aboutsummaryrefslogtreecommitdiff
path: root/llvm/test/CodeGen/AMDGPU/reassoc-mul-add-1-to-mad.ll
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU/reassoc-mul-add-1-to-mad.ll')
-rw-r--r--llvm/test/CodeGen/AMDGPU/reassoc-mul-add-1-to-mad.ll834
1 files changed, 834 insertions, 0 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/reassoc-mul-add-1-to-mad.ll b/llvm/test/CodeGen/AMDGPU/reassoc-mul-add-1-to-mad.ll
index d89e572..25609e8 100644
--- a/llvm/test/CodeGen/AMDGPU/reassoc-mul-add-1-to-mad.ll
+++ b/llvm/test/CodeGen/AMDGPU/reassoc-mul-add-1-to-mad.ll
@@ -5,6 +5,7 @@
; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx900 < %s | FileCheck -check-prefixes=GFX9,GFX900 %s
; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx90a < %s | FileCheck -check-prefixes=GFX9,GFX90A %s
; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx1030 < %s | FileCheck -check-prefixes=GFX10 %s
+; RUN: llc -mtriple=amdgcn-amd-amdpal -mcpu=gfx1250 < %s | FileCheck -check-prefixes=GFX1250 %s
; We want to undo these canonicalizations to enable mad matching:
; (x * y) + x --> x * (y + 1)
@@ -36,6 +37,13 @@ define i32 @v_mul_add_1_i32(i32 %x, i32 %y) {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_mad_u64_u32 v[0:1], null, v0, v1, v[0:1]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_1_i32:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_u32 v0, v0, v1, v0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add i32 %y, 1
%mul = mul i32 %x, %add
ret i32 %mul
@@ -67,6 +75,13 @@ define i32 @v_mul_add_1_i32_commute(i32 %x, i32 %y) {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_mad_u64_u32 v[0:1], null, v0, v1, v[0:1]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_1_i32_commute:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_u32 v0, v0, v1, v0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add i32 %y, 1
%mul = mul i32 %add, %x
ret i32 %mul
@@ -98,6 +113,13 @@ define i32 @v_mul_add_x_i32(i32 %x, i32 %y) {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_mad_u64_u32 v[0:1], null, v0, v1, v[0:1]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_x_i32:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_u32 v0, v0, v1, v0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul i32 %x, %y
%add = add i32 %x, %mul
ret i32 %add
@@ -131,6 +153,15 @@ define i32 @v_mul_sub_1_i32(i32 %x, i32 %y) {
; GFX10-NEXT: v_add_nc_u32_e32 v1, -1, v1
; GFX10-NEXT: v_mul_lo_u32 v0, v0, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_1_i32:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_add_nc_u32_e32 v1, -1, v1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_mul_lo_u32 v0, v0, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%sub = sub i32 %y, 1
%mul = mul i32 %x, %sub
ret i32 %mul
@@ -164,6 +195,15 @@ define i32 @v_mul_sub_1_i32_commute(i32 %x, i32 %y) {
; GFX10-NEXT: v_add_nc_u32_e32 v1, -1, v1
; GFX10-NEXT: v_mul_lo_u32 v0, v1, v0
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_1_i32_commute:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_add_nc_u32_e32 v1, -1, v1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_mul_lo_u32 v0, v1, v0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%sub = sub i32 %y, 1
%mul = mul i32 %sub, %x
ret i32 %mul
@@ -197,6 +237,15 @@ define i32 @v_mul_sub_x_i32(i32 %x, i32 %y) {
; GFX10-NEXT: v_mul_lo_u32 v1, v0, v1
; GFX10-NEXT: v_sub_nc_u32_e32 v0, v1, v0
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_x_i32:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mul_lo_u32 v1, v0, v1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_sub_nc_u32_e32 v0, v1, v0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul i32 %x, %y
%sub = sub i32 %mul, %x
ret i32 %sub
@@ -230,6 +279,15 @@ define i32 @v_mul_add_2_i32(i32 %x, i32 %y) {
; GFX10-NEXT: v_add_nc_u32_e32 v1, 2, v1
; GFX10-NEXT: v_mul_lo_u32 v0, v0, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_2_i32:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_add_nc_u32_e32 v1, 2, v1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_mul_lo_u32 v0, v0, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add i32 %y, 2
%mul = mul i32 %x, %add
ret i32 %mul
@@ -263,6 +321,15 @@ define i32 @v_mul_sub_2_i32(i32 %x, i32 %y) {
; GFX10-NEXT: v_add_nc_u32_e32 v1, -2, v1
; GFX10-NEXT: v_mul_lo_u32 v0, v0, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_2_i32:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_add_nc_u32_e32 v1, -2, v1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_mul_lo_u32 v0, v0, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%sub = sub i32 %y, 2
%mul = mul i32 %x, %sub
ret i32 %mul
@@ -296,6 +363,15 @@ define i32 @v_mul_add_65_i32(i32 %x, i32 %y) {
; GFX10-NEXT: v_add_nc_u32_e32 v1, 0x41, v1
; GFX10-NEXT: v_mul_lo_u32 v0, v0, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_65_i32:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_add_nc_u32_e32 v1, 0x41, v1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_mul_lo_u32 v0, v0, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add i32 %y, 65
%mul = mul i32 %x, %add
ret i32 %mul
@@ -329,6 +405,15 @@ define i32 @v_mul_sub_65_i32(i32 %x, i32 %y) {
; GFX10-NEXT: v_add_nc_u32_e32 v1, 0xffffffbf, v1
; GFX10-NEXT: v_mul_lo_u32 v0, v0, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_65_i32:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_add_nc_u32_e32 v1, 0xffffffbf, v1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_mul_lo_u32 v0, v0, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%sub = sub i32 %y, 65
%mul = mul i32 %x, %sub
ret i32 %mul
@@ -362,6 +447,15 @@ define i24 @v_mul_add_1_i24_zext(i24 zeroext %x, i24 zeroext %y) {
; GFX10-NEXT: v_add_nc_u32_e32 v1, 1, v1
; GFX10-NEXT: v_mul_u32_u24_e32 v0, v0, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_1_i24_zext:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_add_nc_u32_e32 v1, 1, v1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_mul_u32_u24_e32 v0, v0, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add i24 %y, 1
%mul = mul i24 %x, %add
ret i24 %mul
@@ -395,6 +489,15 @@ define i24 @v_mul_sub_1_i24_zext(i24 zeroext %x, i24 zeroext %y) {
