1 ; RUN: opt %loadPolly -polly-scops -polly-invariant-load-hoisting=true -analyze < %s | FileCheck %s
2 ; RUN: opt %loadPolly -polly-codegen -polly-invariant-load-hoisting=true -S < %s | FileCheck %s --check-prefix=CODEGEN
6 ; for (int i = 0; i < 1000; i++)
7 ; A[i] = (*(int *)&U) + (int)(*(float *)&U);
10 ; CHECK: Invariant Accesses: {
11 ; CHECK-NEXT: ReadAccess := [Reduction Type: NONE] [Scalar: 0]
12 ; CHECK-NEXT: { Stmt_for_body[i0] -> MemRef_U[0] };
13 ; CHECK-NEXT: Execution Context: { : }
14 ; CHECK-NEXT: ReadAccess := [Reduction Type: NONE] [Scalar: 0]
15 ; CHECK-NEXT: { Stmt_for_body[i0] -> MemRef_U[0] };
16 ; CHECK-NEXT: Execution Context: { : }
20 ; CHECK-NEXT: Stmt_for_body
21 ; CHECK-NEXT: Domain :=
22 ; CHECK-NEXT: { Stmt_for_body[i0] : 0 <= i0 <= 999 };
23 ; CHECK-NEXT: Schedule :=
24 ; CHECK-NEXT: { Stmt_for_body[i0] -> [i0] };
25 ; CHECK-NEXT: MustWriteAccess := [Reduction Type: NONE] [Scalar: 0]
26 ; CHECK-NEXT: { Stmt_for_body[i0] -> MemRef_A[i0] };
30 ; CODEGEN-DAG: %U.f.preload.s2a = alloca float
31 ; CODEGEN-DAG: %U.i.preload.s2a = alloca i32
32 ; CODEGEN: br label %polly.split_new_and_old
34 ; CODEGEN: polly.preload.begin:
35 ; CODEGEN-DAG: %U.load[[f:[.0-9]*]] = load float, float* bitcast (i32* @U to float*)
36 ; CODEGEN-DAG: store float %U.load[[f]], float* %U.f.preload.s2a
37 ; CODEGEN-DAG: %U.load[[i:[.0-9]*]] = load i32, i32* @U
38 ; CODEGEN-DAG: store i32 %U.load[[i]], i32* %U.i.preload.s2a
40 ; CODEGEN: polly.merge_new_and_old:
41 ; CODEGEN-NOT: merge = phi
43 ; CODEGEN: polly.loop_exit:
44 ; CODEGEN-NOT: final_reload
46 ; CODEGEN: polly.stmt.for.body:
47 ; CODEGEN: %p_add = add nsw i32 %U.load[[i]], %p_conv
49 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
51 @U = common global i32 0, align 4
53 define void @f(i32* %A) {
57 for.cond: ; preds = %for.inc, %entry
58 %indvars.iv = phi i64 [ %indvars.iv.next, %for.inc ], [ 0, %entry ]
59 %exitcond = icmp ne i64 %indvars.iv, 1000
60 br i1 %exitcond, label %for.body, label %for.end
62 for.body: ; preds = %for.cond
63 %U.i = load i32, i32* @U, align 4
64 %U.cast = bitcast i32 *@U to float*
65 %U.f = load float, float* %U.cast, align 4
66 %conv = fptosi float %U.f to i32
67 %add = add nsw i32 %U.i, %conv
68 %arrayidx = getelementptr inbounds i32, i32* %A, i64 %indvars.iv
69 store i32 %add, i32* %arrayidx, align 4
72 for.inc: ; preds = %for.body
73 %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
76 for.end: ; preds = %for.cond