1 ; RUN: opt -loop-accesses -analyze < %s | FileCheck %s 2 ; RUN: opt -passes='require<scalar-evolution>,require<aa>,loop(print-access-info)' -disable-output < %s 2>&1 | FileCheck %s 3 4 ; Check that loop-indepedent forward dependences are discovered properly. 5 ; 6 ; FIXME: This does not actually always work which is pretty confusing. Right 7 ; now there is hack in LAA that tries to figure out loop-indepedent forward 8 ; dependeces *outside* of the MemoryDepChecker logic (i.e. proper dependence 9 ; analysis). 10 ; 11 ; Therefore if there is only loop-independent dependences for an array 12 ; (i.e. the same index is used), we don't discover the forward dependence. 13 ; So, at ***, we add another non-I-based access of A to trigger 14 ; MemoryDepChecker analysis for accesses of A. 15 ; 16 ; for (unsigned i = 0; i < 100; i++) { 17 ; A[i + 1] = B[i] + 1; // *** 18 ; A[i] = B[i] + 2; 19 ; C[i] = A[i] * 2; 20 ; } 21 22 target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128" 23 24 define void @f(i32* noalias %A, i32* noalias %B, i32* noalias %C, i64 %N) { 25 26 ; CHECK: Dependences: 27 ; CHECK-NEXT: Forward: 28 ; CHECK-NEXT: store i32 %b_p1, i32* %Aidx, align 4 -> 29 ; CHECK-NEXT: %a = load i32, i32* %Aidx, align 4 30 ; CHECK: ForwardButPreventsForwarding: 31 ; CHECK-NEXT: store i32 %b_p2, i32* %Aidx_next, align 4 -> 32 ; CHECK-NEXT: %a = load i32, i32* %Aidx, align 4 33 ; CHECK: Forward: 34 ; CHECK-NEXT: store i32 %b_p2, i32* %Aidx_next, align 4 -> 35 ; CHECK-NEXT: store i32 %b_p1, i32* %Aidx, align 4 36 37 entry: 38 br label %for.body 39 40 for.body: ; preds = %for.body, %entry 41 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 42 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 43 44 %Bidx = getelementptr inbounds i32, i32* %B, i64 %indvars.iv 45 %Cidx = getelementptr inbounds i32, i32* %C, i64 %indvars.iv 46 %Aidx_next = getelementptr inbounds i32, i32* %A, i64 %indvars.iv.next 47 %Aidx = getelementptr inbounds i32, i32* %A, i64 %indvars.iv 48 49 %b = load i32, i32* %Bidx, align 4 50 %b_p2 = add i32 %b, 1 51 store i32 %b_p2, i32* %Aidx_next, align 4 52 53 %b_p1 = add i32 %b, 2 54 store i32 %b_p1, i32* %Aidx, align 4 55 56 %a = load i32, i32* %Aidx, align 4 57 %c = mul i32 %a, 2 58 store i32 %c, i32* %Cidx, align 4 59 60 %exitcond = icmp eq i64 %indvars.iv.next, %N 61 br i1 %exitcond, label %for.end, label %for.body 62 63 for.end: ; preds = %for.body 64 ret void 65 } 66
