86 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; Make sure we correctly lower in the presence of getelementptr constant5; expressions.6 7@foo = global [5 x <4 x double>] zeroinitializer, align 168 9define void @test(i32 %r, i32 %c) {10; CHECK-LABEL: @test(11; CHECK-NEXT: entry:12; CHECK-NEXT: [[R_ADDR:%.*]] = alloca i32, align 413; CHECK-NEXT: [[C_ADDR:%.*]] = alloca i32, align 414; CHECK-NEXT: store i32 [[R:%.*]], ptr [[R_ADDR]], align 415; CHECK-NEXT: store i32 [[C:%.*]], ptr [[C_ADDR]], align 416; CHECK-NEXT: [[COL_LOAD:%.*]] = load <2 x double>, ptr @foo, align 817; CHECK-NEXT: [[COL_LOAD1:%.*]] = load <2 x double>, ptr getelementptr (double, ptr @foo, i64 2), align 818; CHECK-NEXT: [[BLOCK:%.*]] = shufflevector <2 x double> [[COL_LOAD]], <2 x double> poison, <1 x i32> zeroinitializer19; CHECK-NEXT: [[TMP0:%.*]] = extractelement <2 x double> [[COL_LOAD]], i64 020; CHECK-NEXT: [[SPLAT_SPLATINSERT:%.*]] = insertelement <1 x double> poison, double [[TMP0]], i64 021; CHECK-NEXT: [[SPLAT_SPLAT:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT]], <1 x double> poison, <1 x i32> zeroinitializer22; CHECK-NEXT: [[TMP1:%.*]] = fmul <1 x double> [[BLOCK]], [[SPLAT_SPLAT]]23; CHECK-NEXT: [[BLOCK2:%.*]] = shufflevector <2 x double> [[COL_LOAD1]], <2 x double> poison, <1 x i32> zeroinitializer24; CHECK-NEXT: [[TMP2:%.*]] = extractelement <2 x double> [[COL_LOAD]], i64 125; CHECK-NEXT: [[SPLAT_SPLATINSERT3:%.*]] = insertelement <1 x double> poison, double [[TMP2]], i64 026; CHECK-NEXT: [[SPLAT_SPLAT4:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT3]], <1 x double> poison, <1 x i32> zeroinitializer27; CHECK-NEXT: [[TMP3:%.*]] = fmul <1 x double> [[BLOCK2]], [[SPLAT_SPLAT4]]28; CHECK-NEXT: [[TMP4:%.*]] = fadd <1 x double> [[TMP1]], [[TMP3]]29; CHECK-NEXT: [[TMP5:%.*]] = shufflevector <1 x double> [[TMP4]], <1 x double> poison, <2 x i32> <i32 0, i32 poison>30; CHECK-NEXT: [[TMP6:%.*]] = shufflevector <2 x double> poison, <2 x double> [[TMP5]], <2 x i32> <i32 2, i32 1>31; CHECK-NEXT: [[BLOCK5:%.*]] = shufflevector <2 x double> [[COL_LOAD]], <2 x double> poison, <1 x i32> <i32 1>32; CHECK-NEXT: [[TMP7:%.*]] = extractelement <2 x double> [[COL_LOAD]], i64 033; CHECK-NEXT: [[SPLAT_SPLATINSERT6:%.*]] = insertelement <1 x double> poison, double [[TMP7]], i64 034; CHECK-NEXT: [[SPLAT_SPLAT7:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT6]], <1 x double> poison, <1 x i32> zeroinitializer35; CHECK-NEXT: [[TMP8:%.*]] = fmul <1 x double> [[BLOCK5]], [[SPLAT_SPLAT7]]36; CHECK-NEXT: [[BLOCK8:%.*]] = shufflevector <2 x double> [[COL_LOAD1]], <2 x double> poison, <1 x i32> <i32 1>37; CHECK-NEXT: [[TMP9:%.*]] = extractelement <2 x double> [[COL_LOAD]], i64 138; CHECK-NEXT: [[SPLAT_SPLATINSERT9:%.*]] = insertelement <1 x double> poison, double [[TMP9]], i64 039; CHECK-NEXT: [[SPLAT_SPLAT10:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT9]], <1 x double> poison, <1 x i32> zeroinitializer40; CHECK-NEXT: [[TMP10:%.*]] = fmul <1 x double> [[BLOCK8]], [[SPLAT_SPLAT10]]41; CHECK-NEXT: [[TMP11:%.*]] = fadd <1 x double> [[TMP8]], [[TMP10]]42; CHECK-NEXT: [[TMP12:%.*]] = shufflevector <1 x double> [[TMP11]], <1 x double> poison, <2 x i32> <i32 0, i32 poison>43; CHECK-NEXT: [[TMP13:%.*]] = shufflevector <2 x double> [[TMP6]], <2 x double> [[TMP12]], <2 x i32> <i32 0, i32 2>44; CHECK-NEXT: [[BLOCK11:%.*]] = shufflevector <2 x double> [[COL_LOAD]], <2 x double> poison, <1 x i32> zeroinitializer45; CHECK-NEXT: [[TMP14:%.