1 ; RUN: opt -loop-load-elim -S < %s | \
2 ; RUN: FileCheck %s -check-prefix=ALL -check-prefix=ONE_STRIDE_SPEC \
3 ; RUN: -check-prefix=TWO_STRIDE_SPEC
5 ; RUN: opt -loop-load-elim -S -enable-mem-access-versioning=0 < %s | \
6 ; RUN: FileCheck %s -check-prefix=ALL -check-prefix=NO_ONE_STRIDE_SPEC \
7 ; RUN: -check-prefix=NO_TWO_STRIDE_SPEC
9 ; RUN: opt -loop-load-elim -S -loop-load-elimination-scev-check-threshold=1 < %s | \
10 ; RUN: FileCheck %s -check-prefix=ALL -check-prefix=ONE_STRIDE_SPEC \
11 ; RUN: -check-prefix=NO_TWO_STRIDE_SPEC
13 ; Forwarding in the presence of symbolic strides:
15 ; for (unsigned i = 0; i < 100; i++)
16 ; A[i + 1] = A[Stride * i] + B[i];
18 target datalayout = "e-m:o-i64:64-f80:128-n8:16:32:64-S128"
21 define void @f(i32* noalias nocapture %A, i32* noalias nocapture readonly %B, i64 %N,
24 ; ONE_STRIDE_SPEC: %ident.check = icmp ne i64 %stride, 1
27 ; NO_ONE_STRIDE_SPEC-NOT: %load_initial = load i32, i32* %A
28 ; ONE_STRIDE_SPEC: %load_initial = load i32, i32* %A
31 for.body: ; preds = %for.body, %entry
32 ; NO_ONE_STRIDE_SPEC-NOT: %store_forwarded = phi i32 [ %load_initial, {{.*}} ], [ %add, %for.body ]
33 ; ONE_STRIDE_SPEC: %store_forwarded = phi i32 [ %load_initial, {{.*}} ], [ %add, %for.body ]
34 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
35 %mul = mul i64 %indvars.iv, %stride
36 %arrayidx = getelementptr inbounds i32, i32* %A, i64 %mul
37 %load = load i32, i32* %arrayidx, align 4
38 %arrayidx2 = getelementptr inbounds i32, i32* %B, i64 %indvars.iv
39 %load_1 = load i32, i32* %arrayidx2, align 4
40 ; NO_ONE_STRIDE_SPEC-NOT: %add = add i32 %load_1, %store_forwarded
41 ; ONE_STRIDE_SPEC: %add = add i32 %load_1, %store_forwarded
42 %add = add i32 %load_1, %load
43 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
44 %arrayidx_next = getelementptr inbounds i32, i32* %A, i64 %indvars.iv.next
45 store i32 %add, i32* %arrayidx_next, align 4
46 %exitcond = icmp eq i64 %indvars.iv.next, %N
47 br i1 %exitcond, label %for.end, label %for.body
49 for.end: ; preds = %for.body
53 ; With two symbolic strides:
55 ; for (unsigned i = 0; i < 100; i++)
56 ; A[Stride2 * (i + 1)] = A[Stride1 * i] + B[i];
58 ; ALL-LABEL: @two_strides(
59 define void @two_strides(i32* noalias nocapture %A, i32* noalias nocapture readonly %B, i64 %N,
60 i64 %stride.1, i64 %stride.2) {
62 ; TWO_STRIDE_SPEC: %ident.check = icmp ne i64 %stride.2, 1
63 ; TWO_STRIDE_SPEC: %ident.check1 = icmp ne i64 %stride.1, 1
64 ; NO_TWO_STRIDE_SPEC-NOT: %ident.check{{.*}} = icmp ne i64 %stride{{.*}}, 1
67 ; NO_TWO_STRIDE_SPEC-NOT: %load_initial = load i32, i32* %A
68 ; TWO_STRIDE_SPEC: %load_initial = load i32, i32* %A
71 for.body: ; preds = %for.body, %entry
72 ; NO_TWO_STRIDE_SPEC-NOT: %store_forwarded = phi i32 [ %load_initial, {{.*}} ], [ %add, %for.body ]
73 ; TWO_STRIDE_SPEC: %store_forwarded = phi i32 [ %load_initial, {{.*}} ], [ %add, %for.body ]
74 %indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
75 %mul = mul i64 %indvars.iv, %stride.1
76 %arrayidx = getelementptr inbounds i32, i32* %A, i64 %mul
77 %load = load i32, i32* %arrayidx, align 4
78 %arrayidx2 = getelementptr inbounds i32, i32* %B, i64 %indvars.iv
79 %load_1 = load i32, i32* %arrayidx2, align 4
80 ; NO_TWO_STRIDE_SPEC-NOT: %add = add i32 %load_1, %store_forwarded
81 ; TWO_STRIDE_SPEC: %add = add i32 %load_1, %store_forwarded
82 %add = add i32 %load_1, %load
83 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
84 %mul.2 = mul i64 %indvars.iv.next, %stride.2
85 %arrayidx_next = getelementptr inbounds i32, i32* %A, i64 %mul.2
86 store i32 %add, i32* %arrayidx_next, align 4
87 %exitcond = icmp eq i64 %indvars.iv.next, %N
88 br i1 %exitcond, label %for.end, label %for.body
90 for.end: ; preds = %for.body