diff options
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU/fract-match.ll')
| -rw-r--r-- | llvm/test/CodeGen/AMDGPU/fract-match.ll | 428 |
1 files changed, 428 insertions, 0 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/fract-match.ll b/llvm/test/CodeGen/AMDGPU/fract-match.ll index f6ee007..80b4d64 100644 --- a/llvm/test/CodeGen/AMDGPU/fract-match.ll +++ b/llvm/test/CodeGen/AMDGPU/fract-match.ll @@ -14,6 +14,7 @@ ; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=kaveri < %s | FileCheck -check-prefixes=GCN,GFX7 %s ; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx900 < %s | FileCheck -check-prefixes=GCN,GFX8 %s ; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx1100 < %s | FileCheck -check-prefixes=GCN,GFX11 %s +; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx1200 < %s | FileCheck -check-prefixes=GFX12 %s ; Test patterns to match v_fract_* instructions. @@ -103,6 +104,21 @@ define float @safe_math_fract_f32(float %x, ptr addrspace(1) nocapture writeonly ; GFX11-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo ; GFX11-NEXT: global_store_b32 v[1:2], v4, off ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: safe_math_fract_f32: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_fract_f32_e32 v3, v0 +; GFX12-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0| +; GFX12-NEXT: v_floor_f32_e32 v4, v0 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_3) +; GFX12-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo +; GFX12-NEXT: global_store_b32 v[1:2], v4, off +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call float @llvm.floor.f32(float %x) %sub = fsub float %x, %floor @@ -181,6 +197,18 @@ define float @safe_math_fract_f32_noinf_check(float %x, ptr addrspace(1) nocaptu ; GFX11-NEXT: v_fract_f32_e32 v0, v0 ; GFX11-NEXT: global_store_b32 v[1:2], v3, off ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: safe_math_fract_f32_noinf_check: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_floor_f32_e32 v3, v0 +; GFX12-NEXT: v_fract_f32_e32 v0, v0 +; GFX12-NEXT: global_store_b32 v[1:2], v3, off +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call float @llvm.floor.f32(float %x) %sub = fsub float %x, %floor @@ -263,6 +291,22 @@ define float @no_nan_check_math_fract_f32(float %x, ptr addrspace(1) nocapture w ; GFX11-NEXT: v_min_f32_e32 v4, 0x3f7fffff, v4 ; GFX11-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc_lo ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: no_nan_check_math_fract_f32: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_floor_f32_e32 v3, v0 +; GFX12-NEXT: v_cmp_neq_f32_e64 vcc_lo, 0x7f800000, |v0| +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1) +; GFX12-NEXT: v_sub_f32_e32 v4, v0, v3 +; GFX12-NEXT: global_store_b32 v[1:2], v3, off +; GFX12-NEXT: v_min_num_f32_e32 v4, 0x3f7fffff, v4 +; GFX12-NEXT: v_cndmask_b32_e32 v0, 0, v4, vcc_lo +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call float @llvm.floor.f32(float %x) %sub = fsub float %x, %floor @@ -314,6 +358,16 @@ define float @basic_fract_f32_nonans(float nofpclass(nan) %x) { ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_fract_f32_e32 v0, v0 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: basic_fract_f32_nonans: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_fract_f32_e32 v0, v0 +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call float @llvm.floor.f32(float %x) %sub = fsub float %x, %floor @@ -362,6 +416,19 @@ define float @basic_fract_f32_flags_minnum(float %x) { ; GFX11-NEXT: v_sub_f32_e32 v0, v0, v1 ; GFX11-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: basic_fract_f32_flags_minnum: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_floor_f32_e32 v1, v0 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-NEXT: v_sub_f32_e32 v0, v0, v1 +; GFX12-NEXT: v_min_num_f32_e32 v0, 0x3f7fffff, v0 +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call float @llvm.floor.f32(float %x) %sub = fsub float %x, %floor @@ -409,6 +476,16 @@ define float @basic_fract_f32_flags_fsub(float nofpclass(nan) %x) { ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_fract_f32_e32 