| Sebastian Pop | bf6e1c2 | 2018-03-06 21:55:59 +0000 | [diff] [blame^] | 1 | ; REQUIRES: asserts |
| 2 | ; RUN: opt < %s -basicaa -loop-interchange -verify-dom-info -S -debug 2>&1 | FileCheck %s |
| Florian Hahn | 4284049 | 2017-07-31 09:00:52 +0000 | [diff] [blame] | 3 | |
| 4 | target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128" |
| 5 | target triple = "x86_64-unknown-linux-gnu" |
| 6 | |
| 7 | @A = common global [100 x [100 x i32]] zeroinitializer |
| 8 | |
| 9 | declare void @foo(i64 %a) |
| 10 | declare void @bar(i64 %a) readnone |
| 11 | |
| 12 | ;;--------------------------------------Test case 01------------------------------------ |
| 13 | ;; Not safe to interchange, because the called function `foo` is not marked as |
| 14 | ;; readnone, so it could introduce dependences. |
| 15 | ;; |
| 16 | ;; for(int i=0;i<N;i++) { |
| 17 | ;; for(int j=1;j<N;j++) { |
| 18 | ;; foo(i); |
| 19 | ;; A[j][i] = A[j][i]+k; |
| 20 | ;; } |
| 21 | ;; } |
| 22 | |
| Sebastian Pop | bf6e1c2 | 2018-03-06 21:55:59 +0000 | [diff] [blame^] | 23 | ; CHECK: Not interchanging loops. Cannot prove legality. |
| 24 | |
| Florian Hahn | 4284049 | 2017-07-31 09:00:52 +0000 | [diff] [blame] | 25 | define void @interchange_01(i32 %k, i32 %N) { |
| 26 | entry: |
| 27 | %cmp21 = icmp sgt i32 %N, 0 |
| 28 | br i1 %cmp21, label %for1.ph, label %exit |
| 29 | |
| 30 | for1.ph: |
| 31 | %cmp219 = icmp sgt i32 %N, 1 |
| 32 | %0 = add i32 %N, -1 |
| 33 | br label %for1.header |
| 34 | |
| 35 | for1.header: |
| 36 | %indvars.iv23 = phi i64 [ 0, %for1.ph ], [ %indvars.iv.next24, %for1.inc10 ] |
| 37 | br i1 %cmp219, label %for2.ph, label %for1.inc10 |
| 38 | |
| 39 | for2.ph: |
| 40 | br label %for2 |
| 41 | |
| 42 | for2: |
| 43 | %indvars.iv = phi i64 [ %indvars.iv.next, %for2 ], [ 1, %for2.ph ] |
| 44 | call void @foo(i64 %indvars.iv23) |
| 45 | %arrayidx5 = getelementptr inbounds [100 x [100 x i32]], [100 x [100 x i32]]* @A, i64 0, i64 %indvars.iv, i64 %indvars.iv23 |
| 46 | %1 = load i32, i32* %arrayidx5 |
| 47 | %add = add nsw i32 %1, %k |
| 48 | store i32 %add, i32* %arrayidx5 |
| 49 | %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 |
| 50 | %lftr.wideiv = trunc i64 %indvars.iv to i32 |
| 51 | %exitcond = icmp eq i32 %lftr.wideiv, %0 |
| 52 | br i1 %exitcond, label %for2.loopexit , label %for2 |
| 53 | |
| 54 | for2.loopexit: |
| 55 | br label %for1.inc10 |
| 56 | |
| 57 | for1.inc10: |
| 58 | %indvars.iv.next24 = add nuw nsw i64 %indvars.iv23, 1 |
| 59 | %lftr.wideiv25 = trunc i64 %indvars.iv23 to i32 |
| 60 | %exitcond26 = icmp eq i32 %lftr.wideiv25, %0 |
| 61 | br i1 %exitcond26, label %for1.loopexit, label %for1.header |
| 62 | |
| 63 | for1.loopexit: |
| 64 | br label %exit |
| 65 | |
| 66 | exit: |
| 67 | ret void |
| 68 | } |
| 69 | |
| Florian Hahn | 4284049 | 2017-07-31 09:00:52 +0000 | [diff] [blame] | 70 | ;;--------------------------------------Test case 02------------------------------------ |
| 71 | ;; Safe to interchange, because the called function `bar` is marked as readnone, |
| 72 | ;; so it cannot introduce dependences. |
| 73 | ;; |
| 74 | ;; for(int i=0;i<N;i++) { |
| 75 | ;; for(int j=1;j<N;j++) { |
| 76 | ;; bar(i); |
| 77 | ;; A[j][i] = A[j][i]+k; |
| 78 | ;; } |
| 79 | ;; } |
| 80 | |
| Sebastian Pop | bf6e1c2 | 2018-03-06 21:55:59 +0000 | [diff] [blame^] | 81 | ; CHECK: Not interchanging loops. Cannot prove legality. |
| 82 | |
| Florian Hahn | 4284049 | 2017-07-31 09:00:52 +0000 | [diff] [blame] | 83 | define void @interchange_02(i32 %k, i32 %N) { |
| 84 | entry: |
| 85 | %cmp21 = icmp sgt i32 %N, 0 |
| 86 | br i1 %cmp21, label %for1.ph, label %exit |
| 87 | |
| 88 | for1.ph: |
| 89 | %cmp219 = icmp sgt i32 %N, 1 |
| 90 | %0 = add i32 %N, -1 |
| 91 | br label %for1.header |
| 92 | |
| 93 | for1.header: |
| 94 | %indvars.iv23 = phi i64 [ 0, %for1.ph ], [ %indvars.iv.next24, %for1.inc10 ] |
| 95 | br i1 %cmp219, label %for2.ph, label %for1.inc10 |
| 96 | |
| 97 | for2.ph: |
| 98 | br label %for2 |
| 99 | |
| 100 | for2: |
| 101 | %indvars.iv = phi i64 [ %indvars.iv.next, %for2 ], [ 1, %for2.ph ] |
| 102 | call void @bar(i64 %indvars.iv23) |
| 103 | %arrayidx5 = getelementptr inbounds [100 x [100 x i32]], [100 x [100 x i32]]* @A, i64 0, i64 %indvars.iv, i64 %indvars.iv23 |
| 104 | %1 = load i32, i32* %arrayidx5 |
| 105 | %add = add nsw i32 %1, %k |
| 106 | store i32 %add, i32* %arrayidx5 |
| 107 | %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 |
| 108 | %lftr.wideiv = trunc i64 %indvars.iv to i32 |
| 109 | %exitcond = icmp eq i32 %lftr.wideiv, %0 |
| 110 | br i1 %exitcond, label %for2.loopexit , label %for2 |
| 111 | |
| 112 | for2.loopexit: |
| 113 | br label %for1.inc10 |
| 114 | |
| 115 | for1.inc10: |
| 116 | %indvars.iv.next24 = add nuw nsw i64 %indvars.iv23, 1 |
| 117 | %lftr.wideiv25 = trunc i64 %indvars.iv23 to i32 |
| 118 | %exitcond26 = icmp eq i32 %lftr.wideiv25, %0 |
| 119 | br i1 %exitcond26, label %for1.loopexit, label %for1.header |
| 120 | |
| 121 | for1.loopexit: |
| 122 | br label %exit |
| 123 | |
| 124 | exit: |
| 125 | ret void |
| 126 | } |