115 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -mtriple=x86_64-unknown-unknown --mattr=+avx,+no-bypass-delay-shuffle | FileCheck %s --check-prefixes=CHECK,CHECK-AVX13; RUN: llc < %s -mtriple=x86_64-unknown-unknown --mattr=+avx | FileCheck %s --check-prefixes=CHECK,CHECK-AVX1-DELAY4; RUN: llc < %s -mtriple=x86_64-unknown-unknown --mattr=+avx2,+no-bypass-delay-shuffle | FileCheck %s --check-prefixes=CHECK,CHECK-AVX25; RUN: llc < %s -mtriple=x86_64-unknown-unknown --mattr=+avx2 | FileCheck %s --check-prefixes=CHECK,CHECK-AVX2-DELAY6; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=icelake-server | FileCheck %s --check-prefixes=CHECK,CHECK-ICX,CHECK-ICX-NO-BYPASS-DELAY7; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=icelake-server -mattr=-no-bypass-delay-shuffle | FileCheck %s --check-prefixes=CHECK,CHECK-ICX,CHECK-ICX-BYPASS-DELAY8; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=sandybridge -mattr=+no-bypass-delay-shuffle | FileCheck %s --check-prefixes=CHECK,CHECK-SNB,CHECK-SNB-NO-BYPASS-DELAY9; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mcpu=sandybridge -mattr=-no-bypass-delay-shuffle | FileCheck %s --check-prefixes=CHECK,CHECK-SNB,CHECK-SNB-BYPASS-DELAY10 11define <8 x float> @transform_VPERMILPSYrr(<8 x float> %a) nounwind {12; CHECK-LABEL: transform_VPERMILPSYrr:13; CHECK: # %bb.0:14; CHECK-NEXT: vshufps {{.*#+}} ymm0 = ymm0[3,2,1,0,7,6,5,4]15; CHECK-NEXT: retq16 %shufp = shufflevector <8 x float> %a, <8 x float> poison, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 7, i32 6, i32 5, i32 4>17 ret <8 x float> %shufp18}19 20define <4 x float> @transform_VPERMILPSrr(<4 x float> %a) nounwind {21; CHECK-LABEL: transform_VPERMILPSrr:22; CHECK: # %bb.0:23; CHECK-NEXT: vshufps {{.*#+}} xmm0 = xmm0[3,2,1,0]24; CHECK-NEXT: retq25 %shufp = shufflevector <4 x float> %a, <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>26 ret <4 x float> %shufp27}28 29define <8 x float> @transform_VPERMILPSYrm(ptr %ap) nounwind {30; CHECK-AVX1-LABEL: transform_VPERMILPSYrm:31; CHECK-AVX1: # %bb.0:32; CHECK-AVX1-NEXT: vpermilps {{.*#+}} ymm0 = mem[3,2,1,0,7,6,5,4]33; CHECK-AVX1-NEXT: retq34;35; CHECK-AVX1-DELAY-LABEL: transform_VPERMILPSYrm:36; CHECK-AVX1-DELAY: # %bb.0:37; CHECK-AVX1-DELAY-NEXT: vpermilps {{.*#+}} ymm0 = mem[3,2,1,0,7,6,5,4]38; CHECK-AVX1-DELAY-NEXT: retq39;40; CHECK-AVX2-LABEL: transform_VPERMILPSYrm:41; CHECK-AVX2: # %bb.0:42; CHECK-AVX2-NEXT: vpermilps {{.*#+}} ymm0 = mem[3,2,1,0,7,6,5,4]43; CHECK-AVX2-NEXT: retq44;45; CHECK-AVX2-DELAY-LABEL: transform_VPERMILPSYrm:46; CHECK-AVX2-DELAY: # %bb.0:47; CHECK-AVX2-DELAY-NEXT: vpermilps {{.