1 ; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py
2 ; RUN: llc < %s -mtriple=x86_64-unknown -mcpu=core-avx2 | FileCheck %s
4 %v8_uniform_FVector3 = type { float, float, float }
6 declare <8 x float> @llvm.x86.avx.hadd.ps.256(<8 x float>, <8 x float>)
8 define void @foo(ptr %Out, ptr %In, <8 x i32> %__mask) {
10 ; CHECK: # %bb.0: # %allocas
11 ; CHECK-NEXT: vmovups (%rsi), %xmm0
12 ; CHECK-NEXT: vhaddps 32(%rsi), %xmm0, %xmm0
13 ; CHECK-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,0,1,1]
14 ; CHECK-NEXT: vhaddps %ymm0, %ymm0, %ymm0
15 ; CHECK-NEXT: vextractf128 $1, %ymm0, %xmm1
16 ; CHECK-NEXT: vaddss %xmm1, %xmm0, %xmm0
17 ; CHECK-NEXT: vmovss %xmm0, (%rdi)
18 ; CHECK-NEXT: vzeroupper
21 %ptr_masked_load74 = load <8 x float>, ptr %In, align 4
22 %ptr8096 = getelementptr float, ptr %In, i64 8
23 %ptr80_masked_load82 = load <8 x float>, ptr %ptr8096, align 4
24 %ret_7.i.i = shufflevector <8 x float> %ptr_masked_load74, <8 x float> %ptr80_masked_load82, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 8, i32 9, i32 10, i32 11>
25 %v1.i.i100 = tail call <8 x float> @llvm.x86.avx.hadd.ps.256(<8 x float> %ret_7.i.i, <8 x float> %ret_7.i.i)
26 %v2.i.i101 = tail call <8 x float> @llvm.x86.avx.hadd.ps.256(<8 x float> %v1.i.i100, <8 x float> %v1.i.i100)
27 %scalar1.i.i102 = extractelement <8 x float> %v2.i.i101, i32 0
28 %scalar2.i.i103 = extractelement <8 x float> %v2.i.i101, i32 4
29 %sum.i.i104 = fadd float %scalar1.i.i102, %scalar2.i.i103
30 store float %sum.i.i104, ptr %Out, align 4