1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py 2; RUN: opt -lower-matrix-intrinsics -S < %s | FileCheck %s 3; RUN: opt -passes='lower-matrix-intrinsics' -S < %s | FileCheck %s 4 5; Check that we do not emit shufflevectors to flatten the result of the 6; transpose and store the columns directly. 7define void @transpose_store(<8 x double> %a, <8 x double>* %Ptr) { 8; CHECK-LABEL: @transpose_store( 9; CHECK-NEXT: entry: 10; CHECK-NEXT: [[SPLIT:%.*]] = shufflevector <8 x double> [[A:%.*]], <8 x double> undef, <2 x i32> <i32 0, i32 1> 11; CHECK-NEXT: [[SPLIT1:%.*]] = shufflevector <8 x double> [[A]], <8 x double> undef, <2 x i32> <i32 2, i32 3> 12; CHECK-NEXT: [[SPLIT2:%.*]] = shufflevector <8 x double> [[A]], <8 x double> undef, <2 x i32> <i32 4, i32 5> 13; CHECK-NEXT: [[SPLIT3:%.*]] = shufflevector <8 x double> [[A]], <8 x double> undef, <2 x i32> <i32 6, i32 7> 14; CHECK-NEXT: [[TMP0:%.*]] = extractelement <2 x double> [[SPLIT]], i64 0 15; CHECK-NEXT: [[TMP1:%.*]] = insertelement <4 x double> undef, double [[TMP0]], i64 0 16; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x double> [[SPLIT1]], i64 0 17; CHECK-NEXT: [[TMP3:%.*]] = insertelement <4 x double> [[TMP1]], double [[TMP2]], i64 1 18; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x double> [[SPLIT2]], i64 0 19; CHECK-NEXT: [[TMP5:%.*]] = insertelement <4 x double> [[TMP3]], double [[TMP4]], i64 2 20; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 0 21; CHECK-NEXT: [[TMP7:%.*]] = insertelement <4 x double> [[TMP5]], double [[TMP6]], i64 3 22; CHECK-NEXT: [[TMP8:%.*]] = extractelement <2 x double> [[SPLIT]], i64 1 23; CHECK-NEXT: [[TMP9:%.*]] = insertelement <4 x double> undef, double [[TMP8]], i64 0 24; CHECK-NEXT: [[TMP10:%.*]] = extractelement <2 x double> [[SPLIT1]], i64 1 25; CHECK-NEXT: [[TMP11:%.*]] = insertelement <4 x double> [[TMP9]], double [[TMP10]], i64 1 26; CHECK-NEXT: [[TMP12:%.*]] = extractelement <2 x double> [[SPLIT2]], i64 1 27; CHECK-NEXT: [[TMP13:%.*]] = insertelement <4 x double> [[TMP11]], double [[TMP12]], i64 2 28; CHECK-NEXT: [[TMP14:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 1 29; CHECK-NEXT: [[TMP15:%.*]] = insertelement <4 x double> [[TMP13]], double [[TMP14]], i64 3 30; CHECK-NEXT: [[TMP16:%.*]] = bitcast <8 x double>* [[PTR:%.*]] to double* 31; CHECK-NEXT: [[VEC_CAST:%.*]] = bitcast double* [[TMP16]] to <4 x double>* 32; CHECK-NEXT: store <4 x double> [[TMP7]], <4 x double>* [[VEC_CAST]], align 8 33; CHECK-NEXT: [[VEC_GEP:%.*]] = getelementptr double, double* [[TMP16]], i64 4 34; CHECK-NEXT: [[VEC_CAST4:%.