65 lines · plain
1; RUN: opt %loadNPMPolly '-passes=polly-custom<opt-isl>' -polly-pattern-matching-based-opts=true -debug-only=polly-opt-isl -disable-output -polly-tc-opt=true < %s 2>&1 | FileCheck %s2; REQUIRES: asserts3;4; for (i = 0; i < _PB_NI; i++)5; for (j = 0; j < _PB_NJ; j++)6; {7; for (k = 0; k < _PB_NK; k++)8; {9; double Mul = A[i][k] * B[k][j];10; D[i][j][k] += Mul;11; C[i][j] += Mul;12; }13; }14;15; CHECK-NOT: The matrix multiplication pattern was detected16; CHECK-NOT: The tensor contraction pattern was detected17 18target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"19target triple = "x86_64-unknown-linux-gnu"20 21define void @kernel_gemm(ptr %C, ptr %A, ptr %B, ptr %D) {22entry:23 br label %for.cond1.preheader24 25for.cond1.preheader: ; preds = %for.inc27, %entry26 %indvars.iv7 = phi i64 [ 0, %entry ], [ %indvars.iv.next8, %for.inc27 ]27 br label %for.cond4.preheader28 29for.cond4.preheader: ; preds = %for.inc24, %for.cond1.preheader30 %indvars.iv4 = phi i64 [ 0, %for.cond1.preheader ], [ %indvars.iv.next5, %for.inc24 ]31 %arrayidx22 = getelementptr inbounds [1024 x double], ptr %C, i64 %indvars.iv7, i64 %indvars.iv432 br label %for.body633 34for.body6: ; preds = %for.body6, %for.cond4.preheader35 %indvars.iv = phi i64 [ 0, %for.cond4.preheader ], [ %indvars.iv.next, %for.body6 ]36 %arrayidx8 = getelementptr inbounds [1024 x double], ptr %A, i64 %indvars.iv7, i64 %indvars.iv37 %i = load double, ptr %arrayidx8, align 838 %arrayidx12 = getelementptr inbounds [1024 x double], ptr %B, i64 %indvars.iv, i64 %indvars.iv439 %i1 = load double, ptr %arrayidx12, align 840 %mul = fmul double %i1, %i41 %arrayidx18 = getelementptr inbounds [1024 x [1024 x double]], ptr %D, i64 %indvars.iv7, i64 %indvars.iv4, i64 %indvars.iv42 %i2 = load double, ptr %arrayidx18, align 843 %add = fadd double %i2, %mul44 store double %add, ptr %arrayidx18, align 845 %i3 = load double, ptr %arrayidx22, align 846 %add23 = fadd double %i3, %mul47 store double %add23, ptr %arrayidx22, align 848 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 149 %exitcond.not = icmp eq i64 %indvars.iv.next, 102450 br i1 %exitcond.not, label %for.inc24, label %for.body651 52for.inc24: ; preds = %for.body653 %indvars.iv.next5 = add nuw nsw i64 %indvars.iv4, 154 %exitcond6.not = icmp eq i64 %indvars.iv.next5, 102455 br i1 %exitcond6.not, label %for.inc27, label %for.cond4.preheader56 57for.inc27: ; preds = %for.inc2458 %indvars.iv.next8 = add nuw nsw i64 %indvars.iv7, 159 %exitcond9.not = icmp eq i64 %indvars.iv.next8, 102460 br i1 %exitcond9.not, label %for.end29, label %for.cond1.preheader61 62for.end29: ; preds = %for.inc2763 ret void64}65