1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 3 2; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=avx2 | FileCheck %s --check-prefix=AVX2 3; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=f16c,fma | FileCheck %s --check-prefix=F16C 4; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=avx512vl | FileCheck %s --check-prefix=F16C 5; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=avx512fp16 | FileCheck %s --check-prefix=FP16 6 7define <2 x half> @foo(<2 x half> %0) "unsafe-fp-math"="true" nounwind { 8; AVX2-LABEL: foo: 9; AVX2: # %bb.0: 10; AVX2-NEXT: subq $40, %rsp 11; AVX2-NEXT: vmovdqa %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 16-byte Spill 12; AVX2-NEXT: vpsrld $16, %xmm0, %xmm0 13; AVX2-NEXT: callq __extendhfsf2@PLT 14; AVX2-NEXT: vmulss {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm0, %xmm0 15; AVX2-NEXT: callq __truncsfhf2@PLT 16; AVX2-NEXT: vmovss %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill 17; AVX2-NEXT: vmovaps {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 16-byte Reload 18; AVX2-NEXT: callq __extendhfsf2@PLT 19; AVX2-NEXT: vmovss %xmm0, {{[-0-9]+}}(%r{{[sb]}}p) # 4-byte Spill 20; AVX2-NEXT: vmovss {{[-0-9]+}}(%r{{[sb]}}p), %xmm0 # 4-byte Reload 21; AVX2-NEXT: # xmm0 = mem[0],zero,zero,zero 22; AVX2-NEXT: callq __extendhfsf2@PLT 23; AVX2-NEXT: vsubss {{[-0-9]+}}(%r{{[sb]}}p), %xmm0, %xmm0 # 4-byte Folded Reload 24; AVX2-NEXT: callq __truncsfhf2@PLT 25; AVX2-NEXT: addq $40, %rsp 26; AVX2-NEXT: retq 27; 28; F16C-LABEL: foo: 29; F16C: # %bb.0: 30; F16C-NEXT: vpsrld $16, %xmm0, %xmm1 31; F16C-NEXT: vcvtph2ps %xmm1, %ymm1 32; F16C-NEXT: vmulps {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %ymm1, %ymm1 33; F16C-NEXT: vcvtps2ph $4, %ymm1, %xmm1 34; F16C-NEXT: vcvtph2ps %xmm0, %ymm0 35; F16C-NEXT: vcvtph2ps %xmm1, %ymm1 36; F16C-NEXT: vsubps %ymm0, %ymm1, %ymm2 37; F16C-NEXT: vcvtps2ph $4, %ymm2, %xmm2 38; F16C-NEXT: vaddps %ymm0, %ymm1, %ymm0 39; F16C-NEXT: vcvtps2ph $4, %ymm0, %xmm0 40; F16C-NEXT: vpblendw {{.*#+}} xmm0 = xmm2[0],xmm0[1],xmm2[2,3,4,5,6,7] 41; F16C-NEXT: vzeroupper 42; F16C-NEXT: retq 43; 44; FP16-LABEL: foo: 45; FP16: # %bb.0: 46; FP16-NEXT: vpsrld $16, %xmm0, %xmm1 47; FP16-NEXT: vfmaddsub231ph {{\.?LCPI[0-9]+_[0-9]+}}(%rip), %xmm1, %xmm0 48; FP16-NEXT: retq 49 %2 = shufflevector <2 x half> %0, <2 x half> undef, <2 x i32> <i32 1, i32 2> 50 %3 = fmul fast <2 x half> %2, <half 0xH3D3A, half 0xH3854> 51 %4 = fsub fast <2 x half> %3, %0 52 %5 = fadd fast <2 x half> %3, %0 53 %6 = shufflevector <2 x half> %4, <2 x half> %5, <2 x i32> <i32 0, i32 3> 54 %7 = fadd fast <2 x half> %6, zeroinitializer 55 %8 = shufflevector <2 x half> undef, <2 x half> %7, <2 x i32> <i32 0, i32 3> 56 %9 = fsub fast <2 x half> %8, zeroinitializer 57 ret <2 x half> %9 58} 59