*#+}} ymm0 = mem[3,2,1,0,7,6,5,4]48; CHECK-AVX2-DELAY-NEXT: retq49;50; CHECK-ICX-NO-BYPASS-DELAY-LABEL: transform_VPERMILPSYrm:51; CHECK-ICX-NO-BYPASS-DELAY: # %bb.0:52; CHECK-ICX-NO-BYPASS-DELAY-NEXT: vpshufd {{.*#+}} ymm0 = mem[3,2,1,0,7,6,5,4]53; CHECK-ICX-NO-BYPASS-DELAY-NEXT: retq54;55; CHECK-ICX-BYPASS-DELAY-LABEL: transform_VPERMILPSYrm:56; CHECK-ICX-BYPASS-DELAY: # %bb.0:57; CHECK-ICX-BYPASS-DELAY-NEXT: vpermilps {{.*#+}} ymm0 = mem[3,2,1,0,7,6,5,4]58; CHECK-ICX-BYPASS-DELAY-NEXT: retq59;60; CHECK-SNB-LABEL: transform_VPERMILPSYrm:61; CHECK-SNB: # %bb.0:62; CHECK-SNB-NEXT: vpermilps {{.*#+}} ymm0 = mem[3,2,1,0,7,6,5,4]63; CHECK-SNB-NEXT: retq64 %a = load <8 x float>, ptr %ap65 %shufp = shufflevector <8 x float> %a, <8 x float> poison, <8 x i32> <i32 3, i32 2, i32 1, i32 0, i32 7, i32 6, i32 5, i32 4>66 ret <8 x float> %shufp67}68 69define <4 x float> @transform_VPERMILPSrm(ptr %ap) nounwind {70; CHECK-AVX1-LABEL: transform_VPERMILPSrm:71; CHECK-AVX1: # %bb.0:72; CHECK-AVX1-NEXT: vpermilps {{.*#+}} xmm0 = mem[3,2,1,0]73; CHECK-AVX1-NEXT: retq74;75; CHECK-AVX1-DELAY-LABEL: transform_VPERMILPSrm:76; CHECK-AVX1-DELAY: # %bb.0:77; CHECK-AVX1-DELAY-NEXT: vpermilps {{.*#+}} xmm0 = mem[3,2,1,0]78; CHECK-AVX1-DELAY-NEXT: retq79;80; CHECK-AVX2-LABEL: transform_VPERMILPSrm:81; CHECK-AVX2: # %bb.0:82; CHECK-AVX2-NEXT: vpermilps {{.*#+}} xmm0 = mem[3,2,1,0]83; CHECK-AVX2-NEXT: retq84;85; CHECK-AVX2-DELAY-LABEL: transform_VPERMILPSrm:86; CHECK-AVX2-DELAY: # %bb.0:87; CHECK-AVX2-DELAY-NEXT: vpermilps {{.*#+}} xmm0 = mem[3,2,1,0]88; CHECK-AVX2-DELAY-NEXT: retq89;90; CHECK-ICX-NO-BYPASS-DELAY-LABEL: transform_VPERMILPSrm:91; CHECK-ICX-NO-BYPASS-DELAY: # %bb.0:92; CHECK-ICX-NO-BYPASS-DELAY-NEXT: vpshufd {{.*#+}} xmm0 = mem[3,2,1,0]93; CHECK-ICX-NO-BYPASS-DELAY-NEXT: retq94;95; CHECK-ICX-BYPASS-DELAY-LABEL: transform_VPERMILPSrm:96; CHECK-ICX-BYPASS-DELAY: # %bb.0:97; CHECK-ICX-BYPASS-DELAY-NEXT: vpermilps {{.*#+}} xmm0 = mem[3,2,1,0]98; CHECK-ICX-BYPASS-DELAY-NEXT: retq99;100; CHECK-SNB-NO-BYPASS-DELAY-LABEL: transform_VPERMILPSrm:101; CHECK-SNB-NO-BYPASS-DELAY: # %bb.0:102; CHECK-SNB-NO-BYPASS-DELAY-NEXT: vpshufd {{.*#+}} xmm0 = mem[3,2,1,0]103; CHECK-SNB-NO-BYPASS-DELAY-NEXT: retq104;105; CHECK-SNB-BYPASS-DELAY-LABEL: transform_VPERMILPSrm:106; CHECK-SNB-BYPASS-DELAY: # %bb.0:107; CHECK-SNB-BYPASS-DELAY-NEXT: vpermilps {{.*#+}} xmm0 = mem[3,2,1,0]108; CHECK-SNB-BYPASS-DELAY-NEXT: retq109 %a = load <4 x float>, ptr %ap110 %shufp = shufflevector <4 x float> %a, <4 x float> poison, <4 x i32> <i32 3, i32 2, i32 1, i32 0>111 ret <4 x float> %shufp112}113;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:114; CHECK-ICX: {{.*}}115