1; REQUIRES: asserts 2; RUN: opt < %s -basic-aa -loop-interchange -cache-line-size=64 -verify-dom-info -verify-loop-info \ 3; RUN: -S -debug 2>&1 | FileCheck %s 4 5target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128" 6 7@A = common global [100 x [100 x i32]] zeroinitializer 8@B = common global [100 x i32] zeroinitializer 9@C = common global [100 x [100 x i32]] zeroinitializer 10@D = common global [100 x [100 x [100 x i32]]] zeroinitializer 11 12;; Loops not tightly nested are not interchanged 13;; for(int j=0;j<N;j++) { 14;; B[j] = j+k; 15;; for(int i=0;i<N;i++) 16;; A[j][i] = A[j][i]+B[j]; 17;; } 18 19; CHECK: Not interchanging loops. Cannot prove legality. 20 21define void @interchange_05(i32 %k, i32 %N){ 22entry: 23 %cmp30 = icmp sgt i32 %N, 0 24 br i1 %cmp30, label %for.body.lr.ph, label %for.end17 25 26for.body.lr.ph: 27 %0 = add i32 %N, -1 28 %1 = zext i32 %k to i64 29 br label %for.body 30 31for.body: 32 %indvars.iv32 = phi i64 [ 0, %for.body.lr.ph ], [ %indvars.iv.next33, %for.inc15 ] 33 %2 = add nsw i64 %indvars.iv32, %1 34 %arrayidx = getelementptr inbounds [100 x i32], [100 x i32]* @B, i64 0, i64 %indvars.iv32 35 %3 = trunc i64 %2 to i32 36 store i32 %3, i32* %arrayidx 37 br label %for.body3 38 39for.body3: 40 %indvars.iv = phi i64 [ 0, %for.body ], [ %indvars.iv.next, %for.body3 ] 41 %arrayidx7 = getelementptr inbounds [100 x [100 x i32]], [100 x [100 x i32]]* @A, i64 0, i64 %indvars.iv32, i64 %indvars.iv 42 %4 = load i32, i32* %arrayidx7 43 %add10 = add nsw i32 %3, %4 44 store i32 %add10, i32* %arrayidx7 45 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 46 %lftr.wideiv = trunc i64 %indvars.iv to i32 47 %exitcond = icmp eq i32 %lftr.wideiv, %0 48 br i1 %exitcond, label %for.inc15, label %for.body3 49 50for.inc15: 51 %indvars.iv.next33 = add nuw nsw i64 %indvars.iv32, 1 52 %lftr.wideiv35 = trunc i64 %indvars.iv32 to i32 53 %exitcond36 = icmp eq i32 %lftr.wideiv35, %0 54 br i1 %exitcond36, label %for.end17, label %for.body 55 56for.end17: 57 ret void 58} 59 60declare void @foo(...) readnone 61 62;; Loops not tightly nested are not interchanged 63;; for(int j=0;j<N;j++) { 64;; foo(); 65;; for(int i=2;i<N;i++) 66;; A[j][i] = A[j][i]+k; 67;; } 68 69; CHECK: Not interchanging loops. Cannot prove legality. 70 71define void @interchange_06(i32 %k, i32 %N) { 72entry: 73 %cmp22 = icmp sgt i32 %N, 0 74 br i1 %cmp22, label %for.body.lr.ph, label %for.end12 75 76for.body.lr.ph: 77 %0 = add i32 %N, -1 78 br label %for.body 79 80for.body: 81 %indvars.iv24 = phi i64 [ 0, %for.body.lr.ph ], [ %indvars.iv.next25, %for.inc10 ] 82 tail call void (...) @foo() 83 br label %for.body3 84 85for.body3: 86 %indvars.iv = phi i64 [ %indvars.iv.next, %for.body3 ], [ 2, %for.body ] 87 %arrayidx5 = getelementptr inbounds [100 x [100 x i32]], [100 x [100 x i32]]* @A, i64 0, i64 %indvars.iv24, i64 %indvars.iv 88 %1 = load i32, i32* %arrayidx5 89 %add = add nsw i32 %1, %k 90 store i32 %add, i32* %arrayidx5 91 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 92 %lftr.wideiv = trunc i64 %indvars.iv to i32 93 %exitcond = icmp eq i32 %lftr.wideiv, %0 94 br i1 %exitcond, label %for.inc10, label %for.body3 95 96for.inc10: 97 %indvars.iv.next25 = add nuw nsw i64 %indvars.iv24, 1 98 %lftr.wideiv26 = trunc i64 %indvars.iv24 to i32 99 %exitcond27 = icmp eq i32 %lftr.wideiv26, %0 100 br i1 %exitcond27, label %for.end12, label %for.body 101 102for.end12: 103 ret void 104} 105 106;; The following Loop is not considered tightly nested and is not interchanged. 107;; The outer loop header does not branch to the inner loop preheader, or the 108;; inner loop header, or the outer loop latch. 109; CHECK: Not interchanging loops. Cannot prove legality. 110define void @interchange_07(i32 %k, i32 %N, i64 %ny) { 111entry: 112 br label %for1.header 113 114for1.header: 115 %j23 = phi i64 [ 0, %entry ], [ %j.next24, %for1.inc10 ] 116 %cmp21 = icmp slt i64 0, %ny 117 br label %singleSucc 118 119singleSucc: 120 br i1 %cmp21, label %preheader.j, label %for1.inc10 121 122preheader.j: 123 br label %for2 124 125for2: 126 %j = phi i64 [ %j.next, %for2 ], [ 0, %preheader.j ] 127 %arrayidx5 = getelementptr inbounds [100 x [100 x i32]], [100 x [100 x i32]]* @A, i64 0, i64 %j, i64 %j23 128 %lv = load i32, i32* %arrayidx5 129 %add = add nsw i32 %lv, %k 130 store i32 %add, i32* %arrayidx5 131 %j.next = add nuw nsw i64 %j, 1 132 %exitcond = icmp eq i64 %j, 99 133 br i1 %exitcond, label %for1.inc10, label %for2 134 135for1.inc10: 136 %j.next24 = add nuw nsw i64 %j23, 1 137 %exitcond26 = icmp eq i64 %j23, 99 138 br i1 %exitcond26, label %for.end12, label %for1.header 139 140for.end12: 141 ret void 142} 143