75 lines · plain
1; NOTE: Assertions have been autogenerated by utils/update_test_checks.py2 3; RUN: opt -passes=lower-matrix-intrinsics -matrix-default-layout=row-major -S < %s | FileCheck %s4 5define <4 x double> @multiply_left_transpose_2x2(<4 x double> %a, <4 x double> %b) {6; CHECK-LABEL: @multiply_left_transpose_2x2(7; CHECK-NEXT: entry:8; CHECK-NEXT: [[SPLIT:%.*]] = shufflevector <4 x double> [[A:%.*]], <4 x double> poison, <2 x i32> <i32 0, i32 1>9; CHECK-NEXT: [[SPLIT1:%.*]] = shufflevector <4 x double> [[A]], <4 x double> poison, <2 x i32> <i32 2, i32 3>10; CHECK-NEXT: [[SPLIT2:%.*]] = shufflevector <4 x double> [[B:%.*]], <4 x double> poison, <2 x i32> <i32 0, i32 1>11; CHECK-NEXT: [[SPLIT3:%.*]] = shufflevector <4 x double> [[B]], <4 x double> poison, <2 x i32> <i32 2, i32 3>12; CHECK-NEXT: [[BLOCK:%.*]] = shufflevector <2 x double> [[SPLIT2]], <2 x double> poison, <1 x i32> zeroinitializer13; CHECK-NEXT: [[TMP0:%.*]] = extractelement <2 x double> [[SPLIT]], i64 014; CHECK-NEXT: [[SPLAT_SPLATINSERT:%.*]] = insertelement <1 x double> poison, double [[TMP0]], i64 015; CHECK-NEXT: [[SPLAT_SPLAT:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT]], <1 x double> poison, <1 x i32> zeroinitializer16; CHECK-NEXT: [[TMP1:%.*]] = fmul <1 x double> [[SPLAT_SPLAT]], [[BLOCK]]17; CHECK-NEXT: [[BLOCK4:%.*]] = shufflevector <2 x double> [[SPLIT3]], <2 x double> poison, <1 x i32> zeroinitializer18; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x double> [[SPLIT1]], i64 019; CHECK-NEXT: [[SPLAT_SPLATINSERT5:%.*]] = insertelement <1 x double> poison, double [[TMP2]], i64 020; CHECK-NEXT: [[SPLAT_SPLAT6:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT5]], <1 x double> poison, <1 x i32> zeroinitializer21; CHECK-NEXT: [[TMP3:%.*]] = fmul <1 x double> [[SPLAT_SPLAT6]], [[BLOCK4]]22; CHECK-NEXT: [[TMP4:%.*]] = fadd <1 x double> [[TMP1]], [[TMP3]]23; CHECK-NEXT: [[TMP5:%.*]] = shufflevector <1 x double> [[TMP4]], <1 x double> poison, <2 x i32> <i32 0, i32 poison>24; CHECK-NEXT: [[TMP6:%.*]] = shufflevector <2 x double> poison, <2 x double> [[TMP5]], <2 x i32> <i32 2, i32 1>25; CHECK-NEXT: [[BLOCK7:%.*]] = shufflevector <2 x double> [[SPLIT2]], <2 x double> poison, <1 x i32> <i32 1>26; CHECK-NEXT: [[TMP7:%.*]] = extractelement <2 x double> [[SPLIT]], i64 027; CHECK-NEXT: [[SPLAT_SPLATINSERT8:%.*]] = insertelement <1 x double> poison, double [[TMP7]], i64 028; CHECK-NEXT: [[SPLAT_SPLAT9:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT8]], <1 x double> poison, <1 x i32> zeroinitializer29; CHECK-NEXT: [[TMP8:%.*]] = fmul <1 x double> [[SPLAT_SPLAT9]], [[BLOCK7]]30; CHECK-NEXT: [[BLOCK10:%.*]] = shufflevector <2 x double> [[SPLIT3]], <2 x double> poison, <1 x i32> <i32 1>31; CHECK-NEXT: [[TMP9:%.*]] = extractelement <2 x double> [[SPLIT1]], i64 032; CHECK-NEXT: [[SPLAT_SPLATINSERT11:%.*]] = insertelement <1 x double> poison, double [[TMP9]], i64 033; CHECK-NEXT: [[SPLAT_SPLAT12:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT11]], <1 x double> poison, <1 x i32> zeroinitializer34; CHECK-NEXT: [[TMP10:%.*]] = fmul <1 x double> [[SPLAT_SPLAT12]], [[BLOCK10]]35; CHECK-NEXT: [[TMP11:%.*]] = fadd <1 x double> [[TMP8]], [[TMP10]]36; CHECK-NEXT: [[TMP12:%.*]] = shufflevector <1 x double> [[TMP11]], <1 x double> poison, <2 x i32> <i32 0, i32 poison>37; CHECK-NEXT: [[TMP13:%.