diff --git a/llvm/test/CodeGen/X86/vector-shift-ashr-256.ll b/llvm/test/CodeGen/X86/vector-shift-ashr-256.ll index af3ddcf8048e..eba539976154 100644 --- a/llvm/test/CodeGen/X86/vector-shift-ashr-256.ll +++ b/llvm/test/CodeGen/X86/vector-shift-ashr-256.ll @@ -7,6 +7,10 @@ ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=knl -mattr=+avx512bw | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512BW ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=knl -mattr=+avx512dq,+avx512vl | FileCheck %s --check-prefix=ALL --check-prefix=AVX512VL --check-prefix=AVX512DQVL ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=knl -mattr=+avx512bw,+avx512vl | FileCheck %s --check-prefix=ALL --check-prefix=AVX512VL --check-prefix=AVX512BWVL +; +; 32-bit runs to make sure we do reasonable things for i64 shifts. +; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=ALL --check-prefix=X32-AVX --check-prefix=X32-AVX1 +; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=ALL --check-prefix=X32-AVX --check-prefix=X32-AVX2 ; ; Variable Shifts @@ -81,6 +85,41 @@ define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) nounwind { ; AVX512VL: # BB#0: ; AVX512VL-NEXT: vpsravq %ymm1, %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: var_shift_v4i64: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2 +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [0,2147483648,0,2147483648] +; X32-AVX1-NEXT: vpsrlq %xmm2, %xmm3, %xmm4 +; X32-AVX1-NEXT: vpshufd {{.*#+}} xmm5 = xmm2[2,3,0,1] +; X32-AVX1-NEXT: vpsrlq %xmm5, %xmm3, %xmm6 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm4 = xmm4[0,1,2,3],xmm6[4,5,6,7] +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm6 +; X32-AVX1-NEXT: vpsrlq %xmm2, %xmm6, %xmm2 +; X32-AVX1-NEXT: vpsrlq %xmm5, %xmm6, %xmm5 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm5[4,5,6,7] +; X32-AVX1-NEXT: vpxor %xmm4, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsubq %xmm4, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsrlq %xmm1, %xmm3, %xmm4 +; X32-AVX1-NEXT: vpshufd {{.*#+}} xmm5 = xmm1[2,3,0,1] +; X32-AVX1-NEXT: vpsrlq %xmm5, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm4[0,1,2,3],xmm3[4,5,6,7] +; X32-AVX1-NEXT: vpsrlq %xmm1, %xmm0, %xmm1 +; X32-AVX1-NEXT: vpsrlq %xmm5, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7] +; X32-AVX1-NEXT: vpxor %xmm3, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsubq %xmm3, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: var_shift_v4i64: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vmovdqa {{.*#+}} ymm2 = [0,2147483648,0,2147483648,0,2147483648,0,2147483648] +; X32-AVX2-NEXT: vpsrlvq %ymm1, %ymm2, %ymm3 +; X32-AVX2-NEXT: vpxor %ymm2, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsrlvq %ymm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsubq %ymm3, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = ashr <4 x i64> %a, %b ret <4 x i64> %shift } @@ -147,6 +186,41 @@ define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind { ; AVX512VL: # BB#0: ; AVX512VL-NEXT: vpsravd %ymm1, %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: var_shift_v8i32: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2 +; X32-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm3 +; X32-AVX1-NEXT: vpsrldq {{.*#+}} xmm4 = xmm3[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero +; X32-AVX1-NEXT: vpsrad %xmm4, %xmm2, %xmm4 +; X32-AVX1-NEXT: vpsrlq $32, %xmm3, %xmm5 +; X32-AVX1-NEXT: vpsrad %xmm5, %xmm2, %xmm5 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm4 = xmm5[0,1,2,3],xmm4[4,5,6,7] +; X32-AVX1-NEXT: vpxor %xmm5, %xmm5, %xmm5 +; X32-AVX1-NEXT: vpunpckhdq {{.*#+}} xmm6 = xmm3[2],xmm5[2],xmm3[3],xmm5[3] +; X32-AVX1-NEXT: vpsrad %xmm6, %xmm2, %xmm6 +; X32-AVX1-NEXT: vpmovzxdq {{.*#+}} xmm3 = xmm3[0],zero,xmm3[1],zero +; X32-AVX1-NEXT: vpsrad %xmm3, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm6[4,5,6,7] +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1],xmm4[2,3],xmm2[4,5],xmm4[6,7] +; X32-AVX1-NEXT: vpsrldq {{.*#+}} xmm3 = xmm1[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero +; X32-AVX1-NEXT: vpsrad %xmm3, %xmm0, %xmm3 +; X32-AVX1-NEXT: vpsrlq $32, %xmm1, %xmm4 +; X32-AVX1-NEXT: vpsrad %xmm4, %xmm0, %xmm4 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm4[0,1,2,3],xmm3[4,5,6,7] +; X32-AVX1-NEXT: vpunpckhdq {{.*#+}} xmm4 = xmm1[2],xmm5[2],xmm1[3],xmm5[3] +; X32-AVX1-NEXT: vpsrad %xmm4, %xmm0, %xmm4 +; X32-AVX1-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero +; X32-AVX1-NEXT: vpsrad %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm4[4,5,6,7] +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm3[2,3],xmm0[4,5],xmm3[6,7] +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: var_shift_v8i32: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpsravd %ymm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = ashr <8 x i32> %a, %b ret <8 x i32> %shift } @@ -253,6 +327,55 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind { ; AVX512BWVL: # BB#0: ; AVX512BWVL-NEXT: vpsravw %ymm1, %ymm0, %ymm0 ; AVX512BWVL-NEXT: retq +; +; X32-AVX1-LABEL: var_shift_v16i16: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2 +; X32-AVX1-NEXT: vpsllw $12, %xmm2, %xmm3 +; X32-AVX1-NEXT: vpsllw $4, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpor %xmm3, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpaddw %xmm2, %xmm2, %xmm3 +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm4 +; X32-AVX1-NEXT: vpsraw $8, %xmm4, %xmm5 +; X32-AVX1-NEXT: vpblendvb %xmm2, %xmm5, %xmm4, %xmm2 +; X32-AVX1-NEXT: vpsraw $4, %xmm2, %xmm4 +; X32-AVX1-NEXT: vpblendvb %xmm3, %xmm4, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsraw $2, %xmm2, %xmm4 +; X32-AVX1-NEXT: vpaddw %xmm3, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpblendvb %xmm3, %xmm4, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsraw $1, %xmm2, %xmm4 +; X32-AVX1-NEXT: vpaddw %xmm3, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpblendvb %xmm3, %xmm4, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsllw $12, %xmm1, %xmm3 +; X32-AVX1-NEXT: vpsllw $4, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpor %xmm3, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpaddw %xmm1, %xmm1, %xmm3 +; X32-AVX1-NEXT: vpsraw $8, %xmm0, %xmm4 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm4, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsraw $4, %xmm0, %xmm1 +; X32-AVX1-NEXT: vpblendvb %xmm3, %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsraw $2, %xmm0, %xmm1 +; X32-AVX1-NEXT: vpaddw %xmm3, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpblendvb %xmm3, %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsraw $1, %xmm0, %xmm1 +; X32-AVX1-NEXT: vpaddw %xmm3, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpblendvb %xmm3, %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: var_shift_v16i16: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpxor %ymm2, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpunpckhwd {{.*#+}} ymm3 = ymm1[4],ymm2[4],ymm1[5],ymm2[5],ymm1[6],ymm2[6],ymm1[7],ymm2[7],ymm1[12],ymm2[12],ymm1[13],ymm2[13],ymm1[14],ymm2[14],ymm1[15],ymm2[15] +; X32-AVX2-NEXT: vpunpckhwd {{.*#+}} ymm4 = ymm2[4],ymm0[4],ymm2[5],ymm0[5],ymm2[6],ymm0[6],ymm2[7],ymm0[7],ymm2[12],ymm0[12],ymm2[13],ymm0[13],ymm2[14],ymm0[14],ymm2[15],ymm0[15] +; X32-AVX2-NEXT: vpsravd %ymm3, %ymm4, %ymm3 +; X32-AVX2-NEXT: vpsrld $16, %ymm3, %ymm3 +; X32-AVX2-NEXT: vpunpcklwd {{.*#+}} ymm1 = ymm1[0],ymm2[0],ymm1[1],ymm2[1],ymm1[2],ymm2[2],ymm1[3],ymm2[3],ymm1[8],ymm2[8],ymm1[9],ymm2[9],ymm1[10],ymm2[10],ymm1[11],ymm2[11] +; X32-AVX2-NEXT: vpunpcklwd {{.*#+}} ymm0 = ymm2[0],ymm0[0],ymm2[1],ymm0[1],ymm2[2],ymm0[2],ymm2[3],ymm0[3],ymm2[8],ymm0[8],ymm2[9],ymm0[9],ymm2[10],ymm0[10],ymm2[11],ymm0[11] +; X32-AVX2-NEXT: vpsravd %ymm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsrld $16, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpackusdw %ymm3, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = ashr <16 x i16> %a, %b ret <16 x i16> %shift } @@ -436,6 +559,89 @@ define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind { ; AVX512BWVL-NEXT: vpsravw %zmm1, %zmm0, %zmm0 ; AVX512BWVL-NEXT: vpmovwb %zmm0, %ymm0 ; AVX512BWVL-NEXT: retq +; +; X32-AVX1-LABEL: var_shift_v32i8: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2 +; X32-AVX1-NEXT: vpsllw $5, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpunpckhbw {{.*#+}} xmm3 = xmm0[8],xmm2[8],xmm0[9],xmm2[9],xmm0[10],xmm2[10],xmm0[11],xmm2[11],xmm0[12],xmm2[12],xmm0[13],xmm2[13],xmm0[14],xmm2[14],xmm0[15],xmm2[15] +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm4 +; X32-AVX1-NEXT: vpunpckhbw {{.*#+}} xmm5 = xmm0[8],xmm4[8],xmm0[9],xmm4[9],xmm0[10],xmm4[10],xmm0[11],xmm4[11],xmm0[12],xmm4[12],xmm0[13],xmm4[13],xmm0[14],xmm4[14],xmm0[15],xmm4[15] +; X32-AVX1-NEXT: vpsraw $4, %xmm5, %xmm6 +; X32-AVX1-NEXT: vpblendvb %xmm3, %xmm6, %xmm5, %xmm5 +; X32-AVX1-NEXT: vpsraw $2, %xmm5, %xmm6 +; X32-AVX1-NEXT: vpaddw %xmm3, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpblendvb %xmm3, %xmm6, %xmm5, %xmm5 +; X32-AVX1-NEXT: vpsraw $1, %xmm5, %xmm6 +; X32-AVX1-NEXT: vpaddw %xmm3, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpblendvb %xmm3, %xmm6, %xmm5, %xmm3 +; X32-AVX1-NEXT: vpsrlw $8, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpunpcklbw {{.*#+}} xmm2 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7] +; X32-AVX1-NEXT: vpunpcklbw {{.*#+}} xmm4 = xmm0[0],xmm4[0],xmm0[1],xmm4[1],xmm0[2],xmm4[2],xmm0[3],xmm4[3],xmm0[4],xmm4[4],xmm0[5],xmm4[5],xmm0[6],xmm4[6],xmm0[7],xmm4[7] +; X32-AVX1-NEXT: vpsraw $4, %xmm4, %xmm5 +; X32-AVX1-NEXT: vpblendvb %xmm2, %xmm5, %xmm4, %xmm4 +; X32-AVX1-NEXT: vpsraw $2, %xmm4, %xmm5 +; X32-AVX1-NEXT: vpaddw %xmm2, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpblendvb %xmm2, %xmm5, %xmm4, %xmm4 +; X32-AVX1-NEXT: vpsraw $1, %xmm4, %xmm5 +; X32-AVX1-NEXT: vpaddw %xmm2, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpblendvb %xmm2, %xmm5, %xmm4, %xmm2 +; X32-AVX1-NEXT: vpsrlw $8, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpackuswb %xmm3, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsllw $5, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpunpckhbw {{.*#+}} xmm3 = xmm0[8],xmm1[8],xmm0[9],xmm1[9],xmm0[10],xmm1[10],xmm0[11],xmm1[11],xmm0[12],xmm1[12],xmm0[13],xmm1[13],xmm0[14],xmm1[14],xmm0[15],xmm1[15] +; X32-AVX1-NEXT: vpunpckhbw {{.