*]] = bitcast double* [[VEC_GEP]] to <4 x double>* 35; CHECK-NEXT: store <4 x double> [[TMP15]], <4 x double>* [[VEC_CAST4]], align 8 36; CHECK-NEXT: ret void 37; 38entry: 39 %c = call <8 x double> @llvm.matrix.transpose(<8 x double> %a, i32 2, i32 4) 40 store <8 x double> %c, <8 x double>* %Ptr, align 8 41 ret void 42} 43 44declare <8 x double> @llvm.matrix.transpose(<8 x double>, i32, i32) 45 46define <8 x double> @transpose_fadd(<8 x double> %a) { 47; CHECK-LABEL: @transpose_fadd( 48; CHECK-NEXT: entry: 49; CHECK-NEXT: [[SPLIT:%.*]] = shufflevector <8 x double> [[A:%.*]], <8 x double> undef, <2 x i32> <i32 0, i32 1> 50; CHECK-NEXT: [[SPLIT1:%.*]] = shufflevector <8 x double> [[A]], <8 x double> undef, <2 x i32> <i32 2, i32 3> 51; CHECK-NEXT: [[SPLIT2:%.*]] = shufflevector <8 x double> [[A]], <8 x double> undef, <2 x i32> <i32 4, i32 5> 52; CHECK-NEXT: [[SPLIT3:%.*]] = shufflevector <8 x double> [[A]], <8 x double> undef, <2 x i32> <i32 6, i32 7> 53; CHECK-NEXT: [[TMP0:%.*]] = extractelement <2 x double> [[SPLIT]], i64 0 54; CHECK-NEXT: [[TMP1:%.*]] = insertelement <4 x double> undef, double [[TMP0]], i64 0 55; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x double> [[SPLIT1]], i64 0 56; CHECK-NEXT: [[TMP3:%.*]] = insertelement <4 x double> [[TMP1]], double [[TMP2]], i64 1 57; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x double> [[SPLIT2]], i64 0 58; CHECK-NEXT: [[TMP5:%.*]] = insertelement <4 x double> [[TMP3]], double [[TMP4]], i64 2 59; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 0 60; CHECK-NEXT: [[TMP7:%.*]] = insertelement <4 x double> [[TMP5]], double [[TMP6]], i64 3 61; CHECK-NEXT: [[TMP8:%.*]] = extractelement <2 x double> [[SPLIT]], i64 1 62; CHECK-NEXT: [[TMP9:%.*]] = insertelement <4 x double> undef, double [[TMP8]], i64 0 63; CHECK-NEXT: [[TMP10:%.*]] = extractelement <2 x double> [[SPLIT1]], i64 1 64; CHECK-NEXT: [[TMP11:%.*]] = insertelement <4 x double> [[TMP9]], double [[TMP10]], i64 1 65; CHECK-NEXT: [[TMP12:%.*]] = extractelement <2 x double> [[SPLIT2]], i64 1 66; CHECK-NEXT: [[TMP13:%.*]] = insertelement <4 x double> [[TMP11]], double [[TMP12]], i64 2 67; CHECK-NEXT: [[TMP14:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 1 68; CHECK-NEXT: [[TMP15:%.*]] = insertelement <4 x double> [[TMP13]], double [[TMP14]], i64 3 69; CHECK-NEXT: [[SPLIT4:%.*]] = shufflevector <8 x double> [[A]], <8 x double> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> 70; CHECK-NEXT: [[SPLIT5:%.*]] = shufflevector <8 x double> [[A]], <8 x double> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7> 71; CHECK-NEXT: [[TMP16:%.*]] = fadd <4 x double> [[TMP7]], [[SPLIT4]] 72; CHECK-NEXT: [[TMP17:%.*]] = fadd <4 x double> [[TMP15]], [[SPLIT5]] 73; CHECK-NEXT: [[TMP18:%.