llvm-project

1
; RUN: opt %loadNPMPolly '-passes=polly-delicm,polly-simplify,polly-opt-isl' \
2
; RUN: -polly-pattern-matching-based-opts=true \
3
; RUN: -polly-tc-opt=true -debug -disable-output < %s 2>&1 | FileCheck %s
4
; REQUIRES: asserts
5
;
6
; Check that the pattern matching detects the tensor contraction pattern
7
; after a full run of -polly-delicm. This test case generates the following
8
; schedule, which contans two band nodes. Without DeLICM two statement are
9
; generated.
10
;
11
; domain: "{ Stmt5[i0, i1, i2, i3, i4, i5] : 0 <= i0 <= 31 and 0 <= i1 <= 31 and
12
;                                            0 <= i2 <= 31 and 0 <= i3 <= 31 and
13
;                                            0 <= i4 <= 31 and 0 <= i5 <= 31 }"
14
; child:
15
;   schedule: "[{ Stmt5[i0, i1, i2, i3, i4, i5] -> [(i0)] },
16
;               { Stmt5[i0, i1, i2, i3, i4, i5] -> [(i1)] },
17
;               { Stmt5[i0, i1, i2, i3, i4, i5] -> [(i2)] },
18
;               { Stmt5[i0, i1, i2, i3, i4, i5] -> [(i4)] },
19
;               { Stmt5[i0, i1, i2, i3, i4, i5] -> [(i3)] }]"
20
;   permutable: 1
21
;   coincident: [ 1, 1, 1, 1, 0 ]
22
;   child:
23
;     schedule: "[{ Stmt5[i0, i1, i2, i3, i4, i5] -> [(i5)] }]"
24
;     permutable: 1
25
;     child:
26
;       leaf
27
;
28
;   for (i = 0; i < 32; i++)
29
;     for (j = 0; j < 32; j++)
30
;       for (k = 0; k < 32; ++k)
31
;         for (l = 0; l < 32; ++l)
32
;           for (w = 0; w < 32; ++w)
33
;             for (q = 0; q < 32; ++q)
34
;               C[i][j][k][w] += A[i][l][j][q] * B[q][w][l][k];
35
;
36
; CHECK: The tensor contraction pattern was detected
37
;
38
target datalayout = "e-m:e-p270:32:32-p271:32:32-p272:64:64-i64:64-f80:128-n8:16:32:64-S128"
39
target triple = "x86_64-unknown-linux-gnu"
40

41
define internal fastcc void @kernel_tc(ptr nocapture %C, ptr nocapture readonly %A, ptr nocapture readonly %B) {
42
entry:
43
  br label %for.cond1.preheader
44

45
for.cond1.preheader:                              ; preds = %for.inc50, %entry
46
  %indvars.iv19 = phi i64 [ 0, %entry ], [ %indvars.iv.next20, %for.inc50 ]
47
  br label %for.cond4.preheader
48

49
for.cond4.preheader:                              ; preds = %for.inc47, %for.cond1.preheader
50
  %indvars.iv16 = phi i64 [ 0, %for.cond1.preheader ], [ %indvars.iv.next17, %for.inc47 ]
51
  br label %for.cond7.preheader
52

53
for.cond7.preheader:                              ; preds = %for.inc44, %for.cond4.preheader
54
  %indvars.iv13 = phi i64 [ 0, %for.cond4.preheader ], [ %indvars.iv.next14, %for.inc44 ]
55
  br label %for.cond10.preheader
56

57
for.cond10.preheader:                             ; preds = %for.inc41, %for.cond7.preheader
58
  %indvars.iv10 = phi i64 [ 0, %for.cond7.preheader ], [ %indvars.iv.next11, %for.inc41 ]
59
  br label %for.cond13.preheader
60

61
for.cond13.preheader:                             ; preds = %for.inc38, %for.cond10.preheader
62
  %indvars.iv7 = phi i64 [ 0, %for.cond10.preheader ], [ %indvars.iv.next8, %for.inc38 ]
63
  %arrayidx37 = getelementptr inbounds [32 x [32 x [32 x double]]], ptr %C, i64 %indvars.iv19, i64 %indvars.iv16, i64 %indvars.iv13, i64 %indvars.iv7
64
  %.pre = load double, ptr %arrayidx37, align 8
65
  br label %for.body15
66

67
for.body15:                                       ; preds = %for.body15, %for.cond13.preheader
68
  %i = phi double [ %.pre, %for.cond13.preheader ], [ %add, %for.body15 ]
69
  %indvars.iv = phi i64 [ 0, %for.cond13.preheader ], [ %indvars.iv.next, %for.body15 ]
70
  %arrayidx21 = getelementptr inbounds [32 x [32 x [32 x double]]], ptr %A, i64 %indvars.iv19, i64 %indvars.iv10, i64 %indvars.iv16, i64 %indvars.iv
71
  %i1 = load double, ptr %arrayidx21, align 8
72
  %arrayidx29 = getelementptr inbounds [32 x [32 x [32 x double]]], ptr %B, i64 %indvars.iv, i64 %indvars.iv7, i64 %indvars.iv10, i64 %indvars.iv13
73
  %i2 = load double, ptr %arrayidx29, align 8
74
  %mul = fmul fast double %i2, %i1
75
  %add = fadd fast double %i, %mul
76
  store double %add, ptr %arrayidx37, align 8
77
  %indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
78
  %exitcond.not = icmp eq i64 %indvars.iv.next, 32
79
  br i1 %exitcond.not, label %for.inc38, label %for.body15
80

81
for.inc38:                                        ; preds = %for.body15
82
  %indvars.iv.next8 = add nuw nsw i64 %indvars.iv7, 1
83
  %exitcond9.not = icmp eq i64 %indvars.iv.next8, 32
84
  br i1 %exitcond9.not, label %for.inc41, label %for.cond13.preheader
85

86
for.inc41:                                        ; preds = %for.inc38
87
  %indvars.iv.next11 = add nuw nsw i64 %indvars.iv10, 1
88
  %exitcond12.not = icmp eq i64 %indvars.iv.next11, 32
89
  br i1 %exitcond12.not, label %for.inc44, label %for.cond10.preheader
90

91
for.inc44:                                        ; preds = %for.inc41
92
  %indvars.iv.next14 = add nuw nsw i64 %indvars.iv13, 1
93
  %exitcond15.not = icmp eq i64 %indvars.iv.next14, 32
94
  br i1 %exitcond15.not, label %for.inc47, label %for.cond7.preheader
95

96
for.inc47:                                        ; preds = %for.inc44
97
  %indvars.iv.next17 = add nuw nsw i64 %indvars.iv16, 1
98
  %exitcond18.not = icmp eq i64 %indvars.iv.next17, 32
99
  br i1 %exitcond18.not, label %for.inc50, label %for.cond4.preheader
100

101
for.inc50:                                        ; preds = %for.inc47
102
  %indvars.iv.next20 = add nuw nsw i64 %indvars.iv19, 1
103
  %exitcond21.not = icmp eq i64 %indvars.iv.next20, 32
104
  br i1 %exitcond21.not, label %for.end52, label %for.cond1.preheader
105

106
for.end52:                                        ; preds = %for.inc50
107
  ret void
108
}
109