*#+}} xmm4 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15] +; X32-AVX1-NEXT: vpsraw $4, %xmm4, %xmm5 +; X32-AVX1-NEXT: vpblendvb %xmm3, %xmm5, %xmm4, %xmm4 +; X32-AVX1-NEXT: vpsraw $2, %xmm4, %xmm5 +; X32-AVX1-NEXT: vpaddw %xmm3, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpblendvb %xmm3, %xmm5, %xmm4, %xmm4 +; X32-AVX1-NEXT: vpsraw $1, %xmm4, %xmm5 +; X32-AVX1-NEXT: vpaddw %xmm3, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpblendvb %xmm3, %xmm5, %xmm4, %xmm3 +; X32-AVX1-NEXT: vpsrlw $8, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpunpcklbw {{.*#+}} xmm1 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] +; X32-AVX1-NEXT: vpunpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7] +; X32-AVX1-NEXT: vpsraw $4, %xmm0, %xmm4 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm4, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsraw $2, %xmm0, %xmm4 +; X32-AVX1-NEXT: vpaddw %xmm1, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm4, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsraw $1, %xmm0, %xmm4 +; X32-AVX1-NEXT: vpaddw %xmm1, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm4, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsrlw $8, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpackuswb %xmm3, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: var_shift_v32i8: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpsllw $5, %ymm1, %ymm1 +; X32-AVX2-NEXT: vpunpckhbw {{.*#+}} ymm2 = ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11],ymm0[12],ymm1[12],ymm0[13],ymm1[13],ymm0[14],ymm1[14],ymm0[15],ymm1[15],ymm0[24],ymm1[24],ymm0[25],ymm1[25],ymm0[26],ymm1[26],ymm0[27],ymm1[27],ymm0[28],ymm1[28],ymm0[29],ymm1[29],ymm0[30],ymm1[30],ymm0[31],ymm1[31] +; X32-AVX2-NEXT: vpunpckhbw {{.*#+}} ymm3 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31] +; X32-AVX2-NEXT: vpsraw $4, %ymm3, %ymm4 +; X32-AVX2-NEXT: vpblendvb %ymm2, %ymm4, %ymm3, %ymm3 +; X32-AVX2-NEXT: vpsraw $2, %ymm3, %ymm4 +; X32-AVX2-NEXT: vpaddw %ymm2, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpblendvb %ymm2, %ymm4, %ymm3, %ymm3 +; X32-AVX2-NEXT: vpsraw $1, %ymm3, %ymm4 +; X32-AVX2-NEXT: vpaddw %ymm2, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpblendvb %ymm2, %ymm4, %ymm3, %ymm2 +; X32-AVX2-NEXT: vpsrlw $8, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpunpcklbw {{.*#+}} ymm1 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[4],ymm1[4],ymm0[5],ymm1[5],ymm0[6],ymm1[6],ymm0[7],ymm1[7],ymm0[16],ymm1[16],ymm0[17],ymm1[17],ymm0[18],ymm1[18],ymm0[19],ymm1[19],ymm0[20],ymm1[20],ymm0[21],ymm1[21],ymm0[22],ymm1[22],ymm0[23],ymm1[23] +; X32-AVX2-NEXT: vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23] +; X32-AVX2-NEXT: vpsraw $4, %ymm0, %ymm3 +; X32-AVX2-NEXT: vpblendvb %ymm1, %ymm3, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsraw $2, %ymm0, %ymm3 +; X32-AVX2-NEXT: vpaddw %ymm1, %ymm1, %ymm1 +; X32-AVX2-NEXT: vpblendvb %ymm1, %ymm3, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsraw $1, %ymm0, %ymm3 +; X32-AVX2-NEXT: vpaddw %ymm1, %ymm1, %ymm1 +; X32-AVX2-NEXT: vpblendvb %ymm1, %ymm3, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsrlw $8, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpackuswb %ymm2, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = ashr <32 x i8> %a, %b ret <32 x i8> %shift } @@ -499,6 +705,33 @@ define <4 x i64> @splatvar_shift_v4i64(<4 x i64> %a, <4 x i64> %b) nounwind { ; AVX512VL: # BB#0: ; AVX512VL-NEXT: vpsraq %xmm1, %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: splatvar_shift_v4i64: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vpextrd $1, %xmm1, %eax +; X32-AVX1-NEXT: vpinsrd $1, %eax, %xmm1, %xmm1 +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [0,2147483648,0,2147483648] +; X32-AVX1-NEXT: vpsrlq %xmm1, %xmm2, %xmm2 +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3 +; X32-AVX1-NEXT: vpsrlq %xmm1, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpxor %xmm2, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpsubq %xmm2, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpsrlq %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpxor %xmm2, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsubq %xmm2, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm3, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: splatvar_shift_v4i64: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpextrd $1, %xmm1, %eax +; X32-AVX2-NEXT: vpinsrd $1, %eax, %xmm1, %xmm1 +; X32-AVX2-NEXT: vmovdqa {{.*#+}} ymm2 = [0,2147483648,0,2147483648,0,2147483648,0,2147483648] +; X32-AVX2-NEXT: vpsrlq %xmm1, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpsrlq %xmm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpxor %ymm2, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsubq %ymm2, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %splat = shufflevector <4 x i64> %b, <4 x i64> undef, <4 x i32> zeroinitializer %shift = ashr <4 x i64> %a, %splat ret <4 x i64> %shift @@ -546,6 +779,21 @@ define <8 x i32> @splatvar_shift_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind { ; AVX512VL-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero ; AVX512VL-NEXT: vpsrad %xmm1, %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: splatvar_shift_v8i32: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2 +; X32-AVX1-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero +; X32-AVX1-NEXT: vpsrad %xmm1, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsrad %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: splatvar_shift_v8i32: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero +; X32-AVX2-NEXT: vpsrad %xmm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %splat = shufflevector <8 x i32> %b, <8 x i32> undef, <8 x i32> zeroinitializer %shift = ashr <8 x i32> %a, %splat ret <8 x i32> %shift @@ -593,6 +841,21 @@ define <16 x i16> @splatvar_shift_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind ; AVX512VL-NEXT: vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero ; AVX512VL-NEXT: vpsraw %xmm1, %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: splatvar_shift_v16i16: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2 +; X32-AVX1-NEXT: vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero +; X32-AVX1-NEXT: vpsraw %xmm1, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsraw %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: splatvar_shift_v16i16: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero +; X32-AVX2-NEXT: vpsraw %xmm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %splat = shufflevector <16 x i16> %b, <16 x i16> undef, <16 x i32> zeroinitializer %shift = ashr <16 x i16> %a, %splat ret <16 x i16> %shift @@ -776,6 +1039,84 @@ define <32 x i8> @splatvar_shift_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind { ; AVX512BWVL-NEXT: vpsravw %zmm1, %zmm0, %zmm0 ; AVX512BWVL-NEXT: vpmovwb %zmm0, %ymm0 ; AVX512BWVL-NEXT: retq +; +; X32-AVX1-LABEL: splatvar_shift_v32i8: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpshufb %xmm2, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpsllw $5, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpunpckhbw {{.*#+}} xmm2 = xmm0[8],xmm1[8],xmm0[9],xmm1[9],xmm0[10],xmm1[10],xmm0[11],xmm1[11],xmm0[12],xmm1[12],xmm0[13],xmm1[13],xmm0[14],xmm1[14],xmm0[15],xmm1[15] +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3 +; X32-AVX1-NEXT: vpunpckhbw {{.*#+}} xmm4 = xmm0[8],xmm3[8],xmm0[9],xmm3[9],xmm0[10],xmm3[10],xmm0[11],xmm3[11],xmm0[12],xmm3[12],xmm0[13],xmm3[13],xmm0[14],xmm3[14],xmm0[15],xmm3[15] +; X32-AVX1-NEXT: vpsraw $4, %xmm4, %xmm5 +; X32-AVX1-NEXT: vpblendvb %xmm2, %xmm5, %xmm4, %xmm4 +; X32-AVX1-NEXT: vpunpckhbw {{.*#+}} xmm5 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15] +; X32-AVX1-NEXT: vpsraw $4, %xmm5, %xmm6 +; X32-AVX1-NEXT: vpblendvb %xmm2, %xmm6, %xmm5, %xmm5 +; X32-AVX1-NEXT: vpsraw $2, %xmm4, %xmm6 +; X32-AVX1-NEXT: vpaddw %xmm2, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpblendvb %xmm2, %xmm6, %xmm4, %xmm4 +; X32-AVX1-NEXT: vpsraw $2, %xmm5, %xmm6 +; X32-AVX1-NEXT: vpblendvb %xmm2, %xmm6, %xmm5, %xmm5 +; X32-AVX1-NEXT: vpsraw $1, %xmm4, %xmm6 +; X32-AVX1-NEXT: vpaddw %xmm2, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpblendvb %xmm2, %xmm6, %xmm4, %xmm4 +; X32-AVX1-NEXT: vpsraw $1, %xmm5, %xmm6 +; X32-AVX1-NEXT: vpblendvb %xmm2, %xmm6, %xmm5, %xmm2 +; X32-AVX1-NEXT: vpunpcklbw {{.*#+}} xmm1 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] +; X32-AVX1-NEXT: vpunpcklbw {{.*#+}} xmm3 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3],xmm0[4],xmm3[4],xmm0[5],xmm3[5],xmm0[6],xmm3[6],xmm0[7],xmm3[7] +; X32-AVX1-NEXT: vpsraw $4, %xmm3, %xmm5 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm5, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpunpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7] +; X32-AVX1-NEXT: vpsraw $4, %xmm0, %xmm5 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm5, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsraw $2, %xmm3, %xmm5 +; X32-AVX1-NEXT: vpaddw %xmm1, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm5, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpsraw $2, %xmm0, %xmm5 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm5, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsraw $1, %xmm3, %xmm5 +; X32-AVX1-NEXT: vpaddw %xmm1, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm5, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpsraw $1, %xmm0, %xmm5 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm5, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsrlw $8, %xmm4, %xmm1 +; X32-AVX1-NEXT: vpsrlw $8, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpackuswb %xmm1, %xmm3, %xmm1 +; X32-AVX1-NEXT: vpsrlw $8, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsrlw $8, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpackuswb %xmm2, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: splatvar_shift_v32i8: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpbroadcastb %xmm1, %ymm1 +; X32-AVX2-NEXT: vpsllw $5, %ymm1, %ymm1 +; X32-AVX2-NEXT: vpunpckhbw {{.*#+}} ymm2 = ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11],ymm0[12],ymm1[12],ymm0[13],ymm1[13],ymm0[14],ymm1[14],ymm0[15],ymm1[15],ymm0[24],ymm1[24],ymm0[25],ymm1[25],ymm0[26],ymm1[26],ymm0[27],ymm1[27],ymm0[28],ymm1[28],ymm0[29],ymm1[29],ymm0[30],ymm1[30],ymm0[31],ymm1[31] +; X32-AVX2-NEXT: vpunpckhbw {{.*#+}} ymm3 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31] +; X32-AVX2-NEXT: vpsraw $4, %ymm3, %ymm4 +; X32-AVX2-NEXT: vpblendvb %ymm2, %ymm4, %ymm3, %ymm3 +; X32-AVX2-NEXT: vpsraw $2, %ymm3, %ymm4 +; X32-AVX2-NEXT: vpaddw %ymm2, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpblendvb %ymm2, %ymm4, %ymm3, %ymm3 +; X32-AVX2-NEXT: vpsraw $1, %ymm3, %ymm4 +; X32-AVX2-NEXT: vpaddw %ymm2, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpblendvb %ymm2, %ymm4, %ymm3, %ymm2 +; X32-AVX2-NEXT: vpsrlw $8, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpunpcklbw {{.