aboutsummaryrefslogtreecommitdiff
path: root/llvm/test/CodeGen/AMDGPU/idot4u.ll
diff options
context:
space:
mode:
Diffstat (limited to 'llvm/test/CodeGen/AMDGPU/idot4u.ll')
-rw-r--r--llvm/test/CodeGen/AMDGPU/idot4u.ll13
1 files changed, 5 insertions, 8 deletions
diff --git a/llvm/test/CodeGen/AMDGPU/idot4u.ll b/llvm/test/CodeGen/AMDGPU/idot4u.ll
index 049663a..f80d50b 100644
--- a/llvm/test/CodeGen/AMDGPU/idot4u.ll
+++ b/llvm/test/CodeGen/AMDGPU/idot4u.ll
@@ -2730,18 +2730,15 @@ define amdgpu_kernel void @udot4_acc8_vecMul(ptr addrspace(1) %src1,
; GFX11-DL-TRUE16-NEXT: v_lshrrev_b16 v1.h, 8, v4.l
; GFX11-DL-TRUE16-NEXT: v_mad_u16 v0.l, v3.l, v4.l, v0.l
; GFX11-DL-TRUE16-NEXT: v_mul_lo_u16 v2.l, v2.l, v6.l
-; GFX11-DL-TRUE16-NEXT: v_mov_b16_e32 v6.l, 0
+; GFX11-DL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
; GFX11-DL-TRUE16-NEXT: v_and_b16 v1.l, 0xff, v1.l
; GFX11-DL-TRUE16-NEXT: v_mul_lo_u16 v0.h, v0.h, v1.h
-; GFX11-DL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_4) | instskip(NEXT) | instid1(VALU_DEP_4)
+; GFX11-DL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_2)
; GFX11-DL-TRUE16-NEXT: v_lshlrev_b16 v2.l, 8, v2.l
-; GFX11-DL-TRUE16-NEXT: v_mov_b16_e32 v7.h, v6.l
-; GFX11-DL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_3) | instskip(NEXT) | instid1(VALU_DEP_3)
-; GFX11-DL-TRUE16-NEXT: v_lshlrev_b16 v7.l, 8, v0.h
+; GFX11-DL-TRUE16-NEXT: v_lshlrev_b16 v6.l, 8, v0.h
+; GFX11-DL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(NEXT) | instid1(VALU_DEP_1)
; GFX11-DL-TRUE16-NEXT: v_or_b16 v6.h, v1.l, v2.l
-; GFX11-DL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(NEXT) | instid1(VALU_DEP_1)
-; GFX11-DL-TRUE16-NEXT: v_or_b32_e32 v1, v7, v6
-; GFX11-DL-TRUE16-NEXT: v_lshrrev_b32_e32 v1, 8, v1
+; GFX11-DL-TRUE16-NEXT: v_lshrrev_b32_e32 v1, 8, v6
; GFX11-DL-TRUE16-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_1) | instid1(VALU_DEP_2)
; GFX11-DL-TRUE16-NEXT: v_add_nc_u16 v0.l, v0.l, v1.l
; GFX11-DL-TRUE16-NEXT: v_lshrrev_b32_e32 v1, 8, v2