94 lines · plain
1; RUN: opt %loadNPMPolly '-passes=polly-custom<opt-isl>' -polly-pattern-matching-based-opts=true -polly-tc-opt=true -debug -disable-output < %s 2>&1 | FileCheck %s2; REQUIRES: asserts3;4; for (i = 0; i < 16; i++)5; for (j = 0; j < 16; j++)6; for (k = 0; k < 8; ++k)7; for (l = 0; l < 1024; ++l)8; for (w = 0; w < 8; ++w)9; for (q = 0; q < 8; ++q)10; for (x = 0; x < 8; ++x)11; C[i][j][k][w][q][x] += A[l][x][j][k] * B[w][q][l][i];12;13; CHECK: The tensor contraction pattern was detected14;15target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"16target triple = "x86_64-unknown-linux-gnu"17 18define internal void @kernel_tc(ptr %C, ptr %A, ptr %B) {19entry:20 br label %for.cond1.preheader21 22for.cond1.preheader: ; preds = %for.inc60, %entry23 %indvars.iv85 = phi i64 [ 0, %entry ], [ %indvars.iv.next86, %for.inc60 ]24 br label %for.cond4.preheader25 26for.cond4.preheader: ; preds = %for.inc57, %for.cond1.preheader27 %indvars.iv82 = phi i64 [ 0, %for.cond1.preheader ], [ %indvars.iv.next83, %for.inc57 ]28 br label %for.cond7.preheader29 30for.cond7.preheader: ; preds = %for.inc54, %for.cond4.preheader31 %indvars.iv79 = phi i64 [ 0, %for.cond4.preheader ], [ %indvars.iv.next80, %for.inc54 ]32 br label %for.cond10.preheader33 34for.cond10.preheader: ; preds = %for.inc51, %for.cond7.preheader35 %indvars.iv76 = phi i64 [ 0, %for.cond7.preheader ], [ %indvars.iv.next77, %for.inc51 ]36 br label %for.cond13.preheader37 38for.cond13.preheader: ; preds = %for.inc48, %for.cond10.preheader39 %indvars.iv73 = phi i64 [ 0, %for.cond10.preheader ], [ %indvars.iv.next74, %for.inc48 ]40 br label %for.cond16.preheader41 42for.cond16.preheader: ; preds = %for.inc45, %for.cond13.preheader43 %indvars.iv70 = phi i64 [ 0, %for.cond13.preheader ], [ %indvars.iv.next71, %for.inc45 ]44 br label %for.body1845 46for.body18: ; preds = %for.body18, %for.cond16.preheader47 %indvars.iv = phi i64 [ 0, %for.cond16.preheader ], [ %indvars.iv.next, %for.body18 ]48 %arrayidx24 = getelementptr inbounds [8 x [16 x [8 x double]]], ptr %A, i64 %indvars.iv76, i64 %indvars.iv, i64 %indvars.iv82, i64 %indvars.iv7949 %i = load double, ptr %arrayidx24, align 850 %arrayidx32 = getelementptr inbounds [8 x [1024 x [16 x double]]], ptr %B, i64 %indvars.iv73, i64 %indvars.iv70, i64 %indvars.iv76, i64 %indvars.iv8551 %i1 = load double, ptr %arrayidx32, align 852 %mul = fmul fast double %i1, %i53 %arrayidx44 = getelementptr inbounds [16 x [8 x [8 x [8 x [8 x double]]]]], ptr %C, i64 %indvars.iv85, i64 %indvars.iv82, i64 %indvars.iv79, i64 %indvars.iv73, i64 %indvars.iv70, i64 %indvars.iv54 %i2 = load double, ptr %arrayidx44, align 855 %add = fadd fast double %i2, %mul56 store double %add, ptr %arrayidx44, align 857 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 158 %exitcond = icmp ne i64 %indvars.iv.next, 859 br i1 %exitcond, label %for.body18, label %for.inc4560 61for.inc45: ; preds = %for.body1862 %indvars.iv.next71 = add nuw nsw i64 %indvars.iv70, 163 %exitcond72 = icmp ne i64 %indvars.iv.next71, 864 br i1 %exitcond72, label %for.cond16.preheader, label %for.inc4865 66for.inc48: ; preds = %for.inc4567 %indvars.iv.next74 = add nuw nsw i64 %indvars.iv73, 168 %exitcond75 = icmp ne i64 %indvars.iv.next74, 869 br i1 %exitcond75, label %for.cond13.preheader, label %for.inc5170 71for.inc51: ; preds = %for.inc4872 %indvars.iv.next77 = add nuw nsw i64 %indvars.iv76, 173 %exitcond78 = icmp ne i64 %indvars.iv.next77, 102474 br i1 %exitcond78, label %for.cond10.preheader, label %for.inc5475 76for.inc54: ; preds = %for.inc5177 %indvars.iv.next80 = add nuw nsw i64 %indvars.iv79, 178 %exitcond81 = icmp ne i64 %indvars.iv.next80, 879 br i1 %exitcond81, label %for.cond7.preheader, label %for.inc5780 81for.inc57: ; preds = %for.inc5482 %indvars.iv.next83 = add nuw nsw i64 %indvars.iv82, 183 %exitcond84 = icmp ne i64 %indvars.iv.next83, 1684 br i1 %exitcond84, label %for.cond4.preheader, label %for.inc6085 86for.inc60: ; preds = %for.inc5787 %indvars.iv.next86 = add nuw nsw i64 %indvars.iv85, 188 %exitcond87 = icmp ne i64 %indvars.iv.next86, 1689 br i1 %exitcond87, label %for.cond1.preheader, label %for.end6290 91for.end62: ; preds = %for.inc6092 ret void93}94