107 lines · plain
1; RUN: opt %loadNPMPolly '-passes=polly-custom<delicm;simplify-1;opt-isl>' -polly-pattern-matching-based-opts=true -polly-tc-opt=true -debug -disable-output < %s 2>&1 | FileCheck %s2; REQUIRES: asserts3;4; Check that the pattern matching detects the tensor contraction pattern5; after a full run of -polly-delicm. This test case generates the following6; schedule, which contains two band nodes. Without DeLICM two statement are7; generated.8;9; domain: "{ Stmt5[i0, i1, i2, i3, i4, i5] : 0 <= i0 <= 31 and 0 <= i1 <= 31 and10; 0 <= i2 <= 31 and 0 <= i3 <= 31 and11; 0 <= i4 <= 31 and 0 <= i5 <= 31 }"12; child:13; schedule: "[{ Stmt5[i0, i1, i2, i3, i4, i5] -> [(i0)] },14; { Stmt5[i0, i1, i2, i3, i4, i5] -> [(i1)] },15; { Stmt5[i0, i1, i2, i3, i4, i5] -> [(i2)] },16; { Stmt5[i0, i1, i2, i3, i4, i5] -> [(i4)] },17; { Stmt5[i0, i1, i2, i3, i4, i5] -> [(i3)] }]"18; permutable: 119; coincident: [ 1, 1, 1, 1, 0 ]20; child:21; schedule: "[{ Stmt5[i0, i1, i2, i3, i4, i5] -> [(i5)] }]"22; permutable: 123; child:24; leaf25;26; for (i = 0; i < 32; i++)27; for (j = 0; j < 32; j++)28; for (k = 0; k < 32; ++k)29; for (l = 0; l < 32; ++l)30; for (w = 0; w < 32; ++w)31; for (q = 0; q < 32; ++q)32; C[i][j][k][w] += A[i][l][j][q] * B[q][w][l][k];33;34; CHECK: The tensor contraction pattern was detected35;36target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"37target triple = "x86_64-unknown-linux-gnu"38 39define internal fastcc void @kernel_tc(ptr nocapture %C, ptr nocapture readonly %A, ptr nocapture readonly %B) {40entry:41 br label %for.cond1.preheader42 43for.cond1.preheader: ; preds = %for.inc50, %entry44 %indvars.iv19 = phi i64 [ 0, %entry ], [ %indvars.iv.next20, %for.inc50 ]45 br label %for.cond4.preheader46 47for.cond4.preheader: ; preds = %for.inc47, %for.cond1.preheader48 %indvars.iv16 = phi i64 [ 0, %for.cond1.preheader ], [ %indvars.iv.next17, %for.inc47 ]49 br label %for.cond7.preheader50 51for.cond7.preheader: ; preds = %for.inc44, %for.cond4.preheader52 %indvars.iv13 = phi i64 [ 0, %for.cond4.preheader ], [ %indvars.iv.next14, %for.inc44 ]53 br label %for.cond10.preheader54 55for.cond10.preheader: ; preds = %for.inc41, %for.cond7.preheader56 %indvars.iv10 = phi i64 [ 0, %for.cond7.preheader ], [ %indvars.iv.next11, %for.inc41 ]57 br label %for.cond13.preheader58 59for.cond13.preheader: ; preds = %for.inc38, %for.cond10.preheader60 %indvars.iv7 = phi i64 [ 0, %for.cond10.preheader ], [ %indvars.iv.next8, %for.inc38 ]61 %arrayidx37 = getelementptr inbounds [32 x [32 x [32 x double]]], ptr %C, i64 %indvars.iv19, i64 %indvars.iv16, i64 %indvars.iv13, i64 %indvars.iv762 %.pre = load double, ptr %arrayidx37, align 863 br label %for.body1564 65for.body15: ; preds = %for.body15, %for.cond13.preheader66 %i = phi double [ %.pre, %for.cond13.preheader ], [ %add, %for.body15 ]67 %indvars.iv = phi i64 [ 0, %for.cond13.preheader ], [ %indvars.iv.next, %for.body15 ]68 %arrayidx21 = getelementptr inbounds [32 x [32 x [32 x double]]], ptr %A, i64 %indvars.iv19, i64 %indvars.iv10, i64 %indvars.iv16, i64 %indvars.iv69 %i1 = load double, ptr %arrayidx21, align 870 %arrayidx29 = getelementptr inbounds [32 x [32 x [32 x double]]], ptr %B, i64 %indvars.iv, i64 %indvars.iv7, i64 %indvars.iv10, i64 %indvars.iv1371 %i2 = load double, ptr %arrayidx29, align 872 %mul = fmul fast double %i2, %i173 %add = fadd fast double %i, %mul74 store double %add, ptr %arrayidx37, align 875 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 176 %exitcond.not = icmp eq i64 %indvars.iv.next, 3277 br i1 %exitcond.not, label %for.inc38, label %for.body1578 79for.inc38: ; preds = %for.body1580 %indvars.iv.next8 = add nuw nsw i64 %indvars.iv7, 181 %exitcond9.not = icmp eq i64 %indvars.iv.next8, 3282 br i1 %exitcond9.not, label %for.inc41, label %for.cond13.preheader83 84for.inc41: ; preds = %for.inc3885 %indvars.iv.next11 = add nuw nsw i64 %indvars.iv10, 186 %exitcond12.not = icmp eq i64 %indvars.iv.next11, 3287 br i1 %exitcond12.not, label %for.inc44, label %for.cond10.preheader88 89for.inc44: ; preds = %for.inc4190 %indvars.iv.next14 = add nuw nsw i64 %indvars.iv13, 191 %exitcond15.not = icmp eq i64 %indvars.iv.next14, 3292 br i1 %exitcond15.not, label %for.inc47, label %for.cond7.preheader93 94for.inc47: ; preds = %for.inc4495 %indvars.iv.next17 = add nuw nsw i64 %indvars.iv16, 196 %exitcond18.not = icmp eq i64 %indvars.iv.next17, 3297 br i1 %exitcond18.not, label %for.inc50, label %for.cond4.preheader98 99for.inc50: ; preds = %for.inc47100 %indvars.iv.next20 = add nuw nsw i64 %indvars.iv19, 1101 %exitcond21.not = icmp eq i64 %indvars.iv.next20, 32102 br i1 %exitcond21.not, label %for.end52, label %for.cond1.preheader103 104for.end52: ; preds = %for.inc50105 ret void106}107