*#+}} ymm1 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[4],ymm1[4],ymm0[5],ymm1[5],ymm0[6],ymm1[6],ymm0[7],ymm1[7],ymm0[16],ymm1[16],ymm0[17],ymm1[17],ymm0[18],ymm1[18],ymm0[19],ymm1[19],ymm0[20],ymm1[20],ymm0[21],ymm1[21],ymm0[22],ymm1[22],ymm0[23],ymm1[23] +; X32-AVX2-NEXT: vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23] +; X32-AVX2-NEXT: vpsraw $4, %ymm0, %ymm3 +; X32-AVX2-NEXT: vpblendvb %ymm1, %ymm3, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsraw $2, %ymm0, %ymm3 +; X32-AVX2-NEXT: vpaddw %ymm1, %ymm1, %ymm1 +; X32-AVX2-NEXT: vpblendvb %ymm1, %ymm3, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsraw $1, %ymm0, %ymm3 +; X32-AVX2-NEXT: vpaddw %ymm1, %ymm1, %ymm1 +; X32-AVX2-NEXT: vpblendvb %ymm1, %ymm3, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsrlw $8, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpackuswb %ymm2, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %splat = shufflevector <32 x i8> %b, <32 x i8> undef, <32 x i32> zeroinitializer %shift = ashr <32 x i8> %a, %splat ret <32 x i8> %shift @@ -843,6 +1184,43 @@ define <4 x i64> @constant_shift_v4i64(<4 x i64> %a) nounwind { ; AVX512VL: # BB#0: ; AVX512VL-NEXT: vpsravq {{.*}}(%rip), %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: constant_shift_v4i64: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vmovdqa {{.*#+}} ymm1 = [1,0,7,0,31,0,62,0] +; X32-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2 +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [0,2147483648,0,2147483648] +; X32-AVX1-NEXT: vpsrlq %xmm2, %xmm3, %xmm4 +; X32-AVX1-NEXT: vpshufd {{.*#+}} xmm5 = xmm2[2,3,0,1] +; X32-AVX1-NEXT: vpsrlq %xmm5, %xmm3, %xmm6 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm4 = xmm4[0,1,2,3],xmm6[4,5,6,7] +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm6 +; X32-AVX1-NEXT: vpsrlq %xmm2, %xmm6, %xmm2 +; X32-AVX1-NEXT: vpsrlq %xmm5, %xmm6, %xmm5 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm5[4,5,6,7] +; X32-AVX1-NEXT: vpxor %xmm4, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsubq %xmm4, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsrlq %xmm1, %xmm3, %xmm4 +; X32-AVX1-NEXT: vpshufd {{.*#+}} xmm5 = xmm1[2,3,0,1] +; X32-AVX1-NEXT: vpsrlq %xmm5, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm4[0,1,2,3],xmm3[4,5,6,7] +; X32-AVX1-NEXT: vpsrlq %xmm1, %xmm0, %xmm1 +; X32-AVX1-NEXT: vpsrlq %xmm5, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm1[0,1,2,3],xmm0[4,5,6,7] +; X32-AVX1-NEXT: vpxor %xmm3, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsubq %xmm3, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: constant_shift_v4i64: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [1,0,7,0,31,0,62,0] +; X32-AVX2-NEXT: vmovdqa {{.*#+}} ymm2 = [0,2147483648,0,2147483648,0,2147483648,0,2147483648] +; X32-AVX2-NEXT: vpsrlvq %ymm1, %ymm2, %ymm3 +; X32-AVX2-NEXT: vpxor %ymm2, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsrlvq %ymm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsubq %ymm3, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = ashr <4 x i64> %a, ret <4 x i64> %shift } @@ -893,6 +1271,29 @@ define <8 x i32> @constant_shift_v8i32(<8 x i32> %a) nounwind { ; AVX512VL: # BB#0: ; AVX512VL-NEXT: vpsravd {{.*}}(%rip), %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: constant_shift_v8i32: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vpsrad $7, %xmm0, %xmm1 +; X32-AVX1-NEXT: vpsrad $5, %xmm0, %xmm2 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1,2,3],xmm1[4,5,6,7] +; X32-AVX1-NEXT: vpsrad $6, %xmm0, %xmm2 +; X32-AVX1-NEXT: vpsrad $4, %xmm0, %xmm3 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7] +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7] +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0 +; X32-AVX1-NEXT: vpsrad $7, %xmm0, %xmm2 +; X32-AVX1-NEXT: vpsrad $9, %xmm0, %xmm3 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7] +; X32-AVX1-NEXT: vpsrad $8, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3],xmm0[4,5],xmm2[6,7] +; X32-AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: constant_shift_v8i32: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpsravd {{\.LCPI.*}}, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = ashr <8 x i32> %a, ret <8 x i32> %shift } @@ -980,6 +1381,40 @@ define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) nounwind { ; AVX512BWVL: # BB#0: ; AVX512BWVL-NEXT: vpsravw {{.*}}(%rip), %ymm0, %ymm0 ; AVX512BWVL-NEXT: retq +; +; X32-AVX1-LABEL: constant_shift_v16i16: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1 +; X32-AVX1-NEXT: vpsraw $8, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpsraw $4, %xmm1, %xmm2 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm2[4,5,6,7] +; X32-AVX1-NEXT: vpsraw $2, %xmm1, %xmm2 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7] +; X32-AVX1-NEXT: vpsraw $1, %xmm1, %xmm2 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7] +; X32-AVX1-NEXT: vpsraw $4, %xmm0, %xmm2 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4,5,6,7] +; X32-AVX1-NEXT: vpsraw $2, %xmm0, %xmm2 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3],xmm0[4,5],xmm2[6,7] +; X32-AVX1-NEXT: vpsraw $1, %xmm0, %xmm2 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7] +; X32-AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: constant_shift_v16i16: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpxor %ymm1, %ymm1, %ymm1 +; X32-AVX2-NEXT: vmovdqa {{.*#+}} ymm2 = [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15] +; X32-AVX2-NEXT: vpunpckhwd {{.*#+}} ymm3 = ymm2[4],ymm1[4],ymm2[5],ymm1[5],ymm2[6],ymm1[6],ymm2[7],ymm1[7],ymm2[12],ymm1[12],ymm2[13],ymm1[13],ymm2[14],ymm1[14],ymm2[15],ymm1[15] +; X32-AVX2-NEXT: vpunpckhwd {{.*#+}} ymm4 = ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15] +; X32-AVX2-NEXT: vpsravd %ymm3, %ymm4, %ymm3 +; X32-AVX2-NEXT: vpsrld $16, %ymm3, %ymm3 +; X32-AVX2-NEXT: vpunpcklwd {{.*#+}} ymm2 = ymm2[0],ymm1[0],ymm2[1],ymm1[1],ymm2[2],ymm1[2],ymm2[3],ymm1[3],ymm2[8],ymm1[8],ymm2[9],ymm1[9],ymm2[10],ymm1[10],ymm2[11],ymm1[11] +; X32-AVX2-NEXT: vpunpcklwd {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11] +; X32-AVX2-NEXT: vpsravd %ymm2, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsrld $16, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpackusdw %ymm3, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = ashr <16 x i16> %a, ret <16 x i16> %shift } @@ -1149,6 +1584,81 @@ define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) nounwind { ; AVX512BWVL-NEXT: vpsravw {{.*}}(%rip), %zmm0, %zmm0 ; AVX512BWVL-NEXT: vpmovwb %zmm0, %ymm0 ; AVX512BWVL-NEXT: retq +; +; X32-AVX1-LABEL: constant_shift_v32i8: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm1 = [8192,24640,41088,57536,49376,32928,16480,32] +; X32-AVX1-NEXT: vpunpckhbw {{.*#+}} xmm2 = xmm0[8],xmm1[8],xmm0[9],xmm1[9],xmm0[10],xmm1[10],xmm0[11],xmm1[11],xmm0[12],xmm1[12],xmm0[13],xmm1[13],xmm0[14],xmm1[14],xmm0[15],xmm1[15] +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3 +; X32-AVX1-NEXT: vpunpckhbw {{.*#+}} xmm4 = xmm0[8],xmm3[8],xmm0[9],xmm3[9],xmm0[10],xmm3[10],xmm0[11],xmm3[11],xmm0[12],xmm3[12],xmm0[13],xmm3[13],xmm0[14],xmm3[14],xmm0[15],xmm3[15] +; X32-AVX1-NEXT: vpsraw $4, %xmm4, %xmm5 +; X32-AVX1-NEXT: vpblendvb %xmm2, %xmm5, %xmm4, %xmm4 +; X32-AVX1-NEXT: vpunpckhbw {{.*#+}} xmm5 = xmm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15] +; X32-AVX1-NEXT: vpsraw $4, %xmm5, %xmm6 +; X32-AVX1-NEXT: vpblendvb %xmm2, %xmm6, %xmm5, %xmm5 +; X32-AVX1-NEXT: vpsraw $2, %xmm4, %xmm6 +; X32-AVX1-NEXT: vpaddw %xmm2, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpblendvb %xmm2, %xmm6, %xmm4, %xmm4 +; X32-AVX1-NEXT: vpsraw $2, %xmm5, %xmm6 +; X32-AVX1-NEXT: vpblendvb %xmm2, %xmm6, %xmm5, %xmm5 +; X32-AVX1-NEXT: vpsraw $1, %xmm4, %xmm6 +; X32-AVX1-NEXT: vpaddw %xmm2, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpblendvb %xmm2, %xmm6, %xmm4, %xmm4 +; X32-AVX1-NEXT: vpsraw $1, %xmm5, %xmm6 +; X32-AVX1-NEXT: vpblendvb %xmm2, %xmm6, %xmm5, %xmm2 +; X32-AVX1-NEXT: vpunpcklbw {{.*#+}} xmm1 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7] +; X32-AVX1-NEXT: vpunpcklbw {{.*#+}} xmm3 = xmm0[0],xmm3[0],xmm0[1],xmm3[1],xmm0[2],xmm3[2],xmm0[3],xmm3[3],xmm0[4],xmm3[4],xmm0[5],xmm3[5],xmm0[6],xmm3[6],xmm0[7],xmm3[7] +; X32-AVX1-NEXT: vpsraw $4, %xmm3, %xmm5 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm5, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpunpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7] +; X32-AVX1-NEXT: vpsraw $4, %xmm0, %xmm5 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm5, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsraw $2, %xmm3, %xmm5 +; X32-AVX1-NEXT: vpaddw %xmm1, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm5, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpsraw $2, %xmm0, %xmm5 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm5, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsraw $1, %xmm3, %xmm5 +; X32-AVX1-NEXT: vpaddw %xmm1, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm5, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpsraw $1, %xmm0, %xmm5 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm5, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsrlw $8, %xmm4, %xmm1 +; X32-AVX1-NEXT: vpsrlw $8, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpackuswb %xmm1, %xmm3, %xmm1 +; X32-AVX1-NEXT: vpsrlw $8, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsrlw $8, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpackuswb %xmm2, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: constant_shift_v32i8: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [8192,24640,41088,57536,49376,32928,16480,32,8192,24640,41088,57536,49376,32928,16480,32] +; X32-AVX2-NEXT: vpunpckhbw {{.*#+}} ymm2 = ymm0[8],ymm1[8],ymm0[9],ymm1[9],ymm0[10],ymm1[10],ymm0[11],ymm1[11],ymm0[12],ymm1[12],ymm0[13],ymm1[13],ymm0[14],ymm1[14],ymm0[15],ymm1[15],ymm0[24],ymm1[24],ymm0[25],ymm1[25],ymm0[26],ymm1[26],ymm0[27],ymm1[27],ymm0[28],ymm1[28],ymm0[29],ymm1[29],ymm0[30],ymm1[30],ymm0[31],ymm1[31] +; X32-AVX2-NEXT: vpunpckhbw {{.*#+}} ymm3 = ymm0[8,8,9,9,10,10,11,11,12,12,13,13,14,14,15,15,24,24,25,25,26,26,27,27,28,28,29,29,30,30,31,31] +; X32-AVX2-NEXT: vpsraw $4, %ymm3, %ymm4 +; X32-AVX2-NEXT: vpblendvb %ymm2, %ymm4, %ymm3, %ymm3 +; X32-AVX2-NEXT: vpsraw $2, %ymm3, %ymm4 +; X32-AVX2-NEXT: vpaddw %ymm2, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpblendvb %ymm2, %ymm4, %ymm3, %ymm3 +; X32-AVX2-NEXT: vpsraw $1, %ymm3, %ymm4 +; X32-AVX2-NEXT: vpaddw %ymm2, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpblendvb %ymm2, %ymm4, %ymm3, %ymm2 +; X32-AVX2-NEXT: vpsrlw $8, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpunpcklbw {{.