|
236
|
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
|
|
150
|
5 ;
|
|
|
6 ; Verify that the accesses are correctly expanded for MemoryKind::PHI
|
|
|
7 ; tmp_05 and tmp2_06 are not expanded because they need copy-in.
|
|
|
8 ;
|
|
|
9 ; Original source code :
|
|
|
10 ;
|
|
|
11 ; #define Ni 10000
|
|
|
12 ; #define Nj 10000
|
|
|
13 ;
|
|
|
14 ; void mse(double A[Ni], double B[Nj]) {
|
|
|
15 ; int i,j;
|
|
|
16 ; double tmp = 6;
|
|
|
17 ; double tmp2 = 9;
|
|
|
18 ; for (i = 0; i < Ni; i++) {
|
|
|
19 ; for(j = 0; j < Nj; j++) {
|
|
|
20 ; tmp = tmp + tmp2;
|
|
|
21 ; tmp2 = i*j;
|
|
|
22 ; }
|
|
|
23 ; }
|
|
|
24 ; }
|
|
|
25 ;
|
|
|
26 ; Check that the pass detects that tmp_05 and tmp2_06 read from their original values.
|
|
|
27 ;
|
|
|
28 ; MSE-DAG: MemRef_tmp_05__phi read from its original value.
|
|
|
29 ; MSE-DAG: MemRef_tmp2_06__phi read from its original value.
|
|
|
30 ;
|
|
|
31 ; Check that the SAI are created except the expanded SAI of tmp_05 and tmp2_06.
|
|
|
32 ;
|
|
|
33 ; CHECK-DAG: double MemRef_add_lcssa__phi_Stmt_for_inc4_expanded[10000]; // Element size 8
|
|
|
34 ; CHECK-DAG: double MemRef_tmp2_13__phi_Stmt_for_inc_expanded[10000][10000]; // Element size
|
|
|
35 ; CHECK-DAG: double MemRef_conv_lcssa__phi_Stmt_for_inc4_expanded[10000]; // Element size 8
|
|
|
36 ; CHECK-DAG: double MemRef_tmp_12__phi_Stmt_for_inc_expanded[10000][10000]; // Element size 8
|
|
|
37 ; CHECK-NOT: double MemRef_tmp_05__phi_Stmt_for_body_expanded[10000]; // Element size 8
|
|
|
38 ; CHECK-NOT: double MemRef_tmp2_06__phi_Stmt_for_body_expanded[10000]; // Element size 8
|
|
|
39 ;
|
|
|
40 ; Check that the memory accesses are modified except those related to tmp_05 and tmp_06.
|
|
|
41 ;
|
|
|
42 ; CHECK-NOT: new: { Stmt_for_body[i0] -> MemRef_tmp2_06__phi_Stmt_for_body_expanded[i0] };
|
|
|
43 ; CHECK-NOT: new: { Stmt_for_body[i0] -> MemRef_tmp_05__phi_Stmt_for_body_expanded[i0] };
|
|
|
44 ; CHECK: new: { Stmt_for_body[i0] -> MemRef_tmp2_13__phi_Stmt_for_inc_expanded[i0, 0] };
|
|
|
45 ; CHECK: new: { Stmt_for_body[i0] -> MemRef_tmp_12__phi_Stmt_for_inc_expanded[i0, 0] };
|
|
|
46 ; CHECK: new: { Stmt_for_inc[i0, i1] -> MemRef_tmp2_13__phi_Stmt_for_inc_expanded[i0, 1 + i1] : i1 <= 9998 };
|
|
|
47 ; CHECK: new: { Stmt_for_inc[i0, i1] -> MemRef_tmp2_13__phi_Stmt_for_inc_expanded[i0, i1] };
|
|
|
