210 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 4define <4 x float> @preserve_fmf_fast(<4 x float> %m, <4 x float> %n) {5; CHECK-LABEL: @preserve_fmf_fast(6; CHECK-NEXT: [[SPLIT:%.*]] = shufflevector <4 x float> [[M:%.*]], <4 x float> poison, <2 x i32> <i32 0, i32 1>7; CHECK-NEXT: [[SPLIT1:%.*]] = shufflevector <4 x float> [[M]], <4 x float> poison, <2 x i32> <i32 2, i32 3>8; CHECK-NEXT: [[SPLIT2:%.*]] = shufflevector <4 x float> [[N:%.*]], <4 x float> poison, <2 x i32> <i32 0, i32 1>9; CHECK-NEXT: [[SPLIT3:%.*]] = shufflevector <4 x float> [[N]], <4 x float> poison, <2 x i32> <i32 2, i32 3>10; CHECK-NEXT: [[BLOCK:%.*]] = shufflevector <2 x float> [[SPLIT]], <2 x float> poison, <1 x i32> zeroinitializer11; CHECK-NEXT: [[TMP1:%.*]] = extractelement <2 x float> [[SPLIT2]], i64 012; CHECK-NEXT: [[SPLAT_SPLATINSERT:%.*]] = insertelement <1 x float> poison, float [[TMP1]], i64 013; CHECK-NEXT: [[SPLAT_SPLAT:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT]], <1 x float> poison, <1 x i32> zeroinitializer14; CHECK-NEXT: [[TMP2:%.*]] = fmul fast <1 x float> [[BLOCK]], [[SPLAT_SPLAT]]15; CHECK-NEXT: [[BLOCK4:%.*]] = shufflevector <2 x float> [[SPLIT1]], <2 x float> poison, <1 x i32> zeroinitializer16; CHECK-NEXT: [[TMP3:%.*]] = extractelement <2 x float> [[SPLIT2]], i64 117; CHECK-NEXT: [[SPLAT_SPLATINSERT5:%.*]] = insertelement <1 x float> poison, float [[TMP3]], i64 018; CHECK-NEXT: [[SPLAT_SPLAT6:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT5]], <1 x float> poison, <1 x i32> zeroinitializer19; CHECK-NEXT: [[TMP4:%.*]] = call fast <1 x float> @llvm.fmuladd.v1f32(<1 x float> [[BLOCK4]], <1 x float> [[SPLAT_SPLAT6]], <1 x float> [[TMP2]])20; CHECK-NEXT: [[TMP5:%.*]] = shufflevector <1 x float> [[TMP4]], <1 x float> poison, <2 x i32> <i32 0, i32 poison>21; CHECK-NEXT: [[TMP6:%.*]] = shufflevector <2 x float> poison, <2 x float> [[TMP5]], <2 x i32> <i32 2, i32 1>22; CHECK-NEXT: [[BLOCK7:%.*]] = shufflevector <2 x float> [[SPLIT]], <2 x float> poison, <1 x i32> <i32 1>23; CHECK-NEXT: [[TMP7:%.*]] = extractelement <2 x float> [[SPLIT2]], i64 024; CHECK-NEXT: [[SPLAT_SPLATINSERT8:%.*]] = insertelement <1 x float> poison, float [[TMP7]], i64 025; CHECK-NEXT: [[SPLAT_SPLAT9:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT8]], <1 x float> poison, <1 x i32> zeroinitializer26; CHECK-NEXT: [[TMP8:%.*]] = fmul fast <1 x float> [[BLOCK7]], [[SPLAT_SPLAT9]]27; CHECK-NEXT: [[BLOCK10:%.*]] = shufflevector <2 x float> [[SPLIT1]], <2 x float> poison, <1 x i32> <i32 1>28; CHECK-NEXT: [[TMP9:%.*]] = extractelement <2 x float> [[SPLIT2]], i64 129; CHECK-NEXT: [[SPLAT_SPLATINSERT11:%.*]] = insertelement <1 x float> poison, float [[TMP9]], i64 030; CHECK-NEXT: [[SPLAT_SPLAT12:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT11]], <1 x float> poison, <1 x i32> zeroinitializer31; CHECK-NEXT: [[TMP10:%.