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
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
|
// =================================================================================================
// This file is part of the CLBlast project. The project is licensed under Apache Version 2.0. This
// project loosely follows the Google C++ styleguide and uses a tab-size of two spaces and a max-
// width of 100 characters per line.
//
// Author(s):
// Cedric Nugteren <www.cedricnugteren.nl>
//
// This file implements the Xgemm class (see the header for information about the class).
//
// =================================================================================================
#include "internal/routines/level3/xgemm.h"
#include <string>
#include <vector>
namespace clblast {
// =================================================================================================
// Specific implementations to get the memory-type based on a template argument
template <> const Precision Xgemm<float>::precision_ = Precision::kSingle;
template <> const Precision Xgemm<double>::precision_ = Precision::kDouble;
template <> const Precision Xgemm<float2>::precision_ = Precision::kComplexSingle;
template <> const Precision Xgemm<double2>::precision_ = Precision::kComplexDouble;
// =================================================================================================
// Constructor: forwards to base class constructor
template <typename T>
Xgemm<T>::Xgemm(Queue &queue, EventPointer event, const std::string &name):
Routine<T>(queue, event, name, {"Copy","Pad","Transpose","Padtranspose","Xgemm"}, precision_) {
source_string_ =
#include "../../kernels/level3/copy.opencl"
#include "../../kernels/level3/pad.opencl"
#include "../../kernels/level3/transpose.opencl"
#include "../../kernels/level3/padtranspose.opencl"
#include "../../kernels/level3/xgemm_part1.opencl"
#include "../../kernels/level3/xgemm_part2.opencl"
;
}
// =================================================================================================
// The main routine
template <typename T>
StatusCode Xgemm<T>::DoGemm(const Layout layout,
const Transpose a_transpose, const Transpose b_transpose,
const size_t m, const size_t n, const size_t k,
const T alpha,
const Buffer<T> &a_buffer, const size_t a_offset, const size_t a_ld,
const Buffer<T> &b_buffer, const size_t b_offset, const size_t b_ld,
const T beta,
const Buffer<T> &c_buffer, const size_t c_offset, const size_t c_ld) {
// Makes sure all dimensions are larger than zero
if ((m == 0) || (n == 0) || (k == 0)) { return StatusCode::kInvalidDimension; }
// Computes whether or not the matrices are transposed in memory. This is based on their layout
// (row or column-major) and whether or not they are requested to be pre-transposed. Note
// that the Xgemm kernel expects either matrices A and C (in case of row-major) or B (in case of
// col-major) to be transformed, so transposing requirements are not the same as whether or not
// the matrix is actually transposed in memory.
auto a_rotated = (layout == Layout::kColMajor && a_transpose != Transpose::kNo) ||
(layout == Layout::kRowMajor && a_transpose == Transpose::kNo);
auto b_rotated = (layout == Layout::kColMajor && b_transpose != Transpose::kNo) ||
(layout == Layout::kRowMajor && b_transpose == Transpose::kNo);
auto c_rotated = (layout == Layout::kRowMajor);
auto a_do_transpose = a_rotated;
auto b_do_transpose = !b_rotated;
auto c_do_transpose = c_rotated;
// In case of complex data-types, the transpose can also become a conjugate transpose
auto a_conjugate = (a_transpose == Transpose::kConjugate);
auto b_conjugate = (b_transpose == Transpose::kConjugate);
// Computes the first and second dimensions of the 3 matrices taking into account whether the
// matrices are rotated or not
auto a_one = (a_rotated) ? k : m;
auto a_two = (a_rotated) ? m : k;
auto b_one = (b_rotated) ? n : k;
auto b_two = (b_rotated) ? k : n;
auto c_one = (c_rotated) ? n : m;
auto c_two = (c_rotated) ? m : n;
// Tests three matrices (A, B, C) for validity, first from a perspective of the OpenCL buffers and
// their sizes, and then from a perspective of parameter values (e.g. m, n, k). Tests whether the
// OpenCL buffers are valid and non-zero and whether the OpenCL buffers have sufficient storage
// space. Also tests that the leading dimensions of:
// matrix A cannot be less than K when rotated, or less than M when not-rotated
// matrix B cannot be less than N when rotated, or less than K when not-rotated
// matrix C cannot be less than N when rotated, or less than M when not-rotated
auto status = TestMatrixA(a_one, a_two, a_buffer, a_offset, a_ld, sizeof(T));
