| 1 |
; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5 |
| 2 |
; RUN: opt -passes=loop-interchange -cache-line-size=64 -S < %s | FileCheck %s |
| 3 |
|
| 4 |
; Check that nsw/nuw flags are dropped when interchanging loops. |
| 5 |
; |
| 6 |
; int sum = 0; |
| 7 |
; for (int i = 0; i < 2; i++) |
| 8 |
; for (int j = 0; j < 2; j++) |
| 9 |
; sum += A[j][i]; |
| 10 |
; |
| 11 |
define void @reduction_add(ptr %A) { |
| 12 |
; CHECK-LABEL: define void @reduction_add( |
| 13 |
; CHECK-SAME: ptr [[A:%.*]]) { |
| 14 |
; CHECK-NEXT: [[ENTRY:.*:]] |
| 15 |
; CHECK-NEXT: br label %[[FOR_J_PREHEADER:.*]] |
| 16 |
; CHECK: [[FOR_I_HEADER_PREHEADER:.*]]: |
| 17 |
; CHECK-NEXT: br label %[[FOR_I_HEADER:.*]] |
| 18 |
; CHECK: [[FOR_I_HEADER]]: |
| 19 |
; CHECK-NEXT: [[I:%.*]] = phi i32 [ [[I_INC:%.*]], %[[FOR_I_LATCH:.*]] ], [ 0, %[[FOR_I_HEADER_PREHEADER]] ] |
| 20 |
; CHECK-NEXT: [[SUM_J:%.*]] = phi i32 [ [[SUM_J_NEXT:%.*]], %[[FOR_I_LATCH]] ], [ [[SUM_I:%.*]], %[[FOR_I_HEADER_PREHEADER]] ] |
| 21 |
; CHECK-NEXT: br label %[[FOR_J_SPLIT1:.*]] |
| 22 |
; CHECK: [[FOR_J_PREHEADER]]: |
| 23 |
; CHECK-NEXT: br label %[[FOR_J:.*]] |
| 24 |
; CHECK: [[FOR_J]]: |
| 25 |
; CHECK-NEXT: [[J:%.*]] = phi i32 [ [[TMP0:%.*]], %[[FOR_J_SPLIT:.*]] ], [ 0, %[[FOR_J_PREHEADER]] ] |
| 26 |
; CHECK-NEXT: [[SUM_I]] = phi i32 [ [[SUM_I_LCSSA:%.*]], %[[FOR_J_SPLIT]] ], [ 0, %[[FOR_J_PREHEADER]] ] |
| 27 |
; CHECK-NEXT: br label %[[FOR_I_HEADER_PREHEADER]] |
| 28 |
; CHECK: [[FOR_J_SPLIT1]]: |
| 29 |
; CHECK-NEXT: [[IDX:%.*]] = getelementptr inbounds [2 x [2 x i32]], ptr [[A]], i32 0, i32 [[J]], i32 [[I]] |
| 30 |
; CHECK-NEXT: [[A:%.*]] = load i32, ptr [[IDX]], align 4 |
| 31 |
; CHECK-NEXT: [[SUM_J_NEXT]] = add i32 [[SUM_J]], [[A]] |
| 32 |
; CHECK-NEXT: [[J_INC:%.*]] = add i32 [[J]], 1 |
| 33 |
; CHECK-NEXT: [[CMP_J:%.*]] = icmp slt i32 [[J_INC]], 2 |
| 34 |
; CHECK-NEXT: br label %[[FOR_I_LATCH]] |
| 35 |
; CHECK: [[FOR_J_SPLIT]]: |
| 36 |
; CHECK-NEXT: [[SUM_I_LCSSA]] = phi i32 [ [[SUM_J_NEXT]], %[[FOR_I_LATCH]] ] |
| 37 |
; CHECK-NEXT: [[TMP0]] = add i32 [[J]], 1 |
| 38 |
; CHECK-NEXT: [[TMP1:%.*]] = icmp slt i32 [[TMP0]], 2 |
| 39 |
; CHECK-NEXT: br i1 [[TMP1]], label %[[FOR_J]], label %[[EXIT:.*]] |
| 40 |
; CHECK: [[FOR_I_LATCH]]: |
| 41 |
; CHECK-NEXT: [[I_INC]] = add i32 [[I]], 1 |
| 42 |
