191 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2; RUN: opt -passes='lower-matrix-intrinsics' -S < %s | FileCheck %s3 4define <4 x double> @multiply_right_transpose_2x2(<4 x double> %a, <4 x double> %b) {5; CHECK-LABEL: @multiply_right_transpose_2x2(6; CHECK-NEXT: entry:7; CHECK-NEXT: [[SPLIT:%.*]] = shufflevector <4 x double> [[A:%.*]], <4 x double> poison, <2 x i32> <i32 0, i32 1>8; CHECK-NEXT: [[SPLIT1:%.*]] = shufflevector <4 x double> [[A]], <4 x double> poison, <2 x i32> <i32 2, i32 3>9; CHECK-NEXT: [[SPLIT2:%.*]] = shufflevector <4 x double> [[B:%.*]], <4 x double> poison, <2 x i32> <i32 0, i32 1>10; CHECK-NEXT: [[SPLIT3:%.*]] = shufflevector <4 x double> [[B]], <4 x double> poison, <2 x i32> <i32 2, i32 3>11; CHECK-NEXT: [[BLOCK:%.*]] = shufflevector <2 x double> [[SPLIT]], <2 x double> poison, <1 x i32> zeroinitializer12; CHECK-NEXT: [[TMP0:%.*]] = extractelement <2 x double> [[SPLIT2]], i64 013; CHECK-NEXT: [[SPLAT_SPLATINSERT:%.*]] = insertelement <1 x double> poison, double [[TMP0]], i64 014; CHECK-NEXT: [[SPLAT_SPLAT:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT]], <1 x double> poison, <1 x i32> zeroinitializer15; CHECK-NEXT: [[TMP1:%.*]] = fmul <1 x double> [[BLOCK]], [[SPLAT_SPLAT]]16; CHECK-NEXT: [[BLOCK4:%.*]] = shufflevector <2 x double> [[SPLIT1]], <2 x double> poison, <1 x i32> zeroinitializer17; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 018; CHECK-NEXT: [[SPLAT_SPLATINSERT5:%.*]] = insertelement <1 x double> poison, double [[TMP2]], i64 019; CHECK-NEXT: [[SPLAT_SPLAT6:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT5]], <1 x double> poison, <1 x i32> zeroinitializer20; CHECK-NEXT: [[TMP3:%.*]] = fmul <1 x double> [[BLOCK4]], [[SPLAT_SPLAT6]]21; CHECK-NEXT: [[TMP4:%.*]] = fadd <1 x double> [[TMP1]], [[TMP3]]22; CHECK-NEXT: [[TMP5:%.*]] = shufflevector <1 x double> [[TMP4]], <1 x double> poison, <2 x i32> <i32 0, i32 poison>23; CHECK-NEXT: [[TMP6:%.*]] = shufflevector <2 x double> poison, <2 x double> [[TMP5]], <2 x i32> <i32 2, i32 1>24; CHECK-NEXT: [[BLOCK7:%.*]] = shufflevector <2 x double> [[SPLIT]], <2 x double> poison, <1 x i32> <i32 1>25; CHECK-NEXT: [[TMP7:%.*]] = extractelement <2 x double> [[SPLIT2]], i64 026; CHECK-NEXT: [[SPLAT_SPLATINSERT8:%.*]] = insertelement <1 x double> poison, double [[TMP7]], i64 027; CHECK-NEXT: [[SPLAT_SPLAT9:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT8]], <1 x double> poison, <1 x i32> zeroinitializer28; CHECK-NEXT: [[TMP8:%.*]] = fmul <1 x double> [[BLOCK7]], [[SPLAT_SPLAT9]]29; CHECK-NEXT: [[BLOCK10:%.*]] = shufflevector <2 x double> [[SPLIT1]], <2 x double> poison, <1 x i32> <i32 1>30; CHECK-NEXT: [[TMP9:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 031; CHECK-NEXT: [[SPLAT_SPLATINSERT11:%.*]] = insertelement <1 x double> poison, double [[TMP9]], i64 032; CHECK-NEXT: [[SPLAT_SPLAT12:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT11]], <1 x double> poison, <1 x i32> zeroinitializer33; CHECK-NEXT: [[TMP10:%.