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
140
141
142
143
144
145
146
/*
* Copyright 2015-2017 Philippe Tillet
* Copyright (c) 2017, Intel Corporation
*
* Permission is hereby granted, free of charge, to any person obtaining
* a copy of this software and associated documentation files
* (the "Software"), to deal in the Software without restriction,
* including without limitation the rights to use, copy, modify, merge,
* publish, distribute, sublicense, and/or sell copies of the Software,
* and to permit persons to whom the Software is furnished to do so,
* subject to the following conditions:
*
* The above copyright notice and this permission notice shall be
* included in all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
* IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY
* CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
*/
#ifdef HAVE_OPENCL
#include <sstream>
#include "precomp.hpp"
#include "opencl_kernels_core.hpp"
#include "opencv2/core/opencl/runtime/opencl_clamdblas.hpp"
#include "opencv2/core/opencl/runtime/opencl_core.hpp"
namespace cv
{
static bool intel_gpu_gemm(
UMat A, Size sizeA,
UMat B, Size sizeB,
UMat D, Size sizeD,
double alpha, double beta,
bool atrans, bool btrans)
{
CV_UNUSED(sizeB);
int M = sizeD.height, N = sizeD.width, K = ((atrans)? sizeA.height : sizeA.width);
std::string kernelName;
bool ret = true;
size_t lx = 8, ly = 4;
size_t dx = 4, dy = 8;
if(!atrans && !btrans)
{
if (M % 32 == 0 && N % 32 == 0 && K % 16 == 0)
{
kernelName = "intelblas_gemm_buffer_NN_sp";
}
else
{
kernelName = "intelblas_gemm_buffer_NN";
}
}
else if(atrans && !btrans)
{
kernelName = "intelblas_gemm_buffer_TN";
}
else if(!atrans && btrans)
{
kernelName = "intelblas_gemm_buffer_NT";
ly = 16;
dx = 1;
}
else
{
kernelName = "intelblas_gemm_buffer_TT";
}
const size_t gx = (size_t)(N + dx - 1) / dx;
const size_t gy = (size_t)(M + dy - 1) / dy;
size_t local[] = {lx, ly, 1};
size_t global[] = {(gx + lx - 1) / lx * lx, (gy + ly - 1) / ly * ly, 1};
int stride = (M * N < 1024 * 1024) ? 10000000 : 256;
ocl::Queue q;
String errmsg;
const ocl::Program program = ocl::Context::getDefault().getProg(ocl::core::intel_gemm_oclsrc, "", errmsg);
if(!atrans && btrans)
{
ocl::Kernel k(kernelName.c_str(), program);
if (k.empty())
{
return false;
}
k.args(ocl::KernelArg::PtrReadOnly(A),
(int) (A.offset / sizeof(float)),
ocl::KernelArg::PtrReadOnly(B),
(int) (B.offset / sizeof(float)),
ocl::KernelArg::PtrWriteOnly(D),
(int) (D.offset / sizeof(float)),
M, N, K,
(float)alpha,
(float)beta,
(int)(A.step / sizeof(float)),
(int)(B.step / sizeof(float)),
(int)(D.step / sizeof(float))
);
ret = k.run(2, global, local, false, q);
}
else
{
for(int start_index = 0; start_index < K; start_index += stride)
{
ocl::Kernel k(kernelName.c_str(), program);
k.args(ocl::KernelArg::PtrReadOnly(A),
(int) (A.offset / sizeof(float)),
ocl::KernelArg::PtrReadOnly(B),
(int) (B.offset / sizeof(float)),
ocl::KernelArg::PtrWriteOnly(D),
(int) (D.offset / sizeof(float)),
M, N, K,
(float)alpha,
(float)beta,
(int)(A.step / sizeof(float)),
(int)(B.step / sizeof(float)),
(int)(D.step / sizeof(float)),
(int) start_index, // 14 start_index
stride);
ret = k.run(2, global, local, false, q);
if (!ret) return ret;
}
}
return ret;
}
} // namespace cv
#endif