64 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 < 32; i++)5; for (j = 0; j < 1024; j++)6; for (k = 0; k < 32; ++k)7; for (l = 0; l < 1024; ++l)8; C[i][j][k] += A[i][k][l] * B[l][j];9;10; CHECK: The tensor contraction pattern was detected11;12target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"13target triple = "x86_64-unknown-linux-gnu"14 15define internal void @kernel_tc(i32 %ni, i32 %nj, i32 %nk, i32 %nl, double %alpha, double %beta, ptr %C, ptr %A, ptr %B) {16entry:17 br label %for.cond1.preheader18 19for.cond1.preheader: ; preds = %for.inc30, %entry20 %indvars.iv43 = phi i64 [ 0, %entry ], [ %indvars.iv.next44, %for.inc30 ]21 br label %for.cond4.preheader22 23for.cond4.preheader: ; preds = %for.inc27, %for.cond1.preheader24 %indvars.iv40 = phi i64 [ 0, %for.cond1.preheader ], [ %indvars.iv.next41, %for.inc27 ]25 br label %for.cond7.preheader26 27for.cond7.preheader: ; preds = %for.inc24, %for.cond4.preheader28 %indvars.iv37 = phi i64 [ 0, %for.cond4.preheader ], [ %indvars.iv.next38, %for.inc24 ]29 br label %for.body930 31for.body9: ; preds = %for.body9, %for.cond7.preheader32 %indvars.iv = phi i64 [ 0, %for.cond7.preheader ], [ %indvars.iv.next, %for.body9 ]33 %arrayidx13 = getelementptr inbounds [32 x [1024 x double]], ptr %A, i64 %indvars.iv43, i64 %indvars.iv37, i64 %indvars.iv34 %i = load double, ptr %arrayidx13, align 835 %arrayidx17 = getelementptr inbounds [1024 x double], ptr %B, i64 %indvars.iv, i64 %indvars.iv4036 %i1 = load double, ptr %arrayidx17, align 837 %mul = fmul fast double %i1, %i38 %arrayidx23 = getelementptr inbounds [1024 x [32 x double]], ptr %C, i64 %indvars.iv43, i64 %indvars.iv40, i64 %indvars.iv3739 %i2 = load double, ptr %arrayidx23, align 840 %add = fadd fast double %i2, %mul41 store double %add, ptr %arrayidx23, align 842 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 143 %exitcond = icmp ne i64 %indvars.iv.next, 102444 br i1 %exitcond, label %for.body9, label %for.inc2445 46for.inc24: ; preds = %for.body947 %indvars.iv.next38 = add nuw nsw i64 %indvars.iv37, 148 %exitcond39 = icmp ne i64 %indvars.iv.next38, 3249 br i1 %exitcond39, label %for.cond7.preheader, label %for.inc2750 51for.inc27: ; preds = %for.inc2452 %indvars.iv.next41 = add nuw nsw i64 %indvars.iv40, 153 %exitcond42 = icmp ne i64 %indvars.iv.next41, 102454 br i1 %exitcond42, label %for.cond4.preheader, label %for.inc3055 56for.inc30: ; preds = %for.inc2757 %indvars.iv.next44 = add nuw nsw i64 %indvars.iv43, 158 %exitcond45 = icmp ne i64 %indvars.iv.next44, 3259 br i1 %exitcond45, label %for.cond1.preheader, label %for.end3260 61for.end32: ; preds = %for.inc3062 ret void63}64