v0, v0 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: basic_fract_f32_flags_fsub: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_fract_f32_e32 v0, v0 +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call float @llvm.floor.f32(float %x) %sub = fsub nsz float %x, %floor @@ -467,6 +544,17 @@ define <2 x float> @basic_fract_v2f32_nonans(<2 x float> nofpclass(nan) %x) { ; GFX11-NEXT: v_fract_f32_e32 v0, v0 ; GFX11-NEXT: v_fract_f32_e32 v1, v1 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: basic_fract_v2f32_nonans: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_fract_f32_e32 v0, v0 +; GFX12-NEXT: v_fract_f32_e32 v1, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call <2 x float> @llvm.floor.v2f32(<2 x float> %x) %sub = fsub <2 x float> %x, %floor @@ -540,6 +628,20 @@ define float @basic_fract_f32_multi_use_fsub_nonans(float nofpclass(nan) %x, ptr ; GFX11-NEXT: v_fract_f32_e32 v0, v0 ; GFX11-NEXT: global_store_b32 v[1:2], v3, off ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: basic_fract_f32_multi_use_fsub_nonans: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_floor_f32_e32 v3, v0 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-NEXT: v_sub_f32_e32 v3, v0, v3 +; GFX12-NEXT: v_fract_f32_e32 v0, v0 +; GFX12-NEXT: global_store_b32 v[1:2], v3, off +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call float @llvm.floor.f32(float %x) %sub = fsub float %x, %floor @@ -588,6 +690,16 @@ define float @nnan_minnum_fract_f32(float %x) { ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_fract_f32_e32 v0, v0 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: nnan_minnum_fract_f32: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_fract_f32_e32 v0, v0 +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call float @llvm.floor.f32(float %x) %sub = fsub float %x, %floor @@ -638,6 +750,19 @@ define float @nnan_fsub_fract_f32(float %x) { ; GFX11-NEXT: v_sub_f32_e32 v0, v0, v1 ; GFX11-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: nnan_fsub_fract_f32: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_floor_f32_e32 v1, v0 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-NEXT: v_sub_f32_e32 v0, v0, v1 +; GFX12-NEXT: v_min_num_f32_e32 v0, 0x3f7fffff, v0 +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call float @llvm.floor.f32(float %x) %sub = fsub nnan float %x, %floor @@ -686,6 +811,19 @@ define float @nnan_floor_fract_f32(float %x) { ; GFX11-NEXT: v_sub_f32_e32 v0, v0, v1 ; GFX11-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: nnan_floor_fract_f32: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_floor_f32_e32 v1, v0 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-NEXT: v_sub_f32_e32 v0, v0, v1 +; GFX12-NEXT: v_min_num_f32_e32 v0, 0x3f7fffff, v0 +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call nnan float @llvm.floor.f32(float %x) %sub = fsub float %x, %floor @@ -733,6 +871,16 @@ define float @nnan_src_fract_f32(float nofpclass(nan) %x) { ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_fract_f32_e32 v0, v0 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: nnan_src_fract_f32: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_fract_f32_e32 v0, v0 +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call float @llvm.floor.f32(float %x) %sub = fsub float %x, %floor @@ -782,6 +930,19 @@ define float @not_fract_f32_wrong_const(float nofpclass(nan) %x) { ; GFX11-NEXT: v_sub_f32_e32 v0, v0, v1 ; GFX11-NEXT: v_min_f32_e32 v0, 0x3f7ffffe, v0 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: not_fract_f32_wrong_const: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_floor_f32_e32 v1, v0 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-NEXT: v_sub_f32_e32 v0, v0, v1 +; GFX12-NEXT: v_min_num_f32_e32 v0, 0x3f7ffffe, v0 +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call float @llvm.floor.f32(float %x) %sub = fsub float %x, %floor @@ -831,6 +992,19 @@ define float @not_fract_f32_swapped_fsub(float nofpclass(nan) %x) { ; GFX11-NEXT: v_sub_f32_e32 v0, v1, v0 ; GFX11-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: not_fract_f32_swapped_fsub: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_floor_f32_e32 v1, v0 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-NEXT: v_sub_f32_e32 v0, v1, v0 +; GFX12-NEXT: v_min_num_f32_e32 v0, 0x3f7fffff, v0 +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call float @llvm.floor.f32(float %x) %sub = fsub float %floor, %x @@ -880,6 +1054,19 @@ define float @not_fract_f32_not_floor(float nofpclass(nan) %x) { ; GFX11-NEXT: v_sub_f32_e32 v0, v0, v1 ; GFX11-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: not_fract_f32_not_floor: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_trunc_f32_e32 v1, v0 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-NEXT: v_sub_f32_e32 v0, v0, v1 +; GFX12-NEXT: v_min_num_f32_e32 v0, 0x3f7fffff, v0 +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call float @llvm.trunc.f32(float %x) %sub = fsub float %x, %floor @@ -929,6 +1116,19 @@ define float @not_fract_f32_different_floor(float %x, float %y) { ; GFX11-NEXT: v_sub_f32_e32 v0, v0, v1 ; GFX11-NEXT: v_min_f32_e32 v0, 0x3f7fffff, v0 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: not_fract_f32_different_floor: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_floor_f32_e32 v1, v1 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-NEXT: v_sub_f32_e32 v0, v0, v1 +; GFX12-NEXT: v_min_num_f32_e32 v0, 0x3f7fffff, v0 +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call float @llvm.floor.f32(float %y) %sub = fsub float %x, %floor @@ -978,6 +1178,19 @@ define float @not_fract_f32_maxnum(float nofpclass(nan) %x) { ; GFX11-NEXT: v_sub_f32_e32 v0, v0, v1 ; GFX11-NEXT: v_max_f32_e32 v0, 0x3f7fffff, v0 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: not_fract_f32_maxnum: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_floor_f32_e32 v1, v0 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-NEXT: v_sub_f32_e32 v0, v0, v1 +; GFX12-NEXT: v_max_num_f32_e32 v0, 0x3f7fffff, v0 +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call float @llvm.floor.f32(float %x) %sub = fsub float %x, %floor @@ -1000,6 +1213,15 @@ define float @fcmp_uno_check_is_nan_f32(float %x) { ; GCN: ; %bb.0: ; %entry ; GCN-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GCN-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: fcmp_uno_check_is_nan_f32: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call float @llvm.floor.f32(float %x) %sub = fsub float %x, %floor @@ -1054,6 +1276,16 @@ define float @select_nan_fract_f32(float %x) { ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_fract_f32_e32 v0, v0 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: select_nan_fract_f32: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_fract_f32_e32 v0, v0 +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call float @llvm.floor.f32(float %x) %sub = fsub float %x, %floor @@ -1107,6 +1339,16 @@ define float @commuted_select_nan_fract_f32(float %x) { ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_fract_f32_e32 v0, v0 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: commuted_select_nan_fract_f32: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_fract_f32_e32 v0, v0 +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call float @llvm.floor.f32(float %x) %sub = fsub float %x, %floor @@ -1168,6 +1410,22 @@ define float @wrong_commuted_nan_select_f32(float %x) { ; GFX11-NEXT: s_delay_alu instid0(VALU_DEP_1) ; GFX11-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc_lo ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: wrong_commuted_nan_select_f32: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_floor_f32_e32 v1, v0 +; GFX12-NEXT: v_cmp_u_f32_e32 vcc_lo, v0, v0 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-NEXT: v_sub_f32_e32 v1, v0, v1 +; GFX12-NEXT: v_min_num_f32_e32 v1, 0x3f7fffff, v1 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) +; GFX12-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc_lo +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call float @llvm.floor.f32(float %x) %sub = fsub float %x, %floor @@ -1231,6 +1489,16 @@ define half @basic_fract_f16_nonan(half