16_e32 v8, v8 +; SI-NEXT: v_max_legacy_f32_e32 v0, v8, v0 +; SI-NEXT: v_max_legacy_f32_e32 v1, v9, v1 +; SI-NEXT: v_max_legacy_f32_e32 v2, v10, v2 +; SI-NEXT: v_max_legacy_f32_e32 v3, v11, v3 +; SI-NEXT: v_max_legacy_f32_e32 v4, v12, v4 +; SI-NEXT: v_max_legacy_f32_e32 v5, v13, v5 +; SI-NEXT: v_max_legacy_f32_e32 v6, v14, v6 +; SI-NEXT: v_max_legacy_f32_e32 v7, v15, v7 +; SI-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-TRUE16-LABEL: test_fmax_legacy_ugt_v8f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v0.h, v4.h +; GFX11-TRUE16-NEXT: v_cmp_nle_f16_e64 s0, v1.h, v5.h +; GFX11-TRUE16-NEXT: v_cmp_nle_f16_e64 s1, v2.h, v6.h +; GFX11-TRUE16-NEXT: v_cmp_nle_f16_e64 s2, v3.h, v7.h +; GFX11-TRUE16-NEXT: v_cmp_nle_f16_e64 s3, v0.l, v4.l +; GFX11-TRUE16-NEXT: v_cmp_nle_f16_e64 s4, v1.l, v5.l +; GFX11-TRUE16-NEXT: v_cmp_nle_f16_e64 s5, v2.l, v6.l +; GFX11-TRUE16-NEXT: v_cmp_nle_f16_e64 s6, v3.l, v7.l +; GFX11-TRUE16-NEXT: v_cndmask_b16 v3.h, v7.h, v3.h, s2 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v2.h, v6.h, v2.h, s1 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v1.h, v5.h, v1.h, s0 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v4.h, v0.h, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v4.l, v0.l, s3 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v1.l, v5.l, v1.l, s4 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v2.l, v6.l, v2.l, s5 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v3.l, v7.l, v3.l, s6 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: test_fmax_legacy_ugt_v8f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v10, 16, v7 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v11, 16, v3 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v12, 16, v6 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v13, 16, v2 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v14, 16, v5 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v15, 16, v1 +; GFX11-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v11, v10 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v8, 16, v4 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v9, 16, v0 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v10, v10, v11, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v13, v12 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v11, v12, v13, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v15, v14 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v12, v14, v15, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v9, v8 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v8, v8, v9, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v2, v6 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v2, v6, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v0, v4 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(VALU_DEP_2) +; GFX11-FAKE16-NEXT: v_perm_b32 v2, v11, v2, 0x5040100 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, v4, v0, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v1, v5 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, v5, v1, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_nle_f16_e32 vcc_lo, v3, v7 +; GFX11-FAKE16-NEXT: v_perm_b32 v1, v12, v1, 0x5040100 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v3, v7, v3, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v8, v0, 0x5040100 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX11-FAKE16-NEXT: v_perm_b32 v3, v10, v3, 0x5040100 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = fcmp ugt <8 x half> %a, %b %val = select <8 x i1> %cmp, <8 x half> %a, <8 x half> %b ret <8 x half> %val } +define <8 x half> @test_fmax_legacy_ugt_v8f16_fast(<8 x half> %a, <8 x half> %b) #0 { +; GFX9-LABEL: test_fmax_legacy_ugt_v8f16_fast: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_pk_max_f16 v0, v0, v4 +; GFX9-NEXT: v_pk_max_f16 v1, v1, v5 +; GFX9-NEXT: v_pk_max_f16 v2, v2, v6 +; GFX9-NEXT: v_pk_max_f16 v3, v3, v7 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; VI-LABEL: test_fmax_legacy_ugt_v8f16_fast: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_max_f16_sdwa v8, v3, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; VI-NEXT: v_max_f16_sdwa v9, v2, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; VI-NEXT: v_max_f16_sdwa v10, v1, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; VI-NEXT: v_max_f16_sdwa v11, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; VI-NEXT: v_max_f16_e32 v3, v3, v7 +; VI-NEXT: v_max_f16_e32 v2, v2, v6 +; VI-NEXT: v_max_f16_e32 v1, v1, v5 +; VI-NEXT: v_max_f16_e32 v0, v0, v4 +; VI-NEXT: v_or_b32_e32 v0, v0, v11 +; VI-NEXT: v_or_b32_e32 v1, v1, v10 +; VI-NEXT: v_or_b32_e32 v2, v2, v9 +; VI-NEXT: v_or_b32_e32 v3, v3, v8 +; VI-NEXT: s_setpc_b64 s[30:31] +; +; SI-LABEL: test_fmax_legacy_ugt_v8f16_fast: +; SI: ; %bb.0: +; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; SI-NEXT: v_cvt_f16_f32_e32 v15, v15 +; SI-NEXT: v_cvt_f16_f32_e32 v7, v7 +; SI-NEXT: v_cvt_f16_f32_e32 v14, v14 +; SI-NEXT: v_cvt_f16_f32_e32 v6, v6 +; SI-NEXT: v_cvt_f16_f32_e32 v13, v13 +; SI-NEXT: v_cvt_f16_f32_e32 v5, v5 +; SI-NEXT: v_cvt_f16_f32_e32 v12, v12 +; SI-NEXT: v_cvt_f16_f32_e32 v4, v4 +; SI-NEXT: v_cvt_f16_f32_e32 v11, v11 +; SI-NEXT: v_cvt_f16_f32_e32 v3, v3 +; SI-NEXT: v_cvt_f16_f32_e32 v10, v10 +; SI-NEXT: v_cvt_f16_f32_e32 v2, v2 +; SI-NEXT: v_cvt_f16_f32_e32 v9, v9 +; SI-NEXT: v_cvt_f16_f32_e32 v1, v1 +; SI-NEXT: v_cvt_f16_f32_e32 v8, v8 +; SI-NEXT: v_cvt_f16_f32_e32 v0, v0 +; SI-NEXT: v_cvt_f32_f16_e32 v15, v15 +; SI-NEXT: v_cvt_f32_f16_e32 v7, v7 +; SI-NEXT: v_cvt_f32_f16_e32 v14, v14 +; SI-NEXT: v_cvt_f32_f16_e32 v6, v6 +; SI-NEXT: v_cvt_f32_f16_e32 v13, v13 +; SI-NEXT: v_cvt_f32_f16_e32 v5, v5 +; SI-NEXT: v_cvt_f32_f16_e32 v12, v12 +; SI-NEXT: v_cvt_f32_f16_e32 v4, v4 +; SI-NEXT: v_cvt_f32_f16_e32 v11, v11 +; SI-NEXT: v_cvt_f32_f16_e32 v3, v3 +; SI-NEXT: v_cvt_f32_f16_e32 v10, v10 +; SI-NEXT: v_cvt_f32_f16_e32 v2, v2 +; SI-NEXT: v_cvt_f32_f16_e32 v9, v9 +; SI-NEXT: v_cvt_f32_f16_e32 v1, v1 +; SI-NEXT: v_cvt_f32_f16_e32 v8, v8 +; SI-NEXT: v_cvt_f32_f16_e32 v0, v0 +; SI-NEXT: v_max_f32_e32 v0, v0, v8 +; SI-NEXT: v_max_f32_e32 v1, v1, v9 +; SI-NEXT: v_max_f32_e32 v2, v2, v10 +; SI-NEXT: v_max_f32_e32 v3, v3, v11 +; SI-NEXT: v_max_f32_e32 v4, v4, v12 +; SI-NEXT: v_max_f32_e32 v5, v5, v13 +; SI-NEXT: v_max_f32_e32 v6, v6, v14 +; SI-NEXT: v_max_f32_e32 v7, v7, v15 +; SI-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-LABEL: test_fmax_legacy_ugt_v8f16_fast: +; GFX11: ; %bb.0: +; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-NEXT: v_pk_max_f16 v0, v0, v4 +; GFX11-NEXT: v_pk_max_f16 v1, v1, v5 +; GFX11-NEXT: v_pk_max_f16 v2, v2, v6 +; GFX11-NEXT: v_pk_max_f16 v3, v3, v7 +; GFX11-NEXT: s_setpc_b64 s[30:31] + %cmp = fcmp ugt <8 x half> %a, %b + %val = select nnan nsz <8 x i1> %cmp, <8 x half> %a, <8 x half> %b + ret <8 x half> %val +} + attributes #0 = { nounwind } diff --git a/llvm/test/CodeGen/AMDGPU/fmax_legacy.ll b/llvm/test/CodeGen/AMDGPU/fmax_legacy.ll index eee2bd1..f3a84e6 100644 --- a/llvm/test/CodeGen/AMDGPU/fmax_legacy.ll +++ b/llvm/test/CodeGen/AMDGPU/fmax_legacy.ll @@ -1,8 +1,6 @@ -; RUN: llc -mtriple=amdgcn < %s | FileCheck -enable-var-scope -check-prefixes=SI-SAFE,GCN,FUNC %s -; RUN: llc -enable-no-nans-fp-math -enable-no-signed-zeros-fp-math -mtriple=amdgcn < %s | FileCheck -enable-var-scope -check-prefixes=GCN-NONAN,GCN,FUNC %s +; RUN: llc -mtriple=amdgcn < %s | FileCheck -enable-var-scope -check-prefixes=SI,GCN,FUNC %s -; RUN: llc -mtriple=amdgcn -mcpu=fiji < %s | FileCheck -enable-var-scope -check-prefixes=VI-SAFE,GCN,FUNC %s -; RUN: llc -enable-no-nans-fp-math -enable-no-signed-zeros-fp-math -mtriple=amdgcn -mcpu=fiji < %s | FileCheck -enable-var-scope -check-prefixes=GCN-NONAN,GCN,FUNC %s +; RUN: llc -mtriple=amdgcn -mcpu=fiji < %s | FileCheck -enable-var-scope -check-prefixes=VI,GCN,FUNC %s ; RUN: llc -mtriple=r600 -mcpu=redwood < %s | FileCheck -enable-var-scope --check-prefixes=EG,FUNC %s @@ -12,12 +10,10 @@ declare i32 @llvm.amdgcn.workitem.id.x() #1 ; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]] ; GCN: {{buffer|flat}}_load_dword [[B:v[0-9]+]] -; SI-SAFE: v_max_legacy_f32_e32 {{v[0-9]+}}, [[B]], [[A]] +; SI: v_max_legacy_f32_e32 {{v[0-9]+}}, [[B]], [[A]] -; VI-SAFE: v_cmp_nlt_f32_e32 vcc, [[A]], [[B]] -; VI-SAFE: v_cndmask_b32_e32 v{{[0-9]+}}, [[B]], [[A]] - -; GCN-NONAN: v_max_f32_e32 {{v[0-9]+}}, [[A]], [[B]] +; VI: v_cmp_nlt_f32_e32 vcc, [[A]], [[B]] +; VI: v_cndmask_b32_e32 v{{[0-9]+}}, [[B]], [[A]] ; EG: MAX define amdgpu_kernel void @test_fmax_legacy_uge_f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { @@ -34,18 +30,38 @@ define amdgpu_kernel void @test_fmax_legacy_uge_f32(ptr addrspace(1) %out, ptr a ret void } +; FUNC-LABEL: {{^}}test_fmax_legacy_uge_f32_fast: +; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]] +; GCN: {{buffer|flat}}_load_dword [[B:v[0-9]+]] + +; GCN: v_max_f32_e32 {{v[0-9]+}}, [[A]], [[B]] + +; EG: MAX +define amdgpu_kernel void @test_fmax_legacy_uge_f32_fast(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { + %tid = call i32 @llvm.amdgcn.workitem.id.x() #1 + %gep.0 = getelementptr float, ptr addrspace(1) %in, i32 %tid + %gep.1 = getelementptr float, ptr addrspace(1) %gep.0, i32 1 + + %a = load volatile float, ptr addrspace(1) %gep.0, align 4 + %b = load volatile float, ptr addrspace(1) %gep.1, align 4 + + %cmp = fcmp uge float %a, %b + %val = select nnan nsz i1 %cmp, float %a, float %b + store float %val, ptr addrspace(1) %out, align 4 + ret void +} + ; FUNC-LABEL: {{^}}test_fmax_legacy_uge_f32_nnan_src: ; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]] ; GCN: {{buffer|flat}}_load_dword [[B:v[0-9]+]] ; GCN-DAG: v_add_f32_e32 [[ADD_A:v[0-9]+]], 1.0, [[A]] ; GCN-DAG: v_add_f32_e32 [[ADD_B:v[0-9]+]], 2.0, [[B]] -; SI-SAFE: v_max_legacy_f32_e32 {{v[0-9]+}}, [[ADD_B]], [[ADD_A]] +; SI: v_max_legacy_f32_e32 {{v[0-9]+}}, [[ADD_B]], [[ADD_A]] -; VI-SAFE: v_cmp_nlt_f32_e32 vcc, [[ADD_A]], [[ADD_B]] -; VI-SAFE: v_cndmask_b32_e32 v{{[0-9]+}}, [[ADD_B]], [[ADD_A]] +; VI: v_cmp_nlt_f32_e32 vcc, [[ADD_A]], [[ADD_B]] +; VI: v_cndmask_b32_e32 v{{[0-9]+}}, [[ADD_B]], [[ADD_A]] -; GCN-NONAN: v_max_f32_e32 {{v[0-9]+}}, [[ADD_A]], [[ADD_B]] ; EG: MAX define amdgpu_kernel void @test_fmax_legacy_uge_f32_nnan_src(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { @@ -64,16 +80,40 @@ define amdgpu_kernel void @test_fmax_legacy_uge_f32_nnan_src(ptr addrspace(1) %o ret void } +; FUNC-LABEL: {{^}}test_fmax_legacy_uge_f32_nnan_src_fast: +; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]] +; GCN: {{buffer|flat}}_load_dword [[B:v[0-9]+]] +; GCN-DAG: v_add_f32_e32 [[ADD_A:v[0-9]+]], 1.0, [[A]] +; GCN-DAG: v_add_f32_e32 [[ADD_B:v[0-9]+]], 2.0, [[B]] + +; GCN: v_max_f32_e32 {{v[0-9]+}}, [[ADD_A]], [[ADD_B]] + +; EG: MAX +define amdgpu_kernel void @test_fmax_legacy_uge_f32_nnan_src_fast(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { + %tid = call i32 @llvm.amdgcn.workitem.id.x() #1 + %gep.0 = getelementptr float, ptr addrspace(1) %in, i32 %tid + %gep.1 = getelementptr float, ptr addrspace(1) %gep.0, i32 1 + + %a = load volatile float, ptr addrspace(1) %gep.0, align 4 + %b = load volatile float, ptr addrspace(1) %gep.1, align 4 + %a.nnan = fadd nnan float %a, 1.0 + %b.nnan = fadd nnan float %b, 2.0 + + %cmp = fcmp uge float %a.nnan, %b.nnan + %val = select nnan nsz i1 %cmp, float %a.nnan, float %b.nnan + store float %val, ptr addrspace(1) %out, align 4 + ret void +} + ; FUNC-LABEL: {{^}}test_fmax_legacy_oge_f32: ; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]] ; GCN: {{buffer|flat}}_load_dword [[B:v[0-9]+]] -; SI-SAFE: v_max_legacy_f32_e32 {{v[0-9]+}}, [[A]], [[B]] +; SI: v_max_legacy_f32_e32 {{v[0-9]+}}, [[A]], [[B]] -; VI-SAFE: v_cmp_ge_f32_e32 vcc, [[A]], [[B]] -; VI-SAFE: v_cndmask_b32_e32 v{{[0-9]+}}, [[B]], [[A]] +; VI: v_cmp_ge_f32_e32 vcc, [[A]], [[B]] +; VI: v_cndmask_b32_e32 v{{[0-9]+}}, [[B]], [[A]] -; GCN-NONAN: v_max_f32_e32 {{v[0-9]+}}, [[A]], [[B]] ; EG: MAX define amdgpu_kernel void @test_fmax_legacy_oge_f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { %tid = call i32 @llvm.amdgcn.workitem.id.x() #1 @@ -89,17 +129,35 @@ define amdgpu_kernel void @test_fmax_legacy_oge_f32(ptr addrspace(1) %out, ptr a ret void } -; FUNC-LABEL: {{^}}test_fmax_legacy_ugt_f32: +; FUNC-LABEL: {{^}}test_fmax_legacy_oge_f32_fast: ; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]] ; GCN: {{buffer|flat}}_load_dword [[B:v[0-9]+]] -; SI-SAFE: v_max_legacy_f32_e32 {{v[0-9]+}}, [[B]], [[A]] +; GCN: v_max_f32_e32 {{v[0-9]+}}, [[A]], [[B]] +; EG: MAX +define amdgpu_kernel void @test_fmax_legacy_oge_f32_fast(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { + %tid = call i32 @llvm.amdgcn.workitem.id.x() #1 + %gep.0 = getelementptr float, ptr addrspace(1) %in, i32 %tid + %gep.1 = getelementptr float, ptr addrspace(1) %gep.0, i32 1 + + %a = load volatile float, ptr addrspace(1) %gep.0, align 4 + %b = load volatile float, ptr addrspace(1) %gep.1, align 4 -; VI-SAFE: v_cmp_nle_f32_e32 vcc, [[A]], [[B]] -; VI-SAFE: v_cndmask_b32_e32 v{{[0-9]+}}, [[B]], [[A]] + %cmp = fcmp oge float %a, %b + %val = select nnan nsz i1 %cmp, float %a, float %b + store float %val, ptr addrspace(1) %out, align 4 + ret void +} +; FUNC-LABEL: {{^}}test_fmax_legacy_ugt_f32: +; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]] +; GCN: {{buffer|flat}}_load_dword [[B:v[0-9]+]] + +; SI: v_max_legacy_f32_e32 {{v[0-9]+}}, [[B]], [[A]] + +; VI: v_cmp_nle_f32_e32 vcc, [[A]], [[B]] +; VI: v_cndmask_b32_e32 v{{[0-9]+}}, [[B]], [[A]] -; GCN-NONAN: v_max_f32_e32 {{v[0-9]+}}, [[A]], [[B]] ; EG: MAX define amdgpu_kernel void @test_fmax_legacy_ugt_f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { %tid = call i32 @llvm.amdgcn.workitem.id.x() #1 @@ -115,16 +173,35 @@ define amdgpu_kernel void @test_fmax_legacy_ugt_f32(ptr addrspace(1) %out, ptr a ret void } +; FUNC-LABEL: {{^}}test_fmax_legacy_ugt_f32_fast: +; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]] +; GCN: {{buffer|flat}}_load_dword [[B:v[0-9]+]] + +; GCN: v_max_f32_e32 {{v[0-9]+}}, [[A]], [[B]] +; EG: MAX +define amdgpu_kernel void @test_fmax_legacy_ugt_f32_fast(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { + %tid = call i32 @llvm.amdgcn.workitem.id.x() #1 + %gep.0 = getelementptr float, ptr addrspace(1) %in, i32 %tid + %gep.1 = getelementptr float, ptr addrspace(1) %gep.0, i32 1 + + %a = load volatile float, ptr addrspace(1) %gep.0, align 4 + %b = load volatile float, ptr addrspace(1) %gep.1, align 4 + + %cmp = fcmp ugt float %a, %b + %val = select nnan nsz i1 %cmp, float %a, float %b + store float %val, ptr addrspace(1) %out, align 4 + ret void +} + ; FUNC-LABEL: {{^}}test_fmax_legacy_ogt_f32: ; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]] ; GCN: {{buffer|flat}}_load_dword [[B:v[0-9]+]] -; SI-SAFE: v_max_legacy_f32_e32 {{v[0-9]+}}, [[A]], [[B]] +; SI: v_max_legacy_f32_e32 {{v[0-9]+}}, [[A]], [[B]] -; VI-SAFE: v_cmp_gt_f32_e32 vcc, [[A]], [[B]] -; VI-SAFE: v_cndmask_b32_e32 v{{[0-9]+}}, [[B]], [[A]] +; VI: v_cmp_gt_f32_e32 vcc, [[A]], [[B]] +; VI: v_cndmask_b32_e32 v{{[0-9]+}}, [[B]], [[A]] -; GCN-NONAN: v_max_f32_e32 {{v[0-9]+}}, [[A]], [[B]] ; EG: MAX define amdgpu_kernel void @test_fmax_legacy_ogt_f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { %tid = call i32 @llvm.amdgcn.workitem.id.x() #1 @@ -140,17 +217,35 @@ define amdgpu_kernel void @test_fmax_legacy_ogt_f32(ptr addrspace(1) %out, ptr a ret void } -; FUNC-LABEL: {{^}}test_fmax_legacy_ogt_v1f32: +; FUNC-LABEL: {{^}}test_fmax_legacy_ogt_f32_fast: ; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]] ; GCN: {{buffer|flat}}_load_dword [[B:v[0-9]+]] -; SI-SAFE: v_max_legacy_f32_e32 {{v[0-9]+}}, [[A]], [[B]] +; GCN: v_max_f32_e32 {{v[0-9]+}}, [[A]], [[B]] +; EG: MAX +define amdgpu_kernel void @test_fmax_legacy_ogt_f32_fast(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { + %tid = call i32 @llvm.amdgcn.workitem.id.x() #1 + %gep.0 = getelementptr float, ptr addrspace(1) %in, i32 %tid + %gep.1 = getelementptr float, ptr addrspace(1) %gep.0, i32 1 + + %a = load volatile float, ptr addrspace(1) %gep.0, align 4 + %b = load volatile float, ptr addrspace(1) %gep.1, align 4 -; VI-SAFE: v_cmp_gt_f32_e32 vcc, [[A]], [[B]] -; VI-SAFE: v_cndmask_b32_e32 v{{[0-9]+}}, [[B]], [[A]] + %cmp = fcmp ogt float %a, %b + %val = select nnan nsz i1 %cmp, float %a, float %b + store float %val, ptr addrspace(1) %out, align 4 + ret void +} +; FUNC-LABEL: {{^}}test_fmax_legacy_ogt_v1f32: +; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]] +; GCN: {{buffer|flat}}_load_dword [[B:v[0-9]+]] + +; SI: v_max_legacy_f32_e32 {{v[0-9]+}}, [[A]], [[B]] + +; VI: v_cmp_gt_f32_e32 vcc, [[A]], [[B]] +; VI: v_cndmask_b32_e32 v{{[0-9]+}}, [[B]], [[A]] -; GCN-NONAN: v_max_f32_e32 {{v[0-9]+}}, [[A]], [[B]] ; EG: MAX define amdgpu_kernel void @test_fmax_legacy_ogt_v1f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { %tid = call i32 @llvm.amdgcn.workitem.id.x() #1 @@ -166,23 +261,39 @@ define amdgpu_kernel void @test_fmax_legacy_ogt_v1f32(ptr addrspace(1) %out, ptr ret void } +; FUNC-LABEL: {{^}}test_fmax_legacy_ogt_v1f32_fast: +; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]] +; GCN: {{buffer|flat}}_load_dword [[B:v[0-9]+]] + +; GCN: v_max_f32_e32 {{v[0-9]+}}, [[A]], [[B]] +; EG: MAX +define amdgpu_kernel void @test_fmax_legacy_ogt_v1f32_fast(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { + %tid = call i32 @llvm.amdgcn.workitem.id.x() #1 + %gep.0 = getelementptr <1 x float>, ptr addrspace(1) %in, i32 %tid + %gep.1 = getelementptr <1 x float>, ptr addrspace(1) %gep.0, i32 1 + + %a = load volatile <1 x float>, ptr addrspace(1) %gep.0 + %b = load volatile <1 x float>, ptr addrspace(1) %gep.1 + + %cmp = fcmp ogt <1 x float> %a, %b + %val = select nnan nsz <1 x i1> %cmp, <1 x float> %a, <1 x float> %b + store <1 x float> %val, ptr addrspace(1) %out + ret void +} + ; FUNC-LABEL: {{^}}test_fmax_legacy_ogt_v3f32: -; SI-SAFE: v_max_legacy_f32_e32 -; SI-SAFE: v_max_legacy_f32_e32 -; SI-SAFE: v_max_legacy_f32_e32 - -; VI-SAFE: v_cmp_gt_f32_e32 -; VI-SAFE: v_cndmask_b32_e32 -; VI-SAFE: v_cmp_gt_f32_e32 -; VI-SAFE: v_cndmask_b32_e32 -; VI-SAFE: v_cmp_gt_f32_e32 -; VI-SAFE: v_cndmask_b32_e32 -; VI-SAFE-NOT: v_cmp -; VI-SAFE-NOT: v_cndmask - -; GCN-NONAN: v_max_f32_e32 -; GCN-NONAN: v_max_f32_e32 -; GCN-NONAN: v_max_f32_e32 +; SI: v_max_legacy_f32_e32 +; SI: v_max_legacy_f32_e32 +; SI: v_max_legacy_f32_e32 + +; VI: v_cmp_gt_f32_e32 +; VI: v_cndmask_b32_e32 +; VI: v_cmp_gt_f32_e32 +; VI: v_cndmask_b32_e32 +; VI: v_cmp_gt_f32_e32 +; VI: v_cndmask_b32_e32 +; VI-NOT: v_cmp +; VI-NOT: v_cndmask ; GCN-NOT: v_max define amdgpu_kernel void @test_fmax_legacy_ogt_v3f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { @@ -199,6 +310,27 @@ define amdgpu_kernel void @test_fmax_legacy_ogt_v3f32(ptr addrspace(1) %out, ptr ret void } +; FUNC-LABEL: {{^}}test_fmax_legacy_ogt_v3f32_fast: + +; GCN: v_max_f32_e32 +; GCN: v_max_f32_e32 +; GCN: v_max_f32_e32 + +; GCN-NOT: v_max +define amdgpu_kernel void @test_fmax_legacy_ogt_v3f32_fast(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { + %tid = call i32 @llvm.amdgcn.workitem.id.x() #1 + %gep.0 = getelementptr <3 x float>, ptr addrspace(1) %in, i32 %tid + %gep.1 = getelementptr <3 x float>, ptr addrspace(1) %gep.0, i32 1 + + %a = load <3 x float>, ptr addrspace(1) %gep.0 + %b = load <3 x float>, ptr addrspace(1) %gep.1 + + %cmp = fcmp ogt <3 x float> %a, %b + %val = select nnan nsz <3 x i1> %cmp, <3 x float> %a, <3 x float> %b + store <3 x float> %val, ptr addrspace(1) %out + ret void +} + ; FUNC-LABEL: {{^}}test_fmax_legacy_ogt_f32_multi_use: ; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]] ; GCN: {{buffer|flat}}_load_dword [[B:v[0-9]+]] diff --git a/llvm/test/CodeGen/AMDGPU/fmin_fmax_legacy.amdgcn.ll b/llvm/test/CodeGen/AMDGPU/fmin_fmax_legacy.amdgcn.ll index 2ac5891..37f077d5 100644 --- a/llvm/test/CodeGen/AMDGPU/fmin_fmax_legacy.amdgcn.ll +++ b/llvm/test/CodeGen/AMDGPU/fmin_fmax_legacy.amdgcn.ll @@ -1,16 +1,12 @@ -; RUN: llc -mtriple=amdgcn < %s | FileCheck -enable-var-scope -check-prefixes=SI-SAFE,GCN %s -; RUN: llc -enable-no-nans-fp-math -enable-no-signed-zeros-fp-math -mtriple=amdgcn < %s | FileCheck -enable-var-scope --check-prefixes=GCN %s +; RUN: llc -mtriple=amdgcn < %s | FileCheck -enable-var-scope -check-prefixes=SI,GCN %s -; RUN: llc -mtriple=amdgcn -mcpu=fiji < %s | FileCheck -enable-var-scope -check-prefixes=VI-SAFE,GCN %s -; RUN: llc -enable-no-nans-fp-math -enable-no-signed-zeros-fp-math -mtriple=amdgcn -mcpu=fiji < %s | FileCheck -enable-var-scope --check-prefixes=GCN,VI-NNAN %s +; RUN: llc -mtriple=amdgcn -mcpu=fiji < %s | FileCheck -enable-var-scope -check-prefixes=VI,GCN %s ; GCN-LABEL: {{^}}min_fneg_select_regression_0: ; GCN-NOT: v_mul -; SI: v_max_legacy_f32_e64 [[MIN:v[0-9]+]], -1.0, -v0 - -; VI-SAFE: v_cmp_nle_f32_e32 vcc, 1.0, v0 -; VI-SAFE-NEXT: v_cndmask_b32_e64 v0, -1.0, -v0, vcc +; VI: v_cmp_nle_f32_e32 vcc, 1.0, v0 +; VI-NEXT: v_cndmask_b32_e64 v0, -1.0, -v0, vcc define amdgpu_ps float @min_fneg_select_regression_0(float %a, float %b) #0 { %fneg.a = fsub float -0.0, %a %cmp.a = fcmp ult float %a, 1.0 @@ -18,15 +14,23 @@ define amdgpu_ps float @min_fneg_select_regression_0(float %a, float %b) #0 { ret float %min.a } +; GCN-LABEL: {{^}}min_fneg_select_regression_0_fast: +; GCN-NOT: v_mul + +define amdgpu_ps float @min_fneg_select_regression_0_fast(float %a, float %b) #0 { + %fneg.a = fsub float -0.0, %a + %cmp.a = fcmp ult float %a, 1.0 + %min.a = select nnan nsz i1 %cmp.a, float %fneg.a, float -1.0 + ret float %min.a +} + ; GCN-LABEL: {{^}}min_fneg_select_regression_posk_0: ; GCN-NOT: v_mul ; SI: v_max_legacy_f32_e64 [[MIN:v[0-9]+]], 1.0, -v0 -; VI-SAFE: v_cmp_nle_f32_e32 vcc, -1.0, v0 -; VI-SAFE-NEXT: v_cndmask_b32_e64 v0, 1.0, -v0, vcc - -; VI-NNAN: v_max_f32_e64 v{{[0-9]+}}, -v0, 1.0 +; VI: v_cmp_nle_f32_e32 vcc, -1.0, v0 +; VI-NEXT: v_cndmask_b32_e64 v0, 1.0, -v0, vcc define amdgpu_ps float @min_fneg_select_regression_posk_0(float %a, float %b) #0 { %fneg.a = fsub float -0.0, %a %cmp.a = fcmp ult float %a, -1.0 @@ -34,15 +38,24 @@ define amdgpu_ps float @min_fneg_select_regression_posk_0(float %a, float %b) #0 ret float %min.a } -; GCN-LABEL: {{^}}max_fneg_select_regression_0: +; GCN-LABEL: {{^}}min_fneg_select_regression_posk_0_fast: ; GCN-NOT: v_mul -; SI-SAFE: v_min_legacy_f32_e64 [[MIN:v[0-9]+]], -1.0, -v0 +; VI: v_max_f32_e64 v{{[0-9]+}}, -v0, 1.0 +define amdgpu_ps float @min_fneg_select_regression_posk_0_fast(float %a, float %b) #0 { + %fneg.a = fsub float -0.0, %a + %cmp.a = fcmp ult float %a, -1.0 + %min.a = select nnan nsz i1 %cmp.a, float %fneg.a, float 1.0 + ret float %min.a +} + +; GCN-LABEL: {{^}}max_fneg_select_regression_0: +; GCN-NOT: v_mul -; VI-SAFE: v_cmp_nge_f32_e32 vcc, 1.0, v0 -; VI-SAFE-NEXT: v_cndmask_b32_e64 v0, -1.0, -v0, vcc +; SI: v_min_legacy_f32_e64 [[MIN:v[0-9]+]], -1.0, -v0 -; GCN-NONAN: v_min_f32_e64 [[MIN:v[0-9]+]], -v0, -1.0 +; VI: v_cmp_nge_f32_e32 vcc, 1.0, v0 +; VI-NEXT: v_cndmask_b32_e64 v0, -1.0, -v0, vcc define amdgpu_ps float @max_fneg_select_regression_0(float %a) #0 { %fneg.a = fsub float -0.0, %a %cmp.a = fcmp ugt float %a, 1.0 @@ -50,15 +63,24 @@ define amdgpu_ps float @max_fneg_select_regression_0(float %a) #0 { ret float %min.a } -; GCN-LABEL: {{^}}max_fneg_select_regression_posk_0: +; GCN-LABEL: {{^}}max_fneg_select_regression_0_fast: ; GCN-NOT: v_mul -; SI-SAFE: v_min_legacy_f32_e64 [[MIN:v[0-9]+]], 1.0, -v0 +; GCN: v_min_f32_e64 [[MIN:v[0-9]+]], -v0, -1.0 +define amdgpu_ps float @max_fneg_select_regression_0_fast(float %a) #0 { + %fneg.a = fsub float -0.0, %a + %cmp.a = fcmp ugt float %a, 1.0 + %min.a = select nnan nsz i1 %cmp.a, float %fneg.a, float -1.0 + ret float %min.a +} + +; GCN-LABEL: {{^}}max_fneg_select_regression_posk_0: +; GCN-NOT: v_mul -; VI-SAFE: v_cmp_nge_f32_e32 vcc, -1.0, v0 -; VI-SAFE-NEXT: v_cndmask_b32_e64 v0, 1.0, -v0, vcc +; SI: v_min_legacy_f32_e64 [[MIN:v[0-9]+]], 1.0, -v0 -; GCN-NONAN: v_min_f32_e64 [[MIN:v[0-9]+]], -v0, 1.0 +; VI: v_cmp_nge_f32_e32 vcc, -1.0, v0 +; VI-NEXT: v_cndmask_b32_e64 v0, 1.0, -v0, vcc define amdgpu_ps float @max_fneg_select_regression_posk_0(float %a) #0 { %fneg.a = fsub float -0.0, %a %cmp.a = fcmp ugt float %a, -1.0 @@ -66,13 +88,22 @@ define amdgpu_ps float @max_fneg_select_regression_posk_0(float %a) #0 { ret float %min.a } +; GCN-LABEL: {{^}}max_fneg_select_regression_posk_0_fast: +; GCN-NOT: v_mul + +; GCN: v_min_f32_e64 [[MIN:v[0-9]+]], -v0, 1.0 +define amdgpu_ps float @max_fneg_select_regression_posk_0_fast(float %a) #0 { + %fneg.a = fsub float -0.0, %a + %cmp.a = fcmp ugt float %a, -1.0 + %min.a = select nnan nsz i1 %cmp.a, float %fneg.a, float 1.0 + ret float %min.a +} + ; GCN-LABEL: {{^}}select_fneg_a_or_q_cmp_ugt_a_neg1: ; SI: v_min_legacy_f32_e64 v0, 1.0, -v0 -; VI-SAFE: v_cmp_nge_f32_e32 vcc, -1.0, v0 -; VI-SAFE-NEXT: v_cndmask_b32_e64 v0, 1.0, -v0, vcc - -; VI-NNAN: v_min_f32_e64 v0, -v0, 1.0 +; VI: v_cmp_nge_f32_e32 vcc, -1.0, v0 +; VI-NEXT: v_cndmask_b32_e64 v0, 1.0, -v0, vcc define amdgpu_ps float @select_fneg_a_or_q_cmp_ugt_a_neg1(float %a, float %b) #0 { %fneg.a = fneg float %a %cmp.a = fcmp ugt float %a, -1.0 @@ -80,13 +111,21 @@ define amdgpu_ps float @select_fneg_a_or_q_cmp_ugt_a_neg1(float %a, float %b) #0 ret float %min.a } +; GCN-LABEL: {{^}}select_fneg_a_or_q_cmp_ugt_a_neg1_fast: + +; VI: v_min_f32_e64 v0, -v0, 1.0 +define amdgpu_ps float @select_fneg_a_or_q_cmp_ugt_a_neg1_fast(float %a, float %b) #0 { + %fneg.a = fneg float %a + %cmp.a = fcmp ugt float %a, -1.0 + %min.a = select nnan nsz i1 %cmp.a, float %fneg.a, float 1.0 + ret float %min.a +} + ; GCN-LABEL: {{^}}select_fneg_a_or_q_cmp_ult_a_neg1: ; SI: v_max_legacy_f32_e64 v0, 1.0, -v0 -; VI-SAFE: v_cmp_nle_f32_e32 vcc, -1.0, v0 -; VI-SAFE-NEXT: v_cndmask_b32_e64 v0, 1.0, -v0, vcc - -; VI-NNAN: v_max_f32_e64 v0, -v0, 1.0 +; VI: v_cmp_nle_f32_e32 vcc, -1.0, v0 +; VI-NEXT: v_cndmask_b32_e64 v0, 1.0, -v0, vcc define amdgpu_ps float @select_fneg_a_or_q_cmp_ult_a_neg1(float %a, float %b) #0 { %fneg.a = fneg float %a %cmp.a = fcmp ult float %a, -1.0 @@ -94,13 +133,21 @@ define amdgpu_ps float @select_fneg_a_or_q_cmp_ult_a_neg1(float %a, float %b) #0 ret float %min.a } +; GCN-LABEL: {{^}}select_fneg_a_or_q_cmp_ult_a_neg1_fast: + +; VI: v_max_f32_e64 v0, -v0, 1.0 +define amdgpu_ps float @select_fneg_a_or_q_cmp_ult_a_neg1_fast(float %a, float %b) #0 { + %fneg.a = fneg float %a + %cmp.a = fcmp ult float %a, -1.0 + %min.a = select nnan nsz i1 %cmp.a, float %fneg.a, float 1.0 + ret float %min.a +} + ; GCN-LABEL: {{^}}select_fneg_a_or_q_cmp_ogt_a_neg1: ; SI: v_min_legacy_f32_e64 v0, -v0, 1.0 -; VI-SAFE: v_cmp_lt_f32_e32 vcc, -1.0, v0 -; VI-SAFE-NEXT: v_cndmask_b32_e64 v0, 1.0, -v0, vcc - -; VI-NNAN: v_min_f32_e64 v0, -v0, 1.0 +; VI: v_cmp_lt_f32_e32 vcc, -1.0, v0 +; VI-NEXT: v_cndmask_b32_e64 v0, 1.0, -v0, vcc define amdgpu_ps float @select_fneg_a_or_q_cmp_ogt_a_neg1(float %a, float %b) #0 { %fneg.a = fneg float %a %cmp.a = fcmp ogt float %a, -1.0 @@ -108,13 +155,21 @@ define amdgpu_ps float @select_fneg_a_or_q_cmp_ogt_a_neg1(float %a, float %b) #0 ret float %min.a } +; GCN-LABEL: {{^}}select_fneg_a_or_q_cmp_ogt_a_neg1_fast: + +; VI: v_min_f32_e64 v0, -v0, 1.0 +define amdgpu_ps float @select_fneg_a_or_q_cmp_ogt_a_neg1_fast(float %a, float %b) #0 { + %fneg.a = fneg float %a + %cmp.a = fcmp ogt float %a, -1.0 + %min.a = select nnan nsz i1 %cmp.a, float %fneg.a, float 1.0 + ret float %min.a +} + ; GCN-LABEL: {{^}}select_fneg_a_or_q_cmp_olt_a_neg1: ; SI: v_max_legacy_f32_e64 v0, -v0, 1.0 -; VI-SAFE: v_cmp_gt_f32_e32 vcc, -1.0, v0 -; VI-SAFE-NEXT: v_cndmask_b32_e64 v0, 1.0, -v0, vcc - -; VI-NANN: v_max_f32_e64 v0, -v0, 1.0 +; VI: v_cmp_gt_f32_e32 vcc, -1.0, v0 +; VI-NEXT: v_cndmask_b32_e64 v0, 1.0, -v0, vcc define amdgpu_ps float @select_fneg_a_or_q_cmp_olt_a_neg1(float %a, float %b) #0 { %fneg.a = fneg float %a %cmp.a = fcmp olt float %a, -1.0 @@ -122,17 +177,24 @@ define amdgpu_ps float @select_fneg_a_or_q_cmp_olt_a_neg1(float %a, float %b) #0 ret float %min.a } +; GCN-LABEL: {{^}}select_fneg_a_or_q_cmp_olt_a_neg1_fast: + +; VI-NANN: v_max_f32_e64 v0, -v0, 1.0 +define amdgpu_ps float @select_fneg_a_or_q_cmp_olt_a_neg1_fast(float %a, float %b) #0 { + %fneg.a = fneg float %a + %cmp.a = fcmp olt float %a, -1.0 + %min.a = select nnan nsz i1 %cmp.a, float %fneg.a, float 1.0 + ret float %min.a +} + ; GCN-LABEL: {{^}}select_fneg_a_or_q_cmp_ugt_a_neg8: ; SI: s_mov_b32 [[K:s[0-9]+]], 0x41000000 ; SI-NEXT: v_min_legacy_f32_e64 v0, [[K]], -v0 -; VI-SAFE-DAG: s_mov_b32 [[K0:s[0-9]+]], 0xc1000000 -; VI-SAFE-DAG: v_mov_b32_e32 [[K1:v[0-9]+]], 0x41000000 -; VI-SAFE: v_cmp_nge_f32_e32 vcc, [[K0]], v0 -; VI-SAFE-NEXT: v_cndmask_b32_e64 v0, [[K1]], -v0, vcc - -; VI-NNAN: s_mov_b32 [[K:s[0-9]+]], 0x41000000 -; VI-NNAN-NEXT: v_min_f32_e64 v0, -v0, [[K]] +; VI-DAG: s_mov_b32 [[K0:s[0-9]+]], 0xc1000000 +; VI-DAG: v_mov_b32_e32 [[K1:v[0-9]+]], 0x41000000 +; VI: v_cmp_nge_f32_e32 vcc, [[K0]], v0 +; VI-NEXT: v_cndmask_b32_e64 v0, [[K1]], -v0, vcc define amdgpu_ps float @select_fneg_a_or_q_cmp_ugt_a_neg8(float %a, float %b) #0 { %fneg.a = fneg float %a %cmp.a = fcmp ugt float %a, -8.0 @@ -140,17 +202,25 @@ define amdgpu_ps float @select_fneg_a_or_q_cmp_ugt_a_neg8(float %a, float %b) #0 ret float %min.a } +; GCN-LABEL: {{^}}select_fneg_a_or_q_cmp_ugt_a_neg8_fast: + +; VI: s_mov_b32 [[K:s[0-9]+]], 0x41000000 +; VI-NEXT: v_min_f32_e64 v0, -v0, [[K]] +define amdgpu_ps float @select_fneg_a_or_q_cmp_ugt_a_neg8_fast(float %a, float %b) #0 { + %fneg.a = fneg float %a + %cmp.a = fcmp ugt float %a, -8.0 + %min.a = select nnan nsz i1 %cmp.a, float %fneg.a, float 8.0 + ret float %min.a +} + ; GCN-LABEL: {{^}}select_fneg_a_or_q_cmp_ult_a_neg8: ; SI: s_mov_b32 [[K:s[0-9]+]], 0x41000000 ; SI-NEXT: v_max_legacy_f32_e64 v0, [[K]], -v0 -; VI-SAFE-DAG: s_mov_b32 [[K0:s[0-9]+]], 0xc1000000 -; VI-SAFE-DAG: v_mov_b32_e32 [[K1:v[0-9]+]], 0x41000000 -; VI-SAFE: v_cmp_nle_f32_e32 vcc, [[K0]], v0 -; VI-SAFE-NEXT: v_cndmask_b32_e64 v0, [[K1]], -v0, vcc - -; VI-NNAN: s_mov_b32 [[K:s[0-9]+]], 0x41000000 -; VI-NNAN-NEXT: v_max_f32_e64 v0, -v0, [[K]] +; VI-DAG: s_mov_b32 [[K0:s[0-9]+]], 0xc1000000 +; VI-DAG: v_mov_b32_e32 [[K1:v[0-9]+]], 0x41000000 +; VI: v_cmp_nle_f32_e32 vcc, [[K0]], v0 +; VI-NEXT: v_cndmask_b32_e64 v0, [[K1]], -v0, vcc define amdgpu_ps float @select_fneg_a_or_q_cmp_ult_a_neg8(float %a, float %b) #0 { %fneg.a = fneg float %a %cmp.a = fcmp ult float %a, -8.0 @@ -158,17 +228,25 @@ define amdgpu_ps float @select_fneg_a_or_q_cmp_ult_a_neg8(float %a, float %b) #0 ret float %min.a } +; GCN-LABEL: {{^}}select_fneg_a_or_q_cmp_ult_a_neg8_fast: + +; VI: s_mov_b32 [[K:s[0-9]+]], 0x41000000 +; VI-NEXT: v_max_f32_e64 v0, -v0, [[K]] +define amdgpu_ps float @select_fneg_a_or_q_cmp_ult_a_neg8_fast(float %a, float %b) #0 { + %fneg.a = fneg float %a + %cmp.a = fcmp ult float %a, -8.0 + %min.a = select nnan nsz i1 %cmp.a, float %fneg.a, float 8.0 + ret float %min.a +} + ; GCN-LABEL: {{^}}select_fneg_a_or_q_cmp_ogt_a_neg8: ; SI: s_mov_b32 [[K:s[0-9]+]], 0x41000000 ; SI-NEXT: v_min_legacy_f32_e64 v0, -v0, [[K]] -; VI-SAFE-DAG: s_mov_b32 [[K0:s[0-9]+]], 0xc1000000 -; VI-SAFE-DAG: v_mov_b32_e32 [[K1:v[0-9]+]], 0x41000000 -; VI-SAFE: v_cmp_lt_f32_e32 vcc, [[K0]], v0 -; VI-SAFE-NEXT: v_cndmask_b32_e64 v0, [[K1]], -v0, vcc - -; VI-NNAN: s_mov_b32 [[K:s[0-9]+]], 0x41000000 -; VI-NNAN-NEXT: v_min_f32_e64 v0, -v0, [[K]] +; VI-DAG: s_mov_b32 [[K0:s[0-9]+]], 0xc1000000 +; VI-DAG: v_mov_b32_e32 [[K1:v[0-9]+]], 0x41000000 +; VI: v_cmp_lt_f32_e32 vcc, [[K0]], v0 +; VI-NEXT: v_cndmask_b32_e64 v0, [[K1]], -v0, vcc define amdgpu_ps float @select_fneg_a_or_q_cmp_ogt_a_neg8(float %a, float %b) #0 { %fneg.a = fneg float %a %cmp.a = fcmp ogt float %a, -8.0 @@ -176,18 +254,26 @@ define amdgpu_ps float @select_fneg_a_or_q_cmp_ogt_a_neg8(float %a, float %b) #0 ret float %min.a } +; GCN-LABEL: {{^}}select_fneg_a_or_q_cmp_ogt_a_neg8_fast: + +; VI: s_mov_b32 [[K:s[0-9]+]], 0x41000000 +; VI-NEXT: v_min_f32_e64 v0, -v0, [[K]] +define amdgpu_ps float @select_fneg_a_or_q_cmp_ogt_a_neg8_fast(float %a, float %b) #0 { + %fneg.a = fneg float %a + %cmp.a = fcmp ogt float %a, -8.0 + %min.a = select nnan nsz i1 %cmp.a, float %fneg.a, float 8.0 + ret float %min.a +} + ; GCN-LABEL: {{^}}select_fneg_a_or_q_cmp_olt_a_neg8: ; SI: s_mov_b32 [[K:s[0-9]+]], 0x41000000 ; SI-NEXT: v_max_legacy_f32_e64 v0, -v0, [[K]] -; VI-SAFE-DAG: s_mov_b32 [[K0:s[0-9]+]], 0xc1000000 -; VI-SAFE-DAG: v_mov_b32_e32 [[K1:v[0-9]+]], 0x41000000 -; VI-SAFE: v_cmp_gt_f32_e32 vcc, [[K0]], v0 -; VI-SAFE-NEXT: v_cndmask_b32_e64 v0, [[K1]], -v0, vcc - -; VI-NNAN: s_mov_b32 [[K:s[0-9]+]], 0x41000000 -; VI-NNAN-NEXT: v_max_f32_e64 v0, -v0, [[K]] +; VI-DAG: s_mov_b32 [[K0:s[0-9]+]], 0xc1000000 +; VI-DAG: v_mov_b32_e32 [[K1:v[0-9]+]], 0x41000000 +; VI: v_cmp_gt_f32_e32 vcc, [[K0]], v0 +; VI-NEXT: v_cndmask_b32_e64 v0, [[K1]], -v0, vcc define amdgpu_ps float @select_fneg_a_or_q_cmp_olt_a_neg8(float %a, float %b) #0 { %fneg.a = fneg float %a %cmp.a = fcmp olt float %a, -8.0 @@ -195,13 +281,22 @@ define amdgpu_ps float @select_fneg_a_or_q_cmp_olt_a_neg8(float %a, float %b) #0 ret float %min.a } +; GCN-LABEL: {{^}}select_fneg_a_or_q_cmp_olt_a_neg8_fast: + +; VI: s_mov_b32 [[K:s[0-9]+]], 0x41000000 +; VI-NEXT: v_max_f32_e64 v0, -v0, [[K]] +define amdgpu_ps float @select_fneg_a_or_q_cmp_olt_a_neg8_fast(float %a, float %b) #0 { + %fneg.a = fneg float %a + %cmp.a = fcmp olt float %a, -8.0 + %min.a = select nnan nsz i1 %cmp.a, float %fneg.a, float 8.0 + ret float %min.a +} + ; GCN-LABEL: {{^}}select_fneg_a_or_neg1_cmp_olt_a_1: ; SI: v_max_legacy_f32_e64 v0, -v0, -1.0 -; VI-SAFE: v_cmp_gt_f32_e32 vcc, 1.0, v0 -; VI-SAFE-NEXT: v_cndmask_b32_e64 v0, -1.0, -v0, vcc - -; VI-NNAN: v_max_f32_e64 v0, -v0, -1.0 +; VI: v_cmp_gt_f32_e32 vcc, 1.0, v0 +; VI-NEXT: v_cndmask_b32_e64 v0, -1.0, -v0, vcc define amdgpu_ps float @select_fneg_a_or_neg1_cmp_olt_a_1(float %a, float %b) #0 { %fneg.a = fneg float %a %cmp.a = fcmp olt float %a, 1.0 @@ -209,15 +304,22 @@ define amdgpu_ps float @select_fneg_a_or_neg1_cmp_olt_a_1(float %a, float %b) #0 ret float %min.a } +; GCN-LABEL: {{^}}select_fneg_a_or_neg1_cmp_olt_a_1_fast: + +; VI: v_max_f32_e64 v0, -v0, -1.0 +define amdgpu_ps float @select_fneg_a_or_neg1_cmp_olt_a_1_fast(float %a, float %b) #0 { + %fneg.a = fneg float %a + %cmp.a = fcmp olt float %a, 1.0 + %min.a = select nnan nsz i1 %cmp.a, float %fneg.a, float -1.0 + ret float %min.a +} + ; GCN-LABEL: {{^}}ult_a_select_fneg_a_b: ; SI: v_cmp_nge_f32_e32 vcc, v0, v1 ; SI-NEXT: v_cndmask_b32_e64 v0, v1, -v0, vcc -; VI-SAFE: v_cmp_nge_f32_e32 vcc, v0, v1 -; VI-SAFE-NEXT: v_cndmask_b32_e64 v0, v1, -v0, vcc - -; VI-NNAN: v_cmp_lt_f32_e32 vcc, v0, v1 -; VI-NNAN-NEXT: v_cndmask_b32_e64 v0, v1, -v0, vcc +; VI: v_cmp_nge_f32_e32 vcc, v0, v1 +; VI-NEXT: v_cndmask_b32_e64 v0, v1, -v0, vcc define amdgpu_ps float @ult_a_select_fneg_a_b(float %a, float %b) #0 { %fneg.a = fneg float %a %cmp.a = fcmp ult float %a, %b @@ -225,15 +327,23 @@ define amdgpu_ps float @ult_a_select_fneg_a_b(float %a, float %b) #0 { ret float %min.a } +; GCN-LABEL: {{^}}ult_a_select_fneg_a_b_fast: + +; VI: v_cmp_lt_f32_e32 vcc, v0, v1 +; VI-NEXT: v_cndmask_b32_e64 v0, v1, -v0, vcc +define amdgpu_ps float @ult_a_select_fneg_a_b_fast(float %a, float %b) #0 { + %fneg.a = fneg float %a + %cmp.a = fcmp nnan nsz ult float %a, %b + %min.a = select nnan nsz i1 %cmp.a, float %fneg.a, float %b + ret float %min.a +} + ; GCN-LABEL: {{^}}ugt_a_select_fneg_a_b: ; SI: v_cmp_nle_f32_e32 vcc, v0, v1 ; SI-NEXT: v_cndmask_b32_e64 v0, v1, -v0, vcc -; VI-SAFE: v_cmp_nle_f32_e32 vcc, v0, v1 -; VI-SAFE-NEXT: v_cndmask_b32_e64 v0, v1, -v0, vcc - -; VI-NNAN: v_cmp_gt_f32_e32 vcc, v0, v1 -; VI-NNAN-NEXT: v_cndmask_b32_e64 v0, v1, -v0, vcc +; VI: v_cmp_nle_f32_e32 vcc, v0, v1 +; VI-NEXT: v_cndmask_b32_e64 v0, v1, -v0, vcc define amdgpu_ps float @ugt_a_select_fneg_a_b(float %a, float %b) #0 { %fneg.a = fneg float %a %cmp.a = fcmp ugt float %a, %b @@ -241,5 +351,16 @@ define amdgpu_ps float @ugt_a_select_fneg_a_b(float %a, float %b) #0 { ret float %min.a } +; GCN-LABEL: {{^}}ugt_a_select_fneg_a_b_fast: + +; VI: v_cmp_gt_f32_e32 vcc, v0, v1 +; VI-NEXT: v_cndmask_b32_e64 v0, v1, -v0, vcc +define amdgpu_ps float @ugt_a_select_fneg_a_b_fast(float %a, float %b) #0 { + %fneg.a = fneg float %a + %cmp.a = fcmp nnan nsz ugt float %a, %b + %min.a = select nnan nsz i1 %cmp.a, float %fneg.a, float %b + ret float %min.a +} + attributes #0 = { nounwind } attributes #1 = { nounwind readnone } diff --git a/llvm/test/CodeGen/AMDGPU/fmin_legacy.f16.ll b/llvm/test/CodeGen/AMDGPU/fmin_legacy.f16.ll index 34cb0b1..40c2ec0 100644 --- a/llvm/test/CodeGen/AMDGPU/fmin_legacy.f16.ll +++ b/llvm/test/CodeGen/AMDGPU/fmin_legacy.f16.ll @@ -1,735 +1,760 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py -; RUN: llc -mtriple=amdgcn-- -mcpu=gfx900 < %s | FileCheck -enable-var-scope -check-prefixes=GFX9-SAFE %s -; RUN: llc -enable-no-nans-fp-math -enable-no-signed-zeros-fp-math -mtriple=amdgcn-- -mcpu=gfx900 < %s | FileCheck -enable-var-scope -check-prefixes=GFX9-NNAN %s +; RUN: llc -mtriple=amdgcn-- -mcpu=gfx900 < %s | FileCheck -enable-var-scope -check-prefixes=GFX9 %s -; RUN: llc -mtriple=amdgcn-- -mcpu=fiji < %s | FileCheck -enable-var-scope -check-prefix=VI-SAFE %s -; RUN: llc -enable-no-nans-fp-math -enable-no-signed-zeros-fp-math -mtriple=amdgcn-- -mcpu=fiji < %s | FileCheck -enable-var-scope -check-prefixes=VI-NNAN %s +; RUN: llc -mtriple=amdgcn-- -mcpu=fiji < %s | FileCheck -enable-var-scope -check-prefix=VI %s -; RUN: llc -mtriple=amdgcn-- < %s | FileCheck -enable-var-scope -check-prefix=SI-SAFE %s -; RUN: llc -enable-no-nans-fp-math -enable-no-signed-zeros-fp-math -mtriple=amdgcn-- < %s | FileCheck -enable-var-scope -check-prefixes=SI-NNAN %s +; RUN: llc -mtriple=amdgcn-- < %s | FileCheck -enable-var-scope -check-prefix=SI %s -; RUN: llc -mtriple=amdgcn-- -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX11-SAFE-TRUE16 %s -; RUN: llc -mtriple=amdgcn-- -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX11-SAFE-FAKE16 %s -; RUN: llc -enable-no-nans-fp-math -enable-no-signed-zeros-fp-math -mtriple=amdgcn-- -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX11-NNAN,GFX11-NNAN-TRUE16 %s -; RUN: llc -enable-no-nans-fp-math -enable-no-signed-zeros-fp-math -mtriple=amdgcn-- -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX11-NNAN,GFX11-NNAN-FAKE16 %s +; RUN: llc -mtriple=amdgcn-- -mcpu=gfx1100 -mattr=+real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX11,GFX11-TRUE16 %s +; RUN: llc -mtriple=amdgcn-- -mcpu=gfx1100 -mattr=-real-true16 < %s | FileCheck -enable-var-scope -check-prefixes=GFX11,GFX11-FAKE16 %s define half @test_fmin_legacy_ule_f16(half %a, half %b) #0 { -; GFX9-SAFE-LABEL: test_fmin_legacy_ule_f16: -; GFX9-SAFE: ; %bb.0: -; GFX9-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v0, v1 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc -; GFX9-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; GFX9-NNAN-LABEL: test_fmin_legacy_ule_f16: -; GFX9-NNAN: ; %bb.0: -; GFX9-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-NNAN-NEXT: v_min_f16_e32 v0, v0, v1 -; GFX9-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; VI-SAFE-LABEL: test_fmin_legacy_ule_f16: -; VI-SAFE: ; %bb.0: -; VI-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v0, v1 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc -; VI-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; VI-NNAN-LABEL: test_fmin_legacy_ule_f16: -; VI-NNAN: ; %bb.0: -; VI-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-NNAN-NEXT: v_min_f16_e32 v0, v0, v1 -; VI-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; SI-SAFE-LABEL: test_fmin_legacy_ule_f16: -; SI-SAFE: ; %bb.0: -; SI-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v0, v0 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v1, v1 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v1, v1 -; SI-SAFE-NEXT: v_min_legacy_f32_e32 v0, v1, v0 -; SI-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; SI-NNAN-LABEL: test_fmin_legacy_ule_f16: -; SI-NNAN: ; %bb.0: -; SI-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v1, v1 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v0, v0 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v1, v1 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-NNAN-NEXT: v_min_f32_e32 v0, v0, v1 -; SI-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-TRUE16-LABEL: test_fmin_legacy_ule_f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v0.l, v1.l -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v1.l, v0.l, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: test_fmin_legacy_ule_f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v0, v1 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NNAN-TRUE16-LABEL: test_fmin_legacy_ule_f16: -; GFX11-NNAN-TRUE16: ; %bb.0: -; GFX11-NNAN-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NNAN-TRUE16-NEXT: v_min_f16_e32 v0.l, v0.l, v1.l -; GFX11-NNAN-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NNAN-FAKE16-LABEL: test_fmin_legacy_ule_f16: -; GFX11-NNAN-FAKE16: ; %bb.0: -; GFX11-NNAN-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NNAN-FAKE16-NEXT: v_min_f16_e32 v0, v0, v1 -; GFX11-NNAN-FAKE16-NEXT: s_setpc_b64 s[30:31] +; GFX9-LABEL: test_fmin_legacy_ule_f16: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_cmp_ngt_f16_e32 vcc, v0, v1 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; VI-LABEL: test_fmin_legacy_ule_f16: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_cmp_ngt_f16_e32 vcc, v0, v1 +; VI-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc +; VI-NEXT: s_setpc_b64 s[30:31] +; +; SI-LABEL: test_fmin_legacy_ule_f16: +; SI: ; %bb.0: +; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; SI-NEXT: v_cvt_f16_f32_e32 v0, v0 +; SI-NEXT: v_cvt_f16_f32_e32 v1, v1 +; SI-NEXT: v_cvt_f32_f16_e32 v0, v0 +; SI-NEXT: v_cvt_f32_f16_e32 v1, v1 +; SI-NEXT: v_min_legacy_f32_e32 v0, v1, v0 +; SI-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-TRUE16-LABEL: test_fmin_legacy_ule_f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v0.l, v1.l +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v1.l, v0.l, vcc_lo +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: test_fmin_legacy_ule_f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v0, v1 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc_lo +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = fcmp ule half %a, %b %val = select i1 %cmp, half %a, half %b ret half %val } +define half @test_fmin_legacy_ule_f16_fast(half %a, half %b) #0 { +; GFX9-LABEL: test_fmin_legacy_ule_f16_fast: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_min_f16_e32 v0, v0, v1 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; VI-LABEL: test_fmin_legacy_ule_f16_fast: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_min_f16_e32 v0, v0, v1 +; VI-NEXT: s_setpc_b64 s[30:31] +; +; SI-LABEL: test_fmin_legacy_ule_f16_fast: +; SI: ; %bb.0: +; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; SI-NEXT: v_cvt_f16_f32_e32 v1, v1 +; SI-NEXT: v_cvt_f16_f32_e32 v0, v0 +; SI-NEXT: v_cvt_f32_f16_e32 v1, v1 +; SI-NEXT: v_cvt_f32_f16_e32 v0, v0 +; SI-NEXT: v_min_f32_e32 v0, v0, v1 +; SI-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-TRUE16-LABEL: test_fmin_legacy_ule_f16_fast: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_min_f16_e32 v0.l, v0.l, v1.l +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: test_fmin_legacy_ule_f16_fast: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_min_f16_e32 v0, v0, v1 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] + %cmp = fcmp ule half %a, %b + %val = select nnan nsz i1 %cmp, half %a, half %b + ret half %val +} + define <2 x half> @test_fmin_legacy_ule_v2f16(<2 x half> %a, <2 x half> %b) #0 { -; GFX9-SAFE-LABEL: test_fmin_legacy_ule_v2f16: -; GFX9-SAFE: ; %bb.0: -; GFX9-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v2, 16, v1 -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v3, 16, v0 -; GFX9-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v3, v2 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v2, v2, v3, vcc -; GFX9-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v0, v1 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc -; GFX9-SAFE-NEXT: s_mov_b32 s4, 0x5040100 -; GFX9-SAFE-NEXT: v_perm_b32 v0, v2, v0, s4 -; GFX9-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; GFX9-NNAN-LABEL: test_fmin_legacy_ule_v2f16: -; GFX9-NNAN: ; %bb.0: -; GFX9-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-NNAN-NEXT: v_pk_min_f16 v0, v0, v1 -; GFX9-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; VI-SAFE-LABEL: test_fmin_legacy_ule_v2f16: -; VI-SAFE: ; %bb.0: -; VI-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v2, 16, v1 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v3, 16, v0 -; VI-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v3, v2 -; VI-SAFE-NEXT: v_cndmask_b32_sdwa v2, v2, v3, vcc dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; VI-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v0, v1 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc -; VI-SAFE-NEXT: v_or_b32_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD -; VI-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; VI-NNAN-LABEL: test_fmin_legacy_ule_v2f16: -; VI-NNAN: ; %bb.0: -; VI-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-NNAN-NEXT: v_min_f16_sdwa v2, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; VI-NNAN-NEXT: v_min_f16_e32 v0, v0, v1 -; VI-NNAN-NEXT: v_or_b32_e32 v0, v0, v2 -; VI-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; SI-SAFE-LABEL: test_fmin_legacy_ule_v2f16: -; SI-SAFE: ; %bb.0: -; SI-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v1, v1 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v3, v3 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v0, v0 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v2, v2 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v1, v1 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v3, v3 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v2, v2 -; SI-SAFE-NEXT: v_min_legacy_f32_e32 v0, v2, v0 -; SI-SAFE-NEXT: v_min_legacy_f32_e32 v1, v3, v1 -; SI-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; SI-NNAN-LABEL: test_fmin_legacy_ule_v2f16: -; SI-NNAN: ; %bb.0: -; SI-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v3, v3 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v1, v1 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v2, v2 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v0, v0 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v3, v3 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v1, v1 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v2, v2 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-NNAN-NEXT: v_min_f32_e32 v0, v0, v2 -; SI-NNAN-NEXT: v_min_f32_e32 v1, v1, v3 -; SI-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-TRUE16-LABEL: test_fmin_legacy_ule_v2f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v0.h, v1.h -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ngt_f16_e64 s0, v0.l, v1.l -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v1.h, v0.h, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v1.l, v0.l, s0 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: test_fmin_legacy_ule_v2f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v2, 16, v1 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_1) -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v3, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v2, v2, v3, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v0, v1 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v2, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NNAN-LABEL: test_fmin_legacy_ule_v2f16: -; GFX11-NNAN: ; %bb.0: -; GFX11-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NNAN-NEXT: v_pk_min_f16 v0, v0, v1 -; GFX11-NNAN-NEXT: s_setpc_b64 s[30:31] +; GFX9-LABEL: test_fmin_legacy_ule_v2f16: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_lshrrev_b32_e32 v2, 16, v1 +; GFX9-NEXT: v_lshrrev_b32_e32 v3, 16, v0 +; GFX9-NEXT: v_cmp_ngt_f16_e32 vcc, v3, v2 +; GFX9-NEXT: v_cndmask_b32_e32 v2, v2, v3, vcc +; GFX9-NEXT: v_cmp_ngt_f16_e32 vcc, v0, v1 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc +; GFX9-NEXT: s_mov_b32 s4, 0x5040100 +; GFX9-NEXT: v_perm_b32 v0, v2, v0, s4 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; VI-LABEL: test_fmin_legacy_ule_v2f16: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_lshrrev_b32_e32 v2, 16, v1 +; VI-NEXT: v_lshrrev_b32_e32 v3, 16, v0 +; VI-NEXT: v_cmp_ngt_f16_e32 vcc, v3, v2 +; VI-NEXT: v_cndmask_b32_sdwa v2, v2, v3, vcc dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; VI-NEXT: v_cmp_ngt_f16_e32 vcc, v0, v1 +; VI-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc +; VI-NEXT: v_or_b32_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD +; VI-NEXT: s_setpc_b64 s[30:31] +; +; SI-LABEL: test_fmin_legacy_ule_v2f16: +; SI: ; %bb.0: +; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; SI-NEXT: v_cvt_f16_f32_e32 v1, v1 +; SI-NEXT: v_cvt_f16_f32_e32 v3, v3 +; SI-NEXT: v_cvt_f16_f32_e32 v0, v0 +; SI-NEXT: v_cvt_f16_f32_e32 v2, v2 +; SI-NEXT: v_cvt_f32_f16_e32 v1, v1 +; SI-NEXT: v_cvt_f32_f16_e32 v3, v3 +; SI-NEXT: v_cvt_f32_f16_e32 v0, v0 +; SI-NEXT: v_cvt_f32_f16_e32 v2, v2 +; SI-NEXT: v_min_legacy_f32_e32 v0, v2, v0 +; SI-NEXT: v_min_legacy_f32_e32 v1, v3, v1 +; SI-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-TRUE16-LABEL: test_fmin_legacy_ule_v2f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v0.h, v1.h +; GFX11-TRUE16-NEXT: v_cmp_ngt_f16_e64 s0, v0.l, v1.l +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v1.h, v0.h, vcc_lo +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v1.l, v0.l, s0 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: test_fmin_legacy_ule_v2f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v2, 16, v1 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v3, 16, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_1) +; GFX11-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v3, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v2, v2, v3, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v0, v1 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, v1, v0, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v2, v0, 0x5040100 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = fcmp ule <2 x half> %a, %b %val = select <2 x i1> %cmp, <2 x half> %a, <2 x half> %b ret <2 x half> %val } +define <2 x half> @test_fmin_legacy_ule_v2f16_fast(<2 x half> %a, <2 x half> %b) #0 { +; GFX9-LABEL: test_fmin_legacy_ule_v2f16_fast: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_pk_min_f16 v0, v0, v1 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; VI-LABEL: test_fmin_legacy_ule_v2f16_fast: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_min_f16_sdwa v2, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; VI-NEXT: v_min_f16_e32 v0, v0, v1 +; VI-NEXT: v_or_b32_e32 v0, v0, v2 +; VI-NEXT: s_setpc_b64 s[30:31] +; +; SI-LABEL: test_fmin_legacy_ule_v2f16_fast: +; SI: ; %bb.0: +; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; SI-NEXT: v_cvt_f16_f32_e32 v3, v3 +; SI-NEXT: v_cvt_f16_f32_e32 v1, v1 +; SI-NEXT: v_cvt_f16_f32_e32 v2, v2 +; SI-NEXT: v_cvt_f16_f32_e32 v0, v0 +; SI-NEXT: v_cvt_f32_f16_e32 v3, v3 +; SI-NEXT: v_cvt_f32_f16_e32 v1, v1 +; SI-NEXT: v_cvt_f32_f16_e32 v2, v2 +; SI-NEXT: v_cvt_f32_f16_e32 v0, v0 +; SI-NEXT: v_min_f32_e32 v0, v0, v2 +; SI-NEXT: v_min_f32_e32 v1, v1, v3 +; SI-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-LABEL: test_fmin_legacy_ule_v2f16_fast: +; GFX11: ; %bb.0: +; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-NEXT: v_pk_min_f16 v0, v0, v1 +; GFX11-NEXT: s_setpc_b64 s[30:31] + %cmp = fcmp ule <2 x half> %a, %b + %val = select nnan nsz <2 x i1> %cmp, <2 x half> %a, <2 x half> %b + ret <2 x half> %val +} + define <3 x half> @test_fmin_legacy_ule_v3f16(<3 x half> %a, <3 x half> %b) #0 { -; GFX9-SAFE-LABEL: test_fmin_legacy_ule_v3f16: -; GFX9-SAFE: ; %bb.0: -; GFX9-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v5, 16, v0 -; GFX9-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v5, v4 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v4, v4, v5, vcc -; GFX9-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v1, v3 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc -; GFX9-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v0, v2 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc -; GFX9-SAFE-NEXT: s_mov_b32 s4, 0x5040100 -; GFX9-SAFE-NEXT: v_perm_b32 v0, v4, v0, s4 -; GFX9-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; GFX9-NNAN-LABEL: test_fmin_legacy_ule_v3f16: -; GFX9-NNAN: ; %bb.0: -; GFX9-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-NNAN-NEXT: v_pk_min_f16 v1, v1, v3 -; GFX9-NNAN-NEXT: v_pk_min_f16 v0, v0, v2 -; GFX9-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; VI-SAFE-LABEL: test_fmin_legacy_ule_v3f16: -; VI-SAFE: ; %bb.0: -; VI-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v5, 16, v0 -; VI-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v5, v4 -; VI-SAFE-NEXT: v_cndmask_b32_sdwa v4, v4, v5, vcc dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD -; VI-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v1, v3 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc -; VI-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v0, v2 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc -; VI-SAFE-NEXT: v_or_b32_sdwa v0, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD -; VI-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; VI-NNAN-LABEL: test_fmin_legacy_ule_v3f16: -; VI-NNAN: ; %bb.0: -; VI-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-NNAN-NEXT: v_min_f16_sdwa v4, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; VI-NNAN-NEXT: v_min_f16_e32 v0, v0, v2 -; VI-NNAN-NEXT: v_min_f16_e32 v1, v1, v3 -; VI-NNAN-NEXT: v_or_b32_e32 v0, v0, v4 -; VI-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; SI-SAFE-LABEL: test_fmin_legacy_ule_v3f16: -; SI-SAFE: ; %bb.0: -; SI-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v2, v2 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v5, v5 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v1, v1 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v4, v4 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v0, v0 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v3, v3 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v2, v2 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v5, v5 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v1, v1 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v4, v4 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v3, v3 -; SI-SAFE-NEXT: v_min_legacy_f32_e32 v0, v3, v0 -; SI-SAFE-NEXT: v_min_legacy_f32_e32 v1, v4, v1 -; SI-SAFE-NEXT: v_min_legacy_f32_e32 v2, v5, v2 -; SI-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; SI-NNAN-LABEL: test_fmin_legacy_ule_v3f16: -; SI-NNAN: ; %bb.0: -; SI-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v5, v5 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v2, v2 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v4, v4 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v1, v1 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v3, v3 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v0, v0 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v5, v5 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v2, v2 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v4, v4 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v1, v1 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v3, v3 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-NNAN-NEXT: v_min_f32_e32 v0, v0, v3 -; SI-NNAN-NEXT: v_min_f32_e32 v1, v1, v4 -; SI-NNAN-NEXT: v_min_f32_e32 v2, v2, v5 -; SI-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-TRUE16-LABEL: test_fmin_legacy_ule_v3f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v0.h, v2.h -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ngt_f16_e64 s0, v0.l, v2.l -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ngt_f16_e64 s1, v1.l, v3.l -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v0.h, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, s0 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v1.l, v3.l, v1.l, s1 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: test_fmin_legacy_ule_v3f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v0 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v0, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_3) | instid1(VALU_DEP_3) -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v5, v4 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v2, v4, v5, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v1, v3 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v2, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NNAN-LABEL: test_fmin_legacy_ule_v3f16: -; GFX11-NNAN: ; %bb.0: -; GFX11-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NNAN-NEXT: v_pk_min_f16 v0, v0, v2 -; GFX11-NNAN-NEXT: v_pk_min_f16 v1, v1, v3 -; GFX11-NNAN-NEXT: s_setpc_b64 s[30:31] +; GFX9-LABEL: test_fmin_legacy_ule_v3f16: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; GFX9-NEXT: v_lshrrev_b32_e32 v5, 16, v0 +; GFX9-NEXT: v_cmp_ngt_f16_e32 vcc, v5, v4 +; GFX9-NEXT: v_cndmask_b32_e32 v4, v4, v5, vcc +; GFX9-NEXT: v_cmp_ngt_f16_e32 vcc, v1, v3 +; GFX9-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc +; GFX9-NEXT: v_cmp_ngt_f16_e32 vcc, v0, v2 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc +; GFX9-NEXT: s_mov_b32 s4, 0x5040100 +; GFX9-NEXT: v_perm_b32 v0, v4, v0, s4 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; VI-LABEL: test_fmin_legacy_ule_v3f16: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; VI-NEXT: v_lshrrev_b32_e32 v5, 16, v0 +; VI-NEXT: v_cmp_ngt_f16_e32 vcc, v5, v4 +; VI-NEXT: v_cndmask_b32_sdwa v4, v4, v5, vcc dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:DWORD +; VI-NEXT: v_cmp_ngt_f16_e32 vcc, v1, v3 +; VI-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc +; VI-NEXT: v_cmp_ngt_f16_e32 vcc, v0, v2 +; VI-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc +; VI-NEXT: v_or_b32_sdwa v0, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD +; VI-NEXT: s_setpc_b64 s[30:31] +; +; SI-LABEL: test_fmin_legacy_ule_v3f16: +; SI: ; %bb.0: +; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; SI-NEXT: v_cvt_f16_f32_e32 v2, v2 +; SI-NEXT: v_cvt_f16_f32_e32 v5, v5 +; SI-NEXT: v_cvt_f16_f32_e32 v1, v1 +; SI-NEXT: v_cvt_f16_f32_e32 v4, v4 +; SI-NEXT: v_cvt_f16_f32_e32 v0, v0 +; SI-NEXT: v_cvt_f16_f32_e32 v3, v3 +; SI-NEXT: v_cvt_f32_f16_e32 v2, v2 +; SI-NEXT: v_cvt_f32_f16_e32 v5, v5 +; SI-NEXT: v_cvt_f32_f16_e32 v1, v1 +; SI-NEXT: v_cvt_f32_f16_e32 v4, v4 +; SI-NEXT: v_cvt_f32_f16_e32 v0, v0 +; SI-NEXT: v_cvt_f32_f16_e32 v3, v3 +; SI-NEXT: v_min_legacy_f32_e32 v0, v3, v0 +; SI-NEXT: v_min_legacy_f32_e32 v1, v4, v1 +; SI-NEXT: v_min_legacy_f32_e32 v2, v5, v2 +; SI-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-TRUE16-LABEL: test_fmin_legacy_ule_v3f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v0.h, v2.h +; GFX11-TRUE16-NEXT: v_cmp_ngt_f16_e64 s0, v0.l, v2.l +; GFX11-TRUE16-NEXT: v_cmp_ngt_f16_e64 s1, v1.l, v3.l +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v0.h, vcc_lo +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, s0 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v1.l, v3.l, v1.l, s1 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: test_fmin_legacy_ule_v3f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v0 +; GFX11-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v0, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc_lo +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_3) | instid1(VALU_DEP_3) +; GFX11-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v5, v4 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v2, v4, v5, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v1, v3 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v2, v0, 0x5040100 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = fcmp ule <3 x half> %a, %b %val = select <3 x i1> %cmp, <3 x half> %a, <3 x half> %b ret <3 x half> %val } +define <3 x half> @test_fmin_legacy_ule_v3f16_fast(<3 x half> %a, <3 x half> %b) #0 { +; GFX9-LABEL: test_fmin_legacy_ule_v3f16_fast: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_pk_max_f16 v3, v3, v3 +; GFX9-NEXT: v_pk_max_f16 v1, v1, v1 +; GFX9-NEXT: v_pk_min_f16 v1, v1, v3 +; GFX9-NEXT: v_pk_min_f16 v0, v0, v2 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; VI-LABEL: test_fmin_legacy_ule_v3f16_fast: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_min_f16_sdwa v4, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; VI-NEXT: v_min_f16_e32 v0, v0, v2 +; VI-NEXT: v_min_f16_e32 v1, v1, v3 +; VI-NEXT: v_or_b32_e32 v0, v0, v4 +; VI-NEXT: s_setpc_b64 s[30:31] +; +; SI-LABEL: test_fmin_legacy_ule_v3f16_fast: +; SI: ; %bb.0: +; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; SI-NEXT: v_cvt_f16_f32_e32 v5, v5 +; SI-NEXT: v_cvt_f16_f32_e32 v2, v2 +; SI-NEXT: v_cvt_f16_f32_e32 v4, v4 +; SI-NEXT: v_cvt_f16_f32_e32 v1, v1 +; SI-NEXT: v_cvt_f16_f32_e32 v3, v3 +; SI-NEXT: v_cvt_f16_f32_e32 v0, v0 +; SI-NEXT: v_cvt_f32_f16_e32 v5, v5 +; SI-NEXT: v_cvt_f32_f16_e32 v2, v2 +; SI-NEXT: v_cvt_f32_f16_e32 v4, v4 +; SI-NEXT: v_cvt_f32_f16_e32 v1, v1 +; SI-NEXT: v_cvt_f32_f16_e32 v3, v3 +; SI-NEXT: v_cvt_f32_f16_e32 v0, v0 +; SI-NEXT: v_min_f32_e32 v0, v0, v3 +; SI-NEXT: v_min_f32_e32 v1, v1, v4 +; SI-NEXT: v_min_f32_e32 v2, v2, v5 +; SI-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-LABEL: test_fmin_legacy_ule_v3f16_fast: +; GFX11: ; %bb.0: +; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-NEXT: v_pk_max_f16 v3, v3, v3 +; GFX11-NEXT: v_pk_max_f16 v1, v1, v1 +; GFX11-NEXT: v_pk_min_f16 v0, v0, v2 +; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX11-NEXT: v_pk_min_f16 v1, v1, v3 +; GFX11-NEXT: s_setpc_b64 s[30:31] + %cmp = fcmp ule <3 x half> %a, %b + %val = select nnan nsz <3 x i1> %cmp, <3 x half> %a, <3 x half> %b + ret <3 x half> %val +} + define <4 x half> @test_fmin_legacy_ule_v4f16(<4 x half> %a, <4 x half> %b) #0 { -; GFX9-SAFE-LABEL: test_fmin_legacy_ule_v4f16: -; GFX9-SAFE: ; %bb.0: -; GFX9-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v6, 16, v3 -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v7, 16, v1 -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v5, 16, v0 -; GFX9-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v7, v6 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v6, v6, v7, vcc -; GFX9-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v5, v4 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v4, v4, v5, vcc -; GFX9-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v1, v3 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc -; GFX9-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v0, v2 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc -; GFX9-SAFE-NEXT: s_mov_b32 s4, 0x5040100 -; GFX9-SAFE-NEXT: v_perm_b32 v0, v4, v0, s4 -; GFX9-SAFE-NEXT: v_perm_b32 v1, v6, v1, s4 -; GFX9-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; GFX9-NNAN-LABEL: test_fmin_legacy_ule_v4f16: -; GFX9-NNAN: ; %bb.0: -; GFX9-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-NNAN-NEXT: v_pk_min_f16 v0, v0, v2 -; GFX9-NNAN-NEXT: v_pk_min_f16 v1, v1, v3 -; GFX9-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; VI-SAFE-LABEL: test_fmin_legacy_ule_v4f16: -; VI-SAFE: ; %bb.0: -; VI-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v6, 16, v3 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v7, 16, v1 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v4, 16, v2 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v5, 16, v0 -; VI-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v7, v6 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v6, v6, v7, vcc -; VI-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v5, v4 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v4, v4, v5, vcc -; VI-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v1, v3 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc -; VI-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v0, v2 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc -; VI-SAFE-NEXT: v_lshlrev_b32_e32 v2, 16, v4 -; VI-SAFE-NEXT: v_or_b32_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD -; VI-SAFE-NEXT: v_lshlrev_b32_e32 v2, 16, v6 -; VI-SAFE-NEXT: v_or_b32_sdwa v1, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD -; VI-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; VI-NNAN-LABEL: test_fmin_legacy_ule_v4f16: -; VI-NNAN: ; %bb.0: -; VI-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-NNAN-NEXT: v_min_f16_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; VI-NNAN-NEXT: v_min_f16_sdwa v5, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; VI-NNAN-NEXT: v_min_f16_e32 v1, v1, v3 -; VI-NNAN-NEXT: v_min_f16_e32 v0, v0, v2 -; VI-NNAN-NEXT: v_or_b32_e32 v0, v0, v5 -; VI-NNAN-NEXT: v_or_b32_e32 v1, v1, v4 -; VI-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; SI-SAFE-LABEL: test_fmin_legacy_ule_v4f16: -; SI-SAFE: ; %bb.0: -; SI-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v3, v3 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v7, v7 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v2, v2 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v6, v6 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v1, v1 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v5, v5 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v0, v0 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v4, v4 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v3, v3 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v7, v7 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v2, v2 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v6, v6 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v1, v1 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v5, v5 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v4, v4 -; SI-SAFE-NEXT: v_min_legacy_f32_e32 v0, v4, v0 -; SI-SAFE-NEXT: v_min_legacy_f32_e32 v1, v5, v1 -; SI-SAFE-NEXT: v_min_legacy_f32_e32 v2, v6, v2 -; SI-SAFE-NEXT: v_min_legacy_f32_e32 v3, v7, v3 -; SI-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; SI-NNAN-LABEL: test_fmin_legacy_ule_v4f16: -; SI-NNAN: ; %bb.0: -; SI-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v7, v7 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v3, v3 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v6, v6 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v2, v2 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v5, v5 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v1, v1 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v4, v4 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v0, v0 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v7, v7 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v3, v3 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v6, v6 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v2, v2 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v5, v5 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v1, v1 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v4, v4 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-NNAN-NEXT: v_min_f32_e32 v0, v0, v4 -; SI-NNAN-NEXT: v_min_f32_e32 v1, v1, v5 -; SI-NNAN-NEXT: v_min_f32_e32 v2, v2, v6 -; SI-NNAN-NEXT: v_min_f32_e32 v3, v3, v7 -; SI-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-TRUE16-LABEL: test_fmin_legacy_ule_v4f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v1.h, v3.h -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ngt_f16_e64 s0, v0.h, v2.h -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ngt_f16_e64 s1, v0.l, v2.l -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ngt_f16_e64 s2, v1.l, v3.l -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v1.h, v3.h, v1.h, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v0.h, s0 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, s1 -; GFX11-SAFE-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v1.l, v3.l, v1.l, s2 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: test_fmin_legacy_ule_v4f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v3 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v1 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v7, 16, v0 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_3) -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v5, v4 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v4, v4, v5, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v7, v6 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v5, v6, v7, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v0, v2 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v1, v3 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v5, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v1, v4, v1, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NNAN-LABEL: test_fmin_legacy_ule_v4f16: -; GFX11-NNAN: ; %bb.0: -; GFX11-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NNAN-NEXT: v_pk_min_f16 v0, v0, v2 -; GFX11-NNAN-NEXT: v_pk_min_f16 v1, v1, v3 -; GFX11-NNAN-NEXT: s_setpc_b64 s[30:31] +; GFX9-LABEL: test_fmin_legacy_ule_v4f16: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_lshrrev_b32_e32 v6, 16, v3 +; GFX9-NEXT: v_lshrrev_b32_e32 v7, 16, v1 +; GFX9-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; GFX9-NEXT: v_lshrrev_b32_e32 v5, 16, v0 +; GFX9-NEXT: v_cmp_ngt_f16_e32 vcc, v7, v6 +; GFX9-NEXT: v_cndmask_b32_e32 v6, v6, v7, vcc +; GFX9-NEXT: v_cmp_ngt_f16_e32 vcc, v5, v4 +; GFX9-NEXT: v_cndmask_b32_e32 v4, v4, v5, vcc +; GFX9-NEXT: v_cmp_ngt_f16_e32 vcc, v1, v3 +; GFX9-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc +; GFX9-NEXT: v_cmp_ngt_f16_e32 vcc, v0, v2 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc +; GFX9-NEXT: s_mov_b32 s4, 0x5040100 +; GFX9-NEXT: v_perm_b32 v0, v4, v0, s4 +; GFX9-NEXT: v_perm_b32 v1, v6, v1, s4 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; VI-LABEL: test_fmin_legacy_ule_v4f16: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_lshrrev_b32_e32 v6, 16, v3 +; VI-NEXT: v_lshrrev_b32_e32 v7, 16, v1 +; VI-NEXT: v_lshrrev_b32_e32 v4, 16, v2 +; VI-NEXT: v_lshrrev_b32_e32 v5, 16, v0 +; VI-NEXT: v_cmp_ngt_f16_e32 vcc, v7, v6 +; VI-NEXT: v_cndmask_b32_e32 v6, v6, v7, vcc +; VI-NEXT: v_cmp_ngt_f16_e32 vcc, v5, v4 +; VI-NEXT: v_cndmask_b32_e32 v4, v4, v5, vcc +; VI-NEXT: v_cmp_ngt_f16_e32 vcc, v1, v3 +; VI-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc +; VI-NEXT: v_cmp_ngt_f16_e32 vcc, v0, v2 +; VI-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc +; VI-NEXT: v_lshlrev_b32_e32 v2, 16, v4 +; VI-NEXT: v_or_b32_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD +; VI-NEXT: v_lshlrev_b32_e32 v2, 16, v6 +; VI-NEXT: v_or_b32_sdwa v1, v1, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD +; VI-NEXT: s_setpc_b64 s[30:31] +; +; SI-LABEL: test_fmin_legacy_ule_v4f16: +; SI: ; %bb.0: +; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; SI-NEXT: v_cvt_f16_f32_e32 v3, v3 +; SI-NEXT: v_cvt_f16_f32_e32 v7, v7 +; SI-NEXT: v_cvt_f16_f32_e32 v2, v2 +; SI-NEXT: v_cvt_f16_f32_e32 v6, v6 +; SI-NEXT: v_cvt_f16_f32_e32 v1, v1 +; SI-NEXT: v_cvt_f16_f32_e32 v5, v5 +; SI-NEXT: v_cvt_f16_f32_e32 v0, v0 +; SI-NEXT: v_cvt_f16_f32_e32 v4, v4 +; SI-NEXT: v_cvt_f32_f16_e32 v3, v3 +; SI-NEXT: v_cvt_f32_f16_e32 v7, v7 +; SI-NEXT: v_cvt_f32_f16_e32 v2, v2 +; SI-NEXT: v_cvt_f32_f16_e32 v6, v6 +; SI-NEXT: v_cvt_f32_f16_e32 v1, v1 +; SI-NEXT: v_cvt_f32_f16_e32 v5, v5 +; SI-NEXT: v_cvt_f32_f16_e32 v0, v0 +; SI-NEXT: v_cvt_f32_f16_e32 v4, v4 +; SI-NEXT: v_min_legacy_f32_e32 v0, v4, v0 +; SI-NEXT: v_min_legacy_f32_e32 v1, v5, v1 +; SI-NEXT: v_min_legacy_f32_e32 v2, v6, v2 +; SI-NEXT: v_min_legacy_f32_e32 v3, v7, v3 +; SI-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-TRUE16-LABEL: test_fmin_legacy_ule_v4f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v1.h, v3.h +; GFX11-TRUE16-NEXT: v_cmp_ngt_f16_e64 s0, v0.h, v2.h +; GFX11-TRUE16-NEXT: v_cmp_ngt_f16_e64 s1, v0.l, v2.l +; GFX11-TRUE16-NEXT: v_cmp_ngt_f16_e64 s2, v1.l, v3.l +; GFX11-TRUE16-NEXT: v_cndmask_b16 v1.h, v3.h, v1.h, vcc_lo +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v2.h, v0.h, s0 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v2.l, v0.l, s1 +; GFX11-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) +; GFX11-TRUE16-NEXT: v_cndmask_b16 v1.l, v3.l, v1.l, s2 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: test_fmin_legacy_ule_v4f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v4, 16, v3 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v5, 16, v1 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v6, 16, v2 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v7, 16, v0 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_1) | instid1(VALU_DEP_3) +; GFX11-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v5, v4 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v4, v4, v5, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v7, v6 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v5, v6, v7, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v0, v2 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, v2, v0, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v1, v3 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, v3, v1, vcc_lo +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v5, v0, 0x5040100 +; GFX11-FAKE16-NEXT: v_perm_b32 v1, v4, v1, 0x5040100 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = fcmp ule <4 x half> %a, %b %val = select <4 x i1> %cmp, <4 x half> %a, <4 x half> %b ret <4 x half> %val } +define <4 x half> @test_fmin_legacy_ule_v4f16_fast(<4 x half> %a, <4 x half> %b) #0 { +; GFX9-LABEL: test_fmin_legacy_ule_v4f16_fast: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_pk_min_f16 v0, v0, v2 +; GFX9-NEXT: v_pk_min_f16 v1, v1, v3 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; VI-LABEL: test_fmin_legacy_ule_v4f16_fast: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_min_f16_sdwa v4, v1, v3 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; VI-NEXT: v_min_f16_sdwa v5, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; VI-NEXT: v_min_f16_e32 v1, v1, v3 +; VI-NEXT: v_min_f16_e32 v0, v0, v2 +; VI-NEXT: v_or_b32_e32 v0, v0, v5 +; VI-NEXT: v_or_b32_e32 v1, v1, v4 +; VI-NEXT: s_setpc_b64 s[30:31] +; +; SI-LABEL: test_fmin_legacy_ule_v4f16_fast: +; SI: ; %bb.0: +; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; SI-NEXT: v_cvt_f16_f32_e32 v7, v7 +; SI-NEXT: v_cvt_f16_f32_e32 v3, v3 +; SI-NEXT: v_cvt_f16_f32_e32 v6, v6 +; SI-NEXT: v_cvt_f16_f32_e32 v2, v2 +; SI-NEXT: v_cvt_f16_f32_e32 v5, v5 +; SI-NEXT: v_cvt_f16_f32_e32 v1, v1 +; SI-NEXT: v_cvt_f16_f32_e32 v4, v4 +; SI-NEXT: v_cvt_f16_f32_e32 v0, v0 +; SI-NEXT: v_cvt_f32_f16_e32 v7, v7 +; SI-NEXT: v_cvt_f32_f16_e32 v3, v3 +; SI-NEXT: v_cvt_f32_f16_e32 v6, v6 +; SI-NEXT: v_cvt_f32_f16_e32 v2, v2 +; SI-NEXT: v_cvt_f32_f16_e32 v5, v5 +; SI-NEXT: v_cvt_f32_f16_e32 v1, v1 +; SI-NEXT: v_cvt_f32_f16_e32 v4, v4 +; SI-NEXT: v_cvt_f32_f16_e32 v0, v0 +; SI-NEXT: v_min_f32_e32 v0, v0, v4 +; SI-NEXT: v_min_f32_e32 v1, v1, v5 +; SI-NEXT: v_min_f32_e32 v2, v2, v6 +; SI-NEXT: v_min_f32_e32 v3, v3, v7 +; SI-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-LABEL: test_fmin_legacy_ule_v4f16_fast: +; GFX11: ; %bb.0: +; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-NEXT: v_pk_min_f16 v0, v0, v2 +; GFX11-NEXT: v_pk_min_f16 v1, v1, v3 +; GFX11-NEXT: s_setpc_b64 s[30:31] + %cmp = fcmp ule <4 x half> %a, %b + %val = select nnan nsz <4 x i1> %cmp, <4 x half> %a, <4 x half> %b + ret <4 x half> %val +} + define <8 x half> @test_fmin_legacy_ule_v8f16(<8 x half> %a, <8 x half> %b) #0 { -; GFX9-SAFE-LABEL: test_fmin_legacy_ule_v8f16: -; GFX9-SAFE: ; %bb.0: -; GFX9-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v14, 16, v7 -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v15, 16, v3 -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v12, 16, v6 -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v13, 16, v2 -; GFX9-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v15, v14 -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v10, 16, v5 -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v11, 16, v1 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v14, v14, v15, vcc -; GFX9-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v13, v12 -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v8, 16, v4 -; GFX9-SAFE-NEXT: v_lshrrev_b32_e32 v9, 16, v0 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v12, v12, v13, vcc -; GFX9-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v11, v10 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v10, v10, v11, vcc -; GFX9-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v9, v8 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v8, v8, v9, vcc -; GFX9-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v3, v7 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v3, v7, v3, vcc -; GFX9-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v2, v6 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v2, v6, v2, vcc -; GFX9-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v1, v5 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v1, v5, v1, vcc -; GFX9-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v0, v4 -; GFX9-SAFE-NEXT: v_cndmask_b32_e32 v0, v4, v0, vcc -; GFX9-SAFE-NEXT: s_mov_b32 s4, 0x5040100 -; GFX9-SAFE-NEXT: v_perm_b32 v0, v8, v0, s4 -; GFX9-SAFE-NEXT: v_perm_b32 v1, v10, v1, s4 -; GFX9-SAFE-NEXT: v_perm_b32 v2, v12, v2, s4 -; GFX9-SAFE-NEXT: v_perm_b32 v3, v14, v3, s4 -; GFX9-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; GFX9-NNAN-LABEL: test_fmin_legacy_ule_v8f16: -; GFX9-NNAN: ; %bb.0: -; GFX9-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX9-NNAN-NEXT: v_pk_min_f16 v0, v0, v4 -; GFX9-NNAN-NEXT: v_pk_min_f16 v1, v1, v5 -; GFX9-NNAN-NEXT: v_pk_min_f16 v2, v2, v6 -; GFX9-NNAN-NEXT: v_pk_min_f16 v3, v3, v7 -; GFX9-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; VI-SAFE-LABEL: test_fmin_legacy_ule_v8f16: -; VI-SAFE: ; %bb.0: -; VI-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v14, 16, v7 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v15, 16, v3 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v12, 16, v6 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v13, 16, v2 -; VI-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v15, v14 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v10, 16, v5 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v11, 16, v1 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v14, v14, v15, vcc -; VI-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v13, v12 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v8, 16, v4 -; VI-SAFE-NEXT: v_lshrrev_b32_e32 v9, 16, v0 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v12, v12, v13, vcc -; VI-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v11, v10 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v10, v10, v11, vcc -; VI-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v9, v8 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v8, v8, v9, vcc -; VI-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v3, v7 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v3, v7, v3, vcc -; VI-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v2, v6 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v2, v6, v2, vcc -; VI-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v1, v5 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v1, v5, v1, vcc -; VI-SAFE-NEXT: v_cmp_ngt_f16_e32 vcc, v0, v4 -; VI-SAFE-NEXT: v_cndmask_b32_e32 v0, v4, v0, vcc -; VI-SAFE-NEXT: v_lshlrev_b32_e32 v4, 16, v8 -; VI-SAFE-NEXT: v_or_b32_sdwa v0, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD -; VI-SAFE-NEXT: v_lshlrev_b32_e32 v4, 16, v10 -; VI-SAFE-NEXT: v_or_b32_sdwa v1, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD -; VI-SAFE-NEXT: v_lshlrev_b32_e32 v4, 16, v12 -; VI-SAFE-NEXT: v_or_b32_sdwa v2, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD -; VI-SAFE-NEXT: v_lshlrev_b32_e32 v4, 16, v14 -; VI-SAFE-NEXT: v_or_b32_sdwa v3, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD -; VI-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; VI-NNAN-LABEL: test_fmin_legacy_ule_v8f16: -; VI-NNAN: ; %bb.0: -; VI-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; VI-NNAN-NEXT: v_min_f16_sdwa v8, v3, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; VI-NNAN-NEXT: v_min_f16_sdwa v9, v2, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; VI-NNAN-NEXT: v_min_f16_sdwa v10, v1, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; VI-NNAN-NEXT: v_min_f16_sdwa v11, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 -; VI-NNAN-NEXT: v_min_f16_e32 v3, v3, v7 -; VI-NNAN-NEXT: v_min_f16_e32 v2, v2, v6 -; VI-NNAN-NEXT: v_min_f16_e32 v1, v1, v5 -; VI-NNAN-NEXT: v_min_f16_e32 v0, v0, v4 -; VI-NNAN-NEXT: v_or_b32_e32 v0, v0, v11 -; VI-NNAN-NEXT: v_or_b32_e32 v1, v1, v10 -; VI-NNAN-NEXT: v_or_b32_e32 v2, v2, v9 -; VI-NNAN-NEXT: v_or_b32_e32 v3, v3, v8 -; VI-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; SI-SAFE-LABEL: test_fmin_legacy_ule_v8f16: -; SI-SAFE: ; %bb.0: -; SI-SAFE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v7, v7 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v15, v15 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v6, v6 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v14, v14 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v5, v5 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v13, v13 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v4, v4 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v12, v12 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v3, v3 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v11, v11 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v2, v2 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v10, v10 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v1, v1 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v9, v9 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v0, v0 -; SI-SAFE-NEXT: v_cvt_f16_f32_e32 v8, v8 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v7, v7 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v15, v15 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v6, v6 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v14, v14 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v5, v5 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v13, v13 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v4, v4 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v12, v12 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v3, v3 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v11, v11 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v2, v2 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v10, v10 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v1, v1 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v9, v9 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-SAFE-NEXT: v_cvt_f32_f16_e32 v8, v8 -; SI-SAFE-NEXT: v_min_legacy_f32_e32 v0, v8, v0 -; SI-SAFE-NEXT: v_min_legacy_f32_e32 v1, v9, v1 -; SI-SAFE-NEXT: v_min_legacy_f32_e32 v2, v10, v2 -; SI-SAFE-NEXT: v_min_legacy_f32_e32 v3, v11, v3 -; SI-SAFE-NEXT: v_min_legacy_f32_e32 v4, v12, v4 -; SI-SAFE-NEXT: v_min_legacy_f32_e32 v5, v13, v5 -; SI-SAFE-NEXT: v_min_legacy_f32_e32 v6, v14, v6 -; SI-SAFE-NEXT: v_min_legacy_f32_e32 v7, v15, v7 -; SI-SAFE-NEXT: s_setpc_b64 s[30:31] -; -; SI-NNAN-LABEL: test_fmin_legacy_ule_v8f16: -; SI-NNAN: ; %bb.0: -; SI-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v15, v15 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v7, v7 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v14, v14 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v6, v6 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v13, v13 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v5, v5 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v12, v12 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v4, v4 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v11, v11 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v3, v3 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v10, v10 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v2, v2 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v9, v9 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v1, v1 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v8, v8 -; SI-NNAN-NEXT: v_cvt_f16_f32_e32 v0, v0 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v15, v15 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v7, v7 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v14, v14 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v6, v6 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v13, v13 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v5, v5 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v12, v12 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v4, v4 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v11, v11 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v3, v3 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v10, v10 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v2, v2 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v9, v9 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v1, v1 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v8, v8 -; SI-NNAN-NEXT: v_cvt_f32_f16_e32 v0, v0 -; SI-NNAN-NEXT: v_min_f32_e32 v0, v0, v8 -; SI-NNAN-NEXT: v_min_f32_e32 v1, v1, v9 -; SI-NNAN-NEXT: v_min_f32_e32 v2, v2, v10 -; SI-NNAN-NEXT: v_min_f32_e32 v3, v3, v11 -; SI-NNAN-NEXT: v_min_f32_e32 v4, v4, v12 -; SI-NNAN-NEXT: v_min_f32_e32 v5, v5, v13 -; SI-NNAN-NEXT: v_min_f32_e32 v6, v6, v14 -; SI-NNAN-NEXT: v_min_f32_e32 v7, v7, v15 -; SI-NNAN-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-TRUE16-LABEL: test_fmin_legacy_ule_v8f16: -; GFX11-SAFE-TRUE16: ; %bb.0: -; GFX11-SAFE-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v0.h, v4.h -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ngt_f16_e64 s0, v1.h, v5.h -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ngt_f16_e64 s1, v2.h, v6.h -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ngt_f16_e64 s2, v3.h, v7.h -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ngt_f16_e64 s3, v0.l, v4.l -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ngt_f16_e64 s4, v1.l, v5.l -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ngt_f16_e64 s5, v2.l, v6.l -; GFX11-SAFE-TRUE16-NEXT: v_cmp_ngt_f16_e64 s6, v3.l, v7.l -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v3.h, v7.h, v3.h, s2 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v2.h, v6.h, v2.h, s1 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v1.h, v5.h, v1.h, s0 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.h, v4.h, v0.h, vcc_lo -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v0.l, v4.l, v0.l, s3 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v1.l, v5.l, v1.l, s4 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v2.l, v6.l, v2.l, s5 -; GFX11-SAFE-TRUE16-NEXT: v_cndmask_b16 v3.l, v7.l, v3.l, s6 -; GFX11-SAFE-TRUE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-SAFE-FAKE16-LABEL: test_fmin_legacy_ule_v8f16: -; GFX11-SAFE-FAKE16: ; %bb.0: -; GFX11-SAFE-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v10, 16, v7 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v11, 16, v3 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v12, 16, v6 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v13, 16, v2 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v14, 16, v5 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v15, 16, v1 -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v11, v10 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v8, 16, v4 -; GFX11-SAFE-FAKE16-NEXT: v_lshrrev_b32_e32 v9, 16, v0 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v10, v10, v11, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v13, v12 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v11, v12, v13, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v15, v14 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v12, v14, v15, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v9, v8 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v8, v8, v9, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v2, v6 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v2, v6, v2, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v0, v4 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(VALU_DEP_2) -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v2, v11, v2, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v0, v4, v0, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v1, v5 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v1, v5, v1, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v3, v7 -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v1, v12, v1, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: v_cndmask_b32_e32 v3, v7, v3, vcc_lo -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v0, v8, v0, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) -; GFX11-SAFE-FAKE16-NEXT: v_perm_b32 v3, v10, v3, 0x5040100 -; GFX11-SAFE-FAKE16-NEXT: s_setpc_b64 s[30:31] -; -; GFX11-NNAN-LABEL: test_fmin_legacy_ule_v8f16: -; GFX11-NNAN: ; %bb.0: -; GFX11-NNAN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) -; GFX11-NNAN-NEXT: v_pk_min_f16 v0, v0, v4 -; GFX11-NNAN-NEXT: v_pk_min_f16 v1, v1, v5 -; GFX11-NNAN-NEXT: v_pk_min_f16 v2, v2, v6 -; GFX11-NNAN-NEXT: v_pk_min_f16 v3, v3, v7 -; GFX11-NNAN-NEXT: s_setpc_b64 s[30:31] +; GFX9-LABEL: test_fmin_legacy_ule_v8f16: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_lshrrev_b32_e32 v14, 16, v7 +; GFX9-NEXT: v_lshrrev_b32_e32 v15, 16, v3 +; GFX9-NEXT: v_lshrrev_b32_e32 v12, 16, v6 +; GFX9-NEXT: v_lshrrev_b32_e32 v13, 16, v2 +; GFX9-NEXT: v_cmp_ngt_f16_e32 vcc, v15, v14 +; GFX9-NEXT: v_lshrrev_b32_e32 v10, 16, v5 +; GFX9-NEXT: v_lshrrev_b32_e32 v11, 16, v1 +; GFX9-NEXT: v_cndmask_b32_e32 v14, v14, v15, vcc +; GFX9-NEXT: v_cmp_ngt_f16_e32 vcc, v13, v12 +; GFX9-NEXT: v_lshrrev_b32_e32 v8, 16, v4 +; GFX9-NEXT: v_lshrrev_b32_e32 v9, 16, v0 +; GFX9-NEXT: v_cndmask_b32_e32 v12, v12, v13, vcc +; GFX9-NEXT: v_cmp_ngt_f16_e32 vcc, v11, v10 +; GFX9-NEXT: v_cndmask_b32_e32 v10, v10, v11, vcc +; GFX9-NEXT: v_cmp_ngt_f16_e32 vcc, v9, v8 +; GFX9-NEXT: v_cndmask_b32_e32 v8, v8, v9, vcc +; GFX9-NEXT: v_cmp_ngt_f16_e32 vcc, v3, v7 +; GFX9-NEXT: v_cndmask_b32_e32 v3, v7, v3, vcc +; GFX9-NEXT: v_cmp_ngt_f16_e32 vcc, v2, v6 +; GFX9-NEXT: v_cndmask_b32_e32 v2, v6, v2, vcc +; GFX9-NEXT: v_cmp_ngt_f16_e32 vcc, v1, v5 +; GFX9-NEXT: v_cndmask_b32_e32 v1, v5, v1, vcc +; GFX9-NEXT: v_cmp_ngt_f16_e32 vcc, v0, v4 +; GFX9-NEXT: v_cndmask_b32_e32 v0, v4, v0, vcc +; GFX9-NEXT: s_mov_b32 s4, 0x5040100 +; GFX9-NEXT: v_perm_b32 v0, v8, v0, s4 +; GFX9-NEXT: v_perm_b32 v1, v10, v1, s4 +; GFX9-NEXT: v_perm_b32 v2, v12, v2, s4 +; GFX9-NEXT: v_perm_b32 v3, v14, v3, s4 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; VI-LABEL: test_fmin_legacy_ule_v8f16: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_lshrrev_b32_e32 v14, 16, v7 +; VI-NEXT: v_lshrrev_b32_e32 v15, 16, v3 +; VI-NEXT: v_lshrrev_b32_e32 v12, 16, v6 +; VI-NEXT: v_lshrrev_b32_e32 v13, 16, v2 +; VI-NEXT: v_cmp_ngt_f16_e32 vcc, v15, v14 +; VI-NEXT: v_lshrrev_b32_e32 v10, 16, v5 +; VI-NEXT: v_lshrrev_b32_e32 v11, 16, v1 +; VI-NEXT: v_cndmask_b32_e32 v14, v14, v15, vcc +; VI-NEXT: v_cmp_ngt_f16_e32 vcc, v13, v12 +; VI-NEXT: v_lshrrev_b32_e32 v8, 16, v4 +; VI-NEXT: v_lshrrev_b32_e32 v9, 16, v0 +; VI-NEXT: v_cndmask_b32_e32 v12, v12, v13, vcc +; VI-NEXT: v_cmp_ngt_f16_e32 vcc, v11, v10 +; VI-NEXT: v_cndmask_b32_e32 v10, v10, v11, vcc +; VI-NEXT: v_cmp_ngt_f16_e32 vcc, v9, v8 +; VI-NEXT: v_cndmask_b32_e32 v8, v8, v9, vcc +; VI-NEXT: v_cmp_ngt_f16_e32 vcc, v3, v7 +; VI-NEXT: v_cndmask_b32_e32 v3, v7, v3, vcc +; VI-NEXT: v_cmp_ngt_f16_e32 vcc, v2, v6 +; VI-NEXT: v_cndmask_b32_e32 v2, v6, v2, vcc +; VI-NEXT: v_cmp_ngt_f16_e32 vcc, v1, v5 +; VI-NEXT: v_cndmask_b32_e32 v1, v5, v1, vcc +; VI-NEXT: v_cmp_ngt_f16_e32 vcc, v0, v4 +; VI-NEXT: v_cndmask_b32_e32 v0, v4, v0, vcc +; VI-NEXT: v_lshlrev_b32_e32 v4, 16, v8 +; VI-NEXT: v_or_b32_sdwa v0, v0, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD +; VI-NEXT: v_lshlrev_b32_e32 v4, 16, v10 +; VI-NEXT: v_or_b32_sdwa v1, v1, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD +; VI-NEXT: v_lshlrev_b32_e32 v4, 16, v12 +; VI-NEXT: v_or_b32_sdwa v2, v2, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD +; VI-NEXT: v_lshlrev_b32_e32 v4, 16, v14 +; VI-NEXT: v_or_b32_sdwa v3, v3, v4 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:WORD_0 src1_sel:DWORD +; VI-NEXT: s_setpc_b64 s[30:31] +; +; SI-LABEL: test_fmin_legacy_ule_v8f16: +; SI: ; %bb.0: +; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; SI-NEXT: v_cvt_f16_f32_e32 v7, v7 +; SI-NEXT: v_cvt_f16_f32_e32 v15, v15 +; SI-NEXT: v_cvt_f16_f32_e32 v6, v6 +; SI-NEXT: v_cvt_f16_f32_e32 v14, v14 +; SI-NEXT: v_cvt_f16_f32_e32 v5, v5 +; SI-NEXT: v_cvt_f16_f32_e32 v13, v13 +; SI-NEXT: v_cvt_f16_f32_e32 v4, v4 +; SI-NEXT: v_cvt_f16_f32_e32 v12, v12 +; SI-NEXT: v_cvt_f16_f32_e32 v3, v3 +; SI-NEXT: v_cvt_f16_f32_e32 v11, v11 +; SI-NEXT: v_cvt_f16_f32_e32 v2, v2 +; SI-NEXT: v_cvt_f16_f32_e32 v10, v10 +; SI-NEXT: v_cvt_f16_f32_e32 v1, v1 +; SI-NEXT: v_cvt_f16_f32_e32 v9, v9 +; SI-NEXT: v_cvt_f16_f32_e32 v0, v0 +; SI-NEXT: v_cvt_f16_f32_e32 v8, v8 +; SI-NEXT: v_cvt_f32_f16_e32 v7, v7 +; SI-NEXT: v_cvt_f32_f16_e32 v15, v15 +; SI-NEXT: v_cvt_f32_f16_e32 v6, v6 +; SI-NEXT: v_cvt_f32_f16_e32 v14, v14 +; SI-NEXT: v_cvt_f32_f16_e32 v5, v5 +; SI-NEXT: v_cvt_f32_f16_e32 v13, v13 +; SI-NEXT: v_cvt_f32_f16_e32 v4, v4 +; SI-NEXT: v_cvt_f32_f16_e32 v12, v12 +; SI-NEXT: v_cvt_f32_f16_e32 v3, v3 +; SI-NEXT: v_cvt_f32_f16_e32 v11, v11 +; SI-NEXT: v_cvt_f32_f16_e32 v2, v2 +; SI-NEXT: v_cvt_f32_f16_e32 v10, v10 +; SI-NEXT: v_cvt_f32_f16_e32 v1, v1 +; SI-NEXT: v_cvt_f32_f16_e32 v9, v9 +; SI-NEXT: v_cvt_f32_f16_e32 v0, v0 +; SI-NEXT: v_cvt_f32_f16_e32 v8, v8 +; SI-NEXT: v_min_legacy_f32_e32 v0, v8, v0 +; SI-NEXT: v_min_legacy_f32_e32 v1, v9, v1 +; SI-NEXT: v_min_legacy_f32_e32 v2, v10, v2 +; SI-NEXT: v_min_legacy_f32_e32 v3, v11, v3 +; SI-NEXT: v_min_legacy_f32_e32 v4, v12, v4 +; SI-NEXT: v_min_legacy_f32_e32 v5, v13, v5 +; SI-NEXT: v_min_legacy_f32_e32 v6, v14, v6 +; SI-NEXT: v_min_legacy_f32_e32 v7, v15, v7 +; SI-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-TRUE16-LABEL: test_fmin_legacy_ule_v8f16: +; GFX11-TRUE16: ; %bb.0: +; GFX11-TRUE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-TRUE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v0.h, v4.h +; GFX11-TRUE16-NEXT: v_cmp_ngt_f16_e64 s0, v1.h, v5.h +; GFX11-TRUE16-NEXT: v_cmp_ngt_f16_e64 s1, v2.h, v6.h +; GFX11-TRUE16-NEXT: v_cmp_ngt_f16_e64 s2, v3.h, v7.h +; GFX11-TRUE16-NEXT: v_cmp_ngt_f16_e64 s3, v0.l, v4.l +; GFX11-TRUE16-NEXT: v_cmp_ngt_f16_e64 s4, v1.l, v5.l +; GFX11-TRUE16-NEXT: v_cmp_ngt_f16_e64 s5, v2.l, v6.l +; GFX11-TRUE16-NEXT: v_cmp_ngt_f16_e64 s6, v3.l, v7.l +; GFX11-TRUE16-NEXT: v_cndmask_b16 v3.h, v7.h, v3.h, s2 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v2.h, v6.h, v2.h, s1 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v1.h, v5.h, v1.h, s0 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.h, v4.h, v0.h, vcc_lo +; GFX11-TRUE16-NEXT: v_cndmask_b16 v0.l, v4.l, v0.l, s3 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v1.l, v5.l, v1.l, s4 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v2.l, v6.l, v2.l, s5 +; GFX11-TRUE16-NEXT: v_cndmask_b16 v3.l, v7.l, v3.l, s6 +; GFX11-TRUE16-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-FAKE16-LABEL: test_fmin_legacy_ule_v8f16: +; GFX11-FAKE16: ; %bb.0: +; GFX11-FAKE16-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v10, 16, v7 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v11, 16, v3 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v12, 16, v6 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v13, 16, v2 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v14, 16, v5 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v15, 16, v1 +; GFX11-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v11, v10 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v8, 16, v4 +; GFX11-FAKE16-NEXT: v_lshrrev_b32_e32 v9, 16, v0 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v10, v10, v11, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v13, v12 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v11, v12, v13, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v15, v14 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v12, v14, v15, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v9, v8 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v8, v8, v9, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v2, v6 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v2, v6, v2, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v0, v4 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_4) | instid1(VALU_DEP_2) +; GFX11-FAKE16-NEXT: v_perm_b32 v2, v11, v2, 0x5040100 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v0, v4, v0, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v1, v5 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v1, v5, v1, vcc_lo +; GFX11-FAKE16-NEXT: v_cmp_ngt_f16_e32 vcc_lo, v3, v7 +; GFX11-FAKE16-NEXT: v_perm_b32 v1, v12, v1, 0x5040100 +; GFX11-FAKE16-NEXT: v_cndmask_b32_e32 v3, v7, v3, vcc_lo +; GFX11-FAKE16-NEXT: v_perm_b32 v0, v8, v0, 0x5040100 +; GFX11-FAKE16-NEXT: s_delay_alu instid0(VALU_DEP_2) +; GFX11-FAKE16-NEXT: v_perm_b32 v3, v10, v3, 0x5040100 +; GFX11-FAKE16-NEXT: s_setpc_b64 s[30:31] %cmp = fcmp ule <8 x half> %a, %b %val = select <8 x i1> %cmp, <8 x half> %a, <8 x half> %b ret <8 x half> %val } +define <8 x half> @test_fmin_legacy_ule_v8f16_fast(<8 x half> %a, <8 x half> %b) #0 { +; GFX9-LABEL: test_fmin_legacy_ule_v8f16_fast: +; GFX9: ; %bb.0: +; GFX9-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX9-NEXT: v_pk_min_f16 v0, v0, v4 +; GFX9-NEXT: v_pk_min_f16 v1, v1, v5 +; GFX9-NEXT: v_pk_min_f16 v2, v2, v6 +; GFX9-NEXT: v_pk_min_f16 v3, v3, v7 +; GFX9-NEXT: s_setpc_b64 s[30:31] +; +; VI-LABEL: test_fmin_legacy_ule_v8f16_fast: +; VI: ; %bb.0: +; VI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; VI-NEXT: v_min_f16_sdwa v8, v3, v7 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; VI-NEXT: v_min_f16_sdwa v9, v2, v6 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; VI-NEXT: v_min_f16_sdwa v10, v1, v5 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; VI-NEXT: v_min_f16_sdwa v11, v0, v4 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:WORD_1 +; VI-NEXT: v_min_f16_e32 v3, v3, v7 +; VI-NEXT: v_min_f16_e32 v2, v2, v6 +; VI-NEXT: v_min_f16_e32 v1, v1, v5 +; VI-NEXT: v_min_f16_e32 v0, v0, v4 +; VI-NEXT: v_or_b32_e32 v0, v0, v11 +; VI-NEXT: v_or_b32_e32 v1, v1, v10 +; VI-NEXT: v_or_b32_e32 v2, v2, v9 +; VI-NEXT: v_or_b32_e32 v3, v3, v8 +; VI-NEXT: s_setpc_b64 s[30:31] +; +; SI-LABEL: test_fmin_legacy_ule_v8f16_fast: +; SI: ; %bb.0: +; SI-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; SI-NEXT: v_cvt_f16_f32_e32 v15, v15 +; SI-NEXT: v_cvt_f16_f32_e32 v7, v7 +; SI-NEXT: v_cvt_f16_f32_e32 v14, v14 +; SI-NEXT: v_cvt_f16_f32_e32 v6, v6 +; SI-NEXT: v_cvt_f16_f32_e32 v13, v13 +; SI-NEXT: v_cvt_f16_f32_e32 v5, v5 +; SI-NEXT: v_cvt_f16_f32_e32 v12, v12 +; SI-NEXT: v_cvt_f16_f32_e32 v4, v4 +; SI-NEXT: v_cvt_f16_f32_e32 v11, v11 +; SI-NEXT: v_cvt_f16_f32_e32 v3, v3 +; SI-NEXT: v_cvt_f16_f32_e32 v10, v10 +; SI-NEXT: v_cvt_f16_f32_e32 v2, v2 +; SI-NEXT: v_cvt_f16_f32_e32 v9, v9 +; SI-NEXT: v_cvt_f16_f32_e32 v1, v1 +; SI-NEXT: v_cvt_f16_f32_e32 v8, v8 +; SI-NEXT: v_cvt_f16_f32_e32 v0, v0 +; SI-NEXT: v_cvt_f32_f16_e32 v15, v15 +; SI-NEXT: v_cvt_f32_f16_e32 v7, v7 +; SI-NEXT: v_cvt_f32_f16_e32 v14, v14 +; SI-NEXT: v_cvt_f32_f16_e32 v6, v6 +; SI-NEXT: v_cvt_f32_f16_e32 v13, v13 +; SI-NEXT: v_cvt_f32_f16_e32 v5, v5 +; SI-NEXT: v_cvt_f32_f16_e32 v12, v12 +; SI-NEXT: v_cvt_f32_f16_e32 v4, v4 +; SI-NEXT: v_cvt_f32_f16_e32 v11, v11 +; SI-NEXT: v_cvt_f32_f16_e32 v3, v3 +; SI-NEXT: v_cvt_f32_f16_e32 v10, v10 +; SI-NEXT: v_cvt_f32_f16_e32 v2, v2 +; SI-NEXT: v_cvt_f32_f16_e32 v9, v9 +; SI-NEXT: v_cvt_f32_f16_e32 v1, v1 +; SI-NEXT: v_cvt_f32_f16_e32 v8, v8 +; SI-NEXT: v_cvt_f32_f16_e32 v0, v0 +; SI-NEXT: v_min_f32_e32 v0, v0, v8 +; SI-NEXT: v_min_f32_e32 v1, v1, v9 +; SI-NEXT: v_min_f32_e32 v2, v2, v10 +; SI-NEXT: v_min_f32_e32 v3, v3, v11 +; SI-NEXT: v_min_f32_e32 v4, v4, v12 +; SI-NEXT: v_min_f32_e32 v5, v5, v13 +; SI-NEXT: v_min_f32_e32 v6, v6, v14 +; SI-NEXT: v_min_f32_e32 v7, v7, v15 +; SI-NEXT: s_setpc_b64 s[30:31] +; +; GFX11-LABEL: test_fmin_legacy_ule_v8f16_fast: +; GFX11: ; %bb.0: +; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) +; GFX11-NEXT: v_pk_min_f16 v0, v0, v4 +; GFX11-NEXT: v_pk_min_f16 v1, v1, v5 +; GFX11-NEXT: v_pk_min_f16 v2, v2, v6 +; GFX11-NEXT: v_pk_min_f16 v3, v3, v7 +; GFX11-NEXT: s_setpc_b64 s[30:31] + %cmp = fcmp ule <8 x half> %a, %b + %val = select nnan nsz <8 x i1> %cmp, <8 x half> %a, <8 x half> %b + ret <8 x half> %val +} + attributes #0 = { nounwind } diff --git a/llvm/test/CodeGen/AMDGPU/fmin_legacy.ll b/llvm/test/CodeGen/AMDGPU/fmin_legacy.ll index ec4dd85..defcffa 100644 --- a/llvm/test/CodeGen/AMDGPU/fmin_legacy.ll +++ b/llvm/test/CodeGen/AMDGPU/fmin_legacy.ll @@ -1,8 +1,6 @@ -; RUN: llc -mtriple=amdgcn < %s | FileCheck -enable-var-scope -check-prefixes=SI-SAFE,GCN,FUNC %s -; RUN: llc -enable-no-nans-fp-math -enable-no-signed-zeros-fp-math -mtriple=amdgcn < %s | FileCheck -enable-var-scope -check-prefixes=SI-NONAN,GCN-NONAN,GCN,FUNC %s +; RUN: llc -mtriple=amdgcn < %s | FileCheck -enable-var-scope -check-prefixes=SI,GCN,FUNC %s -; RUN: llc -mtriple=amdgcn -mcpu=fiji < %s | FileCheck -enable-var-scope -check-prefixes=VI-SAFE,GCN,FUNC %s -; RUN: llc -enable-no-nans-fp-math -enable-no-signed-zeros-fp-math -mtriple=amdgcn -mcpu=fiji < %s | FileCheck -enable-var-scope -check-prefixes=VI-NONAN,GCN-NONAN,GCN,FUNC %s +; RUN: llc -mtriple=amdgcn -mcpu=fiji < %s | FileCheck -enable-var-scope -check-prefixes=VI,GCN,FUNC %s ; RUN: llc -mtriple=r600 -mcpu=redwood < %s | FileCheck -enable-var-scope --check-prefixes=EG,FUNC %s @@ -14,13 +12,9 @@ declare i32 @llvm.amdgcn.workitem.id.x() #1 ; FUNC-LABEL: {{^}}s_test_fmin_legacy_subreg_inputs_f32: ; EG: MIN * -; SI-SAFE: v_min_legacy_f32_e32 v{{[0-9]+}}, s{{[0-9]+}}, v{{[0-9]+}} +; SI: v_min_legacy_f32_e32 v{{[0-9]+}}, s{{[0-9]+}}, v{{[0-9]+}} -; SI-NONAN: v_min_f32_e32 v{{[0-9]+}}, s{{[0-9]+}}, v{{[0-9]+}} - -; VI-SAFE: v_cmp_nlt_f32_e32 vcc, s{{[0-9]+}}, v{{[0-9]+}} - -; VI-NONAN: v_min_f32_e32 v{{[0-9]+}}, s{{[0-9]+}}, v{{[0-9]+}} +; VI: v_cmp_nlt_f32_e32 vcc, s{{[0-9]+}}, v{{[0-9]+}} define amdgpu_kernel void @s_test_fmin_legacy_subreg_inputs_f32(ptr addrspace(1) %out, <4 x float> %reg0) #0 { %r0 = extractelement <4 x float> %reg0, i32 0 %r1 = extractelement <4 x float> %reg0, i32 1 @@ -30,22 +24,32 @@ define amdgpu_kernel void @s_test_fmin_legacy_subreg_inputs_f32(ptr addrspace(1) ret void } -; FUNC-LABEL: {{^}}s_test_fmin_legacy_ule_f32: -; GCN-DAG: s_load_dwordx4 s[[[#LOAD:]]:{{[0-9]+}}], s{{\[[0-9]+:[0-9]+\]}}, {{0x9|0x24}} +; FUNC-LABEL: {{^}}s_test_fmin_legacy_subreg_inputs_f32_fast: -; SI-SAFE: v_mov_b32_e32 [[VA:v[0-9]+]], s[[#LOAD + 2]] +; SI: v_min_f32_e32 v{{[0-9]+}}, s{{[0-9]+}}, v{{[0-9]+}} -; GCN-NONAN: v_mov_b32_e32 [[VB:v[0-9]+]], s[[#LOAD + 3]] +; VI: v_min_f32_e32 v{{[0-9]+}}, s{{[0-9]+}}, v{{[0-9]+}} +define amdgpu_kernel void @s_test_fmin_legacy_subreg_inputs_f32_fast(ptr addrspace(1) %out, <4 x float> %reg0) #0 { + %r0 = extractelement <4 x float> %reg0, i32 0 + %r1 = extractelement <4 x float> %reg0, i32 1 + %r2 = fcmp nnan nsz uge float %r0, %r1 + %r3 = select nnan nsz i1 %r2, float %r1, float %r0 + store float %r3, ptr addrspace(1) %out + ret void +} -; VI-SAFE: v_mov_b32_e32 [[VB:v[0-9]+]], s[[#LOAD + 3]] +; FUNC-LABEL: {{^}}s_test_fmin_legacy_ule_f32: +; GCN-DAG: s_load_dwordx4 s[[[#LOAD:]]:{{[0-9]+}}], s{{\[[0-9]+:[0-9]+\]}}, {{0x9|0x24}} + +; SI: v_mov_b32_e32 [[VA:v[0-9]+]], s[[#LOAD + 2]] -; SI-SAFE: v_min_legacy_f32_e32 {{v[0-9]+}}, s[[#LOAD + 3]], [[VA]] +; VI: v_mov_b32_e32 [[VB:v[0-9]+]], s[[#LOAD + 3]] -; VI-SAFE: v_mov_b32_e32 [[VA:v[0-9]+]], s[[#LOAD + 2]] -; VI-SAFE: v_cmp_ngt_f32_e32 vcc, s[[#LOAD + 2]], [[VB]] -; VI-SAFE: v_cndmask_b32_e32 v{{[0-9]+}}, [[VB]], [[VA]] +; SI: v_min_legacy_f32_e32 {{v[0-9]+}}, s[[#LOAD + 3]], [[VA]] -; GCN-NONAN: v_min_f32_e32 {{v[0-9]+}}, s[[#LOAD + 2]], [[VB]] +; VI: v_mov_b32_e32 [[VA:v[0-9]+]], s[[#LOAD + 2]] +; VI: v_cmp_ngt_f32_e32 vcc, s[[#LOAD + 2]], [[VB]] +; VI: v_cndmask_b32_e32 v{{[0-9]+}}, [[VB]], [[VA]] define amdgpu_kernel void @s_test_fmin_legacy_ule_f32(ptr addrspace(1) %out, float %a, float %b) #0 { %cmp = fcmp ule float %a, %b %val = select i1 %cmp, float %a, float %b @@ -53,6 +57,19 @@ define amdgpu_kernel void @s_test_fmin_legacy_ule_f32(ptr addrspace(1) %out, flo ret void } +; FUNC-LABEL: {{^}}s_test_fmin_legacy_ule_f32_fast: +; GCN-DAG: s_load_dwordx4 s[[[#LOAD:]]:{{[0-9]+}}], s{{\[[0-9]+:[0-9]+\]}}, {{0x9|0x24}} + +; GCN: v_mov_b32_e32 [[VB:v[0-9]+]], s[[#LOAD + 3]] + +; GCN: v_min_f32_e32 {{v[0-9]+}}, s[[#LOAD + 2]], [[VB]] +define amdgpu_kernel void @s_test_fmin_legacy_ule_f32_fast(ptr addrspace(1) %out, float %a, float %b) #0 { + %cmp = fcmp ule float %a, %b + %val = select nnan nsz i1 %cmp, float %a, float %b + store float %val, ptr addrspace(1) %out, align 4 + ret void +} + ; Nsz also needed ; FIXME: Should separate tests ; GCN-LABEL: {{^}}s_test_fmin_legacy_ule_f32_nnan_src: @@ -61,12 +78,10 @@ define amdgpu_kernel void @s_test_fmin_legacy_ule_f32(ptr addrspace(1) %out, flo ; GCN-DAG: v_add_f32_e64 [[ADD_A:v[0-9]+]], s[[#LOAD + 2]], 1.0 ; GCN-DAG: v_add_f32_e64 [[ADD_B:v[0-9]+]], s[[#LOAD + 3]], 2.0 -; SI-SAFE: v_min_legacy_f32_e32 {{v[0-9]+}}, [[ADD_B]], [[ADD_A]] - -; VI-SAFE: v_cmp_ngt_f32_e32 vcc, [[ADD_A]], [[ADD_B]] -; VI-SAFE: v_cndmask_b32_e32 {{v[0-9]+}}, [[ADD_B]], [[ADD_A]], vcc +; SI: v_min_legacy_f32_e32 {{v[0-9]+}}, [[ADD_B]], [[ADD_A]] -; GCN-NONAN: v_min_f32_e32 {{v[0-9]+}}, [[ADD_A]], [[ADD_B]] +; VI: v_cmp_ngt_f32_e32 vcc, [[ADD_A]], [[ADD_B]] +; VI: v_cndmask_b32_e32 {{v[0-9]+}}, [[ADD_B]], [[ADD_A]], vcc define amdgpu_kernel void @s_test_fmin_legacy_ule_f32_nnan_src(ptr addrspace(1) %out, float %a, float %b) #0 { %a.nnan = fadd nnan float %a, 1.0 %b.nnan = fadd nnan float %b, 2.0 @@ -76,16 +91,32 @@ define amdgpu_kernel void @s_test_fmin_legacy_ule_f32_nnan_src(ptr addrspace(1) ret void } +; Nsz also needed +; FIXME: Should separate tests +; GCN-LABEL: {{^}}s_test_fmin_legacy_ule_f32_nnan_src_fast: +; GCN: s_load_dwordx4 s[[[#LOAD:]]:{{[0-9]+}}], s{{\[[0-9]+:[0-9]+\]}}, {{0x9|0x24}} + +; GCN-DAG: v_add_f32_e64 [[ADD_A:v[0-9]+]], s[[#LOAD + 2]], 1.0 +; GCN-DAG: v_add_f32_e64 [[ADD_B:v[0-9]+]], s[[#LOAD + 3]], 2.0 + +; GCN: v_min_f32_e32 {{v[0-9]+}}, [[ADD_A]], [[ADD_B]] +define amdgpu_kernel void @s_test_fmin_legacy_ule_f32_nnan_src_fast(ptr addrspace(1) %out, float %a, float %b) #0 { + %a.nnan = fadd nnan float %a, 1.0 + %b.nnan = fadd nnan float %b, 2.0 + %cmp = fcmp ule float %a.nnan, %b.nnan + %val = select nnan nsz i1 %cmp, float %a.nnan, float %b.nnan + store float %val, ptr addrspace(1) %out, align 4 + ret void +} + ; FUNC-LABEL: {{^}}test_fmin_legacy_ule_f32: ; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]] ; GCN: {{buffer|flat}}_load_dword [[B:v[0-9]+]] -; SI-SAFE: v_min_legacy_f32_e32 {{v[0-9]+}}, [[B]], [[A]] +; SI: v_min_legacy_f32_e32 {{v[0-9]+}}, [[B]], [[A]] -; VI-SAFE: v_cmp_ngt_f32_e32 vcc, [[A]], [[B]] -; VI-SAFE: v_cndmask_b32_e32 v{{[0-9]+}}, [[B]], [[A]] - -; GCN-NONAN: v_min_f32_e32 {{v[0-9]+}}, [[A]], [[B]] +; VI: v_cmp_ngt_f32_e32 vcc, [[A]], [[B]] +; VI: v_cndmask_b32_e32 v{{[0-9]+}}, [[B]], [[A]] define amdgpu_kernel void @test_fmin_legacy_ule_f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { %tid = call i32 @llvm.amdgcn.workitem.id.x() #1 %gep.0 = getelementptr float, ptr addrspace(1) %in, i32 %tid @@ -100,16 +131,33 @@ define amdgpu_kernel void @test_fmin_legacy_ule_f32(ptr addrspace(1) %out, ptr a ret void } -; FUNC-LABEL: {{^}}test_fmin_legacy_ole_f32: +; FUNC-LABEL: {{^}}test_fmin_legacy_ule_f32_fast: ; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]] ; GCN: {{buffer|flat}}_load_dword [[B:v[0-9]+]] -; SI-SAFE: v_min_legacy_f32_e32 {{v[0-9]+}}, [[A]], [[B]] +; GCN: v_min_f32_e32 {{v[0-9]+}}, [[A]], [[B]] +define amdgpu_kernel void @test_fmin_legacy_ule_f32_fast(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { + %tid = call i32 @llvm.amdgcn.workitem.id.x() #1 + %gep.0 = getelementptr float, ptr addrspace(1) %in, i32 %tid + %gep.1 = getelementptr float, ptr addrspace(1) %gep.0, i32 1 + + %a = load volatile float, ptr addrspace(1) %gep.0, align 4 + %b = load volatile float, ptr addrspace(1) %gep.1, align 4 -; VI-SAFE: v_cmp_le_f32_e32 vcc, [[A]], [[B]] -; VI-SAFE: v_cndmask_b32_e32 v{{[0-9]+}}, [[B]], [[A]] + %cmp = fcmp ule float %a, %b + %val = select nnan nsz i1 %cmp, float %a, float %b + store float %val, ptr addrspace(1) %out, align 4 + ret void +} -; GCN-NONAN: v_min_f32_e32 {{v[0-9]+}}, [[A]], [[B]] +; FUNC-LABEL: {{^}}test_fmin_legacy_ole_f32: +; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]] +; GCN: {{buffer|flat}}_load_dword [[B:v[0-9]+]] + +; SI: v_min_legacy_f32_e32 {{v[0-9]+}}, [[A]], [[B]] + +; VI: v_cmp_le_f32_e32 vcc, [[A]], [[B]] +; VI: v_cndmask_b32_e32 v{{[0-9]+}}, [[B]], [[A]] define amdgpu_kernel void @test_fmin_legacy_ole_f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { %tid = call i32 @llvm.amdgcn.workitem.id.x() #1 %gep.0 = getelementptr float, ptr addrspace(1) %in, i32 %tid @@ -124,16 +172,33 @@ define amdgpu_kernel void @test_fmin_legacy_ole_f32(ptr addrspace(1) %out, ptr a ret void } -; FUNC-LABEL: {{^}}test_fmin_legacy_olt_f32: +; FUNC-LABEL: {{^}}test_fmin_legacy_ole_f32_fast: ; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]] ; GCN: {{buffer|flat}}_load_dword [[B:v[0-9]+]] -; SI-SAFE: v_min_legacy_f32_e32 {{v[0-9]+}}, [[A]], [[B]] +; GCN: v_min_f32_e32 {{v[0-9]+}}, [[A]], [[B]] +define amdgpu_kernel void @test_fmin_legacy_ole_f32_fast(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { + %tid = call i32 @llvm.amdgcn.workitem.id.x() #1 + %gep.0 = getelementptr float, ptr addrspace(1) %in, i32 %tid + %gep.1 = getelementptr float, ptr addrspace(1) %gep.0, i32 1 -; VI-SAFE: v_cmp_lt_f32_e32 vcc, [[A]], [[B]] -; VI-SAFE: v_cndmask_b32_e32 v{{[0-9]+}}, [[B]], [[A]] + %a = load volatile float, ptr addrspace(1) %gep.0, align 4 + %b = load volatile float, ptr addrspace(1) %gep.1, align 4 + + %cmp = fcmp ole float %a, %b + %val = select nnan nsz i1 %cmp, float %a, float %b + store float %val, ptr addrspace(1) %out, align 4 + ret void +} + +; FUNC-LABEL: {{^}}test_fmin_legacy_olt_f32: +; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]] +; GCN: {{buffer|flat}}_load_dword [[B:v[0-9]+]] + +; SI: v_min_legacy_f32_e32 {{v[0-9]+}}, [[A]], [[B]] -; GCN-NONAN: v_min_f32_e32 {{v[0-9]+}}, [[A]], [[B]] +; VI: v_cmp_lt_f32_e32 vcc, [[A]], [[B]] +; VI: v_cndmask_b32_e32 v{{[0-9]+}}, [[B]], [[A]] define amdgpu_kernel void @test_fmin_legacy_olt_f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { %tid = call i32 @llvm.amdgcn.workitem.id.x() #1 %gep.0 = getelementptr float, ptr addrspace(1) %in, i32 %tid @@ -148,16 +213,33 @@ define amdgpu_kernel void @test_fmin_legacy_olt_f32(ptr addrspace(1) %out, ptr a ret void } -; FUNC-LABEL: {{^}}test_fmin_legacy_ult_f32: +; FUNC-LABEL: {{^}}test_fmin_legacy_olt_f32_fast: ; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]] ; GCN: {{buffer|flat}}_load_dword [[B:v[0-9]+]] -; SI-SAFE: v_min_legacy_f32_e32 {{v[0-9]+}}, [[B]], [[A]] +; GCN: v_min_f32_e32 {{v[0-9]+}}, [[A]], [[B]] +define amdgpu_kernel void @test_fmin_legacy_olt_f32_fast(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { + %tid = call i32 @llvm.amdgcn.workitem.id.x() #1 + %gep.0 = getelementptr float, ptr addrspace(1) %in, i32 %tid + %gep.1 = getelementptr float, ptr addrspace(1) %gep.0, i32 1 -; VI-SAFE: v_cmp_nge_f32_e32 vcc, [[A]], [[B]] -; VI-SAFE: v_cndmask_b32_e32 v{{[0-9]+}}, [[B]], [[A]] + %a = load volatile float, ptr addrspace(1) %gep.0, align 4 + %b = load volatile float, ptr addrspace(1) %gep.1, align 4 -; GCN-NONAN: v_min_f32_e32 {{v[0-9]+}}, [[A]], [[B]] + %cmp = fcmp olt float %a, %b + %val = select nnan nsz i1 %cmp, float %a, float %b + store float %val, ptr addrspace(1) %out, align 4 + ret void +} + +; FUNC-LABEL: {{^}}test_fmin_legacy_ult_f32: +; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]] +; GCN: {{buffer|flat}}_load_dword [[B:v[0-9]+]] + +; SI: v_min_legacy_f32_e32 {{v[0-9]+}}, [[B]], [[A]] + +; VI: v_cmp_nge_f32_e32 vcc, [[A]], [[B]] +; VI: v_cndmask_b32_e32 v{{[0-9]+}}, [[B]], [[A]] define amdgpu_kernel void @test_fmin_legacy_ult_f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { %tid = call i32 @llvm.amdgcn.workitem.id.x() #1 %gep.0 = getelementptr float, ptr addrspace(1) %in, i32 %tid @@ -172,16 +254,33 @@ define amdgpu_kernel void @test_fmin_legacy_ult_f32(ptr addrspace(1) %out, ptr a ret void } -; FUNC-LABEL: {{^}}test_fmin_legacy_ult_v1f32: +; FUNC-LABEL: {{^}}test_fmin_legacy_ult_f32_fast: ; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]] ; GCN: {{buffer|flat}}_load_dword [[B:v[0-9]+]] -; SI-SAFE: v_min_legacy_f32_e32 {{v[0-9]+}}, [[B]], [[A]] +; GCN: v_min_f32_e32 {{v[0-9]+}}, [[A]], [[B]] +define amdgpu_kernel void @test_fmin_legacy_ult_f32_fast(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { + %tid = call i32 @llvm.amdgcn.workitem.id.x() #1 + %gep.0 = getelementptr float, ptr addrspace(1) %in, i32 %tid + %gep.1 = getelementptr