*#+}} ymm1 = ymm0[0],ymm1[0],ymm0[1],ymm1[1],ymm0[2],ymm1[2],ymm0[3],ymm1[3],ymm0[4],ymm1[4],ymm0[5],ymm1[5],ymm0[6],ymm1[6],ymm0[7],ymm1[7],ymm0[16],ymm1[16],ymm0[17],ymm1[17],ymm0[18],ymm1[18],ymm0[19],ymm1[19],ymm0[20],ymm1[20],ymm0[21],ymm1[21],ymm0[22],ymm1[22],ymm0[23],ymm1[23] +; X32-AVX2-NEXT: vpunpcklbw {{.*#+}} ymm0 = ymm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7,16,16,17,17,18,18,19,19,20,20,21,21,22,22,23,23] +; X32-AVX2-NEXT: vpsraw $4, %ymm0, %ymm3 +; X32-AVX2-NEXT: vpblendvb %ymm1, %ymm3, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsraw $2, %ymm0, %ymm3 +; X32-AVX2-NEXT: vpaddw %ymm1, %ymm1, %ymm1 +; X32-AVX2-NEXT: vpblendvb %ymm1, %ymm3, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsraw $1, %ymm0, %ymm3 +; X32-AVX2-NEXT: vpaddw %ymm1, %ymm1, %ymm1 +; X32-AVX2-NEXT: vpblendvb %ymm1, %ymm3, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsrlw $8, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpackuswb %ymm2, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = ashr <32 x i8> %a, ret <32 x i8> %shift } @@ -1206,6 +1716,30 @@ define <4 x i64> @splatconstant_shift_v4i64(<4 x i64> %a) nounwind { ; AVX512VL: # BB#0: ; AVX512VL-NEXT: vpsraq $7, %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: splatconstant_shift_v4i64: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vmovdqa {{.*#+}} ymm1 = [7,0,7,0,7,0,7,0] +; X32-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2 +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [0,2147483648,0,2147483648] +; X32-AVX1-NEXT: vpsrlq %xmm2, %xmm3, %xmm4 +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm5 +; X32-AVX1-NEXT: vpsrlq %xmm2, %xmm5, %xmm2 +; X32-AVX1-NEXT: vpxor %xmm4, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsubq %xmm4, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsrlq %xmm1, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpsrlq %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpxor %xmm3, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsubq %xmm3, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: splatconstant_shift_v4i64: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpsrad $7, %ymm0, %ymm1 +; X32-AVX2-NEXT: vpsrlq $7, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0],ymm1[1],ymm0[2],ymm1[3],ymm0[4],ymm1[5],ymm0[6],ymm1[7] +; X32-AVX2-NEXT: retl %shift = ashr <4 x i64> %a, ret <4 x i64> %shift } @@ -1246,6 +1780,19 @@ define <8 x i32> @splatconstant_shift_v8i32(<8 x i32> %a) nounwind { ; AVX512VL: # BB#0: ; AVX512VL-NEXT: vpsrad $5, %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: splatconstant_shift_v8i32: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vpsrad $5, %xmm0, %xmm1 +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0 +; X32-AVX1-NEXT: vpsrad $5, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: splatconstant_shift_v8i32: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpsrad $5, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = ashr <8 x i32> %a, ret <8 x i32> %shift } @@ -1286,6 +1833,19 @@ define <16 x i16> @splatconstant_shift_v16i16(<16 x i16> %a) nounwind { ; AVX512VL: # BB#0: ; AVX512VL-NEXT: vpsraw $3, %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: splatconstant_shift_v16i16: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vpsraw $3, %xmm0, %xmm1 +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0 +; X32-AVX1-NEXT: vpsraw $3, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: splatconstant_shift_v16i16: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpsraw $3, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = ashr <16 x i16> %a, ret <16 x i16> %shift } @@ -1352,6 +1912,31 @@ define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) nounwind { ; AVX512VL-NEXT: vpxor %ymm1, %ymm0, %ymm0 ; AVX512VL-NEXT: vpsubb %ymm1, %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: splatconstant_shift_v32i8: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1 +; X32-AVX1-NEXT: vpsrlw $3, %xmm1, %xmm1 +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31] +; X32-AVX1-NEXT: vpand %xmm2, %xmm1, %xmm1 +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16] +; X32-AVX1-NEXT: vpxor %xmm3, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpsubb %xmm3, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpsrlw $3, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpxor %xmm3, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsubb %xmm3, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: splatconstant_shift_v32i8: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpsrlw $3, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpand {{\.LCPI.*}}, %ymm0, %ymm0 +; X32-AVX2-NEXT: vmovdqa {{.*#+}} ymm1 = [16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16,16] +; X32-AVX2-NEXT: vpxor %ymm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsubb %ymm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = ashr <32 x i8> %a, ret <32 x i8> %shift } diff --git a/llvm/test/CodeGen/X86/vector-shift-lshr-256.ll b/llvm/test/CodeGen/X86/vector-shift-lshr-256.ll index 60575250d713..d342292f253a 100644 --- a/llvm/test/CodeGen/X86/vector-shift-lshr-256.ll +++ b/llvm/test/CodeGen/X86/vector-shift-lshr-256.ll @@ -7,6 +7,10 @@ ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=knl -mattr=+avx512bw | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512BW ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=knl -mattr=+avx512dq,+avx512vl | FileCheck %s --check-prefix=ALL --check-prefix=AVX512VL --check-prefix=AVX512DQVL ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=knl -mattr=+avx512bw,+avx512vl | FileCheck %s --check-prefix=ALL --check-prefix=AVX512VL --check-prefix=AVX512BWVL +; +; 32-bit runs to make sure we do reasonable things for i64 shifts. +; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=ALL --check-prefix=X32-AVX --check-prefix=X32-AVX1 +; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=ALL --check-prefix=X32-AVX --check-prefix=X32-AVX2 ; ; Variable Shifts @@ -59,6 +63,26 @@ define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) nounwind { ; AVX512VL: # BB#0: ; AVX512VL-NEXT: vpsrlvq %ymm1, %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: var_shift_v4i64: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2 +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3 +; X32-AVX1-NEXT: vpsrlq %xmm2, %xmm3, %xmm4 +; X32-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[2,3,0,1] +; X32-AVX1-NEXT: vpsrlq %xmm2, %xmm3, %xmm2 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm4[0,1,2,3],xmm2[4,5,6,7] +; X32-AVX1-NEXT: vpsrlq %xmm1, %xmm0, %xmm3 +; X32-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; X32-AVX1-NEXT: vpsrlq %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm3[0,1,2,3],xmm0[4,5,6,7] +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: var_shift_v4i64: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpsrlvq %ymm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = lshr <4 x i64> %a, %b ret <4 x i64> %shift } @@ -125,6 +149,41 @@ define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind { ; AVX512VL: # BB#0: ; AVX512VL-NEXT: vpsrlvd %ymm1, %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: var_shift_v8i32: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2 +; X32-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm3 +; X32-AVX1-NEXT: vpsrldq {{.*#+}} xmm4 = xmm3[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero +; X32-AVX1-NEXT: vpsrld %xmm4, %xmm2, %xmm4 +; X32-AVX1-NEXT: vpsrlq $32, %xmm3, %xmm5 +; X32-AVX1-NEXT: vpsrld %xmm5, %xmm2, %xmm5 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm4 = xmm5[0,1,2,3],xmm4[4,5,6,7] +; X32-AVX1-NEXT: vpxor %xmm5, %xmm5, %xmm5 +; X32-AVX1-NEXT: vpunpckhdq {{.*#+}} xmm6 = xmm3[2],xmm5[2],xmm3[3],xmm5[3] +; X32-AVX1-NEXT: vpsrld %xmm6, %xmm2, %xmm6 +; X32-AVX1-NEXT: vpmovzxdq {{.*#+}} xmm3 = xmm3[0],zero,xmm3[1],zero +; X32-AVX1-NEXT: vpsrld %xmm3, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1,2,3],xmm6[4,5,6,7] +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm2[0,1],xmm4[2,3],xmm2[4,5],xmm4[6,7] +; X32-AVX1-NEXT: vpsrldq {{.*#+}} xmm3 = xmm1[12,13,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero +; X32-AVX1-NEXT: vpsrld %xmm3, %xmm0, %xmm3 +; X32-AVX1-NEXT: vpsrlq $32, %xmm1, %xmm4 +; X32-AVX1-NEXT: vpsrld %xmm4, %xmm0, %xmm4 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm3 = xmm4[0,1,2,3],xmm3[4,5,6,7] +; X32-AVX1-NEXT: vpunpckhdq {{.*#+}} xmm4 = xmm1[2],xmm5[2],xmm1[3],xmm5[3] +; X32-AVX1-NEXT: vpsrld %xmm4, %xmm0, %xmm4 +; X32-AVX1-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero +; X32-AVX1-NEXT: vpsrld %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm4[4,5,6,7] +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm3[2,3],xmm0[4,5],xmm3[6,7] +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: var_shift_v8i32: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpsrlvd %ymm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = lshr <8 x i32> %a, %b ret <8 x i32> %shift } @@ -231,6 +290,55 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind { ; AVX512BWVL: # BB#0: ; AVX512BWVL-NEXT: vpsrlvw %ymm1, %ymm0, %ymm0 ; AVX512BWVL-NEXT: retq +; +; X32-AVX1-LABEL: var_shift_v16i16: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2 +; X32-AVX1-NEXT: vpsllw $12, %xmm2, %xmm3 +; X32-AVX1-NEXT: vpsllw $4, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpor %xmm3, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpaddw %xmm2, %xmm2, %xmm3 +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm4 +; X32-AVX1-NEXT: vpsrlw $8, %xmm4, %xmm5 +; X32-AVX1-NEXT: vpblendvb %xmm2, %xmm5, %xmm4, %xmm2 +; X32-AVX1-NEXT: vpsrlw $4, %xmm2, %xmm4 +; X32-AVX1-NEXT: vpblendvb %xmm3, %xmm4, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsrlw $2, %xmm2, %xmm4 +; X32-AVX1-NEXT: vpaddw %xmm3, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpblendvb %xmm3, %xmm4, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsrlw $1, %xmm2, %xmm4 +; X32-AVX1-NEXT: vpaddw %xmm3, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpblendvb %xmm3, %xmm4, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsllw $12, %xmm1, %xmm3 +; X32-AVX1-NEXT: vpsllw $4, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpor %xmm3, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpaddw %xmm1, %xmm1, %xmm3 +; X32-AVX1-NEXT: vpsrlw $8, %xmm0, %xmm4 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm4, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsrlw $4, %xmm0, %xmm1 +; X32-AVX1-NEXT: vpblendvb %xmm3, %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsrlw $2, %xmm0, %xmm1 +; X32-AVX1-NEXT: vpaddw %xmm3, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpblendvb %xmm3, %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsrlw $1, %xmm0, %xmm1 +; X32-AVX1-NEXT: vpaddw %xmm3, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpblendvb %xmm3, %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: var_shift_v16i16: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpxor %ymm2, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpunpckhwd {{.