diff options
Diffstat (limited to 'llvm/test/CodeGen/X86/pmulh.ll')
| -rw-r--r-- | llvm/test/CodeGen/X86/pmulh.ll | 84 |
1 files changed, 74 insertions, 10 deletions
diff --git a/llvm/test/CodeGen/X86/pmulh.ll b/llvm/test/CodeGen/X86/pmulh.ll index 269d0b51c7b..7068650110e 100644 --- a/llvm/test/CodeGen/X86/pmulh.ll +++ b/llvm/test/CodeGen/X86/pmulh.ll @@ -8,14 +8,45 @@ ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx512bw | FileCheck %s --check-prefix=AVX --check-prefix=AVX512 --check-prefix=AVX512BW define <4 x i16> @mulhuw_v4i16(<4 x i16> %a, <4 x i16> %b) { -; SSE-LABEL: mulhuw_v4i16: -; SSE: # %bb.0: -; SSE-NEXT: pmulhuw %xmm1, %xmm0 -; SSE-NEXT: retq +; SSE2-PROMOTE-LABEL: mulhuw_v4i16: +; SSE2-PROMOTE: # %bb.0: +; SSE2-PROMOTE-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7] +; SSE2-PROMOTE-NEXT: pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,6,6,7] +; SSE2-PROMOTE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3] +; SSE2-PROMOTE-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7] +; SSE2-PROMOTE-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7] +; SSE2-PROMOTE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] +; SSE2-PROMOTE-NEXT: pmulhuw %xmm1, %xmm0 +; SSE2-PROMOTE-NEXT: pxor %xmm1, %xmm1 +; SSE2-PROMOTE-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3] +; SSE2-PROMOTE-NEXT: retq +; +; SSE2-WIDEN-LABEL: mulhuw_v4i16: +; SSE2-WIDEN: # %bb.0: +; SSE2-WIDEN-NEXT: pmulhuw %xmm1, %xmm0 +; SSE2-WIDEN-NEXT: retq +; +; SSE41-PROMOTE-LABEL: mulhuw_v4i16: +; SSE41-PROMOTE: # %bb.0: +; SSE41-PROMOTE-NEXT: pxor %xmm2, %xmm2 +; SSE41-PROMOTE-NEXT: pblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7] +; SSE41-PROMOTE-NEXT: pblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7] +; SSE41-PROMOTE-NEXT: pmulld %xmm1, %xmm0 +; SSE41-PROMOTE-NEXT: psrld $16, %xmm0 +; SSE41-PROMOTE-NEXT: retq +; +; SSE41-WIDEN-LABEL: mulhuw_v4i16: +; SSE41-WIDEN: # %bb.0: +; SSE41-WIDEN-NEXT: pmulhuw %xmm1, %xmm0 +; SSE41-WIDEN-NEXT: retq ; ; AVX-LABEL: mulhuw_v4i16: ; AVX: # %bb.0: -; AVX-NEXT: vpmulhuw %xmm1, %xmm0, %xmm0 +; AVX-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; AVX-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7] +; AVX-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7] +; AVX-NEXT: vpmulld %xmm1, %xmm0, %xmm0 +; AVX-NEXT: vpsrld $16, %xmm0, %xmm0 ; AVX-NEXT: retq %a1 = zext <4 x i16> %a to <4 x i32> %b1 = zext <4 x i16> %b to <4 x i32> @@ -26,14 +57,47 @@ define <4 x i16> @mulhuw_v4i16(<4 x i16> %a, <4 x i16> %b) { } define <4 x i16> @mulhw_v4i16(<4 x i16> %a, <4 x i16> %b) { -; SSE-LABEL: mulhw_v4i16: -; SSE: # %bb.0: -; SSE-NEXT: pmulhw %xmm1, %xmm0 -; SSE-NEXT: retq +; SSE2-PROMOTE-LABEL: mulhw_v4i16: +; SSE2-PROMOTE: # %bb.0: +; SSE2-PROMOTE-NEXT: pshuflw {{.*#+}} xmm1 = xmm1[0,2,2,3,4,5,6,7] +; SSE2-PROMOTE-NEXT: pshufhw {{.*#+}} xmm1 = xmm1[0,1,2,3,4,6,6,7] +; SSE2-PROMOTE-NEXT: pshufd {{.*#+}} xmm1 = xmm1[0,2,2,3] +; SSE2-PROMOTE-NEXT: pshuflw {{.*#+}} xmm0 = xmm0[0,2,2,3,4,5,6,7] +; SSE2-PROMOTE-NEXT: pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,4,6,6,7] +; SSE2-PROMOTE-NEXT: pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3] +; SSE2-PROMOTE-NEXT: pmulhw %xmm1, %xmm0 +; SSE2-PROMOTE-NEXT: pxor %xmm1, %xmm1 +; SSE2-PROMOTE-NEXT: punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3] +; SSE2-PROMOTE-NEXT: retq +; +; SSE2-WIDEN-LABEL: mulhw_v4i16: +; SSE2-WIDEN: # %bb.0: +; SSE2-WIDEN-NEXT: pmulhw %xmm1, %xmm0 +; SSE2-WIDEN-NEXT: retq +; +; SSE41-PROMOTE-LABEL: mulhw_v4i16: +; SSE41-PROMOTE: # %bb.0: +; SSE41-PROMOTE-NEXT: pslld $16, %xmm0 +; SSE41-PROMOTE-NEXT: psrad $16, %xmm0 +; SSE41-PROMOTE-NEXT: pslld $16, %xmm1 +; SSE41-PROMOTE-NEXT: psrad $16, %xmm1 +; SSE41-PROMOTE-NEXT: pmulld %xmm1, %xmm0 +; SSE41-PROMOTE-NEXT: psrld $16, %xmm0 +; SSE41-PROMOTE-NEXT: retq +; +; SSE41-WIDEN-LABEL: mulhw_v4i16: +; SSE41-WIDEN: # %bb.0: +; SSE41-WIDEN-NEXT: pmulhw %xmm1, %xmm0 +; SSE41-WIDEN-NEXT: retq ; ; AVX-LABEL: mulhw_v4i16: ; AVX: # %bb.0: -; AVX-NEXT: vpmulhw %xmm1, %xmm0, %xmm0 +; AVX-NEXT: vpslld $16, %xmm0, %xmm0 +; AVX-NEXT: vpsrad $16, %xmm0, %xmm0 +; AVX-NEXT: vpslld $16, %xmm1, %xmm1 +; AVX-NEXT: vpsrad $16, %xmm1, %xmm1 +; AVX-NEXT: vpmulld %xmm1, %xmm0, %xmm0 +; AVX-NEXT: vpsrld $16, %xmm0, %xmm0 ; AVX-NEXT: retq %a1 = sext <4 x i16> %a to <4 x i32> %b1 = sext <4 x i16> %b to <4 x i32> |