float, ptr addrspace(1) %gep.0, i32 1 -; VI-SAFE: v_cmp_nge_f32_e32 vcc, [[A]], [[B]] -; VI-SAFE: v_cndmask_b32_e32 v{{[0-9]+}}, [[B]], [[A]] + %a = load volatile float, ptr addrspace(1) %gep.0, align 4 + %b = load volatile float, ptr addrspace(1) %gep.1, align 4 -; GCN-NONAN: v_min_f32_e32 {{v[0-9]+}}, [[A]], [[B]] + %cmp = fcmp ult float %a, %b + %val = select nnan nsz i1 %cmp, float %a, float %b + store float %val, ptr addrspace(1) %out, align 4 + ret void +} + +; FUNC-LABEL: {{^}}test_fmin_legacy_ult_v1f32: +; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]] +; GCN: {{buffer|flat}}_load_dword [[B:v[0-9]+]] + +; SI: v_min_legacy_f32_e32 {{v[0-9]+}}, [[B]], [[A]] + +; VI: v_cmp_nge_f32_e32 vcc, [[A]], [[B]] +; VI: v_cndmask_b32_e32 v{{[0-9]+}}, [[B]], [[A]] define amdgpu_kernel void @test_fmin_legacy_ult_v1f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { %tid = call i32 @llvm.amdgcn.workitem.id.x() #1 %gep.0 = getelementptr <1 x float>, ptr addrspace(1) %in, i32 %tid @@ -196,19 +295,35 @@ define amdgpu_kernel void @test_fmin_legacy_ult_v1f32(ptr addrspace(1) %out, ptr ret void } +; FUNC-LABEL: {{^}}test_fmin_legacy_ult_v1f32_fast: +; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]] +; GCN: {{buffer|flat}}_load_dword [[B:v[0-9]+]] + +; GCN: v_min_f32_e32 {{v[0-9]+}}, [[A]], [[B]] +define amdgpu_kernel void @test_fmin_legacy_ult_v1f32_fast(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { + %tid = call i32 @llvm.amdgcn.workitem.id.x() #1 + %gep.0 = getelementptr <1 x float>, ptr addrspace(1) %in, i32 %tid + %gep.1 = getelementptr <1 x float>, ptr addrspace(1) %gep.0, i32 1 + + %a = load volatile <1 x float>, ptr addrspace(1) %gep.0 + %b = load volatile <1 x float>, ptr addrspace(1) %gep.1 + + %cmp = fcmp ult <1 x float> %a, %b + %val = select nnan nsz <1 x i1> %cmp, <1 x float> %a, <1 x float> %b + store <1 x float> %val, ptr addrspace(1) %out + ret void +} + ; FUNC-LABEL: {{^}}test_fmin_legacy_ult_v2f32: ; GCN: {{buffer|flat}}_load_dwordx2 ; GCN: {{buffer|flat}}_load_dwordx2 -; SI-SAFE: v_min_legacy_f32_e32 -; SI-SAFE: v_min_legacy_f32_e32 - -; VI-SAFE: v_cmp_nge_f32_e32 -; VI-SAFE: v_cndmask_b32_e32 -; VI-SAFE: v_cmp_nge_f32_e32 -; VI-SAFE: v_cndmask_b32_e32 +; SI: v_min_legacy_f32_e32 +; SI: v_min_legacy_f32_e32 -; GCN-NONAN: v_min_f32_e32 -; GCN-NONAN: v_min_f32_e32 +; VI: v_cmp_nge_f32_e32 +; VI: v_cndmask_b32_e32 +; VI: v_cmp_nge_f32_e32 +; VI: v_cndmask_b32_e32 define amdgpu_kernel void @test_fmin_legacy_ult_v2f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { %tid = call i32 @llvm.amdgcn.workitem.id.x() #1 %gep.0 = getelementptr <2 x float>, ptr addrspace(1) %in, i32 %tid @@ -223,25 +338,40 @@ define amdgpu_kernel void @test_fmin_legacy_ult_v2f32(ptr addrspace(1) %out, ptr ret void } +; FUNC-LABEL: {{^}}test_fmin_legacy_ult_v2f32_fast: +; GCN: {{buffer|flat}}_load_dwordx2 +; GCN: {{buffer|flat}}_load_dwordx2 + +; GCN: v_min_f32_e32 +; GCN: v_min_f32_e32 +define amdgpu_kernel void @test_fmin_legacy_ult_v2f32_fast(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { + %tid = call i32 @llvm.amdgcn.workitem.id.x() #1 + %gep.0 = getelementptr <2 x float>, ptr addrspace(1) %in, i32 %tid + %gep.1 = getelementptr <2 x float>, ptr addrspace(1) %gep.0, i32 1 + + %a = load volatile <2 x float>, ptr addrspace(1) %gep.0 + %b = load volatile <2 x float>, ptr addrspace(1) %gep.1 + + %cmp = fcmp ult <2 x float> %a, %b + %val = select nnan nsz <2 x i1> %cmp, <2 x float> %a, <2 x float> %b + store <2 x float> %val, ptr addrspace(1) %out + ret void +} + ; FUNC-LABEL: {{^}}test_fmin_legacy_ult_v3f32: -; SI-SAFE: v_min_legacy_f32_e32 -; SI-SAFE: v_min_legacy_f32_e32 -; SI-SAFE: v_min_legacy_f32_e32 -; SI-SAFE-NOT: v_min_ - -; VI-SAFE: v_cmp_nge_f32_e32 -; VI-SAFE: v_cndmask_b32_e32 -; VI-SAFE: v_cmp_nge_f32_e32 -; VI-SAFE: v_cndmask_b32_e32 -; VI-SAFE: v_cmp_nge_f32_e32 -; VI-SAFE: v_cndmask_b32_e32 +; SI: v_min_legacy_f32_e32 +; SI: v_min_legacy_f32_e32 +; SI: v_min_legacy_f32_e32 +; SI-NOT: v_min_ + +; VI: v_cmp_nge_f32_e32 +; VI: v_cndmask_b32_e32 +; VI: v_cmp_nge_f32_e32 +; VI: v_cndmask_b32_e32 +; VI: v_cmp_nge_f32_e32 +; VI: v_cndmask_b32_e32 ; VI-NOT: v_cmp ; VI-NOT: v_cndmask - -; GCN-NONAN: v_min_f32_e32 -; GCN-NONAN: v_min_f32_e32 -; GCN-NONAN: v_min_f32_e32 -; GCN-NONAN-NOT: v_min_ define amdgpu_kernel void @test_fmin_legacy_ult_v3f32(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { %tid = call i32 @llvm.amdgcn.workitem.id.x() #1 %gep.0 = getelementptr <3 x float>, ptr addrspace(1) %in, i32 %tid @@ -256,6 +386,28 @@ define amdgpu_kernel void @test_fmin_legacy_ult_v3f32(ptr addrspace(1) %out, ptr ret void } +; FUNC-LABEL: {{^}}test_fmin_legacy_ult_v3f32_fast: +; VI-NOT: v_cmp +; VI-NOT: v_cndmask + +; GCN: v_min_f32_e32 +; GCN: v_min_f32_e32 +; GCN: v_min_f32_e32 +; GCN-NOT: v_min_ +define amdgpu_kernel void @test_fmin_legacy_ult_v3f32_fast(ptr addrspace(1) %out, ptr addrspace(1) %in) #0 { + %tid = call i32 @llvm.amdgcn.workitem.id.x() #1 + %gep.0 = getelementptr <3 x float>, ptr addrspace(1) %in, i32 %tid + %gep.1 = getelementptr <3 x float>, ptr addrspace(1) %gep.0, i32 1 + + %a = load <3 x float>, ptr addrspace(1) %gep.0 + %b = load <3 x float>, ptr addrspace(1) %gep.1 + + %cmp = fcmp ult <3 x float> %a, %b + %val = select nnan nsz <3 x i1> %cmp, <3 x float> %a, <3 x float> %b + store <3 x float> %val, ptr addrspace(1) %out + ret void +} + ; FUNC-LABEL: {{^}}test_fmin_legacy_ole_f32_multi_use: ; GCN: {{buffer|flat}}_load_dword [[A:v[0-9]+]] ; GCN: {{buffer|flat}}_load_dword [[B:v[0-9]+]] diff --git a/llvm/test/CodeGen/LoongArch/lasx/ir-instruction/shuffle-as-xvinsve0.ll b/llvm/test/CodeGen/LoongArch/lasx/ir-instruction/shuffle-as-xvinsve0.ll new file mode 100644 index 0000000..e1784f8 --- /dev/null +++ b/llvm/test/CodeGen/LoongArch/lasx/ir-instruction/shuffle-as-xvinsve0.ll @@ -0,0 +1,197 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6 +; RUN: llc --mtriple=loongarch32 --mattr=+32s,+lasx < %s | FileCheck %s +; RUN: llc --mtriple=loongarch64 --mattr=+lasx < %s | FileCheck %s + +;; xvinsve0.w +define void @xvinsve0_v8i32_l_0(ptr %d, ptr %a, ptr %b) nounwind { +; CHECK-LABEL: xvinsve0_v8i32_l_0: +; CHECK: # %bb.0: # %entry +; CHECK-NEXT: xvld $xr0, $a1, 0 +; CHECK-NEXT: xvld $xr1, $a2, 0 +; CHECK-NEXT: xvinsve0.w $xr0, $xr1, 0 +; CHECK-NEXT: xvst $xr0, $a0, 0 +; CHECK-NEXT: ret +entry: + %va = load <8 x i32>, ptr %a + %vb = load <8 x i32>, ptr %b + %vc = shufflevector <8 x i32> %va, <8 x i32> %vb, <8 x i32> <i32 8, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7> + store <8 x i32> %vc, ptr %d + ret void +} + +define void @xvinsve0_v8i32_l_4(ptr %d, ptr %a, ptr %b) nounwind { +; CHECK-LABEL: xvinsve0_v8i32_l_4: +; CHECK: # %bb.0: # %entry +; CHECK-NEXT: xvld $xr0, $a1, 0 +; CHECK-NEXT: xvld $xr1, $a2, 0 +; CHECK-NEXT: xvinsve0.w $xr0, $xr1, 4 +; CHECK-NEXT: xvst $xr0, $a0, 0 +; CHECK-NEXT: ret +entry: + %va = load <8 x i32>, ptr %a + %vb = load <8 x i32>, ptr %b + %vc = shufflevector <8 x i32> %va, <8 x i32> %vb, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 5, i32 6, i32 7> + store <8 x i32> %vc, ptr %d + ret void +} + +define void @xvinsve0_v8f32_l(ptr %d, ptr %a, ptr %b) nounwind { +; CHECK-LABEL: xvinsve0_v8f32_l: +; CHECK: # %bb.0: # %entry +; CHECK-NEXT: xvld $xr0, $a1, 0 +; CHECK-NEXT: xvld $xr1, $a2, 0 +; CHECK-NEXT: xvinsve0.w $xr0, $xr1, 0 +; CHECK-NEXT: xvst $xr0, $a0, 0 +; CHECK-NEXT: ret +entry: + %va = load <8 x float>, ptr %a + %vb = load <8 x float>, ptr %b + %vc = shufflevector <8 x float> %va, <8 x float> %vb, <8 x i32> <i32 8, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7> + store <8 x float> %vc, ptr %d + ret void +} + +define void @xvinsve0_v8i32_h_1(ptr %d, ptr %a, ptr %b) nounwind { +; CHECK-LABEL: xvinsve0_v8i32_h_1: +; CHECK: # %bb.0: # %entry +; CHECK-NEXT: xvld $xr0, $a1, 0 +; CHECK-NEXT: xvld $xr1, $a2, 0 +; CHECK-NEXT: xvinsve0.w $xr1, $xr0, 1 +; CHECK-NEXT: xvst $xr1, $a0, 0 +; CHECK-NEXT: ret +entry: + %va = load <8 x i32>, ptr %a + %vb = load <8 x i32>, ptr %b + %vc = shufflevector <8 x i32> %va, <8 x i32> %vb, <8 x i32> <i32 8, i32 0, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15> + store <8 x i32> %vc, ptr %d + ret void +} + +define void @xvinsve0_v8i32_h_6(ptr %d, ptr %a, ptr %b) nounwind { +; CHECK-LABEL: xvinsve0_v8i32_h_6: +; CHECK: # %bb.0: # %entry +; CHECK-NEXT: xvld $xr0, $a1, 0 +; CHECK-NEXT: xvld $xr1, $a2, 0 +; CHECK-NEXT: xvinsve0.w $xr1, $xr0, 6 +; CHECK-NEXT: xvst $xr1, $a0, 0 +; CHECK-NEXT: ret +entry: + %va = load <8 x i32>, ptr %a + %vb = load <8 x i32>, ptr %b + %vc = shufflevector <8 x i32> %va, <8 x i32> %vb, <8 x i32> <i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 0, i32 15> + store <8 x i32> %vc, ptr %d + ret void +} + +define void @xvinsve0_v8f32_h(ptr %d, ptr %a, ptr %b) nounwind { +; CHECK-LABEL: xvinsve0_v8f32_h: +; CHECK: # %bb.0: # %entry +; CHECK-NEXT: xvld $xr0, $a1, 0 +; CHECK-NEXT: xvld $xr1, $a2, 0 +; CHECK-NEXT: xvinsve0.w $xr1, $xr0, 0 +; CHECK-NEXT: xvst $xr1, $a0, 0 +; CHECK-NEXT: ret +entry: + %va = load <8 x float>, ptr %a + %vb = load <8 x float>, ptr %b + %vc = shufflevector <8 x float> %va, <8 x float> %vb, <8 x i32> <i32 0, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15> + store <8 x float> %vc, ptr %d + ret void +} + +;; xvinsve0.d +define void @xvinsve0_v4i64_l_1(ptr %d, ptr %a, ptr %b) nounwind { +; CHECK-LABEL: xvinsve0_v4i64_l_1: +; CHECK: # %bb.0: # %entry +; CHECK-NEXT: xvld $xr0, $a1, 0 +; CHECK-NEXT: xvld $xr1, $a2, 0 +; CHECK-NEXT: xvinsve0.d $xr0, $xr1, 1 +; CHECK-NEXT: xvst $xr0, $a0, 0 +; CHECK-NEXT: ret +entry: + %va = load <4 x i64>, ptr %a + %vb = load <4 x i64>, ptr %b + %vc = shufflevector <4 x i64> %va, <4 x i64> %vb, <4 x i32> <i32 0, i32 4, i32 2, i32 3> + store <4 x i64> %vc, ptr %d + ret void +} + +define void @xvinsve0_v4i64_l_2(ptr %d, ptr %a, ptr %b) nounwind { +; CHECK-LABEL: xvinsve0_v4i64_l_2: +; CHECK: # %bb.0: # %entry +; CHECK-NEXT: xvld $xr0, $a1, 0 +; CHECK-NEXT: xvld $xr1, $a2, 0 +; CHECK-NEXT: xvinsve0.d $xr0, $xr1, 2 +; CHECK-NEXT: xvst $xr0, $a0, 0 +; CHECK-NEXT: ret +entry: + %va = load <4 x i64>, ptr %a + %vb = load <4 x i64>, ptr %b + %vc = shufflevector <4 x i64> %va, <4 x i64> %vb, <4 x i32> <i32 0, i32 1, i32 4, i32 3> + store <4 x i64> %vc, ptr %d + ret void +} + +define void @xvinsve0_v4f64_l(ptr %d, ptr %a, ptr %b) nounwind { +; CHECK-LABEL: xvinsve0_v4f64_l: +; CHECK: # %bb.0: # %entry +; CHECK-NEXT: xvld $xr0, $a1, 0 +; CHECK-NEXT: xvld $xr1, $a2, 0 +; CHECK-NEXT: xvinsve0.d $xr0, $xr1, 0 +; CHECK-NEXT: xvst $xr0, $a0, 0 +; CHECK-NEXT: ret +entry: + %va = load <4 x double>, ptr %a + %vb = load <4 x double>, ptr %b + %vc = shufflevector <4 x double> %va, <4 x double> %vb, <4 x i32> <i32 4, i32 1, i32 2, i32 3> + store <4 x double> %vc, ptr %d + ret void +} + +define void @xvinsve0_v4i64_h_0(ptr %d, ptr %a, ptr %b) nounwind { +; CHECK-LABEL: xvinsve0_v4i64_h_0: +; CHECK: # %bb.0: # %entry +; CHECK-NEXT: xvld $xr0, $a1, 0 +; CHECK-NEXT: xvld $xr1, $a2, 0 +; CHECK-NEXT: xvinsve0.d $xr1, $xr0, 0 +; CHECK-NEXT: xvst $xr1, $a0, 0 +; CHECK-NEXT: ret +entry: + %va = load <4 x i64>, ptr %a + %vb = load <4 x i64>, ptr %b + %vc = shufflevector <4 x i64> %va, <4 x i64> %vb, <4 x i32> <i32 0, i32 5, i32 6, i32 7> + store <4 x i64> %vc, ptr %d + ret void +} + +define void @xvinsve0_v4i64_h_2(ptr %d, ptr %a, ptr %b) nounwind { +; CHECK-LABEL: xvinsve0_v4i64_h_2: +; CHECK: # %bb.0: # %entry +; CHECK-NEXT: xvld $xr0, $a1, 0 +; CHECK-NEXT: xvld $xr1, $a2, 0 +; CHECK-NEXT: xvinsve0.d $xr1, $xr0, 2 +; CHECK-NEXT: xvst $xr1, $a0, 0 +; CHECK-NEXT: ret +entry: + %va = load <4 x i64>, ptr %a + %vb = load <4 x i64>, ptr %b + %vc = shufflevector <4 x i64> %va, <4 x i64> %vb, <4 x i32> <i32 4, i32 5, i32 0, i32 7> + store <4 x i64> %vc, ptr %d + ret void +} + +define void @xvinsve0_v4f64_h(ptr %d, ptr %a, ptr %b) nounwind { +; CHECK-LABEL: xvinsve0_v4f64_h: +; CHECK: # %bb.0: # %entry +; CHECK-NEXT: xvld $xr0, $a1, 0 +; CHECK-NEXT: xvld $xr1, $a2, 0 +; CHECK-NEXT: xvinsve0.d $xr1, $xr0, 0 +; CHECK-NEXT: xvst $xr1, $a0, 0 +; CHECK-NEXT: ret +entry: + %va = load <4 x double>, ptr %a + %vb = load <4 x double>, ptr %b + %vc = shufflevector <4 x double> %va, <4 x double> %vb, <4 x i32> <i32 0, i32 5, i32 6, i32 7> + store <4 x double> %vc, ptr %d + ret void +} diff --git a/llvm/test/CodeGen/NVPTX/bug22322.ll b/llvm/test/CodeGen/NVPTX/bug22322.ll index 055c512..71e180b 100644 --- a/llvm/test/CodeGen/NVPTX/bug22322.ll +++ b/llvm/test/CodeGen/NVPTX/bug22322.ll @@ -20,12 +20,12 @@ _ZL11compute_vecRK6float3jb.exit: call void @llvm.lifetime.start.p0(i64 4, ptr %ret_vec.sroa.8.i) %6 = and i32 %4, 15 %7 = icmp eq i32 %6, 0 - %8 = select i1 %7, float 0.000000e+00, float -1.000000e+00 + %8 = select nnan nsz i1 %7, float 0.000000e+00, float -1.000000e+00 store float %8, ptr %ret_vec.sroa.8.i, align 4 ; CHECK: max.f32 %r{{[0-9]+}}, %r{{[0-9]+}}, 0f00000000 %9 = fcmp olt float %8, 0.000000e+00 %ret_vec.sroa.8.i.val = load float, ptr %ret_vec.sroa.8.i, align 4 - %10 = select i1 %9, float 0.000000e+00, float %ret_vec.sroa.8.i.val + %10 = select nnan nsz i1 %9, float 0.000000e+00, float %ret_vec.sroa.8.i.val call void @llvm.lifetime.end.p0(i64 4, ptr %ret_vec.sroa.8.i) %11 = getelementptr inbounds %class.float3, ptr %dst, i64 %5, i32 0 store float 0.000000e+00, ptr %11, align 4 @@ -51,7 +51,7 @@ declare void @llvm.lifetime.start.p0(i64, ptr nocapture) #2 ; Function Attrs: nounwind declare void @llvm.lifetime.end.p0(i64, ptr nocapture) #2 -attributes #0 = { nounwind "less-precise-fpmad"="false" "frame-pointer"="all" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "stack-protector-buffer-size"="8" "no-signed-zeros-fp-math"="true" "use-soft-float"="false" } +attributes #0 = { nounwind "less-precise-fpmad"="false" "frame-pointer"="all" "no-infs-fp-math"="true" "no-nans-fp-math"="true" "stack-protector-buffer-size"="8" "use-soft-float"="false" } attributes #1 = { nounwind readnone } attributes #2 = { nounwind } diff --git a/llvm/test/CodeGen/PowerPC/scalar-min-max.ll b/llvm/test/CodeGen/PowerPC/scalar-min-max.ll index 216d498..5f637e3 100644 --- a/llvm/test/CodeGen/PowerPC/scalar-min-max.ll +++ b/llvm/test/CodeGen/PowerPC/scalar-min-max.ll @@ -1,36 +1,23 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py -; RUN: llc -mcpu=pwr8 -ppc-asm-full-reg-names --enable-unsafe-fp-math \ -; RUN: -verify-machineinstrs --enable-no-signed-zeros-fp-math \ -; RUN: --enable-no-nans-fp-math \ -; RUN: -mtriple=powerpc64le-unknown-unknown < %s | FileCheck %s -; RUN: llc -mcpu=pwr9 -ppc-asm-full-reg-names --enable-unsafe-fp-math \ -; RUN: -verify-machineinstrs --enable-no-signed-zeros-fp-math \ -; RUN: --enable-no-nans-fp-math \ -; RUN: -mtriple=powerpc64le-unknown-unknown < %s | FileCheck %s ; RUN: llc -mcpu=pwr9 -ppc-asm-full-reg-names -verify-machineinstrs \ ; RUN: -mtriple=powerpc64le-unknown-unknown < %s | FileCheck %s \ -; RUN: --check-prefix=NO-FAST-P9 +; RUN: --check-prefix=P9 ; RUN: llc -mcpu=pwr8 -ppc-asm-full-reg-names -verify-machineinstrs \ ; RUN: -mtriple=powerpc64le-unknown-unknown < %s | FileCheck %s \ -; RUN: --check-prefix=NO-FAST-P8 +; RUN: --check-prefix=P8 define dso_local float @testfmax(float %a, float %b) local_unnamed_addr { -; CHECK-LABEL: testfmax: -; CHECK: # %bb.0: # %entry -; CHECK-NEXT: xsmaxdp f1, f1, f2 -; CHECK-NEXT: blr -; -; NO-FAST-P9-LABEL: testfmax: -; NO-FAST-P9: # %bb.0: # %entry -; NO-FAST-P9-NEXT: xsmaxcdp f1, f1, f2 -; NO-FAST-P9-NEXT: blr -; -; NO-FAST-P8-LABEL: testfmax: -; NO-FAST-P8: # %bb.0: # %entry -; NO-FAST-P8-NEXT: fcmpu cr0, f1, f2 -; NO-FAST-P8-NEXT: bgtlr cr0 -; NO-FAST-P8-NEXT: # %bb.1: # %entry -; NO-FAST-P8-NEXT: fmr f1, f2 -; NO-FAST-P8-NEXT: blr +; P9-LABEL: testfmax: +; P9: # %bb.0: # %entry +; P9-NEXT: xsmaxcdp f1, f1, f2 +; P9-NEXT: blr +; +; P8-LABEL: testfmax: +; P8: # %bb.0: # %entry +; P8-NEXT: fcmpu cr0, f1, f2 +; P8-NEXT: bgtlr cr0 +; P8-NEXT: # %bb.1: # %entry +; P8-NEXT: fmr f1, f2 +; P8-NEXT: blr entry: %cmp = fcmp ogt float %a, %b %cond = select i1 %cmp, float %a, float %b @@ -38,23 +25,18 @@ entry: } define dso_local double @testdmax(double %a, double %b) local_unnamed_addr { -; CHECK-LABEL: testdmax: -; CHECK: # %bb.0: # %entry -; CHECK-NEXT: xsmaxdp f1, f1, f2 -; CHECK-NEXT: blr -; -; NO-FAST-P9-LABEL: testdmax: -; NO-FAST-P9: # %bb.0: # %entry -; NO-FAST-P9-NEXT: xsmaxcdp f1, f1, f2 -; NO-FAST-P9-NEXT: blr -; -; NO-FAST-P8-LABEL: testdmax: -; NO-FAST-P8: # %bb.0: # %entry -; NO-FAST-P8-NEXT: xscmpudp cr0, f1, f2 -; NO-FAST-P8-NEXT: bgtlr cr0 -; NO-FAST-P8-NEXT: # %bb.1: # %entry -; NO-FAST-P8-NEXT: fmr f1, f2 -; NO-FAST-P8-NEXT: blr +; P9-LABEL: testdmax: +; P9: # %bb.0: # %entry +; P9-NEXT: xsmaxcdp f1, f1, f2 +; P9-NEXT: blr +; +; P8-LABEL: testdmax: +; P8: # %bb.0: # %entry +; P8-NEXT: xscmpudp cr0, f1, f2 +; P8-NEXT: bgtlr cr0 +; P8-NEXT: # %bb.1: # %entry +; P8-NEXT: fmr f1, f2 +; P8-NEXT: blr entry: %cmp = fcmp ogt double %a, %b %cond = select i1 %cmp, double %a, double %b @@ -62,23 +44,18 @@ entry: } define dso_local float @testfmin(float %a, float %b) local_unnamed_addr { -; CHECK-LABEL: testfmin: -; CHECK: # %bb.0: # %entry -; CHECK-NEXT: xsmindp f1, f1, f2 -; CHECK-NEXT: blr -; -; NO-FAST-P9-LABEL: testfmin: -; NO-FAST-P9: # %bb.0: # %entry -; NO-FAST-P9-NEXT: xsmincdp f1, f1, f2 -; NO-FAST-P9-NEXT: blr -; -; NO-FAST-P8-LABEL: testfmin: -; NO-FAST-P8: # %bb.0: # %entry -; NO-FAST-P8-NEXT: fcmpu cr0, f1, f2 -; NO-FAST-P8-NEXT: bltlr cr0 -; NO-FAST-P8-NEXT: # %bb.1: # %entry -; NO-FAST-P8-NEXT: fmr f1, f2 -; NO-FAST-P8-NEXT: blr +; P9-LABEL: testfmin: +; P9: # %bb.0: # %entry +; P9-NEXT: xsmincdp f1, f1, f2 +; P9-NEXT: blr +; +; P8-LABEL: testfmin: +; P8: # %bb.0: # %entry +; P8-NEXT: fcmpu cr0, f1, f2 +; P8-NEXT: bltlr cr0 +; P8-NEXT: # %bb.1: # %entry +; P8-NEXT: fmr f1, f2 +; P8-NEXT: blr entry: %cmp = fcmp olt float %a, %b %cond = select i1 %cmp, float %a, float %b @@ -86,23 +63,18 @@ entry: } define dso_local double @testdmin(double %a, double %b) local_unnamed_addr { -; CHECK-LABEL: testdmin: -; CHECK: # %bb.0: # %entry -; CHECK-NEXT: xsmindp f1, f1, f2 -; CHECK-NEXT: blr -; -; NO-FAST-P9-LABEL: testdmin: -; NO-FAST-P9: # %bb.0: # %entry -; NO-FAST-P9-NEXT: xsmincdp f1, f1, f2 -; NO-FAST-P9-NEXT: blr -; -; NO-FAST-P8-LABEL: testdmin: -; NO-FAST-P8: # %bb.0: # %entry -; NO-FAST-P8-NEXT: xscmpudp cr0, f1, f2 -; NO-FAST-P8-NEXT: bltlr cr0 -; NO-FAST-P8-NEXT: # %bb.1: # %entry -; NO-FAST-P8-NEXT: fmr f1, f2 -; NO-FAST-P8-NEXT: blr +; P9-LABEL: testdmin: +; P9: # %bb.0: # %entry +; P9-NEXT: xsmincdp f1, f1, f2 +; P9-NEXT: blr +; +; P8-LABEL: testdmin: +; P8: # %bb.0: # %entry +; P8-NEXT: xscmpudp cr0, f1, f2 +; P8-NEXT: bltlr cr0 +; P8-NEXT: # %bb.1: # %entry +; P8-NEXT: fmr f1, f2 +; P8-NEXT: blr entry: %cmp = fcmp olt double %a, %b %cond = select i1 %cmp, double %a, double %b @@ -110,86 +82,62 @@ entry: } define dso_local float @testfmax_fast(float %a, float %b) local_unnamed_addr { -; CHECK-LABEL: testfmax_fast: -; CHECK: # %bb.0: # %entry -; CHECK-NEXT: xsmaxdp f1, f1, f2 -; CHECK-NEXT: blr -; -; NO-FAST-P9-LABEL: testfmax_fast: -; NO-FAST-P9: # %bb.0: # %entry -; NO-FAST-P9-NEXT: xsmaxcdp f1, f1, f2 -; NO-FAST-P9-NEXT: blr -; -; NO-FAST-P8-LABEL: testfmax_fast: -; NO-FAST-P8: # %bb.0: # %entry -; NO-FAST-P8-NEXT: xssubsp f0, f2, f1 -; NO-FAST-P8-NEXT: fsel f1, f0, f2, f1 -; NO-FAST-P8-NEXT: blr +; P9-LABEL: testfmax_fast: +; P9: # %bb.0: # %entry +; P9-NEXT: xsmaxdp f1, f1, f2 +; P9-NEXT: blr +; +; P8-LABEL: testfmax_fast: +; P8: # %bb.0: # %entry +; P8-NEXT: xsmaxdp f1, f1, f2 +; P8-NEXT: blr entry: %cmp = fcmp nnan ninf ogt float %a, %b - %cond = select i1 %cmp, float %a, float %b + %cond = select nnan nsz i1 %cmp, float %a, float %b ret float %cond } define dso_local double @testdmax_fast(double %a, double %b) local_unnamed_addr { -; CHECK-LABEL: testdmax_fast: -; CHECK: # %bb.0: # %entry -; CHECK-NEXT: xsmaxdp f1, f1, f2 -; CHECK-NEXT: blr -; -; NO-FAST-P9-LABEL: testdmax_fast: -; NO-FAST-P9: # %bb.0: # %entry -; NO-FAST-P9-NEXT: xsmaxcdp f1, f1, f2 -; NO-FAST-P9-NEXT: blr -; -; NO-FAST-P8-LABEL: testdmax_fast: -; NO-FAST-P8: # %bb.0: # %entry -; NO-FAST-P8-NEXT: xssubdp f0, f2, f1 -; NO-FAST-P8-NEXT: fsel f1, f0, f2, f1 -; NO-FAST-P8-NEXT: blr +; P9-LABEL: testdmax_fast: +; P9: # %bb.0: # %entry +; P9-NEXT: xsmaxdp f1, f1, f2 +; P9-NEXT: blr +; +; P8-LABEL: testdmax_fast: +; P8: # %bb.0: # %entry +; P8-NEXT: xsmaxdp f1, f1, f2 +; P8-NEXT: blr entry: %cmp = fcmp nnan ninf ogt double %a, %b - %cond = select i1 %cmp, double %a, double %b + %cond = select nnan nsz i1 %cmp, double %a, double %b ret double %cond } define dso_local float @testfmin_fast(float %a, float %b) local_unnamed_addr { -; CHECK-LABEL: testfmin_fast: -; CHECK: # %bb.0: # %entry -; CHECK-NEXT: xsmindp f1, f1, f2 -; CHECK-NEXT: blr -; -; NO-FAST-P9-LABEL: testfmin_fast: -; NO-FAST-P9: # %bb.0: # %entry -; NO-FAST-P9-NEXT: xsmincdp f1, f1, f2 -; NO-FAST-P9-NEXT: blr -; -; NO-FAST-P8-LABEL: testfmin_fast: -; NO-FAST-P8: # %bb.0: # %entry -; NO-FAST-P8-NEXT: xssubsp f0, f1, f2 -; NO-FAST-P8-NEXT: fsel f1, f0, f2, f1 -; NO-FAST-P8-NEXT: blr +; P9-LABEL: testfmin_fast: +; P9: # %bb.0: # %entry +; P9-NEXT: xsmindp f1, f1, f2 +; P9-NEXT: blr +; +; P8-LABEL: testfmin_fast: +; P8: # %bb.0: # %entry +; P8-NEXT: xsmindp f1, f1, f2 +; P8-NEXT: blr entry: %cmp = fcmp nnan ninf olt float %a, %b - %cond = select i1 %cmp, float %a, float %b + %cond = select nnan nsz i1 %cmp, float %a, float %b ret float %cond } define dso_local double @testdmin_fast(double %a, double %b) local_unnamed_addr { -; CHECK-LABEL: testdmin_fast: -; CHECK: # %bb.0: # %entry -; CHECK-NEXT: xsmindp f1, f1, f2 -; CHECK-NEXT: blr -; -; NO-FAST-P9-LABEL: testdmin_fast: -; NO-FAST-P9: # %bb.0: # %entry -; NO-FAST-P9-NEXT: xsmincdp f1, f1, f2 -; NO-FAST-P9-NEXT: blr -; -; NO-FAST-P8-LABEL: testdmin_fast: -; NO-FAST-P8: # %bb.0: # %entry -; NO-FAST-P8-NEXT: xssubdp f0, f1, f2 -; NO-FAST-P8-NEXT: fsel f1, f0, f2, f1 -; NO-FAST-P8-NEXT: blr +; P9-LABEL: testdmin_fast: +; P9: # %bb.0: # %entry +; P9-NEXT: xsmindp f1, f1, f2 +; P9-NEXT: blr +; +; P8-LABEL: testdmin_fast: +; P8: # %bb.0: # %entry +; P8-NEXT: xsmindp f1, f1, f2 +; P8-NEXT: blr entry: %cmp = fcmp nnan ninf olt double %a, %b - %cond = select i1 %cmp, double %a, double %b + %cond = select nnan nsz i1 %cmp, double %a, double %b ret double %cond } diff --git a/llvm/test/CodeGen/RISCV/select-bare.ll b/llvm/test/CodeGen/RISCV/select-bare.ll index 796121a..44028a7 100644 --- a/llvm/test/CodeGen/RISCV/select-bare.ll +++ b/llvm/test/CodeGen/RISCV/select-bare.ll @@ -26,8 +26,8 @@ define i32 @bare_select(i1 %a, i32 %b, i32 %c) nounwind { ; RV32IXQCI-LABEL: bare_select: ; RV32IXQCI: # %bb.0: ; RV32IXQCI-NEXT: andi a0, a0, 1 -; RV32IXQCI-NEXT: qc.mvnei a2, a0, 0, a1 -; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: qc.mveqi a1, a0, 0, a2 +; RV32IXQCI-NEXT: mv a0, a1 ; RV32IXQCI-NEXT: ret %1 = select i1 %a, i32 %b, i32 %c ret i32 %1 @@ -53,8 +53,8 @@ define float @bare_select_float(i1 %a, float %b, float %c) nounwind { ; RV32IXQCI-LABEL: bare_select_float: ; RV32IXQCI: # %bb.0: ; RV32IXQCI-NEXT: andi a0, a0, 1 -; RV32IXQCI-NEXT: qc.mvnei a2, a0, 0, a1 -; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: qc.mveqi a1, a0, 0, a2 +; RV32IXQCI-NEXT: mv a0, a1 ; RV32IXQCI-NEXT: ret %1 = select i1 %a, float %b, float %c ret float %1 diff --git a/llvm/test/CodeGen/RISCV/select-cc.ll b/llvm/test/CodeGen/RISCV/select-cc.ll index 14055df..b57f625 100644 --- a/llvm/test/CodeGen/RISCV/select-cc.ll +++ b/llvm/test/CodeGen/RISCV/select-cc.ll @@ -87,40 +87,40 @@ define signext i32 @foo(i32 signext %a, ptr %b) nounwind { ; ; RV32IXQCI-LABEL: foo: ; RV32IXQCI: # %bb.0: -; RV32IXQCI-NEXT: lw a5, 0(a1) ; RV32IXQCI-NEXT: lw a2, 0(a1) ; RV32IXQCI-NEXT: lw a4, 0(a1) ; RV32IXQCI-NEXT: lw t5, 0(a1) ; RV32IXQCI-NEXT: lw t4, 0(a1) +; RV32IXQCI-NEXT: lw t3, 0(a1) ; RV32IXQCI-NEXT: lw t2, 0(a1) -; RV32IXQCI-NEXT: lw t1, 0(a1) ; RV32IXQCI-NEXT: lw t0, 0(a1) ; RV32IXQCI-NEXT: lw a7, 0(a1) ; RV32IXQCI-NEXT: lw a6, 0(a1) -; RV32IXQCI-NEXT: lw t3, 0(a1) ; RV32IXQCI-NEXT: lw a3, 0(a1) -; RV32IXQCI-NEXT: bltz t3, .LBB0_2 +; RV32IXQCI-NEXT: lw t1, 0(a1) +; RV32IXQCI-NEXT: lw a5, 0(a1) +; RV32IXQCI-NEXT: bltz t1, .LBB0_2 ; RV32IXQCI-NEXT: # %bb.1: -; RV32IXQCI-NEXT: li t6, 0 -; RV32IXQCI-NEXT: qc.mveq a5, a0, a5, a0 -; RV32IXQCI-NEXT: qc.mvne a2, a5, a2, a5 -; RV32IXQCI-NEXT: qc.mvltu a4, a4, a2, a2 -; RV32IXQCI-NEXT: qc.mvgeu t5, a4, t5, a4 -; RV32IXQCI-NEXT: qc.mvltu t4, t5, t4, t5 -; RV32IXQCI-NEXT: qc.mvgeu t2, t2, t4, t4 -; RV32IXQCI-NEXT: qc.mvlt t1, t1, t2, t2 -; RV32IXQCI-NEXT: qc.mvge t0, t1, t0, t1 -; RV32IXQCI-NEXT: qc.mvlt a7, t0, a7, t0 -; RV32IXQCI-NEXT: qc.mvge a6, a6, a7, a7 -; RV32IXQCI-NEXT: mv a3, t3 -; RV32IXQCI-NEXT: qc.mvge a3, t6, t3, a6 +; RV32IXQCI-NEXT: li a5, 0 +; RV32IXQCI-NEXT: qc.mveq a2, a0, a2, a0 +; RV32IXQCI-NEXT: qc.mvne a4, a2, a4, a2 +; RV32IXQCI-NEXT: qc.mvltu t5, t5, a4, a4 +; RV32IXQCI-NEXT: qc.mvgeu t4, t5, t4, t5 +; RV32IXQCI-NEXT: qc.mvltu t3, t4, t3, t4 +; RV32IXQCI-NEXT: qc.mvgeu t2, t2, t3, t3 +; RV32IXQCI-NEXT: qc.mvlt t0, t0, t2, t2 +; RV32IXQCI-NEXT: qc.mvge a7, t0, a7, t0 +; RV32IXQCI-NEXT: qc.mvlt a6, a7, a6, a7 +; RV32IXQCI-NEXT: qc.mvge a3, a3, a6, a6 +; RV32IXQCI-NEXT: qc.mvlt a3, a5, t1, t1 +; RV32IXQCI-NEXT: mv a5, a3 ; RV32IXQCI-NEXT: .LBB0_2: ; RV32IXQCI-NEXT: lw a2, 0(a1) ; RV32IXQCI-NEXT: lw a0, 0(a1) ; RV32IXQCI-NEXT: li a1, 1024 -; RV32IXQCI-NEXT: qc.mvlt a2, a1, a2, a3 +; RV32IXQCI-NEXT: qc.mvlt a2, a1, a2, a5 ; RV32IXQCI-NEXT: li a1, 2046 -; RV32IXQCI-NEXT: qc.mvltu a0, a1, t3, a2 +; RV32IXQCI-NEXT: qc.mvltu a0, a1, t1, a2 ; RV32IXQCI-NEXT: ret ; ; RV64I-LABEL: foo: @@ -417,8 +417,8 @@ define i32 @select_sge_int16min(i32 signext %x, i32 signext %y, i32 signext %z) ; RV32IXQCI: # %bb.0: ; RV32IXQCI-NEXT: lui a3, 1048560 ; RV32IXQCI-NEXT: addi a3, a3, -1 -; RV32IXQCI-NEXT: qc.mvlt a2, a3, a0, a1 -; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: qc.mvge a1, a3, a0, a2 +; RV32IXQCI-NEXT: mv a0, a1 ; RV32IXQCI-NEXT: ret ; ; RV64I-LABEL: select_sge_int16min: @@ -471,10 +471,10 @@ define i64 @select_sge_int32min(i64 %x, i64 %y, i64 %z) { ; RV32IXQCI-NEXT: srli a0, a1, 31 ; RV32IXQCI-NEXT: xori a0, a0, 1 ; RV32IXQCI-NEXT: qc.mveqi a0, a1, -1, a6 -; RV32IXQCI-NEXT: qc.mvnei a4, a0, 0, a2 -; RV32IXQCI-NEXT: qc.mvnei a5, a0, 0, a3 -; RV32IXQCI-NEXT: mv a0, a4 -; RV32IXQCI-NEXT: mv a1, a5 +; RV32IXQCI-NEXT: qc.mveqi a2, a0, 0, a4 +; RV32IXQCI-NEXT: qc.mveqi a3, a0, 0, a5 +; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: mv a1, a3 ; RV32IXQCI-NEXT: ret ; ; RV64I-LABEL: select_sge_int32min: diff --git a/llvm/test/CodeGen/RISCV/select-cond.ll b/llvm/test/CodeGen/RISCV/select-cond.ll index b88fe9a..3ca0f46 100644 --- a/llvm/test/CodeGen/RISCV/select-cond.ll +++ b/llvm/test/CodeGen/RISCV/select-cond.ll @@ -35,8 +35,8 @@ define signext i32 @select_i32_trunc(i32 signext %cond, i32 signext %x, i32 sign ; RV32-XQCICM-LABEL: select_i32_trunc: ; RV32-XQCICM: # %bb.0: ; RV32-XQCICM-NEXT: andi a0, a0, 1 -; RV32-XQCICM-NEXT: qc.mvnei a2, a0, 0, a1 -; RV32-XQCICM-NEXT: mv a0, a2 +; RV32-XQCICM-NEXT: qc.mveqi a1, a0, 0, a2 +; RV32-XQCICM-NEXT: mv a0, a1 ; RV32-XQCICM-NEXT: ret ; ; RV32-XQCICS-LABEL: select_i32_trunc: @@ -48,8 +48,8 @@ define signext i32 @select_i32_trunc(i32 signext %cond, i32 signext %x, i32 sign ; RV32IXQCI-LABEL: select_i32_trunc: ; RV32IXQCI: # %bb.0: ; RV32IXQCI-NEXT: andi a0, a0, 1 -; RV32IXQCI-NEXT: qc.mvnei a2, a0, 0, a1 -; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: qc.mveqi a1, a0, 0, a2 +; RV32IXQCI-NEXT: mv a0, a1 ; RV32IXQCI-NEXT: ret ; ; RV64-LABEL: select_i32_trunc: @@ -93,8 +93,8 @@ define signext i32 @select_i32_param(i1 signext %cond, i32 signext %x, i32 signe ; RV32-XQCICM-LABEL: select_i32_param: ; RV32-XQCICM: # %bb.0: ; RV32-XQCICM-NEXT: andi a0, a0, 1 -; RV32-XQCICM-NEXT: qc.mvnei a2, a0, 0, a1 -; RV32-XQCICM-NEXT: mv a0, a2 +; RV32-XQCICM-NEXT: qc.mveqi a1, a0, 0, a2 +; RV32-XQCICM-NEXT: mv a0, a1 ; RV32-XQCICM-NEXT: ret ; ; RV32-XQCICS-LABEL: select_i32_param: @@ -106,8 +106,8 @@ define signext i32 @select_i32_param(i1 signext %cond, i32 signext %x, i32 signe ; RV32IXQCI-LABEL: select_i32_param: ; RV32IXQCI: # %bb.0: ; RV32IXQCI-NEXT: andi a0, a0, 1 -; RV32IXQCI-NEXT: qc.mvnei a2, a0, 0, a1 -; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: qc.mveqi a1, a0, 0, a2 +; RV32IXQCI-NEXT: mv a0, a1 ; RV32IXQCI-NEXT: ret ; ; RV64-LABEL: select_i32_param: @@ -148,8 +148,8 @@ define signext i32 @select_i32_eq(i32 signext %a, i32 signext %b, i32 signext %x ; ; RV32-XQCICM-LABEL: select_i32_eq: ; RV32-XQCICM: # %bb.0: -; RV32-XQCICM-NEXT: qc.mveq a3, a0, a1, a2 -; RV32-XQCICM-NEXT: mv a0, a3 +; RV32-XQCICM-NEXT: qc.mvne a2, a0, a1, a3 +; RV32-XQCICM-NEXT: mv a0, a2 ; RV32-XQCICM-NEXT: ret ; ; RV32-XQCICS-LABEL: select_i32_eq: @@ -163,8 +163,8 @@ define signext i32 @select_i32_eq(i32 signext %a, i32 signext %b, i32 signext %x ; ; RV32IXQCI-LABEL: select_i32_eq: ; RV32IXQCI: # %bb.0: -; RV32IXQCI-NEXT: qc.mveq a3, a0, a1, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvne a2, a0, a1, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret ; ; RV64-LABEL: select_i32_eq: @@ -205,8 +205,8 @@ define signext i32 @select_i32_ne(i32 signext %a, i32 signext %b, i32 signext %x ; ; RV32-XQCICM-LABEL: select_i32_ne: ; RV32-XQCICM: # %bb.0: -; RV32-XQCICM-NEXT: qc.mvne a3, a0, a1, a2 -; RV32-XQCICM-NEXT: mv a0, a3 +; RV32-XQCICM-NEXT: qc.mveq a2, a0, a1, a3 +; RV32-XQCICM-NEXT: mv a0, a2 ; RV32-XQCICM-NEXT: ret ; ; RV32-XQCICS-LABEL: select_i32_ne: @@ -220,8 +220,8 @@ define signext i32 @select_i32_ne(i32 signext %a, i32 signext %b, i32 signext %x ; ; RV32IXQCI-LABEL: select_i32_ne: ; RV32IXQCI: # %bb.0: -; RV32IXQCI-NEXT: qc.mvne a3, a0, a1, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mveq a2, a0, a1, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret ; ; RV64-LABEL: select_i32_ne: @@ -262,8 +262,8 @@ define signext i32 @select_i32_ugt(i32 signext %a, i32 signext %b, i32 signext % ; ; RV32-XQCICM-LABEL: select_i32_ugt: ; RV32-XQCICM: # %bb.0: -; RV32-XQCICM-NEXT: qc.mvltu a3, a1, a0, a2 -; RV32-XQCICM-NEXT: mv a0, a3 +; RV32-XQCICM-NEXT: qc.mvgeu a2, a1, a0, a3 +; RV32-XQCICM-NEXT: mv a0, a2 ; RV32-XQCICM-NEXT: ret ; ; RV32-XQCICS-LABEL: select_i32_ugt: @@ -277,8 +277,8 @@ define signext i32 @select_i32_ugt(i32 signext %a, i32 signext %b, i32 signext % ; ; RV32IXQCI-LABEL: select_i32_ugt: ; RV32IXQCI: # %bb.0: -; RV32IXQCI-NEXT: qc.mvltu a3, a1, a0, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvgeu a2, a1, a0, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret ; ; RV64-LABEL: select_i32_ugt: @@ -319,8 +319,8 @@ define signext i32 @select_i32_uge(i32 signext %a, i32 signext %b, i32 signext % ; ; RV32-XQCICM-LABEL: select_i32_uge: ; RV32-XQCICM: # %bb.0: -; RV32-XQCICM-NEXT: qc.mvgeu a3, a0, a1, a2 -; RV32-XQCICM-NEXT: mv a0, a3 +; RV32-XQCICM-NEXT: qc.mvltu a2, a0, a1, a3 +; RV32-XQCICM-NEXT: mv a0, a2 ; RV32-XQCICM-NEXT: ret ; ; RV32-XQCICS-LABEL: select_i32_uge: @@ -334,8 +334,8 @@ define signext i32 @select_i32_uge(i32 signext %a, i32 signext %b, i32 signext % ; ; RV32IXQCI-LABEL: select_i32_uge: ; RV32IXQCI: # %bb.0: -; RV32IXQCI-NEXT: qc.mvgeu a3, a0, a1, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvltu a2, a0, a1, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret ; ; RV64-LABEL: select_i32_uge: @@ -376,8 +376,8 @@ define signext i32 @select_i32_ult(i32 signext %a, i32 signext %b, i32 signext % ; ; RV32-XQCICM-LABEL: select_i32_ult: ; RV32-XQCICM: # %bb.0: -; RV32-XQCICM-NEXT: qc.mvltu a3, a0, a1, a2 -; RV32-XQCICM-NEXT: mv a0, a3 +; RV32-XQCICM-NEXT: qc.mvgeu a2, a0, a1, a3 +; RV32-XQCICM-NEXT: mv a0, a2 ; RV32-XQCICM-NEXT: ret ; ; RV32-XQCICS-LABEL: select_i32_ult: @@ -391,8 +391,8 @@ define signext i32 @select_i32_ult(i32 signext %a, i32 signext %b, i32 signext % ; ; RV32IXQCI-LABEL: select_i32_ult: ; RV32IXQCI: # %bb.0: -; RV32IXQCI-NEXT: qc.mvltu a3, a0, a1, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvgeu a2, a0, a1, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret ; ; RV64-LABEL: select_i32_ult: @@ -433,8 +433,8 @@ define signext i32 @select_i32_ule(i32 signext %a, i32 signext %b, i32 signext % ; ; RV32-XQCICM-LABEL: select_i32_ule: ; RV32-XQCICM: # %bb.0: -; RV32-XQCICM-NEXT: qc.mvgeu a3, a1, a0, a2 -; RV32-XQCICM-NEXT: mv a0, a3 +; RV32-XQCICM-NEXT: qc.mvltu a2, a1, a0, a3 +; RV32-XQCICM-NEXT: mv a0, a2 ; RV32-XQCICM-NEXT: ret ; ; RV32-XQCICS-LABEL: select_i32_ule: @@ -448,8 +448,8 @@ define signext i32 @select_i32_ule(i32 signext %a, i32 signext %b, i32 signext % ; ; RV32IXQCI-LABEL: select_i32_ule: ; RV32IXQCI: # %bb.0: -; RV32IXQCI-NEXT: qc.mvgeu a3, a1, a0, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvltu a2, a1, a0, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret ; ; RV64-LABEL: select_i32_ule: @@ -490,8 +490,8 @@ define signext i32 @select_i32_sgt(i32 signext %a, i32 signext %b, i32 signext % ; ; RV32-XQCICM-LABEL: select_i32_sgt: ; RV32-XQCICM: # %bb.0: -; RV32-XQCICM-NEXT: qc.mvlt a3, a1, a0, a2 -; RV32-XQCICM-NEXT: mv a0, a3 +; RV32-XQCICM-NEXT: qc.mvge a2, a1, a0, a3 +; RV32-XQCICM-NEXT: mv a0, a2 ; RV32-XQCICM-NEXT: ret ; ; RV32-XQCICS-LABEL: select_i32_sgt: @@ -505,8 +505,8 @@ define signext i32 @select_i32_sgt(i32 signext %a, i32 signext %b, i32 signext % ; ; RV32IXQCI-LABEL: select_i32_sgt: ; RV32IXQCI: # %bb.0: -; RV32IXQCI-NEXT: qc.mvlt a3, a1, a0, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvge a2, a1, a0, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret ; ; RV64-LABEL: select_i32_sgt: @@ -547,8 +547,8 @@ define signext i32 @select_i32_sge(i32 signext %a, i32 signext %b, i32 signext % ; ; RV32-XQCICM-LABEL: select_i32_sge: ; RV32-XQCICM: # %bb.0: -; RV32-XQCICM-NEXT: qc.mvge a3, a0, a1, a2 -; RV32-XQCICM-NEXT: mv a0, a3 +; RV32-XQCICM-NEXT: qc.mvlt a2, a0, a1, a3 +; RV32-XQCICM-NEXT: mv a0, a2 ; RV32-XQCICM-NEXT: ret ; ; RV32-XQCICS-LABEL: select_i32_sge: @@ -562,8 +562,8 @@ define signext i32 @select_i32_sge(i32 signext %a, i32 signext %b, i32 signext % ; ; RV32IXQCI-LABEL: select_i32_sge: ; RV32IXQCI: # %bb.0: -; RV32IXQCI-NEXT: qc.mvge a3, a0, a1, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvlt a2, a0, a1, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret ; ; RV64-LABEL: select_i32_sge: @@ -604,8 +604,8 @@ define signext i32 @select_i32_slt(i32 signext %a, i32 signext %b, i32 signext % ; ; RV32-XQCICM-LABEL: select_i32_slt: ; RV32-XQCICM: # %bb.0: -; RV32-XQCICM-NEXT: qc.mvlt a3, a0, a1, a2 -; RV32-XQCICM-NEXT: mv a0, a3 +; RV32-XQCICM-NEXT: qc.mvge a2, a0, a1, a3 +; RV32-XQCICM-NEXT: mv a0, a2 ; RV32-XQCICM-NEXT: ret ; ; RV32-XQCICS-LABEL: select_i32_slt: @@ -619,8 +619,8 @@ define signext i32 @select_i32_slt(i32 signext %a, i32 signext %b, i32 signext % ; ; RV32IXQCI-LABEL: select_i32_slt: ; RV32IXQCI: # %bb.0: -; RV32IXQCI-NEXT: qc.mvlt a3, a0, a1, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvge a2, a0, a1, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret ; ; RV64-LABEL: select_i32_slt: @@ -661,8 +661,8 @@ define signext i32 @select_i32_sle(i32 signext %a, i32 signext %b, i32 signext % ; ; RV32-XQCICM-LABEL: select_i32_sle: ; RV32-XQCICM: # %bb.0: -; RV32-XQCICM-NEXT: qc.mvge a3, a1, a0, a2 -; RV32-XQCICM-NEXT: mv a0, a3 +; RV32-XQCICM-NEXT: qc.mvlt a2, a1, a0, a3 +; RV32-XQCICM-NEXT: mv a0, a2 ; RV32-XQCICM-NEXT: ret ; ; RV32-XQCICS-LABEL: select_i32_sle: @@ -676,8 +676,8 @@ define signext i32 @select_i32_sle(i32 signext %a, i32 signext %b, i32 signext % ; ; RV32IXQCI-LABEL: select_i32_sle: ; RV32IXQCI: # %bb.0: -; RV32IXQCI-NEXT: qc.mvge a3, a1, a0, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvlt a2, a1, a0, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret ; ; RV64-LABEL: select_i32_sle: @@ -723,11 +723,11 @@ define i64 @select_i64_trunc(i64 %cond, i64 %x, i64 %y) nounwind { ; ; RV32-XQCICM-LABEL: select_i64_trunc: ; RV32-XQCICM: # %bb.0: -; RV32-XQCICM-NEXT: mv a1, a5 +; RV32-XQCICM-NEXT: mv a1, a3 ; RV32-XQCICM-NEXT: andi a0, a0, 1 -; RV32-XQCICM-NEXT: qc.mvnei a4, a0, 0, a2 -; RV32-XQCICM-NEXT: qc.mvnei a1, a0, 0, a3 -; RV32-XQCICM-NEXT: mv a0, a4 +; RV32-XQCICM-NEXT: qc.mveqi a2, a0, 0, a4 +; RV32-XQCICM-NEXT: qc.mveqi a1, a0, 0, a5 +; RV32-XQCICM-NEXT: mv a0, a2 ; RV32-XQCICM-NEXT: ret ; ; RV32-XQCICS-LABEL: select_i64_trunc: @@ -740,11 +740,11 @@ define i64 @select_i64_trunc(i64 %cond, i64 %x, i64 %y) nounwind { ; ; RV32IXQCI-LABEL: select_i64_trunc: ; RV32IXQCI: # %bb.0: -; RV32IXQCI-NEXT: mv a1, a5 +; RV32IXQCI-NEXT: mv a1, a3 ; RV32IXQCI-NEXT: andi a0, a0, 1 -; RV32IXQCI-NEXT: qc.mvnei a4, a0, 0, a2 -; RV32IXQCI-NEXT: qc.mvnei a1, a0, 0, a3 -; RV32IXQCI-NEXT: mv a0, a4 +; RV32IXQCI-NEXT: qc.mveqi a2, a0, 0, a4 +; RV32IXQCI-NEXT: qc.mveqi a1, a0, 0, a5 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret ; ; RV64-LABEL: select_i64_trunc: @@ -792,10 +792,10 @@ define i64 @select_i64_param(i1 %cond, i64 %x, i64 %y) nounwind { ; RV32-XQCICM-LABEL: select_i64_param: ; RV32-XQCICM: # %bb.0: ; RV32-XQCICM-NEXT: andi a0, a0, 1 -; RV32-XQCICM-NEXT: qc.mvnei a3, a0, 0, a1 -; RV32-XQCICM-NEXT: qc.mvnei a4, a0, 0, a2 -; RV32-XQCICM-NEXT: mv a0, a3 -; RV32-XQCICM-NEXT: mv a1, a4 +; RV32-XQCICM-NEXT: qc.mveqi a1, a0, 0, a3 +; RV32-XQCICM-NEXT: qc.mveqi a2, a0, 0, a4 +; RV32-XQCICM-NEXT: mv a0, a1 +; RV32-XQCICM-NEXT: mv a1, a2 ; RV32-XQCICM-NEXT: ret ; ; RV32-XQCICS-LABEL: select_i64_param: @@ -810,10 +810,10 @@ define i64 @select_i64_param(i1 %cond, i64 %x, i64 %y) nounwind { ; RV32IXQCI-LABEL: select_i64_param: ; RV32IXQCI: # %bb.0: ; RV32IXQCI-NEXT: andi a0, a0, 1 -; RV32IXQCI-NEXT: qc.mvnei a3, a0, 0, a1 -; RV32IXQCI-NEXT: qc.mvnei a4, a0, 0, a2 -; RV32IXQCI-NEXT: mv a0, a3 -; RV32IXQCI-NEXT: mv a1, a4 +; RV32IXQCI-NEXT: qc.mveqi a1, a0, 0, a3 +; RV32IXQCI-NEXT: qc.mveqi a2, a0, 0, a4 +; RV32IXQCI-NEXT: mv a0, a1 +; RV32IXQCI-NEXT: mv a1, a2 ; RV32IXQCI-NEXT: ret ; ; RV64-LABEL: select_i64_param: @@ -866,10 +866,10 @@ define i64 @select_i64_eq(i64 %a, i64 %b, i64 %x, i64 %y) nounwind { ; RV32-XQCICM-NEXT: xor a1, a1, a3 ; RV32-XQCICM-NEXT: xor a0, a0, a2 ; RV32-XQCICM-NEXT: or a0, a0, a1 -; RV32-XQCICM-NEXT: qc.mveqi a6, a0, 0, a4 -; RV32-XQCICM-NEXT: qc.mveqi a7, a0, 0, a5 -; RV32-XQCICM-NEXT: mv a0, a6 -; RV32-XQCICM-NEXT: mv a1, a7 +; RV32-XQCICM-NEXT: qc.mvnei a4, a0, 0, a6 +; RV32-XQCICM-NEXT: qc.mvnei a5, a0, 0, a7 +; RV32-XQCICM-NEXT: mv a0, a4 +; RV32-XQCICM-NEXT: mv a1, a5 ; RV32-XQCICM-NEXT: ret ; ; RV32-XQCICS-LABEL: select_i64_eq: @@ -887,10 +887,10 @@ define i64 @select_i64_eq(i64 %a, i64 %b, i64 %x, i64 %y) nounwind { ; RV32IXQCI-NEXT: xor a1, a1, a3 ; RV32IXQCI-NEXT: xor a0, a0, a2 ; RV32IXQCI-NEXT: or a0, a0, a1 -; RV32IXQCI-NEXT: qc.mveqi a6, a0, 0, a4 -; RV32IXQCI-NEXT: qc.mveqi a7, a0, 0, a5 -; RV32IXQCI-NEXT: mv a0, a6 -; RV32IXQCI-NEXT: mv a1, a7 +; RV32IXQCI-NEXT: qc.mvnei a4, a0, 0, a6 +; RV32IXQCI-NEXT: qc.mvnei a5, a0, 0, a7 +; RV32IXQCI-NEXT: mv a0, a4 +; RV32IXQCI-NEXT: mv a1, a5 ; RV32IXQCI-NEXT: ret ; ; RV64-LABEL: select_i64_eq: @@ -943,10 +943,10 @@ define i64 @select_i64_ne(i64 %a, i64 %b, i64 %x, i64 %y) nounwind { ; RV32-XQCICM-NEXT: xor a1, a1, a3 ; RV32-XQCICM-NEXT: xor a0, a0, a2 ; RV32-XQCICM-NEXT: or a0, a0, a1 -; RV32-XQCICM-NEXT: qc.mvnei a6, a0, 0, a4 -; RV32-XQCICM-NEXT: qc.mvnei a7, a0, 0, a5 -; RV32-XQCICM-NEXT: mv a0, a6 -; RV32-XQCICM-NEXT: mv a1, a7 +; RV32-XQCICM-NEXT: qc.mveqi a4, a0, 0, a6 +; RV32-XQCICM-NEXT: qc.mveqi a5, a0, 0, a7 +; RV32-XQCICM-NEXT: mv a0, a4 +; RV32-XQCICM-NEXT: mv a1, a5 ; RV32-XQCICM-NEXT: ret ; ; RV32-XQCICS-LABEL: select_i64_ne: @@ -964,10 +964,10 @@ define i64 @select_i64_ne(i64 %a, i64 %b, i64 %x, i64 %y) nounwind { ; RV32IXQCI-NEXT: xor a1, a1, a3 ; RV32IXQCI-NEXT: xor a0, a0, a2 ; RV32IXQCI-NEXT: or a0, a0, a1 -; RV32IXQCI-NEXT: qc.mvnei a6, a0, 0, a4 -; RV32IXQCI-NEXT: qc.mvnei a7, a0, 0, a5 -; RV32IXQCI-NEXT: mv a0, a6 -; RV32IXQCI-NEXT: mv a1, a7 +; RV32IXQCI-NEXT: qc.mveqi a4, a0, 0, a6 +; RV32IXQCI-NEXT: qc.mveqi a5, a0, 0, a7 +; RV32IXQCI-NEXT: mv a0, a4 +; RV32IXQCI-NEXT: mv a1, a5 ; RV32IXQCI-NEXT: ret ; ; RV64-LABEL: select_i64_ne: @@ -1025,10 +1025,10 @@ define i64 @select_i64_ugt(i64 %a, i64 %b, i64 %x, i64 %y) nounwind { ; RV32-XQCICM-NEXT: sltu a0, a2, a0 ; RV32-XQCICM-NEXT: sltu a2, a3, a1 ; RV32-XQCICM-NEXT: qc.mveq a2, a1, a3, a0 -; RV32-XQCICM-NEXT: qc.mvnei a6, a2, 0, a4 -; RV32-XQCICM-NEXT: qc.mvnei a7, a2, 0, a5 -; RV32-XQCICM-NEXT: mv a0, a6 -; RV32-XQCICM-NEXT: mv a1, a7 +; RV32-XQCICM-NEXT: qc.mveqi a4, a2, 0, a6 +; RV32-XQCICM-NEXT: qc.mveqi a5, a2, 0, a7 +; RV32-XQCICM-NEXT: mv a0, a4 +; RV32-XQCICM-NEXT: mv a1, a5 ; RV32-XQCICM-NEXT: ret ; ; RV32-XQCICS-LABEL: select_i64_ugt: @@ -1050,10 +1050,10 @@ define i64 @select_i64_ugt(i64 %a, i64 %b, i64 %x, i64 %y) nounwind { ; RV32IXQCI-NEXT: sltu a0, a2, a0 ; RV32IXQCI-NEXT: sltu a2, a3, a1 ; RV32IXQCI-NEXT: qc.mveq a2, a1, a3, a0 -; RV32IXQCI-NEXT: qc.mvnei a6, a2, 0, a4 -; RV32IXQCI-NEXT: qc.mvnei a7, a2, 0, a5 -; RV32IXQCI-NEXT: mv a0, a6 -; RV32IXQCI-NEXT: mv a1, a7 +; RV32IXQCI-NEXT: qc.mveqi a4, a2, 0, a6 +; RV32IXQCI-NEXT: qc.mveqi a5, a2, 0, a7 +; RV32IXQCI-NEXT: mv a0, a4 +; RV32IXQCI-NEXT: mv a1, a5 ; RV32IXQCI-NEXT: ret ; ; RV64-LABEL: select_i64_ugt: @@ -1111,10 +1111,10 @@ define i64 @select_i64_uge(i64 %a, i64 %b, i64 %x, i64 %y) nounwind { ; RV32-XQCICM-NEXT: sltu a0, a0, a2 ; RV32-XQCICM-NEXT: sltu a2, a1, a3 ; RV32-XQCICM-NEXT: qc.mveq a2, a1, a3, a0 -; RV32-XQCICM-NEXT: qc.mveqi a6, a2, 0, a4 -; RV32-XQCICM-NEXT: qc.mveqi a7, a2, 0, a5 -; RV32-XQCICM-NEXT: mv a0, a6 -; RV32-XQCICM-NEXT: mv a1, a7 +; RV32-XQCICM-NEXT: qc.mvnei a4, a2, 0, a6 +; RV32-XQCICM-NEXT: qc.mvnei a5, a2, 0, a7 +; RV32-XQCICM-NEXT: mv a0, a4 +; RV32-XQCICM-NEXT: mv a1, a5 ; RV32-XQCICM-NEXT: ret ; ; RV32-XQCICS-LABEL: select_i64_uge: @@ -1136,10 +1136,10 @@ define i64 @select_i64_uge(i64 %a, i64 %b, i64 %x, i64 %y) nounwind { ; RV32IXQCI-NEXT: sltu a0, a0, a2 ; RV32IXQCI-NEXT: sltu a2, a1, a3 ; RV32IXQCI-NEXT: qc.mveq a2, a1, a3, a0 -; RV32IXQCI-NEXT: qc.mveqi a6, a2, 0, a4 -; RV32IXQCI-NEXT: qc.mveqi a7, a2, 0, a5 -; RV32IXQCI-NEXT: mv a0, a6 -; RV32IXQCI-NEXT: mv a1, a7 +; RV32IXQCI-NEXT: qc.mvnei a4, a2, 0, a6 +; RV32IXQCI-NEXT: qc.mvnei a5, a2, 0, a7 +; RV32IXQCI-NEXT: mv a0, a4 +; RV32IXQCI-NEXT: mv a1, a5 ; RV32IXQCI-NEXT: ret ; ; RV64-LABEL: select_i64_uge: @@ -1197,10 +1197,10 @@ define i64 @select_i64_ult(i64 %a, i64 %b, i64 %x, i64 %y) nounwind { ; RV32-XQCICM-NEXT: sltu a0, a0, a2 ; RV32-XQCICM-NEXT: sltu a2, a1, a3 ; RV32-XQCICM-NEXT: qc.mveq a2, a1, a3, a0 -; RV32-XQCICM-NEXT: qc.mvnei a6, a2, 0, a4 -; RV32-XQCICM-NEXT: qc.mvnei a7, a2, 0, a5 -; RV32-XQCICM-NEXT: mv a0, a6 -; RV32-XQCICM-NEXT: mv a1, a7 +; RV32-XQCICM-NEXT: qc.mveqi a4, a2, 0, a6 +; RV32-XQCICM-NEXT: qc.mveqi a5, a2, 0, a7 +; RV32-XQCICM-NEXT: mv a0, a4 +; RV32-XQCICM-NEXT: mv a1, a5 ; RV32-XQCICM-NEXT: ret ; ; RV32-XQCICS-LABEL: select_i64_ult: @@ -1222,10 +1222,10 @@ define i64 @select_i64_ult(i64 %a, i64 %b, i64 %x, i64 %y) nounwind { ; RV32IXQCI-NEXT: sltu a0, a0, a2 ; RV32IXQCI-NEXT: sltu a2, a1, a3 ; RV32IXQCI-NEXT: qc.mveq a2, a1, a3, a0 -; RV32IXQCI-NEXT: qc.mvnei a6, a2, 0, a4 -; RV32IXQCI-NEXT: qc.mvnei a7, a2, 0, a5 -; RV32IXQCI-NEXT: mv a0, a6 -; RV32IXQCI-NEXT: mv a1, a7 +; RV32IXQCI-NEXT: qc.mveqi a4, a2, 0, a6 +; RV32IXQCI-NEXT: qc.mveqi a5, a2, 0, a7 +; RV32IXQCI-NEXT: mv a0, a4 +; RV32IXQCI-NEXT: mv a1, a5 ; RV32IXQCI-NEXT: ret ; ; RV64-LABEL: select_i64_ult: @@ -1283,10 +1283,10 @@ define i64 @select_i64_ule(i64 %a, i64 %b, i64 %x, i64 %y) nounwind { ; RV32-XQCICM-NEXT: sltu a0, a2, a0 ; RV32-XQCICM-NEXT: sltu a2, a3, a1 ; RV32-XQCICM-NEXT: qc.mveq a2, a1, a3, a0 -; RV32-XQCICM-NEXT: qc.mveqi a6, a2, 0, a4 -; RV32-XQCICM-NEXT: qc.mveqi a7, a2, 0, a5 -; RV32-XQCICM-NEXT: mv a0, a6 -; RV32-XQCICM-NEXT: mv a1, a7 +; RV32-XQCICM-NEXT: qc.mvnei a4, a2, 0, a6 +; RV32-XQCICM-NEXT: qc.mvnei a5, a2, 0, a7 +; RV32-XQCICM-NEXT: mv a0, a4 +; RV32-XQCICM-NEXT: mv a1, a5 ; RV32-XQCICM-NEXT: ret ; ; RV32-XQCICS-LABEL: select_i64_ule: @@ -1308,10 +1308,10 @@ define i64 @select_i64_ule(i64 %a, i64 %b, i64 %x, i64 %y) nounwind { ; RV32IXQCI-NEXT: sltu a0, a2, a0 ; RV32IXQCI-NEXT: sltu a2, a3, a1 ; RV32IXQCI-NEXT: qc.mveq a2, a1, a3, a0 -; RV32IXQCI-NEXT: qc.mveqi a6, a2, 0, a4 -; RV32IXQCI-NEXT: qc.mveqi a7, a2, 0, a5 -; RV32IXQCI-NEXT: mv a0, a6 -; RV32IXQCI-NEXT: mv a1, a7 +; RV32IXQCI-NEXT: qc.mvnei a4, a2, 0, a6 +; RV32IXQCI-NEXT: qc.mvnei a5, a2, 0, a7 +; RV32IXQCI-NEXT: mv a0, a4 +; RV32IXQCI-NEXT: mv a1, a5 ; RV32IXQCI-NEXT: ret ; ; RV64-LABEL: select_i64_ule: @@ -1369,10 +1369,10 @@ define i64 @select_i64_sgt(i64 %a, i64 %b, i64 %x, i64 %y) nounwind { ; RV32-XQCICM-NEXT: sltu a0, a2, a0 ; RV32-XQCICM-NEXT: slt a2, a3, a1 ; RV32-XQCICM-NEXT: qc.mveq a2, a1, a3, a0 -; RV32-XQCICM-NEXT: qc.mvnei a6, a2, 0, a4 -; RV32-XQCICM-NEXT: qc.mvnei a7, a2, 0, a5 -; RV32-XQCICM-NEXT: mv a0, a6 -; RV32-XQCICM-NEXT: mv a1, a7 +; RV32-XQCICM-NEXT: qc.mveqi a4, a2, 0, a6 +; RV32-XQCICM-NEXT: qc.mveqi a5, a2, 0, a7 +; RV32-XQCICM-NEXT: mv a0, a4 +; RV32-XQCICM-NEXT: mv a1, a5 ; RV32-XQCICM-NEXT: ret ; ; RV32-XQCICS-LABEL: select_i64_sgt: @@ -1394,10 +1394,10 @@ define i64 @select_i64_sgt(i64 %a, i64 %b, i64 %x, i64 %y) nounwind { ; RV32IXQCI-NEXT: sltu a0, a2, a0 ; RV32IXQCI-NEXT: slt a2, a3, a1 ; RV32IXQCI-NEXT: qc.mveq a2, a1, a3, a0 -; RV32IXQCI-NEXT: qc.mvnei a6, a2, 0, a4 -; RV32IXQCI-NEXT: qc.mvnei a7, a2, 0, a5 -; RV32IXQCI-NEXT: mv a0, a6 -; RV32IXQCI-NEXT: mv a1, a7 +; RV32IXQCI-NEXT: qc.mveqi a4, a2, 0, a6 +; RV32IXQCI-NEXT: qc.mveqi a5, a2, 0, a7 +; RV32IXQCI-NEXT: mv a0, a4 +; RV32IXQCI-NEXT: mv a1, a5 ; RV32IXQCI-NEXT: ret ; ; RV64-LABEL: select_i64_sgt: @@ -1455,10 +1455,10 @@ define i64 @select_i64_sge(i64 %a, i64 %b, i64 %x, i64 %y) nounwind { ; RV32-XQCICM-NEXT: sltu a0, a0, a2 ; RV32-XQCICM-NEXT: slt a2, a1, a3 ; RV32-XQCICM-NEXT: qc.mveq a2, a1, a3, a0 -; RV32-XQCICM-NEXT: qc.mveqi a6, a2, 0, a4 -; RV32-XQCICM-NEXT: qc.mveqi a7, a2, 0, a5 -; RV32-XQCICM-NEXT: mv a0, a6 -; RV32-XQCICM-NEXT: mv a1, a7 +; RV32-XQCICM-NEXT: qc.mvnei a4, a2, 0, a6 +; RV32-XQCICM-NEXT: qc.mvnei a5, a2, 0, a7 +; RV32-XQCICM-NEXT: mv a0, a4 +; RV32-XQCICM-NEXT: mv a1, a5 ; RV32-XQCICM-NEXT: ret ; ; RV32-XQCICS-LABEL: select_i64_sge: @@ -1480,10 +1480,10 @@ define i64 @select_i64_sge(i64 %a, i64 %b, i64 %x, i64 %y) nounwind { ; RV32IXQCI-NEXT: sltu a0, a0, a2 ; RV32IXQCI-NEXT: slt a2, a1, a3 ; RV32IXQCI-NEXT: qc.mveq a2, a1, a3, a0 -; RV32IXQCI-NEXT: qc.mveqi a6, a2, 0, a4 -; RV32IXQCI-NEXT: qc.mveqi a7, a2, 0, a5 -; RV32IXQCI-NEXT: mv a0, a6 -; RV32IXQCI-NEXT: mv a1, a7 +; RV32IXQCI-NEXT: qc.mvnei a4, a2, 0, a6 +; RV32IXQCI-NEXT: qc.mvnei a5, a2, 0, a7 +; RV32IXQCI-NEXT: mv a0, a4 +; RV32IXQCI-NEXT: mv a1, a5 ; RV32IXQCI-NEXT: ret ; ; RV64-LABEL: select_i64_sge: @@ -1541,10 +1541,10 @@ define i64 @select_i64_slt(i64 %a, i64 %b, i64 %x, i64 %y) nounwind { ; RV32-XQCICM-NEXT: sltu a0, a0, a2 ; RV32-XQCICM-NEXT: slt a2, a1, a3 ; RV32-XQCICM-NEXT: qc.mveq a2, a1, a3, a0 -; RV32-XQCICM-NEXT: qc.mvnei a6, a2, 0, a4 -; RV32-XQCICM-NEXT: qc.mvnei a7, a2, 0, a5 -; RV32-XQCICM-NEXT: mv a0, a6 -; RV32-XQCICM-NEXT: mv a1, a7 +; RV32-XQCICM-NEXT: qc.mveqi a4, a2, 0, a6 +; RV32-XQCICM-NEXT: qc.mveqi a5, a2, 0, a7 +; RV32-XQCICM-NEXT: mv a0, a4 +; RV32-XQCICM-NEXT: mv a1, a5 ; RV32-XQCICM-NEXT: ret ; ; RV32-XQCICS-LABEL: select_i64_slt: @@ -1566,10 +1566,10 @@ define i64 @select_i64_slt(i64 %a, i64 %b, i64 %x, i64 %y) nounwind { ; RV32IXQCI-NEXT: sltu a0, a0, a2 ; RV32IXQCI-NEXT: slt a2, a1, a3 ; RV32IXQCI-NEXT: qc.mveq a2, a1, a3, a0 -; RV32IXQCI-NEXT: qc.mvnei a6, a2, 0, a4 -; RV32IXQCI-NEXT: qc.mvnei a7, a2, 0, a5 -; RV32IXQCI-NEXT: mv a0, a6 -; RV32IXQCI-NEXT: mv a1, a7 +; RV32IXQCI-NEXT: qc.mveqi a4, a2, 0, a6 +; RV32IXQCI-NEXT: qc.mveqi a5, a2, 0, a7 +; RV32IXQCI-NEXT: mv a0, a4 +; RV32IXQCI-NEXT: mv a1, a5 ; RV32IXQCI-NEXT: ret ; ; RV64-LABEL: select_i64_slt: @@ -1627,10 +1627,10 @@ define i64 @select_i64_sle(i64 %a, i64 %b, i64 %x, i64 %y) nounwind { ; RV32-XQCICM-NEXT: sltu a0, a2, a0 ; RV32-XQCICM-NEXT: slt a2, a3, a1 ; RV32-XQCICM-NEXT: qc.mveq a2, a1, a3, a0 -; RV32-XQCICM-NEXT: qc.mveqi a6, a2, 0, a4 -; RV32-XQCICM-NEXT: qc.mveqi a7, a2, 0, a5 -; RV32-XQCICM-NEXT: mv a0, a6 -; RV32-XQCICM-NEXT: mv a1, a7 +; RV32-XQCICM-NEXT: qc.mvnei a4, a2, 0, a6 +; RV32-XQCICM-NEXT: qc.mvnei a5, a2, 0, a7 +; RV32-XQCICM-NEXT: mv a0, a4 +; RV32-XQCICM-NEXT: mv a1, a5 ; RV32-XQCICM-NEXT: ret ; ; RV32-XQCICS-LABEL: select_i64_sle: @@ -1652,10 +1652,10 @@ define i64 @select_i64_sle(i64 %a, i64 %b, i64 %x, i64 %y) nounwind { ; RV32IXQCI-NEXT: sltu a0, a2, a0 ; RV32IXQCI-NEXT: slt a2, a3, a1 ; RV32IXQCI-NEXT: qc.mveq a2, a1, a3, a0 -; RV32IXQCI-NEXT: qc.mveqi a6, a2, 0, a4 -; RV32IXQCI-NEXT: qc.mveqi a7, a2, 0, a5 -; RV32IXQCI-NEXT: mv a0, a6 -; RV32IXQCI-NEXT: mv a1, a7 +; RV32IXQCI-NEXT: qc.mvnei a4, a2, 0, a6 +; RV32IXQCI-NEXT: qc.mvnei a5, a2, 0, a7 +; RV32IXQCI-NEXT: mv a0, a4 +; RV32IXQCI-NEXT: mv a1, a5 ; RV32IXQCI-NEXT: ret ; ; RV64-LABEL: select_i64_sle: diff --git a/llvm/test/CodeGen/RISCV/select.ll b/llvm/test/CodeGen/RISCV/select.ll index 19fade6..8273c65 100644 --- a/llvm/test/CodeGen/RISCV/select.ll +++ b/llvm/test/CodeGen/RISCV/select.ll @@ -1153,8 +1153,8 @@ define i32 @select_sub_1(i1 zeroext %cond, i32 %a, i32 %b) { ; RV32IXQCI-LABEL: select_sub_1: ; RV32IXQCI: # %bb.0: # %entry ; RV32IXQCI-NEXT: sub a1, a1, a2 -; RV32IXQCI-NEXT: qc.mvnei a2, a0, 0, a1 -; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: qc.mveqi a1, a0, 0, a2 +; RV32IXQCI-NEXT: mv a0, a1 ; RV32IXQCI-NEXT: ret entry: %c = sub i32 %a, %b @@ -1301,9 +1301,9 @@ define i32 @select_sub_4(i1 zeroext %cond, i32 %x) { ; ; RV32IXQCI-LABEL: select_sub_4: ; RV32IXQCI: # %bb.0: -; RV32IXQCI-NEXT: addi a1, a1, -128 -; RV32IXQCI-NEXT: li a2, 128 -; RV32IXQCI-NEXT: qc.mvnei a1, a0, 0, a2 +; RV32IXQCI-NEXT: addi a2, a1, -128 +; RV32IXQCI-NEXT: li a1, 128 +; RV32IXQCI-NEXT: qc.mveqi a1, a0, 0, a2 ; RV32IXQCI-NEXT: mv a0, a1 ; RV32IXQCI-NEXT: ret %add = sub i32 %x, 128 @@ -1348,8 +1348,8 @@ define i32 @select_and_1(i1 zeroext %cond, i32 %a, i32 %b) { ; RV32IXQCI-LABEL: select_and_1: ; RV32IXQCI: # %bb.0: # %entry ; RV32IXQCI-NEXT: and a1, a1, a2 -; RV32IXQCI-NEXT: qc.mvnei a2, a0, 0, a1 -; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: qc.mveqi a1, a0, 0, a2 +; RV32IXQCI-NEXT: mv a0, a1 ; RV32IXQCI-NEXT: ret entry: %c = and i32 %a, %b @@ -1493,8 +1493,8 @@ define i32 @select_udiv_1(i1 zeroext %cond, i32 %a, i32 %b) { ; RV32IXQCI-LABEL: select_udiv_1: ; RV32IXQCI: # %bb.0: # %entry ; RV32IXQCI-NEXT: divu a1, a1, a2 -; RV32IXQCI-NEXT: qc.mvnei a2, a0, 0, a1 -; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: qc.mveqi a1, a0, 0, a2 +; RV32IXQCI-NEXT: mv a0, a1 ; RV32IXQCI-NEXT: ret entry: %c = udiv i32 %a, %b @@ -1682,8 +1682,8 @@ define i32 @select_shl_1(i1 zeroext %cond, i32 %a, i32 %b) { ; RV32IXQCI-LABEL: select_shl_1: ; RV32IXQCI: # %bb.0: # %entry ; RV32IXQCI-NEXT: sll a1, a1, a2 -; RV32IXQCI-NEXT: qc.mvnei a2, a0, 0, a1 -; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: qc.mveqi a1, a0, 0, a2 +; RV32IXQCI-NEXT: mv a0, a1 ; RV32IXQCI-NEXT: ret entry: %c = shl i32 %a, %b @@ -1798,8 +1798,8 @@ define i32 @select_ashr_1(i1 zeroext %cond, i32 %a, i32 %b) { ; RV32IXQCI-LABEL: select_ashr_1: ; RV32IXQCI: # %bb.0: # %entry ; RV32IXQCI-NEXT: sra a1, a1, a2 -; RV32IXQCI-NEXT: qc.mvnei a2, a0, 0, a1 -; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: qc.mveqi a1, a0, 0, a2 +; RV32IXQCI-NEXT: mv a0, a1 ; RV32IXQCI-NEXT: ret entry: %c = ashr i32 %a, %b @@ -1914,8 +1914,8 @@ define i32 @select_lshr_1(i1 zeroext %cond, i32 %a, i32 %b) { ; RV32IXQCI-LABEL: select_lshr_1: ; RV32IXQCI: # %bb.0: # %entry ; RV32IXQCI-NEXT: srl a1, a1, a2 -; RV32IXQCI-NEXT: qc.mvnei a2, a0, 0, a1 -; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: qc.mveqi a1, a0, 0, a2 +; RV32IXQCI-NEXT: mv a0, a1 ; RV32IXQCI-NEXT: ret entry: %c = lshr i32 %a, %b @@ -2371,9 +2371,9 @@ define i32 @select_cst5(i1 zeroext %cond) { ; RV32IXQCI-LABEL: select_cst5: ; RV32IXQCI: # %bb.0: ; RV32IXQCI-NEXT: lui a1, 1 -; RV32IXQCI-NEXT: addi a1, a1, -2047 -; RV32IXQCI-NEXT: li a2, 2047 -; RV32IXQCI-NEXT: qc.mvnei a1, a0, 0, a2 +; RV32IXQCI-NEXT: addi a2, a1, -2047 +; RV32IXQCI-NEXT: li a1, 2047 +; RV32IXQCI-NEXT: qc.mveqi a1, a0, 0, a2 ; RV32IXQCI-NEXT: mv a0, a1 ; RV32IXQCI-NEXT: ret %ret = select i1 %cond, i32 2047, i32 2049 @@ -2870,8 +2870,8 @@ define void @select_redundant_czero_eqz1(ptr %0, ptr %1) { ; RV32IXQCI: # %bb.0: # %entry ; RV32IXQCI-NEXT: lui a2, %hi(select_redundant_czero_eqz_data) ; RV32IXQCI-NEXT: addi a2, a2, %lo(select_redundant_czero_eqz_data) -; RV32IXQCI-NEXT: qc.mveqi a0, a0, 0, a2 -; RV32IXQCI-NEXT: sw a0, 0(a1) +; RV32IXQCI-NEXT: qc.mvnei a2, a0, 0, a0 +; RV32IXQCI-NEXT: sw a2, 0(a1) ; RV32IXQCI-NEXT: ret entry: %3 = icmp eq ptr %0, null diff --git a/llvm/test/CodeGen/RISCV/xqcicm.ll b/llvm/test/CodeGen/RISCV/xqcicm.ll index 1741be7..fb48301 100644 --- a/llvm/test/CodeGen/RISCV/xqcicm.ll +++ b/llvm/test/CodeGen/RISCV/xqcicm.ll @@ -23,15 +23,15 @@ define i32 @select_example(i32 %cond, i32 %x, i32 %y) { ; RV32IXQCICM-LABEL: select_example: ; RV32IXQCICM: # %bb.0: # %entry ; RV32IXQCICM-NEXT: andi a0, a0, 1 -; RV32IXQCICM-NEXT: qc.mvnei a2, a0, 0, a1 -; RV32IXQCICM-NEXT: mv a0, a2 +; RV32IXQCICM-NEXT: qc.mveqi a1, a0, 0, a2 +; RV32IXQCICM-NEXT: mv a0, a1 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_example: ; RV32IXQCI: # %bb.0: # %entry ; RV32IXQCI-NEXT: andi a0, a0, 1 -; RV32IXQCI-NEXT: qc.mvnei a2, a0, 0, a1 -; RV32IXQCI-NEXT: mv a0, a2 +; RV32IXQCI-NEXT: qc.mveqi a1, a0, 0, a2 +; RV32IXQCI-NEXT: mv a0, a1 ; RV32IXQCI-NEXT: ret entry: %cond_trunc = trunc i32 %cond to i1 @@ -52,14 +52,14 @@ define i32 @select_cc_example_eq(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_eq: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mveqi a3, a0, 11, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvnei a2, a0, 11, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_eq: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mveqi a3, a0, 11, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvnei a2, a0, 11, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp eq i32 %a, 11 @@ -80,14 +80,14 @@ define i32 @select_cc_example_eq1(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_eq1: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mveqi a3, a0, 11, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvnei a2, a0, 11, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_eq1: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mveqi a3, a0, 11, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvnei a2, a0, 11, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp eq i32 11, %a @@ -108,14 +108,14 @@ define i32 @select_cc_example_ne(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_ne: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvnei a3, a0, 11, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mveqi a2, a0, 11, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_ne: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvnei a3, a0, 11, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mveqi a2, a0, 11, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ne i32 %a, 11 @@ -136,14 +136,14 @@ define i32 @select_cc_example_ne1(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_ne1: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvnei a3, a0, 11, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mveqi a2, a0, 11, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_ne1: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvnei a3, a0, 11, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mveqi a2, a0, 11, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ne i32 11, %a @@ -164,14 +164,14 @@ define i32 @select_cc_example_slt(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_slt: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvlti a3, a0, 11, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvgei a2, a0, 11, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_slt: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvlti a3, a0, 11, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvgei a2, a0, 11, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp slt i32 %a, 11 @@ -192,14 +192,14 @@ define i32 @select_cc_example_slt1(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_slt1: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvgei a3, a0, 12, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvlti a2, a0, 12, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_slt1: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvgei a3, a0, 12, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvlti a2, a0, 12, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp slt i32 11, %a @@ -220,14 +220,14 @@ define i32 @select_cc_example_sle(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_sle: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvlti a3, a0, 12, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvgei a2, a0, 12, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_sle: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvlti