69 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py2; RUN: llc < %s -mtriple=i686-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=X863; RUN: llc < %s -mtriple=x86_64-unknown-unknown -mattr=+avx2 | FileCheck %s --check-prefix=X644 5define <8 x i32> @perm_cl_int_8x32(<8 x i32> %A) nounwind readnone {6; X86-LABEL: perm_cl_int_8x32:7; X86: # %bb.0: # %entry8; X86-NEXT: vmovaps {{.*#+}} ymm1 = [0,7,2,1,2,7,6,0]9; X86-NEXT: vpermps %ymm0, %ymm1, %ymm010; X86-NEXT: retl11;12; X64-LABEL: perm_cl_int_8x32:13; X64: # %bb.0: # %entry14; X64-NEXT: vmovaps {{.*#+}} ymm1 = [0,7,2,1,2,7,6,0]15; X64-NEXT: vpermps %ymm0, %ymm1, %ymm016; X64-NEXT: retq17entry:18 %B = shufflevector <8 x i32> %A, <8 x i32> undef, <8 x i32> <i32 0, i32 7, i32 2, i32 1, i32 2, i32 7, i32 6, i32 0>19 ret <8 x i32> %B20}21 22 23define <8 x float> @perm_cl_fp_8x32(<8 x float> %A) nounwind readnone {24; X86-LABEL: perm_cl_fp_8x32:25; X86: # %bb.0: # %entry26; X86-NEXT: vmovaps {{.*#+}} ymm1 = [u,7,2,u,4,u,1,6]27; X86-NEXT: vpermps %ymm0, %ymm1, %ymm028; X86-NEXT: retl29;30; X64-LABEL: perm_cl_fp_8x32:31; X64: # %bb.0: # %entry32; X64-NEXT: vmovaps {{.*#+}} ymm1 = [u,7,2,u,4,u,1,6]33; X64-NEXT: vpermps %ymm0, %ymm1, %ymm034; X64-NEXT: retq35entry:36 %B = shufflevector <8 x float> %A, <8 x float> undef, <8 x i32> <i32 undef, i32 7, i32 2, i32 undef, i32 4, i32 undef, i32 1, i32 6>37 ret <8 x float> %B38}39 40define <4 x i64> @perm_cl_int_4x64(<4 x i64> %A) nounwind readnone {41; X86-LABEL: perm_cl_int_4x64:42; X86: # %bb.0: # %entry43; X86-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,1]44; X86-NEXT: retl45;46; X64-LABEL: perm_cl_int_4x64:47; X64: # %bb.0: # %entry48; X64-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,1]49; X64-NEXT: retq50entry:51 %B = shufflevector <4 x i64> %A, <4 x i64> undef, <4 x i32> <i32 0, i32 3, i32 2, i32 1>52 ret <4 x i64> %B53}54 55define <4 x double> @perm_cl_fp_4x64(<4 x double> %A) nounwind readnone {56; X86-LABEL: perm_cl_fp_4x64:57; X86: # %bb.0: # %entry58; X86-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,1]59; X86-NEXT: retl60;61; X64-LABEL: perm_cl_fp_4x64:62; X64: # %bb.0: # %entry63; X64-NEXT: vpermpd {{.*#+}} ymm0 = ymm0[0,3,2,1]64; X64-NEXT: retq65entry:66 %B = shufflevector <4 x double> %A, <4 x double> undef, <4 x i32> <i32 0, i32 3, i32 2, i32 1>67 ret <4 x double> %B68}69