147 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -matrix-allow-contract=false -passes='lower-matrix-intrinsics' -S < %s | FileCheck %s3 4; Check that we do not emit shufflevectors to flatten the result of the5; transpose and store the columns directly.6define void @transpose_store(<8 x double> %a, ptr %Ptr) {7; CHECK-LABEL: @transpose_store(8; CHECK-NEXT: entry:9; CHECK-NEXT: [[SPLIT:%.*]] = shufflevector <8 x double> [[A:%.*]], <8 x double> poison, <2 x i32> <i32 0, i32 1>10; CHECK-NEXT: [[SPLIT1:%.*]] = shufflevector <8 x double> [[A]], <8 x double> poison, <2 x i32> <i32 2, i32 3>11; CHECK-NEXT: [[SPLIT2:%.*]] = shufflevector <8 x double> [[A]], <8 x double> poison, <2 x i32> <i32 4, i32 5>12; CHECK-NEXT: [[SPLIT3:%.*]] = shufflevector <8 x double> [[A]], <8 x double> poison, <2 x i32> <i32 6, i32 7>13; CHECK-NEXT: [[TMP0:%.*]] = extractelement <2 x double> [[SPLIT]], i64 014; CHECK-NEXT: [[TMP1:%.*]] = insertelement <4 x double> poison, double [[TMP0]], i64 015; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x double> [[SPLIT1]], i64 016; CHECK-NEXT: [[TMP3:%.*]] = insertelement <4 x double> [[TMP1]], double [[TMP2]], i64 117; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x double> [[SPLIT2]], i64 018; CHECK-NEXT: [[TMP5:%.*]] = insertelement <4 x double> [[TMP3]], double [[TMP4]], i64 219; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 020; CHECK-NEXT: [[TMP7:%.*]] = insertelement <4 x double> [[TMP5]], double [[TMP6]], i64 321; CHECK-NEXT: [[TMP8:%.*]] = extractelement <2 x double> [[SPLIT]], i64 122; CHECK-NEXT: [[TMP9:%.*]] = insertelement <4 x double> poison, double [[TMP8]], i64 023; CHECK-NEXT: [[TMP10:%.*]] = extractelement <2 x double> [[SPLIT1]], i64 124; CHECK-NEXT: [[TMP11:%.*]] = insertelement <4 x double> [[TMP9]], double [[TMP10]], i64 125; CHECK-NEXT: [[TMP12:%.*]] = extractelement <2 x double> [[SPLIT2]], i64 126; CHECK-NEXT: [[TMP13:%.*]] = insertelement <4 x double> [[TMP11]], double [[TMP12]], i64 227; CHECK-NEXT: [[TMP14:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 128; CHECK-NEXT: [[TMP15:%.*]] = insertelement <4 x double> [[TMP13]], double [[TMP14]], i64 329; CHECK-NEXT: store <4 x double> [[TMP7]], ptr [[PTR:%.*]], align 830; CHECK-NEXT: [[VEC_GEP:%.*]] = getelementptr double, ptr [[PTR]], i64 431; CHECK-NEXT: store <4 x double> [[TMP15]], ptr [[VEC_GEP]], align 832; CHECK-NEXT: ret void33;34entry:35 %c = call <8 x double> @llvm.matrix.transpose(<8 x double> %a, i32 2, i32 4)36 store <8 x double> %c, ptr %Ptr, align 837 ret void38}39 40declare <8 x double> @llvm.matrix.transpose(<8 x double>, i32, i32)41 42define <8 x double> @transpose_fadd(<8 x double> %a) {43; CHECK-LABEL: @transpose_fadd(44; CHECK-NEXT: entry:45; CHECK-NEXT: [[SPLIT:%.*]] = shufflevector <8 x double> [[A:%.*]], <8 x double> poison, <2 x i32> <i32 0, i32 1>46; CHECK-NEXT: [[SPLIT1:%.*]] = shufflevector <8 x double> [[A]], <8 x double> poison, <2 x i32> <i32 2, i32 3>47; CHECK-NEXT: [[SPLIT2:%.*]] = shufflevector <8 x double> [[A]], <8 x double> poison, <2 x i32> <i32 4, i32 5>48; CHECK-NEXT: [[SPLIT3:%.*]] = shufflevector <8 x double> [[A]], <8 x double> poison, <2 x i32> <i32 6, i32 7>49; CHECK-NEXT: [[TMP0:%.*]] = extractelement <2 x double> [[SPLIT]], i64 050; CHECK-NEXT: [[TMP1:%.