1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
|
; RUN: opt %loadPolly %defaultOpts -polly-import-jscop -polly-import-jscop-dir=%S -polly-codegen %vector-opt -S -dce %s | FileCheck %s
;#define M 1024
;#define N 1024
;#define K 1024
;float A[K][M];
;float B[N][K];
;float C[M][N];
;/*
;void matmul_vec(void) {
; int i, j, k;
;
;
; /* With much unrolling
; for (i=0;i<=M;i++)
; for (j=0;j<=N;j+=4)
; for (k=0;k<=K;k+=8)
; for (kk=k;kk<=k+7;kk++)
; for (jj=j;jj<=j+3;jj++)
; C[i][jj] += A[kk][i] * B[jj][kk];
; vec_load splat scalar_load
; */
; /* Without unrolling
; for (i=0;i<=M;i++)
; for (j=0;j<=N;j+=4)
; for (k=0;k<=K;k++)
; for (jj=j;jj<=j+3;jj++)
; C[i][jj] += A[k][i] * B[jj][kk];
; vec_load splat scalar_load
; /
;
;}
;i*/
;int main()
;{
; int i, j, k;
; //matmul_vec();
; for(i=0; i<M/4; i++)
; for(k=0; k<K; k++) {
; for(j=0; j<N; j++)
; C[i+0][j] += A[k][i+0] * B[j][k];
; C[i+1][j] += A[k][i+1] * B[j][k];
; C[i+2][j] += A[k][i+2] * B[j][k];
; C[i+3][j] += A[k][i+3] * B[j][k];
; }
;
; return A[42][42];
;}
; ModuleID = 'matmul_vec.s'
target datalayout = "e-p:64:64:64-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:64:64-f32:32:32-f64:64:64-v64:64:64-v128:128:128-a0:0:64-s0:64:64-f80:128:128-n8:16:32:64"
target triple = "x86_64-unknown-linux-gnu"
@A = common global [1024 x [1024 x float]] zeroinitializer, align 16
@B = common global [1024 x [1024 x float]] zeroinitializer, align 16
@C = common global [1024 x [1024 x float]] zeroinitializer, align 16
define void @matmul_vec() nounwind {
; <label>:0
br label %1
; <label>:1 ; preds = %16, %0
%indvar3 = phi i64 [ %indvar.next4, %16 ], [ 0, %0 ]
%exitcond9 = icmp ne i64 %indvar3, 1024
br i1 %exitcond9, label %2, label %17
; <label>:2 ; preds = %1
br label %3
; <label>:3 ; preds = %14, %2
%indvar1 = phi i64 [ %indvar.next2, %14 ], [ 0, %2 ]
%scevgep8 = getelementptr [1024 x [1024 x float]]* @C, i64 0, i64 %indvar3, i64 %indvar1
%exitcond6 = icmp ne i64 %indvar1, 1024
br i1 %exitcond6, label %4, label %15
; <label>:4 ; preds = %3
br label %5
; <label>:5 ; preds = %12, %4
%indvar = phi i64 [ %indvar.next, %12 ], [ 0, %4 ]
%scevgep5 = getelementptr [1024 x [1024 x float]]* @A, i64 0, i64 %indvar, i64 %indvar3
%scevgep = getelementptr [1024 x [1024 x float]]* @B, i64 0, i64 %indvar1, i64 %indvar
%exitcond = icmp ne i64 %indvar, 1024
br i1 %exitcond, label %6, label %13
; <label>:6 ; preds = %5
%7 = load float* %scevgep5, align 4
%8 = load float* %scevgep, align 4
%9 = fmul float %7, %8
%10 = load float* %scevgep8, align 4
%11 = fadd float %10, %9
store float %11, float* %scevgep8, align 4
br label %12
; <label>:12 ; preds = %6
%indvar.next = add i64 %indvar, 1
br label %5
; <label>:13 ; preds = %5
br label %14
; <label>:14 ; preds = %13
%indvar.next2 = add i64 %indvar1, 1
br label %3
; <label>:15 ; preds = %3
br label %16
; <label>:16 ; preds = %15
%indvar.next4 = add i64 %indvar3, 1
br label %1
; <label>:17 ; preds = %1
ret void
}
define i32 @main() nounwind {
call void @matmul_vec()
%1 = load float* getelementptr inbounds ([1024 x [1024 x float]]* @A, i64 0, i64 42, i64 42), align 8
%2 = fptosi float %1 to i32
ret i32 %2
}
; CHECK: load <1 x float>*
; CHECK: shufflevector <1 x float>
; CHECK: load float*
; CHECK: insertelement <4 x float>
; CHECK: load float*
; CHECK: insertelement <4 x float>
; CHECK: load float*
; CHECK: insertelement <4 x float>
; CHECK: load float*
; CHECK: insertelement <4 x float>
; CHECK: fmul <4 x float>
; CHECK: bitcast float*
; CHECK: load <4 x float>*
; CHECK: fadd <4 x float>
; CHECK: bitcast float*
; CHECK: store <4 x float>
|