; GFX10-NEXT: v_add_nc_u32_e32 v1, -1, v1
; GFX10-NEXT: v_mul_u32_u24_e32 v0, v0, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_1_i24_zext:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_add_nc_u32_e32 v1, -1, v1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_mul_u32_u24_e32 v0, v0, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%sub = sub i24 %y, 1
%mul = mul i24 %x, %sub
ret i24 %mul
@@ -424,6 +527,13 @@ define i24 @v_add_mul_i24_zext_1(i24 zeroext %x, i24 zeroext %y) {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_mad_u32_u24 v0, v0, v1, v0
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_add_mul_i24_zext_1:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_u32_u24 v0, v0, v1, v0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul i24 %x, %y
%add = add i24 %mul, %x
ret i24 %add
@@ -457,6 +567,15 @@ define i24 @v_mul_add_1_i24_sext(i24 signext %x, i24 signext %y) {
; GFX10-NEXT: v_add_nc_u32_e32 v1, 1, v1
; GFX10-NEXT: v_mul_u32_u24_e32 v0, v0, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_1_i24_sext:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_add_nc_u32_e32 v1, 1, v1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_mul_u32_u24_e32 v0, v0, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add i24 %y, 1
%mul = mul i24 %x, %add
ret i24 %mul
@@ -486,6 +605,13 @@ define i24 @v_add_mul_i24_sext_1(i24 signext %x, i24 signext %y) {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_mad_u32_u24 v0, v0, v1, v0
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_add_mul_i24_sext_1:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_u32_u24 v0, v0, v1, v0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul i24 %x, %y
%add = add i24 %mul, %x
ret i24 %add
@@ -519,6 +645,15 @@ define i24 @v_mul_sub_1_i24_sext(i24 signext %x, i24 signext %y) {
; GFX10-NEXT: v_add_nc_u32_e32 v1, -1, v1
; GFX10-NEXT: v_mul_u32_u24_e32 v0, v0, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_1_i24_sext:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_add_nc_u32_e32 v1, -1, v1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_mul_u32_u24_e32 v0, v0, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%sub = sub i24 %y, 1
%mul = mul i24 %x, %sub
ret i24 %mul
@@ -550,6 +685,13 @@ define i25 @v_mul_add_1_i25_zext(i25 zeroext %x, i25 zeroext %y) {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_mad_u64_u32 v[0:1], null, v0, v1, v[0:1]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_1_i25_zext:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_u32 v0, v0, v1, v0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add i25 %y, 1
%mul = mul i25 %x, %add
ret i25 %mul
@@ -583,6 +725,15 @@ define i25 @v_mul_sub_1_i25_zext(i25 zeroext %x, i25 zeroext %y) {
; GFX10-NEXT: v_add_nc_u32_e32 v1, 0x1ffffff, v1
; GFX10-NEXT: v_mul_lo_u32 v0, v0, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_1_i25_zext:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_add_nc_u32_e32 v1, 0x1ffffff, v1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_mul_lo_u32 v0, v0, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%sub = sub i25 %y, 1
%mul = mul i25 %x, %sub
ret i25 %mul
@@ -614,6 +765,13 @@ define i25 @v_mul_add_1_i25_sext(i25 signext %x, i25 signext %y) {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_mad_u64_u32 v[0:1], null, v0, v1, v[0:1]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_1_i25_sext:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_u32 v0, v0, v1, v0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add i25 %y, 1
%mul = mul i25 %x, %add
ret i25 %mul
@@ -647,6 +805,15 @@ define i25 @v_mul_sub_1_i25_sext(i25 signext %x, i25 signext %y) {
; GFX10-NEXT: v_add_nc_u32_e32 v1, 0x1ffffff, v1
; GFX10-NEXT: v_mul_lo_u32 v0, v0, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_1_i25_sext:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_add_nc_u32_e32 v1, 0x1ffffff, v1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_mul_lo_u32 v0, v0, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%sub = sub i25 %y, 1
%mul = mul i25 %x, %sub
ret i25 %mul
@@ -679,6 +846,13 @@ define i16 @v_mul_add_1_i16(i16 %x, i16 %y) {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_mad_u16 v0, v0, v1, v0
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_1_i16:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_u16 v0, v0, v1, v0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add i16 %y, 1
%mul = mul i16 %x, %add
ret i16 %mul
@@ -713,6 +887,15 @@ define i32 @v_mul_add_1_i16_zext_result(i16 %x, i16 %y) {
; GFX10-NEXT: v_mad_u16 v0, v0, v1, v0
; GFX10-NEXT: v_and_b32_e32 v0, 0xffff, v0
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_1_i16_zext_result:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_u16 v0, v0, v1, v0
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_and_b32_e32 v0, 0xffff, v0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add i16 %y, 1
%mul = mul i16 %x, %add
%zext = zext i16 %mul to i32
@@ -746,6 +929,13 @@ define i16 @v_mul_add_1_i16_commute(i16 %x, i16 %y) {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_mad_u16 v0, v0, v1, v0
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_1_i16_commute:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_u16 v0, v0, v1, v0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add i16 %y, 1
%mul = mul i16 %add, %x
ret i16 %mul
@@ -777,6 +967,13 @@ define i16 @v_mul_add_x_i16(i16 %x, i16 %y) {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_mad_u16 v0, v0, v1, v0
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_x_i16:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_u16 v0, v0, v1, v0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul i16 %x, %y
%add = add i16 %x, %mul
ret i16 %add
@@ -812,6 +1009,15 @@ define i16 @v_mul_sub_1_i16(i16 %x, i16 %y) {
; GFX10-NEXT: v_add_nc_u16 v1, v1, -1
; GFX10-NEXT: v_mul_lo_u16 v0, v0, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_1_i16:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_add_nc_u16 v1, v1, -1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_mul_lo_u16 v0, v0, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%sub = sub i16 %y, 1