*]] = call fast <1 x float> @llvm.fmuladd.v1f32(<1 x float> [[BLOCK10]], <1 x float> [[SPLAT_SPLAT12]], <1 x float> [[TMP8]])32; CHECK-NEXT: [[TMP11:%.*]] = shufflevector <1 x float> [[TMP10]], <1 x float> poison, <2 x i32> <i32 0, i32 poison>33; CHECK-NEXT: [[TMP12:%.*]] = shufflevector <2 x float> [[TMP6]], <2 x float> [[TMP11]], <2 x i32> <i32 0, i32 2>34; CHECK-NEXT: [[BLOCK13:%.*]] = shufflevector <2 x float> [[SPLIT]], <2 x float> poison, <1 x i32> zeroinitializer35; CHECK-NEXT: [[TMP13:%.*]] = extractelement <2 x float> [[SPLIT3]], i64 036; CHECK-NEXT: [[SPLAT_SPLATINSERT14:%.*]] = insertelement <1 x float> poison, float [[TMP13]], i64 037; CHECK-NEXT: [[SPLAT_SPLAT15:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT14]], <1 x float> poison, <1 x i32> zeroinitializer38; CHECK-NEXT: [[TMP14:%.*]] = fmul fast <1 x float> [[BLOCK13]], [[SPLAT_SPLAT15]]39; CHECK-NEXT: [[BLOCK16:%.*]] = shufflevector <2 x float> [[SPLIT1]], <2 x float> poison, <1 x i32> zeroinitializer40; CHECK-NEXT: [[TMP15:%.*]] = extractelement <2 x float> [[SPLIT3]], i64 141; CHECK-NEXT: [[SPLAT_SPLATINSERT17:%.*]] = insertelement <1 x float> poison, float [[TMP15]], i64 042; CHECK-NEXT: [[SPLAT_SPLAT18:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT17]], <1 x float> poison, <1 x i32> zeroinitializer43; CHECK-NEXT: [[TMP16:%.*]] = call fast <1 x float> @llvm.fmuladd.v1f32(<1 x float> [[BLOCK16]], <1 x float> [[SPLAT_SPLAT18]], <1 x float> [[TMP14]])44; CHECK-NEXT: [[TMP17:%.*]] = shufflevector <1 x float> [[TMP16]], <1 x float> poison, <2 x i32> <i32 0, i32 poison>45; CHECK-NEXT: [[TMP18:%.*]] = shufflevector <2 x float> poison, <2 x float> [[TMP17]], <2 x i32> <i32 2, i32 1>46; CHECK-NEXT: [[BLOCK19:%.*]] = shufflevector <2 x float> [[SPLIT]], <2 x float> poison, <1 x i32> <i32 1>47; CHECK-NEXT: [[TMP19:%.*]] = extractelement <2 x float> [[SPLIT3]], i64 048; CHECK-NEXT: [[SPLAT_SPLATINSERT20:%.*]] = insertelement <1 x float> poison, float [[TMP19]], i64 049; CHECK-NEXT: [[SPLAT_SPLAT21:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT20]], <1 x float> poison, <1 x i32> zeroinitializer50; CHECK-NEXT: [[TMP20:%.*]] = fmul fast <1 x float> [[BLOCK19]], [[SPLAT_SPLAT21]]51; CHECK-NEXT: [[BLOCK22:%.*]] = shufflevector <2 x float> [[SPLIT1]], <2 x float> poison, <1 x i32> <i32 1>52; CHECK-NEXT: [[TMP21:%.*]] = extractelement <2 x float> [[SPLIT3]], i64 153; CHECK-NEXT: [[SPLAT_SPLATINSERT23:%.*]] = insertelement <1 x float> poison, float [[TMP21]], i64 054; CHECK-NEXT: [[SPLAT_SPLAT24:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT23]], <1 x float> poison, <1 x i32> zeroinitializer55; CHECK-NEXT: [[TMP22:%.*]] = call fast <1 x float> @llvm.fmuladd.v1f32(<1 x float> [[BLOCK22]], <1 x float> [[SPLAT_SPLAT24]], <1 x float> [[TMP20]])56; CHECK-NEXT: [[TMP23:%.*]] = shufflevector <1 x float> [[TMP22]], <1 x float> poison, <2 x i32> <i32 0, i32 poison>57; CHECK-NEXT: [[TMP24:%.*]] = shufflevector <2 x float> [[TMP18]], <2 x float> [[TMP23]], <2 x i32> <i32 0, i32 2>58; CHECK-NEXT: [[SPLIT25:%.