1 ; RUN: opt -basicaa -loop-load-elim -S < %s | FileCheck %s 2 3 ; In this case the later store forward to the load: 4 ; 5 ; for (unsigned i = 0; i < 100; i++) { 6 ; B[i] = A[i] + 1; 7 ; A[i+1] = C[i] + 2; 8 ; A[i+1] = D[i] + 3; 9 ; } 10 11 target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128" 12 13 define void @f(i32* noalias nocapture %A, i32* noalias nocapture readonly %B, 14 i32* noalias nocapture %C, i32* noalias nocapture readonly %D, 15 i64 %N) { 16 entry: 17 ; CHECK: %load_initial = load i32, i32* %A 18 br label %for.body 19 20 for.body: ; preds = %for.body, %entry 21 ; CHECK: %store_forwarded = phi i32 [ %load_initial, %entry ], [ %addD, %for.body ] 22 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ] 23 %arrayidxA = getelementptr inbounds i32, i32* %A, i64 %indvars.iv 24 %loadA = load i32, i32* %arrayidxA, align 4 25 ; CHECK: %addA = add i32 %store_forwarded, 1 26 %addA = add i32 %loadA, 1 27 28 %arrayidxB = getelementptr inbounds i32, i32* %B, i64 %indvars.iv 29 store i32 %addA, i32* %arrayidxB, align 4 30 31 %arrayidxC = getelementptr inbounds i32, i32* %C, i64 %indvars.iv 32 %loadC = load i32, i32* %arrayidxC, align 4 33 %addC = add i32 %loadC, 2 34 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1 35 %arrayidxA_next = getelementptr inbounds i32, i32* %A, i64 %indvars.iv.next 36 store i32 %addC, i32* %arrayidxA_next, align 4 37 38 %arrayidxD = getelementptr inbounds i32, i32* %D, i64 %indvars.iv 39 %loadD = load i32, i32* %arrayidxD, align 4 40 %addD = add i32 %loadD, 3 41 store i32 %addD, i32* %arrayidxA_next, align 4 42 43 %exitcond = icmp eq i64 %indvars.iv.next, %N 44 br i1 %exitcond, label %for.end, label %for.body 45 46 for.end: ; preds = %for.body 47 ret void 48 } 49