*]] = fmul <1 x double> [[BLOCK10]], [[SPLAT_SPLAT12]]34; CHECK-NEXT: [[TMP11:%.*]] = fadd <1 x double> [[TMP8]], [[TMP10]]35; CHECK-NEXT: [[TMP12:%.*]] = shufflevector <1 x double> [[TMP11]], <1 x double> poison, <2 x i32> <i32 0, i32 poison>36; CHECK-NEXT: [[TMP13:%.*]] = shufflevector <2 x double> [[TMP6]], <2 x double> [[TMP12]], <2 x i32> <i32 0, i32 2>37; CHECK-NEXT: [[BLOCK13:%.*]] = shufflevector <2 x double> [[SPLIT]], <2 x double> poison, <1 x i32> zeroinitializer38; CHECK-NEXT: [[TMP14:%.*]] = extractelement <2 x double> [[SPLIT2]], i64 139; CHECK-NEXT: [[SPLAT_SPLATINSERT14:%.*]] = insertelement <1 x double> poison, double [[TMP14]], i64 040; CHECK-NEXT: [[SPLAT_SPLAT15:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT14]], <1 x double> poison, <1 x i32> zeroinitializer41; CHECK-NEXT: [[TMP15:%.*]] = fmul <1 x double> [[BLOCK13]], [[SPLAT_SPLAT15]]42; CHECK-NEXT: [[BLOCK16:%.*]] = shufflevector <2 x double> [[SPLIT1]], <2 x double> poison, <1 x i32> zeroinitializer43; CHECK-NEXT: [[TMP16:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 144; CHECK-NEXT: [[SPLAT_SPLATINSERT17:%.*]] = insertelement <1 x double> poison, double [[TMP16]], i64 045; CHECK-NEXT: [[SPLAT_SPLAT18:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT17]], <1 x double> poison, <1 x i32> zeroinitializer46; CHECK-NEXT: [[TMP17:%.*]] = fmul <1 x double> [[BLOCK16]], [[SPLAT_SPLAT18]]47; CHECK-NEXT: [[TMP18:%.*]] = fadd <1 x double> [[TMP15]], [[TMP17]]48; CHECK-NEXT: [[TMP19:%.*]] = shufflevector <1 x double> [[TMP18]], <1 x double> poison, <2 x i32> <i32 0, i32 poison>49; CHECK-NEXT: [[TMP20:%.*]] = shufflevector <2 x double> poison, <2 x double> [[TMP19]], <2 x i32> <i32 2, i32 1>50; CHECK-NEXT: [[BLOCK19:%.*]] = shufflevector <2 x double> [[SPLIT]], <2 x double> poison, <1 x i32> <i32 1>51; CHECK-NEXT: [[TMP21:%.*]] = extractelement <2 x double> [[SPLIT2]], i64 152; CHECK-NEXT: [[SPLAT_SPLATINSERT20:%.*]] = insertelement <1 x double> poison, double [[TMP21]], i64 053; CHECK-NEXT: [[SPLAT_SPLAT21:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT20]], <1 x double> poison, <1 x i32> zeroinitializer54; CHECK-NEXT: [[TMP22:%.*]] = fmul <1 x double> [[BLOCK19]], [[SPLAT_SPLAT21]]55; CHECK-NEXT: [[BLOCK22:%.*]] = shufflevector <2 x double> [[SPLIT1]], <2 x double> poison, <1 x i32> <i32 1>56; CHECK-NEXT: [[TMP23:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 157; CHECK-NEXT: [[SPLAT_SPLATINSERT23:%.*]] = insertelement <1 x double> poison, double [[TMP23]], i64 058; CHECK-NEXT: [[SPLAT_SPLAT24:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT23]], <1 x double> poison, <1 x i32> zeroinitializer59; CHECK-NEXT: [[TMP24:%.*]] = fmul <1 x double> [[BLOCK22]], [[SPLAT_SPLAT24]]60; CHECK-NEXT: [[TMP25:%.*]] = fadd <1 x double> [[TMP22]], [[TMP24]]61; CHECK-NEXT: [[TMP26:%.