*]] = shufflevector <4 x float> [[N]], <4 x float> poison, <2 x i32> <i32 0, i32 1>59; CHECK-NEXT: [[SPLIT26:%.*]] = shufflevector <4 x float> [[N]], <4 x float> poison, <2 x i32> <i32 2, i32 3>60; CHECK-NEXT: [[TMP25:%.*]] = fadd fast <2 x float> [[TMP12]], [[SPLIT25]]61; CHECK-NEXT: [[TMP26:%.*]] = fadd fast <2 x float> [[TMP24]], [[SPLIT26]]62; CHECK-NEXT: [[TMP27:%.*]] = shufflevector <2 x float> [[TMP25]], <2 x float> [[TMP26]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>63; CHECK-NEXT: ret <4 x float> [[TMP27]]64;65 %res.0 = tail call fast <4 x float> @llvm.matrix.multiply.v4f32.v4f32.v4f32(<4 x float> %m, <4 x float> %n, i32 2, i32 2, i32 2)66 %res.1 = fadd fast <4 x float> %res.0, %n67 ret <4 x float> %res.168}69 70define <4 x float> @preserve_fmf_reassoc(<4 x float> %m, <4 x float> %n) {71; CHECK-LABEL: @preserve_fmf_reassoc(72; CHECK-NEXT: [[SPLIT:%.*]] = shufflevector <4 x float> [[M:%.*]], <4 x float> poison, <2 x i32> <i32 0, i32 1>73; CHECK-NEXT: [[SPLIT1:%.*]] = shufflevector <4 x float> [[M]], <4 x float> poison, <2 x i32> <i32 2, i32 3>74; CHECK-NEXT: [[SPLIT2:%.*]] = shufflevector <4 x float> [[N:%.*]], <4 x float> poison, <2 x i32> <i32 0, i32 1>75; CHECK-NEXT: [[SPLIT3:%.*]] = shufflevector <4 x float> [[N]], <4 x float> poison, <2 x i32> <i32 2, i32 3>76; CHECK-NEXT: [[BLOCK:%.*]] = shufflevector <2 x float> [[SPLIT]], <2 x float> poison, <1 x i32> zeroinitializer77; CHECK-NEXT: [[TMP1:%.*]] = extractelement <2 x float> [[SPLIT2]], i64 078; CHECK-NEXT: [[SPLAT_SPLATINSERT:%.*]] = insertelement <1 x float> poison, float [[TMP1]], i64 079; CHECK-NEXT: [[SPLAT_SPLAT:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT]], <1 x float> poison, <1 x i32> zeroinitializer80; CHECK-NEXT: [[TMP2:%.*]] = fmul reassoc <1 x float> [[BLOCK]], [[SPLAT_SPLAT]]81; CHECK-NEXT: [[BLOCK4:%.*]] = shufflevector <2 x float> [[SPLIT1]], <2 x float> poison, <1 x i32> zeroinitializer82; CHECK-NEXT: [[TMP3:%.*]] = extractelement <2 x float> [[SPLIT2]], i64 183; CHECK-NEXT: [[SPLAT_SPLATINSERT5:%.*]] = insertelement <1 x float> poison, float [[TMP3]], i64 084; CHECK-NEXT: [[SPLAT_SPLAT6:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT5]], <1 x float> poison, <1 x i32> zeroinitializer85; CHECK-NEXT: [[TMP4:%.*]] = fmul reassoc <1 x float> [[BLOCK4]], [[SPLAT_SPLAT6]]86; CHECK-NEXT: [[TMP5:%.*]] = fadd reassoc <1 x float> [[TMP2]], [[TMP4]]87; CHECK-NEXT: [[TMP6:%.*]] = shufflevector <1 x float> [[TMP5]], <1 x float> poison, <2 x i32> <i32 0, i32 poison>88; CHECK-NEXT: [[TMP7:%.*]] = shufflevector <2 x float> poison, <2 x float> [[TMP6]], <2 x i32> <i32 2, i32 1>89; CHECK-NEXT: [[BLOCK7:%.*]] = shufflevector <2 x float> [[SPLIT]], <2 x float> poison, <1 x i32> <i32 1>90; CHECK-NEXT: [[TMP8:%.*]] = extractelement <2 x float> [[SPLIT2]], i64 091; CHECK-NEXT: [[SPLAT_SPLATINSERT8:%.