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;
4; Verify that the accesses are correctly expanded
5;
6; Original source code :
7;
8; #define Ni 2000
9; #define Nj 3000
10;
11; void mse(double A[Ni], double B[Nj], double C[Nj], double D[Nj]) {
12;   int i,j;
13;   for (j = 0; j < Nj; j++) {
14;     for (int i = 0; i<Ni; i++) {
15;       B[i] = i;
16;       D[i] = i;
17;     }
18;     A[j] = B[j];
19;     C[j] = D[j];
20;   }
21; }
22;
23; Check that expanded SAI are created
24; CHECK: double MemRef_B_Stmt_for_body4_expanded[10000][10000]; // Element size 8
25; CHECK: double MemRef_D_Stmt_for_body4_expanded[10000][10000]; // Element size 8
26; CHECK: i64 MemRef_A_Stmt_for_end_expanded[10000]; // Element size 8
27; CHECK: i64 MemRef_C_Stmt_for_end_expanded[10000]; // Element size 8
28;
29; Check that the memory access are modified
30;
31; CHECK: new: { Stmt_for_body4[i0, i1] -> MemRef_B_Stmt_for_body4_expanded[i0, i1] };
32; CHECK: new: { Stmt_for_body4[i0, i1] -> MemRef_D_Stmt_for_body4_expanded[i0, i1] };
33; CHECK: new: { Stmt_for_end[i0] -> MemRef_B_Stmt_for_body4_expanded[i0, i0] };
34; CHECK: new: { Stmt_for_end[i0] -> MemRef_A_Stmt_for_end_expanded[i0] };
35; CHECK: new: { Stmt_for_end[i0] -> MemRef_D_Stmt_for_body4_expanded[i0, i0] };
36; CHECK: new: { Stmt_for_end[i0] -> MemRef_C_Stmt_for_end_expanded[i0] };
37;
38target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
39target triple = "x86_64-unknown-linux-gnu"
40
41define void @mse(double* %A, double* %B, double* %C, double* %D) {
42entry:
43  br label %entry.split
44
45entry.split:                                      ; preds = %entry
46  br label %for.body
47
48for.body:                                         ; preds = %entry.split, %for.end
49  %indvars.iv3 = phi i64 [ 0, %entry.split ], [ %indvars.iv.next4, %for.end ]
50  br label %for.body4
51
52for.body4:                                        ; preds = %for.body, %for.body4
53  %indvars.iv = phi i64 [ 0, %for.body ], [ %indvars.iv.next, %for.body4 ]
54  %0 = trunc i64 %indvars.iv to i32
55  %conv = sitofp i32 %0 to double
56  %arrayidx = getelementptr inbounds double, double* %B, i64 %indvars.iv
57  store double %conv, double* %arrayidx, align 8
58  %1 = trunc i64 %indvars.iv to i32
59  %conv5 = sitofp i32 %1 to double
60  %arrayidx7 = getelementptr inbounds double, double* %D, i64 %indvars.iv
61  store double %conv5, double* %arrayidx7, align 8
62  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
63  %exitcond = icmp ne i64 %indvars.iv.next, 10000
64  br i1 %exitcond, label %for.body4, label %for.end
65
66for.end:                                          ; preds = %for.body4
67  %arrayidx9 = getelementptr inbounds double, double* %B, i64 %indvars.iv3
68  %2 = bitcast double* %arrayidx9 to i64*
69  %3 = load i64, i64* %2, align 8
70  %arrayidx11 = getelementptr inbounds double, double* %A, i64 %indvars.iv3
71  %4 = bitcast double* %arrayidx11 to i64*
72  store i64 %3, i64* %4, align 8
73  %arrayidx13 = getelementptr inbounds double, double* %D, i64 %indvars.iv3
74  %5 = bitcast double* %arrayidx13 to i64*
75  %6 = load i64, i64* %5, align 8
76  %arrayidx15 = getelementptr inbounds double, double* %C, i64 %indvars.iv3
77  %7 = bitcast double* %arrayidx15 to i64*
78  store i64 %6, i64* %7, align 8
79  %indvars.iv.next4 = add nuw nsw i64 %indvars.iv3, 1
80  %exitcond5 = icmp ne i64 %indvars.iv.next4, 10000
81  br i1 %exitcond5, label %for.body, label %for.end18
82
83for.end18:                                        ; preds = %for.end
84  ret void
85}
86