*]] = shufflevector <1 x double> [[TMP25]], <1 x double> poison, <2 x i32> <i32 0, i32 poison>62; CHECK-NEXT: [[TMP27:%.*]] = shufflevector <2 x double> [[TMP20]], <2 x double> [[TMP26]], <2 x i32> <i32 0, i32 2>63; CHECK-NEXT: [[TMP28:%.*]] = shufflevector <2 x double> [[TMP13]], <2 x double> [[TMP27]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>64; CHECK-NEXT: ret <4 x double> [[TMP28]]65;66entry:67 %bt = call <4 x double> @llvm.matrix.transpose.v4f64(<4 x double> %b, i32 2, i32 2)68 %c = call <4 x double> @llvm.matrix.multiply.v4f64.v4f64.v4f64(<4 x double> %a, <4 x double> %bt, i32 2, i32 2, i32 2)69 ret <4 x double> %c70}71 72; Lower transpose if not all uses are folded.73define <4 x double> @multiply_right_transpose_2x3x2(<6 x double> %a, <6 x double> %b, ptr %p) {74; CHECK-LABEL: @multiply_right_transpose_2x3x2(75; CHECK-NEXT: entry:76; CHECK-NEXT: [[SPLIT39:%.*]] = shufflevector <6 x double> [[B:%.*]], <6 x double> poison, <2 x i32> <i32 0, i32 1>77; CHECK-NEXT: [[SPLIT40:%.*]] = shufflevector <6 x double> [[B]], <6 x double> poison, <2 x i32> <i32 2, i32 3>78; CHECK-NEXT: [[SPLIT41:%.*]] = shufflevector <6 x double> [[B]], <6 x double> poison, <2 x i32> <i32 4, i32 5>79; CHECK-NEXT: [[TMP0:%.*]] = extractelement <2 x double> [[SPLIT39]], i64 080; CHECK-NEXT: [[TMP1:%.*]] = insertelement <3 x double> poison, double [[TMP0]], i64 081; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x double> [[SPLIT40]], i64 082; CHECK-NEXT: [[TMP3:%.*]] = insertelement <3 x double> [[TMP1]], double [[TMP2]], i64 183; CHECK-NEXT: [[TMP4:%.*]] = extractelement <2 x double> [[SPLIT41]], i64 084; CHECK-NEXT: [[TMP5:%.*]] = insertelement <3 x double> [[TMP3]], double [[TMP4]], i64 285; CHECK-NEXT: [[TMP6:%.*]] = extractelement <2 x double> [[SPLIT39]], i64 186; CHECK-NEXT: [[TMP7:%.*]] = insertelement <3 x double> poison, double [[TMP6]], i64 087; CHECK-NEXT: [[TMP8:%.*]] = extractelement <2 x double> [[SPLIT40]], i64 188; CHECK-NEXT: [[TMP9:%.*]] = insertelement <3 x double> [[TMP7]], double [[TMP8]], i64 189; CHECK-NEXT: [[TMP10:%.*]] = extractelement <2 x double> [[SPLIT41]], i64 190; CHECK-NEXT: [[TMP11:%.*]] = insertelement <3 x double> [[TMP9]], double [[TMP10]], i64 291; CHECK-NEXT: store <3 x double> [[TMP5]], ptr [[P:%.*]], align 1692; CHECK-NEXT: [[VEC_GEP:%.*]] = getelementptr double, ptr [[P]], i64 393; CHECK-NEXT: store <3 x double> [[TMP11]], ptr [[VEC_GEP]], align 894; CHECK-NEXT: [[SPLIT:%.*]] = shufflevector <6 x double> [[A:%.*]], <6 x double> poison, <2 x i32> <i32 0, i32 1>95; CHECK-NEXT: [[SPLIT1:%.*]] = shufflevector <6 x double> [[A]], <6 x double> poison, <2 x i32> <i32 2, i32 3>96; CHECK-NEXT: [[SPLIT2:%.*]] = shufflevector <6 x double> [[A]], <6 x double> poison, <2 x i32> <i32 4, i32 5>97; CHECK-NEXT: [[SPLIT3:%.*]] = shufflevector <6 x double> [[B]], <6 x double> poison, <2 x i32> <i32 0, i32 1>98; CHECK-NEXT: [[SPLIT4:%.