*]] = insertelement <1 x float> poison, float [[TMP8]], i64 092; CHECK-NEXT: [[SPLAT_SPLAT9:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT8]], <1 x float> poison, <1 x i32> zeroinitializer93; CHECK-NEXT: [[TMP9:%.*]] = fmul reassoc <1 x float> [[BLOCK7]], [[SPLAT_SPLAT9]]94; CHECK-NEXT: [[BLOCK10:%.*]] = shufflevector <2 x float> [[SPLIT1]], <2 x float> poison, <1 x i32> <i32 1>95; CHECK-NEXT: [[TMP10:%.*]] = extractelement <2 x float> [[SPLIT2]], i64 196; CHECK-NEXT: [[SPLAT_SPLATINSERT11:%.*]] = insertelement <1 x float> poison, float [[TMP10]], i64 097; CHECK-NEXT: [[SPLAT_SPLAT12:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT11]], <1 x float> poison, <1 x i32> zeroinitializer98; CHECK-NEXT: [[TMP11:%.*]] = fmul reassoc <1 x float> [[BLOCK10]], [[SPLAT_SPLAT12]]99; CHECK-NEXT: [[TMP12:%.*]] = fadd reassoc <1 x float> [[TMP9]], [[TMP11]]100; CHECK-NEXT: [[TMP13:%.*]] = shufflevector <1 x float> [[TMP12]], <1 x float> poison, <2 x i32> <i32 0, i32 poison>101; CHECK-NEXT: [[TMP14:%.*]] = shufflevector <2 x float> [[TMP7]], <2 x float> [[TMP13]], <2 x i32> <i32 0, i32 2>102; CHECK-NEXT: [[BLOCK13:%.*]] = shufflevector <2 x float> [[SPLIT]], <2 x float> poison, <1 x i32> zeroinitializer103; CHECK-NEXT: [[TMP15:%.*]] = extractelement <2 x float> [[SPLIT3]], i64 0104; CHECK-NEXT: [[SPLAT_SPLATINSERT14:%.*]] = insertelement <1 x float> poison, float [[TMP15]], i64 0105; CHECK-NEXT: [[SPLAT_SPLAT15:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT14]], <1 x float> poison, <1 x i32> zeroinitializer106; CHECK-NEXT: [[TMP16:%.*]] = fmul reassoc <1 x float> [[BLOCK13]], [[SPLAT_SPLAT15]]107; CHECK-NEXT: [[BLOCK16:%.*]] = shufflevector <2 x float> [[SPLIT1]], <2 x float> poison, <1 x i32> zeroinitializer108; CHECK-NEXT: [[TMP17:%.*]] = extractelement <2 x float> [[SPLIT3]], i64 1109; CHECK-NEXT: [[SPLAT_SPLATINSERT17:%.*]] = insertelement <1 x float> poison, float [[TMP17]], i64 0110; CHECK-NEXT: [[SPLAT_SPLAT18:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT17]], <1 x float> poison, <1 x i32> zeroinitializer111; CHECK-NEXT: [[TMP18:%.*]] = fmul reassoc <1 x float> [[BLOCK16]], [[SPLAT_SPLAT18]]112; CHECK-NEXT: [[TMP19:%.*]] = fadd reassoc <1 x float> [[TMP16]], [[TMP18]]113; CHECK-NEXT: [[TMP20:%.*]] = shufflevector <1 x float> [[TMP19]], <1 x float> poison, <2 x i32> <i32 0, i32 poison>114; CHECK-NEXT: [[TMP21:%.*]] = shufflevector <2 x float> poison, <2 x float> [[TMP20]], <2 x i32> <i32 2, i32 1>115; CHECK-NEXT: [[BLOCK19:%.*]] = shufflevector <2 x float> [[SPLIT]], <2 x float> poison, <1 x i32> <i32 1>116; CHECK-NEXT: [[TMP22:%.*]] = extractelement <2 x float> [[SPLIT3]], i64 0117; CHECK-NEXT: [[SPLAT_SPLATINSERT20:%.*]] = insertelement <1 x float> poison, float [[TMP22]], i64 0118; CHECK-NEXT: [[SPLAT_SPLAT21:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT20]], <1 x float> poison, <1 x i32> zeroinitializer119; CHECK-NEXT: [[TMP23:%.