%mul = mul i16 %x, %sub
ret i16 %mul
@@ -847,6 +1053,15 @@ define i16 @v_mul_sub_1_i16_commute(i16 %x, i16 %y) {
; GFX10-NEXT: v_add_nc_u16 v1, v1, -1
; GFX10-NEXT: v_mul_lo_u16 v0, v1, v0
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_1_i16_commute:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_add_nc_u16 v1, v1, -1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_mul_lo_u16 v0, v1, v0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%sub = sub i16 %y, 1
%mul = mul i16 %sub, %x
ret i16 %mul
@@ -882,6 +1097,15 @@ define i16 @v_mul_sub_x_i16(i16 %x, i16 %y) {
; GFX10-NEXT: v_mul_lo_u16 v1, v0, v1
; GFX10-NEXT: v_sub_nc_u16 v0, v1, v0
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_x_i16:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mul_lo_u16 v1, v0, v1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_sub_nc_u16 v0, v1, v0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul i16 %x, %y
%sub = sub i16 %mul, %x
ret i16 %sub
@@ -917,6 +1141,15 @@ define i16 @v_mul_add_2_i16(i16 %x, i16 %y) {
; GFX10-NEXT: v_add_nc_u16 v1, v1, 2
; GFX10-NEXT: v_mul_lo_u16 v0, v0, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_2_i16:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_add_nc_u16 v1, v1, 2
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_mul_lo_u16 v0, v0, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add i16 %y, 2
%mul = mul i16 %x, %add
ret i16 %mul
@@ -952,6 +1185,15 @@ define i16 @v_mul_sub_2_i16(i16 %x, i16 %y) {
; GFX10-NEXT: v_add_nc_u16 v1, v1, -2
; GFX10-NEXT: v_mul_lo_u16 v0, v0, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_2_i16:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_add_nc_u16 v1, v1, -2
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_mul_lo_u16 v0, v0, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%sub = sub i16 %y, 2
%mul = mul i16 %x, %sub
ret i16 %mul
@@ -1012,6 +1254,18 @@ define i64 @v_mul_add_1_i64(i64 %x, i64 %y) {
; GFX10-NEXT: v_add3_u32 v1, v1, v5, v0
; GFX10-NEXT: v_mov_b32_e32 v0, v4
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_1_i64:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_nc_u64_u32 v[4:5], v0, v2, v[0:1]
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT: v_mad_u32 v1, v1, v2, v5
+; GFX1250-NEXT: v_mad_u32 v1, v0, v3, v1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3)
+; GFX1250-NEXT: v_mov_b32_e32 v0, v4
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add i64 %y, 1
%mul = mul i64 %x, %add
ret i64 %mul
@@ -1072,6 +1326,18 @@ define i64 @v_mul_add_1_i64_commute(i64 %x, i64 %y) {
; GFX10-NEXT: v_add3_u32 v1, v1, v5, v0
; GFX10-NEXT: v_mov_b32_e32 v0, v4
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_1_i64_commute:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_nc_u64_u32 v[4:5], v0, v2, v[0:1]
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT: v_mad_u32 v1, v1, v2, v5
+; GFX1250-NEXT: v_mad_u32 v1, v0, v3, v1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3)
+; GFX1250-NEXT: v_mov_b32_e32 v0, v4
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add i64 %y, 1
%mul = mul i64 %add, %x
ret i64 %mul
@@ -1132,6 +1398,18 @@ define i64 @v_mul_add_x_i64(i64 %x, i64 %y) {
; GFX10-NEXT: v_add3_u32 v1, v1, v5, v0
; GFX10-NEXT: v_mov_b32_e32 v0, v4
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_x_i64:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_nc_u64_u32 v[4:5], v0, v2, v[0:1]
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT: v_mad_u32 v1, v1, v2, v5
+; GFX1250-NEXT: v_mad_u32 v1, v0, v3, v1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_3)
+; GFX1250-NEXT: v_mov_b32_e32 v0, v4
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul i64 %x, %y
%add = add i64 %x, %mul
ret i64 %add
@@ -1198,6 +1476,15 @@ define i64 @v_mul_sub_1_i64(i64 %x, i64 %y) {
; GFX10-NEXT: v_mad_u64_u32 v[0:1], null, v0, v2, 0
; GFX10-NEXT: v_add3_u32 v1, v1, v3, v4
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_1_i64:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_add_nc_u64_e32 v[2:3], -1, v[2:3]
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_mul_u64_e32 v[0:1], v[0:1], v[2:3]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%sub = sub i64 %y, 1
%mul = mul i64 %x, %sub
ret i64 %mul
@@ -1264,6 +1551,15 @@ define i64 @v_mul_sub_1_i64_commute(i64 %x, i64 %y) {
; GFX10-NEXT: v_mad_u64_u32 v[0:1], null, v2, v0, 0
; GFX10-NEXT: v_add3_u32 v1, v1, v4, v3
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_1_i64_commute:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_add_nc_u64_e32 v[2:3], -1, v[2:3]
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_mul_u64_e32 v[0:1], v[2:3], v[0:1]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%sub = sub i64 %y, 1
%mul = mul i64 %sub, %x
ret i64 %mul
@@ -1328,6 +1624,15 @@ define i64 @v_mul_sub_x_i64(i64 %x, i64 %y) {
; GFX10-NEXT: v_sub_co_u32 v0, vcc_lo, v2, v0
; GFX10-NEXT: v_sub_co_ci_u32_e64 v1, null, v3, v1, vcc_lo
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_x_i64:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mul_u64_e32 v[2:3], v[0:1], v[2:3]
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_sub_nc_u64_e32 v[0:1], v[2:3], v[0:1]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul i64 %x, %y
%sub = sub i64 %mul, %x
ret i64 %sub
@@ -1394,6 +1699,15 @@ define i64 @v_mul_add_2_i64(i64 %x, i64 %y) {
; GFX10-NEXT: v_mad_u64_u32 v[0:1], null, v0, v2, 0
; GFX10-NEXT: v_add3_u32 v1, v1, v3, v4
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_2_i64:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_add_nc_u64_e32 v[2:3], 2, v[2:3]
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_mul_u64_e32 v[0:1], v[0:1], v[2:3]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add i64 %y, 2
%mul = mul i64 %x, %add
ret i64 %mul
@@ -1460,6 +1774,15 @@ define i64 @v_mul_sub_2_i64(i64 %x, i64 %y) {
; GFX10-NEXT: v_mad_u64_u32 v[0:1], null, v0, v2, 0
; GFX10-NEXT: v_add3_u32 v1, v1, v3, v4
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_2_i64:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_add_nc_u64_e32 v[2:3], -2, v[2:3]