a3, a0, 12, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvgei a2, a0, 12, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp sle i32 %a, 11 @@ -248,14 +248,14 @@ define i32 @select_cc_example_sle1(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_sle1: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvgei a3, a0, 11, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvlti a2, a0, 11, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_sle1: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvgei a3, a0, 11, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvlti a2, a0, 11, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp sle i32 11, %a @@ -276,14 +276,14 @@ define i32 @select_cc_example_sgt(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_sgt: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvgei a3, a0, 12, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvlti a2, a0, 12, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_sgt: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvgei a3, a0, 12, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvlti a2, a0, 12, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp sgt i32 %a, 11 @@ -304,14 +304,14 @@ define i32 @select_cc_example_sgt1(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_sgt1: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvlti a3, a0, 11, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvgei a2, a0, 11, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_sgt1: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvlti a3, a0, 11, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvgei a2, a0, 11, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp sgt i32 11, %a @@ -332,14 +332,14 @@ define i32 @select_cc_example_sge(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_sge: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvgei a3, a0, 11, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvlti a2, a0, 11, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_sge: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvgei a3, a0, 11, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvlti a2, a0, 11, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp sge i32 %a, 11 @@ -360,14 +360,14 @@ define i32 @select_cc_example_sge1(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_sge1: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvlti a3, a0, 12, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvgei a2, a0, 12, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_sge1: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvlti a3, a0, 12, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvgei a2, a0, 12, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp sge i32 11, %a @@ -388,14 +388,14 @@ define i32 @select_cc_example_ule(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_ule: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvltui a3, a0, 12, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvgeui a2, a0, 12, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_ule: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvltui a3, a0, 12, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvgeui a2, a0, 12, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ule i32 %a, 11 @@ -416,14 +416,14 @@ define i32 @select_cc_example_ule1(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_ule1: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvgeui a3, a0, 11, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvltui a2, a0, 11, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_ule1: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvgeui a3, a0, 11, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvltui a2, a0, 11, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ule i32 11, %a @@ -444,14 +444,14 @@ define i32 @select_cc_example_ugt(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_ugt: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvgeui a3, a0, 12, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvltui a2, a0, 12, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_ugt: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvgeui a3, a0, 12, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvltui a2, a0, 12, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ugt i32 %a, 11 @@ -472,14 +472,14 @@ define i32 @select_cc_example_ugt1(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_ugt1: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvltui a3, a0, 11, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvgeui a2, a0, 11, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_ugt1: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvltui a3, a0, 11, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvgeui a2, a0, 11, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ugt i32 11, %a @@ -500,14 +500,14 @@ define i32 @select_cc_example_ult(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_ult: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvltui a3, a0, 11, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvgeui a2, a0, 11, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_ult: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvltui a3, a0, 11, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvgeui a2, a0, 11, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ult i32 %a, 11 @@ -528,14 +528,14 @@ define i32 @select_cc_example_ult1(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_ult1: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvgeui a3, a0, 12, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvltui a2, a0, 12, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_ult1: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvgeui a3, a0, 12, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvltui a2, a0, 12, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ult i32 11, %a @@ -556,14 +556,14 @@ define i32 @select_cc_example_uge(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_uge: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvgeui a3, a0, 11, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvltui a2, a0, 11, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_uge: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvgeui a3, a0, 11, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvltui a2, a0, 11, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp uge i32 %a, 11 @@ -584,14 +584,14 @@ define i32 @select_cc_example_uge1(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_uge1: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvltui a3, a0, 12, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvgeui a2, a0, 12, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_uge1: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvltui a3, a0, 12, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvgeui a2, a0, 12, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp uge i32 11, %a @@ -611,14 +611,14 @@ define i32 @select_cc_example_eq_reg(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_eq_reg: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mveq a3, a0, a1, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvne a2, a0, a1, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_eq_reg: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mveq a3, a0, a1, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvne a2, a0, a1, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp eq i32 %a, %b @@ -638,14 +638,14 @@ define i32 @select_cc_example_ne_reg(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_ne_reg: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvne a3, a0, a1, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mveq a2, a0, a1, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_ne_reg: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvne a3, a0, a1, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mveq a2, a0, a1, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ne i32 %a, %b @@ -665,14 +665,14 @@ define i32 @select_cc_example_slt_reg(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_slt_reg: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvlt a3, a0, a1, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvge a2, a0, a1, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_slt_reg: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvlt a3, a0, a1, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvge a2, a0, a1, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp slt i32 %a, %b @@ -692,14 +692,14 @@ define i32 @select_cc_example_sge_reg(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_sge_reg: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvge a3, a0, a1, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvlt a2, a0, a1, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_sge_reg: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvge a3, a0, a1, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvlt a2, a0, a1, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp sge i32 %a, %b @@ -719,14 +719,14 @@ define i32 @select_cc_example_sgt_reg(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_sgt_reg: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvlt a3, a1, a0, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvge a2, a1, a0, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_sgt_reg: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvlt a3, a1, a0, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvge a2, a1, a0, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp sgt i32 %a, %b @@ -746,14 +746,14 @@ define i32 @select_cc_example_sle_reg(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_sle_reg: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvge a3, a1, a0, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvlt a2, a1, a0, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_sle_reg: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvge a3, a1, a0, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvlt a2, a1, a0, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp sle i32 %a, %b @@ -773,14 +773,14 @@ define i32 @select_cc_example_ugt_reg(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_ugt_reg: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvltu a3, a1, a0, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvgeu a2, a1, a0, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_ugt_reg: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvltu a3, a1, a0, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvgeu a2, a1, a0, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ugt i32 %a, %b @@ -800,14 +800,14 @@ define i32 @select_cc_example_ult_reg(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_ult_reg: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvltu a3, a0, a1, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvgeu a2, a0, a1, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_ult_reg: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvltu a3, a0, a1, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvgeu a2, a0, a1, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ult i32 %a, %b @@ -827,14 +827,14 @@ define i32 @select_cc_example_uge_reg(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_uge_reg: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvgeu a3, a0, a1, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvltu a2, a0, a1, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_uge_reg: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvgeu a3, a0, a1, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvltu a2, a0, a1, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp uge i32 %a, %b @@ -854,14 +854,14 @@ define i32 @select_cc_example_ule_reg(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_ule_reg: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvgeu a3, a1, a0, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvltu a2, a1, a0, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_ule_reg: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvgeu a3, a1, a0, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvltu a2, a1, a0, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ule i32 %a, %b @@ -883,18 +883,263 @@ define i32 @select_cc_example_ule_neg(i32 %a, i32 %b, i32 %x, i32 %y) { ; RV32IXQCICM-LABEL: select_cc_example_ule_neg: ; RV32IXQCICM: # %bb.0: # %entry ; RV32IXQCICM-NEXT: li a1, -10 -; RV32IXQCICM-NEXT: qc.mvltu a3, a0, a1, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvgeu a2, a0, a1, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_ule_neg: ; RV32IXQCI: # %bb.0: # %entry ; RV32IXQCI-NEXT: li a1, -10 -; RV32IXQCI-NEXT: qc.mvltu a3, a0, a1, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvgeu a2, a0, a1, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ule i32 %a, -11 %sel = select i1 %cmp, i32 %x, i32 %y ret i32 %sel } + +define i32 @select_cc_example_eq_mv(i32 %a, i32 %b, i32 %x, i32 %y) { +; RV32I-LABEL: select_cc_example_eq_mv: +; RV32I: # %bb.0: # %entry +; RV32I-NEXT: beq a2, a1, .LBB32_2 +; RV32I-NEXT: # %bb.1: # %entry +; RV32I-NEXT: mv a0, a3 +; RV32I-NEXT: .LBB32_2: # %entry +; RV32I-NEXT: ret +; +; RV32IXQCICM-LABEL: select_cc_example_eq_mv: +; RV32IXQCICM: # %bb.0: # %entry +; RV32IXQCICM-NEXT: qc.mvne a0, a2, a1, a3 +; RV32IXQCICM-NEXT: ret +; +; RV32IXQCI-LABEL: select_cc_example_eq_mv: +; RV32IXQCI: # %bb.0: # %entry +; RV32IXQCI-NEXT: qc.mvne a0, a2, a1, a3 +; RV32IXQCI-NEXT: ret +entry: + %cmp = icmp eq i32 %x, %b + %sel = select i1 %cmp, i32 %a, i32 %y + ret i32 %sel +} + +define i32 @select_cc_example_lt_mv(i32 %a, i32 %b, i32 %x, i32 %y) { +; RV32I-LABEL: select_cc_example_lt_mv: +; RV32I: # %bb.0: # %entry +; RV32I-NEXT: blt a2, a1, .LBB33_2 +; RV32I-NEXT: # %bb.1: # %entry +; RV32I-NEXT: mv a0, a3 +; RV32I-NEXT: .LBB33_2: # %entry +; RV32I-NEXT: ret +; +; RV32IXQCICM-LABEL: select_cc_example_lt_mv: +; RV32IXQCICM: # %bb.0: # %entry +; RV32IXQCICM-NEXT: qc.mvge a0, a2, a1, a3 +; RV32IXQCICM-NEXT: ret +; +; RV32IXQCI-LABEL: select_cc_example_lt_mv: +; RV32IXQCI: # %bb.0: # %entry +; RV32IXQCI-NEXT: qc.mvge a0, a2, a1, a3 +; RV32IXQCI-NEXT: ret +entry: + %cmp = icmp slt i32 %x, %b + %sel = select i1 %cmp, i32 %a, i32 %y + ret i32 %sel +} + +define i32 @select_cc_example_ge_mv(i32 %a, i32 %b, i32 %x, i32 %y) { +; RV32I-LABEL: select_cc_example_ge_mv: +; RV32I: # %bb.0: # %entry +; RV32I-NEXT: bge a2, a1, .LBB34_2 +; RV32I-NEXT: # %bb.1: # %entry +; RV32I-NEXT: mv a0, a3 +; RV32I-NEXT: .LBB34_2: # %entry +; RV32I-NEXT: ret +; +; RV32IXQCICM-LABEL: select_cc_example_ge_mv: +; RV32IXQCICM: # %bb.0: # %entry +; RV32IXQCICM-NEXT: qc.mvlt a0, a2, a1, a3 +; RV32IXQCICM-NEXT: ret +; +; RV32IXQCI-LABEL: select_cc_example_ge_mv: +; RV32IXQCI: # %bb.0: # %entry +; RV32IXQCI-NEXT: qc.mvlt a0, a2, a1, a3 +; RV32IXQCI-NEXT: ret +entry: + %cmp = icmp sge i32 %x, %b + %sel = select i1 %cmp, i32 %a, i32 %y + ret i32 %sel +} + +define i32 @select_cc_example_ult_mv(i32 %a, i32 %b, i32 %x, i32 %y) { +; RV32I-LABEL: select_cc_example_ult_mv: +; RV32I: # %bb.0: # %entry +; RV32I-NEXT: bltu a2, a1, .LBB35_2 +; RV32I-NEXT: # %bb.1: # %entry +; RV32I-NEXT: mv a0, a3 +; RV32I-NEXT: .LBB35_2: # %entry +; RV32I-NEXT: ret +; +; RV32IXQCICM-LABEL: select_cc_example_ult_mv: +; RV32IXQCICM: # %bb.0: # %entry +; RV32IXQCICM-NEXT: qc.mvgeu a0, a2, a1, a3 +; RV32IXQCICM-NEXT: ret +; +; RV32IXQCI-LABEL: select_cc_example_ult_mv: +; RV32IXQCI: # %bb.0: # %entry +; RV32IXQCI-NEXT: qc.mvgeu a0, a2, a1, a3 +; RV32IXQCI-NEXT: ret +entry: + %cmp = icmp ult i32 %x, %b + %sel = select i1 %cmp, i32 %a, i32 %y + ret i32 %sel +} + +define i32 @select_cc_example_uge_mv(i32 %a, i32 %b, i32 %x, i32 %y) { +; RV32I-LABEL: select_cc_example_uge_mv: +; RV32I: # %bb.0: # %entry +; RV32I-NEXT: bgeu a2, a1, .LBB36_2 +; RV32I-NEXT: # %bb.1: # %entry +; RV32I-NEXT: mv a0, a3 +; RV32I-NEXT: .LBB36_2: # %entry +; RV32I-NEXT: ret +; +; RV32IXQCICM-LABEL: select_cc_example_uge_mv: +; RV32IXQCICM: # %bb.0: # %entry +; RV32IXQCICM-NEXT: qc.mvltu a0, a2, a1, a3 +; RV32IXQCICM-NEXT: ret +; +; RV32IXQCI-LABEL: select_cc_example_uge_mv: +; RV32IXQCI: # %bb.0: # %entry +; RV32IXQCI-NEXT: qc.mvltu a0, a2, a1, a3 +; RV32IXQCI-NEXT: ret +entry: + %cmp = icmp uge i32 %x, %b + %sel = select i1 %cmp, i32 %a, i32 %y + ret i32 %sel +} + +define i32 @select_cc_example_eq_imm_mv(i32 %a, i32 %b, i32 %x, i32 %y) { +; RV32I-LABEL: select_cc_example_eq_imm_mv: +; RV32I: # %bb.0: # %entry +; RV32I-NEXT: li a1, 11 +; RV32I-NEXT: beq a2, a1, .LBB37_2 +; RV32I-NEXT: # %bb.1: # %entry +; RV32I-NEXT: mv a0, a3 +; RV32I-NEXT: .LBB37_2: # %entry +; RV32I-NEXT: ret +; +; RV32IXQCICM-LABEL: select_cc_example_eq_imm_mv: +; RV32IXQCICM: # %bb.0: # %entry +; RV32IXQCICM-NEXT: qc.mvnei a0, a2, 11, a3 +; RV32IXQCICM-NEXT: ret +; +; RV32IXQCI-LABEL: select_cc_example_eq_imm_mv: +; RV32IXQCI: # %bb.0: # %entry +; RV32IXQCI-NEXT: qc.mvnei a0, a2, 11, a3 +; RV32IXQCI-NEXT: ret +entry: + %cmp = icmp eq i32 %x, 11 + %sel = select i1 %cmp, i32 %a, i32 %y + ret i32 %sel +} + +define i32 @select_cc_example_lt_imm_mv(i32 %a, i32 %b, i32 %x, i32 %y) { +; RV32I-LABEL: select_cc_example_lt_imm_mv: +; RV32I: # %bb.0: # %entry +; RV32I-NEXT: li a1, 11 +; RV32I-NEXT: blt a2, a1, .LBB38_2 +; RV32I-NEXT: # %bb.1: # %entry +; RV32I-NEXT: mv a0, a3 +; RV32I-NEXT: .LBB38_2: # %entry +; RV32I-NEXT: ret +; +; RV32IXQCICM-LABEL: select_cc_example_lt_imm_mv: +; RV32IXQCICM: # %bb.0: # %entry +; RV32IXQCICM-NEXT: qc.mvgei a0, a2, 11, a3 +; RV32IXQCICM-NEXT: ret +; +; RV32IXQCI-LABEL: select_cc_example_lt_imm_mv: +; RV32IXQCI: # %bb.0: # %entry +; RV32IXQCI-NEXT: qc.mvgei a0, a2, 11, a3 +; RV32IXQCI-NEXT: ret +entry: + %cmp = icmp slt i32 %x, 11 + %sel = select i1 %cmp, i32 %a, i32 %y + ret i32 %sel +} + +define i32 @select_cc_example_ge_imm_mv(i32 %a, i32 %b, i32 %x, i32 %y) { +; RV32I-LABEL: select_cc_example_ge_imm_mv: +; RV32I: # %bb.0: # %entry +; RV32I-NEXT: li a1, 10 +; RV32I-NEXT: blt a1, a2, .LBB39_2 +; RV32I-NEXT: # %bb.1: # %entry +; RV32I-NEXT: mv a0, a3 +; RV32I-NEXT: .LBB39_2: # %entry +; RV32I-NEXT: ret +; +; RV32IXQCICM-LABEL: select_cc_example_ge_imm_mv: +; RV32IXQCICM: # %bb.0: # %entry +; RV32IXQCICM-NEXT: qc.mvlti a0, a2, 11, a3 +; RV32IXQCICM-NEXT: ret +; +; RV32IXQCI-LABEL: select_cc_example_ge_imm_mv: +; RV32IXQCI: # %bb.0: # %entry +; RV32IXQCI-NEXT: qc.mvlti a0, a2, 11, a3 +; RV32IXQCI-NEXT: ret +entry: + %cmp = icmp sge i32 %x, 11 + %sel = select i1 %cmp, i32 %a, i32 %y + ret i32 %sel +} + +define i32 @select_cc_example_ult_imm_mv(i32 %a, i32 %b, i32 %x, i32 %y) { +; RV32I-LABEL: select_cc_example_ult_imm_mv: +; RV32I: # %bb.0: # %entry +; RV32I-NEXT: li a1, 11 +; RV32I-NEXT: bltu a2, a1, .LBB40_2 +; RV32I-NEXT: # %bb.1: # %entry +; RV32I-NEXT: mv a0, a3 +; RV32I-NEXT: .LBB40_2: # %entry +; RV32I-NEXT: ret +; +; RV32IXQCICM-LABEL: select_cc_example_ult_imm_mv: +; RV32IXQCICM: # %bb.0: # %entry +; RV32IXQCICM-NEXT: qc.mvgeui a0, a2, 11, a3 +; RV32IXQCICM-NEXT: ret +; +; RV32IXQCI-LABEL: select_cc_example_ult_imm_mv: +; RV32IXQCI: # %bb.0: # %entry +; RV32IXQCI-NEXT: qc.mvgeui a0, a2, 11, a3 +; RV32IXQCI-NEXT: ret +entry: + %cmp = icmp ult i32 %x, 11 + %sel = select i1 %cmp, i32 %a, i32 %y + ret i32 %sel +} + +define i32 @select_cc_example_uge_imm_mv(i32 %a, i32 %b, i32 %x, i32 %y) { +; RV32I-LABEL: select_cc_example_uge_imm_mv: +; RV32I: # %bb.0: # %entry +; RV32I-NEXT: li a1, 10 +; RV32I-NEXT: bltu a1, a2, .LBB41_2 +; RV32I-NEXT: # %bb.1: # %entry +; RV32I-NEXT: mv a0, a3 +; RV32I-NEXT: .LBB41_2: # %entry +; RV32I-NEXT: ret +; +; RV32IXQCICM-LABEL: select_cc_example_uge_imm_mv: +; RV32IXQCICM: # %bb.0: # %entry +; RV32IXQCICM-NEXT: qc.mvltui a0, a2, 11, a3 +; RV32IXQCICM-NEXT: ret +; +; RV32IXQCI-LABEL: select_cc_example_uge_imm_mv: +; RV32IXQCI: # %bb.0: # %entry +; RV32IXQCI-NEXT: qc.mvltui a0, a2, 11, a3 +; RV32IXQCI-NEXT: ret +entry: + %cmp = icmp uge i32 %x, 11 + %sel = select i1 %cmp, i32 %a, i32 %y + ret i32 %sel +} diff --git a/llvm/test/CodeGen/RISCV/xqcics.ll b/llvm/test/CodeGen/RISCV/xqcics.ll index 38de8fb..5b7ca9e7 100644 --- a/llvm/test/CodeGen/RISCV/xqcics.ll +++ b/llvm/test/CodeGen/RISCV/xqcics.ll @@ -134,14 +134,14 @@ define i32 @select_cc_example_eq(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_eq: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mveqi a3, a0, 11, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvnei a2, a0, 11, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_eq: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mveqi a3, a0, 11, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvnei a2, a0, 11, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp eq i32 %a, 11 @@ -167,14 +167,14 @@ define i32 @select_cc_example_eq_c(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_eq_c: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mveqi a3, a0, 11, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mvnei a2, a0, 11, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_eq_c: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mveqi a3, a0, 11, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mvnei a2, a0, 11, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp eq i32 11, %a @@ -200,14 +200,14 @@ define i32 @select_cc_example_ne(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_ne: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvnei a3, a0, 11, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mveqi a2, a0, 11, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_ne: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvnei a3, a0, 11, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mveqi a2, a0, 11, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ne i32 %a, 11 @@ -233,14 +233,14 @@ define i32 @select_cc_example_ne_c(i32 %a, i32 %b, i32 %x, i32 %y) { ; ; RV32IXQCICM-LABEL: select_cc_example_ne_c: ; RV32IXQCICM: # %bb.0: # %entry -; RV32IXQCICM-NEXT: qc.mvnei a3, a0, 11, a2 -; RV32IXQCICM-NEXT: mv a0, a3 +; RV32IXQCICM-NEXT: qc.mveqi a2, a0, 11, a3 +; RV32IXQCICM-NEXT: mv a0, a2 ; RV32IXQCICM-NEXT: ret ; ; RV32IXQCI-LABEL: select_cc_example_ne_c: ; RV32IXQCI: # %bb.0: # %entry -; RV32IXQCI-NEXT: qc.mvnei a3, a0, 11, a2 -; RV32IXQCI-NEXT: mv a0, a3 +; RV32IXQCI-NEXT: qc.mveqi a2, a0, 11, a3 +; RV32IXQCI-NEXT: mv a0, a2 ; RV32IXQCI-NEXT: ret entry: %cmp = icmp ne i32 11, %a diff --git a/llvm/test/CodeGen/VE/Scalar/max.ll b/llvm/test/CodeGen/VE/Scalar/max.ll index 51da557..7950842 100644 --- a/llvm/test/CodeGen/VE/Scalar/max.ll +++ b/llvm/test/CodeGen/VE/Scalar/max.ll @@ -1,7 +1,5 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py ; RUN: llc < %s -mtriple=ve-unknown-unknown | FileCheck %s -; RUN: llc < %s -mtriple=ve-unknown-unknown -enable-no-signed-zeros-fp-math \ -; RUN: -enable-no-nans-fp-math | FileCheck %s -check-prefix=OPT define double @maxf64(double, double) { ; CHECK-LABEL: maxf64: @@ -10,16 +8,21 @@ define double @maxf64(double, double) { ; CHECK-NEXT: cmov.d.gt %s1, %s0, %s2 ; CHECK-NEXT: or %s0, 0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: maxf64: -; OPT: # %bb.0: -; OPT-NEXT: fmax.d %s0, %s0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = fcmp ogt double %0, %1 %4 = select i1 %3, double %0, double %1 ret double %4 } +define double @maxf64_fast(double, double) { +; CHECK-LABEL: maxf64_fast: +; CHECK: # %bb.0: +; CHECK-NEXT: fmax.d %s0, %s0, %s1 +; CHECK-NEXT: b.l.t (, %s10) + %3 = fcmp ogt double %0, %1 + %4 = select nnan nsz i1 %3, double %0, double %1 + ret double %4 +} + define double @max2f64(double, double) { ; CHECK-LABEL: max2f64: ; CHECK: # %bb.0: @@ -27,16 +30,21 @@ define double @max2f64(double, double) { ; CHECK-NEXT: cmov.d.ge %s1, %s0, %s2 ; CHECK-NEXT: or %s0, 0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: max2f64: -; OPT: # %bb.0: -; OPT-NEXT: fmax.d %s0, %s0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = fcmp oge double %0, %1 %4 = select i1 %3, double %0, double %1 ret double %4 } +define double @max2f64_fast(double, double) { +; CHECK-LABEL: max2f64_fast: +; CHECK: # %bb.0: +; CHECK-NEXT: fmax.d %s0, %s0, %s1 +; CHECK-NEXT: b.l.t (, %s10) + %3 = fcmp oge double %0, %1 + %4 = select nnan nsz i1 %3, double %0, double %1 + ret double %4 +} + ; VE has no max for unordered comparison define double @maxuf64(double, double) { ; CHECK-LABEL: maxuf64: @@ -45,16 +53,21 @@ define double @maxuf64(double, double) { ; CHECK-NEXT: cmov.d.gtnan %s1, %s0, %s2 ; CHECK-NEXT: or %s0, 0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: maxuf64: -; OPT: # %bb.0: -; OPT-NEXT: fmax.d %s0, %s0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = fcmp ugt double %0, %1 %4 = select i1 %3, double %0, double %1 ret double %4 } +define double @maxuf64_fast(double, double) { +; CHECK-LABEL: maxuf64_fast: +; CHECK: # %bb.0: +; CHECK-NEXT: fmax.d %s0, %s0, %s1 +; CHECK-NEXT: b.l.t (, %s10) + %3 = fcmp ugt double %0, %1 + %4 = select nnan nsz i1 %3, double %0, double %1 + ret double %4 +} + ; VE has no max for unordered comparison define double @max2uf64(double, double) { ; CHECK-LABEL: max2uf64: @@ -63,16 +76,21 @@ define double @max2uf64(double, double) { ; CHECK-NEXT: cmov.d.genan %s1, %s0, %s2 ; CHECK-NEXT: or %s0, 0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: max2uf64: -; OPT: # %bb.0: -; OPT-NEXT: fmax.d %s0, %s0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = fcmp uge double %0, %1 %4 = select i1 %3, double %0, double %1 ret double %4 } +define double @max2uf64_fast(double, double) { +; CHECK-LABEL: max2uf64_fast: +; CHECK: # %bb.0: +; CHECK-NEXT: fmax.d %s0, %s0, %s1 +; CHECK-NEXT: b.l.t (, %s10) + %3 = fcmp uge double %0, %1 + %4 = select nnan nsz i1 %3, double %0, double %1 + ret double %4 +} + define float @maxf32(float, float) { ; CHECK-LABEL: maxf32: ; CHECK: # %bb.0: @@ -80,16 +98,21 @@ define float @maxf32(float, float) { ; CHECK-NEXT: cmov.s.gt %s1, %s0, %s2 ; CHECK-NEXT: or %s0, 0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: maxf32: -; OPT: # %bb.0: -; OPT-NEXT: fmax.s %s0, %s0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = fcmp ogt float %0, %1 %4 = select i1 %3, float %0, float %1 ret float %4 } +define float @maxf32_fast(float, float) { +; CHECK-LABEL: maxf32_fast: +; CHECK: # %bb.0: +; CHECK-NEXT: fmax.s %s0, %s0, %s1 +; CHECK-NEXT: b.l.t (, %s10) + %3 = fcmp ogt float %0, %1 + %4 = select nnan nsz i1 %3, float %0, float %1 + ret float %4 +} + define float @max2f32(float, float) { ; CHECK-LABEL: max2f32: ; CHECK: # %bb.0: @@ -97,16 +120,21 @@ define float @max2f32(float, float) { ; CHECK-NEXT: cmov.s.ge %s1, %s0, %s2 ; CHECK-NEXT: or %s0, 0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: max2f32: -; OPT: # %bb.0: -; OPT-NEXT: fmax.s %s0, %s0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = fcmp oge float %0, %1 %4 = select i1 %3, float %0, float %1 ret float %4 } +define float @max2f32_fast(float, float) { +; CHECK-LABEL: max2f32_fast: +; CHECK: # %bb.0: +; CHECK-NEXT: fmax.s %s0, %s0, %s1 +; CHECK-NEXT: b.l.t (, %s10) + %3 = fcmp oge float %0, %1 + %4 = select nnan nsz i1 %3, float %0, float %1 + ret float %4 +} + define float @maxuf32(float, float) { ; CHECK-LABEL: maxuf32: ; CHECK: # %bb.0: @@ -114,16 +142,21 @@ define float @maxuf32(float, float) { ; CHECK-NEXT: cmov.s.gtnan %s1, %s0, %s2 ; CHECK-NEXT: or %s0, 0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: maxuf32: -; OPT: # %bb.0: -; OPT-NEXT: fmax.s %s0, %s0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = fcmp ugt float %0, %1 %4 = select i1 %3, float %0, float %1 ret float %4 } +define float @maxuf32_fast(float, float) { +; CHECK-LABEL: maxuf32_fast: +; CHECK: # %bb.0: +; CHECK-NEXT: fmax.s %s0, %s0, %s1 +; CHECK-NEXT: b.l.t (, %s10) + %3 = fcmp ugt float %0, %1 + %4 = select nnan nsz i1 %3, float %0, float %1 + ret float %4 +} + define float @max2uf32(float, float) { ; CHECK-LABEL: max2uf32: ; CHECK: # %bb.0: @@ -131,26 +164,26 @@ define float @max2uf32(float, float) { ; CHECK-NEXT: cmov.s.genan %s1, %s0, %s2 ; CHECK-NEXT: or %s0, 0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: max2uf32: -; OPT: # %bb.0: -; OPT-NEXT: fmax.s %s0, %s0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = fcmp uge float %0, %1 %4 = select i1 %3, float %0, float %1 ret float %4 } +define float @max2uf32_fast(float, float) { +; CHECK-LABEL: max2uf32_fast: +; CHECK: # %bb.0: +; CHECK-NEXT: fmax.s %s0, %s0, %s1 +; CHECK-NEXT: b.l.t (, %s10) + %3 = fcmp uge float %0, %1 + %4 = select nnan nsz i1 %3, float %0, float %1 + ret float %4 +} + define i64 @maxi64(i64, i64) { ; CHECK-LABEL: maxi64: ; CHECK: # %bb.0: ; CHECK-NEXT: maxs.l %s0, %s0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: maxi64: -; OPT: # %bb.0: -; OPT-NEXT: maxs.l %s0, %s0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = icmp sgt i64 %0, %1 %4 = select i1 %3, i64 %0, i64 %1 ret i64 %4 @@ -161,11 +194,6 @@ define i64 @max2i64(i64, i64) { ; CHECK: # %bb.0: ; CHECK-NEXT: maxs.l %s0, %s0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: max2i64: -; OPT: # %bb.0: -; OPT-NEXT: maxs.l %s0, %s0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = icmp sge i64 %0, %1 %4 = select i1 %3, i64 %0, i64 %1 ret i64 %4 @@ -178,13 +206,6 @@ define i64 @maxu64(i64, i64) { ; CHECK-NEXT: cmov.l.gt %s1, %s0, %s2 ; CHECK-NEXT: or %s0, 0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: maxu64: -; OPT: # %bb.0: -; OPT-NEXT: cmpu.l %s2, %s0, %s1 -; OPT-NEXT: cmov.l.gt %s1, %s0, %s2 -; OPT-NEXT: or %s0, 0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = icmp ugt i64 %0, %1 %4 = select i1 %3, i64 %0, i64 %1 ret i64 %4 @@ -197,13 +218,6 @@ define i64 @max2u64(i64, i64) { ; CHECK-NEXT: cmov.l.ge %s1, %s0, %s2 ; CHECK-NEXT: or %s0, 0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: max2u64: -; OPT: # %bb.0: -; OPT-NEXT: cmpu.l %s2, %s0, %s1 -; OPT-NEXT: cmov.l.ge %s1, %s0, %s2 -; OPT-NEXT: or %s0, 0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = icmp uge i64 %0, %1 %4 = select i1 %3, i64 %0, i64 %1 ret i64 %4 @@ -214,11 +228,6 @@ define i32 @maxi32(i32, i32) { ; CHECK: # %bb.0: ; CHECK-NEXT: maxs.w.sx %s0, %s0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: maxi32: -; OPT: # %bb.0: -; OPT-NEXT: maxs.w.sx %s0, %s0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = icmp sgt i32 %0, %1 %4 = select i1 %3, i32 %0, i32 %1 ret i32 %4 @@ -229,11 +238,6 @@ define i32 @max2i32(i32, i32) { ; CHECK: # %bb.0: ; CHECK-NEXT: maxs.w.sx %s0, %s0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: max2i32: -; OPT: # %bb.0: -; OPT-NEXT: maxs.w.sx %s0, %s0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = icmp sge i32 %0, %1 %4 = select i1 %3, i32 %0, i32 %1 ret i32 %4 @@ -246,13 +250,6 @@ define i32 @maxu32(i32, i32) { ; CHECK-NEXT: cmov.w.gt %s1, %s0, %s2 ; CHECK-NEXT: or %s0, 0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: maxu32: -; OPT: # %bb.0: -; OPT-NEXT: cmpu.w %s2, %s0, %s1 -; OPT-NEXT: cmov.w.gt %s1, %s0, %s2 -; OPT-NEXT: or %s0, 0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = icmp ugt i32 %0, %1 %4 = select i1 %3, i32 %0, i32 %1 ret i32 %4 @@ -265,13 +262,6 @@ define i32 @max2u32(i32, i32) { ; CHECK-NEXT: cmov.w.ge %s1, %s0, %s2 ; CHECK-NEXT: or %s0, 0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: max2u32: -; OPT: # %bb.0: -; OPT-NEXT: cmpu.w %s2, %s0, %s1 -; OPT-NEXT: cmov.w.ge %s1, %s0, %s2 -; OPT-NEXT: or %s0, 0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = icmp uge i32 %0, %1 %4 = select i1 %3, i32 %0, i32 %1 ret i32 %4 @@ -283,12 +273,6 @@ define zeroext i1 @maxi1(i1 zeroext, i1 zeroext) { ; CHECK-NEXT: or %s0, %s0, %s1 ; CHECK-NEXT: and %s0, 1, %s0 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: maxi1: -; OPT: # %bb.0: -; OPT-NEXT: or %s0, %s0, %s1 -; OPT-NEXT: and %s0, 1, %s0 -; OPT-NEXT: b.l.t (, %s10) %3 = xor i1 %1, true %4 = and i1 %3, %0 %5 = select i1 %4, i1 %0, i1 %1 diff --git a/llvm/test/CodeGen/VE/Scalar/min.ll b/llvm/test/CodeGen/VE/Scalar/min.ll index e8f4939..36a2e06 100644 --- a/llvm/test/CodeGen/VE/Scalar/min.ll +++ b/llvm/test/CodeGen/VE/Scalar/min.ll @@ -1,7 +1,5 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py ; RUN: llc < %s -mtriple=ve-unknown-unknown | FileCheck %s -; RUN: llc < %s -mtriple=ve-unknown-unknown -enable-no-signed-zeros-fp-math \ -; RUN: -enable-no-nans-fp-math | FileCheck %s -check-prefix=OPT define double @minf64(double, double) { ; CHECK-LABEL: minf64: @@ -10,16 +8,21 @@ define double @minf64(double, double) { ; CHECK-NEXT: cmov.d.lt %s1, %s0, %s2 ; CHECK-NEXT: or %s0, 0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: minf64: -; OPT: # %bb.0: -; OPT-NEXT: fmin.d %s0, %s0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = fcmp olt double %0, %1 %4 = select i1 %3, double %0, double %1 ret double %4 } +define double @minf64_fast(double, double) { +; CHECK-LABEL: minf64_fast: +; CHECK: # %bb.0: +; CHECK-NEXT: fmin.d %s0, %s0, %s1 +; CHECK-NEXT: b.l.t (, %s10) + %3 = fcmp olt double %0, %1 + %4 = select nnan nsz i1 %3, double %0, double %1 + ret double %4 +} + define double @min2f64(double, double) { ; CHECK-LABEL: min2f64: ; CHECK: # %bb.0: @@ -27,16 +30,21 @@ define double @min2f64(double, double) { ; CHECK-NEXT: cmov.d.le %s1, %s0, %s2 ; CHECK-NEXT: or %s0, 0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: min2f64: -; OPT: # %bb.0: -; OPT-NEXT: fmin.d %s0, %s0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = fcmp ole double %0, %1 %4 = select i1 %3, double %0, double %1 ret double %4 } +define double @min2f64_fast(double, double) { +; CHECK-LABEL: min2f64_fast: +; CHECK: # %bb.0: +; CHECK-NEXT: fmin.d %s0, %s0, %s1 +; CHECK-NEXT: b.l.t (, %s10) + %3 = fcmp ole double %0, %1 + %4 = select nnan nsz i1 %3, double %0, double %1 + ret double %4 +} + define double @minuf64(double, double) { ; CHECK-LABEL: minuf64: ; CHECK: # %bb.0: @@ -44,16 +52,21 @@ define double @minuf64(double, double) { ; CHECK-NEXT: cmov.d.ltnan %s1, %s0, %s2 ; CHECK-NEXT: or %s0, 0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: minuf64: -; OPT: # %bb.0: -; OPT-NEXT: fmin.d %s0, %s0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = fcmp ult double %0, %1 %4 = select i1 %3, double %0, double %1 ret double %4 } +define double @minuf64_fast(double, double) { +; CHECK-LABEL: minuf64_fast: +; CHECK: # %bb.0: +; CHECK-NEXT: fmin.d %s0, %s0, %s1 +; CHECK-NEXT: b.l.t (, %s10) + %3 = fcmp ult double %0, %1 + %4 = select nnan nsz i1 %3, double %0, double %1 + ret double %4 +} + define double @min2uf64(double, double) { ; CHECK-LABEL: min2uf64: ; CHECK: # %bb.0: @@ -61,16 +74,21 @@ define double @min2uf64(double, double) { ; CHECK-NEXT: cmov.d.lenan %s1, %s0, %s2 ; CHECK-NEXT: or %s0, 0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: min2uf64: -; OPT: # %bb.0: -; OPT-NEXT: fmin.d %s0, %s0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = fcmp ule double %0, %1 %4 = select i1 %3, double %0, double %1 ret double %4 } +define double @min2uf64_fast(double, double) { +; CHECK-LABEL: min2uf64_fast: +; CHECK: # %bb.0: +; CHECK-NEXT: fmin.d %s0, %s0, %s1 +; CHECK-NEXT: b.l.t (, %s10) + %3 = fcmp ule double %0, %1 + %4 = select nnan nsz i1 %3, double %0, double %1 + ret double %4 +} + define float @minf32(float, float) { ; CHECK-LABEL: minf32: ; CHECK: # %bb.0: @@ -78,16 +96,21 @@ define float @minf32(float, float) { ; CHECK-NEXT: cmov.s.lt %s1, %s0, %s2 ; CHECK-NEXT: or %s0, 0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: minf32: -; OPT: # %bb.0: -; OPT-NEXT: fmin.s %s0, %s0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = fcmp olt float %0, %1 %4 = select i1 %3, float %0, float %1 ret float %4 } +define float @minf32_fast(float, float) { +; CHECK-LABEL: minf32_fast: +; CHECK: # %bb.0: +; CHECK-NEXT: fmin.s %s0, %s0, %s1 +; CHECK-NEXT: b.l.t (, %s10) + %3 = fcmp olt float %0, %1 + %4 = select nnan nsz i1 %3, float %0, float %1 + ret float %4 +} + define float @min2f32(float, float) { ; CHECK-LABEL: min2f32: ; CHECK: # %bb.0: @@ -95,16 +118,21 @@ define float @min2f32(float, float) { ; CHECK-NEXT: cmov.s.le %s1, %s0, %s2 ; CHECK-NEXT: or %s0, 0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: min2f32: -; OPT: # %bb.0: -; OPT-NEXT: fmin.s %s0, %s0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = fcmp ole float %0, %1 %4 = select i1 %3, float %0, float %1 ret float %4 } +define float @min2f32_fast(float, float) { +; CHECK-LABEL: min2f32_fast: +; CHECK: # %bb.0: +; CHECK-NEXT: fmin.s %s0, %s0, %s1 +; CHECK-NEXT: b.l.t (, %s10) + %3 = fcmp ole float %0, %1 + %4 = select nnan nsz i1 %3, float %0, float %1 + ret float %4 +} + define float @minuf32(float, float) { ; CHECK-LABEL: minuf32: ; CHECK: # %bb.0: @@ -112,16 +140,21 @@ define float @minuf32(float, float) { ; CHECK-NEXT: cmov.s.ltnan %s1, %s0, %s2 ; CHECK-NEXT: or %s0, 0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: minuf32: -; OPT: # %bb.0: -; OPT-NEXT: fmin.s %s0, %s0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = fcmp ult float %0, %1 %4 = select i1 %3, float %0, float %1 ret float %4 } +define float @minuf32_fast(float, float) { +; CHECK-LABEL: minuf32_fast: +; CHECK: # %bb.0: +; CHECK-NEXT: fmin.s %s0, %s0, %s1 +; CHECK-NEXT: b.l.t (, %s10) + %3 = fcmp ult float %0, %1 + %4 = select nnan nsz i1 %3, float %0, float %1 + ret float %4 +} + define float @min2uf32(float, float) { ; CHECK-LABEL: min2uf32: ; CHECK: # %bb.0: @@ -129,26 +162,26 @@ define float @min2uf32(float, float) { ; CHECK-NEXT: cmov.s.lenan %s1, %s0, %s2 ; CHECK-NEXT: or %s0, 0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: min2uf32: -; OPT: # %bb.0: -; OPT-NEXT: fmin.s %s0, %s0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = fcmp ule float %0, %1 %4 = select i1 %3, float %0, float %1 ret float %4 } +define float @min2uf32_fast(float, float) { +; CHECK-LABEL: min2uf32_fast: +; CHECK: # %bb.0: +; CHECK-NEXT: fmin.s %s0, %s0, %s1 +; CHECK-NEXT: b.l.t (, %s10) + %3 = fcmp ule float %0, %1 + %4 = select nnan nsz i1 %3, float %0, float %1 + ret float %4 +} + define i64 @mini64(i64, i64) { ; CHECK-LABEL: mini64: ; CHECK: # %bb.0: ; CHECK-NEXT: mins.l %s0, %s0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: mini64: -; OPT: # %bb.0: -; OPT-NEXT: mins.l %s0, %s0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = icmp slt i64 %0, %1 %4 = select i1 %3, i64 %0, i64 %1 ret i64 %4 @@ -159,11 +192,6 @@ define i64 @min2i64(i64, i64) { ; CHECK: # %bb.0: ; CHECK-NEXT: mins.l %s0, %s0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: min2i64: -; OPT: # %bb.0: -; OPT-NEXT: mins.l %s0, %s0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = icmp sle i64 %0, %1 %4 = select i1 %3, i64 %0, i64 %1 ret i64 %4 @@ -176,13 +204,6 @@ define i64 @minu64(i64, i64) { ; CHECK-NEXT: cmov.l.lt %s1, %s0, %s2 ; CHECK-NEXT: or %s0, 0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: minu64: -; OPT: # %bb.0: -; OPT-NEXT: cmpu.l %s2, %s0, %s1 -; OPT-NEXT: cmov.l.lt %s1, %s0, %s2 -; OPT-NEXT: or %s0, 0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = icmp ult i64 %0, %1 %4 = select i1 %3, i64 %0, i64 %1 ret i64 %4 @@ -195,13 +216,6 @@ define i64 @min2u64(i64, i64) { ; CHECK-NEXT: cmov.l.le %s1, %s0, %s2 ; CHECK-NEXT: or %s0, 0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: min2u64: -; OPT: # %bb.0: -; OPT-NEXT: cmpu.l %s2, %s0, %s1 -; OPT-NEXT: cmov.l.le %s1, %s0, %s2 -; OPT-NEXT: or %s0, 0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = icmp ule i64 %0, %1 %4 = select i1 %3, i64 %0, i64 %1 ret i64 %4 @@ -212,11 +226,6 @@ define i32 @mini32(i32, i32) { ; CHECK: # %bb.0: ; CHECK-NEXT: mins.w.sx %s0, %s0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: mini32: -; OPT: # %bb.0: -; OPT-NEXT: mins.w.sx %s0, %s0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = icmp slt i32 %0, %1 %4 = select i1 %3, i32 %0, i32 %1 ret i32 %4 @@ -227,11 +236,6 @@ define i32 @min2i32(i32, i32) { ; CHECK: # %bb.0: ; CHECK-NEXT: mins.w.sx %s0, %s0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: min2i32: -; OPT: # %bb.0: -; OPT-NEXT: mins.w.sx %s0, %s0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = icmp sle i32 %0, %1 %4 = select i1 %3, i32 %0, i32 %1 ret i32 %4 @@ -244,13 +248,6 @@ define i32 @minu32(i32, i32) { ; CHECK-NEXT: cmov.w.lt %s1, %s0, %s2 ; CHECK-NEXT: or %s0, 0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: minu32: -; OPT: # %bb.0: -; OPT-NEXT: cmpu.w %s2, %s0, %s1 -; OPT-NEXT: cmov.w.lt %s1, %s0, %s2 -; OPT-NEXT: or %s0, 0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = icmp ult i32 %0, %1 %4 = select i1 %3, i32 %0, i32 %1 ret i32 %4 @@ -263,13 +260,6 @@ define i32 @min2u32(i32, i32) { ; CHECK-NEXT: cmov.w.le %s1, %s0, %s2 ; CHECK-NEXT: or %s0, 0, %s1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: min2u32: -; OPT: # %bb.0: -; OPT-NEXT: cmpu.w %s2, %s0, %s1 -; OPT-NEXT: cmov.w.le %s1, %s0, %s2 -; OPT-NEXT: or %s0, 0, %s1 -; OPT-NEXT: b.l.t (, %s10) %3 = icmp ule i32 %0, %1 %4 = select i1 %3, i32 %0, i32 %1 ret i32 %4 @@ -283,14 +273,6 @@ define zeroext i1 @mini1(i1 zeroext, i1 zeroext) { ; CHECK-NEXT: cmov.w.ne %s0, %s1, %s2 ; CHECK-NEXT: adds.w.zx %s0, %s0, (0)1 ; CHECK-NEXT: b.l.t (, %s10) -; -; OPT-LABEL: mini1: -; OPT: # %bb.0: -; OPT-NEXT: and %s2, 1, %s0 -; OPT-NEXT: and %s0, %s1, %s0 -; OPT-NEXT: cmov.w.ne %s0, %s1, %s2 -; OPT-NEXT: adds.w.zx %s0, %s0, (0)1 -; OPT-NEXT: b.l.t (, %s10) %3 = xor i1 %0, true %4 = and i1 %3, %1 %5 = select i1 %4, i1 %0, i1 %1 diff --git a/llvm/test/CodeGen/WebAssembly/partial-reduce-accumulate.ll b/llvm/test/CodeGen/WebAssembly/partial-reduce-accumulate.ll index 47ea762..a599f46 100644 --- a/llvm/test/CodeGen/WebAssembly/partial-reduce-accumulate.ll +++ b/llvm/test/CodeGen/WebAssembly/partial-reduce-accumulate.ll @@ -19,11 +19,11 @@ define hidden i32 @accumulate_add_u8_u8(ptr noundef readonly %a, ptr noundef re ; MAX-BANDWIDTH: v128.load ; MAX-BANDWIDTH: i16x8.extadd_pairwise_i8x16_u ; MAX-BANDWIDTH: i32x4.extadd_pairwise_i16x8_u -; MAX-BANDWIDTH: i32x4.add ; MAX-BANDWIDTH: v128.load ; MAX-BANDWIDTH: i16x8.extadd_pairwise_i8x16_u ; MAX-BANDWIDTH: i32x4.extadd_pairwise_i16x8_u ; MAX-BANDWIDTH: i32x4.add +; MAX-BANDWIDTH: i32x4.add entry: %cmp8.not = icmp eq i32 %N, 0 @@ -65,11 +65,11 @@ define hidden i32 @accumulate_add_s8_s8(ptr noundef readonly %a, ptr noundef re ; MAX-BANDWIDTH: v128.load ; MAX-BANDWIDTH: i16x8.extadd_pairwise_i8x16_s ; MAX-BANDWIDTH: i32x4.extadd_pairwise_i16x8_s -; MAX-BANDWIDTH: i32x4.add ; MAX-BANDWIDTH: v128.load ; MAX-BANDWIDTH: i16x8.extadd_pairwise_i8x16_s ; MAX-BANDWIDTH: i32x4.extadd_pairwise_i16x8_s ; MAX-BANDWIDTH: i32x4.add +; MAX-BANDWIDTH: i32x4.add entry: %cmp8.not = icmp eq i32 %N, 0 br i1 %cmp8.not, label %for.cond.cleanup, label %for.body @@ -108,12 +108,11 @@ define hidden i32 @accumulate_add_s8_u8(ptr noundef readonly %a, ptr noundef re ; MAX-BANDWIDTH: loop ; MAX-BANDWIDTH: v128.load -; MAX-BANDWIDTH: i16x8.extadd_pairwise_i8x16_s -; MAX-BANDWIDTH: i32x4.extadd_pairwise_i16x8_s -; MAX-BANDWIDTH: i32x4.add -; MAX-BANDWIDTH: v128.load ; MAX-BANDWIDTH: i16x8.extadd_pairwise_i8x16_u ; MAX-BANDWIDTH: i32x4.extadd_pairwise_i16x8_u +; MAX-BANDWIDTH: v128.load +; MAX-BANDWIDTH: i16x8.extadd_pairwise_i8x16_s +; MAX-BANDWIDTH: i32x4.extadd_pairwise_i16x8_s ; MAX-BANDWIDTH: i32x4.add entry: %cmp8.not = icmp eq i32 %N, 0 @@ -363,10 +362,10 @@ define hidden i32 @accumulate_add_u16_u16(ptr noundef readonly %a, ptr noundef ; MAX-BANDWIDTH: loop ; MAX-BANDWIDTH: v128.load ; MAX-BANDWIDTH: i32x4.extadd_pairwise_i16x8_u -; MAX-BANDWIDTH: i32x4.add ; MAX-BANDWIDTH: v128.load ; MAX-BANDWIDTH: i32x4.extadd_pairwise_i16x8_u ; MAX-BANDWIDTH: i32x4.add +; MAX-BANDWIDTH: i32x4.add entry: %cmp8.not = icmp eq i32 %N, 0 br i1 %cmp8.not, label %for.cond.cleanup, label %for.body @@ -402,10 +401,10 @@ define hidden i32 @accumulate_add_s16_s16(ptr noundef readonly %a, ptr noundef ; MAX-BANDWIDTH: loop ; MAX-BANDWIDTH: v128.load ; MAX-BANDWIDTH: i32x4.extadd_pairwise_i16x8_s -; MAX-BANDWIDTH: i32x4.add ; MAX-BANDWIDTH: v128.load ; MAX-BANDWIDTH: i32x4.extadd_pairwise_i16x8_s ; MAX-BANDWIDTH: i32x4.add +; MAX-BANDWIDTH: i32x4.add entry: %cmp8.not = icmp eq i32 %N, 0 br i1 %cmp8.not, label %for.cond.cleanup, label %for.body diff --git a/llvm/test/CodeGen/X86/avx10.2-intrinsic-upgrade.ll b/llvm/test/CodeGen/X86/avx10.2-intrinsic-upgrade.ll new file mode 100644 index 0000000..76d84c1 --- /dev/null +++ b/llvm/test/CodeGen/X86/avx10.2-intrinsic-upgrade.ll @@ -0,0 +1,99 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6 +; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx10.2 --show-mc-encoding | FileCheck %s --check-prefixes=X86 +; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx10.2 --show-mc-encoding | FileCheck %s --check-prefixes=X64 + +declare <16 x i32> @llvm.x86.avx10.vpdpbssd.512(<16 x i32>, <16 x i32>, <16 x i32>) + +define <16 x i32>@test_int_x86_avx10_vpdpbssd_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) { +; X86-LABEL: test_int_x86_avx10_vpdpbssd_512: +; X86: # %bb.0: +; X86-NEXT: vpdpbssd %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x77,0x48,0x50,0xc2] +; X86-NEXT: retl # encoding: [0xc3] +; +; X64-LABEL: test_int_x86_avx10_vpdpbssd_512: +; X64: # %bb.0: +; X64-NEXT: vpdpbssd %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x77,0x48,0x50,0xc2] +; X64-NEXT: retq # encoding: [0xc3] + %res = call <16 x i32> @llvm.x86.avx10.vpdpbssd.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) + ret <16 x i32> %res +} + +declare <16 x i32> @llvm.x86.avx10.vpdpbssds.512(<16 x i32>, <16 x i32>, <16 x i32>) + +define <16 x i32>@test_int_x86_avx10_vpdpbssds_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) { +; X86-LABEL: test_int_x86_avx10_vpdpbssds_512: +; X86: # %bb.0: +; X86-NEXT: vpdpbssds %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x77,0x48,0x51,0xc2] +; X86-NEXT: retl # encoding: [0xc3] +; +; X64-LABEL: test_int_x86_avx10_vpdpbssds_512: +; X64: # %bb.0: +; X64-NEXT: vpdpbssds %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x77,0x48,0x51,0xc2] +; X64-NEXT: retq # encoding: [0xc3] + %res = call <16 x i32> @llvm.x86.avx10.vpdpbssds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) + ret <16 x i32> %res +} + +declare <16 x i32> @llvm.x86.avx10.vpdpbsud.512(<16 x i32>, <16 x i32>, <16 x i32>) + +define <16 x i32>@test_int_x86_avx10_vpdpbsud_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) { +; X86-LABEL: test_int_x86_avx10_vpdpbsud_512: +; X86: # %bb.0: +; X86-NEXT: vpdpbsud %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x76,0x48,0x50,0xc2] +; X86-NEXT: retl # encoding: [0xc3] +; +; X64-LABEL: test_int_x86_avx10_vpdpbsud_512: +; X64: # %bb.0: +; X64-NEXT: vpdpbsud %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x76,0x48,0x50,0xc2] +; X64-NEXT: retq # encoding: [0xc3] + %res = call <16 x i32> @llvm.x86.avx10.vpdpbsud.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) + ret <16 x i32> %res +} + +declare <16 x i32> @llvm.x86.avx10.vpdpbsuds.512(<16 x i32>, <16 x i32>, <16 x i32>) + +define <16 x i32>@test_int_x86_avx10_vpdpbsuds_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) { +; X86-LABEL: test_int_x86_avx10_vpdpbsuds_512: +; X86: # %bb.0: +; X86-NEXT: vpdpbsuds %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x76,0x48,0x51,0xc2] +; X86-NEXT: retl # encoding: [0xc3] +; +; X64-LABEL: test_int_x86_avx10_vpdpbsuds_512: +; X64: # %bb.0: +; X64-NEXT: vpdpbsuds %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x76,0x48,0x51,0xc2] +; X64-NEXT: retq # encoding: [0xc3] + %res = call <16 x i32> @llvm.x86.avx10.vpdpbsuds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) + ret <16 x i32> %res +} + +declare <16 x i32> @llvm.x86.avx10.vpdpbuud.512(<16 x i32>, <16 x i32>, <16 x i32>) + +define <16 x i32>@test_int_x86_avx10_vpdpbuud_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) { +; X86-LABEL: test_int_x86_avx10_vpdpbuud_512: +; X86: # %bb.0: +; X86-NEXT: vpdpbuud %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x74,0x48,0x50,0xc2] +; X86-NEXT: retl # encoding: [0xc3] +; +; X64-LABEL: test_int_x86_avx10_vpdpbuud_512: +; X64: # %bb.0: +; X64-NEXT: vpdpbuud %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x74,0x48,0x50,0xc2] +; X64-NEXT: retq # encoding: [0xc3] + %res = call <16 x i32> @llvm.x86.avx10.vpdpbuud.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) + ret <16 x i32> %res +} + +declare <16 x i32> @llvm.x86.avx10.vpdpbuuds.512(<16 x i32>, <16 x i32>, <16 x i32>) + +define <16 x i32>@test_int_x86_avx10_vpdpbuuds_512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) { +; X86-LABEL: test_int_x86_avx10_vpdpbuuds_512: +; X86: # %bb.0: +; X86-NEXT: vpdpbuuds %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x74,0x48,0x51,0xc2] +; X86-NEXT: retl # encoding: [0xc3] +; +; X64-LABEL: test_int_x86_avx10_vpdpbuuds_512: +; X64: # %bb.0: +; X64-NEXT: vpdpbuuds %zmm2, %zmm1, %zmm0 # encoding: [0x62,0xf2,0x74,0x48,0x51,0xc2] +; X64-NEXT: retq # encoding: [0xc3] + %res = call <16 x i32> @llvm.x86.avx10.vpdpbuuds.512(<16 x i32> %x0, <16 x i32> %x1, <16 x i32> %x2) + ret <16 x i32> %res +} diff --git a/llvm/test/CodeGen/X86/avx10_2_512ni-intrinsics.ll b/llvm/test/CodeGen/X86/avx10_2_512ni-intrinsics.ll index 09eb53f..a2aad60 100644 --- a/llvm/test/CodeGen/X86/avx10_2_512ni-intrinsics.ll +++ b/llvm/test/CodeGen/X86/avx10_2_512ni-intrinsics.ll @@ -53,7 +53,7 @@ declare <16 x float> @llvm.x86.avx10.vdpphps.512(<16 x float>, <32 x half>, <32 ; VNNI INT8 -define <16 x i32> @test_mm512_dpbssd_epi32(<16 x i32> %__W, <16 x i32> %__A, ptr %pB) { +define <16 x i32> @test_mm512_dpbssd_epi32(<16 x i32> %__W, <64 x i8> %__A, ptr %pB) { ; X86-LABEL: test_mm512_dpbssd_epi32: ; X86: # %bb.0: ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04] @@ -64,12 +64,12 @@ define <16 x i32> @test_mm512_dpbssd_epi32(<16 x i32> %__W, <16 x i32> %__A, ptr ; X64: # %bb.0: ; X64-NEXT: vpdpbssd (%rdi), %zmm1, %zmm0 # encoding: [0x62,0xf2,0x77,0x48,0x50,0x07] ; X64-NEXT: retq # encoding: [0xc3] - %__B = load <16 x i32>, ptr %pB - %res = tail call <16 x i32> @llvm.x86.avx10.vpdpbssd.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) + %__B = load <64 x i8>, ptr %pB + %res = tail call <16 x i32> @llvm.x86.avx10.vpdpbssd.512(<16 x i32> %__W, <64 x i8> %__A, <64 x i8> %__B) ret <16 x i32> %res } -define <16 x i32> @test_mm512_mask_dpbssds_epi32(<16 x i32> %__W, i16 zeroext %__U, <16 x i32> %__A, <16 x i32> %__B) { +define <16 x i32> @test_mm512_mask_dpbssds_epi32(<16 x i32> %__W, i16 zeroext %__U, <64 x i8> %__A, <64 x i8> %__B) { ; X86-LABEL: test_mm512_mask_dpbssds_epi32: ; X86: # %bb.0: ; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04] @@ -81,13 +81,13 @@ define <16 x i32> @test_mm512_mask_dpbssds_epi32(<16 x i32> %__W, i16 zeroext %_ ; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf] ; X64-NEXT: vpdpbssds %zmm2, %zmm1, %zmm0 {%k1} # encoding: [0x62,0xf2,0x77,0x49,0x51,0xc2] ; X64-NEXT: retq # encoding: [0xc3] - %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpbssds.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) + %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpbssds.512(<16 x i32> %__W, <64 x i8> %__A, <64 x i8> %__B) %bst = bitcast i16 %__U to <16 x i1> %res = select <16 x i1> %bst, <16 x i32> %dpi, <16 x i32> %__W ret <16 x i32> %res } -define <16 x i32> @test_mm512_maskz_dpbssd_epi32(i16 zeroext %__U, <16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) { +define <16 x i32> @test_mm512_maskz_dpbssd_epi32(i16 zeroext %__U, <16 x i32> %__W, <64 x i8> %__A, <64 x i8> %__B) { ; X86-LABEL: test_mm512_maskz_dpbssd_epi32: ; X86: # %bb.0: ; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04] @@ -99,16 +99,16 @@ define <16 x i32> @test_mm512_maskz_dpbssd_epi32(i16 zeroext %__U, <16 x i32> %_ ; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf] ; X64-NEXT: vpdpbssd %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x77,0xc9,0x50,0xc2] ; X64-NEXT: retq # encoding: [0xc3] - %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpbssd.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) + %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpbssd.512(<16 x i32> %__W, <64 x i8> %__A, <64 x i8> %__B) %bst = bitcast i16 %__U to <16 x i1> %res = select <16 x i1> %bst, <16 x i32> %dpi, <16 x i32> zeroinitializer ret <16 x i32> %res } -declare <16 x i32> @llvm.x86.avx10.vpdpbssd.512(<16 x i32>, <16 x i32>, <16 x i32>) -declare <16 x i32> @llvm.x86.avx10.vpdpbssds.512(<16 x i32>, <16 x i32>, <16 x i32>) +declare <16 x i32> @llvm.x86.avx10.vpdpbssd.512(<16 x i32>, <64 x i8>, <64 x i8>) +declare <16 x i32> @llvm.x86.avx10.vpdpbssds.512(<16 x i32>, <64 x i8>, <64 x i8>) -define <16 x i32> @test_mm512_dpbsud_epi32(<16 x i32> %__W, <16 x i32> %__A, ptr %pB) { +define <16 x i32> @test_mm512_dpbsud_epi32(<16 x i32> %__W, <64 x i8> %__A, ptr %pB) { ; X86-LABEL: test_mm512_dpbsud_epi32: ; X86: # %bb.0: ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04] @@ -119,12 +119,12 @@ define <16 x i32> @test_mm512_dpbsud_epi32(<16 x i32> %__W, <16 x i32> %__A, ptr ; X64: # %bb.0: ; X64-NEXT: vpdpbsud (%rdi), %zmm1, %zmm0 # encoding: [0x62,0xf2,0x76,0x48,0x50,0x07] ; X64-NEXT: retq # encoding: [0xc3] - %__B = load <16 x i32>, ptr %pB - %res = tail call <16 x i32> @llvm.x86.avx10.vpdpbsud.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) + %__B = load <64 x i8>, ptr %pB + %res = tail call <16 x i32> @llvm.x86.avx10.vpdpbsud.512(<16 x i32> %__W, <64 x i8> %__A, <64 x i8> %__B) ret <16 x i32> %res } -define <16 x i32> @test_mm512_mask_dpbsuds_epi32(<16 x i32> %__W, i16 zeroext %__U, <16 x i32> %__A, <16 x i32> %__B) { +define <16 x i32> @test_mm512_mask_dpbsuds_epi32(<16 x i32> %__W, i16 zeroext %__U, <64 x i8> %__A, <64 x i8> %__B) { ; X86-LABEL: test_mm512_mask_dpbsuds_epi32: ; X86: # %bb.0: ; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04] @@ -136,13 +136,13 @@ define <16 x i32> @test_mm512_mask_dpbsuds_epi32(<16 x i32> %__W, i16 zeroext %_ ; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf] ; X64-NEXT: vpdpbsuds %zmm2, %zmm1, %zmm0 {%k1} # encoding: [0x62,0xf2,0x76,0x49,0x51,0xc2] ; X64-NEXT: retq # encoding: [0xc3] - %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpbsuds.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) + %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpbsuds.512(<16 x i32> %__W, <64 x i8> %__A, <64 x i8> %__B) %bst = bitcast i16 %__U to <16 x i1> %res = select <16 x i1> %bst, <16 x i32> %dpi, <16 x i32> %__W ret <16 x i32> %res } -define <16 x i32> @test_mm512_maskz_dpbsud_epi32(i16 zeroext %__U, <16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) { +define <16 x i32> @test_mm512_maskz_dpbsud_epi32(i16 zeroext %__U, <16 x i32> %__W, <64 x i8> %__A, <64 x i8> %__B) { ; X86-LABEL: test_mm512_maskz_dpbsud_epi32: ; X86: # %bb.0: ; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04] @@ -154,16 +154,16 @@ define <16 x i32> @test_mm512_maskz_dpbsud_epi32(i16 zeroext %__U, <16 x i32> %_ ; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf] ; X64-NEXT: vpdpbsud %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x76,0xc9,0x50,0xc2] ; X64-NEXT: retq # encoding: [0xc3] - %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpbsud.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) + %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpbsud.512(<16 x i32> %__W, <64 x i8> %__A, <64 x i8> %__B) %bst = bitcast i16 %__U to <16 x i1> %res = select <16 x i1> %bst, <16 x i32> %dpi, <16 x i32> zeroinitializer ret <16 x i32> %res } -declare <16 x i32> @llvm.x86.avx10.vpdpbsud.512(<16 x i32>, <16 x i32>, <16 x i32>) -declare <16 x i32> @llvm.x86.avx10.vpdpbsuds.512(<16 x i32>, <16 x i32>, <16 x i32>) +declare <16 x i32> @llvm.x86.avx10.vpdpbsud.512(<16 x i32>, <64 x i8>, <64 x i8>) +declare <16 x i32> @llvm.x86.avx10.vpdpbsuds.512(<16 x i32>, <64 x i8>, <64 x i8>) -define <16 x i32> @test_mm512_dpbuud_epi32(<16 x i32> %__W, <16 x i32> %__A, ptr %pB) { +define <16 x i32> @test_mm512_dpbuud_epi32(<16 x i32> %__W, <64 x i8> %__A, ptr %pB) { ; X86-LABEL: test_mm512_dpbuud_epi32: ; X86: # %bb.0: ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04] @@ -174,12 +174,12 @@ define <16 x i32> @test_mm512_dpbuud_epi32(<16 x i32> %__W, <16 x i32> %__A, ptr ; X64: # %bb.0: ; X64-NEXT: vpdpbuud (%rdi), %zmm1, %zmm0 # encoding: [0x62,0xf2,0x74,0x48,0x50,0x07] ; X64-NEXT: retq # encoding: [0xc3] - %__B = load <16 x i32>, ptr %pB - %res = tail call <16 x i32> @llvm.x86.avx10.vpdpbuud.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) + %__B = load <64 x i8>, ptr %pB + %res = tail call <16 x i32> @llvm.x86.avx10.vpdpbuud.512(<16 x i32> %__W, <64 x i8> %__A, <64 x i8> %__B) ret <16 x i32> %res } -define <16 x i32> @test_mm512_mask_dpbuuds_epi32(<16 x i32> %__W, i16 zeroext %__U, <16 x i32> %__A, <16 x i32> %__B) { +define <16 x i32> @test_mm512_mask_dpbuuds_epi32(<16 x i32> %__W, i16 zeroext %__U, <64 x i8> %__A, <64 x i8> %__B) { ; X86-LABEL: test_mm512_mask_dpbuuds_epi32: ; X86: # %bb.0: ; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04] @@ -191,13 +191,13 @@ define <16 x i32> @test_mm512_mask_dpbuuds_epi32(<16 x i32> %__W, i16 zeroext %_ ; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf] ; X64-NEXT: vpdpbuuds %zmm2, %zmm1, %zmm0 {%k1} # encoding: [0x62,0xf2,0x74,0x49,0x51,0xc2] ; X64-NEXT: retq # encoding: [0xc3] - %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpbuuds.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) + %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpbuuds.512(<16 x i32> %__W, <64 x i8> %__A, <64 x i8> %__B) %bst = bitcast i16 %__U to <16 x i1> %res = select <16 x i1> %bst, <16 x i32> %dpi, <16 x i32> %__W ret <16 x i32> %res } -define <16 x i32> @test_mm512_maskz_dpbuud_epi32(i16 zeroext %__U, <16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) { +define <16 x i32> @test_mm512_maskz_dpbuud_epi32(i16 zeroext %__U, <16 x i32> %__W, <64 x i8> %__A, <64 x i8> %__B) { ; X86-LABEL: test_mm512_maskz_dpbuud_epi32: ; X86: # %bb.0: ; X86-NEXT: kmovw {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf8,0x90,0x4c,0x24,0x04] @@ -209,14 +209,14 @@ define <16 x i32> @test_mm512_maskz_dpbuud_epi32(i16 zeroext %__U, <16 x i32> %_ ; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf] ; X64-NEXT: vpdpbuud %zmm2, %zmm1, %zmm0 {%k1} {z} # encoding: [0x62,0xf2,0x74,0xc9,0x50,0xc2] ; X64-NEXT: retq # encoding: [0xc3] - %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpbuud.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) + %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpbuud.512(<16 x i32> %__W, <64 x i8> %__A, <64 x i8> %__B) %bst = bitcast i16 %__U to <16 x i1> %res = select <16 x i1> %bst, <16 x i32> %dpi, <16 x i32> zeroinitializer ret <16 x i32> %res } -declare <16 x i32> @llvm.x86.avx10.vpdpbuud.512(<16 x i32>, <16 x i32>, <16 x i32>) -declare <16 x i32> @llvm.x86.avx10.vpdpbuuds.512(<16 x i32>, <16 x i32>, <16 x i32>) +declare <16 x i32> @llvm.x86.avx10.vpdpbuud.512(<16 x i32>, <64 x i8>, <64 x i8>) +declare <16 x i32> @llvm.x86.avx10.vpdpbuuds.512(<16 x i32>, <64 x i8>, <64 x i8>) ; VNNI INT16 diff --git a/llvm/test/CodeGen/X86/avx10_2ni-intrinsics.ll b/llvm/test/CodeGen/X86/avx10_2ni-intrinsics.ll index 0c5fd3b..1f270d5 100644 --- a/llvm/test/CodeGen/X86/avx10_2ni-intrinsics.ll +++ b/llvm/test/CodeGen/X86/avx10_2ni-intrinsics.ll @@ -101,7 +101,7 @@ declare <8 x float> @llvm.x86.avx10.vdpphps.256(<8 x float>, <16 x half>, <16 x ; VNNI INT8 -define <4 x i32> @test_mm_mask_dpbssd_epi32(<4 x i32> %__W, i4 zeroext %__U, <4 x i32> %__A, <4 x i32> %__B) { +define <4 x i32> @test_mm_mask_dpbssd_epi32(<4 x i32> %__W, i4 zeroext %__U, <16 x i8> %__A, <16 x i8> %__B) { ; X86-LABEL: test_mm_mask_dpbssd_epi32: ; X86: # %bb.0: ; X86-NEXT: kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04] @@ -113,13 +113,13 @@ define <4 x i32> @test_mm_mask_dpbssd_epi32(<4 x i32> %__W, i4 zeroext %__U, <4 ; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf] ; X64-NEXT: vpdpbssd %xmm2, %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0x77,0x09,0x50,0xc2] ; X64-NEXT: retq # encoding: [0xc3] - %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B) + %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32> %__W, <16 x i8> %__A, <16 x i8> %__B) %bst = bitcast i4 %__U to <4 x i1> %res = select <4 x i1> %bst, <4 x i32> %dpi, <4 x i32> %__W ret <4 x i32> %res } -define <4 x i32> @test_mm_maskz_dpbssds_epi32(i4 zeroext %__U, <4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B) { +define <4 x i32> @test_mm_maskz_dpbssds_epi32(i4 zeroext %__U, <4 x i32> %__W, <16 x i8> %__A, <16 x i8> %__B) { ; X86-LABEL: test_mm_maskz_dpbssds_epi32: ; X86: # %bb.0: ; X86-NEXT: kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04] @@ -131,13 +131,13 @@ define <4 x i32> @test_mm_maskz_dpbssds_epi32(i4 zeroext %__U, <4 x i32> %__W, < ; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf] ; X64-NEXT: vpdpbssds %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x77,0x89,0x51,0xc2] ; X64-NEXT: retq # encoding: [0xc3] - %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B) + %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32> %__W, <16 x i8> %__A, <16 x i8> %__B) %bst = bitcast i4 %__U to <4 x i1> %res = select <4 x i1> %bst, <4 x i32> %dpi, <4 x i32> zeroinitializer ret <4 x i32> %res } -define <8 x i32> @test_mm256_maskz_dpbssds_epi32(<8 x i32> %__W, i8 zeroext %__U, <8 x i32> %__A, <8 x i32> %__B) { +define <8 x i32> @test_mm256_maskz_dpbssds_epi32(<8 x i32> %__W, i8 zeroext %__U, <32 x i8> %__A, <32 x i8> %__B) { ; X86-LABEL: test_mm256_maskz_dpbssds_epi32: ; X86: # %bb.0: ; X86-NEXT: kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04] @@ -149,13 +149,13 @@ define <8 x i32> @test_mm256_maskz_dpbssds_epi32(<8 x i32> %__W, i8 zeroext %__U ; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf] ; X64-NEXT: vpdpbssds %ymm2, %ymm1, %ymm0 {%k1} # encoding: [0x62,0xf2,0x77,0x29,0x51,0xc2] ; X64-NEXT: retq # encoding: [0xc3] - %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B) + %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32> %__W, <32 x i8> %__A, <32 x i8> %__B) %bst = bitcast i8 %__U to <8 x i1> %res = select <8 x i1> %bst, <8 x i32> %dpi, <8 x i32> %__W ret <8 x i32> %res } -define <8 x i32> @test_mm256_mask_dpbssd_epi32(i8 zeroext %__U, <8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B) { +define <8 x i32> @test_mm256_mask_dpbssd_epi32(i8 zeroext %__U, <8 x i32> %__W, <32 x i8> %__A, <32 x i8> %__B) { ; X86-LABEL: test_mm256_mask_dpbssd_epi32: ; X86: # %bb.0: ; X86-NEXT: kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04] @@ -167,18 +167,18 @@ define <8 x i32> @test_mm256_mask_dpbssd_epi32(i8 zeroext %__U, <8 x i32> %__W, ; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf] ; X64-NEXT: vpdpbssd %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x77,0xa9,0x50,0xc2] ; X64-NEXT: retq # encoding: [0xc3] - %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B) + %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32> %__W, <32 x i8> %__A, <32 x i8> %__B) %bst = bitcast i8 %__U to <8 x i1> %res = select <8 x i1> %bst, <8 x i32> %dpi, <8 x i32> zeroinitializer ret <8 x i32> %res } -declare <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32>, <4 x i32>, <4 x i32>) -declare <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32>, <4 x i32>, <4 x i32>) -declare <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32>, <8 x i32>, <8 x i32>) -declare <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32>, <8 x i32>, <8 x i32>) +declare <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32>, <16 x i8>, <16 x i8>) +declare <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32>, <16 x i8>, <16 x i8>) +declare <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32>, <32 x i8>, <32 x i8>) +declare <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32>, <32 x i8>, <32 x i8>) -define <4 x i32> @test_mm_mask_dpbsud_epi32(<4 x i32> %__W, i4 zeroext %__U, <4 x i32> %__A, <4 x i32> %__B) { +define <4 x i32> @test_mm_mask_dpbsud_epi32(<4 x i32> %__W, i4 zeroext %__U, <16 x i8> %__A, <16 x i8> %__B) { ; X86-LABEL: test_mm_mask_dpbsud_epi32: ; X86: # %bb.0: ; X86-NEXT: kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04] @@ -190,13 +190,13 @@ define <4 x i32> @test_mm_mask_dpbsud_epi32(<4 x i32> %__W, i4 zeroext %__U, <4 ; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf] ; X64-NEXT: vpdpbsud %xmm2, %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0x76,0x09,0x50,0xc2] ; X64-NEXT: retq # encoding: [0xc3] - %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B) + %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32> %__W, <16 x i8> %__A, <16 x i8> %__B) %bst = bitcast i4 %__U to <4 x i1> %res = select <4 x i1> %bst, <4 x i32> %dpi, <4 x i32> %__W ret <4 x i32> %res } -define <4 x i32> @test_mm_maskz_dpbsuds_epi32(i4 zeroext %__U, <4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B) { +define <4 x i32> @test_mm_maskz_dpbsuds_epi32(i4 zeroext %__U, <4 x i32> %__W, <16 x i8> %__A, <16 x i8> %__B) { ; X86-LABEL: test_mm_maskz_dpbsuds_epi32: ; X86: # %bb.0: ; X86-NEXT: kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04] @@ -208,13 +208,13 @@ define <4 x i32> @test_mm_maskz_dpbsuds_epi32(i4 zeroext %__U, <4 x i32> %__W, < ; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf] ; X64-NEXT: vpdpbsuds %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x76,0x89,0x51,0xc2] ; X64-NEXT: retq # encoding: [0xc3] - %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B) + %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32> %__W, <16 x i8> %__A, <16 x i8> %__B) %bst = bitcast i4 %__U to <4 x i1> %res = select <4 x i1> %bst, <4 x i32> %dpi, <4 x i32> zeroinitializer ret <4 x i32> %res } -define <8 x i32> @test_mm256_maskz_dpbsuds_epi32(<8 x i32> %__W, i8 zeroext %__U, <8 x i32> %__A, <8 x i32> %__B) { +define <8 x i32> @test_mm256_maskz_dpbsuds_epi32(<8 x i32> %__W, i8 zeroext %__U, <32 x i8> %__A, <32 x i8> %__B) { ; X86-LABEL: test_mm256_maskz_dpbsuds_epi32: ; X86: # %bb.0: ; X86-NEXT: kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04] @@ -226,13 +226,13 @@ define <8 x i32> @test_mm256_maskz_dpbsuds_epi32(<8 x i32> %__W, i8 zeroext %__U ; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf] ; X64-NEXT: vpdpbsuds %ymm2, %ymm1, %ymm0 {%k1} # encoding: [0x62,0xf2,0x76,0x29,0x51,0xc2] ; X64-NEXT: retq # encoding: [0xc3] - %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B) + %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32> %__W, <32 x i8> %__A, <32 x i8> %__B) %bst = bitcast i8 %__U to <8 x i1> %res = select <8 x i1> %bst, <8 x i32> %dpi, <8 x i32> %__W ret <8 x i32> %res } -define <8 x i32> @test_mm256_mask_dpbsud_epi32(i8 zeroext %__U, <8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B) { +define <8 x i32> @test_mm256_mask_dpbsud_epi32(i8 zeroext %__U, <8 x i32> %__W, <32 x i8> %__A, <32 x i8> %__B) { ; X86-LABEL: test_mm256_mask_dpbsud_epi32: ; X86: # %bb.0: ; X86-NEXT: kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04] @@ -244,18 +244,18 @@ define <8 x i32> @test_mm256_mask_dpbsud_epi32(i8 zeroext %__U, <8 x i32> %__W, ; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf] ; X64-NEXT: vpdpbsud %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x76,0xa9,0x50,0xc2] ; X64-NEXT: retq # encoding: [0xc3] - %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B) + %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32> %__W, <32 x i8> %__A, <32 x i8> %__B) %bst = bitcast i8 %__U to <8 x i1> %res = select <8 x i1> %bst, <8 x i32> %dpi, <8 x i32> zeroinitializer ret <8 x i32> %res } -declare <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32>, <4 x i32>, <4 x i32>) -declare <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32>, <4 x i32>, <4 x i32>) -declare <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32>, <8 x i32>, <8 x i32>) -declare <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32>, <8 x i32>, <8 x i32>) +declare <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32>, <16 x i8>, <16 x i8>) +declare <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32>, <16 x i8>, <16 x i8>) +declare <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32>, <32 x i8>, <32 x i8>) +declare <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32>, <32 x i8>, <32 x i8>) -define <4 x i32> @test_mm_mask_dpbuud_epi32(<4 x i32> %__W, i4 zeroext %__U, <4 x i32> %__A, <4 x i32> %__B) { +define <4 x i32> @test_mm_mask_dpbuud_epi32(<4 x i32> %__W, i4 zeroext %__U, <16 x i8> %__A, <16 x i8> %__B) { ; X86-LABEL: test_mm_mask_dpbuud_epi32: ; X86: # %bb.0: ; X86-NEXT: kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04] @@ -267,13 +267,13 @@ define <4 x i32> @test_mm_mask_dpbuud_epi32(<4 x i32> %__W, i4 zeroext %__U, <4 ; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf] ; X64-NEXT: vpdpbuud %xmm2, %xmm1, %xmm0 {%k1} # encoding: [0x62,0xf2,0x74,0x09,0x50,0xc2] ; X64-NEXT: retq # encoding: [0xc3] - %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B) + %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32> %__W, <16 x i8> %__A, <16 x i8> %__B) %bst = bitcast i4 %__U to <4 x i1> %res = select <4 x i1> %bst, <4 x i32> %dpi, <4 x i32> %__W ret <4 x i32> %res } -define <4 x i32> @test_mm_maskz_dpbuuds_epi32(i4 zeroext %__U, <4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B) { +define <4 x i32> @test_mm_maskz_dpbuuds_epi32(i4 zeroext %__U, <4 x i32> %__W, <16 x i8> %__A, <16 x i8> %__B) { ; X86-LABEL: test_mm_maskz_dpbuuds_epi32: ; X86: # %bb.0: ; X86-NEXT: kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04] @@ -285,13 +285,13 @@ define <4 x i32> @test_mm_maskz_dpbuuds_epi32(i4 zeroext %__U, <4 x i32> %__W, < ; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf] ; X64-NEXT: vpdpbuuds %xmm2, %xmm1, %xmm0 {%k1} {z} # encoding: [0x62,0xf2,0x74,0x89,0x51,0xc2] ; X64-NEXT: retq # encoding: [0xc3] - %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B) + %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32> %__W, <16 x i8> %__A, <16 x i8> %__B) %bst = bitcast i4 %__U to <4 x i1> %res = select <4 x i1> %bst, <4 x i32> %dpi, <4 x i32> zeroinitializer ret <4 x i32> %res } -define <8 x i32> @test_mm256_maskz_dpbuuds_epi32(<8 x i32> %__W, i8 zeroext %__U, <8 x i32> %__A, <8 x i32> %__B) { +define <8 x i32> @test_mm256_maskz_dpbuuds_epi32(<8 x i32> %__W, i8 zeroext %__U, <32 x i8> %__A, <32 x i8> %__B) { ; X86-LABEL: test_mm256_maskz_dpbuuds_epi32: ; X86: # %bb.0: ; X86-NEXT: kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04] @@ -303,13 +303,13 @@ define <8 x i32> @test_mm256_maskz_dpbuuds_epi32(<8 x i32> %__W, i8 zeroext %__U ; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf] ; X64-NEXT: vpdpbuuds %ymm2, %ymm1, %ymm0 {%k1} # encoding: [0x62,0xf2,0x74,0x29,0x51,0xc2] ; X64-NEXT: retq # encoding: [0xc3] - %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B) + %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32> %__W, <32 x i8> %__A, <32 x i8> %__B) %bst = bitcast i8 %__U to <8 x i1> %res = select <8 x i1> %bst, <8 x i32> %dpi, <8 x i32> %__W ret <8 x i32> %res } -define <8 x i32> @test_mm256_mask_dpbuud_epi32(i8 zeroext %__U, <8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B) { +define <8 x i32> @test_mm256_mask_dpbuud_epi32(i8 zeroext %__U, <8 x i32> %__W, <32 x i8> %__A, <32 x i8> %__B) { ; X86-LABEL: test_mm256_mask_dpbuud_epi32: ; X86: # %bb.0: ; X86-NEXT: kmovb {{[0-9]+}}(%esp), %k1 # encoding: [0xc5,0xf9,0x90,0x4c,0x24,0x04] @@ -321,16 +321,16 @@ define <8 x i32> @test_mm256_mask_dpbuud_epi32(i8 zeroext %__U, <8 x i32> %__W, ; X64-NEXT: kmovd %edi, %k1 # encoding: [0xc5,0xfb,0x92,0xcf] ; X64-NEXT: vpdpbuud %ymm2, %ymm1, %ymm0 {%k1} {z} # encoding: [0x62,0xf2,0x74,0xa9,0x50,0xc2] ; X64-NEXT: retq # encoding: [0xc3] - %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B) + %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32> %__W, <32 x i8> %__A, <32 x i8> %__B) %bst = bitcast i8 %__U to <8 x i1> %res = select <8 x i1> %bst, <8 x i32> %dpi, <8 x i32> zeroinitializer ret <8 x i32> %res } -declare <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32>, <4 x i32>, <4 x i32>) -declare <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32>, <4 x i32>, <4 x i32>) -declare <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32>, <8 x i32>, <8 x i32>) -declare <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32>, <8 x i32>, <8 x i32>) +declare <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32>, <16 x i8>, <16 x i8>) +declare <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32>, <16 x i8>, <16 x i8>) +declare <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32>, <32 x i8>, <32 x i8>) +declare <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32>, <32 x i8>, <32 x i8>) ; VNNI INT16 diff --git a/llvm/test/CodeGen/X86/avxvnniint8-intrinsics-upgrade.ll b/llvm/test/CodeGen/X86/avxvnniint8-intrinsics-upgrade.ll new file mode 100644 index 0000000..ce9a0fb --- /dev/null +++ b/llvm/test/CodeGen/X86/avxvnniint8-intrinsics-upgrade.ll @@ -0,0 +1,318 @@ +; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 6 +; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avxvnniint8 --show-mc-encoding | FileCheck %s --check-prefixes=X86 +; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avxvnniint8 --show-mc-encoding | FileCheck %s --check-prefixes=X64 +; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx10.2 --show-mc-encoding | FileCheck %s --check-prefixes=AVX10-X86 +; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx10.2 --show-mc-encoding | FileCheck %s --check-prefixes=AVX10-X64 + +declare <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32>, <4 x i32>, <4 x i32>) + +define <4 x i32>@test_int_x86_avx2_vpdpbssd_128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) { +; X86-LABEL: test_int_x86_avx2_vpdpbssd_128: +; X86: # %bb.0: +; X86-NEXT: vpdpbssd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x73,0x50,0xc2] +; X86-NEXT: retl # encoding: [0xc3] +; +; X64-LABEL: test_int_x86_avx2_vpdpbssd_128: +; X64: # %bb.0: +; X64-NEXT: vpdpbssd %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x73,0x50,0xc2] +; X64-NEXT: retq # encoding: [0xc3] +; +; AVX10-X86-LABEL: test_int_x86_avx2_vpdpbssd_128: +; AVX10-X86: # %bb.0: +; AVX10-X86-NEXT: vpdpbssd %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x77,0x08,0x50,0xc2] +; AVX10-X86-NEXT: retl # encoding: [0xc3] +; +; AVX10-X64-LABEL: test_int_x86_avx2_vpdpbssd_128: +; AVX10-X64: # %bb.0: +; AVX10-X64-NEXT: vpdpbssd %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x77,0x08,0x50,0xc2] +; AVX10-X64-NEXT: retq # encoding: [0xc3] + %res = call <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) + ret <4 x i32> %res +} + +declare <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32>, <4 x i32>, <4 x i32>) + +define <4 x i32>@test_int_x86_avx2_vpdpbssds_128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) { +; X86-LABEL: test_int_x86_avx2_vpdpbssds_128: +; X86: # %bb.0: +; X86-NEXT: vpdpbssds %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x73,0x51,0xc2] +; X86-NEXT: retl # encoding: [0xc3] +; +; X64-LABEL: test_int_x86_avx2_vpdpbssds_128: +; X64: # %bb.0: +; X64-NEXT: vpdpbssds %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x73,0x51,0xc2] +; X64-NEXT: retq # encoding: [0xc3] +; +; AVX10-X86-LABEL: test_int_x86_avx2_vpdpbssds_128: +; AVX10-X86: # %bb.0: +; AVX10-X86-NEXT: vpdpbssds %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x77,0x08,0x51,0xc2] +; AVX10-X86-NEXT: retl # encoding: [0xc3] +; +; AVX10-X64-LABEL: test_int_x86_avx2_vpdpbssds_128: +; AVX10-X64: # %bb.0: +; AVX10-X64-NEXT: vpdpbssds %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x77,0x08,0x51,0xc2] +; AVX10-X64-NEXT: retq # encoding: [0xc3] + %res = call <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) + ret <4 x i32> %res +} + +declare <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32>, <8 x i32>, <8 x i32>) + +define <8 x i32>@test_int_x86_avx2_vpdpbssd_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) { +; X86-LABEL: test_int_x86_avx2_vpdpbssd_256: +; X86: # %bb.0: +; X86-NEXT: vpdpbssd %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x77,0x50,0xc2] +; X86-NEXT: retl # encoding: [0xc3] +; +; X64-LABEL: test_int_x86_avx2_vpdpbssd_256: +; X64: # %bb.0: +; X64-NEXT: vpdpbssd %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x77,0x50,0xc2] +; X64-NEXT: retq # encoding: [0xc3] +; +; AVX10-X86-LABEL: test_int_x86_avx2_vpdpbssd_256: +; AVX10-X86: # %bb.0: +; AVX10-X86-NEXT: vpdpbssd %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x77,0x28,0x50,0xc2] +; AVX10-X86-NEXT: retl # encoding: [0xc3] +; +; AVX10-X64-LABEL: test_int_x86_avx2_vpdpbssd_256: +; AVX10-X64: # %bb.0: +; AVX10-X64-NEXT: vpdpbssd %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x77,0x28,0x50,0xc2] +; AVX10-X64-NEXT: retq # encoding: [0xc3] + %res = call <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) + ret <8 x i32> %res +} + +declare <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32>, <8 x i32>, <8 x i32>) + +define <8 x i32>@test_int_x86_avx2_vpdpbssds_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) { +; X86-LABEL: test_int_x86_avx2_vpdpbssds_256: +; X86: # %bb.0: +; X86-NEXT: vpdpbssds %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x77,0x51,0xc2] +; X86-NEXT: retl # encoding: [0xc3] +; +; X64-LABEL: test_int_x86_avx2_vpdpbssds_256: +; X64: # %bb.0: +; X64-NEXT: vpdpbssds %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x77,0x51,0xc2] +; X64-NEXT: retq # encoding: [0xc3] +; +; AVX10-X86-LABEL: test_int_x86_avx2_vpdpbssds_256: +; AVX10-X86: # %bb.0: +; AVX10-X86-NEXT: vpdpbssds %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x77,0x28,0x51,0xc2] +; AVX10-X86-NEXT: retl # encoding: [0xc3] +; +; AVX10-X64-LABEL: test_int_x86_avx2_vpdpbssds_256: +; AVX10-X64: # %bb.0: +; AVX10-X64-NEXT: vpdpbssds %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x77,0x28,0x51,0xc2] +; AVX10-X64-NEXT: retq # encoding: [0xc3] + %res = call <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) + ret <8 x i32> %res +} + +declare <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32>, <4 x i32>, <4 x i32>) + +define <4 x i32>@test_int_x86_avx2_vpdpbsud_128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) { +; X86-LABEL: test_int_x86_avx2_vpdpbsud_128: +; X86: # %bb.0: +; X86-NEXT: vpdpbsud %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x72,0x50,0xc2] +; X86-NEXT: retl # encoding: [0xc3] +; +; X64-LABEL: test_int_x86_avx2_vpdpbsud_128: +; X64: # %bb.0: +; X64-NEXT: vpdpbsud %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x72,0x50,0xc2] +; X64-NEXT: retq # encoding: [0xc3] +; +; AVX10-X86-LABEL: test_int_x86_avx2_vpdpbsud_128: +; AVX10-X86: # %bb.0: +; AVX10-X86-NEXT: vpdpbsud %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x76,0x08,0x50,0xc2] +; AVX10-X86-NEXT: retl # encoding: [0xc3] +; +; AVX10-X64-LABEL: test_int_x86_avx2_vpdpbsud_128: +; AVX10-X64: # %bb.0: +; AVX10-X64-NEXT: vpdpbsud %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x76,0x08,0x50,0xc2] +; AVX10-X64-NEXT: retq # encoding: [0xc3] + %res = call <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) + ret <4 x i32> %res +} + +declare <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32>, <4 x i32>, <4 x i32>) + +define <4 x i32>@test_int_x86_avx2_vpdpbsuds_128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) { +; X86-LABEL: test_int_x86_avx2_vpdpbsuds_128: +; X86: # %bb.0: +; X86-NEXT: vpdpbsuds %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x72,0x51,0xc2] +; X86-NEXT: retl # encoding: [0xc3] +; +; X64-LABEL: test_int_x86_avx2_vpdpbsuds_128: +; X64: # %bb.0: +; X64-NEXT: vpdpbsuds %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x72,0x51,0xc2] +; X64-NEXT: retq # encoding: [0xc3] +; +; AVX10-X86-LABEL: test_int_x86_avx2_vpdpbsuds_128: +; AVX10-X86: # %bb.0: +; AVX10-X86-NEXT: vpdpbsuds %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x76,0x08,0x51,0xc2] +; AVX10-X86-NEXT: retl # encoding: [0xc3] +; +; AVX10-X64-LABEL: test_int_x86_avx2_vpdpbsuds_128: +; AVX10-X64: # %bb.0: +; AVX10-X64-NEXT: vpdpbsuds %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x76,0x08,0x51,0xc2] +; AVX10-X64-NEXT: retq # encoding: [0xc3] + %res = call <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) + ret <4 x i32> %res +} + +declare <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32>, <8 x i32>, <8 x i32>) + +define <8 x i32>@test_int_x86_avx2_vpdpbsud_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) { +; X86-LABEL: test_int_x86_avx2_vpdpbsud_256: +; X86: # %bb.0: +; X86-NEXT: vpdpbsud %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x76,0x50,0xc2] +; X86-NEXT: retl # encoding: [0xc3] +; +; X64-LABEL: test_int_x86_avx2_vpdpbsud_256: +; X64: # %bb.0: +; X64-NEXT: vpdpbsud %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x76,0x50,0xc2] +; X64-NEXT: retq # encoding: [0xc3] +; +; AVX10-X86-LABEL: test_int_x86_avx2_vpdpbsud_256: +; AVX10-X86: # %bb.0: +; AVX10-X86-NEXT: vpdpbsud %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x76,0x28,0x50,0xc2] +; AVX10-X86-NEXT: retl # encoding: [0xc3] +; +; AVX10-X64-LABEL: test_int_x86_avx2_vpdpbsud_256: +; AVX10-X64: # %bb.0: +; AVX10-X64-NEXT: vpdpbsud %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x76,0x28,0x50,0xc2] +; AVX10-X64-NEXT: retq # encoding: [0xc3] + %res = call <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) + ret <8 x i32> %res +} + +declare <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32>, <8 x i32>, <8 x i32>) + +define <8 x i32>@test_int_x86_avx2_vpdpbsuds_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) { +; X86-LABEL: test_int_x86_avx2_vpdpbsuds_256: +; X86: # %bb.0: +; X86-NEXT: vpdpbsuds %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x76,0x51,0xc2] +; X86-NEXT: retl # encoding: [0xc3] +; +; X64-LABEL: test_int_x86_avx2_vpdpbsuds_256: +; X64: # %bb.0: +; X64-NEXT: vpdpbsuds %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x76,0x51,0xc2] +; X64-NEXT: retq # encoding: [0xc3] +; +; AVX10-X86-LABEL: test_int_x86_avx2_vpdpbsuds_256: +; AVX10-X86: # %bb.0: +; AVX10-X86-NEXT: vpdpbsuds %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x76,0x28,0x51,0xc2] +; AVX10-X86-NEXT: retl # encoding: [0xc3] +; +; AVX10-X64-LABEL: test_int_x86_avx2_vpdpbsuds_256: +; AVX10-X64: # %bb.0: +; AVX10-X64-NEXT: vpdpbsuds %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x76,0x28,0x51,0xc2] +; AVX10-X64-NEXT: retq # encoding: [0xc3] + %res = call <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) + ret <8 x i32> %res +} + +declare <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32>, <4 x i32>, <4 x i32>) + +define <4 x i32>@test_int_x86_avx2_vpdpbuud(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) { +; X86-LABEL: test_int_x86_avx2_vpdpbuud: +; X86: # %bb.0: +; X86-NEXT: vpdpbuud %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x70,0x50,0xc2] +; X86-NEXT: retl # encoding: [0xc3] +; +; X64-LABEL: test_int_x86_avx2_vpdpbuud: +; X64: # %bb.0: +; X64-NEXT: vpdpbuud %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x70,0x50,0xc2] +; X64-NEXT: retq # encoding: [0xc3] +; +; AVX10-X86-LABEL: test_int_x86_avx2_vpdpbuud: +; AVX10-X86: # %bb.0: +; AVX10-X86-NEXT: vpdpbuud %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x74,0x08,0x50,0xc2] +; AVX10-X86-NEXT: retl # encoding: [0xc3] +; +; AVX10-X64-LABEL: test_int_x86_avx2_vpdpbuud: +; AVX10-X64: # %bb.0: +; AVX10-X64-NEXT: vpdpbuud %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x74,0x08,0x50,0xc2] +; AVX10-X64-NEXT: retq # encoding: [0xc3] + %res = call <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) + ret <4 x i32> %res +} + +declare <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32>, <4 x i32>, <4 x i32>) + +define <4 x i32>@test_int_x86_avx2_vpdpbuuds_128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) { +; X86-LABEL: test_int_x86_avx2_vpdpbuuds_128: +; X86: # %bb.0: +; X86-NEXT: vpdpbuuds %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x70,0x51,0xc2] +; X86-NEXT: retl # encoding: [0xc3] +; +; X64-LABEL: test_int_x86_avx2_vpdpbuuds_128: +; X64: # %bb.0: +; X64-NEXT: vpdpbuuds %xmm2, %xmm1, %xmm0 # encoding: [0xc4,0xe2,0x70,0x51,0xc2] +; X64-NEXT: retq # encoding: [0xc3] +; +; AVX10-X86-LABEL: test_int_x86_avx2_vpdpbuuds_128: +; AVX10-X86: # %bb.0: +; AVX10-X86-NEXT: vpdpbuuds %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x74,0x08,0x51,0xc2] +; AVX10-X86-NEXT: retl # encoding: [0xc3] +; +; AVX10-X64-LABEL: test_int_x86_avx2_vpdpbuuds_128: +; AVX10-X64: # %bb.0: +; AVX10-X64-NEXT: vpdpbuuds %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x74,0x08,0x51,0xc2] +; AVX10-X64-NEXT: retq # encoding: [0xc3] + %res = call <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) + ret <4 x i32> %res +} + +declare <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32>, <8 x i32>, <8 x i32>) + +define <8 x i32>@test_int_x86_avx2_vpdpbuud_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) { +; X86-LABEL: test_int_x86_avx2_vpdpbuud_256: +; X86: # %bb.0: +; X86-NEXT: vpdpbuud %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x74,0x50,0xc2] +; X86-NEXT: retl # encoding: [0xc3] +; +; X64-LABEL: test_int_x86_avx2_vpdpbuud_256: +; X64: # %bb.0: +; X64-NEXT: vpdpbuud %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x74,0x50,0xc2] +; X64-NEXT: retq # encoding: [0xc3] +; +; AVX10-X86-LABEL: test_int_x86_avx2_vpdpbuud_256: +; AVX10-X86: # %bb.0: +; AVX10-X86-NEXT: vpdpbuud %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x74,0x28,0x50,0xc2] +; AVX10-X86-NEXT: retl # encoding: [0xc3] +; +; AVX10-X64-LABEL: test_int_x86_avx2_vpdpbuud_256: +; AVX10-X64: # %bb.0: +; AVX10-X64-NEXT: vpdpbuud %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x74,0x28,0x50,0xc2] +; AVX10-X64-NEXT: retq # encoding: [0xc3] + %res = call <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) + ret <8 x i32> %res +} + +declare <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32>, <8 x i32>, <8 x i32>) + +define <8 x i32>@test_int_x86_avx2_vpdpbuuds_256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) { +; X86-LABEL: test_int_x86_avx2_vpdpbuuds_256: +; X86: # %bb.0: +; X86-NEXT: vpdpbuuds %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x74,0x51,0xc2] +; X86-NEXT: retl # encoding: [0xc3] +; +; X64-LABEL: test_int_x86_avx2_vpdpbuuds_256: +; X64: # %bb.0: +; X64-NEXT: vpdpbuuds %ymm2, %ymm1, %ymm0 # encoding: [0xc4,0xe2,0x74,0x51,0xc2] +; X64-NEXT: retq # encoding: [0xc3] +; +; AVX10-X86-LABEL: test_int_x86_avx2_vpdpbuuds_256: +; AVX10-X86: # %bb.0: +; AVX10-X86-NEXT: vpdpbuuds %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x74,0x28,0x51,0xc2] +; AVX10-X86-NEXT: retl # encoding: [0xc3] +; +; AVX10-X64-LABEL: test_int_x86_avx2_vpdpbuuds_256: +; AVX10-X64: # %bb.0: +; AVX10-X64-NEXT: vpdpbuuds %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x74,0x28,0x51,0xc2] +; AVX10-X64-NEXT: retq # encoding: [0xc3] + %res = call <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) + ret <8 x i32> %res +} + diff --git a/llvm/test/CodeGen/X86/avxvnniint8-intrinsics.ll b/llvm/test/CodeGen/X86/avxvnniint8-intrinsics.ll index 0ddd017..6c3d90aa 100644 --- a/llvm/test/CodeGen/X86/avxvnniint8-intrinsics.ll +++ b/llvm/test/CodeGen/X86/avxvnniint8-intrinsics.ll @@ -5,9 +5,9 @@ ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx10.2 --show-mc-encoding | FileCheck %s --check-prefixes=AVX10-X64 -declare <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32>, <4 x i32>, <4 x i32>) +declare <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32>, <16 x i8>, <16 x i8>) -define <4 x i32>@test_int_x86_avx2_vpdpbssd_128(<4 x i32> %x0, <4 x i32> %x1, ptr %x2p, <4 x i32> %x4) { +define <4 x i32>@test_int_x86_avx2_vpdpbssd_128(<4 x i32> %x0, <16 x i8> %x1, ptr %x2p, <16 x i8> %x4) { ; X86-LABEL: test_int_x86_avx2_vpdpbssd_128: ; X86: # %bb.0: ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04] @@ -41,16 +41,16 @@ define <4 x i32>@test_int_x86_avx2_vpdpbssd_128(<4 x i32> %x0, <4 x i32> %x1, pt ; AVX10-X64-NEXT: vpdpbssd %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x77,0x08,0x50,0xc2] ; AVX10-X64-NEXT: vpaddd %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xfe,0xc0] ; AVX10-X64-NEXT: retq # encoding: [0xc3] - %x2 = load <4 x i32>, ptr %x2p - %1 = call <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) - %2 = call <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x4) + %x2 = load <16 x i8>, ptr %x2p + %1 = call <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x2) + %2 = call <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x4) %res = add <4 x i32> %1, %2 ret <4 x i32> %res } -declare <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32>, <4 x i32>, <4 x i32>) +declare <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32>, <16 x i8>, <16 x i8>) -define <4 x i32>@test_int_x86_avx2_vpdpbssds_128(<4 x i32> %x0, <4 x i32> %x1, ptr %x2p, <4 x i32> %x4) { +define <4 x i32>@test_int_x86_avx2_vpdpbssds_128(<4 x i32> %x0, <16 x i8> %x1, ptr %x2p, <16 x i8> %x4) { ; X86-LABEL: test_int_x86_avx2_vpdpbssds_128: ; X86: # %bb.0: ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04] @@ -84,16 +84,16 @@ define <4 x i32>@test_int_x86_avx2_vpdpbssds_128(<4 x i32> %x0, <4 x i32> %x1, p ; AVX10-X64-NEXT: vpdpbssds %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x77,0x08,0x51,0xc2] ; AVX10-X64-NEXT: vpaddd %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xfe,0xc0] ; AVX10-X64-NEXT: retq # encoding: [0xc3] - %x2 = load <4 x i32>, ptr %x2p - %1 = call <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) - %2 = call <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x4) + %x2 = load <16 x i8>, ptr %x2p + %1 = call <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x2) + %2 = call <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x4) %res = add <4 x i32> %1, %2 ret <4 x i32> %res } -declare <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32>, <8 x i32>, <8 x i32>) +declare <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32>, <32 x i8>, <32 x i8>) -define <8 x i32>@test_int_x86_avx2_vpdpbssd_256(<8 x i32> %x0, <8 x i32> %x1, ptr %x2p, <8 x i32> %x4) { +define <8 x i32>@test_int_x86_avx2_vpdpbssd_256(<8 x i32> %x0, <32 x i8> %x1, ptr %x2p, <32 x i8> %x4) { ; X86-LABEL: test_int_x86_avx2_vpdpbssd_256: ; X86: # %bb.0: ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04] @@ -127,16 +127,16 @@ define <8 x i32>@test_int_x86_avx2_vpdpbssd_256(<8 x i32> %x0, <8 x i32> %x1, pt ; AVX10-X64-NEXT: vpdpbssd %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x77,0x28,0x50,0xc2] ; AVX10-X64-NEXT: vpaddd %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xfe,0xc0] ; AVX10-X64-NEXT: retq # encoding: [0xc3] - %x2 = load <8 x i32>, ptr %x2p - %1 = call <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) - %2 = call <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x4) + %x2 = load <32 x i8>, ptr %x2p + %1 = call <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x2) + %2 = call <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x4) %res = add <8 x i32> %1, %2 ret <8 x i32> %res } -declare <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32>, <8 x i32>, <8 x i32>) +declare <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32>, <32 x i8>, <32 x i8>) -define <8 x i32>@test_int_x86_avx2_vpdpbssds_256(<8 x i32> %x0, <8 x i32> %x1, ptr %x2p, <8 x i32> %x4) { +define <8 x i32>@test_int_x86_avx2_vpdpbssds_256(<8 x i32> %x0, <32 x i8> %x1, ptr %x2p, <32 x i8> %x4) { ; X86-LABEL: test_int_x86_avx2_vpdpbssds_256: ; X86: # %bb.0: ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04] @@ -170,16 +170,16 @@ define <8 x i32>@test_int_x86_avx2_vpdpbssds_256(<8 x i32> %x0, <8 x i32> %x1, p ; AVX10-X64-NEXT: vpdpbssds %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x77,0x28,0x51,0xc2] ; AVX10-X64-NEXT: vpaddd %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xfe,0xc0] ; AVX10-X64-NEXT: retq # encoding: [0xc3] - %x2 = load <8 x i32>, ptr %x2p - %1 = call <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) - %2 = call <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x4) + %x2 = load <32 x i8>, ptr %x2p + %1 = call <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x2) + %2 = call <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x4) %res = add <8 x i32> %1, %2 ret <8 x i32> %res } -declare <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32>, <4 x i32>, <4 x i32>) +declare <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32>, <16 x i8>, <16 x i8>) -define <4 x i32>@test_int_x86_avx2_vpdpbsud_128(<4 x i32> %x0, <4 x i32> %x1, ptr %x2p, <4 x i32> %x4) { +define <4 x i32>@test_int_x86_avx2_vpdpbsud_128(<4 x i32> %x0, <16 x i8> %x1, ptr %x2p, <16 x i8> %x4) { ; X86-LABEL: test_int_x86_avx2_vpdpbsud_128: ; X86: # %bb.0: ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04] @@ -213,16 +213,16 @@ define <4 x i32>@test_int_x86_avx2_vpdpbsud_128(<4 x i32> %x0, <4 x i32> %x1, pt ; AVX10-X64-NEXT: vpdpbsud %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x76,0x08,0x50,0xc2] ; AVX10-X64-NEXT: vpaddd %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xfe,0xc0] ; AVX10-X64-NEXT: retq # encoding: [0xc3] - %x2 = load <4 x i32>, ptr %x2p - %1 = call <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) - %2 = call <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x4) + %x2 = load <16 x i8>, ptr %x2p + %1 = call <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x2) + %2 = call <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x4) %res = add <4 x i32> %1, %2 ret <4 x i32> %res } -declare <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32>, <4 x i32>, <4 x i32>) +declare <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32>, <16 x i8>, <16 x i8>) -define <4 x i32>@test_int_x86_avx2_vpdpbsuds_128(<4 x i32> %x0, <4 x i32> %x1, ptr %x2p, <4 x i32> %x4) { +define <4 x i32>@test_int_x86_avx2_vpdpbsuds_128(<4 x i32> %x0, <16 x i8> %x1, ptr %x2p, <16 x i8> %x4) { ; X86-LABEL: test_int_x86_avx2_vpdpbsuds_128: ; X86: # %bb.0: ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04] @@ -256,16 +256,16 @@ define <4 x i32>@test_int_x86_avx2_vpdpbsuds_128(<4 x i32> %x0, <4 x i32> %x1, p ; AVX10-X64-NEXT: vpdpbsuds %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x76,0x08,0x51,0xc2] ; AVX10-X64-NEXT: vpaddd %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xfe,0xc0] ; AVX10-X64-NEXT: retq # encoding: [0xc3] - %x2 = load <4 x i32>, ptr %x2p - %1 = call <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) - %2 = call <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x4) + %x2 = load <16 x i8>, ptr %x2p + %1 = call <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x2) + %2 = call <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x4) %res = add <4 x i32> %1, %2 ret <4 x i32> %res } -declare <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32>, <8 x i32>, <8 x i32>) +declare <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32>, <32 x i8>, <32 x i8>) -define <8 x i32>@test_int_x86_avx2_vpdpbsud_256(<8 x i32> %x0, <8 x i32> %x1, ptr %x2p, <8 x i32> %x4) { +define <8 x i32>@test_int_x86_avx2_vpdpbsud_256(<8 x i32> %x0, <32 x i8> %x1, ptr %x2p, <32 x i8> %x4) { ; X86-LABEL: test_int_x86_avx2_vpdpbsud_256: ; X86: # %bb.0: ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04] @@ -299,16 +299,16 @@ define <8 x i32>@test_int_x86_avx2_vpdpbsud_256(<8 x i32> %x0, <8 x i32> %x1, pt ; AVX10-X64-NEXT: vpdpbsud %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x76,0x28,0x50,0xc2] ; AVX10-X64-NEXT: vpaddd %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xfe,0xc0] ; AVX10-X64-NEXT: retq # encoding: [0xc3] - %x2 = load <8 x i32>, ptr %x2p - %1 = call <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) - %2 = call <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x4) + %x2 = load <32 x i8>, ptr %x2p + %1 = call <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x2) + %2 = call <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x4) %res = add <8 x i32> %1, %2 ret <8 x i32> %res } -declare <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32>, <8 x i32>, <8 x i32>) +declare <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32>, <32 x i8>, <32 x i8>) -define <8 x i32>@test_int_x86_avx2_vpdpbsuds_256(<8 x i32> %x0, <8 x i32> %x1, ptr %x2p, <8 x i32> %x4) { +define <8 x i32>@test_int_x86_avx2_vpdpbsuds_256(<8 x i32> %x0, <32 x i8> %x1, ptr %x2p, <32 x i8> %x4) { ; X86-LABEL: test_int_x86_avx2_vpdpbsuds_256: ; X86: # %bb.0: ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04] @@ -342,16 +342,16 @@ define <8 x i32>@test_int_x86_avx2_vpdpbsuds_256(<8 x i32> %x0, <8 x i32> %x1, p ; AVX10-X64-NEXT: vpdpbsuds %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x76,0x28,0x51,0xc2] ; AVX10-X64-NEXT: vpaddd %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xfe,0xc0] ; AVX10-X64-NEXT: retq # encoding: [0xc3] - %x2 = load <8 x i32>, ptr %x2p - %1 = call <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) - %2 = call <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x4) + %x2 = load <32 x i8>, ptr %x2p + %1 = call <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x2) + %2 = call <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x4) %res = add <8 x i32> %1, %2 ret <8 x i32> %res } -declare <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32>, <4 x i32>, <4 x i32>) +declare <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32>, <16 x i8>, <16 x i8>) -define <4 x i32>@test_int_x86_avx2_vpdpbuud_128(<4 x i32> %x0, <4 x i32> %x1, ptr %x2p, <4 x i32> %x4) { +define <4 x i32>@test_int_x86_avx2_vpdpbuud_128(<4 x i32> %x0, <16 x i8> %x1, ptr %x2p, <16 x i8> %x4) { ; X86-LABEL: test_int_x86_avx2_vpdpbuud_128: ; X86: # %bb.0: ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04] @@ -385,16 +385,16 @@ define <4 x i32>@test_int_x86_avx2_vpdpbuud_128(<4 x i32> %x0, <4 x i32> %x1, pt ; AVX10-X64-NEXT: vpdpbuud %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x74,0x08,0x50,0xc2] ; AVX10-X64-NEXT: vpaddd %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xfe,0xc0] ; AVX10-X64-NEXT: retq # encoding: [0xc3] - %x2 = load <4 x i32>, ptr %x2p - %1 = call <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) - %2 = call <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x4) + %x2 = load <16 x i8>, ptr %x2p + %1 = call <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x2) + %2 = call <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x4) %res = add <4 x i32> %1, %2 ret <4 x i32> %res } -declare <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32>, <4 x i32>, <4 x i32>) +declare <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32>, <16 x i8>, <16 x i8>) -define <4 x i32>@test_int_x86_avx2_vpdpbuuds_128(<4 x i32> %x0, <4 x i32> %x1, ptr %x2p, <4 x i32> %x4) { +define <4 x i32>@test_int_x86_avx2_vpdpbuuds_128(<4 x i32> %x0, <16 x i8> %x1, ptr %x2p, <16 x i8> %x4) { ; X86-LABEL: test_int_x86_avx2_vpdpbuuds_128: ; X86: # %bb.0: ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04] @@ -428,16 +428,16 @@ define <4 x i32>@test_int_x86_avx2_vpdpbuuds_128(<4 x i32> %x0, <4 x i32> %x1, p ; AVX10-X64-NEXT: vpdpbuuds %xmm2, %xmm1, %xmm0 # encoding: [0x62,0xf2,0x74,0x08,0x51,0xc2] ; AVX10-X64-NEXT: vpaddd %xmm0, %xmm3, %xmm0 # EVEX TO VEX Compression encoding: [0xc5,0xe1,0xfe,0xc0] ; AVX10-X64-NEXT: retq # encoding: [0xc3] - %x2 = load <4 x i32>, ptr %x2p - %1 = call <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) - %2 = call <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x4) + %x2 = load <16 x i8>, ptr %x2p + %1 = call <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x2) + %2 = call <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x4) %res = add <4 x i32> %1, %2 ret <4 x i32> %res } -declare <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32>, <8 x i32>, <8 x i32>) +declare <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32>, <32 x i8>, <32 x i8>) -define <8 x i32>@test_int_x86_avx2_vpdpbuud_256(<8 x i32> %x0, <8 x i32> %x1, ptr %x2p, <8 x i32> %x4) { +define <8 x i32>@test_int_x86_avx2_vpdpbuud_256(<8 x i32> %x0, <32 x i8> %x1, ptr %x2p, <32 x i8> %x4) { ; X86-LABEL: test_int_x86_avx2_vpdpbuud_256: ; X86: # %bb.0: ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04] @@ -471,16 +471,16 @@ define <8 x i32>@test_int_x86_avx2_vpdpbuud_256(<8 x i32> %x0, <8 x i32> %x1, pt ; AVX10-X64-NEXT: vpdpbuud %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x74,0x28,0x50,0xc2] ; AVX10-X64-NEXT: vpaddd %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xfe,0xc0] ; AVX10-X64-NEXT: retq # encoding: [0xc3] - %x2 = load <8 x i32>, ptr %x2p - %1 = call <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) - %2 = call <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x4) + %x2 = load <32 x i8>, ptr %x2p + %1 = call <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x2) + %2 = call <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x4) %res = add <8 x i32> %1, %2 ret <8 x i32> %res } -declare <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32>, <8 x i32>, <8 x i32>) +declare <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32>, <32 x i8>, <32 x i8>) -define <8 x i32>@test_int_x86_avx2_vpdpbuuds_256(<8 x i32> %x0, <8 x i32> %x1, ptr %x2p, <8 x i32> %x4) { +define <8 x i32>@test_int_x86_avx2_vpdpbuuds_256(<8 x i32> %x0, <32 x i8> %x1, ptr %x2p, <32 x i8> %x4) { ; X86-LABEL: test_int_x86_avx2_vpdpbuuds_256: ; X86: # %bb.0: ; X86-NEXT: movl {{[0-9]+}}(%esp), %eax # encoding: [0x8b,0x44,0x24,0x04] @@ -514,9 +514,9 @@ define <8 x i32>@test_int_x86_avx2_vpdpbuuds_256(<8 x i32> %x0, <8 x i32> %x1, p ; AVX10-X64-NEXT: vpdpbuuds %ymm2, %ymm1, %ymm0 # encoding: [0x62,0xf2,0x74,0x28,0x51,0xc2] ; AVX10-X64-NEXT: vpaddd %ymm0, %ymm3, %ymm0 # EVEX TO VEX Compression encoding: [0xc5,0xe5,0xfe,0xc0] ; AVX10-X64-NEXT: retq # encoding: [0xc3] - %x2 = load <8 x i32>, ptr %x2p - %1 = call <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) - %2 = call <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x4) + %x2 = load <32 x i8>, ptr %x2p + %1 = call <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x2) + %2 = call <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x4) %res = add <8 x i32> %1, %2 ret <8 x i32> %res } diff --git a/llvm/test/CodeGen/X86/stack-folding-int-avxvnniint8.ll b/llvm/test/CodeGen/X86/stack-folding-int-avxvnniint8.ll index fd988f7..a49d3a5 100644 --- a/llvm/test/CodeGen/X86/stack-folding-int-avxvnniint8.ll +++ b/llvm/test/CodeGen/X86/stack-folding-int-avxvnniint8.ll @@ -1,20 +1,20 @@ ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py ; RUN: llc -O3 -disable-peephole -mtriple=x86_64-unknown-unknown -mattr=+avxvnniint8 < %s | FileCheck %s -declare <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32>, <4 x i32>, <4 x i32>) -declare <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32>, <4 x i32>, <4 x i32>) -declare <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32>, <8 x i32>, <8 x i32>) -declare <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32>, <8 x i32>, <8 x i32>) -declare <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32>, <4 x i32>, <4 x i32>) -declare <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32>, <4 x i32>, <4 x i32>) -declare <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32>, <8 x i32>, <8 x i32>) -declare <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32>, <8 x i32>, <8 x i32>) -declare <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32>, <4 x i32>, <4 x i32>) -declare <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32>, <4 x i32>, <4 x i32>) -declare <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32>, <8 x i32>, <8 x i32>) -declare <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32>, <8 x i32>, <8 x i32>) +declare <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32>, <16 x i8>, <16 x i8>) +declare <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32>, <16 x i8>, <16 x i8>) +declare <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32>, <32 x i8>, <32 x i8>) +declare <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32>, <32 x i8>, <32 x i8>) +declare <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32>, <16 x i8>, <16 x i8>) +declare <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32>, <16 x i8>, <16 x i8>) +declare <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32>, <32 x i8>, <32 x i8>) +declare <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32>, <32 x i8>, <32 x i8>) +declare <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32>, <16 x i8>, <16 x i8>) +declare <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32>, <16 x i8>, <16 x i8>) +declare <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32>, <32 x i8>, <32 x i8>) +declare <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32>, <32 x i8>, <32 x i8>) -define <4 x i32> @stack_fold_vpdpbssd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) { +define <4 x i32> @stack_fold_vpdpbssd(<4 x i32> %a0, <16 x i8> %a1, <16 x i8> %a2) { ; CHECK-LABEL: stack_fold_vpdpbssd: ; CHECK: # %bb.0: ; CHECK-NEXT: vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill @@ -24,11 +24,11 @@ define <4 x i32> @stack_fold_vpdpbssd(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a ; CHECK-NEXT: vpdpbssd {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm0 # 16-byte Folded Reload ; CHECK-NEXT: retq %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"() - %2 = call <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) + %2 = call <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32> %a0, <16 x i8> %a1, <16 x i8> %a2) ret <4 x i32> %2 } -define <4 x i32> @stack_fold_vpdpbssd_commuted(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) { +define <4 x i32> @stack_fold_vpdpbssd_commuted(<4 x i32> %a0, <16 x i8> %a1, <16 x i8> %a2) { ; CHECK-LABEL: stack_fold_vpdpbssd_commuted: ; CHECK: # %bb.0: ; CHECK-NEXT: vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill @@ -38,11 +38,11 @@ define <4 x i32> @stack_fold_vpdpbssd_commuted(<4 x i32> %a0, <4 x i32> %a1, <4 ; CHECK-NEXT: vpdpbssd {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm0 # 16-byte Folded Reload ; CHECK-NEXT: retq %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"() - %2 = call <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32> %a0, <4 x i32> %a2, <4 x i32> %a1) + %2 = call <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32> %a0, <16 x i8> %a2, <16 x i8> %a1) ret <4 x i32> %2 } -define <8 x i32> @stack_fold_vpdpbssd_256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) { +define <8 x i32> @stack_fold_vpdpbssd_256(<8 x i32> %a0, <32 x i8> %a1, <32 x i8> %a2) { ; CHECK-LABEL: stack_fold_vpdpbssd_256: ; CHECK: # %bb.0: ; CHECK-NEXT: vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill @@ -52,11 +52,11 @@ define <8 x i32> @stack_fold_vpdpbssd_256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32 ; CHECK-NEXT: vpdpbssd {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm0 # 32-byte Folded Reload ; CHECK-NEXT: retq %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"() - %2 = call <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) + %2 = call <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32> %a0, <32 x i8> %a1, <32 x i8> %a2) ret <8 x i32> %2 } -define <8 x i32> @stack_fold_vpdpbssd_256_commuted(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) { +define <8 x i32> @stack_fold_vpdpbssd_256_commuted(<8 x i32> %a0, <32 x i8> %a1, <32 x i8> %a2) { ; CHECK-LABEL: stack_fold_vpdpbssd_256_commuted: ; CHECK: # %bb.0: ; CHECK-NEXT: vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill @@ -66,11 +66,11 @@ define <8 x i32> @stack_fold_vpdpbssd_256_commuted(<8 x i32> %a0, <8 x i32> %a1, ; CHECK-NEXT: vpdpbssd {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm0 # 32-byte Folded Reload ; CHECK-NEXT: retq %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"() - %2 = call <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32> %a0, <8 x i32> %a2, <8 x i32> %a1) + %2 = call <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32> %a0, <32 x i8> %a2, <32 x i8> %a1) ret <8 x i32> %2 } -define <4 x i32> @stack_fold_vpdpbssds(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) { +define <4 x i32> @stack_fold_vpdpbssds(<4 x i32> %a0, <16 x i8> %a1, <16 x i8> %a2) { ; CHECK-LABEL: stack_fold_vpdpbssds: ; CHECK: # %bb.0: ; CHECK-NEXT: vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill @@ -80,11 +80,11 @@ define <4 x i32> @stack_fold_vpdpbssds(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> % ; CHECK-NEXT: vpdpbssds {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm0 # 16-byte Folded Reload ; CHECK-NEXT: retq %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"() - %2 = call <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) + %2 = call <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32> %a0, <16 x i8> %a1, <16 x i8> %a2) ret <4 x i32> %2 } -define <4 x i32> @stack_fold_vpdpbssds_commuted(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) { +define <4 x i32> @stack_fold_vpdpbssds_commuted(<4 x i32> %a0, <16 x i8> %a1, <16 x i8> %a2) { ; CHECK-LABEL: stack_fold_vpdpbssds_commuted: ; CHECK: # %bb.0: ; CHECK-NEXT: vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill @@ -94,11 +94,11 @@ define <4 x i32> @stack_fold_vpdpbssds_commuted(<4 x i32> %a0, <4 x i32> %a1, <4 ; CHECK-NEXT: vpdpbssds {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm0 # 16-byte Folded Reload ; CHECK-NEXT: retq %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"() - %2 = call <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32> %a0, <4 x i32> %a2, <4 x i32> %a1) + %2 = call <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32> %a0, <16 x i8> %a2, <16 x i8> %a1) ret <4 x i32> %2 } -define <8 x i32> @stack_fold_vpdpbssds_256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) { +define <8 x i32> @stack_fold_vpdpbssds_256(<8 x i32> %a0, <32 x i8> %a1, <32 x i8> %a2) { ; CHECK-LABEL: stack_fold_vpdpbssds_256: ; CHECK: # %bb.0: ; CHECK-NEXT: vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill @@ -108,11 +108,11 @@ define <8 x i32> @stack_fold_vpdpbssds_256(<8 x i32> %a0, <8 x i32> %a1, <8 x i3 ; CHECK-NEXT: vpdpbssds {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm0 # 32-byte Folded Reload ; CHECK-NEXT: retq %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"() - %2 = call <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) + %2 = call <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32> %a0, <32 x i8> %a1, <32 x i8> %a2) ret <8 x i32> %2 } -define <8 x i32> @stack_fold_vpdpbssds_256_commuted(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) { +define <8 x i32> @stack_fold_vpdpbssds_256_commuted(<8 x i32> %a0, <32 x i8> %a1, <32 x i8> %a2) { ; CHECK-LABEL: stack_fold_vpdpbssds_256_commuted: ; CHECK: # %bb.0: ; CHECK-NEXT: vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill @@ -122,11 +122,11 @@ define <8 x i32> @stack_fold_vpdpbssds_256_commuted(<8 x i32> %a0, <8 x i32> %a1 ; CHECK-NEXT: vpdpbssds {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm0 # 32-byte Folded Reload ; CHECK-NEXT: retq %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"() - %2 = call <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32> %a0, <8 x i32> %a2, <8 x i32> %a1) + %2 = call <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32> %a0, <32 x i8> %a2, <32 x i8> %a1) ret <8 x i32> %2 } -define <4 x i32> @stack_fold_vpdpbsud(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) { +define <4 x i32> @stack_fold_vpdpbsud(<4 x i32> %a0, <16 x i8> %a1, <16 x i8> %a2) { ; CHECK-LABEL: stack_fold_vpdpbsud: ; CHECK: # %bb.0: ; CHECK-NEXT: vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill @@ -136,11 +136,11 @@ define <4 x i32> @stack_fold_vpdpbsud(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a ; CHECK-NEXT: vpdpbsud {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm0 # 16-byte Folded Reload ; CHECK-NEXT: retq %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"() - %2 = call <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) + %2 = call <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32> %a0, <16 x i8> %a1, <16 x i8> %a2) ret <4 x i32> %2 } -define <4 x i32> @stack_fold_vpdpbsud_commuted(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) { +define <4 x i32> @stack_fold_vpdpbsud_commuted(<4 x i32> %a0, <16 x i8> %a1, <16 x i8> %a2) { ; CHECK-LABEL: stack_fold_vpdpbsud_commuted: ; CHECK: # %bb.0: ; CHECK-NEXT: vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill @@ -151,11 +151,11 @@ define <4 x i32> @stack_fold_vpdpbsud_commuted(<4 x i32> %a0, <4 x i32> %a1, <4 ; CHECK-NEXT: vpdpbsud %xmm1, %xmm2, %xmm0 ; CHECK-NEXT: retq %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"() - %2 = call <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32> %a0, <4 x i32> %a2, <4 x i32> %a1) + %2 = call <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32> %a0, <16 x i8> %a2, <16 x i8> %a1) ret <4 x i32> %2 } -define <8 x i32> @stack_fold_vpdpbsud_256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) { +define <8 x i32> @stack_fold_vpdpbsud_256(<8 x i32> %a0, <32 x i8> %a1, <32 x i8> %a2) { ; CHECK-LABEL: stack_fold_vpdpbsud_256: ; CHECK: # %bb.0: ; CHECK-NEXT: vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill @@ -165,11 +165,11 @@ define <8 x i32> @stack_fold_vpdpbsud_256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32 ; CHECK-NEXT: vpdpbsud {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm0 # 32-byte Folded Reload ; CHECK-NEXT: retq %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"() - %2 = call <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) + %2 = call <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32> %a0, <32 x i8> %a1, <32 x i8> %a2) ret <8 x i32> %2 } -define <8 x i32> @stack_fold_vpdpbsud_256_commuted(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) { +define <8 x i32> @stack_fold_vpdpbsud_256_commuted(<8 x i32> %a0, <32 x i8> %a1, <32 x i8> %a2) { ; CHECK-LABEL: stack_fold_vpdpbsud_256_commuted: ; CHECK: # %bb.0: ; CHECK-NEXT: vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill @@ -180,11 +180,11 @@ define <8 x i32> @stack_fold_vpdpbsud_256_commuted(<8 x i32> %a0, <8 x i32> %a1, ; CHECK-NEXT: vpdpbsud %ymm1, %ymm2, %ymm0 ; CHECK-NEXT: retq %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"() - %2 = call <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32> %a0, <8 x i32> %a2, <8 x i32> %a1) + %2 = call <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32> %a0, <32 x i8> %a2, <32 x i8> %a1) ret <8 x i32> %2 } -define <4 x i32> @stack_fold_vpdpbsuds(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) { +define <4 x i32> @stack_fold_vpdpbsuds(<4 x i32> %a0, <16 x i8> %a1, <16 x i8> %a2) { ; CHECK-LABEL: stack_fold_vpdpbsuds: ; CHECK: # %bb.0: ; CHECK-NEXT: vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill @@ -194,11 +194,11 @@ define <4 x i32> @stack_fold_vpdpbsuds(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> % ; CHECK-NEXT: vpdpbsuds {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm0 # 16-byte Folded Reload ; CHECK-NEXT: retq %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"() - %2 = call <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) + %2 = call <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32> %a0, <16 x i8> %a1, <16 x i8> %a2) ret <4 x i32> %2 } -define <4 x i32> @stack_fold_vpdpbsuds_commuted(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) { +define <4 x i32> @stack_fold_vpdpbsuds_commuted(<4 x i32> %a0, <16 x i8> %a1, <16 x i8> %a2) { ; CHECK-LABEL: stack_fold_vpdpbsuds_commuted: ; CHECK: # %bb.0: ; CHECK-NEXT: vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill @@ -209,11 +209,11 @@ define <4 x i32> @stack_fold_vpdpbsuds_commuted(<4 x i32> %a0, <4 x i32> %a1, <4 ; CHECK-NEXT: vpdpbsuds %xmm1, %xmm2, %xmm0 ; CHECK-NEXT: retq %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"() - %2 = call <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32> %a0, <4 x i32> %a2, <4 x i32> %a1) + %2 = call <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32> %a0, <16 x i8> %a2, <16 x i8> %a1) ret <4 x i32> %2 } -define <8 x i32> @stack_fold_vpdpbsuds_256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) { +define <8 x i32> @stack_fold_vpdpbsuds_256(<8 x i32> %a0, <32 x i8> %a1, <32 x i8> %a2) { ; CHECK-LABEL: stack_fold_vpdpbsuds_256: ; CHECK: # %bb.0: ; CHECK-NEXT: vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill @@ -223,11 +223,11 @@ define <8 x i32> @stack_fold_vpdpbsuds_256(<8 x i32> %a0, <8 x i32> %a1, <8 x i3 ; CHECK-NEXT: vpdpbsuds {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm0 # 32-byte Folded Reload ; CHECK-NEXT: retq %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"() - %2 = call <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) + %2 = call <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32> %a0, <32 x i8> %a1, <32 x i8> %a2) ret <8 x i32> %2 } -define <8 x i32> @stack_fold_vpdpbsuds_256_commuted(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) { +define <8 x i32> @stack_fold_vpdpbsuds_256_commuted(<8 x i32> %a0, <32 x i8> %a1, <32 x i8> %a2) { ; CHECK-LABEL: stack_fold_vpdpbsuds_256_commuted: ; CHECK: # %bb.0: ; CHECK-NEXT: vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill @@ -238,11 +238,11 @@ define <8 x i32> @stack_fold_vpdpbsuds_256_commuted(<8 x i32> %a0, <8 x i32> %a1 ; CHECK-NEXT: vpdpbsuds %ymm1, %ymm2, %ymm0 ; CHECK-NEXT: retq %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"() - %2 = call <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32> %a0, <8 x i32> %a2, <8 x i32> %a1) + %2 = call <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32> %a0, <32 x i8> %a2, <32 x i8> %a1) ret <8 x i32> %2 } -define <4 x i32> @stack_fold_vpdpbuud(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) { +define <4 x i32> @stack_fold_vpdpbuud(<4 x i32> %a0, <16 x i8> %a1, <16 x i8> %a2) { ; CHECK-LABEL: stack_fold_vpdpbuud: ; CHECK: # %bb.0: ; CHECK-NEXT: vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill @@ -252,11 +252,11 @@ define <4 x i32> @stack_fold_vpdpbuud(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a ; CHECK-NEXT: vpdpbuud {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm0 # 16-byte Folded Reload ; CHECK-NEXT: retq %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"() - %2 = call <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) + %2 = call <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32> %a0, <16 x i8> %a1, <16 x i8> %a2) ret <4 x i32> %2 } -define <4 x i32> @stack_fold_vpdpbuud_commuted(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) { +define <4 x i32> @stack_fold_vpdpbuud_commuted(<4 x i32> %a0, <16 x i8> %a1, <16 x i8> %a2) { ; CHECK-LABEL: stack_fold_vpdpbuud_commuted: ; CHECK: # %bb.0: ; CHECK-NEXT: vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill @@ -266,11 +266,11 @@ define <4 x i32> @stack_fold_vpdpbuud_commuted(<4 x i32> %a0, <4 x i32> %a1, <4 ; CHECK-NEXT: vpdpbuud {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm0 # 16-byte Folded Reload ; CHECK-NEXT: retq %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"() - %2 = call <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32> %a0, <4 x i32> %a2, <4 x i32> %a1) + %2 = call <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32> %a0, <16 x i8> %a2, <16 x i8> %a1) ret <4 x i32> %2 } -define <8 x i32> @stack_fold_vpdpbuud_256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) { +define <8 x i32> @stack_fold_vpdpbuud_256(<8 x i32> %a0, <32 x i8> %a1, <32 x i8> %a2) { ; CHECK-LABEL: stack_fold_vpdpbuud_256: ; CHECK: # %bb.0: ; CHECK-NEXT: vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill @@ -280,11 +280,11 @@ define <8 x i32> @stack_fold_vpdpbuud_256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32 ; CHECK-NEXT: vpdpbuud {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm0 # 32-byte Folded Reload ; CHECK-NEXT: retq %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"() - %2 = call <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) + %2 = call <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32> %a0, <32 x i8> %a1, <32 x i8> %a2) ret <8 x i32> %2 } -define <8 x i32> @stack_fold_vpdpbuud_256_commuted(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) { +define <8 x i32> @stack_fold_vpdpbuud_256_commuted(<8 x i32> %a0, <32 x i8> %a1, <32 x i8> %a2) { ; CHECK-LABEL: stack_fold_vpdpbuud_256_commuted: ; CHECK: # %bb.0: ; CHECK-NEXT: vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill @@ -294,11 +294,11 @@ define <8 x i32> @stack_fold_vpdpbuud_256_commuted(<8 x i32> %a0, <8 x i32> %a1, ; CHECK-NEXT: vpdpbuud {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm0 # 32-byte Folded Reload ; CHECK-NEXT: retq %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"() - %2 = call <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32> %a0, <8 x i32> %a2, <8 x i32> %a1) + %2 = call <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32> %a0, <32 x i8> %a2, <32 x i8> %a1) ret <8 x i32> %2 } -define <4 x i32> @stack_fold_vpdpbuuds(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) { +define <4 x i32> @stack_fold_vpdpbuuds(<4 x i32> %a0, <16 x i8> %a1, <16 x i8> %a2) { ; CHECK-LABEL: stack_fold_vpdpbuuds: ; CHECK: # %bb.0: ; CHECK-NEXT: vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill @@ -308,11 +308,11 @@ define <4 x i32> @stack_fold_vpdpbuuds(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> % ; CHECK-NEXT: vpdpbuuds {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm0 # 16-byte Folded Reload ; CHECK-NEXT: retq %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"() - %2 = call <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) + %2 = call <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32> %a0, <16 x i8> %a1, <16 x i8> %a2) ret <4 x i32> %2 } -define <4 x i32> @stack_fold_vpdpbuuds_commuted(<4 x i32> %a0, <4 x i32> %a1, <4 x i32> %a2) { +define <4 x i32> @stack_fold_vpdpbuuds_commuted(<4 x i32> %a0, <16 x i8> %a1, <16 x i8> %a2) { ; CHECK-LABEL: stack_fold_vpdpbuuds_commuted: ; CHECK: # %bb.0: ; CHECK-NEXT: vmovaps %xmm2, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill @@ -322,11 +322,11 @@ define <4 x i32> @stack_fold_vpdpbuuds_commuted(<4 x i32> %a0, <4 x i32> %a1, <4 ; CHECK-NEXT: vpdpbuuds {{[-0-9]+}}(%r{{[sb]}}p), %xmm1, %xmm0 # 16-byte Folded Reload ; CHECK-NEXT: retq %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"() - %2 = call <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32> %a0, <4 x i32> %a2, <4 x i32> %a1) + %2 = call <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32> %a0, <16 x i8> %a2, <16 x i8> %a1) ret <4 x i32> %2 } -define <8 x i32> @stack_fold_vpdpbuuds_256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) { +define <8 x i32> @stack_fold_vpdpbuuds_256(<8 x i32> %a0, <32 x i8> %a1, <32 x i8> %a2) { ; CHECK-LABEL: stack_fold_vpdpbuuds_256: ; CHECK: # %bb.0: ; CHECK-NEXT: vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill @@ -336,11 +336,11 @@ define <8 x i32> @stack_fold_vpdpbuuds_256(<8 x i32> %a0, <8 x i32> %a1, <8 x i3 ; CHECK-NEXT: vpdpbuuds {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm0 # 32-byte Folded Reload ; CHECK-NEXT: retq %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"() - %2 = call <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) + %2 = call <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32> %a0, <32 x i8> %a1, <32 x i8> %a2) ret <8 x i32> %2 } -define <8 x i32> @stack_fold_vpdpbuuds_256_commuted(<8 x i32> %a0, <8 x i32> %a1, <8 x i32> %a2) { +define <8 x i32> @stack_fold_vpdpbuuds_256_commuted(<8 x i32> %a0, <32 x i8> %a1, <32 x i8> %a2) { ; CHECK-LABEL: stack_fold_vpdpbuuds_256_commuted: ; CHECK: # %bb.0: ; CHECK-NEXT: vmovups %ymm2, {{[-0-9]+}}(%r{{[sb]}}p) # 32-byte Spill @@ -350,6 +350,6 @@ define <8 x i32> @stack_fold_vpdpbuuds_256_commuted(<8 x i32> %a0, <8 x i32> %a1 ; CHECK-NEXT: vpdpbuuds {{[-0-9]+}}(%r{{[sb]}}p), %ymm1, %ymm0 # 32-byte Folded Reload ; CHECK-NEXT: retq %1 = tail call <2 x i64> asm sideeffect "nop", "=x,~{xmm3},~{xmm4},~{xmm5},~{xmm6},~{xmm7},~{xmm8},~{xmm9},~{xmm10},~{xmm11},~{xmm12},~{xmm13},~{xmm14},~{xmm15},~{flags}"() - %2 = call <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32> %a0, <8 x i32> %a2, <8 x i32> %a1) + %2 = call <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32> %a0, <32 x i8> %a2, <32 x i8> %a1) ret <8 x i32> %2 } diff --git a/llvm/test/Instrumentation/MemorySanitizer/X86/avx10_2_512ni-intrinsics.ll b/llvm/test/Instrumentation/MemorySanitizer/X86/avx10_2_512ni-intrinsics.ll index 93006ae..991467e 100644 --- a/llvm/test/Instrumentation/MemorySanitizer/X86/avx10_2_512ni-intrinsics.ll +++ b/llvm/test/Instrumentation/MemorySanitizer/X86/avx10_2_512ni-intrinsics.ll @@ -124,11 +124,11 @@ define <16 x float> @test_mm512_maskz_dpph_ps(i16 zeroext %__U, <16 x float> %__ declare <16 x float> @llvm.x86.avx10.vdpphps.512(<16 x float>, <32 x half>, <32 x half>) -define <16 x i32> @test_mm512_dpbssd_epi32(<16 x i32> %__W, <16 x i32> %__A, ptr %pB) sanitize_memory { +define <16 x i32> @test_mm512_dpbssd_epi32(<16 x i32> %__W, <64 x i8> %__A, ptr %pB) sanitize_memory { ; CHECK-LABEL: define <16 x i32> @test_mm512_dpbssd_epi32( -; CHECK-SAME: <16 x i32> [[__W:%.*]], <16 x i32> [[__A:%.*]], ptr [[PB:%.*]]) #[[ATTR0]] { +; CHECK-SAME: <16 x i32> [[__W:%.*]], <64 x i8> [[__A:%.*]], ptr [[PB:%.*]]) #[[ATTR0]] { ; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8 +; CHECK-NEXT: [[TMP12:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8 ; CHECK-NEXT: [[TMP4:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8 ; CHECK-NEXT: call void @llvm.donothing() ; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0 @@ -137,22 +137,18 @@ define <16 x i32> @test_mm512_dpbssd_epi32(<16 x i32> %__W, <16 x i32> %__A, ptr ; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR5]] ; CHECK-NEXT: unreachable ; CHECK: [[BB5]]: -; CHECK-NEXT: [[__B:%.*]] = load <16 x i32>, ptr [[PB]], align 64 +; CHECK-NEXT: [[TMP10:%.*]] = load <64 x i8>, ptr [[PB]], align 64 ; CHECK-NEXT: [[TMP6:%.*]] = ptrtoint ptr [[PB]] to i64 ; CHECK-NEXT: [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080 ; CHECK-NEXT: [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr -; CHECK-NEXT: [[_MSLD:%.*]] = load <16 x i32>, ptr [[TMP8]], align 64 -; CHECK-NEXT: [[TMP9:%.*]] = bitcast <16 x i32> [[__A]] to <64 x i8> -; CHECK-NEXT: [[TMP10:%.*]] = bitcast <16 x i32> [[__B]] to <64 x i8> -; CHECK-NEXT: [[TMP11:%.*]] = bitcast <16 x i32> [[TMP3]] to <64 x i8> -; CHECK-NEXT: [[TMP12:%.*]] = bitcast <16 x i32> [[_MSLD]] to <64 x i8> -; CHECK-NEXT: [[TMP13:%.*]] = icmp ne <64 x i8> [[TMP11]], zeroinitializer +; CHECK-NEXT: [[TMP9:%.*]] = load <64 x i8>, ptr [[TMP8]], align 64 ; CHECK-NEXT: [[TMP14:%.*]] = icmp ne <64 x i8> [[TMP12]], zeroinitializer ; CHECK-NEXT: [[TMP15:%.*]] = icmp ne <64 x i8> [[TMP9]], zeroinitializer +; CHECK-NEXT: [[TMP11:%.*]] = icmp ne <64 x i8> [[__A]], zeroinitializer ; CHECK-NEXT: [[TMP16:%.*]] = icmp ne <64 x i8> [[TMP10]], zeroinitializer -; CHECK-NEXT: [[TMP17:%.*]] = and <64 x i1> [[TMP13]], [[TMP14]] -; CHECK-NEXT: [[TMP18:%.*]] = and <64 x i1> [[TMP15]], [[TMP14]] -; CHECK-NEXT: [[TMP19:%.*]] = and <64 x i1> [[TMP13]], [[TMP16]] +; CHECK-NEXT: [[TMP17:%.*]] = and <64 x i1> [[TMP14]], [[TMP15]] +; CHECK-NEXT: [[TMP18:%.*]] = and <64 x i1> [[TMP11]], [[TMP15]] +; CHECK-NEXT: [[TMP19:%.*]] = and <64 x i1> [[TMP14]], [[TMP16]] ; CHECK-NEXT: [[TMP20:%.*]] = or <64 x i1> [[TMP17]], [[TMP18]] ; CHECK-NEXT: [[TMP21:%.*]] = or <64 x i1> [[TMP20]], [[TMP19]] ; CHECK-NEXT: [[TMP22:%.*]] = sext <64 x i1> [[TMP21]] to <64 x i8> @@ -160,34 +156,30 @@ define <16 x i32> @test_mm512_dpbssd_epi32(<16 x i32> %__W, <16 x i32> %__A, ptr ; CHECK-NEXT: [[TMP24:%.*]] = icmp ne <16 x i32> [[TMP23]], zeroinitializer ; CHECK-NEXT: [[TMP27:%.*]] = sext <16 x i1> [[TMP24]] to <16 x i32> ; CHECK-NEXT: [[_MSPROP1:%.*]] = or <16 x i32> [[TMP27]], [[TMP4]] -; CHECK-NEXT: [[RES:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpbssd.512(<16 x i32> [[__W]], <16 x i32> [[__A]], <16 x i32> [[__B]]) +; CHECK-NEXT: [[RES:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpbssd.512(<16 x i32> [[__W]], <64 x i8> [[__A]], <64 x i8> [[TMP10]]) ; CHECK-NEXT: store <16 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <16 x i32> [[RES]] ; - %__B = load <16 x i32>, ptr %pB - %res = tail call <16 x i32> @llvm.x86.avx10.vpdpbssd.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) + %__B = load <64 x i8>, ptr %pB + %res = tail call <16 x i32> @llvm.x86.avx10.vpdpbssd.512(<16 x i32> %__W, <64 x i8> %__A, <64 x i8> %__B) ret <16 x i32> %res } -define <16 x i32> @test_mm512_mask_dpbssds_epi32(<16 x i32> %__W, i16 zeroext %__U, <16 x i32> %__A, <16 x i32> %__B) sanitize_memory { +define <16 x i32> @test_mm512_mask_dpbssds_epi32(<16 x i32> %__W, i16 zeroext %__U, <64 x i8> %__A, <64 x i8> %__B) sanitize_memory { ; CHECK-LABEL: define <16 x i32> @test_mm512_mask_dpbssds_epi32( -; CHECK-SAME: <16 x i32> [[__W:%.*]], i16 zeroext [[__U:%.*]], <16 x i32> [[__A:%.*]], <16 x i32> [[__B:%.*]]) #[[ATTR0]] { -; CHECK-NEXT: [[TMP2:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8 +; CHECK-SAME: <16 x i32> [[__W:%.*]], i16 zeroext [[__U:%.*]], <64 x i8> [[__A:%.*]], <64 x i8> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[TMP24:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 +; CHECK-NEXT: [[TMP25:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8 ; CHECK-NEXT: [[TMP1:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8 ; CHECK-NEXT: [[TMP4:%.*]] = load i16, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8 ; CHECK-NEXT: call void @llvm.donothing() -; CHECK-NEXT: [[TMP24:%.*]] = bitcast <16 x i32> [[__A]] to <64 x i8> -; CHECK-NEXT: [[TMP25:%.*]] = bitcast <16 x i32> [[__B]] to <64 x i8> -; CHECK-NEXT: [[TMP26:%.*]] = bitcast <16 x i32> [[TMP2]] to <64 x i8> -; CHECK-NEXT: [[TMP27:%.*]] = bitcast <16 x i32> [[TMP3]] to <64 x i8> -; CHECK-NEXT: [[TMP28:%.*]] = icmp ne <64 x i8> [[TMP26]], zeroinitializer -; CHECK-NEXT: [[TMP10:%.*]] = icmp ne <64 x i8> [[TMP27]], zeroinitializer ; CHECK-NEXT: [[TMP11:%.*]] = icmp ne <64 x i8> [[TMP24]], zeroinitializer ; CHECK-NEXT: [[TMP12:%.*]] = icmp ne <64 x i8> [[TMP25]], zeroinitializer -; CHECK-NEXT: [[TMP13:%.*]] = and <64 x i1> [[TMP28]], [[TMP10]] -; CHECK-NEXT: [[TMP14:%.*]] = and <64 x i1> [[TMP11]], [[TMP10]] -; CHECK-NEXT: [[TMP15:%.*]] = and <64 x i1> [[TMP28]], [[TMP12]] +; CHECK-NEXT: [[TMP10:%.*]] = icmp ne <64 x i8> [[__A]], zeroinitializer +; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <64 x i8> [[__B]], zeroinitializer +; CHECK-NEXT: [[TMP13:%.*]] = and <64 x i1> [[TMP11]], [[TMP12]] +; CHECK-NEXT: [[TMP14:%.*]] = and <64 x i1> [[TMP10]], [[TMP12]] +; CHECK-NEXT: [[TMP15:%.*]] = and <64 x i1> [[TMP11]], [[TMP21]] ; CHECK-NEXT: [[TMP16:%.*]] = or <64 x i1> [[TMP13]], [[TMP14]] ; CHECK-NEXT: [[TMP17:%.*]] = or <64 x i1> [[TMP16]], [[TMP15]] ; CHECK-NEXT: [[TMP18:%.*]] = sext <64 x i1> [[TMP17]] to <64 x i8> @@ -195,7 +187,7 @@ define <16 x i32> @test_mm512_mask_dpbssds_epi32(<16 x i32> %__W, i16 zeroext %_ ; CHECK-NEXT: [[TMP20:%.*]] = icmp ne <16 x i32> [[TMP19]], zeroinitializer ; CHECK-NEXT: [[TMP23:%.*]] = sext <16 x i1> [[TMP20]] to <16 x i32> ; CHECK-NEXT: [[_MSPROP1:%.*]] = or <16 x i32> [[TMP23]], [[TMP1]] -; CHECK-NEXT: [[DPI:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpbssds.512(<16 x i32> [[__W]], <16 x i32> [[__A]], <16 x i32> [[__B]]) +; CHECK-NEXT: [[DPI:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpbssds.512(<16 x i32> [[__W]], <64 x i8> [[__A]], <64 x i8> [[__B]]) ; CHECK-NEXT: [[TMP5:%.*]] = bitcast i16 [[TMP4]] to <16 x i1> ; CHECK-NEXT: [[BST:%.*]] = bitcast i16 [[__U]] to <16 x i1> ; CHECK-NEXT: [[TMP6:%.*]] = select <16 x i1> [[BST]], <16 x i32> [[_MSPROP1]], <16 x i32> [[TMP1]] @@ -207,31 +199,27 @@ define <16 x i32> @test_mm512_mask_dpbssds_epi32(<16 x i32> %__W, i16 zeroext %_ ; CHECK-NEXT: store <16 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <16 x i32> [[RES]] ; - %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpbssds.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) + %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpbssds.512(<16 x i32> %__W, <64 x i8> %__A, <64 x i8> %__B) %bst = bitcast i16 %__U to <16 x i1> %res = select <16 x i1> %bst, <16 x i32> %dpi, <16 x i32> %__W ret <16 x i32> %res } -define <16 x i32> @test_mm512_maskz_dpbssd_epi32(i16 zeroext %__U, <16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) sanitize_memory { +define <16 x i32> @test_mm512_maskz_dpbssd_epi32(i16 zeroext %__U, <16 x i32> %__W, <64 x i8> %__A, <64 x i8> %__B) sanitize_memory { ; CHECK-LABEL: define <16 x i32> @test_mm512_maskz_dpbssd_epi32( -; CHECK-SAME: i16 zeroext [[__U:%.*]], <16 x i32> [[__W:%.*]], <16 x i32> [[__A:%.*]], <16 x i32> [[__B:%.*]]) #[[ATTR0]] { -; CHECK-NEXT: [[TMP2:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8 +; CHECK-SAME: i16 zeroext [[__U:%.*]], <16 x i32> [[__W:%.*]], <64 x i8> [[__A:%.*]], <64 x i8> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[TMP25:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 +; CHECK-NEXT: [[TMP26:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8 ; CHECK-NEXT: [[TMP24:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8 ; CHECK-NEXT: [[TMP4:%.*]] = load i16, ptr @__msan_param_tls, align 8 ; CHECK-NEXT: call void @llvm.donothing() -; CHECK-NEXT: [[TMP25:%.*]] = bitcast <16 x i32> [[__A]] to <64 x i8> -; CHECK-NEXT: [[TMP26:%.*]] = bitcast <16 x i32> [[__B]] to <64 x i8> -; CHECK-NEXT: [[TMP27:%.*]] = bitcast <16 x i32> [[TMP2]] to <64 x i8> -; CHECK-NEXT: [[TMP28:%.*]] = bitcast <16 x i32> [[TMP3]] to <64 x i8> -; CHECK-NEXT: [[TMP29:%.*]] = icmp ne <64 x i8> [[TMP27]], zeroinitializer -; CHECK-NEXT: [[TMP10:%.*]] = icmp ne <64 x i8> [[TMP28]], zeroinitializer ; CHECK-NEXT: [[TMP11:%.*]] = icmp ne <64 x i8> [[TMP25]], zeroinitializer ; CHECK-NEXT: [[TMP12:%.*]] = icmp ne <64 x i8> [[TMP26]], zeroinitializer -; CHECK-NEXT: [[TMP13:%.*]] = and <64 x i1> [[TMP29]], [[TMP10]] -; CHECK-NEXT: [[TMP14:%.*]] = and <64 x i1> [[TMP11]], [[TMP10]] -; CHECK-NEXT: [[TMP15:%.*]] = and <64 x i1> [[TMP29]], [[TMP12]] +; CHECK-NEXT: [[TMP10:%.*]] = icmp ne <64 x i8> [[__A]], zeroinitializer +; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <64 x i8> [[__B]], zeroinitializer +; CHECK-NEXT: [[TMP13:%.*]] = and <64 x i1> [[TMP11]], [[TMP12]] +; CHECK-NEXT: [[TMP14:%.*]] = and <64 x i1> [[TMP10]], [[TMP12]] +; CHECK-NEXT: [[TMP15:%.*]] = and <64 x i1> [[TMP11]], [[TMP21]] ; CHECK-NEXT: [[TMP16:%.*]] = or <64 x i1> [[TMP13]], [[TMP14]] ; CHECK-NEXT: [[TMP17:%.*]] = or <64 x i1> [[TMP16]], [[TMP15]] ; CHECK-NEXT: [[TMP18:%.*]] = sext <64 x i1> [[TMP17]] to <64 x i8> @@ -239,7 +227,7 @@ define <16 x i32> @test_mm512_maskz_dpbssd_epi32(i16 zeroext %__U, <16 x i32> %_ ; CHECK-NEXT: [[TMP20:%.*]] = icmp ne <16 x i32> [[TMP19]], zeroinitializer ; CHECK-NEXT: [[TMP23:%.*]] = sext <16 x i1> [[TMP20]] to <16 x i32> ; CHECK-NEXT: [[_MSPROP1:%.*]] = or <16 x i32> [[TMP23]], [[TMP24]] -; CHECK-NEXT: [[DPI:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpbssd.512(<16 x i32> [[__W]], <16 x i32> [[__A]], <16 x i32> [[__B]]) +; CHECK-NEXT: [[DPI:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpbssd.512(<16 x i32> [[__W]], <64 x i8> [[__A]], <64 x i8> [[__B]]) ; CHECK-NEXT: [[TMP5:%.*]] = bitcast i16 [[TMP4]] to <16 x i1> ; CHECK-NEXT: [[BST:%.*]] = bitcast i16 [[__U]] to <16 x i1> ; CHECK-NEXT: [[TMP6:%.*]] = select <16 x i1> [[BST]], <16 x i32> [[_MSPROP1]], <16 x i32> zeroinitializer @@ -251,21 +239,21 @@ define <16 x i32> @test_mm512_maskz_dpbssd_epi32(i16 zeroext %__U, <16 x i32> %_ ; CHECK-NEXT: store <16 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <16 x i32> [[RES]] ; - %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpbssd.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) + %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpbssd.512(<16 x i32> %__W, <64 x i8> %__A, <64 x i8> %__B) %bst = bitcast i16 %__U to <16 x i1> %res = select <16 x i1> %bst, <16 x i32> %dpi, <16 x i32> zeroinitializer ret <16 x i32> %res } -declare <16 x i32> @llvm.x86.avx10.vpdpbssd.512(<16 x i32>, <16 x i32>, <16 x i32>) -declare <16 x i32> @llvm.x86.avx10.vpdpbssds.512(<16 x i32>, <16 x i32>, <16 x i32>) +declare <16 x i32> @llvm.x86.avx10.vpdpbssd.512(<16 x i32>, <64 x i8>, <64 x i8>) +declare <16 x i32> @llvm.x86.avx10.vpdpbssds.512(<16 x i32>, <64 x i8>, <64 x i8>) -define <16 x i32> @test_mm512_dpbsud_epi32(<16 x i32> %__W, <16 x i32> %__A, ptr %pB) sanitize_memory { +define <16 x i32> @test_mm512_dpbsud_epi32(<16 x i32> %__W, <64 x i8> %__A, ptr %pB) sanitize_memory { ; CHECK-LABEL: define <16 x i32> @test_mm512_dpbsud_epi32( -; CHECK-SAME: <16 x i32> [[__W:%.*]], <16 x i32> [[__A:%.*]], ptr [[PB:%.*]]) #[[ATTR0]] { +; CHECK-SAME: <16 x i32> [[__W:%.*]], <64 x i8> [[__A:%.*]], ptr [[PB:%.*]]) #[[ATTR0]] { ; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8 -; CHECK-NEXT: [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8 +; CHECK-NEXT: [[TMP3:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8 +; CHECK-NEXT: [[TMP4:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8 ; CHECK-NEXT: call void @llvm.donothing() ; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0 ; CHECK-NEXT: br i1 [[_MSCMP]], label %[[BB4:.*]], label %[[BB5:.*]], !prof [[PROF1]] @@ -273,87 +261,123 @@ define <16 x i32> @test_mm512_dpbsud_epi32(<16 x i32> %__W, <16 x i32> %__A, ptr ; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR5]] ; CHECK-NEXT: unreachable ; CHECK: [[BB5]]: -; CHECK-NEXT: [[__B:%.*]] = load <16 x i32>, ptr [[PB]], align 64 +; CHECK-NEXT: [[__B:%.*]] = load <64 x i8>, ptr [[PB]], align 64 ; CHECK-NEXT: [[TMP6:%.*]] = ptrtoint ptr [[PB]] to i64 ; CHECK-NEXT: [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080 ; CHECK-NEXT: [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr -; CHECK-NEXT: [[_MSLD:%.*]] = load <16 x i32>, ptr [[TMP8]], align 64 -; CHECK-NEXT: [[_MSPROP:%.*]] = or <16 x i32> [[TMP2]], [[TMP3]] -; CHECK-NEXT: [[_MSPROP1:%.*]] = or <16 x i32> [[_MSPROP]], [[_MSLD]] -; CHECK-NEXT: [[RES:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpbsud.512(<16 x i32> [[__W]], <16 x i32> [[__A]], <16 x i32> [[__B]]) -; CHECK-NEXT: store <16 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8 +; CHECK-NEXT: [[_MSLD:%.*]] = load <64 x i8>, ptr [[TMP8]], align 64 +; CHECK-NEXT: [[TMP9:%.*]] = icmp ne <64 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP10:%.*]] = icmp ne <64 x i8> [[_MSLD]], zeroinitializer +; CHECK-NEXT: [[TMP11:%.*]] = icmp ne <64 x i8> [[__A]], zeroinitializer +; CHECK-NEXT: [[TMP12:%.*]] = icmp ne <64 x i8> [[__B]], zeroinitializer +; CHECK-NEXT: [[TMP13:%.*]] = and <64 x i1> [[TMP9]], [[TMP10]] +; CHECK-NEXT: [[TMP14:%.*]] = and <64 x i1> [[TMP11]], [[TMP10]] +; CHECK-NEXT: [[TMP15:%.*]] = and <64 x i1> [[TMP9]], [[TMP12]] +; CHECK-NEXT: [[TMP16:%.*]] = or <64 x i1> [[TMP13]], [[TMP14]] +; CHECK-NEXT: [[TMP17:%.*]] = or <64 x i1> [[TMP16]], [[TMP15]] +; CHECK-NEXT: [[TMP18:%.*]] = sext <64 x i1> [[TMP17]] to <64 x i8> +; CHECK-NEXT: [[TMP19:%.*]] = bitcast <64 x i8> [[TMP18]] to <16 x i32> +; CHECK-NEXT: [[TMP20:%.*]] = icmp ne <16 x i32> [[TMP19]], zeroinitializer +; CHECK-NEXT: [[TMP21:%.*]] = sext <16 x i1> [[TMP20]] to <16 x i32> +; CHECK-NEXT: [[TMP22:%.*]] = or <16 x i32> [[TMP21]], [[TMP4]] +; CHECK-NEXT: [[RES:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpbsud.512(<16 x i32> [[__W]], <64 x i8> [[__A]], <64 x i8> [[__B]]) +; CHECK-NEXT: store <16 x i32> [[TMP22]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <16 x i32> [[RES]] ; - %__B = load <16 x i32>, ptr %pB - %res = tail call <16 x i32> @llvm.x86.avx10.vpdpbsud.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) + %__B = load <64 x i8>, ptr %pB + %res = tail call <16 x i32> @llvm.x86.avx10.vpdpbsud.512(<16 x i32> %__W, <64 x i8> %__A, <64 x i8> %__B) ret <16 x i32> %res } -define <16 x i32> @test_mm512_mask_dpbsuds_epi32(<16 x i32> %__W, i16 zeroext %__U, <16 x i32> %__A, <16 x i32> %__B) sanitize_memory { +define <16 x i32> @test_mm512_mask_dpbsuds_epi32(<16 x i32> %__W, i16 zeroext %__U, <64 x i8> %__A, <64 x i8> %__B) sanitize_memory { ; CHECK-LABEL: define <16 x i32> @test_mm512_mask_dpbsuds_epi32( -; CHECK-SAME: <16 x i32> [[__W:%.*]], i16 zeroext [[__U:%.*]], <16 x i32> [[__A:%.*]], <16 x i32> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-SAME: <16 x i32> [[__W:%.*]], i16 zeroext [[__U:%.*]], <64 x i8> [[__A:%.*]], <64 x i8> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 +; CHECK-NEXT: [[TMP3:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8 ; CHECK-NEXT: [[TMP1:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8 -; CHECK-NEXT: [[TMP2:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8 ; CHECK-NEXT: [[TMP4:%.*]] = load i16, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8 ; CHECK-NEXT: call void @llvm.donothing() -; CHECK-NEXT: [[_MSPROP:%.*]] = or <16 x i32> [[TMP1]], [[TMP2]] -; CHECK-NEXT: [[_MSPROP1:%.*]] = or <16 x i32> [[_MSPROP]], [[TMP3]] -; CHECK-NEXT: [[DPI:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpbsuds.512(<16 x i32> [[__W]], <16 x i32> [[__A]], <16 x i32> [[__B]]) +; CHECK-NEXT: [[TMP19:%.*]] = icmp ne <64 x i8> [[TMP2]], zeroinitializer +; CHECK-NEXT: [[TMP20:%.*]] = icmp ne <64 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <64 x i8> [[__A]], zeroinitializer +; CHECK-NEXT: [[TMP22:%.*]] = icmp ne <64 x i8> [[__B]], zeroinitializer +; CHECK-NEXT: [[TMP23:%.*]] = and <64 x i1> [[TMP19]], [[TMP20]] +; CHECK-NEXT: [[TMP10:%.*]] = and <64 x i1> [[TMP21]], [[TMP20]] +; CHECK-NEXT: [[TMP11:%.*]] = and <64 x i1> [[TMP19]], [[TMP22]] +; CHECK-NEXT: [[TMP12:%.*]] = or <64 x i1> [[TMP23]], [[TMP10]] +; CHECK-NEXT: [[TMP13:%.*]] = or <64 x i1> [[TMP12]], [[TMP11]] +; CHECK-NEXT: [[TMP14:%.*]] = sext <64 x i1> [[TMP13]] to <64 x i8> +; CHECK-NEXT: [[TMP15:%.*]] = bitcast <64 x i8> [[TMP14]] to <16 x i32> +; CHECK-NEXT: [[TMP16:%.*]] = icmp ne <16 x i32> [[TMP15]], zeroinitializer +; CHECK-NEXT: [[TMP17:%.*]] = sext <16 x i1> [[TMP16]] to <16 x i32> +; CHECK-NEXT: [[TMP18:%.*]] = or <16 x i32> [[TMP17]], [[TMP1]] +; CHECK-NEXT: [[DPI:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpbsuds.512(<16 x i32> [[__W]], <64 x i8> [[__A]], <64 x i8> [[__B]]) ; CHECK-NEXT: [[TMP5:%.*]] = bitcast i16 [[TMP4]] to <16 x i1> ; CHECK-NEXT: [[BST:%.*]] = bitcast i16 [[__U]] to <16 x i1> -; CHECK-NEXT: [[TMP6:%.*]] = select <16 x i1> [[BST]], <16 x i32> [[_MSPROP1]], <16 x i32> [[TMP1]] +; CHECK-NEXT: [[TMP6:%.*]] = select <16 x i1> [[BST]], <16 x i32> [[TMP18]], <16 x i32> [[TMP1]] ; CHECK-NEXT: [[TMP7:%.*]] = xor <16 x i32> [[DPI]], [[__W]] -; CHECK-NEXT: [[TMP8:%.*]] = or <16 x i32> [[TMP7]], [[_MSPROP1]] +; CHECK-NEXT: [[TMP8:%.*]] = or <16 x i32> [[TMP7]], [[TMP18]] ; CHECK-NEXT: [[TMP9:%.*]] = or <16 x i32> [[TMP8]], [[TMP1]] ; CHECK-NEXT: [[_MSPROP_SELECT:%.*]] = select <16 x i1> [[TMP5]], <16 x i32> [[TMP9]], <16 x i32> [[TMP6]] ; CHECK-NEXT: [[RES:%.*]] = select <16 x i1> [[BST]], <16 x i32> [[DPI]], <16 x i32> [[__W]] ; CHECK-NEXT: store <16 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <16 x i32> [[RES]] ; - %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpbsuds.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) + %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpbsuds.512(<16 x i32> %__W, <64 x i8> %__A, <64 x i8> %__B) %bst = bitcast i16 %__U to <16 x i1> %res = select <16 x i1> %bst, <16 x i32> %dpi, <16 x i32> %__W ret <16 x i32> %res } -define <16 x i32> @test_mm512_maskz_dpbsud_epi32(i16 zeroext %__U, <16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) sanitize_memory { +define <16 x i32> @test_mm512_maskz_dpbsud_epi32(i16 zeroext %__U, <16 x i32> %__W, <64 x i8> %__A, <64 x i8> %__B) sanitize_memory { ; CHECK-LABEL: define <16 x i32> @test_mm512_maskz_dpbsud_epi32( -; CHECK-SAME: i16 zeroext [[__U:%.*]], <16 x i32> [[__W:%.*]], <16 x i32> [[__A:%.*]], <16 x i32> [[__B:%.*]]) #[[ATTR0]] { -; CHECK-NEXT: [[TMP1:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8 -; CHECK-NEXT: [[TMP2:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8 +; CHECK-SAME: i16 zeroext [[__U:%.*]], <16 x i32> [[__W:%.*]], <64 x i8> [[__A:%.*]], <64 x i8> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 +; CHECK-NEXT: [[TMP3:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8 +; CHECK-NEXT: [[TMP19:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8 ; CHECK-NEXT: [[TMP4:%.*]] = load i16, ptr @__msan_param_tls, align 8 ; CHECK-NEXT: call void @llvm.donothing() -; CHECK-NEXT: [[_MSPROP:%.*]] = or <16 x i32> [[TMP1]], [[TMP2]] -; CHECK-NEXT: [[_MSPROP1:%.*]] = or <16 x i32> [[_MSPROP]], [[TMP3]] -; CHECK-NEXT: [[DPI:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpbsud.512(<16 x i32> [[__W]], <16 x i32> [[__A]], <16 x i32> [[__B]]) +; CHECK-NEXT: [[TMP20:%.*]] = icmp ne <64 x i8> [[TMP2]], zeroinitializer +; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <64 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP22:%.*]] = icmp ne <64 x i8> [[__A]], zeroinitializer +; CHECK-NEXT: [[TMP23:%.*]] = icmp ne <64 x i8> [[__B]], zeroinitializer +; CHECK-NEXT: [[TMP24:%.*]] = and <64 x i1> [[TMP20]], [[TMP21]] +; CHECK-NEXT: [[TMP10:%.*]] = and <64 x i1> [[TMP22]], [[TMP21]] +; CHECK-NEXT: [[TMP11:%.*]] = and <64 x i1> [[TMP20]], [[TMP23]] +; CHECK-NEXT: [[TMP12:%.*]] = or <64 x i1> [[TMP24]], [[TMP10]] +; CHECK-NEXT: [[TMP13:%.*]] = or <64 x i1> [[TMP12]], [[TMP11]] +; CHECK-NEXT: [[TMP14:%.*]] = sext <64 x i1> [[TMP13]] to <64 x i8> +; CHECK-NEXT: [[TMP15:%.*]] = bitcast <64 x i8> [[TMP14]] to <16 x i32> +; CHECK-NEXT: [[TMP16:%.*]] = icmp ne <16 x i32> [[TMP15]], zeroinitializer +; CHECK-NEXT: [[TMP17:%.*]] = sext <16 x i1> [[TMP16]] to <16 x i32> +; CHECK-NEXT: [[TMP18:%.*]] = or <16 x i32> [[TMP17]], [[TMP19]] +; CHECK-NEXT: [[DPI:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpbsud.512(<16 x i32> [[__W]], <64 x i8> [[__A]], <64 x i8> [[__B]]) ; CHECK-NEXT: [[TMP5:%.*]] = bitcast i16 [[TMP4]] to <16 x i1> ; CHECK-NEXT: [[BST:%.*]] = bitcast i16 [[__U]] to <16 x i1> -; CHECK-NEXT: [[TMP6:%.*]] = select <16 x i1> [[BST]], <16 x i32> [[_MSPROP1]], <16 x i32> zeroinitializer +; CHECK-NEXT: [[TMP6:%.*]] = select <16 x i1> [[BST]], <16 x i32> [[TMP18]], <16 x i32> zeroinitializer ; CHECK-NEXT: [[TMP7:%.*]] = xor <16 x i32> [[DPI]], zeroinitializer -; CHECK-NEXT: [[TMP8:%.*]] = or <16 x i32> [[TMP7]], [[_MSPROP1]] +; CHECK-NEXT: [[TMP8:%.*]] = or <16 x i32> [[TMP7]], [[TMP18]] ; CHECK-NEXT: [[TMP9:%.*]] = or <16 x i32> [[TMP8]], zeroinitializer ; CHECK-NEXT: [[_MSPROP_SELECT:%.*]] = select <16 x i1> [[TMP5]], <16 x i32> [[TMP9]], <16 x i32> [[TMP6]] ; CHECK-NEXT: [[RES:%.*]] = select <16 x i1> [[BST]], <16 x i32> [[DPI]], <16 x i32> zeroinitializer ; CHECK-NEXT: store <16 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <16 x i32> [[RES]] ; - %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpbsud.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) + %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpbsud.512(<16 x i32> %__W, <64 x i8> %__A, <64 x i8> %__B) %bst = bitcast i16 %__U to <16 x i1> %res = select <16 x i1> %bst, <16 x i32> %dpi, <16 x i32> zeroinitializer ret <16 x i32> %res } -declare <16 x i32> @llvm.x86.avx10.vpdpbsud.512(<16 x i32>, <16 x i32>, <16 x i32>) -declare <16 x i32> @llvm.x86.avx10.vpdpbsuds.512(<16 x i32>, <16 x i32>, <16 x i32>) +declare <16 x i32> @llvm.x86.avx10.vpdpbsud.512(<16 x i32>, <64 x i8>, <64 x i8>) +declare <16 x i32> @llvm.x86.avx10.vpdpbsuds.512(<16 x i32>, <64 x i8>, <64 x i8>) -define <16 x i32> @test_mm512_dpbuud_epi32(<16 x i32> %__W, <16 x i32> %__A, ptr %pB) sanitize_memory { +define <16 x i32> @test_mm512_dpbuud_epi32(<16 x i32> %__W, <64 x i8> %__A, ptr %pB) sanitize_memory { ; CHECK-LABEL: define <16 x i32> @test_mm512_dpbuud_epi32( -; CHECK-SAME: <16 x i32> [[__W:%.*]], <16 x i32> [[__A:%.*]], ptr [[PB:%.*]]) #[[ATTR0]] { +; CHECK-SAME: <16 x i32> [[__W:%.*]], <64 x i8> [[__A:%.*]], ptr [[PB:%.*]]) #[[ATTR0]] { ; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 128) to ptr), align 8 -; CHECK-NEXT: [[TMP2:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8 +; CHECK-NEXT: [[TMP3:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8 +; CHECK-NEXT: [[TMP4:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8 ; CHECK-NEXT: call void @llvm.donothing() ; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0 ; CHECK-NEXT: br i1 [[_MSCMP]], label %[[BB4:.*]], label %[[BB5:.*]], !prof [[PROF1]] @@ -361,80 +385,116 @@ define <16 x i32> @test_mm512_dpbuud_epi32(<16 x i32> %__W, <16 x i32> %__A, ptr ; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR5]] ; CHECK-NEXT: unreachable ; CHECK: [[BB5]]: -; CHECK-NEXT: [[__B:%.*]] = load <16 x i32>, ptr [[PB]], align 64 +; CHECK-NEXT: [[__B:%.*]] = load <64 x i8>, ptr [[PB]], align 64 ; CHECK-NEXT: [[TMP6:%.*]] = ptrtoint ptr [[PB]] to i64 ; CHECK-NEXT: [[TMP7:%.*]] = xor i64 [[TMP6]], 87960930222080 ; CHECK-NEXT: [[TMP8:%.*]] = inttoptr i64 [[TMP7]] to ptr -; CHECK-NEXT: [[_MSLD:%.*]] = load <16 x i32>, ptr [[TMP8]], align 64 -; CHECK-NEXT: [[_MSPROP:%.*]] = or <16 x i32> [[TMP2]], [[TMP3]] -; CHECK-NEXT: [[_MSPROP1:%.*]] = or <16 x i32> [[_MSPROP]], [[_MSLD]] -; CHECK-NEXT: [[RES:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpbuud.512(<16 x i32> [[__W]], <16 x i32> [[__A]], <16 x i32> [[__B]]) -; CHECK-NEXT: store <16 x i32> [[_MSPROP1]], ptr @__msan_retval_tls, align 8 +; CHECK-NEXT: [[_MSLD:%.*]] = load <64 x i8>, ptr [[TMP8]], align 64 +; CHECK-NEXT: [[TMP9:%.*]] = icmp ne <64 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP10:%.*]] = icmp ne <64 x i8> [[_MSLD]], zeroinitializer +; CHECK-NEXT: [[TMP11:%.*]] = icmp ne <64 x i8> [[__A]], zeroinitializer +; CHECK-NEXT: [[TMP12:%.*]] = icmp ne <64 x i8> [[__B]], zeroinitializer +; CHECK-NEXT: [[TMP13:%.*]] = and <64 x i1> [[TMP9]], [[TMP10]] +; CHECK-NEXT: [[TMP14:%.*]] = and <64 x i1> [[TMP11]], [[TMP10]] +; CHECK-NEXT: [[TMP15:%.*]] = and <64 x i1> [[TMP9]], [[TMP12]] +; CHECK-NEXT: [[TMP16:%.*]] = or <64 x i1> [[TMP13]], [[TMP14]] +; CHECK-NEXT: [[TMP17:%.*]] = or <64 x i1> [[TMP16]], [[TMP15]] +; CHECK-NEXT: [[TMP18:%.*]] = sext <64 x i1> [[TMP17]] to <64 x i8> +; CHECK-NEXT: [[TMP19:%.*]] = bitcast <64 x i8> [[TMP18]] to <16 x i32> +; CHECK-NEXT: [[TMP20:%.*]] = icmp ne <16 x i32> [[TMP19]], zeroinitializer +; CHECK-NEXT: [[TMP21:%.*]] = sext <16 x i1> [[TMP20]] to <16 x i32> +; CHECK-NEXT: [[TMP22:%.*]] = or <16 x i32> [[TMP21]], [[TMP4]] +; CHECK-NEXT: [[RES:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpbuud.512(<16 x i32> [[__W]], <64 x i8> [[__A]], <64 x i8> [[__B]]) +; CHECK-NEXT: store <16 x i32> [[TMP22]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <16 x i32> [[RES]] ; - %__B = load <16 x i32>, ptr %pB - %res = tail call <16 x i32> @llvm.x86.avx10.vpdpbuud.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) + %__B = load <64 x i8>, ptr %pB + %res = tail call <16 x i32> @llvm.x86.avx10.vpdpbuud.512(<16 x i32> %__W, <64 x i8> %__A, <64 x i8> %__B) ret <16 x i32> %res } -define <16 x i32> @test_mm512_mask_dpbuuds_epi32(<16 x i32> %__W, i16 zeroext %__U, <16 x i32> %__A, <16 x i32> %__B) sanitize_memory { +define <16 x i32> @test_mm512_mask_dpbuuds_epi32(<16 x i32> %__W, i16 zeroext %__U, <64 x i8> %__A, <64 x i8> %__B) sanitize_memory { ; CHECK-LABEL: define <16 x i32> @test_mm512_mask_dpbuuds_epi32( -; CHECK-SAME: <16 x i32> [[__W:%.*]], i16 zeroext [[__U:%.*]], <16 x i32> [[__A:%.*]], <16 x i32> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-SAME: <16 x i32> [[__W:%.*]], i16 zeroext [[__U:%.*]], <64 x i8> [[__A:%.*]], <64 x i8> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 +; CHECK-NEXT: [[TMP3:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8 ; CHECK-NEXT: [[TMP1:%.*]] = load <16 x i32>, ptr @__msan_param_tls, align 8 -; CHECK-NEXT: [[TMP2:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8 ; CHECK-NEXT: [[TMP4:%.*]] = load i16, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8 ; CHECK-NEXT: call void @llvm.donothing() -; CHECK-NEXT: [[_MSPROP:%.*]] = or <16 x i32> [[TMP1]], [[TMP2]] -; CHECK-NEXT: [[_MSPROP1:%.*]] = or <16 x i32> [[_MSPROP]], [[TMP3]] -; CHECK-NEXT: [[DPI:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpbuuds.512(<16 x i32> [[__W]], <16 x i32> [[__A]], <16 x i32> [[__B]]) +; CHECK-NEXT: [[TMP19:%.*]] = icmp ne <64 x i8> [[TMP2]], zeroinitializer +; CHECK-NEXT: [[TMP20:%.*]] = icmp ne <64 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <64 x i8> [[__A]], zeroinitializer +; CHECK-NEXT: [[TMP22:%.*]] = icmp ne <64 x i8> [[__B]], zeroinitializer +; CHECK-NEXT: [[TMP23:%.*]] = and <64 x i1> [[TMP19]], [[TMP20]] +; CHECK-NEXT: [[TMP10:%.*]] = and <64 x i1> [[TMP21]], [[TMP20]] +; CHECK-NEXT: [[TMP11:%.*]] = and <64 x i1> [[TMP19]], [[TMP22]] +; CHECK-NEXT: [[TMP12:%.*]] = or <64 x i1> [[TMP23]], [[TMP10]] +; CHECK-NEXT: [[TMP13:%.*]] = or <64 x i1> [[TMP12]], [[TMP11]] +; CHECK-NEXT: [[TMP14:%.*]] = sext <64 x i1> [[TMP13]] to <64 x i8> +; CHECK-NEXT: [[TMP15:%.*]] = bitcast <64 x i8> [[TMP14]] to <16 x i32> +; CHECK-NEXT: [[TMP16:%.*]] = icmp ne <16 x i32> [[TMP15]], zeroinitializer +; CHECK-NEXT: [[TMP17:%.*]] = sext <16 x i1> [[TMP16]] to <16 x i32> +; CHECK-NEXT: [[TMP18:%.*]] = or <16 x i32> [[TMP17]], [[TMP1]] +; CHECK-NEXT: [[DPI:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpbuuds.512(<16 x i32> [[__W]], <64 x i8> [[__A]], <64 x i8> [[__B]]) ; CHECK-NEXT: [[TMP5:%.*]] = bitcast i16 [[TMP4]] to <16 x i1> ; CHECK-NEXT: [[BST:%.*]] = bitcast i16 [[__U]] to <16 x i1> -; CHECK-NEXT: [[TMP6:%.*]] = select <16 x i1> [[BST]], <16 x i32> [[_MSPROP1]], <16 x i32> [[TMP1]] +; CHECK-NEXT: [[TMP6:%.*]] = select <16 x i1> [[BST]], <16 x i32> [[TMP18]], <16 x i32> [[TMP1]] ; CHECK-NEXT: [[TMP7:%.*]] = xor <16 x i32> [[DPI]], [[__W]] -; CHECK-NEXT: [[TMP8:%.*]] = or <16 x i32> [[TMP7]], [[_MSPROP1]] +; CHECK-NEXT: [[TMP8:%.*]] = or <16 x i32> [[TMP7]], [[TMP18]] ; CHECK-NEXT: [[TMP9:%.*]] = or <16 x i32> [[TMP8]], [[TMP1]] ; CHECK-NEXT: [[_MSPROP_SELECT:%.*]] = select <16 x i1> [[TMP5]], <16 x i32> [[TMP9]], <16 x i32> [[TMP6]] ; CHECK-NEXT: [[RES:%.*]] = select <16 x i1> [[BST]], <16 x i32> [[DPI]], <16 x i32> [[__W]] ; CHECK-NEXT: store <16 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <16 x i32> [[RES]] ; - %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpbuuds.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) + %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpbuuds.512(<16 x i32> %__W, <64 x i8> %__A, <64 x i8> %__B) %bst = bitcast i16 %__U to <16 x i1> %res = select <16 x i1> %bst, <16 x i32> %dpi, <16 x i32> %__W ret <16 x i32> %res } -define <16 x i32> @test_mm512_maskz_dpbuud_epi32(i16 zeroext %__U, <16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) sanitize_memory { +define <16 x i32> @test_mm512_maskz_dpbuud_epi32(i16 zeroext %__U, <16 x i32> %__W, <64 x i8> %__A, <64 x i8> %__B) sanitize_memory { ; CHECK-LABEL: define <16 x i32> @test_mm512_maskz_dpbuud_epi32( -; CHECK-SAME: i16 zeroext [[__U:%.*]], <16 x i32> [[__W:%.*]], <16 x i32> [[__A:%.*]], <16 x i32> [[__B:%.*]]) #[[ATTR0]] { -; CHECK-NEXT: [[TMP1:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8 -; CHECK-NEXT: [[TMP2:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8 +; CHECK-SAME: i16 zeroext [[__U:%.*]], <16 x i32> [[__W:%.*]], <64 x i8> [[__A:%.*]], <64 x i8> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[TMP2:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 +; CHECK-NEXT: [[TMP3:%.*]] = load <64 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 136) to ptr), align 8 +; CHECK-NEXT: [[TMP19:%.*]] = load <16 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8 ; CHECK-NEXT: [[TMP4:%.*]] = load i16, ptr @__msan_param_tls, align 8 ; CHECK-NEXT: call void @llvm.donothing() -; CHECK-NEXT: [[_MSPROP:%.*]] = or <16 x i32> [[TMP1]], [[TMP2]] -; CHECK-NEXT: [[_MSPROP1:%.*]] = or <16 x i32> [[_MSPROP]], [[TMP3]] -; CHECK-NEXT: [[DPI:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpbuud.512(<16 x i32> [[__W]], <16 x i32> [[__A]], <16 x i32> [[__B]]) +; CHECK-NEXT: [[TMP20:%.*]] = icmp ne <64 x i8> [[TMP2]], zeroinitializer +; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <64 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP22:%.*]] = icmp ne <64 x i8> [[__A]], zeroinitializer +; CHECK-NEXT: [[TMP23:%.*]] = icmp ne <64 x i8> [[__B]], zeroinitializer +; CHECK-NEXT: [[TMP24:%.*]] = and <64 x i1> [[TMP20]], [[TMP21]] +; CHECK-NEXT: [[TMP10:%.*]] = and <64 x i1> [[TMP22]], [[TMP21]] +; CHECK-NEXT: [[TMP11:%.*]] = and <64 x i1> [[TMP20]], [[TMP23]] +; CHECK-NEXT: [[TMP12:%.*]] = or <64 x i1> [[TMP24]], [[TMP10]] +; CHECK-NEXT: [[TMP13:%.*]] = or <64 x i1> [[TMP12]], [[TMP11]] +; CHECK-NEXT: [[TMP14:%.*]] = sext <64 x i1> [[TMP13]] to <64 x i8> +; CHECK-NEXT: [[TMP15:%.*]] = bitcast <64 x i8> [[TMP14]] to <16 x i32> +; CHECK-NEXT: [[TMP16:%.*]] = icmp ne <16 x i32> [[TMP15]], zeroinitializer +; CHECK-NEXT: [[TMP17:%.*]] = sext <16 x i1> [[TMP16]] to <16 x i32> +; CHECK-NEXT: [[TMP18:%.*]] = or <16 x i32> [[TMP17]], [[TMP19]] +; CHECK-NEXT: [[DPI:%.*]] = tail call <16 x i32> @llvm.x86.avx10.vpdpbuud.512(<16 x i32> [[__W]], <64 x i8> [[__A]], <64 x i8> [[__B]]) ; CHECK-NEXT: [[TMP5:%.*]] = bitcast i16 [[TMP4]] to <16 x i1> ; CHECK-NEXT: [[BST:%.*]] = bitcast i16 [[__U]] to <16 x i1> -; CHECK-NEXT: [[TMP6:%.*]] = select <16 x i1> [[BST]], <16 x i32> [[_MSPROP1]], <16 x i32> zeroinitializer +; CHECK-NEXT: [[TMP6:%.*]] = select <16 x i1> [[BST]], <16 x i32> [[TMP18]], <16 x i32> zeroinitializer ; CHECK-NEXT: [[TMP7:%.*]] = xor <16 x i32> [[DPI]], zeroinitializer -; CHECK-NEXT: [[TMP8:%.*]] = or <16 x i32> [[TMP7]], [[_MSPROP1]] +; CHECK-NEXT: [[TMP8:%.*]] = or <16 x i32> [[TMP7]], [[TMP18]] ; CHECK-NEXT: [[TMP9:%.*]] = or <16 x i32> [[TMP8]], zeroinitializer ; CHECK-NEXT: [[_MSPROP_SELECT:%.*]] = select <16 x i1> [[TMP5]], <16 x i32> [[TMP9]], <16 x i32> [[TMP6]] ; CHECK-NEXT: [[RES:%.*]] = select <16 x i1> [[BST]], <16 x i32> [[DPI]], <16 x i32> zeroinitializer ; CHECK-NEXT: store <16 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <16 x i32> [[RES]] ; - %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpbuud.512(<16 x i32> %__W, <16 x i32> %__A, <16 x i32> %__B) + %dpi = tail call <16 x i32> @llvm.x86.avx10.vpdpbuud.512(<16 x i32> %__W, <64 x i8> %__A, <64 x i8> %__B) %bst = bitcast i16 %__U to <16 x i1> %res = select <16 x i1> %bst, <16 x i32> %dpi, <16 x i32> zeroinitializer ret <16 x i32> %res } -declare <16 x i32> @llvm.x86.avx10.vpdpbuud.512(<16 x i32>, <16 x i32>, <16 x i32>) -declare <16 x i32> @llvm.x86.avx10.vpdpbuuds.512(<16 x i32>, <16 x i32>, <16 x i32>) +declare <16 x i32> @llvm.x86.avx10.vpdpbuud.512(<16 x i32>, <64 x i8>, <64 x i8>) +declare <16 x i32> @llvm.x86.avx10.vpdpbuuds.512(<16 x i32>, <64 x i8>, <64 x i8>) define <16 x i32> @test_mm512_dpwsud_epi32(<16 x i32> %__W, <16 x i32> %__A, ptr %pB) sanitize_memory { diff --git a/llvm/test/Instrumentation/MemorySanitizer/X86/avx10_2ni-intrinsics.ll b/llvm/test/Instrumentation/MemorySanitizer/X86/avx10_2ni-intrinsics.ll index e121c3b..373eff6 100644 --- a/llvm/test/Instrumentation/MemorySanitizer/X86/avx10_2ni-intrinsics.ll +++ b/llvm/test/Instrumentation/MemorySanitizer/X86/avx10_2ni-intrinsics.ll @@ -243,25 +243,21 @@ declare <4 x float> @llvm.x86.avx10.vdpphps.128(<4 x float>, <8 x half>, <8 x ha declare <8 x float> @llvm.x86.avx10.vdpphps.256(<8 x float>, <16 x half>, <16 x half>) -define <4 x i32> @test_mm_mask_dpbssd_epi32(<4 x i32> %__W, i4 zeroext %__U, <4 x i32> %__A, <4 x i32> %__B) sanitize_memory { +define <4 x i32> @test_mm_mask_dpbssd_epi32(<4 x i32> %__W, i4 zeroext %__U, <16 x i8> %__A, <16 x i8> %__B) sanitize_memory { ; CHECK-LABEL: define <4 x i32> @test_mm_mask_dpbssd_epi32( -; CHECK-SAME: <4 x i32> [[__W:%.*]], i4 zeroext [[__U:%.*]], <4 x i32> [[__A:%.*]], <4 x i32> [[__B:%.*]]) #[[ATTR0]] { -; CHECK-NEXT: [[TMP2:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 24) to ptr), align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 +; CHECK-SAME: <4 x i32> [[__W:%.*]], i4 zeroext [[__U:%.*]], <16 x i8> [[__A:%.*]], <16 x i8> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[TMP24:%.*]] = load <16 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 24) to ptr), align 8 +; CHECK-NEXT: [[TMP25:%.*]] = load <16 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 ; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8 ; CHECK-NEXT: [[TMP4:%.*]] = load i4, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 16) to ptr), align 8 ; CHECK-NEXT: call void @llvm.donothing() -; CHECK-NEXT: [[TMP24:%.*]] = bitcast <4 x i32> [[__A]] to <16 x i8> -; CHECK-NEXT: [[TMP25:%.*]] = bitcast <4 x i32> [[__B]] to <16 x i8> -; CHECK-NEXT: [[TMP26:%.*]] = bitcast <4 x i32> [[TMP2]] to <16 x i8> -; CHECK-NEXT: [[TMP27:%.*]] = bitcast <4 x i32> [[TMP3]] to <16 x i8> -; CHECK-NEXT: [[TMP28:%.*]] = icmp ne <16 x i8> [[TMP26]], zeroinitializer -; CHECK-NEXT: [[TMP10:%.*]] = icmp ne <16 x i8> [[TMP27]], zeroinitializer ; CHECK-NEXT: [[TMP11:%.*]] = icmp ne <16 x i8> [[TMP24]], zeroinitializer ; CHECK-NEXT: [[TMP12:%.*]] = icmp ne <16 x i8> [[TMP25]], zeroinitializer -; CHECK-NEXT: [[TMP13:%.*]] = and <16 x i1> [[TMP28]], [[TMP10]] -; CHECK-NEXT: [[TMP14:%.*]] = and <16 x i1> [[TMP11]], [[TMP10]] -; CHECK-NEXT: [[TMP15:%.*]] = and <16 x i1> [[TMP28]], [[TMP12]] +; CHECK-NEXT: [[TMP10:%.*]] = icmp ne <16 x i8> [[__A]], zeroinitializer +; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <16 x i8> [[__B]], zeroinitializer +; CHECK-NEXT: [[TMP13:%.*]] = and <16 x i1> [[TMP11]], [[TMP12]] +; CHECK-NEXT: [[TMP14:%.*]] = and <16 x i1> [[TMP10]], [[TMP12]] +; CHECK-NEXT: [[TMP15:%.*]] = and <16 x i1> [[TMP11]], [[TMP21]] ; CHECK-NEXT: [[TMP16:%.*]] = or <16 x i1> [[TMP13]], [[TMP14]] ; CHECK-NEXT: [[TMP17:%.*]] = or <16 x i1> [[TMP16]], [[TMP15]] ; CHECK-NEXT: [[TMP18:%.*]] = sext <16 x i1> [[TMP17]] to <16 x i8> @@ -269,7 +265,7 @@ define <4 x i32> @test_mm_mask_dpbssd_epi32(<4 x i32> %__W, i4 zeroext %__U, <4 ; CHECK-NEXT: [[TMP20:%.*]] = icmp ne <4 x i32> [[TMP19]], zeroinitializer ; CHECK-NEXT: [[TMP23:%.*]] = sext <4 x i1> [[TMP20]] to <4 x i32> ; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[TMP23]], [[TMP1]] -; CHECK-NEXT: [[DPI:%.*]] = tail call <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32> [[__W]], <4 x i32> [[__A]], <4 x i32> [[__B]]) +; CHECK-NEXT: [[DPI:%.*]] = tail call <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32> [[__W]], <16 x i8> [[__A]], <16 x i8> [[__B]]) ; CHECK-NEXT: [[TMP5:%.*]] = bitcast i4 [[TMP4]] to <4 x i1> ; CHECK-NEXT: [[BST:%.*]] = bitcast i4 [[__U]] to <4 x i1> ; CHECK-NEXT: [[TMP6:%.*]] = select <4 x i1> [[BST]], <4 x i32> [[_MSPROP1]], <4 x i32> [[TMP1]] @@ -281,31 +277,27 @@ define <4 x i32> @test_mm_mask_dpbssd_epi32(<4 x i32> %__W, i4 zeroext %__U, <4 ; CHECK-NEXT: store <4 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <4 x i32> [[RES]] ; - %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B) + %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32> %__W, <16 x i8> %__A, <16 x i8> %__B) %bst = bitcast i4 %__U to <4 x i1> %res = select <4 x i1> %bst, <4 x i32> %dpi, <4 x i32> %__W ret <4 x i32> %res } -define <4 x i32> @test_mm_maskz_dpbssds_epi32(i4 zeroext %__U, <4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B) sanitize_memory { +define <4 x i32> @test_mm_maskz_dpbssds_epi32(i4 zeroext %__U, <4 x i32> %__W, <16 x i8> %__A, <16 x i8> %__B) sanitize_memory { ; CHECK-LABEL: define <4 x i32> @test_mm_maskz_dpbssds_epi32( -; CHECK-SAME: i4 zeroext [[__U:%.*]], <4 x i32> [[__W:%.*]], <4 x i32> [[__A:%.*]], <4 x i32> [[__B:%.*]]) #[[ATTR0]] { -; CHECK-NEXT: [[TMP2:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 24) to ptr), align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 +; CHECK-SAME: i4 zeroext [[__U:%.*]], <4 x i32> [[__W:%.*]], <16 x i8> [[__A:%.*]], <16 x i8> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[TMP25:%.*]] = load <16 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 24) to ptr), align 8 +; CHECK-NEXT: [[TMP26:%.*]] = load <16 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 ; CHECK-NEXT: [[TMP24:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8 ; CHECK-NEXT: [[TMP4:%.*]] = load i4, ptr @__msan_param_tls, align 8 ; CHECK-NEXT: call void @llvm.donothing() -; CHECK-NEXT: [[TMP25:%.*]] = bitcast <4 x i32> [[__A]] to <16 x i8> -; CHECK-NEXT: [[TMP26:%.*]] = bitcast <4 x i32> [[__B]] to <16 x i8> -; CHECK-NEXT: [[TMP27:%.*]] = bitcast <4 x i32> [[TMP2]] to <16 x i8> -; CHECK-NEXT: [[TMP28:%.*]] = bitcast <4 x i32> [[TMP3]] to <16 x i8> -; CHECK-NEXT: [[TMP29:%.*]] = icmp ne <16 x i8> [[TMP27]], zeroinitializer -; CHECK-NEXT: [[TMP10:%.*]] = icmp ne <16 x i8> [[TMP28]], zeroinitializer ; CHECK-NEXT: [[TMP11:%.*]] = icmp ne <16 x i8> [[TMP25]], zeroinitializer ; CHECK-NEXT: [[TMP12:%.*]] = icmp ne <16 x i8> [[TMP26]], zeroinitializer -; CHECK-NEXT: [[TMP13:%.*]] = and <16 x i1> [[TMP29]], [[TMP10]] -; CHECK-NEXT: [[TMP14:%.*]] = and <16 x i1> [[TMP11]], [[TMP10]] -; CHECK-NEXT: [[TMP15:%.*]] = and <16 x i1> [[TMP29]], [[TMP12]] +; CHECK-NEXT: [[TMP10:%.*]] = icmp ne <16 x i8> [[__A]], zeroinitializer +; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <16 x i8> [[__B]], zeroinitializer +; CHECK-NEXT: [[TMP13:%.*]] = and <16 x i1> [[TMP11]], [[TMP12]] +; CHECK-NEXT: [[TMP14:%.*]] = and <16 x i1> [[TMP10]], [[TMP12]] +; CHECK-NEXT: [[TMP15:%.*]] = and <16 x i1> [[TMP11]], [[TMP21]] ; CHECK-NEXT: [[TMP16:%.*]] = or <16 x i1> [[TMP13]], [[TMP14]] ; CHECK-NEXT: [[TMP17:%.*]] = or <16 x i1> [[TMP16]], [[TMP15]] ; CHECK-NEXT: [[TMP18:%.*]] = sext <16 x i1> [[TMP17]] to <16 x i8> @@ -313,7 +305,7 @@ define <4 x i32> @test_mm_maskz_dpbssds_epi32(i4 zeroext %__U, <4 x i32> %__W, < ; CHECK-NEXT: [[TMP20:%.*]] = icmp ne <4 x i32> [[TMP19]], zeroinitializer ; CHECK-NEXT: [[TMP23:%.*]] = sext <4 x i1> [[TMP20]] to <4 x i32> ; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[TMP23]], [[TMP24]] -; CHECK-NEXT: [[DPI:%.*]] = tail call <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32> [[__W]], <4 x i32> [[__A]], <4 x i32> [[__B]]) +; CHECK-NEXT: [[DPI:%.*]] = tail call <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32> [[__W]], <16 x i8> [[__A]], <16 x i8> [[__B]]) ; CHECK-NEXT: [[TMP5:%.*]] = bitcast i4 [[TMP4]] to <4 x i1> ; CHECK-NEXT: [[BST:%.*]] = bitcast i4 [[__U]] to <4 x i1> ; CHECK-NEXT: [[TMP6:%.*]] = select <4 x i1> [[BST]], <4 x i32> [[_MSPROP1]], <4 x i32> zeroinitializer @@ -325,31 +317,27 @@ define <4 x i32> @test_mm_maskz_dpbssds_epi32(i4 zeroext %__U, <4 x i32> %__W, < ; CHECK-NEXT: store <4 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <4 x i32> [[RES]] ; - %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B) + %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32> %__W, <16 x i8> %__A, <16 x i8> %__B) %bst = bitcast i4 %__U to <4 x i1> %res = select <4 x i1> %bst, <4 x i32> %dpi, <4 x i32> zeroinitializer ret <4 x i32> %res } -define <8 x i32> @test_mm256_maskz_dpbssds_epi32(<8 x i32> %__W, i8 zeroext %__U, <8 x i32> %__A, <8 x i32> %__B) sanitize_memory { +define <8 x i32> @test_mm256_maskz_dpbssds_epi32(<8 x i32> %__W, i8 zeroext %__U, <32 x i8> %__A, <32 x i8> %__B) sanitize_memory { ; CHECK-LABEL: define <8 x i32> @test_mm256_maskz_dpbssds_epi32( -; CHECK-SAME: <8 x i32> [[__W:%.*]], i8 zeroext [[__U:%.*]], <8 x i32> [[__A:%.*]], <8 x i32> [[__B:%.*]]) #[[ATTR0]] { -; CHECK-NEXT: [[TMP2:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 +; CHECK-SAME: <8 x i32> [[__W:%.*]], i8 zeroext [[__U:%.*]], <32 x i8> [[__A:%.*]], <32 x i8> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[TMP24:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 +; CHECK-NEXT: [[TMP25:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 ; CHECK-NEXT: [[TMP1:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8 ; CHECK-NEXT: [[TMP4:%.*]] = load i8, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8 ; CHECK-NEXT: call void @llvm.donothing() -; CHECK-NEXT: [[TMP24:%.*]] = bitcast <8 x i32> [[__A]] to <32 x i8> -; CHECK-NEXT: [[TMP25:%.*]] = bitcast <8 x i32> [[__B]] to <32 x i8> -; CHECK-NEXT: [[TMP26:%.*]] = bitcast <8 x i32> [[TMP2]] to <32 x i8> -; CHECK-NEXT: [[TMP27:%.*]] = bitcast <8 x i32> [[TMP3]] to <32 x i8> -; CHECK-NEXT: [[TMP28:%.*]] = icmp ne <32 x i8> [[TMP26]], zeroinitializer -; CHECK-NEXT: [[TMP10:%.*]] = icmp ne <32 x i8> [[TMP27]], zeroinitializer ; CHECK-NEXT: [[TMP11:%.*]] = icmp ne <32 x i8> [[TMP24]], zeroinitializer ; CHECK-NEXT: [[TMP12:%.*]] = icmp ne <32 x i8> [[TMP25]], zeroinitializer -; CHECK-NEXT: [[TMP13:%.*]] = and <32 x i1> [[TMP28]], [[TMP10]] -; CHECK-NEXT: [[TMP14:%.*]] = and <32 x i1> [[TMP11]], [[TMP10]] -; CHECK-NEXT: [[TMP15:%.*]] = and <32 x i1> [[TMP28]], [[TMP12]] +; CHECK-NEXT: [[TMP10:%.*]] = icmp ne <32 x i8> [[__A]], zeroinitializer +; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <32 x i8> [[__B]], zeroinitializer +; CHECK-NEXT: [[TMP13:%.*]] = and <32 x i1> [[TMP11]], [[TMP12]] +; CHECK-NEXT: [[TMP14:%.*]] = and <32 x i1> [[TMP10]], [[TMP12]] +; CHECK-NEXT: [[TMP15:%.*]] = and <32 x i1> [[TMP11]], [[TMP21]] ; CHECK-NEXT: [[TMP16:%.*]] = or <32 x i1> [[TMP13]], [[TMP14]] ; CHECK-NEXT: [[TMP17:%.*]] = or <32 x i1> [[TMP16]], [[TMP15]] ; CHECK-NEXT: [[TMP18:%.*]] = sext <32 x i1> [[TMP17]] to <32 x i8> @@ -357,7 +345,7 @@ define <8 x i32> @test_mm256_maskz_dpbssds_epi32(<8 x i32> %__W, i8 zeroext %__U ; CHECK-NEXT: [[TMP20:%.*]] = icmp ne <8 x i32> [[TMP19]], zeroinitializer ; CHECK-NEXT: [[TMP23:%.*]] = sext <8 x i1> [[TMP20]] to <8 x i32> ; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[TMP23]], [[TMP1]] -; CHECK-NEXT: [[DPI:%.*]] = tail call <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32> [[__W]], <8 x i32> [[__A]], <8 x i32> [[__B]]) +; CHECK-NEXT: [[DPI:%.*]] = tail call <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32> [[__W]], <32 x i8> [[__A]], <32 x i8> [[__B]]) ; CHECK-NEXT: [[TMP5:%.*]] = bitcast i8 [[TMP4]] to <8 x i1> ; CHECK-NEXT: [[BST:%.*]] = bitcast i8 [[__U]] to <8 x i1> ; CHECK-NEXT: [[TMP6:%.*]] = select <8 x i1> [[BST]], <8 x i32> [[_MSPROP1]], <8 x i32> [[TMP1]] @@ -369,31 +357,27 @@ define <8 x i32> @test_mm256_maskz_dpbssds_epi32(<8 x i32> %__W, i8 zeroext %__U ; CHECK-NEXT: store <8 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <8 x i32> [[RES]] ; - %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B) + %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32> %__W, <32 x i8> %__A, <32 x i8> %__B) %bst = bitcast i8 %__U to <8 x i1> %res = select <8 x i1> %bst, <8 x i32> %dpi, <8 x i32> %__W ret <8 x i32> %res } -define <8 x i32> @test_mm256_mask_dpbssd_epi32(i8 zeroext %__U, <8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B) sanitize_memory { +define <8 x i32> @test_mm256_mask_dpbssd_epi32(i8 zeroext %__U, <8 x i32> %__W, <32 x i8> %__A, <32 x i8> %__B) sanitize_memory { ; CHECK-LABEL: define <8 x i32> @test_mm256_mask_dpbssd_epi32( -; CHECK-SAME: i8 zeroext [[__U:%.