*#+}} ymm3 = ymm1[4],ymm2[4],ymm1[5],ymm2[5],ymm1[6],ymm2[6],ymm1[7],ymm2[7],ymm1[12],ymm2[12],ymm1[13],ymm2[13],ymm1[14],ymm2[14],ymm1[15],ymm2[15] +; X32-AVX2-NEXT: vpunpckhwd {{.*#+}} ymm4 = ymm2[4],ymm0[4],ymm2[5],ymm0[5],ymm2[6],ymm0[6],ymm2[7],ymm0[7],ymm2[12],ymm0[12],ymm2[13],ymm0[13],ymm2[14],ymm0[14],ymm2[15],ymm0[15] +; X32-AVX2-NEXT: vpsrlvd %ymm3, %ymm4, %ymm3 +; X32-AVX2-NEXT: vpsrld $16, %ymm3, %ymm3 +; X32-AVX2-NEXT: vpunpcklwd {{.*#+}} ymm1 = ymm1[0],ymm2[0],ymm1[1],ymm2[1],ymm1[2],ymm2[2],ymm1[3],ymm2[3],ymm1[8],ymm2[8],ymm1[9],ymm2[9],ymm1[10],ymm2[10],ymm1[11],ymm2[11] +; X32-AVX2-NEXT: vpunpcklwd {{.*#+}} ymm0 = ymm2[0],ymm0[0],ymm2[1],ymm0[1],ymm2[2],ymm0[2],ymm2[3],ymm0[3],ymm2[8],ymm0[8],ymm2[9],ymm0[9],ymm2[10],ymm0[10],ymm2[11],ymm0[11] +; X32-AVX2-NEXT: vpsrlvd %ymm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsrld $16, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpackusdw %ymm3, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = lshr <16 x i16> %a, %b ret <16 x i16> %shift } @@ -357,6 +465,56 @@ define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind { ; AVX512BWVL-NEXT: vpsrlvw %zmm1, %zmm0, %zmm0 ; AVX512BWVL-NEXT: vpmovwb %zmm0, %ymm0 ; AVX512BWVL-NEXT: retq +; +; X32-AVX1-LABEL: var_shift_v32i8: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2 +; X32-AVX1-NEXT: vpsrlw $4, %xmm2, %xmm3 +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] +; X32-AVX1-NEXT: vpand %xmm4, %xmm3, %xmm3 +; X32-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm5 +; X32-AVX1-NEXT: vpsllw $5, %xmm5, %xmm5 +; X32-AVX1-NEXT: vpblendvb %xmm5, %xmm3, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsrlw $2, %xmm2, %xmm3 +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm6 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63] +; X32-AVX1-NEXT: vpand %xmm6, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpaddb %xmm5, %xmm5, %xmm5 +; X32-AVX1-NEXT: vpblendvb %xmm5, %xmm3, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsrlw $1, %xmm2, %xmm3 +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm7 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127] +; X32-AVX1-NEXT: vpand %xmm7, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpaddb %xmm5, %xmm5, %xmm5 +; X32-AVX1-NEXT: vpblendvb %xmm5, %xmm3, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsrlw $4, %xmm0, %xmm3 +; X32-AVX1-NEXT: vpand %xmm4, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpsllw $5, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm3, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsrlw $2, %xmm0, %xmm3 +; X32-AVX1-NEXT: vpand %xmm6, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpaddb %xmm1, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm3, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsrlw $1, %xmm0, %xmm3 +; X32-AVX1-NEXT: vpand %xmm7, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpaddb %xmm1, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm3, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: var_shift_v32i8: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpsllw $5, %ymm1, %ymm1 +; X32-AVX2-NEXT: vpsrlw $4, %ymm0, %ymm2 +; X32-AVX2-NEXT: vpand {{\.LCPI.*}}, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpblendvb %ymm1, %ymm2, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsrlw $2, %ymm0, %ymm2 +; X32-AVX2-NEXT: vpand {{\.LCPI.*}}, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpaddb %ymm1, %ymm1, %ymm1 +; X32-AVX2-NEXT: vpblendvb %ymm1, %ymm2, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsrlw $1, %ymm0, %ymm2 +; X32-AVX2-NEXT: vpand {{\.LCPI.*}}, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpaddb %ymm1, %ymm1, %ymm1 +; X32-AVX2-NEXT: vpblendvb %ymm1, %ymm2, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = lshr <32 x i8> %a, %b ret <32 x i8> %shift } @@ -401,6 +559,23 @@ define <4 x i64> @splatvar_shift_v4i64(<4 x i64> %a, <4 x i64> %b) nounwind { ; AVX512VL: # BB#0: ; AVX512VL-NEXT: vpsrlq %xmm1, %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: splatvar_shift_v4i64: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vpextrd $1, %xmm1, %eax +; X32-AVX1-NEXT: vpinsrd $1, %eax, %xmm1, %xmm1 +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2 +; X32-AVX1-NEXT: vpsrlq %xmm1, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsrlq %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: splatvar_shift_v4i64: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpextrd $1, %xmm1, %eax +; X32-AVX2-NEXT: vpinsrd $1, %eax, %xmm1, %xmm1 +; X32-AVX2-NEXT: vpsrlq %xmm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %splat = shufflevector <4 x i64> %b, <4 x i64> undef, <4 x i32> zeroinitializer %shift = lshr <4 x i64> %a, %splat ret <4 x i64> %shift @@ -448,6 +623,21 @@ define <8 x i32> @splatvar_shift_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind { ; AVX512VL-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero ; AVX512VL-NEXT: vpsrld %xmm1, %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: splatvar_shift_v8i32: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2 +; X32-AVX1-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero +; X32-AVX1-NEXT: vpsrld %xmm1, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsrld %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: splatvar_shift_v8i32: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero +; X32-AVX2-NEXT: vpsrld %xmm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %splat = shufflevector <8 x i32> %b, <8 x i32> undef, <8 x i32> zeroinitializer %shift = lshr <8 x i32> %a, %splat ret <8 x i32> %shift @@ -495,6 +685,21 @@ define <16 x i16> @splatvar_shift_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind ; AVX512VL-NEXT: vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero ; AVX512VL-NEXT: vpsrlw %xmm1, %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: splatvar_shift_v16i16: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2 +; X32-AVX1-NEXT: vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero +; X32-AVX1-NEXT: vpsrlw %xmm1, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsrlw %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: splatvar_shift_v16i16: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero +; X32-AVX2-NEXT: vpsrlw %xmm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %splat = shufflevector <16 x i16> %b, <16 x i16> undef, <16 x i32> zeroinitializer %shift = lshr <16 x i16> %a, %splat ret <16 x i16> %shift @@ -625,6 +830,55 @@ define <32 x i8> @splatvar_shift_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind { ; AVX512BWVL-NEXT: vpsrlvw %zmm1, %zmm0, %zmm0 ; AVX512BWVL-NEXT: vpmovwb %zmm0, %ymm0 ; AVX512BWVL-NEXT: retq +; +; X32-AVX1-LABEL: splatvar_shift_v32i8: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpshufb %xmm2, %xmm1, %xmm1 +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2 +; X32-AVX1-NEXT: vpsrlw $4, %xmm2, %xmm3 +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] +; X32-AVX1-NEXT: vpand %xmm4, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpsllw $5, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm3, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsrlw $4, %xmm0, %xmm3 +; X32-AVX1-NEXT: vpand %xmm4, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpsrlw $2, %xmm2, %xmm4 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm3, %xmm0, %xmm0 +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63] +; X32-AVX1-NEXT: vpand %xmm3, %xmm4, %xmm4 +; X32-AVX1-NEXT: vpaddb %xmm1, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm4, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsrlw $2, %xmm0, %xmm4 +; X32-AVX1-NEXT: vpand %xmm3, %xmm4, %xmm3 +; X32-AVX1-NEXT: vpsrlw $1, %xmm2, %xmm4 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm3, %xmm0, %xmm0 +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127] +; X32-AVX1-NEXT: vpand %xmm3, %xmm4, %xmm4 +; X32-AVX1-NEXT: vpaddb %xmm1, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm4, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsrlw $1, %xmm0, %xmm4 +; X32-AVX1-NEXT: vpand %xmm3, %xmm4, %xmm3 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm3, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: splatvar_shift_v32i8: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpbroadcastb %xmm1, %ymm1 +; X32-AVX2-NEXT: vpsrlw $4, %ymm0, %ymm2 +; X32-AVX2-NEXT: vpand {{\.LCPI.*}}, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpsllw $5, %ymm1, %ymm1 +; X32-AVX2-NEXT: vpblendvb %ymm1, %ymm2, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsrlw $2, %ymm0, %ymm2 +; X32-AVX2-NEXT: vpand {{\.LCPI.*}}, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpaddb %ymm1, %ymm1, %ymm1 +; X32-AVX2-NEXT: vpblendvb %ymm1, %ymm2, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsrlw $1, %ymm0, %ymm2 +; X32-AVX2-NEXT: vpand {{\.LCPI.*}}, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpaddb %ymm1, %ymm1, %ymm1 +; X32-AVX2-NEXT: vpblendvb %ymm1, %ymm2, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %splat = shufflevector <32 x i8> %b, <32 x i8> undef, <32 x i32> zeroinitializer %shift = lshr <32 x i8> %a, %splat ret <32 x i8> %shift @@ -677,6 +931,27 @@ define <4 x i64> @constant_shift_v4i64(<4 x i64> %a) nounwind { ; AVX512VL: # BB#0: ; AVX512VL-NEXT: vpsrlvq {{.*}}(%rip), %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: constant_shift_v4i64: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vmovdqa {{.*#+}} ymm1 = [1,0,7,0,31,0,62,0] +; X32-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2 +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3 +; X32-AVX1-NEXT: vpsrlq %xmm2, %xmm3, %xmm4 +; X32-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[2,3,0,1] +; X32-AVX1-NEXT: vpsrlq %xmm2, %xmm3, %xmm2 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm4[0,1,2,3],xmm2[4,5,6,7] +; X32-AVX1-NEXT: vpsrlq %xmm1, %xmm0, %xmm3 +; X32-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; X32-AVX1-NEXT: vpsrlq %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm3[0,1,2,3],xmm0[4,5,6,7] +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: constant_shift_v4i64: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpsrlvq {{\.LCPI.