diff options
Diffstat (limited to 'llvm/test/CodeGen/X86/avx-vbroadcast.ll')
| -rw-r--r-- | llvm/test/CodeGen/X86/avx-vbroadcast.ll | 18 |
1 files changed, 12 insertions, 6 deletions
diff --git a/llvm/test/CodeGen/X86/avx-vbroadcast.ll b/llvm/test/CodeGen/X86/avx-vbroadcast.ll index 77b9c0dc8f1..37b8753097c 100644 --- a/llvm/test/CodeGen/X86/avx-vbroadcast.ll +++ b/llvm/test/CodeGen/X86/avx-vbroadcast.ll @@ -842,12 +842,15 @@ define float @broadcast_lifetime() nounwind { ; X32-NEXT: leal {{[0-9]+}}(%esp), %esi ; X32-NEXT: movl %esi, (%esp) ; X32-NEXT: calll _gfunc -; X32-NEXT: vbroadcastss {{[0-9]+}}(%esp), %xmm0 +; X32-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero ; X32-NEXT: vmovaps %xmm0, {{[0-9]+}}(%esp) ## 16-byte Spill ; X32-NEXT: movl %esi, (%esp) ; X32-NEXT: calll _gfunc -; X32-NEXT: vbroadcastss {{[0-9]+}}(%esp), %xmm0 -; X32-NEXT: vsubss {{[0-9]+}}(%esp), %xmm0, %xmm0 ## 16-byte Folded Reload +; X32-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero +; X32-NEXT: vpermilps $0, {{[0-9]+}}(%esp), %xmm1 ## 16-byte Folded Reload +; X32-NEXT: ## xmm1 = mem[0,0,0,0] +; X32-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,0] +; X32-NEXT: vsubss %xmm1, %xmm0, %xmm0 ; X32-NEXT: vmovss %xmm0, {{[0-9]+}}(%esp) ; X32-NEXT: flds {{[0-9]+}}(%esp) ; X32-NEXT: addl $56, %esp @@ -859,12 +862,15 @@ define float @broadcast_lifetime() nounwind { ; X64-NEXT: subq $40, %rsp ; X64-NEXT: movq %rsp, %rdi ; X64-NEXT: callq _gfunc -; X64-NEXT: vbroadcastss {{[0-9]+}}(%rsp), %xmm0 +; X64-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero ; X64-NEXT: vmovaps %xmm0, {{[0-9]+}}(%rsp) ## 16-byte Spill ; X64-NEXT: movq %rsp, %rdi ; X64-NEXT: callq _gfunc -; X64-NEXT: vbroadcastss {{[0-9]+}}(%rsp), %xmm0 -; X64-NEXT: vsubss {{[0-9]+}}(%rsp), %xmm0, %xmm0 ## 16-byte Folded Reload +; X64-NEXT: vmovss {{.*#+}} xmm0 = mem[0],zero,zero,zero +; X64-NEXT: vpermilps $0, {{[0-9]+}}(%rsp), %xmm1 ## 16-byte Folded Reload +; X64-NEXT: ## xmm1 = mem[0,0,0,0] +; X64-NEXT: vpermilps {{.*#+}} xmm0 = xmm0[0,0,0,0] +; X64-NEXT: vsubss %xmm1, %xmm0, %xmm0 ; X64-NEXT: addq $40, %rsp ; X64-NEXT: retq %1 = alloca <4 x float>, align 16 |
