brintos

brintos / llvm-project-archived public Read only

0
0
Text · 7.9 KiB · 13a7273 Raw
112 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -passes=lower-matrix-intrinsics -matrix-default-layout=row-major -S < %s | FileCheck --check-prefix=RM %s3 4define <8 x double> @transpose(<8 x double> %a) {5; RM-LABEL: @transpose(6; RM-NEXT:  entry:7; RM-NEXT:    [[SPLIT:%.*]] = shufflevector <8 x double> [[A:%.*]], <8 x double> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>8; RM-NEXT:    [[SPLIT1:%.*]] = shufflevector <8 x double> [[A]], <8 x double> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>9; RM-NEXT:    [[TMP0:%.*]] = extractelement <4 x double> [[SPLIT]], i64 010; RM-NEXT:    [[TMP1:%.*]] = insertelement <2 x double> poison, double [[TMP0]], i64 011; RM-NEXT:    [[TMP2:%.*]] = extractelement <4 x double> [[SPLIT1]], i64 012; RM-NEXT:    [[TMP3:%.*]] = insertelement <2 x double> [[TMP1]], double [[TMP2]], i64 113; RM-NEXT:    [[TMP4:%.*]] = extractelement <4 x double> [[SPLIT]], i64 114; RM-NEXT:    [[TMP5:%.*]] = insertelement <2 x double> poison, double [[TMP4]], i64 015; RM-NEXT:    [[TMP6:%.*]] = extractelement <4 x double> [[SPLIT1]], i64 116; RM-NEXT:    [[TMP7:%.*]] = insertelement <2 x double> [[TMP5]], double [[TMP6]], i64 117; RM-NEXT:    [[TMP8:%.*]] = extractelement <4 x double> [[SPLIT]], i64 218; RM-NEXT:    [[TMP9:%.*]] = insertelement <2 x double> poison, double [[TMP8]], i64 019; RM-NEXT:    [[TMP10:%.*]] = extractelement <4 x double> [[SPLIT1]], i64 220; RM-NEXT:    [[TMP11:%.*]] = insertelement <2 x double> [[TMP9]], double [[TMP10]], i64 121; RM-NEXT:    [[TMP12:%.*]] = extractelement <4 x double> [[SPLIT]], i64 322; RM-NEXT:    [[TMP13:%.*]] = insertelement <2 x double> poison, double [[TMP12]], i64 023; RM-NEXT:    [[TMP14:%.*]] = extractelement <4 x double> [[SPLIT1]], i64 324; RM-NEXT:    [[TMP15:%.*]] = insertelement <2 x double> [[TMP13]], double [[TMP14]], i64 125; RM-NEXT:    [[TMP16:%.*]] = shufflevector <2 x double> [[TMP3]], <2 x double> [[TMP7]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>26; RM-NEXT:    [[TMP17:%.*]] = shufflevector <2 x double> [[TMP11]], <2 x double> [[TMP15]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>27; RM-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>28; RM-NEXT:    ret <8 x double> [[TMP18]]29;30entry:31  %c  = call <8 x double> @llvm.matrix.transpose(<8 x double> %a, i32 2, i32 4)32  ret <8 x double> %c33}34 35declare <8 x double> @llvm.matrix.transpose(<8 x double>, i32, i32)36 37define <8 x double> @transpose_single_column(<8 x double> %a) {38; RM-LABEL: @transpose_single_column(39; RM-NEXT:  entry:40; RM-NEXT:    [[SPLIT:%.*]] = shufflevector <8 x double> [[A:%.*]], <8 x double> poison, <1 x i32> zeroinitializer41; RM-NEXT:    [[SPLIT1:%.*]] = shufflevector <8 x double> [[A]], <8 x double> poison, <1 x i32> <i32 1>42; RM-NEXT:    [[SPLIT2:%.*]] = shufflevector <8 x double> [[A]], <8 x double> poison, <1 x i32> <i32 2>43; RM-NEXT:    [[SPLIT3:%.*]] = shufflevector <8 x double> [[A]], <8 x double> poison, <1 x i32> <i32 3>44; RM-NEXT:    [[SPLIT4:%.*]] = shufflevector <8 x double> [[A]], <8 x double> poison, <1 x i32> <i32 4>45; RM-NEXT:    [[SPLIT5:%.*]] = shufflevector <8 x double> [[A]], <8 x double> poison, <1 x i32> <i32 5>46; RM-NEXT:    [[SPLIT6:%.*]] = shufflevector <8 x double> [[A]], <8 x double> poison, <1 x i32> <i32 6>47; RM-NEXT:    [[SPLIT7:%.*]] = shufflevector <8 x double> [[A]], <8 x double> poison, <1 x i32> <i32 7>48; RM-NEXT:    [[TMP0:%.*]] = extractelement <1 x double> [[SPLIT]], i64 049; RM-NEXT:    [[TMP1:%.*]] = insertelement <8 x double> poison, double [[TMP0]], i64 050; RM-NEXT:    [[TMP2:%.*]] = extractelement <1 x double> [[SPLIT1]], i64 051; RM-NEXT:    [[TMP3:%.*]] = insertelement <8 x double> [[TMP1]], double [[TMP2]], i64 152; RM-NEXT:    [[TMP4:%.*]] = extractelement <1 x double> [[SPLIT2]], i64 053; RM-NEXT:    [[TMP5:%.*]] = insertelement <8 x double> [[TMP3]], double [[TMP4]], i64 254; RM-NEXT:    [[TMP6:%.*]] = extractelement <1 x double> [[SPLIT3]], i64 055; RM-NEXT:    [[TMP7:%.