1; RUN: opt %loadPolly -polly-stmt-granularity=bb -polly-mse -polly-print-scops -disable-output < %s | FileCheck %s
2; RUN: opt %loadNPMPolly -polly-stmt-granularity=bb "-passes=scop(print<polly-mse>)" -disable-output < %s | FileCheck %s
3; RUN: opt %loadPolly -polly-stmt-granularity=bb -polly-mse -polly-print-scops -pass-remarks-analysis="polly-mse" -disable-output < %s 2>&1| FileCheck %s --check-prefix=MSE
4; RUN: opt %loadNPMPolly -polly-stmt-granularity=bb "-passes=scop(print<polly-mse>)" -pass-remarks-analysis="polly-mse" -disable-output < %s 2>&1 | FileCheck %s --check-prefix=MSE
5;
6; Verify that the expansion of an array with load after store in a same statement is not done.
7;
8; Original source code :
9;
10; #define Ni 2000
11; #define Nj 3000
12;
13; void mse(double A[Ni], double B[Nj], double C[Nj], double D[Nj]) {
14;   int i,j;
15;   for (i = 0; i < Ni; i++) {
16;     for (int j = 0; j<Nj; j++) {
17;       B[j] = j;
18;       C[j] = B[j];
19;     }
20;   }
21; }
22;
23; Check that C is expanded
24;
25; CHECK: i64 MemRef_C_Stmt_for_body4_expanded[10000][10000]; // Element size 8
26; CHECK: new: { Stmt_for_body4[i0, i1] -> MemRef_C_Stmt_for_body4_expanded[i0, i1] };
27;
28; Check that B is not expanded
29;
30; CHECK-NOT: double MemRef_B_Stmt_for_body4_expanded[10000][10000]; // Element size 8
31; MSE: MemRef_B has read after write to the same element in same statement. The dependences found during analysis may be wrong because Polly is not able to handle such case for now.
32;
33target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
34target triple = "x86_64-unknown-linux-gnu"
35
36define void @mse(double* %A, double* %B, double* %C, double* %D) {
37entry:
38  br label %entry.split
39
40entry.split:                                      ; preds = %entry
41  br label %for.body
42
43for.body:                                         ; preds = %entry.split, %for.inc9
44  %i.02 = phi i32 [ 0, %entry.split ], [ %inc10, %for.inc9 ]
45  br label %for.body4
46
47for.body4:                                        ; preds = %for.body, %for.body4
48  %indvars.iv = phi i64 [ 0, %for.body ], [ %indvars.iv.next, %for.body4 ]
49  %0 = trunc i64 %indvars.iv to i32
50  %conv = sitofp i32 %0 to double
51  %arrayidx = getelementptr inbounds double, double* %B, i64 %indvars.iv
52  store double %conv, double* %arrayidx, align 8
53  %arrayidx6 = getelementptr inbounds double, double* %B, i64 %indvars.iv
54  %1 = bitcast double* %arrayidx6 to i64*
55  %2 = load i64, i64* %1, align 8
56  %arrayidx8 = getelementptr inbounds double, double* %C, i64 %indvars.iv
57  %3 = bitcast double* %arrayidx8 to i64*
58  store i64 %2, i64* %3, align 8
59  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
60  %exitcond = icmp ne i64 %indvars.iv.next, 10000
61  br i1 %exitcond, label %for.body4, label %for.inc9
62
63for.inc9:                                         ; preds = %for.body4
64  %inc10 = add nuw nsw i32 %i.02, 1
65  %exitcond3 = icmp ne i32 %inc10, 10000
66  br i1 %exitcond3, label %for.body, label %for.end11
67
68for.end11:                                        ; preds = %for.inc9
69  ret void
70}
71