*}}, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = lshr <4 x i64> %a, ret <4 x i64> %shift } @@ -727,6 +1002,29 @@ define <8 x i32> @constant_shift_v8i32(<8 x i32> %a) nounwind { ; AVX512VL: # BB#0: ; AVX512VL-NEXT: vpsrlvd {{.*}}(%rip), %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: constant_shift_v8i32: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vpsrld $7, %xmm0, %xmm1 +; X32-AVX1-NEXT: vpsrld $5, %xmm0, %xmm2 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1,2,3],xmm1[4,5,6,7] +; X32-AVX1-NEXT: vpsrld $6, %xmm0, %xmm2 +; X32-AVX1-NEXT: vpsrld $4, %xmm0, %xmm3 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7] +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm2[0,1],xmm1[2,3],xmm2[4,5],xmm1[6,7] +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0 +; X32-AVX1-NEXT: vpsrld $7, %xmm0, %xmm2 +; X32-AVX1-NEXT: vpsrld $9, %xmm0, %xmm3 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm3[0,1,2,3],xmm2[4,5,6,7] +; X32-AVX1-NEXT: vpsrld $8, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3],xmm0[4,5],xmm2[6,7] +; X32-AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: constant_shift_v8i32: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpsrlvd {{\.LCPI.*}}, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = lshr <8 x i32> %a, ret <8 x i32> %shift } @@ -814,6 +1112,40 @@ define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) nounwind { ; AVX512BWVL: # BB#0: ; AVX512BWVL-NEXT: vpsrlvw {{.*}}(%rip), %ymm0, %ymm0 ; AVX512BWVL-NEXT: retq +; +; X32-AVX1-LABEL: constant_shift_v16i16: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1 +; X32-AVX1-NEXT: vpsrlw $8, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpsrlw $4, %xmm1, %xmm2 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1,2,3],xmm2[4,5,6,7] +; X32-AVX1-NEXT: vpsrlw $2, %xmm1, %xmm2 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7] +; X32-AVX1-NEXT: vpsrlw $1, %xmm1, %xmm2 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm1 = xmm1[0],xmm2[1],xmm1[2],xmm2[3],xmm1[4],xmm2[5],xmm1[6],xmm2[7] +; X32-AVX1-NEXT: vpsrlw $4, %xmm0, %xmm2 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1,2,3],xmm2[4,5,6,7] +; X32-AVX1-NEXT: vpsrlw $2, %xmm0, %xmm2 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm2[2,3],xmm0[4,5],xmm2[6,7] +; X32-AVX1-NEXT: vpsrlw $1, %xmm0, %xmm2 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm0[0],xmm2[1],xmm0[2],xmm2[3],xmm0[4],xmm2[5],xmm0[6],xmm2[7] +; X32-AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: constant_shift_v16i16: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpxor %ymm1, %ymm1, %ymm1 +; X32-AVX2-NEXT: vmovdqa {{.*#+}} ymm2 = [0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15] +; X32-AVX2-NEXT: vpunpckhwd {{.*#+}} ymm3 = ymm2[4],ymm1[4],ymm2[5],ymm1[5],ymm2[6],ymm1[6],ymm2[7],ymm1[7],ymm2[12],ymm1[12],ymm2[13],ymm1[13],ymm2[14],ymm1[14],ymm2[15],ymm1[15] +; X32-AVX2-NEXT: vpunpckhwd {{.*#+}} ymm4 = ymm1[4],ymm0[4],ymm1[5],ymm0[5],ymm1[6],ymm0[6],ymm1[7],ymm0[7],ymm1[12],ymm0[12],ymm1[13],ymm0[13],ymm1[14],ymm0[14],ymm1[15],ymm0[15] +; X32-AVX2-NEXT: vpsrlvd %ymm3, %ymm4, %ymm3 +; X32-AVX2-NEXT: vpsrld $16, %ymm3, %ymm3 +; X32-AVX2-NEXT: vpunpcklwd {{.*#+}} ymm2 = ymm2[0],ymm1[0],ymm2[1],ymm1[1],ymm2[2],ymm1[2],ymm2[3],ymm1[3],ymm2[8],ymm1[8],ymm2[9],ymm1[9],ymm2[10],ymm1[10],ymm2[11],ymm1[11] +; X32-AVX2-NEXT: vpunpcklwd {{.*#+}} ymm0 = ymm1[0],ymm0[0],ymm1[1],ymm0[1],ymm1[2],ymm0[2],ymm1[3],ymm0[3],ymm1[8],ymm0[8],ymm1[9],ymm0[9],ymm1[10],ymm0[10],ymm1[11],ymm0[11] +; X32-AVX2-NEXT: vpsrlvd %ymm2, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsrld $16, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpackusdw %ymm3, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = lshr <16 x i16> %a, ret <16 x i16> %shift } @@ -930,6 +1262,52 @@ define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) nounwind { ; AVX512BWVL-NEXT: vpsrlvw {{.*}}(%rip), %zmm0, %zmm0 ; AVX512BWVL-NEXT: vpmovwb %zmm0, %ymm0 ; AVX512BWVL-NEXT: retq +; +; X32-AVX1-LABEL: constant_shift_v32i8: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1 +; X32-AVX1-NEXT: vpsrlw $4, %xmm1, %xmm2 +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [15,15,15,15,15,15,15,15,15,15,15,15,15,15,15,15] +; X32-AVX1-NEXT: vpand %xmm3, %xmm2, %xmm2 +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [8192,24640,41088,57536,49376,32928,16480,32] +; X32-AVX1-NEXT: vpblendvb %xmm4, %xmm2, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpsrlw $4, %xmm0, %xmm2 +; X32-AVX1-NEXT: vpand %xmm3, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsrlw $2, %xmm1, %xmm3 +; X32-AVX1-NEXT: vpblendvb %xmm4, %xmm2, %xmm0, %xmm0 +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [63,63,63,63,63,63,63,63,63,63,63,63,63,63,63,63] +; X32-AVX1-NEXT: vpand %xmm2, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpaddb %xmm4, %xmm4, %xmm4 +; X32-AVX1-NEXT: vpblendvb %xmm4, %xmm3, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpsrlw $2, %xmm0, %xmm3 +; X32-AVX1-NEXT: vpand %xmm2, %xmm3, %xmm2 +; X32-AVX1-NEXT: vpsrlw $1, %xmm1, %xmm3 +; X32-AVX1-NEXT: vpblendvb %xmm4, %xmm2, %xmm0, %xmm0 +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [127,127,127,127,127,127,127,127,127,127,127,127,127,127,127,127] +; X32-AVX1-NEXT: vpand %xmm2, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpaddb %xmm4, %xmm4, %xmm4 +; X32-AVX1-NEXT: vpblendvb %xmm4, %xmm3, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpsrlw $1, %xmm0, %xmm3 +; X32-AVX1-NEXT: vpand %xmm2, %xmm3, %xmm2 +; X32-AVX1-NEXT: vpblendvb %xmm4, %xmm2, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: constant_shift_v32i8: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpsrlw $4, %ymm0, %ymm1 +; X32-AVX2-NEXT: vpand {{\.LCPI.*}}, %ymm1, %ymm1 +; X32-AVX2-NEXT: vmovdqa {{.*#+}} ymm2 = [8192,24640,41088,57536,49376,32928,16480,32,8192,24640,41088,57536,49376,32928,16480,32] +; X32-AVX2-NEXT: vpblendvb %ymm2, %ymm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsrlw $2, %ymm0, %ymm1 +; X32-AVX2-NEXT: vpand {{\.LCPI.*}}, %ymm1, %ymm1 +; X32-AVX2-NEXT: vpaddb %ymm2, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpblendvb %ymm2, %ymm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsrlw $1, %ymm0, %ymm1 +; X32-AVX2-NEXT: vpand {{\.LCPI.*}}, %ymm1, %ymm1 +; X32-AVX2-NEXT: vpaddb %ymm2, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpblendvb %ymm2, %ymm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = lshr <32 x i8> %a, ret <32 x i8> %shift } @@ -974,6 +1352,21 @@ define <4 x i64> @splatconstant_shift_v4i64(<4 x i64> %a) nounwind { ; AVX512VL: # BB#0: ; AVX512VL-NEXT: vpsrlq $7, %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: splatconstant_shift_v4i64: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vmovdqa {{.*#+}} ymm1 = [7,0,7,0,7,0,7,0] +; X32-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2 +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3 +; X32-AVX1-NEXT: vpsrlq %xmm2, %xmm3, %xmm2 +; X32-AVX1-NEXT: vpsrlq %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: splatconstant_shift_v4i64: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpsrlq $7, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = lshr <4 x i64> %a, ret <4 x i64> %shift } @@ -1014,6 +1407,19 @@ define <8 x i32> @splatconstant_shift_v8i32(<8 x i32> %a) nounwind { ; AVX512VL: # BB#0: ; AVX512VL-NEXT: vpsrld $5, %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: splatconstant_shift_v8i32: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vpsrld $5, %xmm0, %xmm1 +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0 +; X32-AVX1-NEXT: vpsrld $5, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: splatconstant_shift_v8i32: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpsrld $5, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = lshr <8 x i32> %a, ret <8 x i32> %shift } @@ -1054,6 +1460,19 @@ define <16 x i16> @splatconstant_shift_v16i16(<16 x i16> %a) nounwind { ; AVX512VL: # BB#0: ; AVX512VL-NEXT: vpsrlw $3, %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: splatconstant_shift_v16i16: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vpsrlw $3, %xmm0, %xmm1 +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0 +; X32-AVX1-NEXT: vpsrlw $3, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: splatconstant_shift_v16i16: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpsrlw $3, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = lshr <16 x i16> %a, ret <16 x i16> %shift } @@ -1103,6 +1522,23 @@ define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) nounwind { ; AVX512VL-NEXT: vpsrlw $3, %ymm0, %ymm0 ; AVX512VL-NEXT: vpand {{.*}}(%rip), %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: splatconstant_shift_v32i8: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1 +; X32-AVX1-NEXT: vpsrlw $3, %xmm1, %xmm1 +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [31,31,31,31,31,31,31,31,31,31,31,31,31,31,31,31] +; X32-AVX1-NEXT: vpand %xmm2, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpsrlw $3, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: splatconstant_shift_v32i8: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpsrlw $3, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpand {{\.LCPI.