*]] = shufflevector <2 x double> [[TMP6]], <2 x double> [[TMP12]], <2 x i32> <i32 0, i32 2>38; CHECK-NEXT: [[BLOCK13:%.*]] = shufflevector <2 x double> [[SPLIT2]], <2 x double> poison, <1 x i32> zeroinitializer39; CHECK-NEXT: [[TMP14:%.*]] = extractelement <2 x double> [[SPLIT]], i64 140; CHECK-NEXT: [[SPLAT_SPLATINSERT14:%.*]] = insertelement <1 x double> poison, double [[TMP14]], i64 041; CHECK-NEXT: [[SPLAT_SPLAT15:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT14]], <1 x double> poison, <1 x i32> zeroinitializer42; CHECK-NEXT: [[TMP15:%.*]] = fmul <1 x double> [[SPLAT_SPLAT15]], [[BLOCK13]]43; CHECK-NEXT: [[BLOCK16:%.*]] = shufflevector <2 x double> [[SPLIT3]], <2 x double> poison, <1 x i32> zeroinitializer44; CHECK-NEXT: [[TMP16:%.*]] = extractelement <2 x double> [[SPLIT1]], i64 145; CHECK-NEXT: [[SPLAT_SPLATINSERT17:%.*]] = insertelement <1 x double> poison, double [[TMP16]], i64 046; CHECK-NEXT: [[SPLAT_SPLAT18:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT17]], <1 x double> poison, <1 x i32> zeroinitializer47; CHECK-NEXT: [[TMP17:%.*]] = fmul <1 x double> [[SPLAT_SPLAT18]], [[BLOCK16]]48; CHECK-NEXT: [[TMP18:%.*]] = fadd <1 x double> [[TMP15]], [[TMP17]]49; CHECK-NEXT: [[TMP19:%.*]] = shufflevector <1 x double> [[TMP18]], <1 x double> poison, <2 x i32> <i32 0, i32 poison>50; CHECK-NEXT: [[TMP20:%.*]] = shufflevector <2 x double> poison, <2 x double> [[TMP19]], <2 x i32> <i32 2, i32 1>51; CHECK-NEXT: [[BLOCK19:%.*]] = shufflevector <2 x double> [[SPLIT2]], <2 x double> poison, <1 x i32> <i32 1>52; CHECK-NEXT: [[TMP21:%.*]] = extractelement <2 x double> [[SPLIT]], i64 153; CHECK-NEXT: [[SPLAT_SPLATINSERT20:%.*]] = insertelement <1 x double> poison, double [[TMP21]], i64 054; CHECK-NEXT: [[SPLAT_SPLAT21:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT20]], <1 x double> poison, <1 x i32> zeroinitializer55; CHECK-NEXT: [[TMP22:%.*]] = fmul <1 x double> [[SPLAT_SPLAT21]], [[BLOCK19]]56; CHECK-NEXT: [[BLOCK22:%.*]] = shufflevector <2 x double> [[SPLIT3]], <2 x double> poison, <1 x i32> <i32 1>57; CHECK-NEXT: [[TMP23:%.*]] = extractelement <2 x double> [[SPLIT1]], i64 158; CHECK-NEXT: [[SPLAT_SPLATINSERT23:%.*]] = insertelement <1 x double> poison, double [[TMP23]], i64 059; CHECK-NEXT: [[SPLAT_SPLAT24:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT23]], <1 x double> poison, <1 x i32> zeroinitializer60; CHECK-NEXT: [[TMP24:%.*]] = fmul <1 x double> [[SPLAT_SPLAT24]], [[BLOCK22]]61; CHECK-NEXT: [[TMP25:%.*]] = fadd <1 x double> [[TMP22]], [[TMP24]]62; CHECK-NEXT: [[TMP26:%.*]] = shufflevector <1 x double> [[TMP25]], <1 x double> poison, <2 x i32> <i32 0, i32 poison>63; CHECK-NEXT: [[TMP27:%.*]] = shufflevector <2 x double> [[TMP20]], <2 x double> [[TMP26]], <2 x i32> <i32 0, i32 2>64; CHECK-NEXT: [[TMP28:%.*]] = shufflevector <2 x double> [[TMP13]], <2 x double> [[TMP27]], <4 x i32> <i32 0, i32 1, i32 2, i32 3>65; CHECK-NEXT: ret <4 x double> [[TMP28]]66;67entry:68 %at = call <4 x double> @llvm.matrix.transpose.v4f64(<4 x double> %a, i32 2, i32 2)69 %c = call <4 x double> @llvm.matrix.multiply.v4f64.v4f64.v4f64(<4 x double> %at, <4 x double> %b, i32 2, i32 2, i32 2)70 ret <4 x double> %c71}72 73declare <4 x double> @llvm.matrix.multiply.v4f64.v4f64.v4f64(<4 x double>, <4 x double>, i32, i32, i32)74declare <4 x double> @llvm.matrix.transpose.v4f64(<4 x double>, i32, i32)75