1 ; RUN: opt -loop-accesses -analyze < %s | FileCheck %s 2 3 ; The runtime memory check code and the access grouping 4 ; algorithm both assume that the start and end values 5 ; for an access range are ordered (start <= stop). 6 ; When generating checks for accesses with negative stride 7 ; we need to take this into account and swap the interval 8 ; ends. 9 ; 10 ; for (i = 0; i < 10000; i++) { 11 ; B[i] = A[15000 - i] * 3; 12 ; } 13 14 target datalayout = "e-m:e-i64:64-i128:128-n32:64-S128" 15 target triple = "aarch64--linux-gnueabi" 16 17 ; CHECK: function 'f': 18 ; CHECK: (Low: (20000 + %a) High: (60000 + %a)) 19 20 @B = common global i32* null, align 8 21 @A = common global i32* null, align 8 22 23 define void @f() { 24 entry: 25 %a = load i32*, i32** @A, align 8 26 %b = load i32*, i32** @B, align 8 27 br label %for.body 28 29 for.body: ; preds = %for.body, %entry 30 %idx = phi i64 [ 0, %entry ], [ %add, %for.body ] 31 %negidx = sub i64 15000, %idx 32 33 %arrayidxA0 = getelementptr inbounds i32, i32* %a, i64 %negidx 34 %loadA0 = load i32, i32* %arrayidxA0, align 2 35 36 %res = mul i32 %loadA0, 3 37 38 %add = add nuw nsw i64 %idx, 1 39 40 %arrayidxB = getelementptr inbounds i32, i32* %b, i64 %idx 41 store i32 %res, i32* %arrayidxB, align 2 42 43 %exitcond = icmp eq i64 %idx, 10000 44 br i1 %exitcond, label %for.end, label %for.body 45 46 for.end: ; preds = %for.body 47 ret void 48 } 49 50 ; CHECK: function 'g': 51 ; When the stride is not constant, we are forced to do umin/umax to get 52 ; the interval limits. 53 54 ; for (i = 0; i < 10000; i++) { 55 ; B[i] = A[15000 - step * i] * 3; 56 ; } 57 58 ; Here it is not obvious what the limits are, since 'step' could be negative. 59 60 ; CHECK: Low: (-1 + (-1 * ((-60001 + (-1 * %a)) umax (-60001 + (40000 * %step) + (-1 * %a))))) 61 ; CHECK: High: ((60000 + %a) umax (60000 + (-40000 * %step) + %a)) 62 63 define void @g(i64 %step) { 64 entry: 65 %a = load i32*, i32** @A, align 8 66 %b = load i32*, i32** @B, align 8 67 br label %for.body 68 69 for.body: ; preds = %for.body, %entry 70 %idx = phi i64 [ 0, %entry ], [ %add, %for.body ] 71 %idx_mul = mul i64 %idx, %step 72 %negidx = sub i64 15000, %idx_mul 73 74 %arrayidxA0 = getelementptr inbounds i32, i32* %a, i64 %negidx 75 %loadA0 = load i32, i32* %arrayidxA0, align 2 76 77 %res = mul i32 %loadA0, 3 78 79 %add = add nuw nsw i64 %idx, 1 80 81 %arrayidxB = getelementptr inbounds i32, i32* %b, i64 %idx 82 store i32 %res, i32* %arrayidxB, align 2 83 84 %exitcond = icmp eq i64 %idx, 10000 85 br i1 %exitcond, label %for.end, label %for.body 86 87 for.end: ; preds = %for.body 88 ret void 89 } 90