*}}, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = lshr <32 x i8> %a, ret <32 x i8> %shift } diff --git a/llvm/test/CodeGen/X86/vector-shift-shl-256.ll b/llvm/test/CodeGen/X86/vector-shift-shl-256.ll index 7f534050b6a7..4e5ff0e9ceec 100644 --- a/llvm/test/CodeGen/X86/vector-shift-shl-256.ll +++ b/llvm/test/CodeGen/X86/vector-shift-shl-256.ll @@ -7,6 +7,10 @@ ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=knl -mattr=+avx512bw | FileCheck %s --check-prefix=ALL --check-prefix=AVX512 --check-prefix=AVX512BW ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=knl -mattr=+avx512dq,+avx512vl | FileCheck %s --check-prefix=ALL --check-prefix=AVX512VL --check-prefix=AVX512DQVL ; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=knl -mattr=+avx512bw,+avx512vl | FileCheck %s --check-prefix=ALL --check-prefix=AVX512VL --check-prefix=AVX512BWVL +; +; 32-bit runs to make sure we do reasonable things for i64 shifts. +; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx | FileCheck %s --check-prefix=ALL --check-prefix=X32-AVX --check-prefix=X32-AVX1 +; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=ALL --check-prefix=X32-AVX --check-prefix=X32-AVX2 ; ; Variable Shifts @@ -56,6 +60,26 @@ define <4 x i64> @var_shift_v4i64(<4 x i64> %a, <4 x i64> %b) nounwind { ; AVX512VL: # BB#0: ; AVX512VL-NEXT: vpsllvq %ymm1, %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: var_shift_v4i64: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2 +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3 +; X32-AVX1-NEXT: vpsllq %xmm2, %xmm3, %xmm4 +; X32-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[2,3,0,1] +; X32-AVX1-NEXT: vpsllq %xmm2, %xmm3, %xmm2 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm4[0,1,2,3],xmm2[4,5,6,7] +; X32-AVX1-NEXT: vpsllq %xmm1, %xmm0, %xmm3 +; X32-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; X32-AVX1-NEXT: vpsllq %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm3[0,1,2,3],xmm0[4,5,6,7] +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: var_shift_v4i64: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpsllvq %ymm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = shl <4 x i64> %a, %b ret <4 x i64> %shift } @@ -105,6 +129,27 @@ define <8 x i32> @var_shift_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind { ; AVX512VL: # BB#0: ; AVX512VL-NEXT: vpsllvd %ymm1, %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: var_shift_v8i32: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2 +; X32-AVX1-NEXT: vpslld $23, %xmm2, %xmm2 +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [1065353216,1065353216,1065353216,1065353216] +; X32-AVX1-NEXT: vpaddd %xmm3, %xmm2, %xmm2 +; X32-AVX1-NEXT: vcvttps2dq %xmm2, %xmm2 +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm4 +; X32-AVX1-NEXT: vpmulld %xmm4, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpslld $23, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpaddd %xmm3, %xmm1, %xmm1 +; X32-AVX1-NEXT: vcvttps2dq %xmm1, %xmm1 +; X32-AVX1-NEXT: vpmulld %xmm0, %xmm1, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: var_shift_v8i32: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpsllvd %ymm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = shl <8 x i32> %a, %b ret <8 x i32> %shift } @@ -205,6 +250,55 @@ define <16 x i16> @var_shift_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind { ; AVX512BWVL: # BB#0: ; AVX512BWVL-NEXT: vpsllvw %ymm1, %ymm0, %ymm0 ; AVX512BWVL-NEXT: retq +; +; X32-AVX1-LABEL: var_shift_v16i16: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2 +; X32-AVX1-NEXT: vpsllw $12, %xmm2, %xmm3 +; X32-AVX1-NEXT: vpsllw $4, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpor %xmm3, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpaddw %xmm2, %xmm2, %xmm3 +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm4 +; X32-AVX1-NEXT: vpsllw $8, %xmm4, %xmm5 +; X32-AVX1-NEXT: vpblendvb %xmm2, %xmm5, %xmm4, %xmm2 +; X32-AVX1-NEXT: vpsllw $4, %xmm2, %xmm4 +; X32-AVX1-NEXT: vpblendvb %xmm3, %xmm4, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsllw $2, %xmm2, %xmm4 +; X32-AVX1-NEXT: vpaddw %xmm3, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpblendvb %xmm3, %xmm4, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsllw $1, %xmm2, %xmm4 +; X32-AVX1-NEXT: vpaddw %xmm3, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpblendvb %xmm3, %xmm4, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsllw $12, %xmm1, %xmm3 +; X32-AVX1-NEXT: vpsllw $4, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpor %xmm3, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpaddw %xmm1, %xmm1, %xmm3 +; X32-AVX1-NEXT: vpsllw $8, %xmm0, %xmm4 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm4, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsllw $4, %xmm0, %xmm1 +; X32-AVX1-NEXT: vpblendvb %xmm3, %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsllw $2, %xmm0, %xmm1 +; X32-AVX1-NEXT: vpaddw %xmm3, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpblendvb %xmm3, %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsllw $1, %xmm0, %xmm1 +; X32-AVX1-NEXT: vpaddw %xmm3, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpblendvb %xmm3, %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: var_shift_v16i16: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpxor %ymm2, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpunpckhwd {{.*#+}} ymm3 = ymm1[4],ymm2[4],ymm1[5],ymm2[5],ymm1[6],ymm2[6],ymm1[7],ymm2[7],ymm1[12],ymm2[12],ymm1[13],ymm2[13],ymm1[14],ymm2[14],ymm1[15],ymm2[15] +; X32-AVX2-NEXT: vpunpckhwd {{.*#+}} ymm4 = ymm2[4],ymm0[4],ymm2[5],ymm0[5],ymm2[6],ymm0[6],ymm2[7],ymm0[7],ymm2[12],ymm0[12],ymm2[13],ymm0[13],ymm2[14],ymm0[14],ymm2[15],ymm0[15] +; X32-AVX2-NEXT: vpsllvd %ymm3, %ymm4, %ymm3 +; X32-AVX2-NEXT: vpsrld $16, %ymm3, %ymm3 +; X32-AVX2-NEXT: vpunpcklwd {{.*#+}} ymm1 = ymm1[0],ymm2[0],ymm1[1],ymm2[1],ymm1[2],ymm2[2],ymm1[3],ymm2[3],ymm1[8],ymm2[8],ymm1[9],ymm2[9],ymm1[10],ymm2[10],ymm1[11],ymm2[11] +; X32-AVX2-NEXT: vpunpcklwd {{.*#+}} ymm0 = ymm2[0],ymm0[0],ymm2[1],ymm0[1],ymm2[2],ymm0[2],ymm2[3],ymm0[3],ymm2[8],ymm0[8],ymm2[9],ymm0[9],ymm2[10],ymm0[10],ymm2[11],ymm0[11] +; X32-AVX2-NEXT: vpsllvd %ymm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsrld $16, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpackusdw %ymm3, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = shl <16 x i16> %a, %b ret <16 x i16> %shift } @@ -319,6 +413,52 @@ define <32 x i8> @var_shift_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind { ; AVX512BWVL-NEXT: vpsllvw %zmm1, %zmm0, %zmm0 ; AVX512BWVL-NEXT: vpmovwb %zmm0, %ymm0 ; AVX512BWVL-NEXT: retq +; +; X32-AVX1-LABEL: var_shift_v32i8: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2 +; X32-AVX1-NEXT: vpsllw $4, %xmm2, %xmm3 +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240] +; X32-AVX1-NEXT: vpand %xmm4, %xmm3, %xmm3 +; X32-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm5 +; X32-AVX1-NEXT: vpsllw $5, %xmm5, %xmm5 +; X32-AVX1-NEXT: vpblendvb %xmm5, %xmm3, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsllw $2, %xmm2, %xmm3 +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm6 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252] +; X32-AVX1-NEXT: vpand %xmm6, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpaddb %xmm5, %xmm5, %xmm5 +; X32-AVX1-NEXT: vpblendvb %xmm5, %xmm3, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpaddb %xmm2, %xmm2, %xmm3 +; X32-AVX1-NEXT: vpaddb %xmm5, %xmm5, %xmm5 +; X32-AVX1-NEXT: vpblendvb %xmm5, %xmm3, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsllw $4, %xmm0, %xmm3 +; X32-AVX1-NEXT: vpand %xmm4, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpsllw $5, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm3, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsllw $2, %xmm0, %xmm3 +; X32-AVX1-NEXT: vpand %xmm6, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpaddb %xmm1, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm3, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpaddb %xmm0, %xmm0, %xmm3 +; X32-AVX1-NEXT: vpaddb %xmm1, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm3, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: var_shift_v32i8: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpsllw $5, %ymm1, %ymm1 +; X32-AVX2-NEXT: vpsllw $4, %ymm0, %ymm2 +; X32-AVX2-NEXT: vpand {{\.LCPI.*}}, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpblendvb %ymm1, %ymm2, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsllw $2, %ymm0, %ymm2 +; X32-AVX2-NEXT: vpand {{\.LCPI.*}}, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpaddb %ymm1, %ymm1, %ymm1 +; X32-AVX2-NEXT: vpblendvb %ymm1, %ymm2, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpaddb %ymm0, %ymm0, %ymm2 +; X32-AVX2-NEXT: vpaddb %ymm1, %ymm1, %ymm1 +; X32-AVX2-NEXT: vpblendvb %ymm1, %ymm2, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = shl <32 x i8> %a, %b ret <32 x i8> %shift } @@ -363,6 +503,23 @@ define <4 x i64> @splatvar_shift_v4i64(<4 x i64> %a, <4 x i64> %b) nounwind { ; AVX512VL: # BB#0: ; AVX512VL-NEXT: vpsllq %xmm1, %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: splatvar_shift_v4i64: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vpextrd $1, %xmm1, %eax +; X32-AVX1-NEXT: vpinsrd $1, %eax, %xmm1, %xmm1 +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2 +; X32-AVX1-NEXT: vpsllq %xmm1, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsllq %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: splatvar_shift_v4i64: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpextrd $1, %xmm1, %eax +; X32-AVX2-NEXT: vpinsrd $1, %eax, %xmm1, %xmm1 +; X32-AVX2-NEXT: vpsllq %xmm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %splat = shufflevector <4 x i64> %b, <4 x i64> undef, <4 x i32> zeroinitializer %shift = shl <4 x i64> %a, %splat ret <4 x i64> %shift @@ -410,6 +567,21 @@ define <8 x i32> @splatvar_shift_v8i32(<8 x i32> %a, <8 x i32> %b) nounwind { ; AVX512VL-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero ; AVX512VL-NEXT: vpslld %xmm1, %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: splatvar_shift_v8i32: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2 +; X32-AVX1-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero +; X32-AVX1-NEXT: vpslld %xmm1, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpslld %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: splatvar_shift_v8i32: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpmovzxdq {{.*#+}} xmm1 = xmm1[0],zero,xmm1[1],zero +; X32-AVX2-NEXT: vpslld %xmm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %splat = shufflevector <8 x i32> %b, <8 x i32> undef, <8 x i32> zeroinitializer %shift = shl <8 x i32> %a, %splat ret <8 x i32> %shift @@ -457,6 +629,21 @@ define <16 x i16> @splatvar_shift_v16i16(<16 x i16> %a, <16 x i16> %b) nounwind ; AVX512VL-NEXT: vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero ; AVX512VL-NEXT: vpsllw %xmm1, %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: splatvar_shift_v16i16: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2 +; X32-AVX1-NEXT: vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero +; X32-AVX1-NEXT: vpsllw %xmm1, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsllw %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: splatvar_shift_v16i16: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpmovzxwq {{.