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_mul_u64_e32 v[0:1], v[0:1], v[2:3]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%sub = sub i64 %y, 2
%mul = mul i64 %x, %sub
ret i64 %mul
@@ -1508,6 +1831,14 @@ define <2 x i32> @v_mul_add_1_i32_multiple(i32 %x, i32 %y, i32 %z) {
; GFX10-NEXT: v_mad_u64_u32 v[0:1], null, v0, v3, v[0:1]
; GFX10-NEXT: v_mad_u64_u32 v[1:2], null, v2, v3, v[2:3]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_1_i32_multiple:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_u32 v0, v0, v1, v0
+; GFX1250-NEXT: v_mad_u32 v1, v2, v1, v2
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add i32 %y, 1
%mul0 = mul i32 %x, %add
%mul1 = mul i32 %z, %add
@@ -1544,6 +1875,15 @@ define <2 x i32> @v_mul_add_1_i32_other_use(i32 %x, i32 %y, i32 %z) {
; GFX10-NEXT: v_add_nc_u32_e32 v1, 1, v1
; GFX10-NEXT: v_mul_lo_u32 v0, v0, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_1_i32_other_use:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_add_nc_u32_e32 v1, 1, v1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_mul_lo_u32 v0, v0, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add i32 %y, 1
%mul0 = mul i32 %x, %add
%mul1 = mul i32 %z, %add
@@ -1594,6 +1934,19 @@ define i32 @v_mul_add_1_i32_chain(i32 %arg0, i32 %arg1, i32 %arg2) {
; GFX10-NEXT: v_mul_lo_u32 v0, v2, v0
; GFX10-NEXT: v_mad_u64_u32 v[0:1], null, v0, v1, v[0:1]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_1_i32_chain:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_add_nc_u32_e32 v2, 1, v0
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT: v_mul_lo_u32 v1, v2, v1
+; GFX1250-NEXT: v_add_nc_u32_e32 v2, v1, v2
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT: v_mul_lo_u32 v0, v2, v0
+; GFX1250-NEXT: v_mad_u32 v0, v0, v1, v0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%i2 = add i32 %arg0, 1
%i3 = mul i32 %i2, %arg1
%i4 = add i32 %i3, %i2
@@ -1640,6 +1993,15 @@ define <2 x i16> @v_mul_add_1_v2i16(<2 x i16> %x, <2 x i16> %y) {
; GFX10-NEXT: v_pk_add_u16 v1, v1, 1 op_sel_hi:[1,0]
; GFX10-NEXT: v_pk_mul_lo_u16 v0, v0, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_1_v2i16:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_pk_add_u16 v1, v1, 1 op_sel_hi:[1,0]
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_pk_mul_lo_u16 v0, v0, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add <2 x i16> %y, <i16 1, i16 1>
%mul = mul <2 x i16> %x, %add
ret <2 x i16> %mul
@@ -1683,6 +2045,15 @@ define <2 x i16> @v_mul_add_1_v2i16_commute(<2 x i16> %x, <2 x i16> %y) {
; GFX10-NEXT: v_pk_add_u16 v1, v1, 1 op_sel_hi:[1,0]
; GFX10-NEXT: v_pk_mul_lo_u16 v0, v1, v0
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_1_v2i16_commute:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_pk_add_u16 v1, v1, 1 op_sel_hi:[1,0]
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_pk_mul_lo_u16 v0, v1, v0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add <2 x i16> %y, <i16 1, i16 1>
%mul = mul <2 x i16> %add, %x
ret <2 x i16> %mul
@@ -1726,6 +2097,13 @@ define <2 x i16> @v_mul_add_x_v2i16(<2 x i16> %x, <2 x i16> %y) {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_pk_mad_u16 v0, v0, v1, v0
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_x_v2i16:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_pk_mad_u16 v0, v0, v1, v0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul <2 x i16> %x, %y
%add = add <2 x i16> %x, %mul
ret <2 x i16> %add
@@ -1769,6 +2147,15 @@ define <2 x i16> @v_mul_sub_1_v2i16(<2 x i16> %x, <2 x i16> %y) {
; GFX10-NEXT: v_pk_sub_i16 v1, v1, 1 op_sel_hi:[1,0]
; GFX10-NEXT: v_pk_mul_lo_u16 v0, v0, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_1_v2i16:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_pk_sub_i16 v1, v1, 1 op_sel_hi:[1,0]
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_pk_mul_lo_u16 v0, v0, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%sub = sub <2 x i16> %y, <i16 1, i16 1>
%mul = mul <2 x i16> %x, %sub
ret <2 x i16> %mul
@@ -1812,6 +2199,15 @@ define <2 x i16> @v_mul_sub_1_v2i16_commute(<2 x i16> %x, <2 x i16> %y) {
; GFX10-NEXT: v_pk_sub_i16 v1, v1, 1 op_sel_hi:[1,0]
; GFX10-NEXT: v_pk_mul_lo_u16 v0, v1, v0
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_1_v2i16_commute:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_pk_sub_i16 v1, v1, 1 op_sel_hi:[1,0]
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_pk_mul_lo_u16 v0, v1, v0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%sub = sub <2 x i16> %y, <i16 1, i16 1>
%mul = mul <2 x i16> %sub, %x
ret <2 x i16> %mul
@@ -1858,6 +2254,15 @@ define <2 x i16> @v_mul_sub_x_v2i16(<2 x i16> %x, <2 x i16> %y) {
; GFX10-NEXT: v_pk_mul_lo_u16 v1, v0, v1
; GFX10-NEXT: v_pk_sub_i16 v0, v1, v0
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_x_v2i16:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_pk_mul_lo_u16 v1, v0, v1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_pk_sub_i16 v0, v1, v0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul <2 x i16> %x, %y
%sub = sub <2 x i16> %mul, %x
ret <2 x i16> %sub
@@ -1901,6 +2306,15 @@ define <2 x i16> @v_mul_add_2_v2i16(<2 x i16> %x, <2 x i16> %y) {
; GFX10-NEXT: v_pk_add_u16 v1, v1, 2 op_sel_hi:[1,0]
; GFX10-NEXT: v_pk_mul_lo_u16 v0, v0, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_2_v2i16:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_pk_add_u16 v1, v1, 2 op_sel_hi:[1,0]
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_pk_mul_lo_u16 v0, v0, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add <2 x i16> %y, <i16 2, i16 2>
%mul = mul <2 x i16> %x, %add
ret <2 x i16> %mul
@@ -1944,6 +2358,15 @@ define <2 x i16> @v_mul_sub_2_v2i16(<2 x i16> %x, <2 x i16> %y) {
; GFX10-NEXT: v_pk_sub_i16 v1, v1, 2 op_sel_hi:[1,0]
; GFX10-NEXT: v_pk_mul_lo_u16 v0, v0, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_2_v2i16:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_pk_sub_i16 v1, v1, 2 op_sel_hi:[1,0]
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_pk_mul_lo_u16 v0, v0, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%sub = sub <2 x i16> %y, <i16 2, i16 2>
%mul = mul <2 x i16> %x, %sub