48 ; CHECK: new: { Stmt_for_inc[i0, i1] -> MemRef_tmp_12__phi_Stmt_for_inc_expanded[i0, 1 + i1] : i1 <= 9998 };
|
|
|
49 ; CHECK: new: { Stmt_for_inc[i0, i1] -> MemRef_tmp_12__phi_Stmt_for_inc_expanded[i0, i1] };
|
|
|
50 ; CHECK: new: { Stmt_for_inc[i0, 9999] -> MemRef_conv_lcssa__phi_Stmt_for_inc4_expanded[i0] };
|
|
|
51 ; CHECK: new: { Stmt_for_inc[i0, 9999] -> MemRef_add_lcssa__phi_Stmt_for_inc4_expanded[i0] };
|
|
|
52 ; CHECK-NOT: new: { Stmt_for_inc4[i0] -> MemRef_tmp2_06__phi_Stmt_for_body_expanded[1 + i0] : i0 <= 9998 };
|
|
|
53 ; CHECK-NOT: new: { Stmt_for_inc4[i0] -> MemRef_tmp_05__phi_Stmt_for_body_expanded[1 + i0] : i0 <= 9998 };
|
|
|
54 ; CHECK: new: { Stmt_for_inc4[i0] -> MemRef_conv_lcssa__phi_Stmt_for_inc4_expanded[i0] };
|
|
|
55 ; CHECK: new: { Stmt_for_inc4[i0] -> MemRef_add_lcssa__phi_Stmt_for_inc4_expanded[i0] };
|
|
236
|
56 ;
|
|
150
|
57 target datalayout = "e-m:e-i64:64-f80:128-n8:16:32:64-S128"
|
|
|
58 target triple = "x86_64-unknown-linux-gnu"
|
|
|
59
|
|
|
60 define void @tmp(double* %A, double* %B) {
|
|
|
61 entry:
|
|
|
62 br label %entry.split
|
|
|
63
|
|
|
64 entry.split: ; preds = %entry
|
|
|
65 br label %for.body
|
|
|
66
|
|
|
67 for.body: ; preds = %entry.split, %for.inc4
|
|
|
68 %tmp2.06 = phi double [ 9.000000e+00, %entry.split ], [ %conv.lcssa, %for.inc4 ]
|
|
|
69 %tmp.05 = phi double [ 6.000000e+00, %entry.split ], [ %add.lcssa, %for.inc4 ]
|
|
|
70 %i.04 = phi i32 [ 0, %entry.split ], [ %inc5, %for.inc4 ]
|
|
|
71 br label %for.inc
|
|
|
72
|
|
|
73 for.inc: ; preds = %for.body, %for.inc
|
|
|
74 %tmp2.13 = phi double [ %tmp2.06, %for.body ], [ %conv, %for.inc ]
|
|
|
75 %tmp.12 = phi double [ %tmp.05, %for.body ], [ %add, %for.inc ]
|
|
|
76 %j.01 = phi i32 [ 0, %for.body ], [ %inc, %for.inc ]
|
|
|
77 %mul = mul nuw nsw i32 %j.01, %i.04
|
|
|
78 %conv = sitofp i32 %mul to double
|
|
|
79 %add = fadd double %tmp.12, %tmp2.13
|
|
|
80 %inc = add nuw nsw i32 %j.01, 1
|
|
|
81 %exitcond = icmp ne i32 %inc, 10000
|
|
|
82 br i1 %exitcond, label %for.inc, label %for.inc4
|
|
|
83
|
|
|
84 for.inc4: ; preds = %for.inc
|
|
|
85 %conv.lcssa = phi double [ %conv, %for.inc ]
|
|
|
86 %add.lcssa = phi double [ %add, %for.inc ]
|
|
|
87 %inc5 = add nuw nsw i32 %i.04, 1
|
|
|
88 %exitcond7 = icmp ne i32 %inc5, 10000
|
|
|
89 br i1 %exitcond7, label %for.body, label %for.end6
|
|
|
90
|
|
|
91 for.end6: ; preds = %for.inc4
|
|
|
92 ret void
|
|
|
93 }
|