*]] = fmul reassoc <1 x float> [[BLOCK19]], [[SPLAT_SPLAT21]]120; CHECK-NEXT: [[BLOCK22:%.*]] = shufflevector <2 x float> [[SPLIT1]], <2 x float> poison, <1 x i32> <i32 1>121; CHECK-NEXT: [[TMP24:%.*]] = extractelement <2 x float> [[SPLIT3]], i64 1122; CHECK-NEXT: [[SPLAT_SPLATINSERT23:%.*]] = insertelement <1 x float> poison, float [[TMP24]], i64 0123; CHECK-NEXT: [[SPLAT_SPLAT24:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT23]], <1 x float> poison, <1 x i32> zeroinitializer124; CHECK-NEXT: [[TMP25:%.*]] = fmul reassoc <1 x float> [[BLOCK22]], [[SPLAT_SPLAT24]]125; CHECK-NEXT: [[TMP26:%.*]] = fadd reassoc <1 x float> [[TMP23]], [[TMP25]]126; CHECK-NEXT: [[TMP27:%.*]] = shufflevector <1 x float> [[TMP26]], <1 x float> poison, <2 x i32> <i32 0, i32 poison>127; CHECK-NEXT: [[TMP28:%.*]] = shufflevector <2 x float> [[TMP21]], <2 x float> [[TMP27]], <2 x i32> <i32 0, i32 2>128; CHECK-NEXT: [[SPLIT25:%.*]] = shufflevector <4 x float> [[N]], <4 x float> poison, <2 x i32> <i32 0, i32 1>129; CHECK-NEXT: [[SPLIT26:%.*]] = shufflevector <4 x float> [[N]], <4 x float> poison, <2 x i32> <i32 2, i32 3>130; CHECK-NEXT: [[TMP29:%.*]] = fsub reassoc <2 x float> [[TMP14]], [[SPLIT25]]131; CHECK-NEXT: [[TMP30:%.*]] = fsub reassoc <2 x float> [[TMP28]], [[SPLIT26]]132; CHECK-NEXT: [[TMP31:%.*]] = shufflevector <2 x float> [[TMP29]], <2 x float> [[TMP30]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>133; CHECK-NEXT: ret <4 x float> [[TMP31]]134;135 %res.0 = tail call reassoc <4 x float> @llvm.matrix.multiply.v4f32.v4f32.v4f32(<4 x float> %m, <4 x float> %n, i32 2, i32 2, i32 2)136 %res.1 = fsub reassoc <4 x float> %res.0, %n137 ret <4 x float> %res.1138}139 140define <4 x float> @preserve_fmf_contract_reassoc(<4 x float> %m, <4 x float> %n) {141; CHECK-LABEL: @preserve_fmf_contract_reassoc(142; CHECK-NEXT: [[SPLIT:%.*]] = shufflevector <4 x float> [[M:%.*]], <4 x float> poison, <2 x i32> <i32 0, i32 1>143; CHECK-NEXT: [[SPLIT1:%.*]] = shufflevector <4 x float> [[M]], <4 x float> poison, <2 x i32> <i32 2, i32 3>144; CHECK-NEXT: [[SPLIT2:%.*]] = shufflevector <4 x float> [[N:%.*]], <4 x float> poison, <2 x i32> <i32 0, i32 1>145; CHECK-NEXT: [[SPLIT3:%.*]] = shufflevector <4 x float> [[N]], <4 x float> poison, <2 x i32> <i32 2, i32 3>146; CHECK-NEXT: [[BLOCK:%.*]] = shufflevector <2 x float> [[SPLIT]], <2 x float> poison, <1 x i32> zeroinitializer147; CHECK-NEXT: [[TMP1:%.*]] = extractelement <2 x float> [[SPLIT2]], i64 0148; CHECK-NEXT: [[SPLAT_SPLATINSERT:%.*]] = insertelement <1 x float> poison, float [[TMP1]], i64 0149; CHECK-NEXT: [[SPLAT_SPLAT:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT]], <1 x float> poison, <1 x i32> zeroinitializer150; CHECK-NEXT: [[TMP2:%.*]] = fmul reassoc contract <1 x float> [[BLOCK]], [[SPLAT_SPLAT]]151; CHECK-NEXT: [[BLOCK4:%.*]] = shufflevector <2 x float> [[SPLIT1]], <2 x float> poison, <1 x i32> zeroinitializer152; CHECK-NEXT: [[TMP3:%.