*]] = insertelement <8 x double> [[TMP5]], double [[TMP6]], i64 356; RM-NEXT:    [[TMP8:%.*]] = extractelement <1 x double> [[SPLIT4]], i64 057; RM-NEXT:    [[TMP9:%.*]] = insertelement <8 x double> [[TMP7]], double [[TMP8]], i64 458; RM-NEXT:    [[TMP10:%.*]] = extractelement <1 x double> [[SPLIT5]], i64 059; RM-NEXT:    [[TMP11:%.*]] = insertelement <8 x double> [[TMP9]], double [[TMP10]], i64 560; RM-NEXT:    [[TMP12:%.*]] = extractelement <1 x double> [[SPLIT6]], i64 061; RM-NEXT:    [[TMP13:%.*]] = insertelement <8 x double> [[TMP11]], double [[TMP12]], i64 662; RM-NEXT:    [[TMP14:%.*]] = extractelement <1 x double> [[SPLIT7]], i64 063; RM-NEXT:    [[TMP15:%.*]] = insertelement <8 x double> [[TMP13]], double [[TMP14]], i64 764; RM-NEXT:    ret <8 x double> [[TMP15]]65;66entry:67  %c  = call <8 x double> @llvm.matrix.transpose(<8 x double> %a, i32 8, i32 1)68  ret <8 x double> %c69}70 71declare <12 x double> @llvm.matrix.transpose.v12f64(<12 x double>, i32, i32)72 73define <12 x double> @transpose_double_3x4(<12 x double> %a) {74; RM-LABEL: @transpose_double_3x4(75; RM-NEXT:  entry:76; RM-NEXT:    [[SPLIT:%.*]] = shufflevector <12 x double> [[A:%.*]], <12 x double> poison, <4 x i32> <i32 0, i32 1, i32 2, i32 3>77; RM-NEXT:    [[SPLIT1:%.*]] = shufflevector <12 x double> [[A]], <12 x double> poison, <4 x i32> <i32 4, i32 5, i32 6, i32 7>78; RM-NEXT:    [[SPLIT2:%.*]] = shufflevector <12 x double> [[A]], <12 x double> poison, <4 x i32> <i32 8, i32 9, i32 10, i32 11>79; RM-NEXT:    [[TMP0:%.*]] = extractelement <4 x double> [[SPLIT]], i64 080; RM-NEXT:    [[TMP1:%.*]] = insertelement <3 x double> poison, double [[TMP0]], i64 081; RM-NEXT:    [[TMP2:%.*]] = extractelement <4 x double> [[SPLIT1]], i64 082; RM-NEXT:    [[TMP3:%.*]] = insertelement <3 x double> [[TMP1]], double [[TMP2]], i64 183; RM-NEXT:    [[TMP4:%.*]] = extractelement <4 x double> [[SPLIT2]], i64 084; RM-NEXT:    [[TMP5:%.*]] = insertelement <3 x double> [[TMP3]], double [[TMP4]], i64 285; RM-NEXT:    [[TMP6:%.*]] = extractelement <4 x double> [[SPLIT]], i64 186; RM-NEXT:    [[TMP7:%.*]] = insertelement <3 x double> poison, double [[TMP6]], i64 087; RM-NEXT:    [[TMP8:%.*]] = extractelement <4 x double> [[SPLIT1]], i64 188; RM-NEXT:    [[TMP9:%.*]] = insertelement <3 x double> [[TMP7]], double [[TMP8]], i64 189; RM-NEXT:    [[TMP10:%.*]] = extractelement <4 x double> [[SPLIT2]], i64 190; RM-NEXT:    [[TMP11:%.*]] = insertelement <3 x double> [[TMP9]], double [[TMP10]], i64 291; RM-NEXT:    [[TMP12:%.*]] = extractelement <4 x double> [[SPLIT]], i64 292; RM-NEXT:    [[TMP13:%.*]] = insertelement <3 x double> poison, double [[TMP12]], i64 093; RM-NEXT:    [[TMP14:%.*]] = extractelement <4 x double> [[SPLIT1]], i64 294; RM-NEXT:    [[TMP15:%.*]] = insertelement <3 x double> [[TMP13]], double [[TMP14]], i64 195; RM-NEXT:    [[TMP16:%.*]] = extractelement <4 x double> [[SPLIT2]], i64 296; RM-NEXT:    [[TMP17:%.*]] = insertelement <3 x double> [[TMP15]], double [[TMP16]], i64 297; RM-NEXT:    [[TMP18:%.*]] = extractelement <4 x double> [[SPLIT]], i64 398; RM-NEXT:    [[TMP19:%.*]] = insertelement <3 x double> poison, double [[TMP18]], i64 099; RM-NEXT:    [[TMP20:%.*]] = extractelement <4 x double> [[SPLIT1]], i64 3100; RM-NEXT:    [[TMP21:%.*]] = insertelement <3 x double> [[TMP19]], double [[TMP20]], i64 1101; RM-NEXT:    [[TMP22:%.*]] = extractelement <4 x double> [[SPLIT2]], i64 3102; RM-NEXT:    [[TMP23:%.*]] = insertelement <3 x double> [[TMP21]], double [[TMP22]], i64 2103; RM-NEXT:    [[TMP24:%.*]] = shufflevector <3 x double> [[TMP5]], <3 x double> [[TMP11]], <6 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5>104; RM-NEXT:    [[TMP25:%.*]] = shufflevector <3 x double> [[TMP17]], <3 x double> [[TMP23]], <6 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5>105; RM-NEXT:    [[TMP26:%.*]] = shufflevector <6 x double> [[TMP24]], <6 x double> [[TMP25]], <12 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11>106; RM-NEXT:    ret <12 x double> [[TMP26]]107;108entry:109  %c  = call <12 x double> @llvm.matrix.transpose.v12f64(<12 x double> %a, i32 3, i32 4)110  ret <12 x double> %c111}112