*]], <8 x i32> [[__W:%.*]], <8 x i32> [[__A:%.*]], <8 x i32> [[__B:%.*]]) #[[ATTR0]] { -; CHECK-NEXT: [[TMP2:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 +; CHECK-SAME: i8 zeroext [[__U:%.*]], <8 x i32> [[__W:%.*]], <32 x i8> [[__A:%.*]], <32 x i8> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[TMP25:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 +; CHECK-NEXT: [[TMP26:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 ; CHECK-NEXT: [[TMP24:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8 ; CHECK-NEXT: [[TMP4:%.*]] = load i8, ptr @__msan_param_tls, align 8 ; CHECK-NEXT: call void @llvm.donothing() -; CHECK-NEXT: [[TMP25:%.*]] = bitcast <8 x i32> [[__A]] to <32 x i8> -; CHECK-NEXT: [[TMP26:%.*]] = bitcast <8 x i32> [[__B]] to <32 x i8> -; CHECK-NEXT: [[TMP27:%.*]] = bitcast <8 x i32> [[TMP2]] to <32 x i8> -; CHECK-NEXT: [[TMP28:%.*]] = bitcast <8 x i32> [[TMP3]] to <32 x i8> -; CHECK-NEXT: [[TMP29:%.*]] = icmp ne <32 x i8> [[TMP27]], zeroinitializer -; CHECK-NEXT: [[TMP10:%.*]] = icmp ne <32 x i8> [[TMP28]], zeroinitializer ; CHECK-NEXT: [[TMP11:%.*]] = icmp ne <32 x i8> [[TMP25]], zeroinitializer ; CHECK-NEXT: [[TMP12:%.*]] = icmp ne <32 x i8> [[TMP26]], zeroinitializer -; CHECK-NEXT: [[TMP13:%.*]] = and <32 x i1> [[TMP29]], [[TMP10]] -; CHECK-NEXT: [[TMP14:%.*]] = and <32 x i1> [[TMP11]], [[TMP10]] -; CHECK-NEXT: [[TMP15:%.*]] = and <32 x i1> [[TMP29]], [[TMP12]] +; CHECK-NEXT: [[TMP10:%.*]] = icmp ne <32 x i8> [[__A]], zeroinitializer +; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <32 x i8> [[__B]], zeroinitializer +; CHECK-NEXT: [[TMP13:%.*]] = and <32 x i1> [[TMP11]], [[TMP12]] +; CHECK-NEXT: [[TMP14:%.*]] = and <32 x i1> [[TMP10]], [[TMP12]] +; CHECK-NEXT: [[TMP15:%.*]] = and <32 x i1> [[TMP11]], [[TMP21]] ; CHECK-NEXT: [[TMP16:%.*]] = or <32 x i1> [[TMP13]], [[TMP14]] ; CHECK-NEXT: [[TMP17:%.*]] = or <32 x i1> [[TMP16]], [[TMP15]] ; CHECK-NEXT: [[TMP18:%.*]] = sext <32 x i1> [[TMP17]] to <32 x i8> @@ -401,7 +385,7 @@ define <8 x i32> @test_mm256_mask_dpbssd_epi32(i8 zeroext %__U, <8 x i32> %__W, ; CHECK-NEXT: [[TMP20:%.*]] = icmp ne <8 x i32> [[TMP19]], zeroinitializer ; CHECK-NEXT: [[TMP23:%.*]] = sext <8 x i1> [[TMP20]] to <8 x i32> ; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[TMP23]], [[TMP24]] -; CHECK-NEXT: [[DPI:%.*]] = tail call <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32> [[__W]], <8 x i32> [[__A]], <8 x i32> [[__B]]) +; CHECK-NEXT: [[DPI:%.*]] = tail call <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32> [[__W]], <32 x i8> [[__A]], <32 x i8> [[__B]]) ; CHECK-NEXT: [[TMP5:%.*]] = bitcast i8 [[TMP4]] to <8 x i1> ; CHECK-NEXT: [[BST:%.*]] = bitcast i8 [[__U]] to <8 x i1> ; CHECK-NEXT: [[TMP6:%.*]] = select <8 x i1> [[BST]], <8 x i32> [[_MSPROP1]], <8 x i32> zeroinitializer @@ -413,28 +397,40 @@ define <8 x i32> @test_mm256_mask_dpbssd_epi32(i8 zeroext %__U, <8 x i32> %__W, ; CHECK-NEXT: store <8 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <8 x i32> [[RES]] ; - %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B) + %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32> %__W, <32 x i8> %__A, <32 x i8> %__B) %bst = bitcast i8 %__U to <8 x i1> %res = select <8 x i1> %bst, <8 x i32> %dpi, <8 x i32> zeroinitializer ret <8 x i32> %res } -declare <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32>, <4 x i32>, <4 x i32>) -declare <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32>, <4 x i32>, <4 x i32>) -declare <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32>, <8 x i32>, <8 x i32>) -declare <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32>, <8 x i32>, <8 x i32>) +declare <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32>, <16 x i8>, <16 x i8>) +declare <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32>, <16 x i8>, <16 x i8>) +declare <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32>, <32 x i8>, <32 x i8>) +declare <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32>, <32 x i8>, <32 x i8>) -define <4 x i32> @test_mm_mask_dpbsud_epi32(<4 x i32> %__W, i4 zeroext %__U, <4 x i32> %__A, <4 x i32> %__B) sanitize_memory { +define <4 x i32> @test_mm_mask_dpbsud_epi32(<4 x i32> %__W, i4 zeroext %__U, <16 x i8> %__A, <16 x i8> %__B) sanitize_memory { ; CHECK-LABEL: define <4 x i32> @test_mm_mask_dpbsud_epi32( -; CHECK-SAME: <4 x i32> [[__W:%.*]], i4 zeroext [[__U:%.*]], <4 x i32> [[__A:%.*]], <4 x i32> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-SAME: <4 x i32> [[__W:%.*]], i4 zeroext [[__U:%.*]], <16 x i8> [[__A:%.*]], <16 x i8> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[TMP3:%.*]] = load <16 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 24) to ptr), align 8 +; CHECK-NEXT: [[TMP2:%.*]] = load <16 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 ; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8 -; CHECK-NEXT: [[TMP2:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 24) to ptr), align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 ; CHECK-NEXT: [[TMP4:%.*]] = load i4, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 16) to ptr), align 8 ; CHECK-NEXT: call void @llvm.donothing() -; CHECK-NEXT: [[_MSPROP:%.*]] = or <4 x i32> [[TMP1]], [[TMP2]] -; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[_MSPROP]], [[TMP3]] -; CHECK-NEXT: [[DPI:%.*]] = tail call <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32> [[__W]], <4 x i32> [[__A]], <4 x i32> [[__B]]) +; CHECK-NEXT: [[TMP18:%.*]] = icmp ne <16 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP19:%.*]] = icmp ne <16 x i8> [[TMP2]], zeroinitializer +; CHECK-NEXT: [[TMP20:%.*]] = icmp ne <16 x i8> [[__A]], zeroinitializer +; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <16 x i8> [[__B]], zeroinitializer +; CHECK-NEXT: [[TMP22:%.*]] = and <16 x i1> [[TMP18]], [[TMP19]] +; CHECK-NEXT: [[TMP10:%.*]] = and <16 x i1> [[TMP20]], [[TMP19]] +; CHECK-NEXT: [[TMP11:%.*]] = and <16 x i1> [[TMP18]], [[TMP21]] +; CHECK-NEXT: [[TMP12:%.*]] = or <16 x i1> [[TMP22]], [[TMP10]] +; CHECK-NEXT: [[TMP13:%.*]] = or <16 x i1> [[TMP12]], [[TMP11]] +; CHECK-NEXT: [[TMP14:%.*]] = sext <16 x i1> [[TMP13]] to <16 x i8> +; CHECK-NEXT: [[TMP15:%.*]] = bitcast <16 x i8> [[TMP14]] to <4 x i32> +; CHECK-NEXT: [[TMP16:%.*]] = icmp ne <4 x i32> [[TMP15]], zeroinitializer +; CHECK-NEXT: [[TMP17:%.*]] = sext <4 x i1> [[TMP16]] to <4 x i32> +; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[TMP17]], [[TMP1]] +; CHECK-NEXT: [[DPI:%.*]] = tail call <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32> [[__W]], <16 x i8> [[__A]], <16 x i8> [[__B]]) ; CHECK-NEXT: [[TMP5:%.*]] = bitcast i4 [[TMP4]] to <4 x i1> ; CHECK-NEXT: [[BST:%.*]] = bitcast i4 [[__U]] to <4 x i1> ; CHECK-NEXT: [[TMP6:%.*]] = select <4 x i1> [[BST]], <4 x i32> [[_MSPROP1]], <4 x i32> [[TMP1]] @@ -446,23 +442,35 @@ define <4 x i32> @test_mm_mask_dpbsud_epi32(<4 x i32> %__W, i4 zeroext %__U, <4 ; CHECK-NEXT: store <4 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <4 x i32> [[RES]] ; - %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B) + %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32> %__W, <16 x i8> %__A, <16 x i8> %__B) %bst = bitcast i4 %__U to <4 x i1> %res = select <4 x i1> %bst, <4 x i32> %dpi, <4 x i32> %__W ret <4 x i32> %res } -define <4 x i32> @test_mm_maskz_dpbsuds_epi32(i4 zeroext %__U, <4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B) sanitize_memory { +define <4 x i32> @test_mm_maskz_dpbsuds_epi32(i4 zeroext %__U, <4 x i32> %__W, <16 x i8> %__A, <16 x i8> %__B) sanitize_memory { ; CHECK-LABEL: define <4 x i32> @test_mm_maskz_dpbsuds_epi32( -; CHECK-SAME: i4 zeroext [[__U:%.*]], <4 x i32> [[__W:%.*]], <4 x i32> [[__A:%.*]], <4 x i32> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-SAME: i4 zeroext [[__U:%.*]], <4 x i32> [[__W:%.*]], <16 x i8> [[__A:%.*]], <16 x i8> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[TMP3:%.*]] = load <16 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 24) to ptr), align 8 +; CHECK-NEXT: [[TMP2:%.*]] = load <16 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 ; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8 -; CHECK-NEXT: [[TMP2:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 24) to ptr), align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 ; CHECK-NEXT: [[TMP4:%.*]] = load i4, ptr @__msan_param_tls, align 8 ; CHECK-NEXT: call void @llvm.donothing() -; CHECK-NEXT: [[_MSPROP:%.*]] = or <4 x i32> [[TMP1]], [[TMP2]] -; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[_MSPROP]], [[TMP3]] -; CHECK-NEXT: [[DPI:%.*]] = tail call <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32> [[__W]], <4 x i32> [[__A]], <4 x i32> [[__B]]) +; CHECK-NEXT: [[TMP18:%.*]] = icmp ne <16 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP19:%.*]] = icmp ne <16 x i8> [[TMP2]], zeroinitializer +; CHECK-NEXT: [[TMP20:%.*]] = icmp ne <16 x i8> [[__A]], zeroinitializer +; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <16 x i8> [[__B]], zeroinitializer +; CHECK-NEXT: [[TMP22:%.*]] = and <16 x i1> [[TMP18]], [[TMP19]] +; CHECK-NEXT: [[TMP10:%.*]] = and <16 x i1> [[TMP20]], [[TMP19]] +; CHECK-NEXT: [[TMP11:%.*]] = and <16 x i1> [[TMP18]], [[TMP21]] +; CHECK-NEXT: [[TMP12:%.*]] = or <16 x i1> [[TMP22]], [[TMP10]] +; CHECK-NEXT: [[TMP13:%.*]] = or <16 x i1> [[TMP12]], [[TMP11]] +; CHECK-NEXT: [[TMP14:%.*]] = sext <16 x i1> [[TMP13]] to <16 x i8> +; CHECK-NEXT: [[TMP15:%.*]] = bitcast <16 x i8> [[TMP14]] to <4 x i32> +; CHECK-NEXT: [[TMP16:%.*]] = icmp ne <4 x i32> [[TMP15]], zeroinitializer +; CHECK-NEXT: [[TMP17:%.*]] = sext <4 x i1> [[TMP16]] to <4 x i32> +; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[TMP17]], [[TMP1]] +; CHECK-NEXT: [[DPI:%.*]] = tail call <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32> [[__W]], <16 x i8> [[__A]], <16 x i8> [[__B]]) ; CHECK-NEXT: [[TMP5:%.*]] = bitcast i4 [[TMP4]] to <4 x i1> ; CHECK-NEXT: [[BST:%.*]] = bitcast i4 [[__U]] to <4 x i1> ; CHECK-NEXT: [[TMP6:%.*]] = select <4 x i1> [[BST]], <4 x i32> [[_MSPROP1]], <4 x i32> zeroinitializer @@ -474,23 +482,35 @@ define <4 x i32> @test_mm_maskz_dpbsuds_epi32(i4 zeroext %__U, <4 x i32> %__W, < ; CHECK-NEXT: store <4 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <4 x i32> [[RES]] ; - %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B) + %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32> %__W, <16 x i8> %__A, <16 x i8> %__B) %bst = bitcast i4 %__U to <4 x i1> %res = select <4 x i1> %bst, <4 x i32> %dpi, <4 x i32> zeroinitializer ret <4 x i32> %res } -define <8 x i32> @test_mm256_maskz_dpbsuds_epi32(<8 x i32> %__W, i8 zeroext %__U, <8 x i32> %__A, <8 x i32> %__B) sanitize_memory { +define <8 x i32> @test_mm256_maskz_dpbsuds_epi32(<8 x i32> %__W, i8 zeroext %__U, <32 x i8> %__A, <32 x i8> %__B) sanitize_memory { ; CHECK-LABEL: define <8 x i32> @test_mm256_maskz_dpbsuds_epi32( -; CHECK-SAME: <8 x i32> [[__W:%.*]], i8 zeroext [[__U:%.*]], <8 x i32> [[__A:%.*]], <8 x i32> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-SAME: <8 x i32> [[__W:%.*]], i8 zeroext [[__U:%.*]], <32 x i8> [[__A:%.*]], <32 x i8> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[TMP3:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 +; CHECK-NEXT: [[TMP2:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 ; CHECK-NEXT: [[TMP1:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8 -; CHECK-NEXT: [[TMP2:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 ; CHECK-NEXT: [[TMP4:%.*]] = load i8, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8 ; CHECK-NEXT: call void @llvm.donothing() -; CHECK-NEXT: [[_MSPROP:%.*]] = or <8 x i32> [[TMP1]], [[TMP2]] -; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[_MSPROP]], [[TMP3]] -; CHECK-NEXT: [[DPI:%.*]] = tail call <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32> [[__W]], <8 x i32> [[__A]], <8 x i32> [[__B]]) +; CHECK-NEXT: [[TMP18:%.*]] = icmp ne <32 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP19:%.*]] = icmp ne <32 x i8> [[TMP2]], zeroinitializer +; CHECK-NEXT: [[TMP20:%.*]] = icmp ne <32 x i8> [[__A]], zeroinitializer +; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <32 x i8> [[__B]], zeroinitializer +; CHECK-NEXT: [[TMP22:%.*]] = and <32 x i1> [[TMP18]], [[TMP19]] +; CHECK-NEXT: [[TMP10:%.*]] = and <32 x i1> [[TMP20]], [[TMP19]] +; CHECK-NEXT: [[TMP11:%.*]] = and <32 x i1> [[TMP18]], [[TMP21]] +; CHECK-NEXT: [[TMP12:%.*]] = or <32 x i1> [[TMP22]], [[TMP10]] +; CHECK-NEXT: [[TMP13:%.*]] = or <32 x i1> [[TMP12]], [[TMP11]] +; CHECK-NEXT: [[TMP14:%.*]] = sext <32 x i1> [[TMP13]] to <32 x i8> +; CHECK-NEXT: [[TMP15:%.*]] = bitcast <32 x i8> [[TMP14]] to <8 x i32> +; CHECK-NEXT: [[TMP16:%.*]] = icmp ne <8 x i32> [[TMP15]], zeroinitializer +; CHECK-NEXT: [[TMP17:%.*]] = sext <8 x i1> [[TMP16]] to <8 x i32> +; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[TMP17]], [[TMP1]] +; CHECK-NEXT: [[DPI:%.*]] = tail call <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32> [[__W]], <32 x i8> [[__A]], <32 x i8> [[__B]]) ; CHECK-NEXT: [[TMP5:%.*]] = bitcast i8 [[TMP4]] to <8 x i1> ; CHECK-NEXT: [[BST:%.*]] = bitcast i8 [[__U]] to <8 x i1> ; CHECK-NEXT: [[TMP6:%.*]] = select <8 x i1> [[BST]], <8 x i32> [[_MSPROP1]], <8 x i32> [[TMP1]] @@ -502,23 +522,35 @@ define <8 x i32> @test_mm256_maskz_dpbsuds_epi32(<8 x i32> %__W, i8 zeroext %__U ; CHECK-NEXT: store <8 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <8 x i32> [[RES]] ; - %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B) + %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32> %__W, <32 x i8> %__A, <32 x i8> %__B) %bst = bitcast i8 %__U to <8 x i1> %res = select <8 x i1> %bst, <8 x i32> %dpi, <8 x i32> %__W ret <8 x i32> %res } -define <8 x i32> @test_mm256_mask_dpbsud_epi32(i8 zeroext %__U, <8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B) sanitize_memory { +define <8 x i32> @test_mm256_mask_dpbsud_epi32(i8 zeroext %__U, <8 x i32> %__W, <32 x i8> %__A, <32 x i8> %__B) sanitize_memory { ; CHECK-LABEL: define <8 x i32> @test_mm256_mask_dpbsud_epi32( -; CHECK-SAME: i8 zeroext [[__U:%.*]], <8 x i32> [[__W:%.*]], <8 x i32> [[__A:%.*]], <8 x i32> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-SAME: i8 zeroext [[__U:%.*]], <8 x i32> [[__W:%.*]], <32 x i8> [[__A:%.*]], <32 x i8> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[TMP3:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 +; CHECK-NEXT: [[TMP2:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 ; CHECK-NEXT: [[TMP1:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8 -; CHECK-NEXT: [[TMP2:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 ; CHECK-NEXT: [[TMP4:%.*]] = load i8, ptr @__msan_param_tls, align 8 ; CHECK-NEXT: call void @llvm.donothing() -; CHECK-NEXT: [[_MSPROP:%.*]] = or <8 x i32> [[TMP1]], [[TMP2]] -; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[_MSPROP]], [[TMP3]] -; CHECK-NEXT: [[DPI:%.*]] = tail call <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32> [[__W]], <8 x i32> [[__A]], <8 x i32> [[__B]]) +; CHECK-NEXT: [[TMP18:%.*]] = icmp ne <32 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP19:%.*]] = icmp ne <32 x i8> [[TMP2]], zeroinitializer +; CHECK-NEXT: [[TMP20:%.*]] = icmp ne <32 x i8> [[__A]], zeroinitializer +; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <32 x i8> [[__B]], zeroinitializer +; CHECK-NEXT: [[TMP22:%.*]] = and <32 x i1> [[TMP18]], [[TMP19]] +; CHECK-NEXT: [[TMP10:%.*]] = and <32 x i1> [[TMP20]], [[TMP19]] +; CHECK-NEXT: [[TMP11:%.*]] = and <32 x i1> [[TMP18]], [[TMP21]] +; CHECK-NEXT: [[TMP12:%.*]] = or <32 x i1> [[TMP22]], [[TMP10]] +; CHECK-NEXT: [[TMP13:%.*]] = or <32 x i1> [[TMP12]], [[TMP11]] +; CHECK-NEXT: [[TMP14:%.*]] = sext <32 x i1> [[TMP13]] to <32 x i8> +; CHECK-NEXT: [[TMP15:%.*]] = bitcast <32 x i8> [[TMP14]] to <8 x i32> +; CHECK-NEXT: [[TMP16:%.*]] = icmp ne <8 x i32> [[TMP15]], zeroinitializer +; CHECK-NEXT: [[TMP17:%.*]] = sext <8 x i1> [[TMP16]] to <8 x i32> +; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[TMP17]], [[TMP1]] +; CHECK-NEXT: [[DPI:%.*]] = tail call <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32> [[__W]], <32 x i8> [[__A]], <32 x i8> [[__B]]) ; CHECK-NEXT: [[TMP5:%.*]] = bitcast i8 [[TMP4]] to <8 x i1> ; CHECK-NEXT: [[BST:%.*]] = bitcast i8 [[__U]] to <8 x i1> ; CHECK-NEXT: [[TMP6:%.*]] = select <8 x i1> [[BST]], <8 x i32> [[_MSPROP1]], <8 x i32> zeroinitializer @@ -530,28 +562,40 @@ define <8 x i32> @test_mm256_mask_dpbsud_epi32(i8 zeroext %__U, <8 x i32> %__W, ; CHECK-NEXT: store <8 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <8 x i32> [[RES]] ; - %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B) + %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32> %__W, <32 x i8> %__A, <32 x i8> %__B) %bst = bitcast i8 %__U to <8 x i1> %res = select <8 x i1> %bst, <8 x i32> %dpi, <8 x i32> zeroinitializer ret <8 x i32> %res } -declare <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32>, <4 x i32>, <4 x i32>) -declare <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32>, <4 x i32>, <4 x i32>) -declare <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32>, <8 x i32>, <8 x i32>) -declare <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32>, <8 x i32>, <8 x i32>) +declare <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32>, <16 x i8>, <16 x i8>) +declare <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32>, <16 x i8>, <16 x i8>) +declare <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32>, <32 x i8>, <32 x i8>) +declare <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32>, <32 x i8>, <32 x i8>) -define <4 x i32> @test_mm_mask_dpbuud_epi32(<4 x i32> %__W, i4 zeroext %__U, <4 x i32> %__A, <4 x i32> %__B) sanitize_memory { +define <4 x i32> @test_mm_mask_dpbuud_epi32(<4 x i32> %__W, i4 zeroext %__U, <16 x i8> %__A, <16 x i8> %__B) sanitize_memory { ; CHECK-LABEL: define <4 x i32> @test_mm_mask_dpbuud_epi32( -; CHECK-SAME: <4 x i32> [[__W:%.*]], i4 zeroext [[__U:%.*]], <4 x i32> [[__A:%.*]], <4 x i32> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-SAME: <4 x i32> [[__W:%.*]], i4 zeroext [[__U:%.*]], <16 x i8> [[__A:%.*]], <16 x i8> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[TMP3:%.*]] = load <16 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 24) to ptr), align 8 +; CHECK-NEXT: [[TMP2:%.*]] = load <16 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 ; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8 -; CHECK-NEXT: [[TMP2:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 24) to ptr), align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 ; CHECK-NEXT: [[TMP4:%.*]] = load i4, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 16) to ptr), align 8 ; CHECK-NEXT: call void @llvm.donothing() -; CHECK-NEXT: [[_MSPROP:%.*]] = or <4 x i32> [[TMP1]], [[TMP2]] -; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[_MSPROP]], [[TMP3]] -; CHECK-NEXT: [[DPI:%.*]] = tail call <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32> [[__W]], <4 x i32> [[__A]], <4 x i32> [[__B]]) +; CHECK-NEXT: [[TMP18:%.*]] = icmp ne <16 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP19:%.*]] = icmp ne <16 x i8> [[TMP2]], zeroinitializer +; CHECK-NEXT: [[TMP20:%.*]] = icmp ne <16 x i8> [[__A]], zeroinitializer +; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <16 x i8> [[__B]], zeroinitializer +; CHECK-NEXT: [[TMP22:%.*]] = and <16 x i1> [[TMP18]], [[TMP19]] +; CHECK-NEXT: [[TMP10:%.*]] = and <16 x i1> [[TMP20]], [[TMP19]] +; CHECK-NEXT: [[TMP11:%.*]] = and <16 x i1> [[TMP18]], [[TMP21]] +; CHECK-NEXT: [[TMP12:%.*]] = or <16 x i1> [[TMP22]], [[TMP10]] +; CHECK-NEXT: [[TMP13:%.*]] = or <16 x i1> [[TMP12]], [[TMP11]] +; CHECK-NEXT: [[TMP14:%.*]] = sext <16 x i1> [[TMP13]] to <16 x i8> +; CHECK-NEXT: [[TMP15:%.*]] = bitcast <16 x i8> [[TMP14]] to <4 x i32> +; CHECK-NEXT: [[TMP16:%.*]] = icmp ne <4 x i32> [[TMP15]], zeroinitializer +; CHECK-NEXT: [[TMP17:%.*]] = sext <4 x i1> [[TMP16]] to <4 x i32> +; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[TMP17]], [[TMP1]] +; CHECK-NEXT: [[DPI:%.*]] = tail call <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32> [[__W]], <16 x i8> [[__A]], <16 x i8> [[__B]]) ; CHECK-NEXT: [[TMP5:%.*]] = bitcast i4 [[TMP4]] to <4 x i1> ; CHECK-NEXT: [[BST:%.*]] = bitcast i4 [[__U]] to <4 x i1> ; CHECK-NEXT: [[TMP6:%.*]] = select <4 x i1> [[BST]], <4 x i32> [[_MSPROP1]], <4 x i32> [[TMP1]] @@ -563,23 +607,35 @@ define <4 x i32> @test_mm_mask_dpbuud_epi32(<4 x i32> %__W, i4 zeroext %__U, <4 ; CHECK-NEXT: store <4 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <4 x i32> [[RES]] ; - %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B) + %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32> %__W, <16 x i8> %__A, <16 x i8> %__B) %bst = bitcast i4 %__U to <4 x i1> %res = select <4 x i1> %bst, <4 x i32> %dpi, <4 x i32> %__W ret <4 x i32> %res } -define <4 x i32> @test_mm_maskz_dpbuuds_epi32(i4 zeroext %__U, <4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B) sanitize_memory { +define <4 x i32> @test_mm_maskz_dpbuuds_epi32(i4 zeroext %__U, <4 x i32> %__W, <16 x i8> %__A, <16 x i8> %__B) sanitize_memory { ; CHECK-LABEL: define <4 x i32> @test_mm_maskz_dpbuuds_epi32( -; CHECK-SAME: i4 zeroext [[__U:%.*]], <4 x i32> [[__W:%.*]], <4 x i32> [[__A:%.*]], <4 x i32> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-SAME: i4 zeroext [[__U:%.*]], <4 x i32> [[__W:%.*]], <16 x i8> [[__A:%.*]], <16 x i8> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[TMP3:%.*]] = load <16 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 24) to ptr), align 8 +; CHECK-NEXT: [[TMP2:%.*]] = load <16 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 ; CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8 -; CHECK-NEXT: [[TMP2:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 24) to ptr), align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 ; CHECK-NEXT: [[TMP4:%.*]] = load i4, ptr @__msan_param_tls, align 8 ; CHECK-NEXT: call void @llvm.donothing() -; CHECK-NEXT: [[_MSPROP:%.*]] = or <4 x i32> [[TMP1]], [[TMP2]] -; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[_MSPROP]], [[TMP3]] -; CHECK-NEXT: [[DPI:%.*]] = tail call <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32> [[__W]], <4 x i32> [[__A]], <4 x i32> [[__B]]) +; CHECK-NEXT: [[TMP18:%.*]] = icmp ne <16 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP19:%.*]] = icmp ne <16 x i8> [[TMP2]], zeroinitializer +; CHECK-NEXT: [[TMP20:%.*]] = icmp ne <16 x i8> [[__A]], zeroinitializer +; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <16 x i8> [[__B]], zeroinitializer +; CHECK-NEXT: [[TMP22:%.*]] = and <16 x i1> [[TMP18]], [[TMP19]] +; CHECK-NEXT: [[TMP10:%.*]] = and <16 x i1> [[TMP20]], [[TMP19]] +; CHECK-NEXT: [[TMP11:%.*]] = and <16 x i1> [[TMP18]], [[TMP21]] +; CHECK-NEXT: [[TMP12:%.*]] = or <16 x i1> [[TMP22]], [[TMP10]] +; CHECK-NEXT: [[TMP13:%.*]] = or <16 x i1> [[TMP12]], [[TMP11]] +; CHECK-NEXT: [[TMP14:%.*]] = sext <16 x i1> [[TMP13]] to <16 x i8> +; CHECK-NEXT: [[TMP15:%.*]] = bitcast <16 x i8> [[TMP14]] to <4 x i32> +; CHECK-NEXT: [[TMP16:%.*]] = icmp ne <4 x i32> [[TMP15]], zeroinitializer +; CHECK-NEXT: [[TMP17:%.*]] = sext <4 x i1> [[TMP16]] to <4 x i32> +; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[TMP17]], [[TMP1]] +; CHECK-NEXT: [[DPI:%.*]] = tail call <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32> [[__W]], <16 x i8> [[__A]], <16 x i8> [[__B]]) ; CHECK-NEXT: [[TMP5:%.*]] = bitcast i4 [[TMP4]] to <4 x i1> ; CHECK-NEXT: [[BST:%.*]] = bitcast i4 [[__U]] to <4 x i1> ; CHECK-NEXT: [[TMP6:%.*]] = select <4 x i1> [[BST]], <4 x i32> [[_MSPROP1]], <4 x i32> zeroinitializer @@ -591,23 +647,35 @@ define <4 x i32> @test_mm_maskz_dpbuuds_epi32(i4 zeroext %__U, <4 x i32> %__W, < ; CHECK-NEXT: store <4 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <4 x i32> [[RES]] ; - %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32> %__W, <4 x i32> %__A, <4 x i32> %__B) + %dpi = tail call <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32> %__W, <16 x i8> %__A, <16 x i8> %__B) %bst = bitcast i4 %__U to <4 x i1> %res = select <4 x i1> %bst, <4 x i32> %dpi, <4 x i32> zeroinitializer ret <4 x i32> %res } -define <8 x i32> @test_mm256_maskz_dpbuuds_epi32(<8 x i32> %__W, i8 zeroext %__U, <8 x i32> %__A, <8 x i32> %__B) sanitize_memory { +define <8 x i32> @test_mm256_maskz_dpbuuds_epi32(<8 x i32> %__W, i8 zeroext %__U, <32 x i8> %__A, <32 x i8> %__B) sanitize_memory { ; CHECK-LABEL: define <8 x i32> @test_mm256_maskz_dpbuuds_epi32( -; CHECK-SAME: <8 x i32> [[__W:%.*]], i8 zeroext [[__U:%.*]], <8 x i32> [[__A:%.*]], <8 x i32> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-SAME: <8 x i32> [[__W:%.*]], i8 zeroext [[__U:%.*]], <32 x i8> [[__A:%.*]], <32 x i8> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[TMP3:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 +; CHECK-NEXT: [[TMP2:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 ; CHECK-NEXT: [[TMP1:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8 -; CHECK-NEXT: [[TMP2:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 ; CHECK-NEXT: [[TMP4:%.*]] = load i8, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8 ; CHECK-NEXT: call void @llvm.donothing() -; CHECK-NEXT: [[_MSPROP:%.*]] = or <8 x i32> [[TMP1]], [[TMP2]] -; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[_MSPROP]], [[TMP3]] -; CHECK-NEXT: [[DPI:%.*]] = tail call <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32> [[__W]], <8 x i32> [[__A]], <8 x i32> [[__B]]) +; CHECK-NEXT: [[TMP18:%.*]] = icmp ne <32 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP19:%.*]] = icmp ne <32 x i8> [[TMP2]], zeroinitializer +; CHECK-NEXT: [[TMP20:%.*]] = icmp ne <32 x i8> [[__A]], zeroinitializer +; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <32 x i8> [[__B]], zeroinitializer +; CHECK-NEXT: [[TMP22:%.*]] = and <32 x i1> [[TMP18]], [[TMP19]] +; CHECK-NEXT: [[TMP10:%.*]] = and <32 x i1> [[TMP20]], [[TMP19]] +; CHECK-NEXT: [[TMP11:%.*]] = and <32 x i1> [[TMP18]], [[TMP21]] +; CHECK-NEXT: [[TMP12:%.*]] = or <32 x i1> [[TMP22]], [[TMP10]] +; CHECK-NEXT: [[TMP13:%.*]] = or <32 x i1> [[TMP12]], [[TMP11]] +; CHECK-NEXT: [[TMP14:%.*]] = sext <32 x i1> [[TMP13]] to <32 x i8> +; CHECK-NEXT: [[TMP15:%.*]] = bitcast <32 x i8> [[TMP14]] to <8 x i32> +; CHECK-NEXT: [[TMP16:%.*]] = icmp ne <8 x i32> [[TMP15]], zeroinitializer +; CHECK-NEXT: [[TMP17:%.*]] = sext <8 x i1> [[TMP16]] to <8 x i32> +; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[TMP17]], [[TMP1]] +; CHECK-NEXT: [[DPI:%.*]] = tail call <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32> [[__W]], <32 x i8> [[__A]], <32 x i8> [[__B]]) ; CHECK-NEXT: [[TMP5:%.*]] = bitcast i8 [[TMP4]] to <8 x i1> ; CHECK-NEXT: [[BST:%.*]] = bitcast i8 [[__U]] to <8 x i1> ; CHECK-NEXT: [[TMP6:%.*]] = select <8 x i1> [[BST]], <8 x i32> [[_MSPROP1]], <8 x i32> [[TMP1]] @@ -619,23 +687,35 @@ define <8 x i32> @test_mm256_maskz_dpbuuds_epi32(<8 x i32> %__W, i8 zeroext %__U ; CHECK-NEXT: store <8 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <8 x i32> [[RES]] ; - %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B) + %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32> %__W, <32 x i8> %__A, <32 x i8> %__B) %bst = bitcast i8 %__U to <8 x i1> %res = select <8 x i1> %bst, <8 x i32> %dpi, <8 x i32> %__W ret <8 x i32> %res } -define <8 x i32> @test_mm256_mask_dpbuud_epi32(i8 zeroext %__U, <8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B) sanitize_memory { +define <8 x i32> @test_mm256_mask_dpbuud_epi32(i8 zeroext %__U, <8 x i32> %__W, <32 x i8> %__A, <32 x i8> %__B) sanitize_memory { ; CHECK-LABEL: define <8 x i32> @test_mm256_mask_dpbuud_epi32( -; CHECK-SAME: i8 zeroext [[__U:%.*]], <8 x i32> [[__W:%.*]], <8 x i32> [[__A:%.*]], <8 x i32> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-SAME: i8 zeroext [[__U:%.*]], <8 x i32> [[__W:%.*]], <32 x i8> [[__A:%.*]], <32 x i8> [[__B:%.*]]) #[[ATTR0]] { +; CHECK-NEXT: [[TMP3:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 +; CHECK-NEXT: [[TMP2:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 ; CHECK-NEXT: [[TMP1:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 8) to ptr), align 8 -; CHECK-NEXT: [[TMP2:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 ; CHECK-NEXT: [[TMP4:%.*]] = load i8, ptr @__msan_param_tls, align 8 ; CHECK-NEXT: call void @llvm.donothing() -; CHECK-NEXT: [[_MSPROP:%.*]] = or <8 x i32> [[TMP1]], [[TMP2]] -; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[_MSPROP]], [[TMP3]] -; CHECK-NEXT: [[DPI:%.*]] = tail call <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32> [[__W]], <8 x i32> [[__A]], <8 x i32> [[__B]]) +; CHECK-NEXT: [[TMP18:%.*]] = icmp ne <32 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP19:%.*]] = icmp ne <32 x i8> [[TMP2]], zeroinitializer +; CHECK-NEXT: [[TMP20:%.*]] = icmp ne <32 x i8> [[__A]], zeroinitializer +; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <32 x i8> [[__B]], zeroinitializer +; CHECK-NEXT: [[TMP22:%.*]] = and <32 x i1> [[TMP18]], [[TMP19]] +; CHECK-NEXT: [[TMP10:%.*]] = and <32 x i1> [[TMP20]], [[TMP19]] +; CHECK-NEXT: [[TMP11:%.*]] = and <32 x i1> [[TMP18]], [[TMP21]] +; CHECK-NEXT: [[TMP12:%.*]] = or <32 x i1> [[TMP22]], [[TMP10]] +; CHECK-NEXT: [[TMP13:%.*]] = or <32 x i1> [[TMP12]], [[TMP11]] +; CHECK-NEXT: [[TMP14:%.*]] = sext <32 x i1> [[TMP13]] to <32 x i8> +; CHECK-NEXT: [[TMP15:%.*]] = bitcast <32 x i8> [[TMP14]] to <8 x i32> +; CHECK-NEXT: [[TMP16:%.*]] = icmp ne <8 x i32> [[TMP15]], zeroinitializer +; CHECK-NEXT: [[TMP17:%.*]] = sext <8 x i1> [[TMP16]] to <8 x i32> +; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[TMP17]], [[TMP1]] +; CHECK-NEXT: [[DPI:%.*]] = tail call <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32> [[__W]], <32 x i8> [[__A]], <32 x i8> [[__B]]) ; CHECK-NEXT: [[TMP5:%.*]] = bitcast i8 [[TMP4]] to <8 x i1> ; CHECK-NEXT: [[BST:%.*]] = bitcast i8 [[__U]] to <8 x i1> ; CHECK-NEXT: [[TMP6:%.*]] = select <8 x i1> [[BST]], <8 x i32> [[_MSPROP1]], <8 x i32> zeroinitializer @@ -647,16 +727,16 @@ define <8 x i32> @test_mm256_mask_dpbuud_epi32(i8 zeroext %__U, <8 x i32> %__W, ; CHECK-NEXT: store <8 x i32> [[_MSPROP_SELECT]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <8 x i32> [[RES]] ; - %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32> %__W, <8 x i32> %__A, <8 x i32> %__B) + %dpi = tail call <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32> %__W, <32 x i8> %__A, <32 x i8> %__B) %bst = bitcast i8 %__U to <8 x i1> %res = select <8 x i1> %bst, <8 x i32> %dpi, <8 x i32> zeroinitializer ret <8 x i32> %res } -declare <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32>, <4 x i32>, <4 x i32>) -declare <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32>, <4 x i32>, <4 x i32>) -declare <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32>, <8 x i32>, <8 x i32>) -declare <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32>, <8 x i32>, <8 x i32>) +declare <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32>, <16 x i8>, <16 x i8>) +declare <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32>, <16 x i8>, <16 x i8>) +declare <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32>, <32 x i8>, <32 x i8>) +declare <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32>, <32 x i8>, <32 x i8>) define <4 x i32> @test_mm_mask_dpwsud_epi32(<4 x i32> %__W, i4 zeroext %__U, <4 x i32> %__A, <4 x i32> %__B) sanitize_memory { diff --git a/llvm/test/Instrumentation/MemorySanitizer/X86/avxvnniint8-intrinsics.ll b/llvm/test/Instrumentation/MemorySanitizer/X86/avxvnniint8-intrinsics.ll index 3df0f1d..d91abea 100644 --- a/llvm/test/Instrumentation/MemorySanitizer/X86/avxvnniint8-intrinsics.ll +++ b/llvm/test/Instrumentation/MemorySanitizer/X86/avxvnniint8-intrinsics.ll @@ -10,15 +10,15 @@ target datalayout = "e-m:o-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128" target triple = "x86_64-unknown-linux-gnu" -declare <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32>, <4 x i32>, <4 x i32>) +declare <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32>, <16 x i8>, <16 x i8>) -define <4 x i32>@test_int_x86_avx2_vpdpbssd_128(<4 x i32> %x0, <4 x i32> %x1, ptr %x2p, <4 x i32> %x4) sanitize_memory { +define <4 x i32>@test_int_x86_avx2_vpdpbssd_128(<4 x i32> %x0, <16 x i8> %x1, ptr %x2p, <16 x i8> %x4) sanitize_memory { ; CHECK-LABEL: define <4 x i32> @test_int_x86_avx2_vpdpbssd_128( -; CHECK-SAME: <4 x i32> [[X0:%.*]], <4 x i32> [[X1:%.*]], ptr [[X2P:%.*]], <4 x i32> [[X4:%.*]]) #[[ATTR1:[0-9]+]] { +; CHECK-SAME: <4 x i32> [[X0:%.*]], <16 x i8> [[X1:%.*]], ptr [[X2P:%.*]], <16 x i8> [[X4:%.*]]) #[[ATTR1:[0-9]+]] { ; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 16) to ptr), align 8 +; CHECK-NEXT: [[TMP13:%.*]] = load <16 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 16) to ptr), align 8 ; CHECK-NEXT: [[TMP5:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8 -; CHECK-NEXT: [[TMP4:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 +; CHECK-NEXT: [[TMP32:%.*]] = load <16 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 ; CHECK-NEXT: call void @llvm.donothing() ; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0 ; CHECK-NEXT: br i1 [[_MSCMP]], label %[[BB5:.*]], label %[[BB6:.*]], !prof [[PROF1:![0-9]+]] @@ -26,22 +26,18 @@ define <4 x i32>@test_int_x86_avx2_vpdpbssd_128(<4 x i32> %x0, <4 x i32> %x1, pt ; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4:[0-9]+]] ; CHECK-NEXT: unreachable ; CHECK: [[BB6]]: -; CHECK-NEXT: [[X2:%.*]] = load <4 x i32>, ptr [[X2P]], align 16 +; CHECK-NEXT: [[TMP30:%.*]] = load <16 x i8>, ptr [[X2P]], align 16 ; CHECK-NEXT: [[TMP7:%.*]] = ptrtoint ptr [[X2P]] to i64 ; CHECK-NEXT: [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080 ; CHECK-NEXT: [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr -; CHECK-NEXT: [[_MSLD:%.*]] = load <4 x i32>, ptr [[TMP9]], align 16 -; CHECK-NEXT: [[TMP29:%.*]] = bitcast <4 x i32> [[X1]] to <16 x i8> -; CHECK-NEXT: [[TMP30:%.*]] = bitcast <4 x i32> [[X2]] to <16 x i8> -; CHECK-NEXT: [[TMP12:%.*]] = bitcast <4 x i32> [[TMP3]] to <16 x i8> -; CHECK-NEXT: [[TMP13:%.*]] = bitcast <4 x i32> [[_MSLD]] to <16 x i8> -; CHECK-NEXT: [[TMP14:%.*]] = icmp ne <16 x i8> [[TMP12]], zeroinitializer +; CHECK-NEXT: [[TMP29:%.*]] = load <16 x i8>, ptr [[TMP9]], align 16 ; CHECK-NEXT: [[TMP15:%.*]] = icmp ne <16 x i8> [[TMP13]], zeroinitializer ; CHECK-NEXT: [[TMP16:%.*]] = icmp ne <16 x i8> [[TMP29]], zeroinitializer +; CHECK-NEXT: [[TMP12:%.*]] = icmp ne <16 x i8> [[X1]], zeroinitializer ; CHECK-NEXT: [[TMP17:%.*]] = icmp ne <16 x i8> [[TMP30]], zeroinitializer -; CHECK-NEXT: [[TMP18:%.*]] = and <16 x i1> [[TMP14]], [[TMP15]] -; CHECK-NEXT: [[TMP19:%.*]] = and <16 x i1> [[TMP16]], [[TMP15]] -; CHECK-NEXT: [[TMP20:%.*]] = and <16 x i1> [[TMP14]], [[TMP17]] +; CHECK-NEXT: [[TMP18:%.*]] = and <16 x i1> [[TMP15]], [[TMP16]] +; CHECK-NEXT: [[TMP19:%.*]] = and <16 x i1> [[TMP12]], [[TMP16]] +; CHECK-NEXT: [[TMP20:%.*]] = and <16 x i1> [[TMP15]], [[TMP17]] ; CHECK-NEXT: [[TMP21:%.*]] = or <16 x i1> [[TMP18]], [[TMP19]] ; CHECK-NEXT: [[TMP22:%.*]] = or <16 x i1> [[TMP21]], [[TMP20]] ; CHECK-NEXT: [[TMP23:%.*]] = sext <16 x i1> [[TMP22]] to <16 x i8> @@ -49,18 +45,14 @@ define <4 x i32>@test_int_x86_avx2_vpdpbssd_128(<4 x i32> %x0, <4 x i32> %x1, pt ; CHECK-NEXT: [[TMP25:%.*]] = icmp ne <4 x i32> [[TMP24]], zeroinitializer ; CHECK-NEXT: [[TMP28:%.*]] = sext <4 x i1> [[TMP25]] to <4 x i32> ; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[TMP28]], [[TMP5]] -; CHECK-NEXT: [[TMP10:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X2]]) -; CHECK-NEXT: [[TMP31:%.*]] = bitcast <4 x i32> [[X1]] to <16 x i8> -; CHECK-NEXT: [[TMP32:%.*]] = bitcast <4 x i32> [[X4]] to <16 x i8> -; CHECK-NEXT: [[TMP33:%.*]] = bitcast <4 x i32> [[TMP3]] to <16 x i8> -; CHECK-NEXT: [[TMP34:%.*]] = bitcast <4 x i32> [[TMP4]] to <16 x i8> -; CHECK-NEXT: [[TMP35:%.*]] = icmp ne <16 x i8> [[TMP33]], zeroinitializer -; CHECK-NEXT: [[TMP36:%.*]] = icmp ne <16 x i8> [[TMP34]], zeroinitializer -; CHECK-NEXT: [[TMP37:%.*]] = icmp ne <16 x i8> [[TMP31]], zeroinitializer +; CHECK-NEXT: [[TMP10:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32> [[X0]], <16 x i8> [[X1]], <16 x i8> [[TMP30]]) +; CHECK-NEXT: [[TMP26:%.*]] = icmp ne <16 x i8> [[TMP13]], zeroinitializer ; CHECK-NEXT: [[TMP38:%.*]] = icmp ne <16 x i8> [[TMP32]], zeroinitializer -; CHECK-NEXT: [[TMP39:%.*]] = and <16 x i1> [[TMP35]], [[TMP36]] -; CHECK-NEXT: [[TMP40:%.*]] = and <16 x i1> [[TMP37]], [[TMP36]] -; CHECK-NEXT: [[TMP41:%.*]] = and <16 x i1> [[TMP35]], [[TMP38]] +; CHECK-NEXT: [[TMP27:%.*]] = icmp ne <16 x i8> [[X1]], zeroinitializer +; CHECK-NEXT: [[TMP31:%.*]] = icmp ne <16 x i8> [[X4]], zeroinitializer +; CHECK-NEXT: [[TMP39:%.*]] = and <16 x i1> [[TMP26]], [[TMP38]] +; CHECK-NEXT: [[TMP40:%.*]] = and <16 x i1> [[TMP27]], [[TMP38]] +; CHECK-NEXT: [[TMP41:%.*]] = and <16 x i1> [[TMP26]], [[TMP31]] ; CHECK-NEXT: [[TMP42:%.*]] = or <16 x i1> [[TMP39]], [[TMP40]] ; CHECK-NEXT: [[TMP43:%.*]] = or <16 x i1> [[TMP42]], [[TMP41]] ; CHECK-NEXT: [[TMP44:%.*]] = sext <16 x i1> [[TMP43]] to <16 x i8> @@ -68,28 +60,28 @@ define <4 x i32>@test_int_x86_avx2_vpdpbssd_128(<4 x i32> %x0, <4 x i32> %x1, pt ; CHECK-NEXT: [[TMP46:%.*]] = icmp ne <4 x i32> [[TMP45]], zeroinitializer ; CHECK-NEXT: [[TMP49:%.*]] = sext <4 x i1> [[TMP46]] to <4 x i32> ; CHECK-NEXT: [[_MSPROP3:%.*]] = or <4 x i32> [[TMP49]], [[TMP5]] -; CHECK-NEXT: [[TMP11:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X4]]) +; CHECK-NEXT: [[TMP11:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32> [[X0]], <16 x i8> [[X1]], <16 x i8> [[X4]]) ; CHECK-NEXT: [[_MSPROP4:%.*]] = or <4 x i32> [[_MSPROP1]], [[_MSPROP3]] ; CHECK-NEXT: [[RES:%.*]] = add <4 x i32> [[TMP10]], [[TMP11]] ; CHECK-NEXT: store <4 x i32> [[_MSPROP4]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <4 x i32> [[RES]] ; - %x2 = load <4 x i32>, ptr %x2p - %1 = call <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) - %2 = call <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x4) + %x2 = load <16 x i8>, ptr %x2p + %1 = call <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x2) + %2 = call <4 x i32> @llvm.x86.avx2.vpdpbssd.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x4) %res = add <4 x i32> %1, %2 ret <4 x i32> %res } -declare <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32>, <4 x i32>, <4 x i32>) +declare <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32>, <16 x i8>, <16 x i8>) -define <4 x i32>@test_int_x86_avx2_vpdpbssds_128(<4 x i32> %x0, <4 x i32> %x1, ptr %x2p, <4 x i32> %x4) sanitize_memory { +define <4 x i32>@test_int_x86_avx2_vpdpbssds_128(<4 x i32> %x0, <16 x i8> %x1, ptr %x2p, <16 x i8> %x4) sanitize_memory { ; CHECK-LABEL: define <4 x i32> @test_int_x86_avx2_vpdpbssds_128( -; CHECK-SAME: <4 x i32> [[X0:%.*]], <4 x i32> [[X1:%.*]], ptr [[X2P:%.*]], <4 x i32> [[X4:%.*]]) #[[ATTR1]] { +; CHECK-SAME: <4 x i32> [[X0:%.*]], <16 x i8> [[X1:%.*]], ptr [[X2P:%.*]], <16 x i8> [[X4:%.*]]) #[[ATTR1]] { ; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 16) to ptr), align 8 +; CHECK-NEXT: [[TMP13:%.*]] = load <16 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 16) to ptr), align 8 ; CHECK-NEXT: [[TMP5:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8 -; CHECK-NEXT: [[TMP4:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 +; CHECK-NEXT: [[TMP32:%.*]] = load <16 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 ; CHECK-NEXT: call void @llvm.donothing() ; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0 ; CHECK-NEXT: br i1 [[_MSCMP]], label %[[BB5:.*]], label %[[BB6:.*]], !prof [[PROF1]] @@ -97,22 +89,18 @@ define <4 x i32>@test_int_x86_avx2_vpdpbssds_128(<4 x i32> %x0, <4 x i32> %x1, p ; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4]] ; CHECK-NEXT: unreachable ; CHECK: [[BB6]]: -; CHECK-NEXT: [[X2:%.*]] = load <4 x i32>, ptr [[X2P]], align 16 +; CHECK-NEXT: [[TMP30:%.*]] = load <16 x i8>, ptr [[X2P]], align 16 ; CHECK-NEXT: [[TMP7:%.*]] = ptrtoint ptr [[X2P]] to i64 ; CHECK-NEXT: [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080 ; CHECK-NEXT: [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr -; CHECK-NEXT: [[_MSLD:%.*]] = load <4 x i32>, ptr [[TMP9]], align 16 -; CHECK-NEXT: [[TMP29:%.*]] = bitcast <4 x i32> [[X1]] to <16 x i8> -; CHECK-NEXT: [[TMP30:%.*]] = bitcast <4 x i32> [[X2]] to <16 x i8> -; CHECK-NEXT: [[TMP12:%.*]] = bitcast <4 x i32> [[TMP3]] to <16 x i8> -; CHECK-NEXT: [[TMP13:%.*]] = bitcast <4 x i32> [[_MSLD]] to <16 x i8> -; CHECK-NEXT: [[TMP14:%.*]] = icmp ne <16 x i8> [[TMP12]], zeroinitializer +; CHECK-NEXT: [[TMP29:%.*]] = load <16 x i8>, ptr [[TMP9]], align 16 ; CHECK-NEXT: [[TMP15:%.*]] = icmp ne <16 x i8> [[TMP13]], zeroinitializer ; CHECK-NEXT: [[TMP16:%.*]] = icmp ne <16 x i8> [[TMP29]], zeroinitializer +; CHECK-NEXT: [[TMP12:%.*]] = icmp ne <16 x i8> [[X1]], zeroinitializer ; CHECK-NEXT: [[TMP17:%.*]] = icmp ne <16 x i8> [[TMP30]], zeroinitializer -; CHECK-NEXT: [[TMP18:%.*]] = and <16 x i1> [[TMP14]], [[TMP15]] -; CHECK-NEXT: [[TMP19:%.*]] = and <16 x i1> [[TMP16]], [[TMP15]] -; CHECK-NEXT: [[TMP20:%.*]] = and <16 x i1> [[TMP14]], [[TMP17]] +; CHECK-NEXT: [[TMP18:%.*]] = and <16 x i1> [[TMP15]], [[TMP16]] +; CHECK-NEXT: [[TMP19:%.*]] = and <16 x i1> [[TMP12]], [[TMP16]] +; CHECK-NEXT: [[TMP20:%.*]] = and <16 x i1> [[TMP15]], [[TMP17]] ; CHECK-NEXT: [[TMP21:%.*]] = or <16 x i1> [[TMP18]], [[TMP19]] ; CHECK-NEXT: [[TMP22:%.*]] = or <16 x i1> [[TMP21]], [[TMP20]] ; CHECK-NEXT: [[TMP23:%.*]] = sext <16 x i1> [[TMP22]] to <16 x i8> @@ -120,18 +108,14 @@ define <4 x i32>@test_int_x86_avx2_vpdpbssds_128(<4 x i32> %x0, <4 x i32> %x1, p ; CHECK-NEXT: [[TMP25:%.*]] = icmp ne <4 x i32> [[TMP24]], zeroinitializer ; CHECK-NEXT: [[TMP28:%.*]] = sext <4 x i1> [[TMP25]] to <4 x i32> ; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[TMP28]], [[TMP5]] -; CHECK-NEXT: [[TMP10:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X2]]) -; CHECK-NEXT: [[TMP31:%.*]] = bitcast <4 x i32> [[X1]] to <16 x i8> -; CHECK-NEXT: [[TMP32:%.*]] = bitcast <4 x i32> [[X4]] to <16 x i8> -; CHECK-NEXT: [[TMP33:%.*]] = bitcast <4 x i32> [[TMP3]] to <16 x i8> -; CHECK-NEXT: [[TMP34:%.*]] = bitcast <4 x i32> [[TMP4]] to <16 x i8> -; CHECK-NEXT: [[TMP35:%.*]] = icmp ne <16 x i8> [[TMP33]], zeroinitializer -; CHECK-NEXT: [[TMP36:%.*]] = icmp ne <16 x i8> [[TMP34]], zeroinitializer -; CHECK-NEXT: [[TMP37:%.*]] = icmp ne <16 x i8> [[TMP31]], zeroinitializer +; CHECK-NEXT: [[TMP10:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32> [[X0]], <16 x i8> [[X1]], <16 x i8> [[TMP30]]) +; CHECK-NEXT: [[TMP26:%.*]] = icmp ne <16 x i8> [[TMP13]], zeroinitializer ; CHECK-NEXT: [[TMP38:%.*]] = icmp ne <16 x i8> [[TMP32]], zeroinitializer -; CHECK-NEXT: [[TMP39:%.*]] = and <16 x i1> [[TMP35]], [[TMP36]] -; CHECK-NEXT: [[TMP40:%.*]] = and <16 x i1> [[TMP37]], [[TMP36]] -; CHECK-NEXT: [[TMP41:%.*]] = and <16 x i1> [[TMP35]], [[TMP38]] +; CHECK-NEXT: [[TMP27:%.*]] = icmp ne <16 x i8> [[X1]], zeroinitializer +; CHECK-NEXT: [[TMP31:%.*]] = icmp ne <16 x i8> [[X4]], zeroinitializer +; CHECK-NEXT: [[TMP39:%.*]] = and <16 x i1> [[TMP26]], [[TMP38]] +; CHECK-NEXT: [[TMP40:%.*]] = and <16 x i1> [[TMP27]], [[TMP38]] +; CHECK-NEXT: [[TMP41:%.*]] = and <16 x i1> [[TMP26]], [[TMP31]] ; CHECK-NEXT: [[TMP42:%.*]] = or <16 x i1> [[TMP39]], [[TMP40]] ; CHECK-NEXT: [[TMP43:%.*]] = or <16 x i1> [[TMP42]], [[TMP41]] ; CHECK-NEXT: [[TMP44:%.*]] = sext <16 x i1> [[TMP43]] to <16 x i8> @@ -139,28 +123,28 @@ define <4 x i32>@test_int_x86_avx2_vpdpbssds_128(<4 x i32> %x0, <4 x i32> %x1, p ; CHECK-NEXT: [[TMP46:%.*]] = icmp ne <4 x i32> [[TMP45]], zeroinitializer ; CHECK-NEXT: [[TMP49:%.*]] = sext <4 x i1> [[TMP46]] to <4 x i32> ; CHECK-NEXT: [[_MSPROP3:%.*]] = or <4 x i32> [[TMP49]], [[TMP5]] -; CHECK-NEXT: [[TMP11:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X4]]) +; CHECK-NEXT: [[TMP11:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32> [[X0]], <16 x i8> [[X1]], <16 x i8> [[X4]]) ; CHECK-NEXT: [[_MSPROP4:%.*]] = or <4 x i32> [[_MSPROP1]], [[_MSPROP3]] ; CHECK-NEXT: [[RES:%.*]] = add <4 x i32> [[TMP10]], [[TMP11]] ; CHECK-NEXT: store <4 x i32> [[_MSPROP4]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <4 x i32> [[RES]] ; - %x2 = load <4 x i32>, ptr %x2p - %1 = call <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) - %2 = call <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x4) + %x2 = load <16 x i8>, ptr %x2p + %1 = call <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x2) + %2 = call <4 x i32> @llvm.x86.avx2.vpdpbssds.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x4) %res = add <4 x i32> %1, %2 ret <4 x i32> %res } -declare <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32>, <8 x i32>, <8 x i32>) +declare <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32>, <32 x i8>, <32 x i8>) -define <8 x i32>@test_int_x86_avx2_vpdpbssd_256(<8 x i32> %x0, <8 x i32> %x1, ptr %x2p, <8 x i32> %x4) sanitize_memory { +define <8 x i32>@test_int_x86_avx2_vpdpbssd_256(<8 x i32> %x0, <32 x i8> %x1, ptr %x2p, <32 x i8> %x4) sanitize_memory { ; CHECK-LABEL: define <8 x i32> @test_int_x86_avx2_vpdpbssd_256( -; CHECK-SAME: <8 x i32> [[X0:%.*]], <8 x i32> [[X1:%.*]], ptr [[X2P:%.*]], <8 x i32> [[X4:%.*]]) #[[ATTR1]] { +; CHECK-SAME: <8 x i32> [[X0:%.*]], <32 x i8> [[X1:%.*]], ptr [[X2P:%.*]], <32 x i8> [[X4:%.*]]) #[[ATTR1]] { ; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8 +; CHECK-NEXT: [[TMP13:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8 ; CHECK-NEXT: [[TMP5:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8 -; CHECK-NEXT: [[TMP4:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 +; CHECK-NEXT: [[TMP32:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 ; CHECK-NEXT: call void @llvm.donothing() ; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0 ; CHECK-NEXT: br i1 [[_MSCMP]], label %[[BB5:.*]], label %[[BB6:.*]], !prof [[PROF1]] @@ -168,22 +152,18 @@ define <8 x i32>@test_int_x86_avx2_vpdpbssd_256(<8 x i32> %x0, <8 x i32> %x1, pt ; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4]] ; CHECK-NEXT: unreachable ; CHECK: [[BB6]]: -; CHECK-NEXT: [[X2:%.*]] = load <8 x i32>, ptr [[X2P]], align 32 +; CHECK-NEXT: [[TMP30:%.*]] = load <32 x i8>, ptr [[X2P]], align 32 ; CHECK-NEXT: [[TMP7:%.*]] = ptrtoint ptr [[X2P]] to i64 ; CHECK-NEXT: [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080 ; CHECK-NEXT: [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr -; CHECK-NEXT: [[_MSLD:%.*]] = load <8 x i32>, ptr [[TMP9]], align 32 -; CHECK-NEXT: [[TMP29:%.*]] = bitcast <8 x i32> [[X1]] to <32 x i8> -; CHECK-NEXT: [[TMP30:%.*]] = bitcast <8 x i32> [[X2]] to <32 x i8> -; CHECK-NEXT: [[TMP12:%.*]] = bitcast <8 x i32> [[TMP3]] to <32 x i8> -; CHECK-NEXT: [[TMP13:%.*]] = bitcast <8 x i32> [[_MSLD]] to <32 x i8> -; CHECK-NEXT: [[TMP14:%.*]] = icmp ne <32 x i8> [[TMP12]], zeroinitializer +; CHECK-NEXT: [[TMP29:%.*]] = load <32 x i8>, ptr [[TMP9]], align 32 ; CHECK-NEXT: [[TMP15:%.*]] = icmp ne <32 x i8> [[TMP13]], zeroinitializer ; CHECK-NEXT: [[TMP16:%.*]] = icmp ne <32 x i8> [[TMP29]], zeroinitializer +; CHECK-NEXT: [[TMP12:%.*]] = icmp ne <32 x i8> [[X1]], zeroinitializer ; CHECK-NEXT: [[TMP17:%.*]] = icmp ne <32 x i8> [[TMP30]], zeroinitializer -; CHECK-NEXT: [[TMP18:%.*]] = and <32 x i1> [[TMP14]], [[TMP15]] -; CHECK-NEXT: [[TMP19:%.*]] = and <32 x i1> [[TMP16]], [[TMP15]] -; CHECK-NEXT: [[TMP20:%.*]] = and <32 x i1> [[TMP14]], [[TMP17]] +; CHECK-NEXT: [[TMP18:%.*]] = and <32 x i1> [[TMP15]], [[TMP16]] +; CHECK-NEXT: [[TMP19:%.*]] = and <32 x i1> [[TMP12]], [[TMP16]] +; CHECK-NEXT: [[TMP20:%.*]] = and <32 x i1> [[TMP15]], [[TMP17]] ; CHECK-NEXT: [[TMP21:%.*]] = or <32 x i1> [[TMP18]], [[TMP19]] ; CHECK-NEXT: [[TMP22:%.*]] = or <32 x i1> [[TMP21]], [[TMP20]] ; CHECK-NEXT: [[TMP23:%.*]] = sext <32 x i1> [[TMP22]] to <32 x i8> @@ -191,18 +171,14 @@ define <8 x i32>@test_int_x86_avx2_vpdpbssd_256(<8 x i32> %x0, <8 x i32> %x1, pt ; CHECK-NEXT: [[TMP25:%.*]] = icmp ne <8 x i32> [[TMP24]], zeroinitializer ; CHECK-NEXT: [[TMP28:%.*]] = sext <8 x i1> [[TMP25]] to <8 x i32> ; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[TMP28]], [[TMP5]] -; CHECK-NEXT: [[TMP10:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X2]]) -; CHECK-NEXT: [[TMP31:%.*]] = bitcast <8 x i32> [[X1]] to <32 x i8> -; CHECK-NEXT: [[TMP32:%.*]] = bitcast <8 x i32> [[X4]] to <32 x i8> -; CHECK-NEXT: [[TMP33:%.*]] = bitcast <8 x i32> [[TMP3]] to <32 x i8> -; CHECK-NEXT: [[TMP34:%.*]] = bitcast <8 x i32> [[TMP4]] to <32 x i8> -; CHECK-NEXT: [[TMP35:%.*]] = icmp ne <32 x i8> [[TMP33]], zeroinitializer -; CHECK-NEXT: [[TMP36:%.*]] = icmp ne <32 x i8> [[TMP34]], zeroinitializer -; CHECK-NEXT: [[TMP37:%.*]] = icmp ne <32 x i8> [[TMP31]], zeroinitializer +; CHECK-NEXT: [[TMP10:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32> [[X0]], <32 x i8> [[X1]], <32 x i8> [[TMP30]]) +; CHECK-NEXT: [[TMP26:%.*]] = icmp ne <32 x i8> [[TMP13]], zeroinitializer ; CHECK-NEXT: [[TMP38:%.*]] = icmp ne <32 x i8> [[TMP32]], zeroinitializer -; CHECK-NEXT: [[TMP39:%.*]] = and <32 x i1> [[TMP35]], [[TMP36]] -; CHECK-NEXT: [[TMP40:%.*]] = and <32 x i1> [[TMP37]], [[TMP36]] -; CHECK-NEXT: [[TMP41:%.*]] = and <32 x i1> [[TMP35]], [[TMP38]] +; CHECK-NEXT: [[TMP27:%.*]] = icmp ne <32 x i8> [[X1]], zeroinitializer +; CHECK-NEXT: [[TMP31:%.*]] = icmp ne <32 x i8> [[X4]], zeroinitializer +; CHECK-NEXT: [[TMP39:%.*]] = and <32 x i1> [[TMP26]], [[TMP38]] +; CHECK-NEXT: [[TMP40:%.*]] = and <32 x i1> [[TMP27]], [[TMP38]] +; CHECK-NEXT: [[TMP41:%.*]] = and <32 x i1> [[TMP26]], [[TMP31]] ; CHECK-NEXT: [[TMP42:%.*]] = or <32 x i1> [[TMP39]], [[TMP40]] ; CHECK-NEXT: [[TMP43:%.*]] = or <32 x i1> [[TMP42]], [[TMP41]] ; CHECK-NEXT: [[TMP44:%.*]] = sext <32 x i1> [[TMP43]] to <32 x i8> @@ -210,28 +186,28 @@ define <8 x i32>@test_int_x86_avx2_vpdpbssd_256(<8 x i32> %x0, <8 x i32> %x1, pt ; CHECK-NEXT: [[TMP46:%.*]] = icmp ne <8 x i32> [[TMP45]], zeroinitializer ; CHECK-NEXT: [[TMP49:%.*]] = sext <8 x i1> [[TMP46]] to <8 x i32> ; CHECK-NEXT: [[_MSPROP3:%.*]] = or <8 x i32> [[TMP49]], [[TMP5]] -; CHECK-NEXT: [[TMP11:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X4]]) +; CHECK-NEXT: [[TMP11:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32> [[X0]], <32 x i8> [[X1]], <32 x i8> [[X4]]) ; CHECK-NEXT: [[_MSPROP4:%.*]] = or <8 x i32> [[_MSPROP1]], [[_MSPROP3]] ; CHECK-NEXT: [[RES:%.*]] = add <8 x i32> [[TMP10]], [[TMP11]] ; CHECK-NEXT: store <8 x i32> [[_MSPROP4]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <8 x i32> [[RES]] ; - %x2 = load <8 x i32>, ptr %x2p - %1 = call <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) - %2 = call <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x4) + %x2 = load <32 x i8>, ptr %x2p + %1 = call <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x2) + %2 = call <8 x i32> @llvm.x86.avx2.vpdpbssd.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x4) %res = add <8 x i32> %1, %2 ret <8 x i32> %res } -declare <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32>, <8 x i32>, <8 x i32>) +declare <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32>, <32 x i8>, <32 x i8>) -define <8 x i32>@test_int_x86_avx2_vpdpbssds_256(<8 x i32> %x0, <8 x i32> %x1, ptr %x2p, <8 x i32> %x4) sanitize_memory { +define <8 x i32>@test_int_x86_avx2_vpdpbssds_256(<8 x i32> %x0, <32 x i8> %x1, ptr %x2p, <32 x i8> %x4) sanitize_memory { ; CHECK-LABEL: define <8 x i32> @test_int_x86_avx2_vpdpbssds_256( -; CHECK-SAME: <8 x i32> [[X0:%.*]], <8 x i32> [[X1:%.*]], ptr [[X2P:%.*]], <8 x i32> [[X4:%.*]]) #[[ATTR1]] { +; CHECK-SAME: <8 x i32> [[X0:%.*]], <32 x i8> [[X1:%.*]], ptr [[X2P:%.*]], <32 x i8> [[X4:%.*]]) #[[ATTR1]] { ; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8 +; CHECK-NEXT: [[TMP13:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8 ; CHECK-NEXT: [[TMP5:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8 -; CHECK-NEXT: [[TMP4:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 +; CHECK-NEXT: [[TMP32:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 ; CHECK-NEXT: call void @llvm.donothing() ; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0 ; CHECK-NEXT: br i1 [[_MSCMP]], label %[[BB5:.*]], label %[[BB6:.*]], !prof [[PROF1]] @@ -239,22 +215,18 @@ define <8 x i32>@test_int_x86_avx2_vpdpbssds_256(<8 x i32> %x0, <8 x i32> %x1, p ; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4]] ; CHECK-NEXT: unreachable ; CHECK: [[BB6]]: -; CHECK-NEXT: [[X2:%.*]] = load <8 x i32>, ptr [[X2P]], align 32 +; CHECK-NEXT: [[TMP30:%.*]] = load <32 x i8>, ptr [[X2P]], align 32 ; CHECK-NEXT: [[TMP7:%.*]] = ptrtoint ptr [[X2P]] to i64 ; CHECK-NEXT: [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080 ; CHECK-NEXT: [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr -; CHECK-NEXT: [[_MSLD:%.*]] = load <8 x i32>, ptr [[TMP9]], align 32 -; CHECK-NEXT: [[TMP29:%.*]] = bitcast <8 x i32> [[X1]] to <32 x i8> -; CHECK-NEXT: [[TMP30:%.*]] = bitcast <8 x i32> [[X2]] to <32 x i8> -; CHECK-NEXT: [[TMP12:%.*]] = bitcast <8 x i32> [[TMP3]] to <32 x i8> -; CHECK-NEXT: [[TMP13:%.*]] = bitcast <8 x i32> [[_MSLD]] to <32 x i8> -; CHECK-NEXT: [[TMP14:%.*]] = icmp ne <32 x i8> [[TMP12]], zeroinitializer +; CHECK-NEXT: [[TMP29:%.*]] = load <32 x i8>, ptr [[TMP9]], align 32 ; CHECK-NEXT: [[TMP15:%.*]] = icmp ne <32 x i8> [[TMP13]], zeroinitializer ; CHECK-NEXT: [[TMP16:%.*]] = icmp ne <32 x i8> [[TMP29]], zeroinitializer +; CHECK-NEXT: [[TMP12:%.*]] = icmp ne <32 x i8> [[X1]], zeroinitializer ; CHECK-NEXT: [[TMP17:%.*]] = icmp ne <32 x i8> [[TMP30]], zeroinitializer -; CHECK-NEXT: [[TMP18:%.*]] = and <32 x i1> [[TMP14]], [[TMP15]] -; CHECK-NEXT: [[TMP19:%.*]] = and <32 x i1> [[TMP16]], [[TMP15]] -; CHECK-NEXT: [[TMP20:%.*]] = and <32 x i1> [[TMP14]], [[TMP17]] +; CHECK-NEXT: [[TMP18:%.*]] = and <32 x i1> [[TMP15]], [[TMP16]] +; CHECK-NEXT: [[TMP19:%.*]] = and <32 x i1> [[TMP12]], [[TMP16]] +; CHECK-NEXT: [[TMP20:%.*]] = and <32 x i1> [[TMP15]], [[TMP17]] ; CHECK-NEXT: [[TMP21:%.*]] = or <32 x i1> [[TMP18]], [[TMP19]] ; CHECK-NEXT: [[TMP22:%.*]] = or <32 x i1> [[TMP21]], [[TMP20]] ; CHECK-NEXT: [[TMP23:%.*]] = sext <32 x i1> [[TMP22]] to <32 x i8> @@ -262,18 +234,14 @@ define <8 x i32>@test_int_x86_avx2_vpdpbssds_256(<8 x i32> %x0, <8 x i32> %x1, p ; CHECK-NEXT: [[TMP25:%.*]] = icmp ne <8 x i32> [[TMP24]], zeroinitializer ; CHECK-NEXT: [[TMP28:%.*]] = sext <8 x i1> [[TMP25]] to <8 x i32> ; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[TMP28]], [[TMP5]] -; CHECK-NEXT: [[TMP10:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X2]]) -; CHECK-NEXT: [[TMP31:%.*]] = bitcast <8 x i32> [[X1]] to <32 x i8> -; CHECK-NEXT: [[TMP32:%.*]] = bitcast <8 x i32> [[X4]] to <32 x i8> -; CHECK-NEXT: [[TMP33:%.*]] = bitcast <8 x i32> [[TMP3]] to <32 x i8> -; CHECK-NEXT: [[TMP34:%.*]] = bitcast <8 x i32> [[TMP4]] to <32 x i8> -; CHECK-NEXT: [[TMP35:%.*]] = icmp ne <32 x i8> [[TMP33]], zeroinitializer -; CHECK-NEXT: [[TMP36:%.*]] = icmp ne <32 x i8> [[TMP34]], zeroinitializer -; CHECK-NEXT: [[TMP37:%.*]] = icmp ne <32 x i8> [[TMP31]], zeroinitializer +; CHECK-NEXT: [[TMP10:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32> [[X0]], <32 x i8> [[X1]], <32 x i8> [[TMP30]]) +; CHECK-NEXT: [[TMP26:%.*]] = icmp ne <32 x i8> [[TMP13]], zeroinitializer ; CHECK-NEXT: [[TMP38:%.*]] = icmp ne <32 x i8> [[TMP32]], zeroinitializer -; CHECK-NEXT: [[TMP39:%.*]] = and <32 x i1> [[TMP35]], [[TMP36]] -; CHECK-NEXT: [[TMP40:%.*]] = and <32 x i1> [[TMP37]], [[TMP36]] -; CHECK-NEXT: [[TMP41:%.*]] = and <32 x i1> [[TMP35]], [[TMP38]] +; CHECK-NEXT: [[TMP27:%.*]] = icmp ne <32 x i8> [[X1]], zeroinitializer +; CHECK-NEXT: [[TMP31:%.*]] = icmp ne <32 x i8> [[X4]], zeroinitializer +; CHECK-NEXT: [[TMP39:%.*]] = and <32 x i1> [[TMP26]], [[TMP38]] +; CHECK-NEXT: [[TMP40:%.*]] = and <32 x i1> [[TMP27]], [[TMP38]] +; CHECK-NEXT: [[TMP41:%.*]] = and <32 x i1> [[TMP26]], [[TMP31]] ; CHECK-NEXT: [[TMP42:%.*]] = or <32 x i1> [[TMP39]], [[TMP40]] ; CHECK-NEXT: [[TMP43:%.*]] = or <32 x i1> [[TMP42]], [[TMP41]] ; CHECK-NEXT: [[TMP44:%.*]] = sext <32 x i1> [[TMP43]] to <32 x i8> @@ -281,28 +249,28 @@ define <8 x i32>@test_int_x86_avx2_vpdpbssds_256(<8 x i32> %x0, <8 x i32> %x1, p ; CHECK-NEXT: [[TMP46:%.*]] = icmp ne <8 x i32> [[TMP45]], zeroinitializer ; CHECK-NEXT: [[TMP49:%.*]] = sext <8 x i1> [[TMP46]] to <8 x i32> ; CHECK-NEXT: [[_MSPROP3:%.*]] = or <8 x i32> [[TMP49]], [[TMP5]] -; CHECK-NEXT: [[TMP11:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X4]]) +; CHECK-NEXT: [[TMP11:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32> [[X0]], <32 x i8> [[X1]], <32 x i8> [[X4]]) ; CHECK-NEXT: [[_MSPROP4:%.*]] = or <8 x i32> [[_MSPROP1]], [[_MSPROP3]] ; CHECK-NEXT: [[RES:%.