*]] = extractelement <2 x float> [[SPLIT2]], i64 1153; CHECK-NEXT: [[SPLAT_SPLATINSERT5:%.*]] = insertelement <1 x float> poison, float [[TMP3]], i64 0154; CHECK-NEXT: [[SPLAT_SPLAT6:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT5]], <1 x float> poison, <1 x i32> zeroinitializer155; CHECK-NEXT: [[TMP4:%.*]] = call reassoc contract <1 x float> @llvm.fmuladd.v1f32(<1 x float> [[BLOCK4]], <1 x float> [[SPLAT_SPLAT6]], <1 x float> [[TMP2]])156; CHECK-NEXT: [[TMP5:%.*]] = shufflevector <1 x float> [[TMP4]], <1 x float> poison, <2 x i32> <i32 0, i32 poison>157; CHECK-NEXT: [[TMP6:%.*]] = shufflevector <2 x float> poison, <2 x float> [[TMP5]], <2 x i32> <i32 2, i32 1>158; CHECK-NEXT: [[BLOCK7:%.*]] = shufflevector <2 x float> [[SPLIT]], <2 x float> poison, <1 x i32> <i32 1>159; CHECK-NEXT: [[TMP7:%.*]] = extractelement <2 x float> [[SPLIT2]], i64 0160; CHECK-NEXT: [[SPLAT_SPLATINSERT8:%.*]] = insertelement <1 x float> poison, float [[TMP7]], i64 0161; CHECK-NEXT: [[SPLAT_SPLAT9:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT8]], <1 x float> poison, <1 x i32> zeroinitializer162; CHECK-NEXT: [[TMP8:%.*]] = fmul reassoc contract <1 x float> [[BLOCK7]], [[SPLAT_SPLAT9]]163; CHECK-NEXT: [[BLOCK10:%.*]] = shufflevector <2 x float> [[SPLIT1]], <2 x float> poison, <1 x i32> <i32 1>164; CHECK-NEXT: [[TMP9:%.*]] = extractelement <2 x float> [[SPLIT2]], i64 1165; CHECK-NEXT: [[SPLAT_SPLATINSERT11:%.*]] = insertelement <1 x float> poison, float [[TMP9]], i64 0166; CHECK-NEXT: [[SPLAT_SPLAT12:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT11]], <1 x float> poison, <1 x i32> zeroinitializer167; CHECK-NEXT: [[TMP10:%.*]] = call reassoc contract <1 x float> @llvm.fmuladd.v1f32(<1 x float> [[BLOCK10]], <1 x float> [[SPLAT_SPLAT12]], <1 x float> [[TMP8]])168; CHECK-NEXT: [[TMP11:%.*]] = shufflevector <1 x float> [[TMP10]], <1 x float> poison, <2 x i32> <i32 0, i32 poison>169; CHECK-NEXT: [[TMP12:%.*]] = shufflevector <2 x float> [[TMP6]], <2 x float> [[TMP11]], <2 x i32> <i32 0, i32 2>170; CHECK-NEXT: [[BLOCK13:%.*]] = shufflevector <2 x float> [[SPLIT]], <2 x float> poison, <1 x i32> zeroinitializer171; CHECK-NEXT: [[TMP13:%.*]] = extractelement <2 x float> [[SPLIT3]], i64 0172; CHECK-NEXT: [[SPLAT_SPLATINSERT14:%.*]] = insertelement <1 x float> poison, float [[TMP13]], i64 0173; CHECK-NEXT: [[SPLAT_SPLAT15:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT14]], <1 x float> poison, <1 x i32> zeroinitializer174; CHECK-NEXT: [[TMP14:%.*]] = fmul reassoc contract <1 x float> [[BLOCK13]], [[SPLAT_SPLAT15]]175; CHECK-NEXT: [[BLOCK16:%.*]] = shufflevector <2 x float> [[SPLIT1]], <2 x float> poison, <1 x i32> zeroinitializer176; CHECK-NEXT: [[TMP15:%.*]] = extractelement <2 x float> [[SPLIT3]], i64 1177; CHECK-NEXT: [[SPLAT_SPLATINSERT17:%.*]] = insertelement <1 x float> poison, float [[TMP15]], i64 0178; CHECK-NEXT: [[SPLAT_SPLAT18:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT17]], <1 x float> poison, <1 x i32> zeroinitializer179; CHECK-NEXT: [[TMP16:%.