84 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 < 32; j++)6; for (k = 0; k < 32; ++k)7; for (l = 0; l < 32; ++l)8; for (w = 0; w < 32; ++w)9; for (q = 0; q < 32; ++q)10; C[i][j][k][w] += A[i][l][j][q] * B[q][w][l][k];11;12; CHECK: The tensor contraction pattern was detected13;14target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"15target triple = "x86_64-unknown-linux-gnu"16 17define internal void @kernel_tc(i32 %ni, i32 %nj, i32 %nk, i32 %nl, i32 %nq, i32 %nw, double %alpha, double %beta, ptr %C, ptr %A, ptr %B) {18entry:19 br label %for.cond1.preheader20 21for.cond1.preheader: ; preds = %for.inc50, %entry22 %indvars.iv71 = phi i64 [ 0, %entry ], [ %indvars.iv.next72, %for.inc50 ]23 br label %for.cond4.preheader24 25for.cond4.preheader: ; preds = %for.inc47, %for.cond1.preheader26 %indvars.iv68 = phi i64 [ 0, %for.cond1.preheader ], [ %indvars.iv.next69, %for.inc47 ]27 br label %for.cond7.preheader28 29for.cond7.preheader: ; preds = %for.inc44, %for.cond4.preheader30 %indvars.iv65 = phi i64 [ 0, %for.cond4.preheader ], [ %indvars.iv.next66, %for.inc44 ]31 br label %for.cond10.preheader32 33for.cond10.preheader: ; preds = %for.inc41, %for.cond7.preheader34 %indvars.iv62 = phi i64 [ 0, %for.cond7.preheader ], [ %indvars.iv.next63, %for.inc41 ]35 br label %for.cond13.preheader36 37for.cond13.preheader: ; preds = %for.inc38, %for.cond10.preheader38 %indvars.iv59 = phi i64 [ 0, %for.cond10.preheader ], [ %indvars.iv.next60, %for.inc38 ]39 br label %for.body1540 41for.body15: ; preds = %for.body15, %for.cond13.preheader42 %indvars.iv = phi i64 [ 0, %for.cond13.preheader ], [ %indvars.iv.next, %for.body15 ]43 %arrayidx21 = getelementptr inbounds [32 x [32 x [32 x double]]], ptr %A, i64 %indvars.iv71, i64 %indvars.iv62, i64 %indvars.iv68, i64 %indvars.iv44 %i = load double, ptr %arrayidx21, align 845 %arrayidx29 = getelementptr inbounds [32 x [32 x [32 x double]]], ptr %B, i64 %indvars.iv, i64 %indvars.iv59, i64 %indvars.iv62, i64 %indvars.iv6546 %i1 = load double, ptr %arrayidx29, align 847 %mul = fmul fast double %i1, %i48 %arrayidx37 = getelementptr inbounds [32 x [32 x [32 x double]]], ptr %C, i64 %indvars.iv71, i64 %indvars.iv68, i64 %indvars.iv65, i64 %indvars.iv5949 %i2 = load double, ptr %arrayidx37, align 850 %add = fadd fast double %i2, %mul51 store double %add, ptr %arrayidx37, align 852 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 153 %exitcond = icmp ne i64 %indvars.iv.next, 3254 br i1 %exitcond, label %for.body15, label %for.inc3855 56for.inc38: ; preds = %for.body1557 %indvars.iv.next60 = add nuw nsw i64 %indvars.iv59, 158 %exitcond61 = icmp ne i64 %indvars.iv.next60, 3259 br i1 %exitcond61, label %for.cond13.preheader, label %for.inc4160 61for.inc41: ; preds = %for.inc3862 %indvars.iv.next63 = add nuw nsw i64 %indvars.iv62, 163 %exitcond64 = icmp ne i64 %indvars.iv.next63, 3264 br i1 %exitcond64, label %for.cond10.preheader, label %for.inc4465 66for.inc44: ; preds = %for.inc4167 %indvars.iv.next66 = add nuw nsw i64 %indvars.iv65, 168 %exitcond67 = icmp ne i64 %indvars.iv.next66, 3269 br i1 %exitcond67, label %for.cond7.preheader, label %for.inc4770 71for.inc47: ; preds = %for.inc4472 %indvars.iv.next69 = add nuw nsw i64 %indvars.iv68, 173 %exitcond70 = icmp ne i64 %indvars.iv.next69, 3274 br i1 %exitcond70, label %for.cond4.preheader, label %for.inc5075 76for.inc50: ; preds = %for.inc4777 %indvars.iv.next72 = add nuw nsw i64 %indvars.iv71, 178 %exitcond73 = icmp ne i64 %indvars.iv.next72, 3279 br i1 %exitcond73, label %for.cond1.preheader, label %for.end5280 81for.end52: ; preds = %for.inc5082 ret void83}84