*]] = shufflevector <6 x double> [[B]], <6 x double> poison, <2 x i32> <i32 2, i32 3>99; CHECK-NEXT: [[SPLIT5:%.*]] = shufflevector <6 x double> [[B]], <6 x double> poison, <2 x i32> <i32 4, i32 5>100; CHECK-NEXT: [[BLOCK:%.*]] = shufflevector <2 x double> [[SPLIT]], <2 x double> poison, <1 x i32> zeroinitializer101; CHECK-NEXT: [[TMP12:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 0102; CHECK-NEXT: [[SPLAT_SPLATINSERT:%.*]] = insertelement <1 x double> poison, double [[TMP12]], i64 0103; CHECK-NEXT: [[SPLAT_SPLAT:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT]], <1 x double> poison, <1 x i32> zeroinitializer104; CHECK-NEXT: [[TMP13:%.*]] = fmul <1 x double> [[BLOCK]], [[SPLAT_SPLAT]]105; CHECK-NEXT: [[BLOCK6:%.*]] = shufflevector <2 x double> [[SPLIT1]], <2 x double> poison, <1 x i32> zeroinitializer106; CHECK-NEXT: [[TMP14:%.*]] = extractelement <2 x double> [[SPLIT4]], i64 0107; CHECK-NEXT: [[SPLAT_SPLATINSERT7:%.*]] = insertelement <1 x double> poison, double [[TMP14]], i64 0108; CHECK-NEXT: [[SPLAT_SPLAT8:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT7]], <1 x double> poison, <1 x i32> zeroinitializer109; CHECK-NEXT: [[TMP15:%.*]] = fmul <1 x double> [[BLOCK6]], [[SPLAT_SPLAT8]]110; CHECK-NEXT: [[TMP16:%.*]] = fadd <1 x double> [[TMP13]], [[TMP15]]111; CHECK-NEXT: [[BLOCK9:%.*]] = shufflevector <2 x double> [[SPLIT2]], <2 x double> poison, <1 x i32> zeroinitializer112; CHECK-NEXT: [[TMP17:%.*]] = extractelement <2 x double> [[SPLIT5]], i64 0113; CHECK-NEXT: [[SPLAT_SPLATINSERT10:%.*]] = insertelement <1 x double> poison, double [[TMP17]], i64 0114; CHECK-NEXT: [[SPLAT_SPLAT11:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT10]], <1 x double> poison, <1 x i32> zeroinitializer115; CHECK-NEXT: [[TMP18:%.*]] = fmul <1 x double> [[BLOCK9]], [[SPLAT_SPLAT11]]116; CHECK-NEXT: [[TMP19:%.*]] = fadd <1 x double> [[TMP16]], [[TMP18]]117; CHECK-NEXT: [[TMP20:%.*]] = shufflevector <1 x double> [[TMP19]], <1 x double> poison, <2 x i32> <i32 0, i32 poison>118; CHECK-NEXT: [[TMP21:%.*]] = shufflevector <2 x double> poison, <2 x double> [[TMP20]], <2 x i32> <i32 2, i32 1>119; CHECK-NEXT: [[BLOCK12:%.*]] = shufflevector <2 x double> [[SPLIT]], <2 x double> poison, <1 x i32> <i32 1>120; CHECK-NEXT: [[TMP22:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 0121; CHECK-NEXT: [[SPLAT_SPLATINSERT13:%.*]] = insertelement <1 x double> poison, double [[TMP22]], i64 0122; CHECK-NEXT: [[SPLAT_SPLAT14:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT13]], <1 x double> poison, <1 x i32> zeroinitializer123; CHECK-NEXT: [[TMP23:%.*]] = fmul <1 x double> [[BLOCK12]], [[SPLAT_SPLAT14]]124; CHECK-NEXT: [[BLOCK15:%.*]] = shufflevector <2 x double> [[SPLIT1]], <2 x double> poison, <1 x i32> <i32 1>125; CHECK-NEXT: [[TMP24:%.*]] = extractelement <2 x double> [[SPLIT4]], i64 0126; CHECK-NEXT: [[SPLAT_SPLATINSERT16:%.