nofpclass(nan) %x) { ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_fract_f16_e32 v0, v0 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: basic_fract_f16_nonan: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_fract_f16_e32 v0, v0 +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call half @llvm.floor.f16(half %x) %sub = fsub half %x, %floor @@ -1313,6 +1581,20 @@ define <2 x half> @basic_fract_v2f16_nonan(<2 x half> nofpclass(nan) %x) { ; GFX11-NEXT: v_fract_f16_e32 v1, v1 ; GFX11-NEXT: v_pack_b32_f16 v0, v0, v1 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: basic_fract_v2f16_nonan: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_lshrrev_b32_e32 v1, 16, v0 +; GFX12-NEXT: v_fract_f16_e32 v0, v0 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1) +; GFX12-NEXT: v_fract_f16_e32 v1, v1 +; GFX12-NEXT: v_pack_b32_f16 v0, v0, v1 +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call <2 x half> @llvm.floor.v2f16(<2 x half> %x) %sub = fsub <2 x half> %x, %floor @@ -1369,6 +1651,16 @@ define double @basic_fract_f64_nanans(double nofpclass(nan) %x) { ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_fract_f64_e32 v[0:1], v[0:1] ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: basic_fract_f64_nanans: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_fract_f64_e32 v[0:1], v[0:1] +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call double @llvm.floor.f64(double %x) %sub = fsub double %x, %floor @@ -1461,6 +1753,18 @@ define half @safe_math_fract_f16_noinf_check(half %x, ptr addrspace(1) nocapture ; GFX11-NEXT: v_fract_f16_e32 v0, v0 ; GFX11-NEXT: global_store_b16 v[1:2], v3, off ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: safe_math_fract_f16_noinf_check: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_floor_f16_e32 v3, v0 +; GFX12-NEXT: v_fract_f16_e32 v0, v0 +; GFX12-NEXT: global_store_b16 v[1:2], v3, off +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call half @llvm.floor.f16(half %x) %sub = fsub half %x, %floor @@ -1546,6 +1850,18 @@ define double @safe_math_fract_f64_noinf_check(double %x, ptr addrspace(1) nocap ; GFX11-NEXT: v_fract_f64_e32 v[0:1], v[0:1] ; GFX11-NEXT: global_store_b64 v[2:3], v[4:5], off ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: safe_math_fract_f64_noinf_check: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_floor_f64_e32 v[4:5], v[0:1] +; GFX12-NEXT: v_fract_f64_e32 v[0:1], v[0:1] +; GFX12-NEXT: global_store_b64 v[2:3], v[4:5], off +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call double @llvm.floor.f64(double %x) %sub = fsub double %x, %floor @@ -1600,6 +1916,16 @@ define float @select_nan_fract_f32_flags_select(float %x) { ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_fract_f32_e32 v0, v0 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: select_nan_fract_f32_flags_select: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_fract_f32_e32 v0, v0 +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call float @llvm.floor.f32(float %x) %sub = fsub float %x, %floor @@ -1653,6 +1979,16 @@ define float @select_nan_fract_f32_flags_minnum(float %x) { ; GFX11-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0) ; GFX11-NEXT: v_fract_f32_e32 v0, v0 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: select_nan_fract_f32_flags_minnum: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_fract_f32_e32 v0, v0 +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call float @llvm.floor.f32(float %x) %sub = fsub float %x, %floor @@ -1769,6 +2105,25 @@ define <2 x float> @safe_math_fract_v2f32(<2 x float> %x, ptr addrspace(1) nocap ; GFX11-NEXT: global_store_b64 v[2:3], v[4:5], off ; GFX11-NEXT: v_cndmask_b32_e64 v1, v7, 0, s0 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: safe_math_fract_v2f32: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_fract_f32_e32 v6, v0 +; GFX12-NEXT: v_cmp_class_f32_e64 s0, v0, 0x204 +; GFX12-NEXT: v_fract_f32_e32 v7, v1 +; GFX12-NEXT: v_floor_f32_e32 v4, v0 +; GFX12-NEXT: v_floor_f32_e32 v5, v1 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_4) +; GFX12-NEXT: v_cndmask_b32_e64 v0, v6, 0, s0 +; GFX12-NEXT: v_cmp_class_f32_e64 s0, v1, 0x204 +; GFX12-NEXT: global_store_b64 v[2:3], v[4:5], off +; GFX12-NEXT: v_cndmask_b32_e64 v1, v7, 0, s0 +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call <2 x float> @llvm.floor.v2f32(<2 x float> %x) %sub = fsub <2 x float> %x, %floor @@ -1881,6 +2236,21 @@ define double @safe_math_fract_f64(double %x, ptr addrspace(1) nocapture writeon ; GFX11-NEXT: v_dual_cndmask_b32 v0, 0, v4 :: v_dual_cndmask_b32 v1, 0, v5 ; GFX11-NEXT: global_store_b64 v[2:3], v[6:7], off ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: safe_math_fract_f64: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_fract_f64_e32 v[4:5], v[0:1] +; GFX12-NEXT: v_cmp_neq_f64_e64 vcc_lo, 0x7ff00000, |v[0:1]| +; GFX12-NEXT: v_floor_f64_e32 v[6:7], v[0:1] +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_3) +; GFX12-NEXT: v_dual_cndmask_b32 v0, 0, v4 :: v_dual_cndmask_b32 v1, 0, v5 +; GFX12-NEXT: global_store_b64 v[2:3], v[6:7], off +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call double @llvm.floor.f64(double %x) %sub = fsub double %x, %floor @@ -2002,6 +2372,21 @@ define half @safe_math_fract_f16(half %x, ptr addrspace(1) nocapture writeonly % ; GFX11-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo ; GFX11-NEXT: global_store_b16 v[1:2], v4, off ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: safe_math_fract_f16: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_fract_f16_e32 v3, v0 +; GFX12-NEXT: v_cmp_neq_f16_e64 vcc_lo, 0x7c00, |v0| +; GFX12-NEXT: v_floor_f16_e32 v4, v0 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_3) +; GFX12-NEXT: v_cndmask_b32_e32 v0, 0, v3, vcc_lo +; GFX12-NEXT: global_store_b16 v[1:2], v4, off +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call half @llvm.floor.f16(half %x) %sub = fsub half %x, %floor @@ -2168,6 +2553,29 @@ define <2 x half> @safe_math_fract_v2f16(<2 x half> %x, ptr addrspace(1) nocaptu ; GFX11-NEXT: global_store_b32 v[1:2], v4, off ; GFX11-NEXT: v_pack_b32_f16 v0, v0, v3 ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: safe_math_fract_v2f16: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_lshrrev_b32_e32 v3, 16, v0 +; GFX12-NEXT: v_fract_f16_e32 v6, v0 +; GFX12-NEXT: v_floor_f16_e32 v5, v0 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(SKIP_2) | instid1(VALU_DEP_2) +; GFX12-NEXT: v_fract_f16_e32 v4, v3 +; GFX12-NEXT: v_cmp_class_f16_e64 s0, v3, 0x204 +; GFX12-NEXT: v_floor_f16_e32 v7, v3 +; GFX12-NEXT: v_cndmask_b32_e64 v3, v4, 0, s0 +; GFX12-NEXT: v_cmp_class_f16_e64 s0, v0, 0x204 +; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2) +; GFX12-NEXT: v_pack_b32_f16 v4, v5, v7 +; GFX12-NEXT: v_cndmask_b32_e64 v0, v6, 0, s0 +; GFX12-NEXT: global_store_b32 v[1:2], v4, off +; GFX12-NEXT: v_pack_b32_f16 v0, v0, v3 +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call <2 x half> @llvm.floor.v2f16(<2 x half> %x) %sub = fsub <2 x half> %x, %floor @@ -2311,6 +2719,26 @@ define <2 x double> @safe_math_fract_v2f64(<2 x double> %x, ptr addrspace(1) noc ; GFX11-NEXT: v_cndmask_b32_e64 v3, v13, 0, s1 ; GFX11-NEXT: global_store_b128 v[4:5], v[6:9], off ; GFX11-NEXT: s_setpc_b64 s[30:31] +; +; GFX12-LABEL: safe_math_fract_v2f64: +; GFX12: ; %bb.0: ; %entry +; GFX12-NEXT: s_wait_loadcnt_dscnt 0x0 +; GFX12-NEXT: s_wait_expcnt 0x0 +; GFX12-NEXT: s_wait_samplecnt 0x0 +; GFX12-NEXT: s_wait_bvhcnt 0x0 +; GFX12-NEXT: s_wait_kmcnt 0x0 +; GFX12-NEXT: v_fract_f64_e32 v[10:11], v[0:1] +; GFX12-NEXT: v_cmp_class_f64_e64 s0, v[0:1], 0x204 +; GFX12-NEXT: v_fract_f64_e32 v[12:13], v[2:3] +; GFX12-NEXT: v_cmp_class_f64_e64 s1, v[2:3], 0x204 +; GFX12-NEXT: v_floor_f64_e32 v[8:9], v[2:3] +; GFX12-NEXT: v_floor_f64_e32 v[6:7], v[0:1] +; GFX12-NEXT: v_cndmask_b32_e64 v0, v10, 0, s0 +; GFX12-NEXT: v_cndmask_b32_e64 v1, v11, 0, s0 +; GFX12-NEXT: v_cndmask_b32_e64 v2, v12, 0, s1 +; GFX12-NEXT: v_cndmask_b32_e64 v3, v13, 0, s1 +; GFX12-NEXT: global_store_b128 v[4:5], v[6:9], off +; GFX12-NEXT: s_setpc_b64 s[30:31] entry: %floor = tail call <2 x double> @llvm.floor.v2f64(<2 x double> %x) %sub = fsub <2 x double> %x, %floor |