*]] = add <8 x i32> [[TMP10]], [[TMP11]] ; CHECK-NEXT: store <8 x i32> [[_MSPROP4]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <8 x i32> [[RES]] ; - %x2 = load <8 x i32>, ptr %x2p - %1 = call <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) - %2 = call <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x4) + %x2 = load <32 x i8>, ptr %x2p + %1 = call <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x2) + %2 = call <8 x i32> @llvm.x86.avx2.vpdpbssds.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x4) %res = add <8 x i32> %1, %2 ret <8 x i32> %res } -declare <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32>, <4 x i32>, <4 x i32>) +declare <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32>, <16 x i8>, <16 x i8>) -define <4 x i32>@test_int_x86_avx2_vpdpbsud_128(<4 x i32> %x0, <4 x i32> %x1, ptr %x2p, <4 x i32> %x4) sanitize_memory { +define <4 x i32>@test_int_x86_avx2_vpdpbsud_128(<4 x i32> %x0, <16 x i8> %x1, ptr %x2p, <16 x i8> %x4) sanitize_memory { ; CHECK-LABEL: define <4 x i32> @test_int_x86_avx2_vpdpbsud_128( -; CHECK-SAME: <4 x i32> [[X0:%.*]], <4 x i32> [[X1:%.*]], ptr [[X2P:%.*]], <4 x i32> [[X4:%.*]]) #[[ATTR1]] { +; CHECK-SAME: <4 x i32> [[X0:%.*]], <16 x i8> [[X1:%.*]], ptr [[X2P:%.*]], <16 x i8> [[X4:%.*]]) #[[ATTR1]] { ; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8 +; CHECK-NEXT: [[TMP3:%.*]] = load <16 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 16) to ptr), align 8 ; CHECK-NEXT: [[TMP2:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 16) to ptr), align 8 -; CHECK-NEXT: [[TMP4:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 +; CHECK-NEXT: [[TMP4:%.*]] = load <16 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 ; CHECK-NEXT: call void @llvm.donothing() ; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0 ; CHECK-NEXT: br i1 [[_MSCMP]], label %[[BB5:.*]], label %[[BB6:.*]], !prof [[PROF1]] @@ -310,38 +278,62 @@ define <4 x i32>@test_int_x86_avx2_vpdpbsud_128(<4 x i32> %x0, <4 x i32> %x1, pt ; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4]] ; CHECK-NEXT: unreachable ; CHECK: [[BB6]]: -; CHECK-NEXT: [[X2:%.*]] = load <4 x i32>, ptr [[X2P]], align 16 +; CHECK-NEXT: [[X2:%.*]] = load <16 x i8>, ptr [[X2P]], align 16 ; CHECK-NEXT: [[TMP7:%.*]] = ptrtoint ptr [[X2P]] to i64 ; CHECK-NEXT: [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080 ; CHECK-NEXT: [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr -; CHECK-NEXT: [[_MSLD:%.*]] = load <4 x i32>, ptr [[TMP9]], align 16 -; CHECK-NEXT: [[_MSPROP:%.*]] = or <4 x i32> [[TMP2]], [[TMP3]] -; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[_MSPROP]], [[_MSLD]] -; CHECK-NEXT: [[TMP10:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X2]]) -; CHECK-NEXT: [[_MSPROP2:%.*]] = or <4 x i32> [[TMP2]], [[TMP3]] -; CHECK-NEXT: [[_MSPROP3:%.*]] = or <4 x i32> [[_MSPROP2]], [[TMP4]] -; CHECK-NEXT: [[TMP11:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X4]]) +; CHECK-NEXT: [[_MSLD:%.*]] = load <16 x i8>, ptr [[TMP9]], align 16 +; CHECK-NEXT: [[TMP23:%.*]] = icmp ne <16 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP24:%.*]] = icmp ne <16 x i8> [[_MSLD]], zeroinitializer +; CHECK-NEXT: [[TMP12:%.*]] = icmp ne <16 x i8> [[X1]], zeroinitializer +; CHECK-NEXT: [[TMP13:%.*]] = icmp ne <16 x i8> [[X2]], zeroinitializer +; CHECK-NEXT: [[TMP14:%.*]] = and <16 x i1> [[TMP23]], [[TMP24]] +; CHECK-NEXT: [[TMP15:%.*]] = and <16 x i1> [[TMP12]], [[TMP24]] +; CHECK-NEXT: [[TMP16:%.*]] = and <16 x i1> [[TMP23]], [[TMP13]] +; CHECK-NEXT: [[TMP17:%.*]] = or <16 x i1> [[TMP14]], [[TMP15]] +; CHECK-NEXT: [[TMP18:%.*]] = or <16 x i1> [[TMP17]], [[TMP16]] +; CHECK-NEXT: [[TMP19:%.*]] = sext <16 x i1> [[TMP18]] to <16 x i8> +; CHECK-NEXT: [[TMP20:%.*]] = bitcast <16 x i8> [[TMP19]] to <4 x i32> +; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <4 x i32> [[TMP20]], zeroinitializer +; CHECK-NEXT: [[TMP22:%.*]] = sext <4 x i1> [[TMP21]] to <4 x i32> +; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[TMP22]], [[TMP2]] +; CHECK-NEXT: [[TMP10:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32> [[X0]], <16 x i8> [[X1]], <16 x i8> [[X2]]) +; CHECK-NEXT: [[TMP25:%.*]] = icmp ne <16 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP26:%.*]] = icmp ne <16 x i8> [[TMP4]], zeroinitializer +; CHECK-NEXT: [[TMP27:%.*]] = icmp ne <16 x i8> [[X1]], zeroinitializer +; CHECK-NEXT: [[TMP28:%.*]] = icmp ne <16 x i8> [[X4]], zeroinitializer +; CHECK-NEXT: [[TMP29:%.*]] = and <16 x i1> [[TMP25]], [[TMP26]] +; CHECK-NEXT: [[TMP30:%.*]] = and <16 x i1> [[TMP27]], [[TMP26]] +; CHECK-NEXT: [[TMP31:%.*]] = and <16 x i1> [[TMP25]], [[TMP28]] +; CHECK-NEXT: [[TMP32:%.*]] = or <16 x i1> [[TMP29]], [[TMP30]] +; CHECK-NEXT: [[TMP33:%.*]] = or <16 x i1> [[TMP32]], [[TMP31]] +; CHECK-NEXT: [[TMP34:%.*]] = sext <16 x i1> [[TMP33]] to <16 x i8> +; CHECK-NEXT: [[TMP35:%.*]] = bitcast <16 x i8> [[TMP34]] to <4 x i32> +; CHECK-NEXT: [[TMP36:%.*]] = icmp ne <4 x i32> [[TMP35]], zeroinitializer +; CHECK-NEXT: [[TMP37:%.*]] = sext <4 x i1> [[TMP36]] to <4 x i32> +; CHECK-NEXT: [[_MSPROP3:%.*]] = or <4 x i32> [[TMP37]], [[TMP2]] +; CHECK-NEXT: [[TMP11:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32> [[X0]], <16 x i8> [[X1]], <16 x i8> [[X4]]) ; CHECK-NEXT: [[_MSPROP4:%.*]] = or <4 x i32> [[_MSPROP1]], [[_MSPROP3]] ; CHECK-NEXT: [[RES:%.*]] = add <4 x i32> [[TMP10]], [[TMP11]] ; CHECK-NEXT: store <4 x i32> [[_MSPROP4]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <4 x i32> [[RES]] ; - %x2 = load <4 x i32>, ptr %x2p - %1 = call <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) - %2 = call <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x4) + %x2 = load <16 x i8>, ptr %x2p + %1 = call <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x2) + %2 = call <4 x i32> @llvm.x86.avx2.vpdpbsud.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x4) %res = add <4 x i32> %1, %2 ret <4 x i32> %res } -declare <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32>, <4 x i32>, <4 x i32>) +declare <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32>, <16 x i8>, <16 x i8>) -define <4 x i32>@test_int_x86_avx2_vpdpbsuds_128(<4 x i32> %x0, <4 x i32> %x1, ptr %x2p, <4 x i32> %x4) sanitize_memory { +define <4 x i32>@test_int_x86_avx2_vpdpbsuds_128(<4 x i32> %x0, <16 x i8> %x1, ptr %x2p, <16 x i8> %x4) sanitize_memory { ; CHECK-LABEL: define <4 x i32> @test_int_x86_avx2_vpdpbsuds_128( -; CHECK-SAME: <4 x i32> [[X0:%.*]], <4 x i32> [[X1:%.*]], ptr [[X2P:%.*]], <4 x i32> [[X4:%.*]]) #[[ATTR1]] { +; CHECK-SAME: <4 x i32> [[X0:%.*]], <16 x i8> [[X1:%.*]], ptr [[X2P:%.*]], <16 x i8> [[X4:%.*]]) #[[ATTR1]] { ; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8 +; CHECK-NEXT: [[TMP3:%.*]] = load <16 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 16) to ptr), align 8 ; CHECK-NEXT: [[TMP2:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 16) to ptr), align 8 -; CHECK-NEXT: [[TMP4:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 +; CHECK-NEXT: [[TMP4:%.*]] = load <16 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 ; CHECK-NEXT: call void @llvm.donothing() ; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0 ; CHECK-NEXT: br i1 [[_MSCMP]], label %[[BB5:.*]], label %[[BB6:.*]], !prof [[PROF1]] @@ -349,38 +341,62 @@ define <4 x i32>@test_int_x86_avx2_vpdpbsuds_128(<4 x i32> %x0, <4 x i32> %x1, p ; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4]] ; CHECK-NEXT: unreachable ; CHECK: [[BB6]]: -; CHECK-NEXT: [[X2:%.*]] = load <4 x i32>, ptr [[X2P]], align 16 +; CHECK-NEXT: [[X2:%.*]] = load <16 x i8>, ptr [[X2P]], align 16 ; CHECK-NEXT: [[TMP7:%.*]] = ptrtoint ptr [[X2P]] to i64 ; CHECK-NEXT: [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080 ; CHECK-NEXT: [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr -; CHECK-NEXT: [[_MSLD:%.*]] = load <4 x i32>, ptr [[TMP9]], align 16 -; CHECK-NEXT: [[_MSPROP:%.*]] = or <4 x i32> [[TMP2]], [[TMP3]] -; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[_MSPROP]], [[_MSLD]] -; CHECK-NEXT: [[TMP10:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X2]]) -; CHECK-NEXT: [[_MSPROP2:%.*]] = or <4 x i32> [[TMP2]], [[TMP3]] -; CHECK-NEXT: [[_MSPROP3:%.*]] = or <4 x i32> [[_MSPROP2]], [[TMP4]] -; CHECK-NEXT: [[TMP11:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X4]]) +; CHECK-NEXT: [[_MSLD:%.*]] = load <16 x i8>, ptr [[TMP9]], align 16 +; CHECK-NEXT: [[TMP23:%.*]] = icmp ne <16 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP24:%.*]] = icmp ne <16 x i8> [[_MSLD]], zeroinitializer +; CHECK-NEXT: [[TMP12:%.*]] = icmp ne <16 x i8> [[X1]], zeroinitializer +; CHECK-NEXT: [[TMP13:%.*]] = icmp ne <16 x i8> [[X2]], zeroinitializer +; CHECK-NEXT: [[TMP14:%.*]] = and <16 x i1> [[TMP23]], [[TMP24]] +; CHECK-NEXT: [[TMP15:%.*]] = and <16 x i1> [[TMP12]], [[TMP24]] +; CHECK-NEXT: [[TMP16:%.*]] = and <16 x i1> [[TMP23]], [[TMP13]] +; CHECK-NEXT: [[TMP17:%.*]] = or <16 x i1> [[TMP14]], [[TMP15]] +; CHECK-NEXT: [[TMP18:%.*]] = or <16 x i1> [[TMP17]], [[TMP16]] +; CHECK-NEXT: [[TMP19:%.*]] = sext <16 x i1> [[TMP18]] to <16 x i8> +; CHECK-NEXT: [[TMP20:%.*]] = bitcast <16 x i8> [[TMP19]] to <4 x i32> +; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <4 x i32> [[TMP20]], zeroinitializer +; CHECK-NEXT: [[TMP22:%.*]] = sext <4 x i1> [[TMP21]] to <4 x i32> +; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[TMP22]], [[TMP2]] +; CHECK-NEXT: [[TMP10:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32> [[X0]], <16 x i8> [[X1]], <16 x i8> [[X2]]) +; CHECK-NEXT: [[TMP25:%.*]] = icmp ne <16 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP26:%.*]] = icmp ne <16 x i8> [[TMP4]], zeroinitializer +; CHECK-NEXT: [[TMP27:%.*]] = icmp ne <16 x i8> [[X1]], zeroinitializer +; CHECK-NEXT: [[TMP28:%.*]] = icmp ne <16 x i8> [[X4]], zeroinitializer +; CHECK-NEXT: [[TMP29:%.*]] = and <16 x i1> [[TMP25]], [[TMP26]] +; CHECK-NEXT: [[TMP30:%.*]] = and <16 x i1> [[TMP27]], [[TMP26]] +; CHECK-NEXT: [[TMP31:%.*]] = and <16 x i1> [[TMP25]], [[TMP28]] +; CHECK-NEXT: [[TMP32:%.*]] = or <16 x i1> [[TMP29]], [[TMP30]] +; CHECK-NEXT: [[TMP33:%.*]] = or <16 x i1> [[TMP32]], [[TMP31]] +; CHECK-NEXT: [[TMP34:%.*]] = sext <16 x i1> [[TMP33]] to <16 x i8> +; CHECK-NEXT: [[TMP35:%.*]] = bitcast <16 x i8> [[TMP34]] to <4 x i32> +; CHECK-NEXT: [[TMP36:%.*]] = icmp ne <4 x i32> [[TMP35]], zeroinitializer +; CHECK-NEXT: [[TMP37:%.*]] = sext <4 x i1> [[TMP36]] to <4 x i32> +; CHECK-NEXT: [[_MSPROP3:%.*]] = or <4 x i32> [[TMP37]], [[TMP2]] +; CHECK-NEXT: [[TMP11:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32> [[X0]], <16 x i8> [[X1]], <16 x i8> [[X4]]) ; CHECK-NEXT: [[_MSPROP4:%.*]] = or <4 x i32> [[_MSPROP1]], [[_MSPROP3]] ; CHECK-NEXT: [[RES:%.*]] = add <4 x i32> [[TMP10]], [[TMP11]] ; CHECK-NEXT: store <4 x i32> [[_MSPROP4]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <4 x i32> [[RES]] ; - %x2 = load <4 x i32>, ptr %x2p - %1 = call <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) - %2 = call <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x4) + %x2 = load <16 x i8>, ptr %x2p + %1 = call <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x2) + %2 = call <4 x i32> @llvm.x86.avx2.vpdpbsuds.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x4) %res = add <4 x i32> %1, %2 ret <4 x i32> %res } -declare <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32>, <8 x i32>, <8 x i32>) +declare <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32>, <32 x i8>, <32 x i8>) -define <8 x i32>@test_int_x86_avx2_vpdpbsud_256(<8 x i32> %x0, <8 x i32> %x1, ptr %x2p, <8 x i32> %x4) sanitize_memory { +define <8 x i32>@test_int_x86_avx2_vpdpbsud_256(<8 x i32> %x0, <32 x i8> %x1, ptr %x2p, <32 x i8> %x4) sanitize_memory { ; CHECK-LABEL: define <8 x i32> @test_int_x86_avx2_vpdpbsud_256( -; CHECK-SAME: <8 x i32> [[X0:%.*]], <8 x i32> [[X1:%.*]], ptr [[X2P:%.*]], <8 x i32> [[X4:%.*]]) #[[ATTR1]] { +; CHECK-SAME: <8 x i32> [[X0:%.*]], <32 x i8> [[X1:%.*]], ptr [[X2P:%.*]], <32 x i8> [[X4:%.*]]) #[[ATTR1]] { ; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8 +; CHECK-NEXT: [[TMP3:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8 ; CHECK-NEXT: [[TMP2:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8 -; CHECK-NEXT: [[TMP4:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 +; CHECK-NEXT: [[TMP4:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 ; CHECK-NEXT: call void @llvm.donothing() ; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0 ; CHECK-NEXT: br i1 [[_MSCMP]], label %[[BB5:.*]], label %[[BB6:.*]], !prof [[PROF1]] @@ -388,38 +404,62 @@ define <8 x i32>@test_int_x86_avx2_vpdpbsud_256(<8 x i32> %x0, <8 x i32> %x1, pt ; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4]] ; CHECK-NEXT: unreachable ; CHECK: [[BB6]]: -; CHECK-NEXT: [[X2:%.*]] = load <8 x i32>, ptr [[X2P]], align 32 +; CHECK-NEXT: [[X2:%.*]] = load <32 x i8>, ptr [[X2P]], align 32 ; CHECK-NEXT: [[TMP7:%.*]] = ptrtoint ptr [[X2P]] to i64 ; CHECK-NEXT: [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080 ; CHECK-NEXT: [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr -; CHECK-NEXT: [[_MSLD:%.*]] = load <8 x i32>, ptr [[TMP9]], align 32 -; CHECK-NEXT: [[_MSPROP:%.*]] = or <8 x i32> [[TMP2]], [[TMP3]] -; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[_MSPROP]], [[_MSLD]] -; CHECK-NEXT: [[TMP10:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X2]]) -; CHECK-NEXT: [[_MSPROP2:%.*]] = or <8 x i32> [[TMP2]], [[TMP3]] -; CHECK-NEXT: [[_MSPROP3:%.*]] = or <8 x i32> [[_MSPROP2]], [[TMP4]] -; CHECK-NEXT: [[TMP11:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X4]]) +; CHECK-NEXT: [[_MSLD:%.*]] = load <32 x i8>, ptr [[TMP9]], align 32 +; CHECK-NEXT: [[TMP23:%.*]] = icmp ne <32 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP24:%.*]] = icmp ne <32 x i8> [[_MSLD]], zeroinitializer +; CHECK-NEXT: [[TMP12:%.*]] = icmp ne <32 x i8> [[X1]], zeroinitializer +; CHECK-NEXT: [[TMP13:%.*]] = icmp ne <32 x i8> [[X2]], zeroinitializer +; CHECK-NEXT: [[TMP14:%.*]] = and <32 x i1> [[TMP23]], [[TMP24]] +; CHECK-NEXT: [[TMP15:%.*]] = and <32 x i1> [[TMP12]], [[TMP24]] +; CHECK-NEXT: [[TMP16:%.*]] = and <32 x i1> [[TMP23]], [[TMP13]] +; CHECK-NEXT: [[TMP17:%.*]] = or <32 x i1> [[TMP14]], [[TMP15]] +; CHECK-NEXT: [[TMP18:%.*]] = or <32 x i1> [[TMP17]], [[TMP16]] +; CHECK-NEXT: [[TMP19:%.*]] = sext <32 x i1> [[TMP18]] to <32 x i8> +; CHECK-NEXT: [[TMP20:%.*]] = bitcast <32 x i8> [[TMP19]] to <8 x i32> +; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <8 x i32> [[TMP20]], zeroinitializer +; CHECK-NEXT: [[TMP22:%.*]] = sext <8 x i1> [[TMP21]] to <8 x i32> +; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[TMP22]], [[TMP2]] +; CHECK-NEXT: [[TMP10:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32> [[X0]], <32 x i8> [[X1]], <32 x i8> [[X2]]) +; CHECK-NEXT: [[TMP25:%.*]] = icmp ne <32 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP26:%.*]] = icmp ne <32 x i8> [[TMP4]], zeroinitializer +; CHECK-NEXT: [[TMP27:%.*]] = icmp ne <32 x i8> [[X1]], zeroinitializer +; CHECK-NEXT: [[TMP28:%.*]] = icmp ne <32 x i8> [[X4]], zeroinitializer +; CHECK-NEXT: [[TMP29:%.*]] = and <32 x i1> [[TMP25]], [[TMP26]] +; CHECK-NEXT: [[TMP30:%.*]] = and <32 x i1> [[TMP27]], [[TMP26]] +; CHECK-NEXT: [[TMP31:%.*]] = and <32 x i1> [[TMP25]], [[TMP28]] +; CHECK-NEXT: [[TMP32:%.*]] = or <32 x i1> [[TMP29]], [[TMP30]] +; CHECK-NEXT: [[TMP33:%.*]] = or <32 x i1> [[TMP32]], [[TMP31]] +; CHECK-NEXT: [[TMP34:%.*]] = sext <32 x i1> [[TMP33]] to <32 x i8> +; CHECK-NEXT: [[TMP35:%.*]] = bitcast <32 x i8> [[TMP34]] to <8 x i32> +; CHECK-NEXT: [[TMP36:%.*]] = icmp ne <8 x i32> [[TMP35]], zeroinitializer +; CHECK-NEXT: [[TMP37:%.*]] = sext <8 x i1> [[TMP36]] to <8 x i32> +; CHECK-NEXT: [[_MSPROP3:%.*]] = or <8 x i32> [[TMP37]], [[TMP2]] +; CHECK-NEXT: [[TMP11:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32> [[X0]], <32 x i8> [[X1]], <32 x i8> [[X4]]) ; CHECK-NEXT: [[_MSPROP4:%.*]] = or <8 x i32> [[_MSPROP1]], [[_MSPROP3]] ; CHECK-NEXT: [[RES:%.*]] = add <8 x i32> [[TMP10]], [[TMP11]] ; CHECK-NEXT: store <8 x i32> [[_MSPROP4]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <8 x i32> [[RES]] ; - %x2 = load <8 x i32>, ptr %x2p - %1 = call <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) - %2 = call <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x4) + %x2 = load <32 x i8>, ptr %x2p + %1 = call <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x2) + %2 = call <8 x i32> @llvm.x86.avx2.vpdpbsud.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x4) %res = add <8 x i32> %1, %2 ret <8 x i32> %res } -declare <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32>, <8 x i32>, <8 x i32>) +declare <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32>, <32 x i8>, <32 x i8>) -define <8 x i32>@test_int_x86_avx2_vpdpbsuds_256(<8 x i32> %x0, <8 x i32> %x1, ptr %x2p, <8 x i32> %x4) sanitize_memory { +define <8 x i32>@test_int_x86_avx2_vpdpbsuds_256(<8 x i32> %x0, <32 x i8> %x1, ptr %x2p, <32 x i8> %x4) sanitize_memory { ; CHECK-LABEL: define <8 x i32> @test_int_x86_avx2_vpdpbsuds_256( -; CHECK-SAME: <8 x i32> [[X0:%.*]], <8 x i32> [[X1:%.*]], ptr [[X2P:%.*]], <8 x i32> [[X4:%.*]]) #[[ATTR1]] { +; CHECK-SAME: <8 x i32> [[X0:%.*]], <32 x i8> [[X1:%.*]], ptr [[X2P:%.*]], <32 x i8> [[X4:%.*]]) #[[ATTR1]] { ; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8 +; CHECK-NEXT: [[TMP3:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8 ; CHECK-NEXT: [[TMP2:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8 -; CHECK-NEXT: [[TMP4:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 +; CHECK-NEXT: [[TMP4:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 ; CHECK-NEXT: call void @llvm.donothing() ; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0 ; CHECK-NEXT: br i1 [[_MSCMP]], label %[[BB5:.*]], label %[[BB6:.*]], !prof [[PROF1]] @@ -427,38 +467,62 @@ define <8 x i32>@test_int_x86_avx2_vpdpbsuds_256(<8 x i32> %x0, <8 x i32> %x1, p ; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4]] ; CHECK-NEXT: unreachable ; CHECK: [[BB6]]: -; CHECK-NEXT: [[X2:%.*]] = load <8 x i32>, ptr [[X2P]], align 32 +; CHECK-NEXT: [[X2:%.*]] = load <32 x i8>, ptr [[X2P]], align 32 ; CHECK-NEXT: [[TMP7:%.*]] = ptrtoint ptr [[X2P]] to i64 ; CHECK-NEXT: [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080 ; CHECK-NEXT: [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr -; CHECK-NEXT: [[_MSLD:%.*]] = load <8 x i32>, ptr [[TMP9]], align 32 -; CHECK-NEXT: [[_MSPROP:%.*]] = or <8 x i32> [[TMP2]], [[TMP3]] -; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[_MSPROP]], [[_MSLD]] -; CHECK-NEXT: [[TMP10:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X2]]) -; CHECK-NEXT: [[_MSPROP2:%.*]] = or <8 x i32> [[TMP2]], [[TMP3]] -; CHECK-NEXT: [[_MSPROP3:%.*]] = or <8 x i32> [[_MSPROP2]], [[TMP4]] -; CHECK-NEXT: [[TMP11:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X4]]) +; CHECK-NEXT: [[_MSLD:%.*]] = load <32 x i8>, ptr [[TMP9]], align 32 +; CHECK-NEXT: [[TMP23:%.*]] = icmp ne <32 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP24:%.*]] = icmp ne <32 x i8> [[_MSLD]], zeroinitializer +; CHECK-NEXT: [[TMP12:%.*]] = icmp ne <32 x i8> [[X1]], zeroinitializer +; CHECK-NEXT: [[TMP13:%.*]] = icmp ne <32 x i8> [[X2]], zeroinitializer +; CHECK-NEXT: [[TMP14:%.*]] = and <32 x i1> [[TMP23]], [[TMP24]] +; CHECK-NEXT: [[TMP15:%.*]] = and <32 x i1> [[TMP12]], [[TMP24]] +; CHECK-NEXT: [[TMP16:%.*]] = and <32 x i1> [[TMP23]], [[TMP13]] +; CHECK-NEXT: [[TMP17:%.*]] = or <32 x i1> [[TMP14]], [[TMP15]] +; CHECK-NEXT: [[TMP18:%.*]] = or <32 x i1> [[TMP17]], [[TMP16]] +; CHECK-NEXT: [[TMP19:%.*]] = sext <32 x i1> [[TMP18]] to <32 x i8> +; CHECK-NEXT: [[TMP20:%.*]] = bitcast <32 x i8> [[TMP19]] to <8 x i32> +; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <8 x i32> [[TMP20]], zeroinitializer +; CHECK-NEXT: [[TMP22:%.*]] = sext <8 x i1> [[TMP21]] to <8 x i32> +; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[TMP22]], [[TMP2]] +; CHECK-NEXT: [[TMP10:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32> [[X0]], <32 x i8> [[X1]], <32 x i8> [[X2]]) +; CHECK-NEXT: [[TMP25:%.*]] = icmp ne <32 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP26:%.*]] = icmp ne <32 x i8> [[TMP4]], zeroinitializer +; CHECK-NEXT: [[TMP27:%.*]] = icmp ne <32 x i8> [[X1]], zeroinitializer +; CHECK-NEXT: [[TMP28:%.*]] = icmp ne <32 x i8> [[X4]], zeroinitializer +; CHECK-NEXT: [[TMP29:%.*]] = and <32 x i1> [[TMP25]], [[TMP26]] +; CHECK-NEXT: [[TMP30:%.*]] = and <32 x i1> [[TMP27]], [[TMP26]] +; CHECK-NEXT: [[TMP31:%.*]] = and <32 x i1> [[TMP25]], [[TMP28]] +; CHECK-NEXT: [[TMP32:%.*]] = or <32 x i1> [[TMP29]], [[TMP30]] +; CHECK-NEXT: [[TMP33:%.*]] = or <32 x i1> [[TMP32]], [[TMP31]] +; CHECK-NEXT: [[TMP34:%.*]] = sext <32 x i1> [[TMP33]] to <32 x i8> +; CHECK-NEXT: [[TMP35:%.*]] = bitcast <32 x i8> [[TMP34]] to <8 x i32> +; CHECK-NEXT: [[TMP36:%.*]] = icmp ne <8 x i32> [[TMP35]], zeroinitializer +; CHECK-NEXT: [[TMP37:%.*]] = sext <8 x i1> [[TMP36]] to <8 x i32> +; CHECK-NEXT: [[_MSPROP3:%.*]] = or <8 x i32> [[TMP37]], [[TMP2]] +; CHECK-NEXT: [[TMP11:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32> [[X0]], <32 x i8> [[X1]], <32 x i8> [[X4]]) ; CHECK-NEXT: [[_MSPROP4:%.*]] = or <8 x i32> [[_MSPROP1]], [[_MSPROP3]] ; CHECK-NEXT: [[RES:%.*]] = add <8 x i32> [[TMP10]], [[TMP11]] ; CHECK-NEXT: store <8 x i32> [[_MSPROP4]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <8 x i32> [[RES]] ; - %x2 = load <8 x i32>, ptr %x2p - %1 = call <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) - %2 = call <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x4) + %x2 = load <32 x i8>, ptr %x2p + %1 = call <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x2) + %2 = call <8 x i32> @llvm.x86.avx2.vpdpbsuds.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x4) %res = add <8 x i32> %1, %2 ret <8 x i32> %res } -declare <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32>, <4 x i32>, <4 x i32>) +declare <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32>, <16 x i8>, <16 x i8>) -define <4 x i32>@test_int_x86_avx2_vpdpbuud_128(<4 x i32> %x0, <4 x i32> %x1, ptr %x2p, <4 x i32> %x4) sanitize_memory { +define <4 x i32>@test_int_x86_avx2_vpdpbuud_128(<4 x i32> %x0, <16 x i8> %x1, ptr %x2p, <16 x i8> %x4) sanitize_memory { ; CHECK-LABEL: define <4 x i32> @test_int_x86_avx2_vpdpbuud_128( -; CHECK-SAME: <4 x i32> [[X0:%.*]], <4 x i32> [[X1:%.*]], ptr [[X2P:%.*]], <4 x i32> [[X4:%.*]]) #[[ATTR1]] { +; CHECK-SAME: <4 x i32> [[X0:%.*]], <16 x i8> [[X1:%.*]], ptr [[X2P:%.*]], <16 x i8> [[X4:%.*]]) #[[ATTR1]] { ; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8 +; CHECK-NEXT: [[TMP3:%.*]] = load <16 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 16) to ptr), align 8 ; CHECK-NEXT: [[TMP2:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 16) to ptr), align 8 -; CHECK-NEXT: [[TMP4:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 +; CHECK-NEXT: [[TMP4:%.*]] = load <16 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 ; CHECK-NEXT: call void @llvm.donothing() ; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0 ; CHECK-NEXT: br i1 [[_MSCMP]], label %[[BB5:.*]], label %[[BB6:.*]], !prof [[PROF1]] @@ -466,38 +530,62 @@ define <4 x i32>@test_int_x86_avx2_vpdpbuud_128(<4 x i32> %x0, <4 x i32> %x1, pt ; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4]] ; CHECK-NEXT: unreachable ; CHECK: [[BB6]]: -; CHECK-NEXT: [[X2:%.*]] = load <4 x i32>, ptr [[X2P]], align 16 +; CHECK-NEXT: [[X2:%.*]] = load <16 x i8>, ptr [[X2P]], align 16 ; CHECK-NEXT: [[TMP7:%.*]] = ptrtoint ptr [[X2P]] to i64 ; CHECK-NEXT: [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080 ; CHECK-NEXT: [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr -; CHECK-NEXT: [[_MSLD:%.*]] = load <4 x i32>, ptr [[TMP9]], align 16 -; CHECK-NEXT: [[_MSPROP:%.*]] = or <4 x i32> [[TMP2]], [[TMP3]] -; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[_MSPROP]], [[_MSLD]] -; CHECK-NEXT: [[TMP10:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X2]]) -; CHECK-NEXT: [[_MSPROP2:%.*]] = or <4 x i32> [[TMP2]], [[TMP3]] -; CHECK-NEXT: [[_MSPROP3:%.*]] = or <4 x i32> [[_MSPROP2]], [[TMP4]] -; CHECK-NEXT: [[TMP11:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X4]]) +; CHECK-NEXT: [[_MSLD:%.*]] = load <16 x i8>, ptr [[TMP9]], align 16 +; CHECK-NEXT: [[TMP23:%.*]] = icmp ne <16 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP24:%.*]] = icmp ne <16 x i8> [[_MSLD]], zeroinitializer +; CHECK-NEXT: [[TMP12:%.*]] = icmp ne <16 x i8> [[X1]], zeroinitializer +; CHECK-NEXT: [[TMP13:%.*]] = icmp ne <16 x i8> [[X2]], zeroinitializer +; CHECK-NEXT: [[TMP14:%.*]] = and <16 x i1> [[TMP23]], [[TMP24]] +; CHECK-NEXT: [[TMP15:%.*]] = and <16 x i1> [[TMP12]], [[TMP24]] +; CHECK-NEXT: [[TMP16:%.*]] = and <16 x i1> [[TMP23]], [[TMP13]] +; CHECK-NEXT: [[TMP17:%.*]] = or <16 x i1> [[TMP14]], [[TMP15]] +; CHECK-NEXT: [[TMP18:%.*]] = or <16 x i1> [[TMP17]], [[TMP16]] +; CHECK-NEXT: [[TMP19:%.*]] = sext <16 x i1> [[TMP18]] to <16 x i8> +; CHECK-NEXT: [[TMP20:%.*]] = bitcast <16 x i8> [[TMP19]] to <4 x i32> +; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <4 x i32> [[TMP20]], zeroinitializer +; CHECK-NEXT: [[TMP22:%.*]] = sext <4 x i1> [[TMP21]] to <4 x i32> +; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[TMP22]], [[TMP2]] +; CHECK-NEXT: [[TMP10:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32> [[X0]], <16 x i8> [[X1]], <16 x i8> [[X2]]) +; CHECK-NEXT: [[TMP25:%.*]] = icmp ne <16 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP26:%.*]] = icmp ne <16 x i8> [[TMP4]], zeroinitializer +; CHECK-NEXT: [[TMP27:%.*]] = icmp ne <16 x i8> [[X1]], zeroinitializer +; CHECK-NEXT: [[TMP28:%.*]] = icmp ne <16 x i8> [[X4]], zeroinitializer +; CHECK-NEXT: [[TMP29:%.*]] = and <16 x i1> [[TMP25]], [[TMP26]] +; CHECK-NEXT: [[TMP30:%.*]] = and <16 x i1> [[TMP27]], [[TMP26]] +; CHECK-NEXT: [[TMP31:%.*]] = and <16 x i1> [[TMP25]], [[TMP28]] +; CHECK-NEXT: [[TMP32:%.*]] = or <16 x i1> [[TMP29]], [[TMP30]] +; CHECK-NEXT: [[TMP33:%.*]] = or <16 x i1> [[TMP32]], [[TMP31]] +; CHECK-NEXT: [[TMP34:%.*]] = sext <16 x i1> [[TMP33]] to <16 x i8> +; CHECK-NEXT: [[TMP35:%.*]] = bitcast <16 x i8> [[TMP34]] to <4 x i32> +; CHECK-NEXT: [[TMP36:%.*]] = icmp ne <4 x i32> [[TMP35]], zeroinitializer +; CHECK-NEXT: [[TMP37:%.*]] = sext <4 x i1> [[TMP36]] to <4 x i32> +; CHECK-NEXT: [[_MSPROP3:%.*]] = or <4 x i32> [[TMP37]], [[TMP2]] +; CHECK-NEXT: [[TMP11:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32> [[X0]], <16 x i8> [[X1]], <16 x i8> [[X4]]) ; CHECK-NEXT: [[_MSPROP4:%.*]] = or <4 x i32> [[_MSPROP1]], [[_MSPROP3]] ; CHECK-NEXT: [[RES:%.*]] = add <4 x i32> [[TMP10]], [[TMP11]] ; CHECK-NEXT: store <4 x i32> [[_MSPROP4]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <4 x i32> [[RES]] ; - %x2 = load <4 x i32>, ptr %x2p - %1 = call <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) - %2 = call <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x4) + %x2 = load <16 x i8>, ptr %x2p + %1 = call <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x2) + %2 = call <4 x i32> @llvm.x86.avx2.vpdpbuud.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x4) %res = add <4 x i32> %1, %2 ret <4 x i32> %res } -declare <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32>, <4 x i32>, <4 x i32>) +declare <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32>, <16 x i8>, <16 x i8>) -define <4 x i32>@test_int_x86_avx2_vpdpbuuds_128(<4 x i32> %x0, <4 x i32> %x1, ptr %x2p, <4 x i32> %x4) sanitize_memory { +define <4 x i32>@test_int_x86_avx2_vpdpbuuds_128(<4 x i32> %x0, <16 x i8> %x1, ptr %x2p, <16 x i8> %x4) sanitize_memory { ; CHECK-LABEL: define <4 x i32> @test_int_x86_avx2_vpdpbuuds_128( -; CHECK-SAME: <4 x i32> [[X0:%.*]], <4 x i32> [[X1:%.*]], ptr [[X2P:%.*]], <4 x i32> [[X4:%.*]]) #[[ATTR1]] { +; CHECK-SAME: <4 x i32> [[X0:%.*]], <16 x i8> [[X1:%.*]], ptr [[X2P:%.*]], <16 x i8> [[X4:%.*]]) #[[ATTR1]] { ; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8 +; CHECK-NEXT: [[TMP3:%.*]] = load <16 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 16) to ptr), align 8 ; CHECK-NEXT: [[TMP2:%.*]] = load <4 x i32>, ptr @__msan_param_tls, align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 16) to ptr), align 8 -; CHECK-NEXT: [[TMP4:%.*]] = load <4 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 +; CHECK-NEXT: [[TMP4:%.*]] = load <16 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 40) to ptr), align 8 ; CHECK-NEXT: call void @llvm.donothing() ; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0 ; CHECK-NEXT: br i1 [[_MSCMP]], label %[[BB5:.*]], label %[[BB6:.*]], !prof [[PROF1]] @@ -505,38 +593,62 @@ define <4 x i32>@test_int_x86_avx2_vpdpbuuds_128(<4 x i32> %x0, <4 x i32> %x1, p ; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4]] ; CHECK-NEXT: unreachable ; CHECK: [[BB6]]: -; CHECK-NEXT: [[X2:%.*]] = load <4 x i32>, ptr [[X2P]], align 16 +; CHECK-NEXT: [[X2:%.*]] = load <16 x i8>, ptr [[X2P]], align 16 ; CHECK-NEXT: [[TMP7:%.*]] = ptrtoint ptr [[X2P]] to i64 ; CHECK-NEXT: [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080 ; CHECK-NEXT: [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr -; CHECK-NEXT: [[_MSLD:%.*]] = load <4 x i32>, ptr [[TMP9]], align 16 -; CHECK-NEXT: [[_MSPROP:%.*]] = or <4 x i32> [[TMP2]], [[TMP3]] -; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[_MSPROP]], [[_MSLD]] -; CHECK-NEXT: [[TMP10:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X2]]) -; CHECK-NEXT: [[_MSPROP2:%.*]] = or <4 x i32> [[TMP2]], [[TMP3]] -; CHECK-NEXT: [[_MSPROP3:%.*]] = or <4 x i32> [[_MSPROP2]], [[TMP4]] -; CHECK-NEXT: [[TMP11:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32> [[X0]], <4 x i32> [[X1]], <4 x i32> [[X4]]) +; CHECK-NEXT: [[_MSLD:%.*]] = load <16 x i8>, ptr [[TMP9]], align 16 +; CHECK-NEXT: [[TMP23:%.*]] = icmp ne <16 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP24:%.*]] = icmp ne <16 x i8> [[_MSLD]], zeroinitializer +; CHECK-NEXT: [[TMP12:%.*]] = icmp ne <16 x i8> [[X1]], zeroinitializer +; CHECK-NEXT: [[TMP13:%.*]] = icmp ne <16 x i8> [[X2]], zeroinitializer +; CHECK-NEXT: [[TMP14:%.*]] = and <16 x i1> [[TMP23]], [[TMP24]] +; CHECK-NEXT: [[TMP15:%.*]] = and <16 x i1> [[TMP12]], [[TMP24]] +; CHECK-NEXT: [[TMP16:%.*]] = and <16 x i1> [[TMP23]], [[TMP13]] +; CHECK-NEXT: [[TMP17:%.*]] = or <16 x i1> [[TMP14]], [[TMP15]] +; CHECK-NEXT: [[TMP18:%.*]] = or <16 x i1> [[TMP17]], [[TMP16]] +; CHECK-NEXT: [[TMP19:%.*]] = sext <16 x i1> [[TMP18]] to <16 x i8> +; CHECK-NEXT: [[TMP20:%.*]] = bitcast <16 x i8> [[TMP19]] to <4 x i32> +; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <4 x i32> [[TMP20]], zeroinitializer +; CHECK-NEXT: [[TMP22:%.*]] = sext <4 x i1> [[TMP21]] to <4 x i32> +; CHECK-NEXT: [[_MSPROP1:%.*]] = or <4 x i32> [[TMP22]], [[TMP2]] +; CHECK-NEXT: [[TMP10:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32> [[X0]], <16 x i8> [[X1]], <16 x i8> [[X2]]) +; CHECK-NEXT: [[TMP25:%.*]] = icmp ne <16 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP26:%.*]] = icmp ne <16 x i8> [[TMP4]], zeroinitializer +; CHECK-NEXT: [[TMP27:%.*]] = icmp ne <16 x i8> [[X1]], zeroinitializer +; CHECK-NEXT: [[TMP28:%.*]] = icmp ne <16 x i8> [[X4]], zeroinitializer +; CHECK-NEXT: [[TMP29:%.*]] = and <16 x i1> [[TMP25]], [[TMP26]] +; CHECK-NEXT: [[TMP30:%.*]] = and <16 x i1> [[TMP27]], [[TMP26]] +; CHECK-NEXT: [[TMP31:%.*]] = and <16 x i1> [[TMP25]], [[TMP28]] +; CHECK-NEXT: [[TMP32:%.*]] = or <16 x i1> [[TMP29]], [[TMP30]] +; CHECK-NEXT: [[TMP33:%.*]] = or <16 x i1> [[TMP32]], [[TMP31]] +; CHECK-NEXT: [[TMP34:%.*]] = sext <16 x i1> [[TMP33]] to <16 x i8> +; CHECK-NEXT: [[TMP35:%.*]] = bitcast <16 x i8> [[TMP34]] to <4 x i32> +; CHECK-NEXT: [[TMP36:%.*]] = icmp ne <4 x i32> [[TMP35]], zeroinitializer +; CHECK-NEXT: [[TMP37:%.*]] = sext <4 x i1> [[TMP36]] to <4 x i32> +; CHECK-NEXT: [[_MSPROP3:%.*]] = or <4 x i32> [[TMP37]], [[TMP2]] +; CHECK-NEXT: [[TMP11:%.*]] = call <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32> [[X0]], <16 x i8> [[X1]], <16 x i8> [[X4]]) ; CHECK-NEXT: [[_MSPROP4:%.*]] = or <4 x i32> [[_MSPROP1]], [[_MSPROP3]] ; CHECK-NEXT: [[RES:%.*]] = add <4 x i32> [[TMP10]], [[TMP11]] ; CHECK-NEXT: store <4 x i32> [[_MSPROP4]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <4 x i32> [[RES]] ; - %x2 = load <4 x i32>, ptr %x2p - %1 = call <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x2) - %2 = call <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32> %x0, <4 x i32> %x1, <4 x i32> %x4) + %x2 = load <16 x i8>, ptr %x2p + %1 = call <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x2) + %2 = call <4 x i32> @llvm.x86.avx2.vpdpbuuds.128(<4 x i32> %x0, <16 x i8> %x1, <16 x i8> %x4) %res = add <4 x i32> %1, %2 ret <4 x i32> %res } -declare <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32>, <8 x i32>, <8 x i32>) +declare <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32>, <32 x i8>, <32 x i8>) -define <8 x i32>@test_int_x86_avx2_vpdpbuud_256(<8 x i32> %x0, <8 x i32> %x1, ptr %x2p, <8 x i32> %x4) sanitize_memory { +define <8 x i32>@test_int_x86_avx2_vpdpbuud_256(<8 x i32> %x0, <32 x i8> %x1, ptr %x2p, <32 x i8> %x4) sanitize_memory { ; CHECK-LABEL: define <8 x i32> @test_int_x86_avx2_vpdpbuud_256( -; CHECK-SAME: <8 x i32> [[X0:%.*]], <8 x i32> [[X1:%.*]], ptr [[X2P:%.*]], <8 x i32> [[X4:%.*]]) #[[ATTR1]] { +; CHECK-SAME: <8 x i32> [[X0:%.*]], <32 x i8> [[X1:%.*]], ptr [[X2P:%.*]], <32 x i8> [[X4:%.*]]) #[[ATTR1]] { ; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8 +; CHECK-NEXT: [[TMP3:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8 ; CHECK-NEXT: [[TMP2:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8 -; CHECK-NEXT: [[TMP4:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 +; CHECK-NEXT: [[TMP4:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 ; CHECK-NEXT: call void @llvm.donothing() ; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0 ; CHECK-NEXT: br i1 [[_MSCMP]], label %[[BB5:.*]], label %[[BB6:.*]], !prof [[PROF1]] @@ -544,38 +656,62 @@ define <8 x i32>@test_int_x86_avx2_vpdpbuud_256(<8 x i32> %x0, <8 x i32> %x1, pt ; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4]] ; CHECK-NEXT: unreachable ; CHECK: [[BB6]]: -; CHECK-NEXT: [[X2:%.*]] = load <8 x i32>, ptr [[X2P]], align 32 +; CHECK-NEXT: [[X2:%.*]] = load <32 x i8>, ptr [[X2P]], align 32 ; CHECK-NEXT: [[TMP7:%.*]] = ptrtoint ptr [[X2P]] to i64 ; CHECK-NEXT: [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080 ; CHECK-NEXT: [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr -; CHECK-NEXT: [[_MSLD:%.*]] = load <8 x i32>, ptr [[TMP9]], align 32 -; CHECK-NEXT: [[_MSPROP:%.*]] = or <8 x i32> [[TMP2]], [[TMP3]] -; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[_MSPROP]], [[_MSLD]] -; CHECK-NEXT: [[TMP10:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X2]]) -; CHECK-NEXT: [[_MSPROP2:%.*]] = or <8 x i32> [[TMP2]], [[TMP3]] -; CHECK-NEXT: [[_MSPROP3:%.*]] = or <8 x i32> [[_MSPROP2]], [[TMP4]] -; CHECK-NEXT: [[TMP11:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X4]]) +; CHECK-NEXT: [[_MSLD:%.*]] = load <32 x i8>, ptr [[TMP9]], align 32 +; CHECK-NEXT: [[TMP23:%.*]] = icmp ne <32 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP24:%.*]] = icmp ne <32 x i8> [[_MSLD]], zeroinitializer +; CHECK-NEXT: [[TMP12:%.*]] = icmp ne <32 x i8> [[X1]], zeroinitializer +; CHECK-NEXT: [[TMP13:%.*]] = icmp ne <32 x i8> [[X2]], zeroinitializer +; CHECK-NEXT: [[TMP14:%.*]] = and <32 x i1> [[TMP23]], [[TMP24]] +; CHECK-NEXT: [[TMP15:%.*]] = and <32 x i1> [[TMP12]], [[TMP24]] +; CHECK-NEXT: [[TMP16:%.*]] = and <32 x i1> [[TMP23]], [[TMP13]] +; CHECK-NEXT: [[TMP17:%.*]] = or <32 x i1> [[TMP14]], [[TMP15]] +; CHECK-NEXT: [[TMP18:%.*]] = or <32 x i1> [[TMP17]], [[TMP16]] +; CHECK-NEXT: [[TMP19:%.*]] = sext <32 x i1> [[TMP18]] to <32 x i8> +; CHECK-NEXT: [[TMP20:%.*]] = bitcast <32 x i8> [[TMP19]] to <8 x i32> +; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <8 x i32> [[TMP20]], zeroinitializer +; CHECK-NEXT: [[TMP22:%.*]] = sext <8 x i1> [[TMP21]] to <8 x i32> +; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[TMP22]], [[TMP2]] +; CHECK-NEXT: [[TMP10:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32> [[X0]], <32 x i8> [[X1]], <32 x i8> [[X2]]) +; CHECK-NEXT: [[TMP25:%.*]] = icmp ne <32 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP26:%.*]] = icmp ne <32 x i8> [[TMP4]], zeroinitializer +; CHECK-NEXT: [[TMP27:%.*]] = icmp ne <32 x i8> [[X1]], zeroinitializer +; CHECK-NEXT: [[TMP28:%.*]] = icmp ne <32 x i8> [[X4]], zeroinitializer +; CHECK-NEXT: [[TMP29:%.*]] = and <32 x i1> [[TMP25]], [[TMP26]] +; CHECK-NEXT: [[TMP30:%.*]] = and <32 x i1> [[TMP27]], [[TMP26]] +; CHECK-NEXT: [[TMP31:%.*]] = and <32 x i1> [[TMP25]], [[TMP28]] +; CHECK-NEXT: [[TMP32:%.*]] = or <32 x i1> [[TMP29]], [[TMP30]] +; CHECK-NEXT: [[TMP33:%.*]] = or <32 x i1> [[TMP32]], [[TMP31]] +; CHECK-NEXT: [[TMP34:%.*]] = sext <32 x i1> [[TMP33]] to <32 x i8> +; CHECK-NEXT: [[TMP35:%.*]] = bitcast <32 x i8> [[TMP34]] to <8 x i32> +; CHECK-NEXT: [[TMP36:%.*]] = icmp ne <8 x i32> [[TMP35]], zeroinitializer +; CHECK-NEXT: [[TMP37:%.*]] = sext <8 x i1> [[TMP36]] to <8 x i32> +; CHECK-NEXT: [[_MSPROP3:%.*]] = or <8 x i32> [[TMP37]], [[TMP2]] +; CHECK-NEXT: [[TMP11:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32> [[X0]], <32 x i8> [[X1]], <32 x i8> [[X4]]) ; CHECK-NEXT: [[_MSPROP4:%.*]] = or <8 x i32> [[_MSPROP1]], [[_MSPROP3]] ; CHECK-NEXT: [[RES:%.*]] = add <8 x i32> [[TMP10]], [[TMP11]] ; CHECK-NEXT: store <8 x i32> [[_MSPROP4]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <8 x i32> [[RES]] ; - %x2 = load <8 x i32>, ptr %x2p - %1 = call <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) - %2 = call <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x4) + %x2 = load <32 x i8>, ptr %x2p + %1 = call <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x2) + %2 = call <8 x i32> @llvm.x86.avx2.vpdpbuud.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x4) %res = add <8 x i32> %1, %2 ret <8 x i32> %res } -declare <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32>, <8 x i32>, <8 x i32>) +declare <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32>, <32 x i8>, <32 x i8>) -define <8 x i32>@test_int_x86_avx2_vpdpbuuds_256(<8 x i32> %x0, <8 x i32> %x1, ptr %x2p, <8 x i32> %x4) sanitize_memory { +define <8 x i32>@test_int_x86_avx2_vpdpbuuds_256(<8 x i32> %x0, <32 x i8> %x1, ptr %x2p, <32 x i8> %x4) sanitize_memory { ; CHECK-LABEL: define <8 x i32> @test_int_x86_avx2_vpdpbuuds_256( -; CHECK-SAME: <8 x i32> [[X0:%.*]], <8 x i32> [[X1:%.*]], ptr [[X2P:%.*]], <8 x i32> [[X4:%.*]]) #[[ATTR1]] { +; CHECK-SAME: <8 x i32> [[X0:%.*]], <32 x i8> [[X1:%.*]], ptr [[X2P:%.*]], <32 x i8> [[X4:%.*]]) #[[ATTR1]] { ; CHECK-NEXT: [[TMP1:%.*]] = load i64, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 64) to ptr), align 8 +; CHECK-NEXT: [[TMP3:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8 ; CHECK-NEXT: [[TMP2:%.*]] = load <8 x i32>, ptr @__msan_param_tls, align 8 -; CHECK-NEXT: [[TMP3:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 32) to ptr), align 8 -; CHECK-NEXT: [[TMP4:%.*]] = load <8 x i32>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 +; CHECK-NEXT: [[TMP4:%.*]] = load <32 x i8>, ptr inttoptr (i64 add (i64 ptrtoint (ptr @__msan_param_tls to i64), i64 72) to ptr), align 8 ; CHECK-NEXT: call void @llvm.donothing() ; CHECK-NEXT: [[_MSCMP:%.*]] = icmp ne i64 [[TMP1]], 0 ; CHECK-NEXT: br i1 [[_MSCMP]], label %[[BB5:.*]], label %[[BB6:.*]], !prof [[PROF1]] @@ -583,25 +719,49 @@ define <8 x i32>@test_int_x86_avx2_vpdpbuuds_256(<8 x i32> %x0, <8 x i32> %x1, p ; CHECK-NEXT: call void @__msan_warning_noreturn() #[[ATTR4]] ; CHECK-NEXT: unreachable ; CHECK: [[BB6]]: -; CHECK-NEXT: [[X2:%.*]] = load <8 x i32>, ptr [[X2P]], align 32 +; CHECK-NEXT: [[X2:%.*]] = load <32 x i8>, ptr [[X2P]], align 32 ; CHECK-NEXT: [[TMP7:%.*]] = ptrtoint ptr [[X2P]] to i64 ; CHECK-NEXT: [[TMP8:%.*]] = xor i64 [[TMP7]], 87960930222080 ; CHECK-NEXT: [[TMP9:%.*]] = inttoptr i64 [[TMP8]] to ptr -; CHECK-NEXT: [[_MSLD:%.*]] = load <8 x i32>, ptr [[TMP9]], align 32 -; CHECK-NEXT: [[_MSPROP:%.*]] = or <8 x i32> [[TMP2]], [[TMP3]] -; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[_MSPROP]], [[_MSLD]] -; CHECK-NEXT: [[TMP10:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X2]]) -; CHECK-NEXT: [[_MSPROP2:%.*]] = or <8 x i32> [[TMP2]], [[TMP3]] -; CHECK-NEXT: [[_MSPROP3:%.*]] = or <8 x i32> [[_MSPROP2]], [[TMP4]] -; CHECK-NEXT: [[TMP11:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32> [[X0]], <8 x i32> [[X1]], <8 x i32> [[X4]]) +; CHECK-NEXT: [[_MSLD:%.*]] = load <32 x i8>, ptr [[TMP9]], align 32 +; CHECK-NEXT: [[TMP23:%.*]] = icmp ne <32 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP24:%.*]] = icmp ne <32 x i8> [[_MSLD]], zeroinitializer +; CHECK-NEXT: [[TMP12:%.*]] = icmp ne <32 x i8> [[X1]], zeroinitializer +; CHECK-NEXT: [[TMP13:%.*]] = icmp ne <32 x i8> [[X2]], zeroinitializer +; CHECK-NEXT: [[TMP14:%.*]] = and <32 x i1> [[TMP23]], [[TMP24]] +; CHECK-NEXT: [[TMP15:%.*]] = and <32 x i1> [[TMP12]], [[TMP24]] +; CHECK-NEXT: [[TMP16:%.*]] = and <32 x i1> [[TMP23]], [[TMP13]] +; CHECK-NEXT: [[TMP17:%.*]] = or <32 x i1> [[TMP14]], [[TMP15]] +; CHECK-NEXT: [[TMP18:%.*]] = or <32 x i1> [[TMP17]], [[TMP16]] +; CHECK-NEXT: [[TMP19:%.*]] = sext <32 x i1> [[TMP18]] to <32 x i8> +; CHECK-NEXT: [[TMP20:%.*]] = bitcast <32 x i8> [[TMP19]] to <8 x i32> +; CHECK-NEXT: [[TMP21:%.*]] = icmp ne <8 x i32> [[TMP20]], zeroinitializer +; CHECK-NEXT: [[TMP22:%.*]] = sext <8 x i1> [[TMP21]] to <8 x i32> +; CHECK-NEXT: [[_MSPROP1:%.*]] = or <8 x i32> [[TMP22]], [[TMP2]] +; CHECK-NEXT: [[TMP10:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32> [[X0]], <32 x i8> [[X1]], <32 x i8> [[X2]]) +; CHECK-NEXT: [[TMP25:%.*]] = icmp ne <32 x i8> [[TMP3]], zeroinitializer +; CHECK-NEXT: [[TMP26:%.*]] = icmp ne <32 x i8> [[TMP4]], zeroinitializer +; CHECK-NEXT: [[TMP27:%.*]] = icmp ne <32 x i8> [[X1]], zeroinitializer +; CHECK-NEXT: [[TMP28:%.*]] = icmp ne <32 x i8> [[X4]], zeroinitializer +; CHECK-NEXT: [[TMP29:%.*]] = and <32 x i1> [[TMP25]], [[TMP26]] +; CHECK-NEXT: [[TMP30:%.*]] = and <32 x i1> [[TMP27]], [[TMP26]] +; CHECK-NEXT: [[TMP31:%.*]] = and <32 x i1> [[TMP25]], [[TMP28]] +; CHECK-NEXT: [[TMP32:%.*]] = or <32 x i1> [[TMP29]], [[TMP30]] +; CHECK-NEXT: [[TMP33:%.*]] = or <32 x i1> [[TMP32]], [[TMP31]] +; CHECK-NEXT: [[TMP34:%.*]] = sext <32 x i1> [[TMP33]] to <32 x i8> +; CHECK-NEXT: [[TMP35:%.*]] = bitcast <32 x i8> [[TMP34]] to <8 x i32> +; CHECK-NEXT: [[TMP36:%.*]] = icmp ne <8 x i32> [[TMP35]], zeroinitializer +; CHECK-NEXT: [[TMP37:%.*]] = sext <8 x i1> [[TMP36]] to <8 x i32> +; CHECK-NEXT: [[_MSPROP3:%.*]] = or <8 x i32> [[TMP37]], [[TMP2]] +; CHECK-NEXT: [[TMP11:%.*]] = call <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32> [[X0]], <32 x i8> [[X1]], <32 x i8> [[X4]]) ; CHECK-NEXT: [[_MSPROP4:%.*]] = or <8 x i32> [[_MSPROP1]], [[_MSPROP3]] ; CHECK-NEXT: [[RES:%.*]] = add <8 x i32> [[TMP10]], [[TMP11]] ; CHECK-NEXT: store <8 x i32> [[_MSPROP4]], ptr @__msan_retval_tls, align 8 ; CHECK-NEXT: ret <8 x i32> [[RES]] ; - %x2 = load <8 x i32>, ptr %x2p - %1 = call <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x2) - %2 = call <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32> %x0, <8 x i32> %x1, <8 x i32> %x4) + %x2 = load <32 x i8>, ptr %x2p + %1 = call <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x2) + %2 = call <8 x i32> @llvm.x86.avx2.vpdpbuuds.256(<8 x i32> %x0, <32 x i8> %x1, <32 x i8> %x4) %res = add <8 x i32> %1, %2 ret <8 x i32> %res } diff --git a/llvm/test/MC/LoongArch/Macros/macros-la.s b/llvm/test/MC/LoongArch/Macros/macros-la.s index a732988..8022d5b 100644 --- a/llvm/test/MC/LoongArch/Macros/macros-la.s +++ b/llvm/test/MC/LoongArch/Macros/macros-la.s @@ -26,6 +26,7 @@ la.abs $a0, sym_abs # ABS-NEXT: lu32i.d $a0, %abs64_lo20(sym_abs) # ABS-NEXT: lu52i.d $a0, $a0, %abs64_hi12(sym_abs) # ABS-EMPTY: +# RELOC-NEXT: R_LARCH_MARK_LA - 0x0 # RELOC-NEXT: R_LARCH_ABS_HI20 sym_abs 0x0 # RELOC-NEXT: R_LARCH_ABS_LO12 sym_abs 0x0 # RELOC-NEXT: R_LARCH_ABS64_LO20 sym_abs 0x0 diff --git a/llvm/test/Transforms/AggressiveInstCombine/memchr.ll b/llvm/test/Transforms/AggressiveInstCombine/memchr.ll index b26320b..6fbe960 100644 --- a/llvm/test/Transforms/AggressiveInstCombine/memchr.ll +++ b/llvm/test/Transforms/AggressiveInstCombine/memchr.ll @@ -6,9 +6,10 @@ declare ptr @memchr(ptr, i32, i64) -define i1 @test_memchr_null(i32 %x) { +define i1 @test_memchr_null(i32 %x) !prof !0 { ; CHECK-LABEL: define i1 @test_memchr_null( -; CHECK-SAME: i32 [[X:%.*]]) { +; CHECK-SAME: i32 [[X:%.*]]) +; CHECK: !prof [[PROF_0:![0-9]+]] { ; CHECK-NEXT: [[ENTRY:.*]]: ; CHECK-NEXT: [[TMP0:%.*]] = trunc i32 [[X]] to i8 ; CHECK-NEXT: switch i8 [[TMP0]], label %[[ENTRY_SPLIT:.*]] [ @@ -40,9 +41,10 @@ entry: ret i1 %isnull } -define ptr @test_memchr(i32 %x) { +define ptr @test_memchr(i32 %x) !prof !0 { ; CHECK-LABEL: define ptr @test_memchr( -; CHECK-SAME: i32 [[X:%.*]]) { +; CHECK-SAME: i32 [[X:%.*]]) +; CHECK: !prof [[PROF_0]] { ; CHECK-NEXT: [[ENTRY:.*]]: ; CHECK-NEXT: [[TMP0:%.*]] = trunc i32 [[X]] to i8 ; CHECK-NEXT: switch i8 [[TMP0]], label %[[ENTRY_SPLIT:.*]] [ @@ -72,16 +74,17 @@ entry: ret ptr %memchr } -define ptr @test_memchr_smaller_n(i32 %x) { +define ptr @test_memchr_smaller_n(i32 %x) !prof !0 { ; CHECK-LABEL: define ptr @test_memchr_smaller_n( -; CHECK-SAME: i32 [[X:%.*]]) { +; CHECK-SAME: i32 [[X:%.*]]) +; CHECK: !prof [[PROF_0]] { ; CHECK-NEXT: [[ENTRY:.*]]: ; CHECK-NEXT: [[TMP0:%.*]] = trunc i32 [[X]] to i8 ; CHECK-NEXT: switch i8 [[TMP0]], label %[[ENTRY_SPLIT:.*]] [ ; CHECK-NEXT: i8 48, label %[[MEMCHR_CASE:.*]] ; CHECK-NEXT: i8 49, label %[[MEMCHR_CASE1:.*]] ; CHECK-NEXT: i8 0, label %[[MEMCHR_CASE2:.*]] -; CHECK-NEXT: ] +; CHECK-NEXT: ], !prof [[PROF_1:![0-9]+]] ; CHECK: [[MEMCHR_CASE]]: ; CHECK-NEXT: br label %[[MEMCHR_SUCCESS:.*]] ; CHECK: [[MEMCHR_CASE1]]: @@ -103,9 +106,10 @@ entry: ; negative tests -define ptr @test_memchr_larger_n(i32 %x) { +define ptr @test_memchr_larger_n(i32 %x) !prof !0 { ; CHECK-LABEL: define ptr @test_memchr_larger_n( -; CHECK-SAME: i32 [[X:%.*]]) { +; CHECK-SAME: i32 [[X:%.*]]) +; CHECK: !prof [[PROF_0]] { ; CHECK-NEXT: [[ENTRY:.*:]] ; CHECK-NEXT: [[MEMCHR:%.*]] = call ptr @memchr(ptr @str, i32 [[X]], i64 6) ; CHECK-NEXT: ret ptr [[MEMCHR]] @@ -115,9 +119,10 @@ entry: ret ptr %memchr } -define ptr @test_memchr_non_constant(i32 %x, ptr %str) { +define ptr @test_memchr_non_constant(i32 %x, ptr %str) !prof !0 { ; CHECK-LABEL: define ptr @test_memchr_non_constant( -; CHECK-SAME: i32 [[X:%.*]], ptr [[STR:%.*]]) { +; CHECK-SAME: i32 [[X:%.*]], ptr [[STR:%.*]]) +; CHECK: !prof [[PROF_0]] { ; CHECK-NEXT: [[ENTRY:.*:]] ; CHECK-NEXT: [[MEMCHR:%.*]] = call ptr @memchr(ptr [[STR]], i32 [[X]], i64 5) ; CHECK-NEXT: ret ptr [[MEMCHR]] @@ -127,8 +132,9 @@ entry: ret ptr %memchr } -define ptr @test_memchr_constant_ch() { -; CHECK-LABEL: define ptr @test_memchr_constant_ch() { +define ptr @test_memchr_constant_ch() !prof !0 { +; CHECK-LABEL: define ptr @test_memchr_constant_ch() +; CHECK: !prof [[PROF_0]] { ; CHECK-NEXT: [[ENTRY:.*:]] ; CHECK-NEXT: [[MEMCHR:%.*]] = call ptr @memchr(ptr @str, i32 49, i64 5) ; CHECK-NEXT: ret ptr [[MEMCHR]] @@ -138,9 +144,10 @@ entry: ret ptr %memchr } -define ptr @test_memchr_dynamic_n(i32 %x, i32 %y) { +define ptr @test_memchr_dynamic_n(i32 %x, i32 %y) !prof !0 { ; CHECK-LABEL: define ptr @test_memchr_dynamic_n( -; CHECK-SAME: i32 [[X:%.*]], i32 [[Y:%.*]]) { +; CHECK-SAME: i32 [[X:%.*]], i32 [[Y:%.*]]) +; CHECK: !prof [[PROF_0]] { ; CHECK-NEXT: [[ENTRY:.*:]] ; CHECK-NEXT: [[MEMCHR:%.*]] = call ptr @memchr(ptr @str, i32 [[X]], i32 [[Y]]) ; CHECK-NEXT: ret ptr [[MEMCHR]] @@ -150,9 +157,10 @@ entry: ret ptr %memchr } -define ptr @test_memchr_long(i32 %x) { +define ptr @test_memchr_long(i32 %x) !prof !0 { ; CHECK-LABEL: define ptr @test_memchr_long( -; CHECK-SAME: i32 [[X:%.*]]) { +; CHECK-SAME: i32 [[X:%.*]]) +; CHECK: !prof [[PROF_0]] { ; CHECK-NEXT: [[ENTRY:.*:]] ; CHECK-NEXT: [[MEMCHR:%.*]] = call ptr @memchr(ptr @str_long, i32 [[X]], i64 8) ; CHECK-NEXT: ret ptr [[MEMCHR]] @@ -163,9 +171,10 @@ entry: } ; We want to check that the compiler still calls memchr if the length is non-constant: -define ptr @test_memchr_non_constant_length2(i32 %x, i64 %len) { +define ptr @test_memchr_non_constant_length2(i32 %x, i64 %len) !prof !0 { ; CHECK-LABEL: define ptr @test_memchr_non_constant_length2( -; CHECK-SAME: i32 [[X:%.*]], i64 [[LEN:%.*]]) { +; CHECK-SAME: i32 [[X:%.*]], i64 [[LEN:%.*]]) +; CHECK: !prof [[PROF_0]] { ; CHECK-NEXT: [[ENTRY:.*:]] ; CHECK-NEXT: [[MEMCHR:%.*]] = call ptr @memchr(ptr @str, i32 [[X]], i64 [[LEN]]) ; CHECK-NEXT: ret ptr [[MEMCHR]] @@ -174,3 +183,7 @@ entry: %memchr = call ptr @memchr(ptr @str, i32 %x, i64 %len) ret ptr %memchr } + +!0 = !{!"function_entry_count", i64 1000} +; CHECK: [[PROF_0]] = !{!"function_entry_count", i64 1000} +; CHECK: [[PROF_1]] = !{!"unknown", !"aggressive-instcombine"}
\ No newline at end of file diff --git a/llvm/test/Transforms/LoopVectorize/vplan-printing-reductions.ll b/llvm/test/Transforms/LoopVectorize/vplan-printing-reductions.ll index 4e6ef0d..5a0c69b 100644 --- a/llvm/test/Transforms/LoopVectorize/vplan-printing-reductions.ll +++ b/llvm/test/Transforms/LoopVectorize/vplan-printing-reductions.ll @@ -580,6 +580,127 @@ exit: ret i32 %add } +define i32 @print_mulacc_negated(ptr %a, ptr %b) { +; CHECK-LABEL: 'print_mulacc_negated' +; CHECK: VPlan 'Initial VPlan for VF={4},UF>=1' { +; CHECK-NEXT: Live-in vp<%0> = VF +; CHECK-NEXT: Live-in vp<%1> = VF * UF +; CHECK-NEXT: Live-in vp<%2> = vector-trip-count +; CHECK-NEXT: Live-in ir<1024> = original trip-count +; CHECK-EMPTY: +; CHECK-NEXT: ir-bb<entry>: +; CHECK-NEXT: Successor(s): scalar.ph, vector.ph +; CHECK-EMPTY: +; CHECK-NEXT: vector.ph: +; CHECK-NEXT: EMIT vp<%3> = reduction-start-vector ir<0>, ir<0>, ir<1> +; CHECK-NEXT: Successor(s): vector loop +; CHECK-EMPTY: +; CHECK-NEXT: <x1> vector loop: { +; CHECK-NEXT: vector.body: +; CHECK-NEXT: EMIT vp<%4> = CANONICAL-INDUCTION ir<0>, vp<%index.next> +; CHECK-NEXT: WIDEN-REDUCTION-PHI ir<%accum> = phi vp<%3>, vp<%8> +; CHECK-NEXT: vp<%5> = SCALAR-STEPS vp<%4>, ir<1>, vp<%0> +; CHECK-NEXT: CLONE ir<%gep.a> = getelementptr ir<%a>, vp<%5> +; CHECK-NEXT: vp<%6> = vector-pointer ir<%gep.a> +; CHECK-NEXT: WIDEN ir<%load.a> = load vp<%6> +; CHECK-NEXT: CLONE ir<%gep.b> = getelementptr ir<%b>, vp<%5> +; CHECK-NEXT: vp<%7> = vector-pointer ir<%gep.b> +; CHECK-NEXT: WIDEN ir<%load.b> = load vp<%7> +; CHECK-NEXT: EXPRESSION vp<%8> = ir<%accum> + reduce.add (sub (0, mul (ir<%load.b> zext to i32), (ir<%load.a> zext to i32))) +; CHECK-NEXT: EMIT vp<%index.next> = add nuw vp<%4>, vp<%1> +; CHECK-NEXT: EMIT branch-on-count vp<%index.next>, vp<%2> +; CHECK-NEXT: No successors +; CHECK-NEXT: } +; CHECK-NEXT: Successor(s): middle.block +; CHECK-EMPTY: +; CHECK-NEXT: middle.block: +; CHECK-NEXT: EMIT vp<%10> = compute-reduction-result ir<%accum>, vp<%8> +; CHECK-NEXT: EMIT vp<%cmp.n> = icmp eq ir<1024>, vp<%2> +; CHECK-NEXT: EMIT branch-on-cond vp<%cmp.n> +; CHECK-NEXT: Successor(s): ir-bb<exit>, scalar.ph +; CHECK-EMPTY: +; CHECK-NEXT: ir-bb<exit>: +; CHECK-NEXT: IR %add.lcssa = phi i32 [ %add, %loop ] (extra operand: vp<%10> from middle.block) +; CHECK-NEXT: No successors +; CHECK-EMPTY: +; CHECK-NEXT: scalar.ph: +; CHECK-NEXT: EMIT-SCALAR vp<%bc.resume.val> = phi [ vp<%2>, middle.block ], [ ir<0>, ir-bb<entry> ] +; CHECK-NEXT: EMIT-SCALAR vp<%bc.merge.rdx> = phi [ vp<%10>, middle.block ], [ ir<0>, ir-bb<entry> ] +; CHECK-NEXT: Successor(s): ir-bb<loop> +; CHECK-EMPTY: +; CHECK-NEXT: ir-bb<loop>: +; CHECK-NEXT: IR %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ] (extra operand: vp<%bc.resume.val> from scalar.ph) +; CHECK-NEXT: IR %accum = phi i32 [ 0, %entry ], [ %add, %loop ] (extra operand: vp<%bc.merge.rdx> from scalar.ph) +; CHECK-NEXT: IR %gep.a = getelementptr i8, ptr %a, i64 %iv +; CHECK-NEXT: IR %load.a = load i8, ptr %gep.a, align 1 +; CHECK-NEXT: IR %ext.a = zext i8 %load.a to i32 +; CHECK-NEXT: IR %gep.b = getelementptr i8, ptr %b, i64 %iv +; CHECK-NEXT: IR %load.b = load i8, ptr %gep.b, align 1 +; CHECK-NEXT: IR %ext.b = zext i8 %load.b to i32 +; CHECK-NEXT: IR %mul = mul i32 %ext.b, %ext.a +; CHECK-NEXT: IR %sub = sub i32 0, %mul +; CHECK-NEXT: IR %add = add i32 %accum, %sub +; CHECK-NEXT: IR %iv.next = add i64 %iv, 1 +; CHECK-NEXT: IR %exitcond.not = icmp eq i64 %iv.next, 1024 +; CHECK-NEXT: No successors +; CHECK-NEXT: } +; CHECK: VPlan 'Final VPlan for VF={4},UF={1}' { +; CHECK-NEXT: Live-in ir<1024> = vector-trip-count +; CHECK-NEXT: Live-in ir<1024> = original trip-count +; CHECK-EMPTY: +; CHECK-NEXT: ir-bb<entry>: +; CHECK-NEXT: Successor(s): vector.ph +; CHECK-EMPTY: +; CHECK-NEXT: vector.ph: +; CHECK-NEXT: Successor(s): vector.body +; CHECK-EMPTY: +; CHECK-NEXT: vector.body: +; CHECK-NEXT: EMIT-SCALAR vp<%index> = phi [ ir<0>, vector.ph ], [ vp<%index.next>, vector.body ] +; CHECK-NEXT: WIDEN-REDUCTION-PHI ir<%accum> = phi ir<0>, ir<%add> +; CHECK-NEXT: CLONE ir<%gep.a> = getelementptr ir<%a>, vp<%index> +; CHECK-NEXT: WIDEN ir<%load.a> = load ir<%gep.a> +; CHECK-NEXT: CLONE ir<%gep.b> = getelementptr ir<%b>, vp<%index> +; CHECK-NEXT: WIDEN ir<%load.b> = load ir<%gep.b> +; CHECK-NEXT: WIDEN-CAST ir<%ext.b> = zext ir<%load.b> to i32 +; CHECK-NEXT: WIDEN-CAST ir<%ext.a> = zext ir<%load.a> to i32 +; CHECK-NEXT: WIDEN ir<%mul> = mul ir<%ext.b>, ir<%ext.a> +; CHECK-NEXT: WIDEN ir<%sub> = sub ir<0>, ir<%mul> +; CHECK-NEXT: REDUCE ir<%add> = ir<%accum> + reduce.add (ir<%sub>) +; CHECK-NEXT: EMIT vp<%index.next> = add nuw vp<%index>, ir<4> +; CHECK-NEXT: EMIT branch-on-count vp<%index.next>, ir<1024> +; CHECK-NEXT: Successor(s): middle.block, vector.body +; CHECK-EMPTY: +; CHECK-NEXT: middle.block: +; CHECK-NEXT: EMIT vp<[[RED_RESULT:%.+]]> = compute-reduction-result ir<%accum>, ir<%add> +; CHECK-NEXT: Successor(s): ir-bb<exit> +; CHECK-EMPTY: +; CHECK-NEXT: ir-bb<exit>: +; CHECK-NEXT: IR %add.lcssa = phi i32 [ %add, %loop ] (extra operand: vp<[[RED_RESULT]]> from middle.block) +; CHECK-NEXT: No successors +; CHECK-NEXT: } +entry: + br label %loop + +loop: + %iv = phi i64 [ 0, %entry ], [ %iv.next, %loop ] + %accum = phi i32 [ 0, %entry ], [ %add, %loop ] + %gep.a = getelementptr i8, ptr %a, i64 %iv + %load.a = load i8, ptr %gep.a, align 1 + %ext.a = zext i8 %load.a to i32 + %gep.b = getelementptr i8, ptr %b, i64 %iv + %load.b = load i8, ptr %gep.b, align 1 + %ext.b = zext i8 %load.b to i32 + %mul = mul i32 %ext.b, %ext.a + %sub = sub i32 0, %mul + %add = add i32 %accum, %sub + %iv.next = add i64 %iv, 1 + %exitcond.not = icmp eq i64 %iv.next, 1024 + br i1 %exitcond.not, label %exit, label %loop + +exit: + ret i32 %add +} + define i64 @print_mulacc_sub_extended(ptr nocapture readonly %x, ptr nocapture readonly %y, i32 %n) { ; CHECK-LABEL: 'print_mulacc_sub_extended' ; CHECK: VPlan 'Initial VPlan for VF={4},UF>=1' { diff --git a/llvm/test/Transforms/VectorCombine/X86/bitop-of-castops.ll b/llvm/test/Transforms/VectorCombine/X86/bitop-of-castops.ll index 79e72aa..38c624e 100644 --- a/llvm/test/Transforms/VectorCombine/X86/bitop-of-castops.ll +++ b/llvm/test/Transforms/VectorCombine/X86/bitop-of-castops.ll @@ -357,7 +357,7 @@ define <4 x i32> @or_sext_v4i8_to_v4i32_constant_with_loss(<4 x i8> %a) { define <4 x i16> @and_trunc_nuw_nsw_constant(<4 x i32> %a) { ; CHECK-LABEL: @and_trunc_nuw_nsw_constant( ; CHECK-NEXT: [[AND_INNER:%.*]] = and <4 x i32> [[A:%.*]], <i32 1, i32 2, i32 3, i32 4> -; CHECK-NEXT: [[AND:%.*]] = trunc <4 x i32> [[AND_INNER]] to <4 x i16> +; CHECK-NEXT: [[AND:%.*]] = trunc nuw nsw <4 x i32> [[AND_INNER]] to <4 x i16> ; CHECK-NEXT: ret <4 x i16> [[AND]] ; %t1 = trunc nuw nsw <4 x i32> %a to <4 x i16> @@ -368,7 +368,7 @@ define <4 x i16> @and_trunc_nuw_nsw_constant(<4 x i32> %a) { define <4 x i8> @and_trunc_nuw_nsw_minus_constant(<4 x i32> %a) { ; CHECK-LABEL: @and_trunc_nuw_nsw_minus_constant( ; CHECK-NEXT: [[AND_INNER:%.*]] = and <4 x i32> [[A:%.*]], <i32 240, i32 241, i32 242, i32 243> -; CHECK-NEXT: [[AND:%.*]] = trunc <4 x i32> [[AND_INNER]] to <4 x i8> +; CHECK-NEXT: [[AND:%.*]] = trunc nuw <4 x i32> [[AND_INNER]] to <4 x i8> ; CHECK-NEXT: ret <4 x i8> [[AND]] ; %t1 = trunc nuw nsw <4 x i32> %a to <4 x i8> @@ -379,7 +379,7 @@ define <4 x i8> @and_trunc_nuw_nsw_minus_constant(<4 x i32> %a) { define <4 x i8> @and_trunc_nuw_nsw_multiconstant(<4 x i32> %a) { ; CHECK-LABEL: @and_trunc_nuw_nsw_multiconstant( ; CHECK-NEXT: [[AND_INNER:%.*]] = and <4 x i32> [[A:%.*]], <i32 240, i32 1, i32 242, i32 3> -; CHECK-NEXT: [[AND:%.*]] = trunc <4 x i32> [[AND_INNER]] to <4 x i8> +; CHECK-NEXT: [[AND:%.*]] = trunc nuw <4 x i32> [[AND_INNER]] to <4 x i8> ; CHECK-NEXT: ret <4 x i8> [[AND]] ; %t1 = trunc nuw nsw <4 x i32> %a to <4 x i8> @@ -391,7 +391,7 @@ define <4 x i8> @and_trunc_nuw_nsw_multiconstant(<4 x i32> %a) { define <4 x i32> @or_zext_nneg_constant(<4 x i16> %a) { ; CHECK-LABEL: @or_zext_nneg_constant( ; CHECK-NEXT: [[OR_INNER:%.*]] = or <4 x i16> [[A:%.*]], <i16 1, i16 2, i16 3, i16 4> -; CHECK-NEXT: [[OR:%.*]] = zext <4 x i16> [[OR_INNER]] to <4 x i32> +; CHECK-NEXT: [[OR:%.*]] = zext nneg <4 x i16> [[OR_INNER]] to <4 x i32> ; CHECK-NEXT: ret <4 x i32> [[OR]] ; %z1 = zext nneg <4 x i16> %a to <4 x i32> |