*]] = extractelement <2 x double> [[COL_LOAD1]], i64 046; CHECK-NEXT: [[SPLAT_SPLATINSERT12:%.*]] = insertelement <1 x double> poison, double [[TMP14]], i64 047; CHECK-NEXT: [[SPLAT_SPLAT13:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT12]], <1 x double> poison, <1 x i32> zeroinitializer48; CHECK-NEXT: [[TMP15:%.*]] = fmul <1 x double> [[BLOCK11]], [[SPLAT_SPLAT13]]49; CHECK-NEXT: [[BLOCK14:%.*]] = shufflevector <2 x double> [[COL_LOAD1]], <2 x double> poison, <1 x i32> zeroinitializer50; CHECK-NEXT: [[TMP16:%.*]] = extractelement <2 x double> [[COL_LOAD1]], i64 151; CHECK-NEXT: [[SPLAT_SPLATINSERT15:%.*]] = insertelement <1 x double> poison, double [[TMP16]], i64 052; CHECK-NEXT: [[SPLAT_SPLAT16:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT15]], <1 x double> poison, <1 x i32> zeroinitializer53; CHECK-NEXT: [[TMP17:%.*]] = fmul <1 x double> [[BLOCK14]], [[SPLAT_SPLAT16]]54; CHECK-NEXT: [[TMP18:%.*]] = fadd <1 x double> [[TMP15]], [[TMP17]]55; CHECK-NEXT: [[TMP19:%.*]] = shufflevector <1 x double> [[TMP18]], <1 x double> poison, <2 x i32> <i32 0, i32 poison>56; CHECK-NEXT: [[TMP20:%.*]] = shufflevector <2 x double> poison, <2 x double> [[TMP19]], <2 x i32> <i32 2, i32 1>57; CHECK-NEXT: [[BLOCK17:%.*]] = shufflevector <2 x double> [[COL_LOAD]], <2 x double> poison, <1 x i32> <i32 1>58; CHECK-NEXT: [[TMP21:%.*]] = extractelement <2 x double> [[COL_LOAD1]], i64 059; CHECK-NEXT: [[SPLAT_SPLATINSERT18:%.*]] = insertelement <1 x double> poison, double [[TMP21]], i64 060; CHECK-NEXT: [[SPLAT_SPLAT19:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT18]], <1 x double> poison, <1 x i32> zeroinitializer61; CHECK-NEXT: [[TMP22:%.*]] = fmul <1 x double> [[BLOCK17]], [[SPLAT_SPLAT19]]62; CHECK-NEXT: [[BLOCK20:%.*]] = shufflevector <2 x double> [[COL_LOAD1]], <2 x double> poison, <1 x i32> <i32 1>63; CHECK-NEXT: [[TMP23:%.*]] = extractelement <2 x double> [[COL_LOAD1]], i64 164; CHECK-NEXT: [[SPLAT_SPLATINSERT21:%.*]] = insertelement <1 x double> poison, double [[TMP23]], i64 065; CHECK-NEXT: [[SPLAT_SPLAT22:%.*]] = shufflevector <1 x double> [[SPLAT_SPLATINSERT21]], <1 x double> poison, <1 x i32> zeroinitializer66; CHECK-NEXT: [[TMP24:%.*]] = fmul <1 x double> [[BLOCK20]], [[SPLAT_SPLAT22]]67; CHECK-NEXT: [[TMP25:%.*]] = fadd <1 x double> [[TMP22]], [[TMP24]]68; CHECK-NEXT: [[TMP26:%.*]] = shufflevector <1 x double> [[TMP25]], <1 x double> poison, <2 x i32> <i32 0, i32 poison>69; CHECK-NEXT: [[TMP27:%.*]] = shufflevector <2 x double> [[TMP20]], <2 x double> [[TMP26]], <2 x i32> <i32 0, i32 2>70; CHECK-NEXT: store <2 x double> [[COL_LOAD]], ptr getelementptr inbounds ([5 x <4 x double>], ptr @foo, i64 0, i64 2), align 871; CHECK-NEXT: store <2 x double> [[COL_LOAD1]], ptr getelementptr (double, ptr getelementptr inbounds ([5 x <4 x double>], ptr @foo, i64 0, i64 2), i64 2), align 872; CHECK-NEXT: ret void73;74entry:75 %r.addr = alloca i32, align 476 %c.addr = alloca i32, align 477 store i32 %r, ptr %r.addr, align 478 store i32 %c, ptr %c.addr, align 479 %0 = load <4 x double>, ptr @foo, align 880 %mul = call <4 x double> @llvm.matrix.multiply(<4 x double> %0, <4 x double> %0, i32 2, i32 2, i32 2)81 store <4 x double> %0, ptr getelementptr inbounds ([5 x <4 x double>], ptr @foo, i64 0, i64 2), align 882 ret void83}84 85declare <4 x double> @llvm.matrix.multiply(<4 x double>, <4 x double>, i32, i32, i32)86