1 ; RUN: opt < %s -analyze -basicaa -lda | FileCheck %s 2 3 @x = common global [256 x i32] zeroinitializer, align 4 4 @y = common global [256 x i32] zeroinitializer, align 4 5 6 ;; for (i = 0; i < 256; i++) 7 ;; x[i] = x[255 - i] + y[i] 8 9 define void @f1(...) nounwind { 10 entry: 11 br label %for.body 12 13 for.body: 14 %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ] 15 %i.255 = sub i64 255, %i 16 %y.ld.addr = getelementptr [256 x i32]* @y, i64 0, i64 %i 17 %x.ld.addr = getelementptr [256 x i32]* @x, i64 0, i64 %i.255 18 %x.st.addr = getelementptr [256 x i32]* @x, i64 0, i64 %i 19 %x = load i32* %x.ld.addr ; 0 20 %y = load i32* %y.ld.addr ; 1 21 %r = add i32 %y, %x 22 store i32 %r, i32* %x.st.addr ; 2 23 ; CHECK: 0,2: dep 24 ; CHECK: 1,2: ind 25 %i.next = add i64 %i, 1 26 %exitcond = icmp eq i64 %i.next, 256 27 br i1 %exitcond, label %for.end, label %for.body 28 29 for.end: 30 ret void 31 } 32 33 ;; for (i = 0; i < 100; i++) 34 ;; x[i] = x[255 - i] + y[i] 35 36 define void @f2(...) nounwind { 37 entry: 38 br label %for.body 39 40 for.body: 41 %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ] 42 %i.255 = sub i64 255, %i 43 %y.ld.addr = getelementptr [256 x i32]* @y, i64 0, i64 %i 44 %x.ld.addr = getelementptr [256 x i32]* @x, i64 0, i64 %i.255 45 %x.st.addr = getelementptr [256 x i32]* @x, i64 0, i64 %i 46 %x = load i32* %x.ld.addr ; 0 47 %y = load i32* %y.ld.addr ; 1 48 %r = add i32 %y, %x 49 store i32 %r, i32* %x.st.addr ; 2 50 ; CHECK: 0,2: dep 51 ; CHECK: 1,2: ind 52 %i.next = add i64 %i, 1 53 %exitcond = icmp eq i64 %i.next, 100 54 br i1 %exitcond, label %for.end, label %for.body 55 56 for.end: 57 ret void 58 } 59 60 ;; // the first iteration (i=0) leads to an out-of-bounds access of x. as the 61 ;; // result of this access is undefined, _any_ dependence result is safe. 62 ;; for (i = 0; i < 256; i++) 63 ;; x[i] = x[256 - i] + y[i] 64 65 define void @f3(...) nounwind { 66 entry: 67 br label %for.body 68 69 for.body: 70 %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ] 71 %i.256 = sub i64 0, %i 72 %y.ld.addr = getelementptr [256 x i32]* @y, i64 0, i64 %i 73 %x.ld.addr = getelementptr [256 x i32]* @x, i64 1, i64 %i.256 74 %x.st.addr = getelementptr [256 x i32]* @x, i64 0, i64 %i 75 %x = load i32* %x.ld.addr ; 0 76 %y = load i32* %y.ld.addr ; 1 77 %r = add i32 %y, %x 78 store i32 %r, i32* %x.st.addr ; 2 79 ; CHECK: 0,2: dep 80 ; CHECK: 1,2: 81 %i.next = add i64 %i, 1 82 %exitcond = icmp eq i64 %i.next, 256 83 br i1 %exitcond, label %for.end, label %for.body 84 85 for.end: 86 ret void 87 } 88 89 ;; // slightly contrived but valid IR for the following loop, where all 90 ;; // accesses in all iterations are within bounds. while this example's first 91 ;; // (ZIV-)subscript is (0, 1), accesses are dependent. 92 ;; for (i = 1; i < 256; i++) 93 ;; x[i] = x[256 - i] + y[i] 94 95 define void @f4(...) nounwind { 96 entry: 97 br label %for.body 98 99 for.body: 100 %i = phi i64 [ 0, %entry ], [ %i.next, %for.body ] 101 %i.1 = add i64 1, %i 102 %i.256 = sub i64 -1, %i 103 %y.ld.addr = getelementptr [256 x i32]* @y, i64 0, i64 %i.1 104 %x.ld.addr = getelementptr [256 x i32]* @x, i64 1, i64 %i.256 105 %x.st.addr = getelementptr [256 x i32]* @x, i64 0, i64 %i.1 106 %x = load i32* %x.ld.addr ; 0 107 %y = load i32* %y.ld.addr ; 1 108 %r = add i32 %y, %x 109 store i32 %r, i32* %x.st.addr ; 2 110 ; CHECK: 0,2: dep 111 ; CHECK: 1,2: ind 112 %i.next = add i64 %i, 1 113 %exitcond = icmp eq i64 %i.next, 256 114 br i1 %exitcond, label %for.end, label %for.body 115 116 for.end: 117 ret void 118 } 119