; CHECK-NEXT: [[CMP_I:%.*]] = icmp slt i32 [[I_INC]], 2 |
| 43 |
; CHECK-NEXT: br i1 [[CMP_I]], label %[[FOR_I_HEADER]], label %[[FOR_J_SPLIT]] |
| 44 |
; CHECK: [[EXIT]]: |
| 45 |
; CHECK-NEXT: ret void |
| 46 |
; |
| 47 |
entry: |
| 48 |
br label %for.i.header |
| 49 |
|
| 50 |
for.i.header: |
| 51 |
%i = phi i32 [ 0, %entry ], [ %i.inc, %for.i.latch ] |
| 52 |
%sum.i = phi i32 [ 0, %entry ], [ %sum.i.lcssa, %for.i.latch ] |
| 53 |
br label %for.j |
| 54 |
|
| 55 |
for.j: |
| 56 |
%j = phi i32 [ 0, %for.i.header ], [ %j.inc, %for.j ] |
| 57 |
%sum.j = phi i32 [ %sum.i, %for.i.header ], [ %sum.j.next, %for.j ] |
| 58 |
%idx = getelementptr inbounds [2 x [2 x i32]], ptr %A, i32 0, i32 %j, i32 %i |
| 59 |
%a = load i32, ptr %idx, align 4 |
| 60 |
%sum.j.next = add nuw nsw i32 %sum.j, %a |
| 61 |
%j.inc = add i32 %j, 1 |
| 62 |
%cmp.j = icmp slt i32 %j.inc, 2 |
| 63 |
br i1 %cmp.j, label %for.j, label %for.i.latch |
| 64 |
|
| 65 |
for.i.latch: |
| 66 |
%sum.i.lcssa = phi i32 [ %sum.j.next, %for.j ] |
| 67 |
%i.inc = add i32 %i, 1 |
| 68 |
%cmp.i = icmp slt i32 %i.inc, 2 |
| 69 |
br i1 %cmp.i, label %for.i.header, label %exit |
| 70 |
|
| 71 |
exit: |
| 72 |
ret void |
| 73 |
} |
| 74 |
|
| 75 |
; Check that nsw/nuw flags are dropped when interchanging loops. |
| 76 |
; |
| 77 |
; int prod = 1; |
| 78 |
; for (int i = 0; i < 2; i++) |
| 79 |
; for (int j = 0; j < 2; j++) |
| 80 |
; prod *= A[j][i]; |
| 81 |
; |
| 82 |
define void @reduction_mul(ptr %A) { |
| 83 |
; CHECK-LABEL: define void @reduction_mul( |
| 84 |
; CHECK-SAME: ptr [[A:%.*]]) { |
| 85 |
; CHECK-NEXT: [[ENTRY:.*:]] |
| 86 |
; CHECK-NEXT: br label %[[FOR_J_PREHEADER:.*]] |
| 87 |
; CHECK: [[FOR_I_HEADER_PREHEADER:.*]]: |
| 88 |
; CHECK-NEXT: br label %[[FOR_I_HEADER:.*]] |
| 89 |
; CHECK: [[FOR_I_HEADER]]: |
| 90 |
; CHECK-NEXT: [[I:%.*]] = phi i32 [ [[I_INC:%.*]], %[[FOR_I_LATCH:.*]] ], [ 0, %[[FOR_I_HEADER_PREHEADER]] ] |
| 91 |
; CHECK-NEXT: [[PROD_J:%.*]] = phi i32 [ [[PROD_J_NEXT:%.*]], %[[FOR_I_LATCH]] ], [ [[PROD_I:%.*]], %[[FOR_I_HEADER_PREHEADER]] ] |
| 92 |
; CHECK-NEXT: br label %[[FOR_J_SPLIT1:.*]] |
| 93 |
; CHECK: [[FOR_J_PREHEADER]]: |
| 94 |
; CHECK-NEXT: br label %[[FOR_J:.*]] |
| 95 |
; CHECK: [[FOR_J]]: |
| 96 |
; CHECK-NEXT: [[J:%.*]] = phi i32 [ [[TMP0:%.*]], %[[FOR_J_SPLIT:.*]] ], [ 0, %[[FOR_J_PREHEADER]] ] |