ret <2 x i16> %mul
@@ -1992,6 +2415,14 @@ define <2 x i32> @v_mul_add_1_v2i32(<2 x i32> %x, <2 x i32> %y) {
; GFX10-NEXT: v_mad_u64_u32 v[1:2], null, v1, v3, v[1:2]
; GFX10-NEXT: v_mov_b32_e32 v0, v4
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_1_v2i32:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_u32 v0, v0, v2, v0
+; GFX1250-NEXT: v_mad_u32 v1, v1, v3, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add <2 x i32> %y, <i32 1, i32 1>
%mul = mul <2 x i32> %x, %add
ret <2 x i32> %mul
@@ -2040,6 +2471,14 @@ define <2 x i32> @v_mul_add_1_v2i32_commute(<2 x i32> %x, <2 x i32> %y) {
; GFX10-NEXT: v_mad_u64_u32 v[1:2], null, v1, v3, v[1:2]
; GFX10-NEXT: v_mov_b32_e32 v0, v4
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_1_v2i32_commute:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_u32 v0, v0, v2, v0
+; GFX1250-NEXT: v_mad_u32 v1, v1, v3, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add <2 x i32> %y, <i32 1, i32 1>
%mul = mul <2 x i32> %add, %x
ret <2 x i32> %mul
@@ -2088,6 +2527,14 @@ define <2 x i32> @v_mul_add_x_v2i32(<2 x i32> %x, <2 x i32> %y) {
; GFX10-NEXT: v_mad_u64_u32 v[1:2], null, v1, v3, v[1:2]
; GFX10-NEXT: v_mov_b32_e32 v0, v4
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_x_v2i32:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_u32 v0, v0, v2, v0
+; GFX1250-NEXT: v_mad_u32 v1, v1, v3, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul <2 x i32> %x, %y
%add = add <2 x i32> %x, %mul
ret <2 x i32> %add
@@ -2129,6 +2576,16 @@ define <2 x i32> @v_mul_sub_1_v2i32(<2 x i32> %x, <2 x i32> %y) {
; GFX10-NEXT: v_mul_lo_u32 v0, v0, v2
; GFX10-NEXT: v_mul_lo_u32 v1, v1, v3
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_1_v2i32:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_dual_add_nc_u32 v2, -1, v2 :: v_dual_add_nc_u32 v3, -1, v3
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1250-NEXT: v_mul_lo_u32 v0, v0, v2
+; GFX1250-NEXT: v_mul_lo_u32 v1, v1, v3
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%sub = sub <2 x i32> %y, <i32 1, i32 1>
%mul = mul <2 x i32> %x, %sub
ret <2 x i32> %mul
@@ -2170,6 +2627,16 @@ define <2 x i32> @v_mul_sub_1_v2i32_commute(<2 x i32> %x, <2 x i32> %y) {
; GFX10-NEXT: v_mul_lo_u32 v0, v2, v0
; GFX10-NEXT: v_mul_lo_u32 v1, v3, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_1_v2i32_commute:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_dual_add_nc_u32 v2, -1, v2 :: v_dual_add_nc_u32 v3, -1, v3
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1250-NEXT: v_mul_lo_u32 v0, v2, v0
+; GFX1250-NEXT: v_mul_lo_u32 v1, v3, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%sub = sub <2 x i32> %y, <i32 1, i32 1>
%mul = mul <2 x i32> %sub, %x
ret <2 x i32> %mul
@@ -2220,6 +2687,16 @@ define <2 x i32> @v_mul_sub_x_v2i32(<2 x i32> %x, <2 x i32> %y) {
; GFX10-NEXT: v_sub_nc_u32_e32 v0, v2, v0
; GFX10-NEXT: v_sub_nc_u32_e32 v1, v3, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_x_v2i32:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mul_lo_u32 v2, v0, v2
+; GFX1250-NEXT: v_mul_lo_u32 v3, v1, v3
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_dual_sub_nc_u32 v0, v2, v0 :: v_dual_sub_nc_u32 v1, v3, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul <2 x i32> %x, %y
%sub = sub <2 x i32> %mul, %x
ret <2 x i32> %sub
@@ -2261,6 +2738,16 @@ define <2 x i32> @v_mul_add_2_v2i32(<2 x i32> %x, <2 x i32> %y) {
; GFX10-NEXT: v_mul_lo_u32 v0, v0, v2
; GFX10-NEXT: v_mul_lo_u32 v1, v1, v3
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_2_v2i32:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_dual_add_nc_u32 v2, 2, v2 :: v_dual_add_nc_u32 v3, 2, v3
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1250-NEXT: v_mul_lo_u32 v0, v0, v2
+; GFX1250-NEXT: v_mul_lo_u32 v1, v1, v3
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add <2 x i32> %y, <i32 2, i32 2>
%mul = mul <2 x i32> %x, %add
ret <2 x i32> %mul
@@ -2302,6 +2789,16 @@ define <2 x i32> @v_mul_sub_2_v2i32(<2 x i32> %x, <2 x i32> %y) {
; GFX10-NEXT: v_mul_lo_u32 v0, v0, v2
; GFX10-NEXT: v_mul_lo_u32 v1, v1, v3
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_2_v2i32:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_dual_add_nc_u32 v2, -2, v2 :: v_dual_add_nc_u32 v3, -2, v3
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1250-NEXT: v_mul_lo_u32 v0, v0, v2
+; GFX1250-NEXT: v_mul_lo_u32 v1, v1, v3
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%sub = sub <2 x i32> %y, <i32 2, i32 2>
%mul = mul <2 x i32> %x, %sub
ret <2 x i32> %mul
@@ -2343,6 +2840,16 @@ define <2 x i24> @v_mul_add_1_v2i24(<2 x i24> %x, <2 x i24> %y) {
; GFX10-NEXT: v_mul_u32_u24_e32 v0, v0, v2
; GFX10-NEXT: v_mul_u32_u24_e32 v1, v1, v3
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_1_v2i24:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_dual_add_nc_u32 v2, 1, v2 :: v_dual_add_nc_u32 v3, 1, v3
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1250-NEXT: v_mul_u32_u24_e32 v0, v0, v2
+; GFX1250-NEXT: v_mul_u32_u24_e32 v1, v1, v3
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add <2 x i24> %y, <i24 1, i24 1>
%mul = mul <2 x i24> %x, %add
ret <2 x i24> %mul
@@ -2384,6 +2891,16 @@ define <2 x i24> @v_mul_add_1_v2i24_commute(<2 x i24> %x, <2 x i24> %y) {
; GFX10-NEXT: v_mul_u32_u24_e32 v0, v2, v0
; GFX10-NEXT: v_mul_u32_u24_e32 v1, v3, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_1_v2i24_commute:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_dual_add_nc_u32 v2, 1, v2 :: v_dual_add_nc_u32 v3, 1, v3
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1250-NEXT: v_mul_u32_u24_e32 v0, v2, v0
+; GFX1250-NEXT: v_mul_u32_u24_e32 v1, v3, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add <2 x i24> %y, <i24 1, i24 1>
%mul = mul <2 x i24> %add, %x
ret <2 x i24> %mul
@@ -2417,6 +2934,14 @@ define <2 x i24> @v_mul_add_x_v2i24(<2 x i24> %x, <2 x i24> %y) {
; GFX10-NEXT: v_mad_u32_u24 v0, v0, v2, v0
; GFX10-NEXT: v_mad_u32_u24 v1, v1, v3, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_x_v2i24:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_u32_u24 v0, v0, v2, v0
