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
216
217
218
219
220
221
222
223
224
225
226
227
228
|
// =================================================================================================
// 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 tunes the Xgemm routine at a high-level: choosing between the direct (single-kernel)
// and the in-direct (kernel plus pre/post-processing) methods.
//
// =================================================================================================
#include <exception>
#include <string>
#include <vector>
#include <iostream>
#include "utilities/utilities.hpp"
#include "test/test_utilities.hpp"
#include "tuning/routines/routine_tuner.hpp"
namespace clblast {
// =================================================================================================
template <typename T>
void RunGemmRoutineMNK(const size_t m, const size_t n, const size_t k,
const Queue& queue, const std::vector<Buffer<T>>& buffers) {
auto queue_plain = queue();
auto event = cl_event{};
auto status = Gemm(Layout::kRowMajor, Transpose::kNo, Transpose::kNo,
m, n, k, ConstantOne<T>(),
buffers[0](), 0, k,
buffers[1](), 0, n, ConstantOne<T>(),
buffers[2](), 0, n,
&queue_plain, &event);
if (status != StatusCode::kSuccess) {
throw RuntimeError("Gemm failed with status " + ToString(status));
}
clWaitForEvents(1, &event);
clReleaseEvent(event);
}
template <typename T>
void RunGemmRoutine(const size_t value, const Queue& queue, const std::vector<Buffer<T>>& buffers) {
RunGemmRoutineMNK(value, value, value, queue, buffers);
}
template <typename T, size_t batch_count>
void RunGemmBatchedRoutine(const size_t value, const Queue& queue, const std::vector<Buffer<T>>& buffers) {
auto offsets = std::vector<size_t>(batch_count);
auto factors = std::vector<T>(batch_count);
for (auto i = size_t{0}; i < batch_count; ++i) {
offsets[i] = batch_count * value;
factors[i] = ConstantOne<T>();
}
auto queue_plain = queue();
auto event = cl_event{};
auto status = GemmBatched(Layout::kRowMajor, Transpose::kNo, Transpose::kNo,
value, value, value, factors.data(),
buffers[0](), offsets.data(), value,
buffers[1](), offsets.data(), value, factors.data(),
buffers[2](), offsets.data(), value, batch_count,
&queue_plain, &event);
if (status != StatusCode::kSuccess) {
throw RuntimeError("GemmBatched failed with status " + ToString(status));
}
clWaitForEvents(1, &event);
clReleaseEvent(event);
}
template <typename T, size_t batch_count>
void RunGemmStridedBatchedRoutine(const size_t value, const Queue& queue, const std::vector<Buffer<T>>& buffers) {
auto queue_plain = queue();
auto event = cl_event{};
auto status = GemmStridedBatched(Layout::kRowMajor, Transpose::kNo, Transpose::kNo,
value, value, value, ConstantOne<T>(),
buffers[0](), 0, value, value * value,
buffers[1](), 0, value, value * value, ConstantOne<T>(),
buffers[2](), 0, value, value * value, batch_count,
&queue_plain, &event);
if (status != StatusCode::kSuccess) {
throw RuntimeError("Gemm failed with status " + ToString(status));
}
clWaitForEvents(1, &event);
clReleaseEvent(event);
}
// =================================================================================================
template <typename T>
void TuneGemmSingleSize(const Platform& platform, const Device& device, const Context& context, Queue& queue,
const size_t m, const size_t n, const size_t k, const size_t num_runs) {
// Buffers
auto buffers = std::vector<Buffer<T>>{
Buffer<T>(context, m * k),
Buffer<T>(context, k * n),
Buffer<T>(context, m * n)
};
const auto FunctionToTune = [&]() { RunGemmRoutineMNK(m, n, k, queue, buffers); };
// Collects the timings for two methods
auto scores = std::vector<TuningResult>();
const auto methods = std::vector<std::string>{"in-direct", "direct"};
for (auto& method: methods) {
printf("* Testing the %s routine\n", method.c_str());
const auto limit = (method == "in-direct") ? 0 : std::max(std::max(m, n), k) + 1; // small or large number
ForceSelectIndirectFrom<T>(limit, device, "GemmRoutine", "XGEMM_MIN_INDIRECT_SIZE");
auto time_ms = -1.0;
try {
time_ms = TimeFunction(num_runs, FunctionToTune);
printf(" --> %9.2lf ms\n", time_ms);
}
catch (...) {
const auto status_code = DispatchExceptionCatchAll(true);
printf(" --> error %-5d\n", static_cast<int>(status_code));
}
auto tuning_results = Configuration();
tuning_results["XGEMM_MIN_INDIRECT_SIZE"] = limit;
tuning_results["PRECISION"] = static_cast<size_t>(PrecisionValue<T>());
scores.push_back(TuningResult{"gemm_kernel_selection_single_size", time_ms, tuning_results});
}
// Outputs the results as JSON to disk, including some meta-data
const auto precision_string = std::to_string(static_cast<size_t>(PrecisionValue<T>()));