*]] = insertelement <1 x double> poison, double [[TMP24]], i64 0127; CHECK-NEXT: [[SPLAT_SPLAT17:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT16]], <1 x double> poison, <1 x i32> zeroinitializer128; CHECK-NEXT: [[TMP25:%.*]] = fmul <1 x double> [[BLOCK15]], [[SPLAT_SPLAT17]]129; CHECK-NEXT: [[TMP26:%.*]] = fadd <1 x double> [[TMP23]], [[TMP25]]130; CHECK-NEXT: [[BLOCK18:%.*]] = shufflevector <2 x double> [[SPLIT2]], <2 x double> poison, <1 x i32> <i32 1>131; CHECK-NEXT: [[TMP27:%.*]] = extractelement <2 x double> [[SPLIT5]], i64 0132; CHECK-NEXT: [[SPLAT_SPLATINSERT19:%.*]] = insertelement <1 x double> poison, double [[TMP27]], i64 0133; CHECK-NEXT: [[SPLAT_SPLAT20:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT19]], <1 x double> poison, <1 x i32> zeroinitializer134; CHECK-NEXT: [[TMP28:%.*]] = fmul <1 x double> [[BLOCK18]], [[SPLAT_SPLAT20]]135; CHECK-NEXT: [[TMP29:%.*]] = fadd <1 x double> [[TMP26]], [[TMP28]]136; CHECK-NEXT: [[TMP30:%.*]] = shufflevector <1 x double> [[TMP29]], <1 x double> poison, <2 x i32> <i32 0, i32 poison>137; CHECK-NEXT: [[TMP31:%.*]] = shufflevector <2 x double> [[TMP21]], <2 x double> [[TMP30]], <2 x i32> <i32 0, i32 2>138; CHECK-NEXT: [[BLOCK21:%.*]] = shufflevector <2 x double> [[SPLIT]], <2 x double> poison, <1 x i32> zeroinitializer139; CHECK-NEXT: [[TMP32:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 1140; CHECK-NEXT: [[SPLAT_SPLATINSERT22:%.*]] = insertelement <1 x double> poison, double [[TMP32]], i64 0141; CHECK-NEXT: [[SPLAT_SPLAT23:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT22]], <1 x double> poison, <1 x i32> zeroinitializer142; CHECK-NEXT: [[TMP33:%.*]] = fmul <1 x double> [[BLOCK21]], [[SPLAT_SPLAT23]]143; CHECK-NEXT: [[BLOCK24:%.*]] = shufflevector <2 x double> [[SPLIT1]], <2 x double> poison, <1 x i32> zeroinitializer144; CHECK-NEXT: [[TMP34:%.*]] = extractelement <2 x double> [[SPLIT4]], i64 1145; CHECK-NEXT: [[SPLAT_SPLATINSERT25:%.*]] = insertelement <1 x double> poison, double [[TMP34]], i64 0146; CHECK-NEXT: [[SPLAT_SPLAT26:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT25]], <1 x double> poison, <1 x i32> zeroinitializer147; CHECK-NEXT: [[TMP35:%.*]] = fmul <1 x double> [[BLOCK24]], [[SPLAT_SPLAT26]]148; CHECK-NEXT: [[TMP36:%.*]] = fadd <1 x double> [[TMP33]], [[TMP35]]149; CHECK-NEXT: [[BLOCK27:%.*]] = shufflevector <2 x double> [[SPLIT2]], <2 x double> poison, <1 x i32> zeroinitializer150; CHECK-NEXT: [[TMP37:%.*]] = extractelement <2 x double> [[SPLIT5]], i64 1151; CHECK-NEXT: [[SPLAT_SPLATINSERT28:%.*]] = insertelement <1 x double> poison, double [[TMP37]], i64 0152; CHECK-NEXT: [[SPLAT_SPLAT29:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT28]], <1 x double> poison, <1 x i32> zeroinitializer153; CHECK-NEXT: [[TMP38:%.*]] = fmul <1 x double> [[BLOCK27]], [[SPLAT_SPLAT29]]154; CHECK-NEXT: [[TMP39:%.*]] = fadd <1 x double> [[TMP36]], [[TMP38]]155; CHECK-NEXT: [[TMP40:%.*]] = shufflevector <1 x double> [[TMP39]], <1 x double> poison, <2 x i32> <i32 0, i32 poison>156; CHECK-NEXT: [[TMP41:%.