90 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+ssse3 | FileCheck %s --check-prefix=SSE3; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+xop | FileCheck %s --check-prefix=XOP4; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+avx2 | FileCheck %s --check-prefix=AVX25; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+avx512f,+avx512vl | FileCheck %s --check-prefix=AVX512F6; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+avx512f,+avx512vl,+avx512bw | FileCheck %s --check-prefix=AVX512BW7; RUN: llc < %s -mtriple=x86_64-pc-linux -mattr=+avx512f,+avx512vl,+avx512vbmi | FileCheck %s --check-prefix=AVX512VBMI8 9define <32 x i8> @foo(ptr %x0) {10; SSE-LABEL: foo:11; SSE: # %bb.0:12; SSE-NEXT: movdqu (%rdi), %xmm013; SSE-NEXT: movdqu 16(%rdi), %xmm214; SSE-NEXT: movdqu 32(%rdi), %xmm115; SSE-NEXT: movdqa %xmm2, %xmm316; SSE-NEXT: pshufb {{.*#+}} xmm3 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,xmm3[0,2,3,5,6]17; SSE-NEXT: pshufb {{.*#+}} xmm0 = xmm0[0,1,3,4,6,7,9,10,12,13,15],zero,zero,zero,zero,zero18; SSE-NEXT: por %xmm3, %xmm019; SSE-NEXT: pshufb {{.*#+}} xmm2 = xmm2[8,9,11,12,14,15],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero20; SSE-NEXT: pshufb {{.*#+}} xmm1 = zero,zero,zero,zero,zero,zero,xmm1[1,2,4,5,7,8,10,11,13,14]21; SSE-NEXT: por %xmm2, %xmm122; SSE-NEXT: retq23;24; XOP-LABEL: foo:25; XOP: # %bb.0:26; XOP-NEXT: vmovdqu (%rdi), %xmm027; XOP-NEXT: vmovdqu 16(%rdi), %xmm128; XOP-NEXT: vmovdqu 32(%rdi), %xmm229; XOP-NEXT: vpperm {{.*#+}} xmm0 = xmm0[0,1,3,4,6,7,9,10,12,13,15],xmm1[0,2,3,5,6]30; XOP-NEXT: vpperm {{.*#+}} xmm1 = xmm1[8,9,11,12,14,15],xmm2[1,2,4,5,7,8,10,11,13,14]31; XOP-NEXT: vinsertf128 $1, %xmm1, %ymm0, %ymm032; XOP-NEXT: retq33;34; AVX2-LABEL: foo:35; AVX2: # %bb.0:36; AVX2-NEXT: vmovdqu 32(%rdi), %xmm037; AVX2-NEXT: vmovdqu (%rdi), %ymm138; AVX2-NEXT: vmovdqu 16(%rdi), %xmm239; AVX2-NEXT: vpshufb {{.*#+}} xmm2 = xmm2[u,u,u,u,u,u,u,u,u,u,u,0,2,3,5,6]40; AVX2-NEXT: vpshufb {{.*#+}} ymm1 = ymm1[0,1,3,4,6,7,9,10,12,13,15,u,u,u,u,u,24,25,27,28,30,31,u,u,u,u,u,u,u,u,u,u]41; AVX2-NEXT: vbroadcasti128 {{.*#+}} ymm3 = [255,255,255,255,255,255,255,255,255,255,255,0,0,0,0,0,255,255,255,255,255,255,255,255,255,255,255,0,0,0,0,0]42; AVX2-NEXT: # ymm3 = mem[0,1,0,1]43; AVX2-NEXT: vpblendvb %ymm3, %ymm1, %ymm2, %ymm144; AVX2-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[u,u,u,u,u,u,1,2,4,5,7,8,10,11,13,14]45; AVX2-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm046; AVX2-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7],ymm1[8,9,10],ymm0[11,12,13,14,15]47; AVX2-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]48; AVX2-NEXT: retq49;50; AVX512F-LABEL: foo:51; AVX512F: # %bb.0:52; AVX512F-NEXT: vmovdqu (%rdi), %ymm053; AVX512F-NEXT: vmovdqu 32(%rdi), %xmm154; AVX512F-NEXT: vpshufb {{.