if (ErrorIn(status)) { return status; }
status = TestMatrixB(b_one, b_two, b_buffer, b_offset, b_ld, sizeof(T));
if (ErrorIn(status)) { return status; }
status = TestMatrixC(c_one, c_two, c_buffer, c_offset, c_ld, sizeof(T));
if (ErrorIn(status)) { return status; }
// Calculates the ceiled versions of m, n, and k
auto m_ceiled = Ceil(m, db_["MWG"]);
auto n_ceiled = Ceil(n, db_["NWG"]);
auto k_ceiled = Ceil(k, db_["KWG"]);
// The padded/transposed input/output matrices: if memory allocation fails, throw an exception
try {
// Loads the program from the database
auto& program = GetProgramFromCache();
// Determines whether or not temporary matrices are needed
auto a_no_temp = a_one == m_ceiled && a_two == k_ceiled && a_ld == m_ceiled && a_offset == 0 &&
a_do_transpose == false && a_conjugate == false;
auto b_no_temp = b_one == n_ceiled && b_two == k_ceiled && b_ld == n_ceiled && b_offset == 0 &&
b_do_transpose == false && b_conjugate == false;
auto c_no_temp = c_one == m_ceiled && c_two == n_ceiled && c_ld == m_ceiled && c_offset == 0 &&
c_do_transpose == false;
// Creates the temporary matrices
auto a_temp = (a_no_temp) ? a_buffer : Buffer<T>(context_, k_ceiled*m_ceiled);
auto b_temp = (b_no_temp) ? b_buffer : Buffer<T>(context_, k_ceiled*n_ceiled);
auto c_temp = (c_no_temp) ? c_buffer : Buffer<T>(context_, m_ceiled*n_ceiled);
// Events of all kernels (including pre/post processing kernels)
auto eventWaitList = std::vector<Event>();
auto emptyEventList = std::vector<Event>();
// Runs the pre-processing kernel for matrix A. This transposes the matrix, but also pads zeros
// to fill it up until it reaches a certain multiple of size (kernel parameter dependent). In
// case nothing has to be done, these kernels can be skipped.
if (!a_no_temp) {
auto eventProcessA = Event();
status = PadCopyTransposeMatrix(eventProcessA.pointer(), emptyEventList,
a_one, a_two, a_ld, a_offset, a_buffer,
m_ceiled, k_ceiled, m_ceiled, 0, a_temp,
program, true, a_do_transpose, a_conjugate);
if (ErrorIn(status)) { return status; }
eventWaitList.push_back(eventProcessA);
}
// As above, but now for matrix B
if (!b_no_temp) {
auto eventProcessB = Event();
status = PadCopyTransposeMatrix(eventProcessB.pointer(), emptyEventList,
b_one, b_two, b_ld, b_offset, b_buffer,
n_ceiled, k_ceiled, n_ceiled, 0, b_temp,
program, true, b_do_transpose, b_conjugate);
if (ErrorIn(status)) { return status; }
eventWaitList.push_back(eventProcessB);
}
// As above, but now for matrix C. This is only necessary if C is used both as input and output.
if (!c_no_temp && beta != static_cast<T>(0)) {
auto eventProcessC = Event();
status = PadCopyTransposeMatrix(eventProcessC.pointer(), emptyEventList,
c_one, c_two, c_ld, c_offset, c_buffer,
m_ceiled, n_ceiled, m_ceiled, 0, c_temp,
program, true, c_do_transpose, false);
if (ErrorIn(status)) { return status; }
eventWaitList.push_back(eventProcessC);
}
// Retrieves the Xgemm kernel from the compiled binary
try {
auto kernel = Kernel(program, "Xgemm");
// Sets the kernel arguments
kernel.SetArgument(0, static_cast<int>(m_ceiled));
kernel.SetArgument(1, static_cast<int>(n_ceiled));
kernel.SetArgument(2, static_cast<int>(k_ceiled));
kernel.SetArgument(3, alpha);
kernel.SetArgument(4, beta);
kernel.SetArgument(5, a_temp());
kernel.SetArgument(6, b_temp());
kernel.SetArgument(7, c_temp());
// Computes the global and local thread sizes
auto global = std::vector<size_t>{
(m_ceiled * db_["MDIMC"]) / db_["MWG"],
(n_ceiled * db_["NDIMC"]) / db_["NWG"]
};
auto local = std::vector<size_t>{db_["MDIMC"], db_["NDIMC"]};
// Launches the kernel
auto eventKernel = Event();
status = RunKernel(kernel, global, local, eventKernel.pointer(), eventWaitList);
if (ErrorIn(status)) { return status; }
eventWaitList.push_back(eventKernel);
// Runs the post-processing kernel if needed
if (!c_no_temp) {
status = PadCopyTransposeMatrix(event_, eventWaitList,
m_ceiled, n_ceiled, m_ceiled, 0, c_temp,
c_one, c_two, c_ld, c_offset, c_buffer,
program, false, c_do_transpose, false);
if (ErrorIn(status)) { return status; }
}
// Successfully finished the computation
return StatusCode::kSuccess;
} catch (...) { return StatusCode::kInvalidKernel; }
} catch (...) { return StatusCode::kTempBufferAllocFailure; }
}
// =================================================================================================
// Compiles the templated class
template class Xgemm<float>;
template class Xgemm<double>;
template class Xgemm<float2>;
template class Xgemm<double2>;
// =================================================================================================
} // namespace clblast
|