*]] = shufflevector <2 x double> poison, <2 x double> [[TMP40]], <2 x i32> <i32 2, i32 1>157; CHECK-NEXT: [[BLOCK30:%.*]] = shufflevector <2 x double> [[SPLIT]], <2 x double> poison, <1 x i32> <i32 1>158; CHECK-NEXT: [[TMP42:%.*]] = extractelement <2 x double> [[SPLIT3]], i64 1159; CHECK-NEXT: [[SPLAT_SPLATINSERT31:%.*]] = insertelement <1 x double> poison, double [[TMP42]], i64 0160; CHECK-NEXT: [[SPLAT_SPLAT32:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT31]], <1 x double> poison, <1 x i32> zeroinitializer161; CHECK-NEXT: [[TMP43:%.*]] = fmul <1 x double> [[BLOCK30]], [[SPLAT_SPLAT32]]162; CHECK-NEXT: [[BLOCK33:%.*]] = shufflevector <2 x double> [[SPLIT1]], <2 x double> poison, <1 x i32> <i32 1>163; CHECK-NEXT: [[TMP44:%.*]] = extractelement <2 x double> [[SPLIT4]], i64 1164; CHECK-NEXT: [[SPLAT_SPLATINSERT34:%.*]] = insertelement <1 x double> poison, double [[TMP44]], i64 0165; CHECK-NEXT: [[SPLAT_SPLAT35:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT34]], <1 x double> poison, <1 x i32> zeroinitializer166; CHECK-NEXT: [[TMP45:%.*]] = fmul <1 x double> [[BLOCK33]], [[SPLAT_SPLAT35]]167; CHECK-NEXT: [[TMP46:%.*]] = fadd <1 x double> [[TMP43]], [[TMP45]]168; CHECK-NEXT: [[BLOCK36:%.*]] = shufflevector <2 x double> [[SPLIT2]], <2 x double> poison, <1 x i32> <i32 1>169; CHECK-NEXT: [[TMP47:%.*]] = extractelement <2 x double> [[SPLIT5]], i64 1170; CHECK-NEXT: [[SPLAT_SPLATINSERT37:%.*]] = insertelement <1 x double> poison, double [[TMP47]], i64 0171; CHECK-NEXT: [[SPLAT_SPLAT38:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT37]], <1 x double> poison, <1 x i32> zeroinitializer172; CHECK-NEXT: [[TMP48:%.*]] = fmul <1 x double> [[BLOCK36]], [[SPLAT_SPLAT38]]173; CHECK-NEXT: [[TMP49:%.*]] = fadd <1 x double> [[TMP46]], [[TMP48]]174; CHECK-NEXT: [[TMP50:%.*]] = shufflevector <1 x double> [[TMP49]], <1 x double> poison, <2 x i32> <i32 0, i32 poison>175; CHECK-NEXT: [[TMP51:%.*]] = shufflevector <2 x double> [[TMP41]], <2 x double> [[TMP50]], <2 x i32> <i32 0, i32 2>176; CHECK-NEXT: [[TMP52:%.*]] = shufflevector <2 x double> [[TMP31]], <2 x double> [[TMP51]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>177; CHECK-NEXT: ret <4 x double> [[TMP52]]178;179entry:180 %bt = call <6 x double> @llvm.matrix.transpose.v6f64(<6 x double> %b, i32 2, i32 3)181 store <6 x double> %bt, ptr %p, align 16182 %c = call <4 x double> @llvm.matrix.multiply.v4f64.v6f64.v6f64(<6 x double> %a, <6 x double> %bt, i32 2, i32 3, i32 2)183 ret <4 x double> %c184}185 186 187declare <4 x double> @llvm.matrix.multiply.v4f64.v4f64.v4f64(<4 x double>, <4 x double>, i32, i32, i32)188declare <4 x double> @llvm.matrix.multiply.v4f64.v6f64.v6f64(<6 x double>, <6 x double>, i32, i32, i32)189declare <4 x double> @llvm.matrix.transpose.v4f64(<4 x double>, i32, i32)190declare <6 x double> @llvm.matrix.transpose.v6f64(<6 x double>, i32, i32)191