|
150
|
1 ; RUN: opt %loadPolly -polly-stmt-granularity=bb -polly-mse -analyze < %s | FileCheck %s
|
|
|
2 ; RUN: opt %loadPolly -polly-stmt-granularity=bb -polly-mse -pass-remarks-analysis="polly-mse" -analyze < %s 2>&1 | FileCheck %s --check-prefix=MSE
|
|
|
3 ;
|
|
|
4 ; Verify that the accesses are correctly expanded for MemoryKind::PHI
|
|
|
5 ; tmp_05 and tmp2_06 are not expanded because they need copy-in.
|
|
|
6 ;
|
|
|
7 ; Original source code :
|
|
|
8 ;
|
|
|
9 ; #define Ni 10000
|
|
|
10 ; #define Nj 10000
|
|
|
11 ;
|
|
|
12 ; void mse(double A[Ni], double B[Nj]) {
|
|
|
13 ; int i,j;
|
|
|
14 ; double tmp = 6;
|
|
|
15 ; double tmp2 = 9;
|
|
|
16 ; for (i = 0; i < Ni; i++) {
|
|
|
17 ; for(j = 0; j < Nj; j++) {
|
|
|
18 ; tmp = tmp + tmp2;
|
|
|
19 ; tmp2 = i*j;
|
|
|
20 ; }
|
|
|
21 ; }
|
|
|
22 ; }
|
|
|
23 ;
|
|
|
24 ; Check that the pass detects that tmp_05 and tmp2_06 read from their original values.
|
|
|
25 ;
|
|
|
26 ; MSE-DAG: MemRef_tmp_05__phi read from its original value.
|
|
|
27 ; MSE-DAG: MemRef_tmp2_06__phi read from its original value.
|
|
|
28 ;
|
|
|
29 ; Check that the SAI are created except the expanded SAI of tmp_05 and tmp2_06.
|
|
|
30 ;
|
|
|
31 ; CHECK-DAG: double MemRef_add_lcssa__phi_Stmt_for_inc4_expanded[10000]; // Element size 8
|
|
|
32 ; CHECK-DAG: double MemRef_tmp2_13__phi_Stmt_for_inc_expanded[10000][10000]; // Element size
|
|
|
33 ; CHECK-DAG: double MemRef_conv_lcssa__phi_Stmt_for_inc4_expanded[10000]; // Element size 8
|
|
|
34 ; CHECK-DAG: double MemRef_tmp_12__phi_Stmt_for_inc_expanded[10000][10000]; // Element size 8
|
|
|
35 ; CHECK-NOT: double MemRef_tmp_05__phi_Stmt_for_body_expanded[10000]; // Element size 8
|
|
|
36 ; CHECK-NOT: double MemRef_tmp2_06__phi_Stmt_for_body_expanded[10000]; // Element size 8
|
|
|
37 ;
|
|
|
38 ; Check that the memory accesses are modified except those related to tmp_05 and tmp_06.