*]] = call reassoc contract <1 x float> @llvm.fmuladd.v1f32(<1 x float> [[BLOCK16]], <1 x float> [[SPLAT_SPLAT18]], <1 x float> [[TMP14]])180; CHECK-NEXT: [[TMP17:%.*]] = shufflevector <1 x float> [[TMP16]], <1 x float> poison, <2 x i32> <i32 0, i32 poison>181; CHECK-NEXT: [[TMP18:%.*]] = shufflevector <2 x float> poison, <2 x float> [[TMP17]], <2 x i32> <i32 2, i32 1>182; CHECK-NEXT: [[BLOCK19:%.*]] = shufflevector <2 x float> [[SPLIT]], <2 x float> poison, <1 x i32> <i32 1>183; CHECK-NEXT: [[TMP19:%.*]] = extractelement <2 x float> [[SPLIT3]], i64 0184; CHECK-NEXT: [[SPLAT_SPLATINSERT20:%.*]] = insertelement <1 x float> poison, float [[TMP19]], i64 0185; CHECK-NEXT: [[SPLAT_SPLAT21:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT20]], <1 x float> poison, <1 x i32> zeroinitializer186; CHECK-NEXT: [[TMP20:%.*]] = fmul reassoc contract <1 x float> [[BLOCK19]], [[SPLAT_SPLAT21]]187; CHECK-NEXT: [[BLOCK22:%.*]] = shufflevector <2 x float> [[SPLIT1]], <2 x float> poison, <1 x i32> <i32 1>188; CHECK-NEXT: [[TMP21:%.*]] = extractelement <2 x float> [[SPLIT3]], i64 1189; CHECK-NEXT: [[SPLAT_SPLATINSERT23:%.*]] = insertelement <1 x float> poison, float [[TMP21]], i64 0190; CHECK-NEXT: [[SPLAT_SPLAT24:%.*]] = shufflevector <1 x float> [[SPLAT_SPLATINSERT23]], <1 x float> poison, <1 x i32> zeroinitializer191; CHECK-NEXT: [[TMP22:%.*]] = call reassoc contract <1 x float> @llvm.fmuladd.v1f32(<1 x float> [[BLOCK22]], <1 x float> [[SPLAT_SPLAT24]], <1 x float> [[TMP20]])192; CHECK-NEXT: [[TMP23:%.*]] = shufflevector <1 x float> [[TMP22]], <1 x float> poison, <2 x i32> <i32 0, i32 poison>193; CHECK-NEXT: [[TMP24:%.*]] = shufflevector <2 x float> [[TMP18]], <2 x float> [[TMP23]], <2 x i32> <i32 0, i32 2>194; CHECK-NEXT: [[SPLIT25:%.*]] = shufflevector <4 x float> [[N]], <4 x float> poison, <2 x i32> <i32 0, i32 1>195; CHECK-NEXT: [[SPLIT26:%.*]] = shufflevector <4 x float> [[N]], <4 x float> poison, <2 x i32> <i32 2, i32 3>196; CHECK-NEXT: [[TMP25:%.*]] = fmul reassoc contract <2 x float> [[TMP12]], [[SPLIT25]]197; CHECK-NEXT: [[TMP26:%.*]] = fmul reassoc contract <2 x float> [[TMP24]], [[SPLIT26]]198; CHECK-NEXT: [[TMP27:%.*]] = fneg contract <2 x float> [[TMP25]]199; CHECK-NEXT: [[TMP28:%.*]] = fneg contract <2 x float> [[TMP26]]200; CHECK-NEXT: [[TMP29:%.*]] = shufflevector <2 x float> [[TMP27]], <2 x float> [[TMP28]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>201; CHECK-NEXT: ret <4 x float> [[TMP29]]202;203 %res.0 = tail call contract reassoc <4 x float> @llvm.matrix.multiply.v4f32.v4f32.v4f32(<4 x float> %m, <4 x float> %n, i32 2, i32 2, i32 2)204 %res.1 = fmul reassoc contract <4 x float> %res.0, %n205 %res.2 = fneg contract <4 x float> %res.1206 ret <4 x float> %res.2207}208 209declare <4 x float> @llvm.matrix.multiply.v4f32.v4f32.v4f32(<4 x float>, <4 x float>, i32, i32, i32)210