*#+}} xmm1 = xmm1[0],zero,zero,zero,xmm1[1],zero,zero,zero +; X32-AVX2-NEXT: vpsllw %xmm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %splat = shufflevector <16 x i16> %b, <16 x i16> undef, <16 x i32> zeroinitializer %shift = shl <16 x i16> %a, %splat ret <16 x i16> %shift @@ -577,6 +764,51 @@ define <32 x i8> @splatvar_shift_v32i8(<32 x i8> %a, <32 x i8> %b) nounwind { ; AVX512BWVL-NEXT: vpsllvw %zmm1, %zmm0, %zmm0 ; AVX512BWVL-NEXT: vpmovwb %zmm0, %ymm0 ; AVX512BWVL-NEXT: retq +; +; X32-AVX1-LABEL: splatvar_shift_v32i8: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vpxor %xmm2, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpshufb %xmm2, %xmm1, %xmm1 +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm2 +; X32-AVX1-NEXT: vpsllw $4, %xmm2, %xmm3 +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240] +; X32-AVX1-NEXT: vpand %xmm4, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpsllw $5, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm3, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsllw $2, %xmm2, %xmm3 +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm5 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252] +; X32-AVX1-NEXT: vpand %xmm5, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpaddb %xmm1, %xmm1, %xmm6 +; X32-AVX1-NEXT: vpblendvb %xmm6, %xmm3, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpaddb %xmm2, %xmm2, %xmm3 +; X32-AVX1-NEXT: vpaddb %xmm6, %xmm6, %xmm7 +; X32-AVX1-NEXT: vpblendvb %xmm7, %xmm3, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpsllw $4, %xmm0, %xmm3 +; X32-AVX1-NEXT: vpand %xmm4, %xmm3, %xmm3 +; X32-AVX1-NEXT: vpblendvb %xmm1, %xmm3, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsllw $2, %xmm0, %xmm1 +; X32-AVX1-NEXT: vpand %xmm5, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpblendvb %xmm6, %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpaddb %xmm0, %xmm0, %xmm1 +; X32-AVX1-NEXT: vpblendvb %xmm7, %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: splatvar_shift_v32i8: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpbroadcastb %xmm1, %ymm1 +; X32-AVX2-NEXT: vpsllw $4, %ymm0, %ymm2 +; X32-AVX2-NEXT: vpand {{\.LCPI.*}}, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpsllw $5, %ymm1, %ymm1 +; X32-AVX2-NEXT: vpblendvb %ymm1, %ymm2, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsllw $2, %ymm0, %ymm2 +; X32-AVX2-NEXT: vpand {{\.LCPI.*}}, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpaddb %ymm1, %ymm1, %ymm1 +; X32-AVX2-NEXT: vpblendvb %ymm1, %ymm2, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpaddb %ymm0, %ymm0, %ymm2 +; X32-AVX2-NEXT: vpaddb %ymm1, %ymm1, %ymm1 +; X32-AVX2-NEXT: vpblendvb %ymm1, %ymm2, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %splat = shufflevector <32 x i8> %b, <32 x i8> undef, <32 x i32> zeroinitializer %shift = shl <32 x i8> %a, %splat ret <32 x i8> %shift @@ -626,6 +858,27 @@ define <4 x i64> @constant_shift_v4i64(<4 x i64> %a) nounwind { ; AVX512VL: # BB#0: ; AVX512VL-NEXT: vpsllvq {{.*}}(%rip), %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: constant_shift_v4i64: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vmovdqa {{.*#+}} ymm1 = [1,0,7,0,31,0,62,0] +; X32-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2 +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3 +; X32-AVX1-NEXT: vpsllq %xmm2, %xmm3, %xmm4 +; X32-AVX1-NEXT: vpshufd {{.*#+}} xmm2 = xmm2[2,3,0,1] +; X32-AVX1-NEXT: vpsllq %xmm2, %xmm3, %xmm2 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm2 = xmm4[0,1,2,3],xmm2[4,5,6,7] +; X32-AVX1-NEXT: vpsllq %xmm1, %xmm0, %xmm3 +; X32-AVX1-NEXT: vpshufd {{.*#+}} xmm1 = xmm1[2,3,0,1] +; X32-AVX1-NEXT: vpsllq %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpblendw {{.*#+}} xmm0 = xmm3[0,1,2,3],xmm0[4,5,6,7] +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: constant_shift_v4i64: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpsllvq {{\.LCPI.*}}, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = shl <4 x i64> %a, ret <4 x i64> %shift } @@ -666,6 +919,19 @@ define <8 x i32> @constant_shift_v8i32(<8 x i32> %a) nounwind { ; AVX512VL: # BB#0: ; AVX512VL-NEXT: vpsllvd {{.*}}(%rip), %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: constant_shift_v8i32: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vpmulld {{\.LCPI.*}}, %xmm0, %xmm1 +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0 +; X32-AVX1-NEXT: vpmulld {{\.LCPI.*}}, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: constant_shift_v8i32: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpsllvd {{\.LCPI.*}}, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = shl <8 x i32> %a, ret <8 x i32> %shift } @@ -719,6 +985,19 @@ define <16 x i16> @constant_shift_v16i16(<16 x i16> %a) nounwind { ; AVX512BWVL: # BB#0: ; AVX512BWVL-NEXT: vpsllvw {{.*}}(%rip), %ymm0, %ymm0 ; AVX512BWVL-NEXT: retq +; +; X32-AVX1-LABEL: constant_shift_v16i16: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vpmullw {{\.LCPI.*}}, %xmm0, %xmm1 +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0 +; X32-AVX1-NEXT: vpmullw {{\.LCPI.*}}, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: constant_shift_v16i16: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpmullw {{\.LCPI.*}}, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = shl <16 x i16> %a, ret <16 x i16> %shift } @@ -827,6 +1106,48 @@ define <32 x i8> @constant_shift_v32i8(<32 x i8> %a) nounwind { ; AVX512BWVL-NEXT: vpsllvw {{.*}}(%rip), %zmm0, %zmm0 ; AVX512BWVL-NEXT: vpmovwb %zmm0, %ymm0 ; AVX512BWVL-NEXT: retq +; +; X32-AVX1-LABEL: constant_shift_v32i8: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1 +; X32-AVX1-NEXT: vpsllw $4, %xmm1, %xmm2 +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm3 = [240,240,240,240,240,240,240,240,240,240,240,240,240,240,240,240] +; X32-AVX1-NEXT: vpand %xmm3, %xmm2, %xmm2 +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm4 = [8192,24640,41088,57536,49376,32928,16480,32] +; X32-AVX1-NEXT: vpblendvb %xmm4, %xmm2, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpsllw $2, %xmm1, %xmm2 +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm5 = [252,252,252,252,252,252,252,252,252,252,252,252,252,252,252,252] +; X32-AVX1-NEXT: vpand %xmm5, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpaddb %xmm4, %xmm4, %xmm6 +; X32-AVX1-NEXT: vpblendvb %xmm6, %xmm2, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpaddb %xmm1, %xmm1, %xmm2 +; X32-AVX1-NEXT: vpaddb %xmm6, %xmm6, %xmm7 +; X32-AVX1-NEXT: vpblendvb %xmm7, %xmm2, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpsllw $4, %xmm0, %xmm2 +; X32-AVX1-NEXT: vpand %xmm3, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpblendvb %xmm4, %xmm2, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpsllw $2, %xmm0, %xmm2 +; X32-AVX1-NEXT: vpand %xmm5, %xmm2, %xmm2 +; X32-AVX1-NEXT: vpblendvb %xmm6, %xmm2, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpaddb %xmm0, %xmm0, %xmm2 +; X32-AVX1-NEXT: vpblendvb %xmm7, %xmm2, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: constant_shift_v32i8: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpsllw $4, %ymm0, %ymm1 +; X32-AVX2-NEXT: vpand {{\.LCPI.*}}, %ymm1, %ymm1 +; X32-AVX2-NEXT: vmovdqa {{.*#+}} ymm2 = [8192,24640,41088,57536,49376,32928,16480,32,8192,24640,41088,57536,49376,32928,16480,32] +; X32-AVX2-NEXT: vpblendvb %ymm2, %ymm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpsllw $2, %ymm0, %ymm1 +; X32-AVX2-NEXT: vpand {{\.LCPI.*}}, %ymm1, %ymm1 +; X32-AVX2-NEXT: vpaddb %ymm2, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpblendvb %ymm2, %ymm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpaddb %ymm0, %ymm0, %ymm1 +; X32-AVX2-NEXT: vpaddb %ymm2, %ymm2, %ymm2 +; X32-AVX2-NEXT: vpblendvb %ymm2, %ymm1, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = shl <32 x i8> %a, ret <32 x i8> %shift } @@ -871,6 +1192,21 @@ define <4 x i64> @splatconstant_shift_v4i64(<4 x i64> %a) nounwind { ; AVX512VL: # BB#0: ; AVX512VL-NEXT: vpsllq $7, %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: splatconstant_shift_v4i64: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vmovdqa {{.*#+}} ymm1 = [7,0,7,0,7,0,7,0] +; X32-AVX1-NEXT: vextractf128 $1, %ymm1, %xmm2 +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm3 +; X32-AVX1-NEXT: vpsllq %xmm2, %xmm3, %xmm2 +; X32-AVX1-NEXT: vpsllq %xmm1, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm2, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: splatconstant_shift_v4i64: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpsllq $7, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = shl <4 x i64> %a, ret <4 x i64> %shift } @@ -911,6 +1247,19 @@ define <8 x i32> @splatconstant_shift_v8i32(<8 x i32> %a) nounwind { ; AVX512VL: # BB#0: ; AVX512VL-NEXT: vpslld $5, %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: splatconstant_shift_v8i32: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vpslld $5, %xmm0, %xmm1 +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0 +; X32-AVX1-NEXT: vpslld $5, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: splatconstant_shift_v8i32: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpslld $5, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = shl <8 x i32> %a, ret <8 x i32> %shift } @@ -951,6 +1300,19 @@ define <16 x i16> @splatconstant_shift_v16i16(<16 x i16> %a) nounwind { ; AVX512VL: # BB#0: ; AVX512VL-NEXT: vpsllw $3, %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: splatconstant_shift_v16i16: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vpsllw $3, %xmm0, %xmm1 +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm0 +; X32-AVX1-NEXT: vpsllw $3, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm0, %ymm1, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: splatconstant_shift_v16i16: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpsllw $3, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = shl <16 x i16> %a, ret <16 x i16> %shift } @@ -999,6 +1361,23 @@ define <32 x i8> @splatconstant_shift_v32i8(<32 x i8> %a) nounwind { ; AVX512VL-NEXT: vpsllw $3, %ymm0, %ymm0 ; AVX512VL-NEXT: vpand {{.*}}(%rip), %ymm0, %ymm0 ; AVX512VL-NEXT: retq +; +; X32-AVX1-LABEL: splatconstant_shift_v32i8: +; X32-AVX1: # BB#0: +; X32-AVX1-NEXT: vextractf128 $1, %ymm0, %xmm1 +; X32-AVX1-NEXT: vpsllw $3, %xmm1, %xmm1 +; X32-AVX1-NEXT: vmovdqa {{.*#+}} xmm2 = [248,248,248,248,248,248,248,248,248,248,248,248,248,248,248,248] +; X32-AVX1-NEXT: vpand %xmm2, %xmm1, %xmm1 +; X32-AVX1-NEXT: vpsllw $3, %xmm0, %xmm0 +; X32-AVX1-NEXT: vpand %xmm2, %xmm0, %xmm0 +; X32-AVX1-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm0 +; X32-AVX1-NEXT: retl +; +; X32-AVX2-LABEL: splatconstant_shift_v32i8: +; X32-AVX2: # BB#0: +; X32-AVX2-NEXT: vpsllw $3, %ymm0, %ymm0 +; X32-AVX2-NEXT: vpand {{\.LCPI.*}}, %ymm0, %ymm0 +; X32-AVX2-NEXT: retl %shift = shl <32 x i8> %a, ret <32 x i8> %shift }