*]] = insertelement <4 x double> poison, double [[TMP0]], i64 051; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x double> [[SPLIT1]], i64 052; CHECK-NEXT: [[TMP3:%.*]] = insertelement <4 x double> [[TMP1]], double [[TMP2]], i64 153; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x double> [[SPLIT2]], i64 054; CHECK-NEXT: [[TMP5:%.*]] = insertelement <4 x double> [[TMP3]], double [[TMP4]], i64 255; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 056; CHECK-NEXT: [[TMP7:%.*]] = insertelement <4 x double> [[TMP5]], double [[TMP6]], i64 357; CHECK-NEXT: [[TMP8:%.*]] = extractelement <2 x double> [[SPLIT]], i64 158; CHECK-NEXT: [[TMP9:%.*]] = insertelement <4 x double> poison, double [[TMP8]], i64 059; CHECK-NEXT: [[TMP10:%.*]] = extractelement <2 x double> [[SPLIT1]], i64 160; CHECK-NEXT: [[TMP11:%.*]] = insertelement <4 x double> [[TMP9]], double [[TMP10]], i64 161; CHECK-NEXT: [[TMP12:%.*]] = extractelement <2 x double> [[SPLIT2]], i64 162; CHECK-NEXT: [[TMP13:%.*]] = insertelement <4 x double> [[TMP11]], double [[TMP12]], i64 263; CHECK-NEXT: [[TMP14:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 164; CHECK-NEXT: [[TMP15:%.*]] = insertelement <4 x double> [[TMP13]], double [[TMP14]], i64 365; CHECK-NEXT: [[SPLIT4:%.*]] = shufflevector <8 x double> [[A]], <8 x double> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>66; CHECK-NEXT: [[SPLIT5:%.*]] = shufflevector <8 x double> [[A]], <8 x double> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>67; CHECK-NEXT: [[TMP16:%.*]] = fadd <4 x double> [[TMP7]], [[SPLIT4]]68; CHECK-NEXT: [[TMP17:%.*]] = fadd <4 x double> [[TMP15]], [[SPLIT5]]69; 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>70; CHECK-NEXT: ret <8 x double> [[TMP18]]71;72entry:73 %c = call <8 x double> @llvm.matrix.transpose(<8 x double> %a, i32 2, i32 4)74 %res = fadd <8 x double> %c, %a75 ret <8 x double> %res76}77 78define <8 x double> @transpose_fmul(<8 x double> %a) {79; CHECK-LABEL: @transpose_fmul(80; CHECK-NEXT: entry:81; CHECK-NEXT: [[SPLIT:%.*]] = shufflevector <8 x double> [[A:%.*]], <8 x double> poison, <2 x i32> <i32 0, i32 1>82; CHECK-NEXT: [[SPLIT1:%.*]] = shufflevector <8 x double> [[A]], <8 x double> poison, <2 x i32> <i32 2, i32 3>83; CHECK-NEXT: [[SPLIT2:%.*]] = shufflevector <8 x double> [[A]], <8 x double> poison, <2 x i32> <i32 4, i32 5>84; CHECK-NEXT: [[SPLIT3:%.*]] = shufflevector <8 x double> [[A]], <8 x double> poison, <2 x i32> <i32 6, i32 7>85; CHECK-NEXT: [[TMP0:%.*]] = extractelement <2 x double> [[SPLIT]], i64 086; CHECK-NEXT: [[TMP1:%.*]] = insertelement <4 x double> poison, double [[TMP0]], i64 087; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x double> [[SPLIT1]], i64 088; CHECK-NEXT: [[TMP3:%.*]] = insertelement <4 x double> [[TMP1]], double [[TMP2]], i64 189; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x double> [[SPLIT2]], i64 090; CHECK-NEXT: [[TMP5:%.*]] = insertelement <4 x double> [[TMP3]], double [[TMP4]], i64 291; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 092; CHECK-NEXT: [[TMP7:%.*]] = insertelement <4 x double> [[TMP5]], double [[TMP6]], i64 393; CHECK-NEXT: [[TMP8:%.*]] = extractelement <2 x double> [[SPLIT]], i64 194; CHECK-NEXT: [[TMP9:%.*]] = insertelement <4 x double> poison, double [[TMP8]], i64 095; CHECK-NEXT: [[TMP10:%.*]] = extractelement <2 x double> [[SPLIT1]], i64 196; CHECK-NEXT: [[TMP11:%.