11455e4758
Followup to D125988 - FPOW is similar to FREM and will most likely scalarize to libcalls, so unroll before widening to prevent use making additional libcalls with UNDEF args.
39 lines
1.6 KiB
LLVM
39 lines
1.6 KiB
LLVM
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
|
|
; RUN: llc -mtriple=x86_64-linux-gnu < %s | FileCheck %s
|
|
|
|
; Ensure vectorized FPOWs are not widened/unrolled such that they get lowered
|
|
; into libcalls on undef elements.
|
|
|
|
define float @test_fpow_v2f32_multiuse(<2 x float> %a0, <2 x float> %a1, <2 x float> *%p3) nounwind {
|
|
; CHECK-LABEL: test_fpow_v2f32_multiuse:
|
|
; CHECK: # %bb.0:
|
|
; CHECK-NEXT: pushq %rbx
|
|
; CHECK-NEXT: subq $48, %rsp
|
|
; CHECK-NEXT: movq %rdi, %rbx
|
|
; CHECK-NEXT: movaps %xmm1, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
|
|
; CHECK-NEXT: movaps %xmm0, (%rsp) # 16-byte Spill
|
|
; CHECK-NEXT: callq powf@PLT
|
|
; CHECK-NEXT: movaps %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill
|
|
; CHECK-NEXT: movaps (%rsp), %xmm0 # 16-byte Reload
|
|
; CHECK-NEXT: shufps {{.*#+}} xmm0 = xmm0[1,1,1,1]
|
|
; CHECK-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm1 # 16-byte Reload
|
|
; CHECK-NEXT: shufps {{.*#+}} xmm1 = xmm1[1,1,1,1]
|
|
; CHECK-NEXT: callq powf@PLT
|
|
; CHECK-NEXT: movaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm2 # 16-byte Reload
|
|
; CHECK-NEXT: movaps %xmm2, %xmm1
|
|
; CHECK-NEXT: unpcklps {{.*#+}} xmm1 = xmm1[0],xmm0[0],xmm1[1],xmm0[1]
|
|
; CHECK-NEXT: addss %xmm2, %xmm0
|
|
; CHECK-NEXT: movlps %xmm1, (%rbx)
|
|
; CHECK-NEXT: addq $48, %rsp
|
|
; CHECK-NEXT: popq %rbx
|
|
; CHECK-NEXT: retq
|
|
%fp = call <2 x float> @llvm.pow.v2f32(<2 x float> %a0, <2 x float> %a1)
|
|
%fp0 = extractelement <2 x float> %fp, i32 0
|
|
%fp1 = extractelement <2 x float> %fp, i32 1
|
|
%res = fadd float %fp0, %fp1
|
|
store <2 x float> %fp, <2 x float> *%p3
|
|
ret float %res
|
|
}
|
|
declare <2 x float> @llvm.pow.v2f32(<2 x float>, <2 x float>)
|
|
|