auto metadata = std::vector<std::pair<std::string,std::string>>{
{"kernel_family", "gemm_routine_single_size"},
{"precision", precision_string},
{"arg_m", ToString(m)},
{"arg_n", ToString(n)},
{"arg_k", ToString(k)},
};
PrintTimingsToFileAsJSON("clblast_gemm_routine_single_size_" + precision_string + ".json",
device, platform, metadata, scores);
}
// =================================================================================================
template <typename T>
void TuneXgemm(int argc, char* argv[]) {
auto command_line_args = RetrieveCommandLineArguments(argc, argv);
auto help = std::string{"* Options given/available:\n"};
const auto platform_id = GetArgument(command_line_args, help, kArgPlatform, ConvertArgument(std::getenv("CLBLAST_PLATFORM"), size_t{0}));
const auto device_id = GetArgument(command_line_args, help, kArgDevice, ConvertArgument(std::getenv("CLBLAST_DEVICE"), size_t{0}));
const auto precision = GetArgument(command_line_args, help, kArgPrecision, Precision::kSingle);
const auto num_runs = GetArgument(command_line_args, help, kArgNumRuns, size_t{10});
const auto arg_m = GetArgument(command_line_args, help, kArgM, -1); // optional
const auto arg_n = GetArgument(command_line_args, help, kArgN, -1); // optional
const auto arg_k = GetArgument(command_line_args, help, kArgK, -1); // optional
fprintf(stdout, "%s\n", help.c_str());
// OpenCL initialisation
const auto platform = Platform(platform_id);
const auto device = Device(platform, device_id);
if (!PrecisionSupported<T>(device)) {
printf("* Unsupported precision, skipping this tuning run\n");
return;
}
const auto context = Context(device);
auto queue = Queue(context, device);
// Pre-load GEMM kernel tuning results if they exist
printf("* The GEMM routine tuner requires already tuned kernels\n");
printf(" Applying tuning results from disk if they exist...\n\n");
const auto kernel_names = {"xgemm_1", "xgemm_direct_1", "copy", "pad", "transpose", "padtranspose"};
for (const auto& kernel_name : kernel_names) {
const auto tuner_file_name = "clblast_" + std::string{kernel_name} + "_" +
ToString(static_cast<int>(precision)) + ".json";
printf("* Looking for tuning results in the current folder: '%s'\n", tuner_file_name.c_str());
if (std::ifstream(tuner_file_name)) { // Checks if the file exists on disk
OverrideParametersFromJSONFiles({tuner_file_name}, device(), precision);
}
else {
printf(" Not found: assuming the kernel '%s' is already tuned\n\n", kernel_name);
}
}
// Test for only one m/n/k size
if (arg_m != -1 || arg_n != -1 || arg_k != -1) {
printf("* Tuning for one specific size: m=%d, n=%d, k=%d\n", arg_m, arg_n, arg_k);
if (arg_m == -1 || arg_n == -1 || arg_k == -1) {
printf("* Error: If one of m/n/k specified, please specify all three\n");
return;
}
TuneGemmSingleSize<T>(platform, device, context, queue, static_cast<size_t>(arg_m),
static_cast<size_t>(arg_n), static_cast<size_t>(arg_k), num_runs);
}
else {
// Run the tuners for the XGEMM routines
TuneKernelSelection<T>(platform, device, context, queue, precision, RunGemmRoutine<T>,
64, 2048, 64, 1, num_runs,
"gemm", "GemmRoutine", "gemm_routine", "XGEMM_MIN_INDIRECT_SIZE");
//TuneKernelSelection<T>(platform, device, context, queue, precision, RunGemmBatchedRoutine<T, 30>,
// 16, 128, 32, 30, num_runs,
// "gemmbatched", "GemmRoutine", "gemm_routine_2", "XGEMMBATCHED_MIN_INDIRECT_SIZE");
//TuneKernelSelection<T>(platform, device, context, queue, precision, RunGemmStridedBatchedRoutine<T, 30>,
// 16, 128, 32, 30, num_runs,
// "gemmstridedbatched", "GemmRoutine", "gemm_routine_3", "XGEMMSTRIDEDBATCHED_MIN_INDIRECT_SIZE");
}
printf("* Completed tuning process\n");
printf("\n");
}
// =================================================================================================
} // namespace clblast
// Shortcuts to the clblast namespace
using half = clblast::half;
using float2 = clblast::float2;
using double2 = clblast::double2;
// Main function (not within the clblast namespace)
int main(int argc, char *argv[]) {
const auto command_line_args = clblast::RetrieveCommandLineArguments(argc, argv);
switch(clblast::GetPrecision(command_line_args)) {
case clblast::Precision::kHalf: clblast::TuneXgemm<half>(argc, argv); break;
case clblast::Precision::kSingle: clblast::TuneXgemm<float>(argc, argv); break;
case clblast::Precision::kDouble: clblast::TuneXgemm<double>(argc, argv); break;
case clblast::Precision::kComplexSingle: clblast::TuneXgemm<float2>(argc, argv); break;
case clblast::Precision::kComplexDouble: clblast::TuneXgemm<double2>(argc, argv); break;
}
return 0;
}
// =================================================================================================
|