*]] = shufflevector <4 x double> [[TMP16]], <4 x double> [[TMP17]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7> 74; CHECK-NEXT: ret <8 x double> [[TMP18]] 75; 76entry: 77 %c = call <8 x double> @llvm.matrix.transpose(<8 x double> %a, i32 2, i32 4) 78 %res = fadd <8 x double> %c, %a 79 ret <8 x double> %res 80} 81 82define <8 x double> @transpose_fmul(<8 x double> %a) { 83; CHECK-LABEL: @transpose_fmul( 84; CHECK-NEXT: entry: 85; CHECK-NEXT: [[SPLIT:%.*]] = shufflevector <8 x double> [[A:%.*]], <8 x double> undef, <2 x i32> <i32 0, i32 1> 86; CHECK-NEXT: [[SPLIT1:%.*]] = shufflevector <8 x double> [[A]], <8 x double> undef, <2 x i32> <i32 2, i32 3> 87; CHECK-NEXT: [[SPLIT2:%.*]] = shufflevector <8 x double> [[A]], <8 x double> undef, <2 x i32> <i32 4, i32 5> 88; CHECK-NEXT: [[SPLIT3:%.*]] = shufflevector <8 x double> [[A]], <8 x double> undef, <2 x i32> <i32 6, i32 7> 89; CHECK-NEXT: [[TMP0:%.*]] = extractelement <2 x double> [[SPLIT]], i64 0 90; CHECK-NEXT: [[TMP1:%.*]] = insertelement <4 x double> undef, double [[TMP0]], i64 0 91; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x double> [[SPLIT1]], i64 0 92; CHECK-NEXT: [[TMP3:%.*]] = insertelement <4 x double> [[TMP1]], double [[TMP2]], i64 1 93; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x double> [[SPLIT2]], i64 0 94; CHECK-NEXT: [[TMP5:%.*]] = insertelement <4 x double> [[TMP3]], double [[TMP4]], i64 2 95; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 0 96; CHECK-NEXT: [[TMP7:%.*]] = insertelement <4 x double> [[TMP5]], double [[TMP6]], i64 3 97; CHECK-NEXT: [[TMP8:%.*]] = extractelement <2 x double> [[SPLIT]], i64 1 98; CHECK-NEXT: [[TMP9:%.*]] = insertelement <4 x double> undef, double [[TMP8]], i64 0 99; CHECK-NEXT: [[TMP10:%.*]] = extractelement <2 x double> [[SPLIT1]], i64 1 100; CHECK-NEXT: [[TMP11:%.*]] = insertelement <4 x double> [[TMP9]], double [[TMP10]], i64 1 101; CHECK-NEXT: [[TMP12:%.*]] = extractelement <2 x double> [[SPLIT2]], i64 1 102; CHECK-NEXT: [[TMP13:%.*]] = insertelement <4 x double> [[TMP11]], double [[TMP12]], i64 2 103; CHECK-NEXT: [[TMP14:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 1 104; CHECK-NEXT: [[TMP15:%.*]] = insertelement <4 x double> [[TMP13]], double [[TMP14]], i64 3 105; CHECK-NEXT: [[SPLIT4:%.*]] = shufflevector <8 x double> [[A]], <8 x double> undef, <4 x i32> <i32 0, i32 1, i32 2, i32 3> 106; CHECK-NEXT: [[SPLIT5:%.*]] = shufflevector <8 x double> [[A]], <8 x double> undef, <4 x i32> <i32 4, i32 5, i32 6, i32 7> 107; CHECK-NEXT: [[TMP16:%.*]] = fmul <4 x double> [[TMP7]], [[SPLIT4]] 108; CHECK-NEXT: [[TMP17:%.*]] = fmul <4 x double> [[TMP15]], [[SPLIT5]] 109; CHECK-NEXT: [[TMP18:%.*]] = shufflevector <4 x double> [[TMP16]], <4 x double> [[TMP17]], <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7> 110; CHECK-NEXT: ret <8 x double> [[TMP18]] 111; 112entry: 113 %c = call <8 x double> @llvm.matrix.transpose(<8 x double> %a, i32 2, i32 4) 114 %res = fmul <8 x double> %c, %a 115 ret <8 x double> %res 116} 117