diff options
| author | Simon Pilgrim <llvm-dev@redking.me.uk> | 2019-06-13 11:42:32 +0000 |
|---|---|---|
| committer | Simon Pilgrim <llvm-dev@redking.me.uk> | 2019-06-13 11:42:32 +0000 |
| commit | a284f4fa7ca1170e65ca5dc58f87ad2989bdc846 (patch) | |
| tree | b1c81a24111e502178c1ac457f355a7c14285add | |
| parent | 0baf136a4d0de7f08d6002bb360d98a57bf7f6aa (diff) | |
| download | bcm5719-llvm-a284f4fa7ca1170e65ca5dc58f87ad2989bdc846.tar.gz bcm5719-llvm-a284f4fa7ca1170e65ca5dc58f87ad2989bdc846.zip | |
[X86][AVX] Add broadcast(v4f64 hadd) test
llvm-svn: 363252
| -rw-r--r-- | llvm/test/CodeGen/X86/haddsub-shuf.ll | 54 |
1 files changed, 54 insertions, 0 deletions
diff --git a/llvm/test/CodeGen/X86/haddsub-shuf.ll b/llvm/test/CodeGen/X86/haddsub-shuf.ll index 0787269072b..d35481161b9 100644 --- a/llvm/test/CodeGen/X86/haddsub-shuf.ll +++ b/llvm/test/CodeGen/X86/haddsub-shuf.ll @@ -263,6 +263,60 @@ define <4 x double> @hadd_v4f64_scalar_splat(<4 x double> %a) { ret <4 x double> %shuf } +define <4 x double> @hadd_v4f64_scalar_broadcast(<4 x double> %a) { +; SSSE3_SLOW-LABEL: hadd_v4f64_scalar_broadcast: +; SSSE3_SLOW: # %bb.0: +; SSSE3_SLOW-NEXT: movapd %xmm0, %xmm1 +; SSSE3_SLOW-NEXT: unpckhpd {{.*#+}} xmm1 = xmm1[1],xmm0[1] +; SSSE3_SLOW-NEXT: addsd %xmm0, %xmm1 +; SSSE3_SLOW-NEXT: movddup {{.*#+}} xmm0 = xmm1[0,0] +; SSSE3_SLOW-NEXT: movapd %xmm0, %xmm1 +; SSSE3_SLOW-NEXT: retq +; +; SSSE3_FAST-LABEL: hadd_v4f64_scalar_broadcast: +; SSSE3_FAST: # %bb.0: +; SSSE3_FAST-NEXT: haddpd %xmm0, %xmm0 +; SSSE3_FAST-NEXT: movapd %xmm0, %xmm1 +; SSSE3_FAST-NEXT: retq +; +; AVX1_SLOW-LABEL: hadd_v4f64_scalar_broadcast: +; AVX1_SLOW: # %bb.0: +; AVX1_SLOW-NEXT: vpermilpd {{.*#+}} xmm1 = xmm0[1,0] +; AVX1_SLOW-NEXT: vaddsd %xmm1, %xmm0, %xmm0 +; AVX1_SLOW-NEXT: vmovddup {{.*#+}} xmm0 = xmm0[0,0] +; AVX1_SLOW-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0 +; AVX1_SLOW-NEXT: retq +; +; AVX1_FAST-LABEL: hadd_v4f64_scalar_broadcast: +; AVX1_FAST: # %bb.0: +; AVX1_FAST-NEXT: vhaddpd %xmm0, %xmm0, %xmm0 +; AVX1_FAST-NEXT: vinsertf128 $1, %xmm0, %ymm0, %ymm0 +; AVX1_FAST-NEXT: retq +; +; AVX2_SLOW-LABEL: hadd_v4f64_scalar_broadcast: +; AVX2_SLOW: # %bb.0: +; AVX2_SLOW-NEXT: vpermilpd {{.*#+}} xmm1 = xmm0[1,0] +; AVX2_SLOW-NEXT: vaddsd %xmm1, %xmm0, %xmm0 +; AVX2_SLOW-NEXT: vbroadcastsd %xmm0, %ymm0 +; AVX2_SLOW-NEXT: retq +; +; AVX2_FAST-LABEL: hadd_v4f64_scalar_broadcast: +; AVX2_FAST: # %bb.0: +; AVX2_FAST-NEXT: vhaddpd %xmm0, %xmm0, %xmm0 +; AVX2_FAST-NEXT: vbroadcastsd %xmm0, %ymm0 +; AVX2_FAST-NEXT: retq + %a0 = extractelement <4 x double> %a, i32 0 + %a1 = extractelement <4 x double> %a, i32 1 + %hop0 = fadd double %a0, %a1 + %a2 = extractelement <4 x double> %a, i32 2 + %a3 = extractelement <4 x double> %a, i32 3 + %hop1 = fadd double %a2, %a3 + %ins = insertelement <4 x double> undef, double %hop0, i32 0 + %ins2 = insertelement <4 x double> %ins, double %hop1, i32 2 + %shuf = shufflevector <4 x double> %ins2, <4 x double> undef, <4 x i32> <i32 0, i32 0, i32 0, i32 0> + ret <4 x double> %shuf +} + define <4 x double> @hadd_v4f64(<4 x double> %a) { ; SSSE3_SLOW-LABEL: hadd_v4f64: ; SSSE3_SLOW: # %bb.0: |
