141 lines · plain
1; RUN: opt %loadNPMPolly -polly-isl-arg=--no-schedule-serialize-sccs '-passes=polly-custom<opt-isl;ast>' -polly-print-ast -disable-output < %s | FileCheck %s2;3;4; void tf(int C[256][256][256], int A0[256][256][256], int A1[256][256][256]) {5; for (int i = 0; i < 256; ++i)6; for (int j = 0; j < 256; ++j)7; for (int k = 0; k < 256; ++k)8; C[i][j][k] += A0[i][j][k];9;10; for (int i = 0; i < 256; ++i)11; for (int j = 0; j < 256; ++j)12; for (int k = 0; k < 256; ++k)13; C[i][j][k] += A1[i][j][k];14; }15;16; The tile_after_fusion.ll test has two statements in separate loop nests and17; checks whether they are tiled after being fused when polly-opt-fusion equals18; "max".19;20; CHECK-LABEL: :: isl ast :: tf :: %for.cond---%for.end56 21; CHECK: 1st level tiling - Tiles22; CHECK-NEXT: for (int c0 = 0; c0 <= 7; c0 += 1)23; CHECK-NEXT: for (int c1 = 0; c1 <= 7; c1 += 1)24; CHECK-NEXT: for (int c2 = 0; c2 <= 7; c2 += 1) {25; CHECK-NEXT: // 1st level tiling - Points26; CHECK-NEXT: for (int c3 = 0; c3 <= 31; c3 += 1)27; CHECK-NEXT: for (int c4 = 0; c4 <= 31; c4 += 1)28; CHECK-NEXT: for (int c5 = 0; c5 <= 31; c5 += 1) {29; CHECK-NEXT: Stmt_for_body6(32 * c0 + c3, 32 * c1 + c4, 32 * c2 + c5);30; CHECK-NEXT: Stmt_for_body34(32 * c0 + c3, 32 * c1 + c4, 32 * c2 + c5);31 32source_filename = "tile_after_fusion.c"33target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"34 35define void @tf(ptr %C, ptr %A0, ptr %A1) {36entry:37 br label %for.cond38 39for.cond: ; preds = %for.inc20, %entry40 %indvars.iv13 = phi i64 [ %indvars.iv.next14, %for.inc20 ], [ 0, %entry ]41 %exitcond15 = icmp ne i64 %indvars.iv13, 25642 br i1 %exitcond15, label %for.body, label %for.end2243 44for.body: ; preds = %for.cond45 br label %for.cond146 47for.cond1: ; preds = %for.inc17, %for.body48 %indvars.iv10 = phi i64 [ %indvars.iv.next11, %for.inc17 ], [ 0, %for.body ]49 %exitcond12 = icmp ne i64 %indvars.iv10, 25650 br i1 %exitcond12, label %for.body3, label %for.end1951 52for.body3: ; preds = %for.cond153 br label %for.cond454 55for.cond4: ; preds = %for.inc, %for.body356 %indvars.iv7 = phi i64 [ %indvars.iv.next8, %for.inc ], [ 0, %for.body3 ]57 %exitcond9 = icmp ne i64 %indvars.iv7, 25658 br i1 %exitcond9, label %for.body6, label %for.end59 60for.body6: ; preds = %for.cond461 %arrayidx10 = getelementptr inbounds [256 x [256 x i32]], ptr %A0, i64 %indvars.iv13, i64 %indvars.iv10, i64 %indvars.iv762 %tmp = load i32, ptr %arrayidx10, align 463 %arrayidx16 = getelementptr inbounds [256 x [256 x i32]], ptr %C, i64 %indvars.iv13, i64 %indvars.iv10, i64 %indvars.iv764 %tmp16 = load i32, ptr %arrayidx16, align 465 %add = add nsw i32 %tmp16, %tmp66 store i32 %add, ptr %arrayidx16, align 467 br label %for.inc68 69for.inc: ; preds = %for.body670 %indvars.iv.next8 = add nuw nsw i64 %indvars.iv7, 171 br label %for.cond472 73for.end: ; preds = %for.cond474 br label %for.inc1775 76for.inc17: ; preds = %for.end77 %indvars.iv.next11 = add nuw nsw i64 %indvars.iv10, 178 br label %for.cond179 80for.end19: ; preds = %for.cond181 br label %for.inc2082 83for.inc20: ; preds = %for.end1984 %indvars.iv.next14 = add nuw nsw i64 %indvars.iv13, 185 br label %for.cond86 87for.end22: ; preds = %for.cond88 br label %for.cond2489 90for.cond24: ; preds = %for.inc54, %for.end2291 %indvars.iv4 = phi i64 [ %indvars.iv.next5, %for.inc54 ], [ 0, %for.end22 ]92 %exitcond6 = icmp ne i64 %indvars.iv4, 25693 br i1 %exitcond6, label %for.body26, label %for.end5694 95for.body26: ; preds = %for.cond2496 br label %for.cond2897 98for.cond28: ; preds = %for.inc51, %for.body2699 %indvars.iv1 = phi i64 [ %indvars.iv.next2, %for.inc51 ], [ 0, %for.body26 ]100 %exitcond3 = icmp ne i64 %indvars.iv1, 256101 br i1 %exitcond3, label %for.body30, label %for.end53102 103for.body30: ; preds = %for.cond28104 br label %for.cond32105 106for.cond32: ; preds = %for.inc48, %for.body30107 %indvars.iv = phi i64 [ %indvars.iv.next, %for.inc48 ], [ 0, %for.body30 ]108 %exitcond = icmp ne i64 %indvars.iv, 256109 br i1 %exitcond, label %for.body34, label %for.end50110 111for.body34: ; preds = %for.cond32112 %arrayidx40 = getelementptr inbounds [256 x [256 x i32]], ptr %A1, i64 %indvars.iv4, i64 %indvars.iv1, i64 %indvars.iv113 %tmp17 = load i32, ptr %arrayidx40, align 4114 %arrayidx46 = getelementptr inbounds [256 x [256 x i32]], ptr %C, i64 %indvars.iv4, i64 %indvars.iv1, i64 %indvars.iv115 %tmp18 = load i32, ptr %arrayidx46, align 4116 %add47 = add nsw i32 %tmp18, %tmp17117 store i32 %add47, ptr %arrayidx46, align 4118 br label %for.inc48119 120for.inc48: ; preds = %for.body34121 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1122 br label %for.cond32123 124for.end50: ; preds = %for.cond32125 br label %for.inc51126 127for.inc51: ; preds = %for.end50128 %indvars.iv.next2 = add nuw nsw i64 %indvars.iv1, 1129 br label %for.cond28130 131for.end53: ; preds = %for.cond28132 br label %for.inc54133 134for.inc54: ; preds = %for.end53135 %indvars.iv.next5 = add nuw nsw i64 %indvars.iv4, 1136 br label %for.cond24137 138for.end56: ; preds = %for.cond24139 ret void140}141