+; GFX1250-NEXT: v_mad_u32_u24 v1, v1, v3, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul <2 x i24> %x, %y
%add = add <2 x i24> %x, %mul
ret <2 x i24> %add
@@ -2458,6 +2983,16 @@ define <2 x i24> @v_mul_sub_1_v2i24(<2 x i24> %x, <2 x i24> %y) {
; GFX10-NEXT: v_mul_u32_u24_e32 v0, v0, v2
; GFX10-NEXT: v_mul_u32_u24_e32 v1, v1, v3
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_1_v2i24:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_dual_add_nc_u32 v2, -1, v2 :: v_dual_add_nc_u32 v3, -1, v3
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1250-NEXT: v_mul_u32_u24_e32 v0, v0, v2
+; GFX1250-NEXT: v_mul_u32_u24_e32 v1, v1, v3
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%sub = sub <2 x i24> %y, <i24 1, i24 1>
%mul = mul <2 x i24> %x, %sub
ret <2 x i24> %mul
@@ -2499,6 +3034,16 @@ define <2 x i24> @v_mul_sub_1_v2i24_commute(<2 x i24> %x, <2 x i24> %y) {
; GFX10-NEXT: v_mul_u32_u24_e32 v0, v2, v0
; GFX10-NEXT: v_mul_u32_u24_e32 v1, v3, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_1_v2i24_commute:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_dual_add_nc_u32 v2, -1, v2 :: v_dual_add_nc_u32 v3, -1, v3
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1250-NEXT: v_mul_u32_u24_e32 v0, v2, v0
+; GFX1250-NEXT: v_mul_u32_u24_e32 v1, v3, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%sub = sub <2 x i24> %y, <i24 1, i24 1>
%mul = mul <2 x i24> %sub, %x
ret <2 x i24> %mul
@@ -2540,6 +3085,16 @@ define <2 x i24> @v_mul_sub_x_v2i24(<2 x i24> %x, <2 x i24> %y) {
; GFX10-NEXT: v_sub_nc_u32_e32 v0, v2, v0
; GFX10-NEXT: v_sub_nc_u32_e32 v1, v3, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_x_v2i24:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mul_u32_u24_e32 v2, v0, v2
+; GFX1250-NEXT: v_mul_u32_u24_e32 v3, v1, v3
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_dual_sub_nc_u32 v0, v2, v0 :: v_dual_sub_nc_u32 v1, v3, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul <2 x i24> %x, %y
%sub = sub <2 x i24> %mul, %x
ret <2 x i24> %sub
@@ -2581,6 +3136,16 @@ define <2 x i24> @v_mul_add_2_v2i24(<2 x i24> %x, <2 x i24> %y) {
; GFX10-NEXT: v_mul_u32_u24_e32 v0, v0, v2
; GFX10-NEXT: v_mul_u32_u24_e32 v1, v1, v3
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_2_v2i24:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_dual_add_nc_u32 v2, 2, v2 :: v_dual_add_nc_u32 v3, 2, v3
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1250-NEXT: v_mul_u32_u24_e32 v0, v0, v2
+; GFX1250-NEXT: v_mul_u32_u24_e32 v1, v1, v3
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add <2 x i24> %y, <i24 2, i24 2>
%mul = mul <2 x i24> %x, %add
ret <2 x i24> %mul
@@ -2622,6 +3187,16 @@ define <2 x i24> @v_mul_sub_2_v2i24(<2 x i24> %x, <2 x i24> %y) {
; GFX10-NEXT: v_mul_u32_u24_e32 v0, v0, v2
; GFX10-NEXT: v_mul_u32_u24_e32 v1, v1, v3
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_sub_2_v2i24:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_dual_add_nc_u32 v2, -2, v2 :: v_dual_add_nc_u32 v3, -2, v3
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1250-NEXT: v_mul_u32_u24_e32 v0, v0, v2
+; GFX1250-NEXT: v_mul_u32_u24_e32 v1, v1, v3
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%sub = sub <2 x i24> %y, <i24 2, i24 2>
%mul = mul <2 x i24> %x, %sub
ret <2 x i24> %mul
@@ -2653,6 +3228,13 @@ define i32 @v_mul_9_add_52_i32(i32 %arg) {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_mad_u64_u32 v[0:1], null, v0, 9, 52
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_9_add_52_i32:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_u32 v0, v0, 9, 52
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul i32 %arg, 9
%add = add i32 %mul, 52
ret i32 %add
@@ -2683,6 +3265,13 @@ define i16 @v_mul_9_add_52_i16(i16 %arg) {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_mad_u16 v0, v0, 9, 52
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_9_add_52_i16:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_u16 v0, v0, 9, 52
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul i16 %arg, 9
%add = add i16 %mul, 52
ret i16 %add
@@ -2723,6 +3312,13 @@ define <2 x i16> @v_mul_9_add_52_v2i16(<2 x i16> %arg) {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_pk_mad_u16 v0, v0, 9, 52 op_sel_hi:[1,0,0]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_9_add_52_v2i16:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_pk_mad_u16 v0, v0, 9, 52 op_sel_hi:[1,0,0]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul <2 x i16> %arg, <i16 9, i16 9>
%add = add <2 x i16> %mul, <i16 52, i16 52>
ret <2 x i16> %add
@@ -2781,6 +3377,16 @@ define i64 @v_mul_9_add_52_i64(i64 %arg) {
; GFX10-NEXT: v_mad_u64_u32 v[0:1], null, v0, 9, 52
; GFX10-NEXT: v_mad_u64_u32 v[1:2], null, v2, 9, v[1:2]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_9_add_52_i64:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mov_b32_e32 v2, v1
+; GFX1250-NEXT: v_mad_nc_u64_u32 v[0:1], v0, 9, 52
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_mad_u32 v1, v2, 9, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul i64 %arg, 9
%add = add i64 %mul, 52
ret i64 %add
@@ -2812,6 +3418,13 @@ define i32 @v_mul_5_add_1_i32(i32 %arg) {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_mad_u64_u32 v[0:1], null, v0, 5, 1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_5_add_1_i32:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_u32 v0, v0, 5, 1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul i32 %arg, 5
%add = add i32 %mul, 1
ret i32 %add
@@ -2848,6 +3461,15 @@ define i32 @v_mul_284_add_82_i32(i32 %arg) {
; GFX10-NEXT: s_movk_i32 s4, 0x11c
; GFX10-NEXT: v_mad_u64_u32 v[0:1], null, v0, s4, 0x52
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_284_add_82_i32:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: s_movk_i32 s0, 0x11c
+; GFX1250-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1250-NEXT: v_mad_u32 v0, v0, s0, 0x52
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul i32 %arg, 284
%add = add i32 %mul, 82
ret i32 %add
@@ -2878,6 +3500,13 @@ define i16 @v_mul_5_add_1_i16(i16 %arg) {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_mad_u16 v0, v0, 5, 1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_5_add_1_i16:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_u16 v0, v0, 5, 1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul i16 %arg, 5