*#+}} xmm1 = xmm1[u,u,u,u,u,u,1,2,4,5,7,8,10,11,13,14]55; AVX512F-NEXT: vinserti128 $1, %xmm1, %ymm0, %ymm156; AVX512F-NEXT: vmovdqu 16(%rdi), %xmm257; AVX512F-NEXT: vpshufb {{.*#+}} ymm2 = zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,ymm2[0,2,3,5,6],zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero,zero58; AVX512F-NEXT: vpshufb {{.*#+}} ymm0 = ymm0[0,1,3,4,6,7,9,10,12,13,15],zero,zero,zero,zero,zero,ymm0[24,25,27,28,30,31,u,u,u,u,u,u,u,u,u,u]59; AVX512F-NEXT: vpor %ymm2, %ymm0, %ymm060; AVX512F-NEXT: vpblendw {{.*#+}} ymm1 = ymm0[0,1,2],ymm1[3,4,5,6,7],ymm0[8,9,10],ymm1[11,12,13,14,15]61; AVX512F-NEXT: vpblendd {{.*#+}} ymm0 = ymm0[0,1,2,3],ymm1[4,5,6,7]62; AVX512F-NEXT: retq63;64; AVX512BW-LABEL: foo:65; AVX512BW: # %bb.0:66; AVX512BW-NEXT: vmovdqu 32(%rdi), %xmm067; AVX512BW-NEXT: vmovdqu (%rdi), %ymm168; AVX512BW-NEXT: vmovdqu 16(%rdi), %xmm269; AVX512BW-NEXT: vpshufb {{.*#+}} xmm2 = xmm2[u,u,u,u,u,u,u,u,u,u,u,0,2,3,5,6]70; AVX512BW-NEXT: vpshufb {{.*#+}} ymm1 = ymm1[0,1,3,4,6,7,9,10,12,13,15,u,u,u,u,u,24,25,27,28,30,31,u,u,u,u,u,u,u,u,u,u]71; AVX512BW-NEXT: movl $63488, %eax # imm = 0xF80072; AVX512BW-NEXT: kmovd %eax, %k173; AVX512BW-NEXT: vmovdqu8 %ymm2, %ymm1 {%k1}74; AVX512BW-NEXT: vpshufb {{.*#+}} xmm0 = xmm0[u,u,u,u,u,u,1,2,4,5,7,8,10,11,13,14]75; AVX512BW-NEXT: vinserti128 $1, %xmm0, %ymm0, %ymm076; AVX512BW-NEXT: vpblendw {{.*#+}} ymm0 = ymm1[0,1,2],ymm0[3,4,5,6,7],ymm1[8,9,10],ymm0[11,12,13,14,15]77; AVX512BW-NEXT: vpblendd {{.*#+}} ymm0 = ymm1[0,1,2,3],ymm0[4,5,6,7]78; AVX512BW-NEXT: retq79;80; AVX512VBMI-LABEL: foo:81; AVX512VBMI: # %bb.0:82; AVX512VBMI-NEXT: vmovdqu 32(%rdi), %xmm183; AVX512VBMI-NEXT: vmovdqa {{.*#+}} ymm0 = [32,33,35,36,38,39,41,42,44,45,47,48,50,51,53,54,56,57,59,60,62,63,1,2,4,5,7,8,10,11,13,14]84; AVX512VBMI-NEXT: vpermi2b (%rdi), %ymm1, %ymm085; AVX512VBMI-NEXT: retq86 %1 = load <48 x i8>, ptr %x0, align 187 %2 = shufflevector <48 x i8> %1, <48 x i8> undef, <32 x i32> <i32 0, i32 1, i32 3, i32 4, i32 6, i32 7, i32 9, i32 10, i32 12, i32 13, i32 15, i32 16, i32 18, i32 19, i32 21, i32 22, i32 24, i32 25, i32 27, i32 28, i32 30, i32 31, i32 33, i32 34, i32 36, i32 37, i32 39, i32 40, i32 42, i32 43, i32 45, i32 46>88 ret <32 x i8> %289}90