1; RUN: opt %loadPolly -polly-opt-isl -polly-ast -polly-opt-fusion=max -analyze < %s | FileCheck %s 2; 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 and 17; checks whether they are tiled after being fused when polly-opt-fusion equals 18; "max". 19; 20; CHECK-LABEL: Printing analysis 'Polly - Generate an AST from the SCoP (isl)' for region: 'for.cond => for.end56' in function 'tf': 21; CHECK: 1st level tiling - Tiles 22; 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 - Points 26; 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([256 x [256 x i32]]* %C, [256 x [256 x i32]]* %A0, [256 x [256 x i32]]* %A1) { 36entry: 37 br label %for.cond 38 39for.cond: ; preds = %for.inc20, %entry 40 %indvars.iv13 = phi i64 [ %indvars.iv.next14, %for.inc20 ], [ 0, %entry ] 41 %exitcond15 = icmp ne i64 %indvars.iv13, 256 42 br i1 %exitcond15, label %for.body, label %for.end22 43 44for.body: ; preds = %for.cond 45 br label %for.cond1 46 47for.cond1: ; preds = %for.inc17, %for.body 48 %indvars.iv10 = phi i64 [ %indvars.iv.next11, %for.inc17 ], [ 0, %for.body ] 49 %exitcond12 = icmp ne i64 %indvars.iv10, 256 50 br i1 %exitcond12, label %for.body3, label %for.end19 51 52for.body3: ; preds = %for.cond1 53 br label %for.cond4 54 55for.cond4: ; preds = %for.inc, %for.body3 56 %indvars.iv7 = phi i64 [ %indvars.iv.next8, %for.inc ], [ 0, %for.body3 ] 57 %exitcond9 = icmp ne i64 %indvars.iv7, 256 58 br i1 %exitcond9, label %for.body6, label %for.end 59 60for.body6: ; preds = %for.cond4 61 %arrayidx10 = getelementptr inbounds [256 x [256 x i32]], [256 x [256 x i32]]* %A0, i64 %indvars.iv13, i64 %indvars.iv10, i64 %indvars.iv7 62 %tmp = load i32, i32* %arrayidx10, align 4 63 %arrayidx16 = getelementptr inbounds [256 x [256 x i32]], [256 x [256 x i32]]* %C, i64 %indvars.iv13, i64 %indvars.iv10, i64 %indvars.iv7 64 %tmp16 = load i32, i32* %arrayidx16, align 4 65 %add = add nsw i32 %tmp16, %tmp 66 store i32 %add, i32* %arrayidx16, align 4 67 br label %for.inc 68 69for.inc: ; preds = %for.body6 70 %indvars.iv.next8 = add nuw nsw i64 %indvars.iv7, 1 71 br label %for.cond4 72 73for.end: ; preds = %for.cond4 74 br label %for.inc17 75 76for.inc17: ; preds = %for.end 77 %indvars.iv.next11 = add nuw nsw i64 %indvars.iv10, 1 78 br label %for.cond1 79 80for.end19: ; preds = %for.cond1 81 br label %for.inc20 82 83for.inc20: ; preds = %for.end19 84 %indvars.iv.next14 = add nuw nsw i64 %indvars.iv13, 1 85 br label %for.cond 86 87for.end22: ; preds = %for.cond 88 br label %for.cond24 89 90for.cond24: ; preds = %for.inc54, %for.end22 91 %indvars.iv4 = phi i64 [ %indvars.iv.next5, %for.inc54 ], [ 0, %for.end22 ] 92 %exitcond6 = icmp ne i64 %indvars.iv4, 256 93 br i1 %exitcond6, label %for.body26, label %for.end56 94 95for.body26: ; preds = %for.cond24 96 br label %for.cond28 97 98for.cond28: ; preds = %for.inc51, %for.body26 99 %indvars.iv1 = phi i64 [ %indvars.iv.next2, %for.inc51 ], [ 0, %for.body26 ] 100 %exitcond3 = icmp ne i64 %indvars.iv1, 256 101 br i1 %exitcond3, label %for.body30, label %for.end53 102 103for.body30: ; preds = %for.cond28 104 br label %for.cond32 105 106for.cond32: ; preds = %for.inc48, %for.body30 107 %indvars.iv = phi i64 [ %indvars.iv.next, %for.inc48 ], [ 0, %for.body30 ] 108 %exitcond = icmp ne i64 %indvars.iv, 256 109 br i1 %exitcond, label %for.body34, label %for.end50 110 111for.body34: ; preds = %for.cond32 112 %arrayidx40 = getelementptr inbounds [256 x [256 x i32]], [256 x [256 x i32]]* %A1, i64 %indvars.iv4, i64 %indvars.iv1, i64 %indvars.iv 113 %tmp17 = load i32, i32* %arrayidx40, align 4 114 %arrayidx46 = getelementptr inbounds [256 x [256 x i32]], [256 x [256 x i32]]* %C, i64 %indvars.iv4, i64 %indvars.iv1, i64 %indvars.iv 115 %tmp18 = load i32, i32* %arrayidx46, align 4 116 %add47 = add nsw i32 %tmp18, %tmp17 117 store i32 %add47, i32* %arrayidx46, align 4 118 br label %for.inc48 119 120for.inc48: ; preds = %for.body34 121 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 122 br label %for.cond32 123 124for.end50: ; preds = %for.cond32 125 br label %for.inc51 126 127for.inc51: ; preds = %for.end50 128 %indvars.iv.next2 = add nuw nsw i64 %indvars.iv1, 1 129 br label %for.cond28 130 131for.end53: ; preds = %for.cond28 132 br label %for.inc54 133 134for.inc54: ; preds = %for.end53 135 %indvars.iv.next5 = add nuw nsw i64 %indvars.iv4, 1 136 br label %for.cond24 137 138for.end56: ; preds = %for.cond24 139 ret void 140} 141