*]] = insertelement <4 x double> [[TMP9]], double [[TMP10]], i64 197; CHECK-NEXT: [[TMP12:%.*]] = extractelement <2 x double> [[SPLIT2]], i64 198; CHECK-NEXT: [[TMP13:%.*]] = insertelement <4 x double> [[TMP11]], double [[TMP12]], i64 299; CHECK-NEXT: [[TMP14:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 1100; CHECK-NEXT: [[TMP15:%.*]] = insertelement <4 x double> [[TMP13]], double [[TMP14]], i64 3101; CHECK-NEXT: [[SPLIT4:%.*]] = shufflevector <8 x double> [[A]], <8 x double> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>102; CHECK-NEXT: [[SPLIT5:%.*]] = shufflevector <8 x double> [[A]], <8 x double> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>103; CHECK-NEXT: [[TMP16:%.*]] = fmul <4 x double> [[TMP7]], [[SPLIT4]]104; CHECK-NEXT: [[TMP17:%.*]] = fmul <4 x double> [[TMP15]], [[SPLIT5]]105; 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>106; CHECK-NEXT: ret <8 x double> [[TMP18]]107;108entry:109 %c = call <8 x double> @llvm.matrix.transpose(<8 x double> %a, i32 2, i32 4)110 %res = fmul <8 x double> %c, %a111 ret <8 x double> %res112}113 114define <8 x double> @transpose_fneg(<8 x double> %a) {115; CHECK-LABEL: @transpose_fneg(116; CHECK-NEXT: entry:117; CHECK-NEXT: [[SPLIT:%.*]] = shufflevector <8 x double> [[A:%.*]], <8 x double> poison, <2 x i32> <i32 0, i32 1>118; CHECK-NEXT: [[SPLIT1:%.*]] = shufflevector <8 x double> [[A]], <8 x double> poison, <2 x i32> <i32 2, i32 3>119; CHECK-NEXT: [[SPLIT2:%.*]] = shufflevector <8 x double> [[A]], <8 x double> poison, <2 x i32> <i32 4, i32 5>120; CHECK-NEXT: [[SPLIT3:%.*]] = shufflevector <8 x double> [[A]], <8 x double> poison, <2 x i32> <i32 6, i32 7>121; CHECK-NEXT: [[TMP0:%.*]] = extractelement <2 x double> [[SPLIT]], i64 0122; CHECK-NEXT: [[TMP1:%.*]] = insertelement <4 x double> poison, double [[TMP0]], i64 0123; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x double> [[SPLIT1]], i64 0124; CHECK-NEXT: [[TMP3:%.*]] = insertelement <4 x double> [[TMP1]], double [[TMP2]], i64 1125; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x double> [[SPLIT2]], i64 0126; CHECK-NEXT: [[TMP5:%.*]] = insertelement <4 x double> [[TMP3]], double [[TMP4]], i64 2127; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 0128; CHECK-NEXT: [[TMP7:%.*]] = insertelement <4 x double> [[TMP5]], double [[TMP6]], i64 3129; CHECK-NEXT: [[TMP8:%.*]] = extractelement <2 x double> [[SPLIT]], i64 1130; CHECK-NEXT: [[TMP9:%.*]] = insertelement <4 x double> poison, double [[TMP8]], i64 0131; CHECK-NEXT: [[TMP10:%.*]] = extractelement <2 x double> [[SPLIT1]], i64 1132; CHECK-NEXT: [[TMP11:%.*]] = insertelement <4 x double> [[TMP9]], double [[TMP10]], i64 1133; CHECK-NEXT: [[TMP12:%.*]] = extractelement <2 x double> [[SPLIT2]], i64 1134; CHECK-NEXT: [[TMP13:%.*]] = insertelement <4 x double> [[TMP11]], double [[TMP12]], i64 2135; CHECK-NEXT: [[TMP14:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 1136; CHECK-NEXT: [[TMP15:%.*]] = insertelement <4 x double> [[TMP13]], double [[TMP14]], i64 3137; CHECK-NEXT: [[TMP16:%.*]] = fneg <4 x double> [[TMP7]]138; CHECK-NEXT: [[TMP17:%.*]] = fneg <4 x double> [[TMP15]]139; 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>140; CHECK-NEXT: ret <8 x double> [[TMP18]]141;142entry:143 %c = call <8 x double> @llvm.matrix.transpose(<8 x double> %a, i32 2, i32 4)144 %res = fneg <8 x double> %c145 ret <8 x double> %res146}147