%add = add i16 %mul, 1
ret i16 %add
@@ -2915,6 +3544,15 @@ define i16 @v_mul_284_add_82_i16(i16 %arg) {
; GFX10-NEXT: s_movk_i32 s4, 0x11c
; GFX10-NEXT: v_mad_u16 v0, v0, s4, 0x52
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_284_add_82_i16:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: s_movk_i32 s0, 0x11c
+; GFX1250-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1250-NEXT: v_mad_u16 v0, v0, s0, 0x52
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul i16 %arg, 284
%add = add i16 %mul, 82
ret i16 %add
@@ -2955,6 +3593,13 @@ define <2 x i16> @v_mul_5_add_1_v2i16(<2 x i16> %arg) {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_pk_mad_u16 v0, v0, 5, 1 op_sel_hi:[1,0,0]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_5_add_1_v2i16:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_pk_mad_u16 v0, v0, 5, 1 op_sel_hi:[1,0,0]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul <2 x i16> %arg, <i16 5, i16 5>
%add = add <2 x i16> %mul, <i16 1, i16 1>
ret <2 x i16> %add
@@ -3002,6 +3647,15 @@ define <2 x i16> @v_mul_284_add_82_v2i16(<2 x i16> %arg) {
; GFX10-NEXT: s_movk_i32 s4, 0x11c
; GFX10-NEXT: v_pk_mad_u16 v0, v0, s4, 0x52 op_sel_hi:[1,0,0]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_284_add_82_v2i16:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: s_movk_i32 s0, 0x11c
+; GFX1250-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1250-NEXT: v_pk_mad_u16 v0, v0, s0, 0x52 op_sel_hi:[1,0,0]
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul <2 x i16> %arg, <i16 284, i16 284>
%add = add <2 x i16> %mul, <i16 82, i16 82>
ret <2 x i16> %add
@@ -3060,6 +3714,16 @@ define i64 @v_mul_5_add_1_i64(i64 %arg) {
; GFX10-NEXT: v_mad_u64_u32 v[0:1], null, v0, 5, 1
; GFX10-NEXT: v_mad_u64_u32 v[1:2], null, v2, 5, v[1:2]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_5_add_1_i64:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mov_b32_e32 v2, v1
+; GFX1250-NEXT: v_mad_nc_u64_u32 v[0:1], v0, 5, 1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_mad_u32 v1, v2, 5, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul i64 %arg, 5
%add = add i64 %mul, 1
ret i64 %add
@@ -3132,6 +3796,17 @@ define i64 @v_mul_284_add_82_i64(i64 %arg) {
; GFX10-NEXT: v_mad_u64_u32 v[0:1], null, v0, s4, 0x52
; GFX10-NEXT: v_mad_u64_u32 v[1:2], null, 0x11c, v2, v[1:2]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_284_add_82_i64:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: s_movk_i32 s0, 0x11c
+; GFX1250-NEXT: v_mov_b32_e32 v2, v1
+; GFX1250-NEXT: v_mad_nc_u64_u32 v[0:1], v0, s0, 0x52
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_mad_u32 v1, 0x11c, v2, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul i64 %arg, 284
%add = add i64 %mul, 82
ret i64 %add
@@ -3204,6 +3879,17 @@ define i64 @v_mul_934584645_add_8234599_i64(i64 %arg) {
; GFX10-NEXT: v_mad_u64_u32 v[0:1], null, v0, s4, 0x7da667
; GFX10-NEXT: v_mad_u64_u32 v[1:2], null, 0x37b4a145, v2, v[1:2]
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_934584645_add_8234599_i64:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: s_mov_b32 s0, 0x37b4a145
+; GFX1250-NEXT: v_mov_b32_e32 v2, v1
+; GFX1250-NEXT: v_mad_nc_u64_u32 v[0:1], v0, s0, 0x7da667
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX1250-NEXT: v_mad_u32 v1, 0x37b4a145, v2, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%mul = mul i64 %arg, 934584645
%add = add i64 %mul, 8234599
ret i64 %add
@@ -3394,6 +4080,44 @@ define amdgpu_kernel void @compute_mad(ptr addrspace(4) %i18, ptr addrspace(4) %
; GFX10-NEXT: v_add_co_ci_u32_e64 v2, null, s5, v3, vcc_lo
; GFX10-NEXT: global_store_dword v[1:2], v0, off
; GFX10-NEXT: s_endpgm
+;
+; GFX1250-LABEL: compute_mad:
+; GFX1250: ; %bb.0: ; %bb
+; GFX1250-NEXT: s_load_b96 s[8:10], s[4:5], 0x10
+; GFX1250-NEXT: v_and_b32_e32 v0, 0x3ff, v0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: s_add_co_i32 s0, s10, 1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
+; GFX1250-NEXT: v_mul_lo_u32 v1, s0, v0
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_1)
+; GFX1250-NEXT: v_dual_add_nc_u32 v2, s0, v1 :: v_dual_add_nc_u32 v1, 1, v1
+; GFX1250-NEXT: s_load_b128 s[0:3], s[4:5], 0x0
+; GFX1250-NEXT: v_mul_lo_u32 v2, v2, v0
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_1)
+; GFX1250-NEXT: v_mul_lo_u32 v3, v2, v1
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: s_load_b32 s2, s[2:3], 0x4
+; GFX1250-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX1250-NEXT: v_add_nc_u32_e32 v1, v3, v1
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_3) | instid1(VALU_DEP_1)
+; GFX1250-NEXT: v_mul_lo_u32 v1, v1, v2
+; GFX1250-NEXT: v_add_nc_u32_e32 v2, 1, v3
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: s_and_b32 s2, s2, 0xffff
+; GFX1250-NEXT: v_mul_lo_u32 v3, v1, v2
+; GFX1250-NEXT: v_mad_u32 v0, ttmp9, s2, v0
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT: v_add_nc_u32_e32 v2, v3, v2
+; GFX1250-NEXT: v_mul_lo_u32 v2, v2, v1
+; GFX1250-NEXT: v_mov_b32_e32 v1, 0
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_3)
+; GFX1250-NEXT: v_add_nc_u64_e32 v[0:1], s[0:1], v[0:1]
+; GFX1250-NEXT: v_mad_u32 v3, v2, v3, v2
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_2)
+; GFX1250-NEXT: v_lshl_add_u64 v[0:1], v[0:1], 2, s[8:9]
+; GFX1250-NEXT: v_mad_u32 v2, v3, v2, v3
+; GFX1250-NEXT: global_store_b32 v[0:1], v2, off
+; GFX1250-NEXT: s_endpgm
bb:
%i = tail call i32 @llvm.amdgcn.workitem.id.x(), !range !0
%i2 = add i32 %arg1, 1
@@ -3450,6 +4174,13 @@ define amdgpu_ps i32 @s_mul_add_1_i32(i32 inreg %x, i32 inreg %y) {
; GFX10-NEXT: s_add_i32 s1, s1, 1
; GFX10-NEXT: s_mul_i32 s0, s0, s1
; GFX10-NEXT: ; return to shader part epilog
+;
+; GFX1250-LABEL: s_mul_add_1_i32:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_add_co_i32 s1, s1, 1
+; GFX1250-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1250-NEXT: s_mul_i32 s0, s0, s1
+; GFX1250-NEXT: ; return to shader part epilog
%add = add i32 %y, 1
%mul = mul i32 %x, %add
ret i32 %mul