|
|
|
39 ;
|
|
|
40 ; CHECK-NOT: new: { Stmt_for_body[i0] -> MemRef_tmp2_06__phi_Stmt_for_body_expanded[i0] };
|
|
|
41 ; CHECK-NOT: new: { Stmt_for_body[i0] -> MemRef_tmp_05__phi_Stmt_for_body_expanded[i0] };
|
|
|
42 ; CHECK: new: { Stmt_for_body[i0] -> MemRef_tmp2_13__phi_Stmt_for_inc_expanded[i0, 0] };
|
|
|
43 ; CHECK: new: { Stmt_for_body[i0] -> MemRef_tmp_12__phi_Stmt_for_inc_expanded[i0, 0] };
|
|
|
44 ; CHECK: new: { Stmt_for_inc[i0, i1] -> MemRef_tmp2_13__phi_Stmt_for_inc_expanded[i0, 1 + i1] : i1 <= 9998 };
|
|
|
45 ; CHECK: new: { Stmt_for_inc[i0, i1] -> MemRef_tmp2_13__phi_Stmt_for_inc_expanded[i0, i1] };
|
|
|
46 ; CHECK: new: { Stmt_for_inc[i0, i1] -> MemRef_tmp_12__phi_Stmt_for_inc_expanded[i0, 1 + i1] : i1 <= 9998 };
|
|
|
47 ; CHECK: new: { Stmt_for_inc[i0, i1] -> MemRef_tmp_12__phi_Stmt_for_inc_expanded[i0, i1] };
|
|
|
48 ; CHECK: new: { Stmt_for_inc[i0, 9999] -> MemRef_conv_lcssa__phi_Stmt_for_inc4_expanded[i0] };
|
|
|
49 ; CHECK: new: { Stmt_for_inc[i0, 9999] -> MemRef_add_lcssa__phi_Stmt_for_inc4_expanded[i0] };
|
|
|
50 ; CHECK-NOT: new: { Stmt_for_inc4[i0] -> MemRef_tmp2_06__phi_Stmt_for_body_expanded[1 + i0] : i0 <= 9998 };
|
|
|
51 ; CHECK-NOT: new: { Stmt_for_inc4[i0] -> MemRef_tmp_05__phi_Stmt_for_body_expanded[1 + i0] : i0 <= 9998 };
|
|
|
52 ; CHECK: new: { Stmt_for_inc4[i0] -> MemRef_conv_lcssa__phi_Stmt_for_inc4_expanded[i0] };
|
|
|
53 ; CHECK: new: { Stmt_for_inc4[i0] -> MemRef_add_lcssa__phi_Stmt_for_inc4_expanded[i0] };
|
|
|
54 ;
|
|
|
55 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
|
|
|
56 target triple = "x86_64-unknown-linux-gnu"
|
|
|
57
|
|
|
58 define void @tmp(double* %A, double* %B) {
|
|
|
59 entry:
|
|
|
60 br label %entry.split
|
|
|
61
|
|
|
62 entry.split: ; preds = %entry
|
|
|
63 br label %for.body
|
|
|
64
|
|
|
65 for.body: ; preds = %entry.split, %for.inc4
|
|
|
66 %tmp2.06 = phi double [ 9.000000e+00, %entry.split ], [ %conv.lcssa, %for.inc4 ]
|
|
|
67 %tmp.05 = phi double [ 6.000000e+00, %entry.split ], [ %add.lcssa, %for.inc4 ]
|
|
|
68 %i.04 = phi i32 [ 0, %entry.split ], [ %inc5, %for.inc4 ]
|
|
|
69 br label %for.inc
|
|
|
70
|
|
|
71 for.inc: ; preds = %for.body, %for.inc
|
|
|
72 %tmp2.13 = phi double [ %tmp2.06, %for.body ], [ %conv, %for.inc ]
|
|
|
73 %tmp.12 = phi double [ %tmp.05, %for.body ], [ %add, %for.inc ]
|
|
|
74 %j.01 = phi i32 [ 0, %for.body ], [ %inc, %for.inc ]
|
|
|
75 %mul = mul nuw nsw i32 %j.01, %i.04
|
|
|
76 %conv = sitofp i32 %mul to double
|
|
|
77 %add = fadd double %tmp.12, %tmp2.13
|
|
|
78 %inc = add nuw nsw i32 %j.01, 1
|
|
|
79 %exitcond = icmp ne i32 %inc, 10000
|
|
|
80 br i1 %exitcond, label %for.inc, label %for.inc4
|
|
|
81
|
|
|
82 for.inc4: ; preds = %for.inc
|
|
|
83 %conv.lcssa = phi double [ %conv, %for.inc ]
|
|
|
84 %add.lcssa = phi double [ %add, %for.inc ]
|
|
|
85 %inc5 = add nuw nsw i32 %i.04, 1
|
|
|
86 %exitcond7 = icmp ne i32 %inc5, 10000
|
|
|
87 br i1 %exitcond7, label %for.body, label %for.end6
|
|
|
88
|
|
|
89 for.end6: ; preds = %for.inc4
|
|
|
90 ret void
|
|
|
91 }
|