| 97 |
; CHECK-NEXT: [[PROD_I]] = phi i32 [ [[PROD_I_LCSSA:%.*]], %[[FOR_J_SPLIT]] ], [ 1, %[[FOR_J_PREHEADER]] ] |
| 98 |
; CHECK-NEXT: br label %[[FOR_I_HEADER_PREHEADER]] |
| 99 |
; CHECK: [[FOR_J_SPLIT1]]: |
| 100 |
; CHECK-NEXT: [[IDX:%.*]] = getelementptr inbounds [2 x [2 x i32]], ptr [[A]], i32 0, i32 [[J]], i32 [[I]] |
| 101 |
; CHECK-NEXT: [[A:%.*]] = load i32, ptr [[IDX]], align 4 |
| 102 |
; CHECK-NEXT: [[PROD_J_NEXT]] = mul i32 [[PROD_J]], [[A]] |
| 103 |
; CHECK-NEXT: [[J_INC:%.*]] = add i32 [[J]], 1 |
| 104 |
; CHECK-NEXT: [[CMP_J:%.*]] = icmp slt i32 [[J_INC]], 2 |
| 105 |
; CHECK-NEXT: br label %[[FOR_I_LATCH]] |
| 106 |
; CHECK: [[FOR_J_SPLIT]]: |
| 107 |
; CHECK-NEXT: [[PROD_I_LCSSA]] = phi i32 [ [[PROD_J_NEXT]], %[[FOR_I_LATCH]] ] |
| 108 |
; CHECK-NEXT: [[TMP0]] = add i32 [[J]], 1 |
| 109 |
; CHECK-NEXT: [[TMP1:%.*]] = icmp slt i32 [[TMP0]], 2 |
| 110 |
; CHECK-NEXT: br i1 [[TMP1]], label %[[FOR_J]], label %[[EXIT:.*]] |
| 111 |
; CHECK: [[FOR_I_LATCH]]: |
| 112 |
; CHECK-NEXT: [[I_INC]] = add i32 [[I]], 1 |
| 113 |
; CHECK-NEXT: [[CMP_I:%.*]] = icmp slt i32 [[I_INC]], 2 |
| 114 |
; CHECK-NEXT: br i1 [[CMP_I]], label %[[FOR_I_HEADER]], label %[[FOR_J_SPLIT]] |
| 115 |
; CHECK: [[EXIT]]: |
| 116 |
; CHECK-NEXT: ret void |
| 117 |
; |
| 118 |
entry: |
| 119 |
br label %for.i.header |
| 120 |
|
| 121 |
for.i.header: |
| 122 |
%i = phi i32 [ 0, %entry ], [ %i.inc, %for.i.latch ] |
| 123 |
%prod.i = phi i32 [ 1, %entry ], [ %prod.i.lcssa, %for.i.latch ] |
| 124 |
br label %for.j |
| 125 |
|
| 126 |
for.j: |
| 127 |
%j = phi i32 [ 0, %for.i.header ], [ %j.inc, %for.j ] |
| 128 |
%prod.j = phi i32 [ %prod.i, %for.i.header ], [ %prod.j.next, %for.j ] |
| 129 |
%idx = getelementptr inbounds [2 x [2 x i32]], ptr %A, i32 0, i32 %j, i32 %i |
| 130 |
%a = load i32, ptr %idx, align 4 |
| 131 |
%prod.j.next = mul nsw nuw i32 %prod.j, %a |
| 132 |
%j.inc = add i32 %j, 1 |
| 133 |
%cmp.j = icmp slt i32 %j.inc, 2 |
| 134 |
br i1 %cmp.j, label %for.j, label %for.i.latch |
| 135 |
|
| 136 |
for.i.latch: |
| 137 |
%prod.i.lcssa = phi i32 [ %prod.j.next, %for.j ] |
| 138 |
%i.inc = add i32 %i, 1 |
| 139 |
%cmp.i = icmp slt i32 %i.inc, 2 |
| 140 |
br i1 %cmp.i, label %for.i.header, label %exit |
| 141 |
|
| 142 |
exit: |
| 143 |
ret void |
| 144 |
} |