@@ -3479,6 +4210,13 @@ define amdgpu_ps i32 @s_mul_add_1_i32_commute(i32 inreg %x, i32 inreg %y) {
; GFX10-NEXT: s_add_i32 s1, s1, 1
; GFX10-NEXT: s_mul_i32 s0, s1, s0
; GFX10-NEXT: ; return to shader part epilog
+;
+; GFX1250-LABEL: s_mul_add_1_i32_commute:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_add_co_i32 s1, s1, 1
+; GFX1250-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX1250-NEXT: s_mul_i32 s0, s1, s0
+; GFX1250-NEXT: ; return to shader part epilog
%add = add i32 %y, 1
%mul = mul i32 %add, %x
ret i32 %mul
@@ -3511,6 +4249,13 @@ define i8 @v_mul_add_1_i8(i8 %x, i8 %y) {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_mad_u16 v0, v0, v1, v0
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_1_i8:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_u16 v0, v0, v1, v0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add i8 %y, 1
%mul = mul i8 %x, %add
ret i8 %mul
@@ -3543,6 +4288,13 @@ define i8 @v_mul_add_1_i8_commute(i8 %x, i8 %y) {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_mad_u16 v0, v0, v1, v0
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_1_i8_commute:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_u16 v0, v0, v1, v0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add i8 %y, 1
%mul = mul i8 %add, %x
ret i8 %mul
@@ -3574,6 +4326,13 @@ define i8 @v_mul_add_1_i8_zext(i8 zeroext %x, i8 zeroext %y) {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_mad_u16 v0, v0, v1, v0
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_1_i8_zext:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_u16 v0, v0, v1, v0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add i8 %y, 1
%mul = mul i8 %x, %add
ret i8 %mul
@@ -3605,6 +4364,13 @@ define i8 @v_mul_add_1_i8_zext_commute(i8 zeroext %x, i8 zeroext %y) {
; GFX10-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
; GFX10-NEXT: v_mad_u16 v0, v0, v1, v0
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_1_i8_zext_commute:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_u16 v0, v0, v1, v0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add i8 %y, 1
%mul = mul i8 %add, %x
ret i8 %mul
@@ -3656,6 +4422,18 @@ define <2 x i8> @v_mul_add_1_v2i8(<2 x i8> %x, <2 x i8> %y) {
; GFX10-NEXT: v_and_b32_e32 v1, 0xff, v1
; GFX10-NEXT: v_or_b32_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_1_v2i8:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_u16 v1, v1, v3, v1
+; GFX1250-NEXT: v_mad_u16 v0, v0, v2, v0
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1250-NEXT: v_lshlrev_b16 v2, 8, v1
+; GFX1250-NEXT: v_and_b32_e32 v1, 0xff, v1
+; GFX1250-NEXT: v_bitop3_b16 v0, v0, v2, 0xff bitop3:0xec
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add <2 x i8> %y, <i8 1, i8 1>
%mul = mul <2 x i8> %x, %add
ret <2 x i8> %mul
@@ -3707,6 +4485,18 @@ define <2 x i8> @v_mul_add_1_v2i8_commute(<2 x i8> %x, <2 x i8> %y) {
; GFX10-NEXT: v_and_b32_e32 v1, 0xff, v1
; GFX10-NEXT: v_or_b32_sdwa v0, v0, v2 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:BYTE_0 src1_sel:DWORD
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: v_mul_add_1_v2i8_commute:
+; GFX1250: ; %bb.0:
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_mad_u16 v1, v1, v3, v1
+; GFX1250-NEXT: v_mad_u16 v0, v0, v2, v0
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX1250-NEXT: v_lshlrev_b16 v2, 8, v1
+; GFX1250-NEXT: v_and_b32_e32 v1, 0xff, v1
+; GFX1250-NEXT: v_bitop3_b16 v0, v0, v2, 0xff bitop3:0xec
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
%add = add <2 x i8> %y, <i8 1, i8 1>
%mul = mul <2 x i8> %add, %x
ret <2 x i8> %mul
@@ -3749,6 +4539,17 @@ define i64 @mul_u24_with_uneven_operands(i32 %z) {
; GFX10-NEXT: v_mul_u32_u24_e32 v0, v1, v0
; GFX10-NEXT: v_mov_b32_e32 v1, 0
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: mul_u24_with_uneven_operands:
+; GFX1250: ; %bb.0: ; %entry
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT: v_add_nc_u32_e32 v1, 1, v0
+; GFX1250-NEXT: v_mul_u32_u24_e32 v0, v1, v0
+; GFX1250-NEXT: v_mov_b32_e32 v1, 0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
entry:
%c = and i32 %z, 1
%d = add nuw nsw i32 %c, 1
@@ -3792,6 +4593,17 @@ define i64 @mul_u24_with_uneven_operands_swapped(i32 %z) {
; GFX10-NEXT: v_mul_u32_u24_e32 v0, v0, v1
; GFX10-NEXT: v_mov_b32_e32 v1, 0
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: mul_u24_with_uneven_operands_swapped:
+; GFX1250: ; %bb.0: ; %entry
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_and_b32_e32 v0, 1, v0
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT: v_add_nc_u32_e32 v1, 1, v0
+; GFX1250-NEXT: v_mul_u32_u24_e32 v0, v0, v1
+; GFX1250-NEXT: v_mov_b32_e32 v1, 0
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
entry:
%c = and i32 %z, 1
%d = add nuw nsw i32 %c, 1
@@ -3836,6 +4648,17 @@ define i64 @mul_i24_with_uneven_operands(i32 %z) {
; GFX10-NEXT: v_mul_i32_i24_e32 v0, v2, v1
; GFX10-NEXT: v_mul_hi_i32_i24_e32 v1, v2, v1
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: mul_i24_with_uneven_operands:
+; GFX1250: ; %bb.0: ; %entry
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_and_b32_e32 v1, 1, v0
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT: v_add_nc_u32_e32 v2, 1, v1
+; GFX1250-NEXT: v_mul_i32_i24_e32 v0, v2, v1
+; GFX1250-NEXT: v_mul_hi_i32_i24_e32 v1, v2, v1
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
entry:
%c = and i32 %z, 1
%d = add nuw nsw i32 %c, 1
@@ -3879,6 +4702,17 @@ define i64 @mul_i24_with_uneven_operands_swapped(i32 %z) {
; GFX10-NEXT: v_mul_i32_i24_e32 v0, v1, v2
; GFX10-NEXT: v_mul_hi_i32_i24_e32 v1, v1, v2
; GFX10-NEXT: s_setpc_b64 s[30:31]
+;
+; GFX1250-LABEL: mul_i24_with_uneven_operands_swapped:
+; GFX1250: ; %bb.0: ; %entry
+; GFX1250-NEXT: s_wait_loadcnt_dscnt 0x0
+; GFX1250-NEXT: s_wait_kmcnt 0x0
+; GFX1250-NEXT: v_and_b32_e32 v1, 1, v0
+; GFX1250-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
+; GFX1250-NEXT: v_add_nc_u32_e32 v2, 1, v1
+; GFX1250-NEXT: v_mul_i32_i24_e32 v0, v1, v2
+; GFX1250-NEXT: v_mul_hi_i32_i24_e32 v1, v1, v2
+; GFX1250-NEXT: s_set_pc_i64 s[30:31]
entry:
%c = and i32 %z, 1
%d = add nuw nsw i32 %c, 1
generated by cgit v1.1 at 2025-07-31 05:55:49 +0000