Repository navigation
Expand file tree
/
Copy pathexecutor.cpp
More file actions
259 lines (211 loc) · 8.53 KB
/
Copy pathexecutor.cpp
File metadata and controls
259 lines (211 loc) · 8.53 KB
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
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
// SPDX-FileCopyrightText: 2017 - 2024 The Ginkgo authors
//
// SPDX-License-Identifier: BSD-3-Clause
#include <ginkgo/core/base/executor.hpp>
#include <iostream>
#include <stdexcept>
#include <thread>
#include <cuda_runtime.h>
#include <ginkgo/config.hpp>
#include <ginkgo/core/base/device.hpp>
#include <ginkgo/core/base/exception.hpp>
#include <ginkgo/core/base/exception_helpers.hpp>
#include <ginkgo/core/base/memory.hpp>
#include "cuda/base/config.hpp"
#include "cuda/base/cublas_bindings.hpp"
#include "cuda/base/cusparse_handle.hpp"
#include "cuda/base/scoped_device_id.hpp"
namespace gko {
#include "common/cuda_hip/base/executor.hpp.inc"
std::unique_ptr<CudaAllocatorBase> cuda_allocator_from_mode(
int device_id, allocation_mode mode)
{
switch (mode) {
case allocation_mode::device:
return std::make_unique<CudaAllocator>();
case allocation_mode::unified_global:
return std::make_unique<CudaUnifiedAllocator>(device_id,
cudaMemAttachGlobal);
case allocation_mode::unified_host:
return std::make_unique<CudaUnifiedAllocator>(device_id,
cudaMemAttachHost);
default:
GKO_NOT_SUPPORTED(mode);
}
}
std::shared_ptr<CudaExecutor> CudaExecutor::create(
int device_id, std::shared_ptr<Executor> master, bool device_reset,
allocation_mode alloc_mode, cudaStream_t stream)
{
return create(device_id, master,
cuda_allocator_from_mode(device_id, alloc_mode), stream);
}
std::shared_ptr<CudaExecutor> CudaExecutor::create(
int device_id, std::shared_ptr<Executor> master,
std::shared_ptr<CudaAllocatorBase> alloc, cudaStream_t stream)
{
if (!alloc->check_environment(device_id, stream)) {
throw Error{__FILE__, __LINE__,
"Allocator uses incorrect stream or device ID."};
}
return std::shared_ptr<CudaExecutor>(new CudaExecutor(
device_id, std::move(master), std::move(alloc), stream));
}
void CudaExecutor::populate_exec_info(const machine_topology* mach_topo)
{
if (this->get_device_id() < this->get_num_devices() &&
this->get_device_id() >= 0) {
detail::cuda_scoped_device_id_guard g(this->get_device_id());
GKO_ASSERT_NO_CUDA_ERRORS(
cudaDeviceGetPCIBusId(&(this->get_exec_info().pci_bus_id.front()),
13, this->get_device_id()));
auto cuda_hwloc_obj =
mach_topo->get_pci_device(this->get_exec_info().pci_bus_id);
if (cuda_hwloc_obj) {
this->get_exec_info().numa_node = cuda_hwloc_obj->closest_numa;
this->get_exec_info().closest_pu_ids =
cuda_hwloc_obj->closest_pu_ids;
}
}
}
void OmpExecutor::raw_copy_to(const CudaExecutor* dest, size_type num_bytes,
const void* src_ptr, void* dest_ptr) const
{
if (num_bytes > 0) {
detail::cuda_scoped_device_id_guard g(dest->get_device_id());
GKO_ASSERT_NO_CUDA_ERRORS(cudaMemcpyAsync(dest_ptr, src_ptr, num_bytes,
cudaMemcpyHostToDevice,
dest->get_stream()));
dest->synchronize();
}
}
void CudaExecutor::raw_free(void* ptr) const noexcept
{
detail::cuda_scoped_device_id_guard g(this->get_device_id());
alloc_->deallocate(ptr);
}
void* CudaExecutor::raw_alloc(size_type num_bytes) const
{
detail::cuda_scoped_device_id_guard g(this->get_device_id());
return alloc_->allocate(num_bytes);
}
void CudaExecutor::raw_copy_to(const OmpExecutor*, size_type num_bytes,
const void* src_ptr, void* dest_ptr) const
{
if (num_bytes > 0) {
detail::cuda_scoped_device_id_guard g(this->get_device_id());
GKO_ASSERT_NO_CUDA_ERRORS(cudaMemcpyAsync(dest_ptr, src_ptr, num_bytes,
cudaMemcpyDeviceToHost,
this->get_stream()));
this->synchronize();
}
}
void CudaExecutor::raw_copy_to(const HipExecutor* dest, size_type num_bytes,
const void* src_ptr, void* dest_ptr) const
{
#if GINKGO_HIP_PLATFORM_NVCC == 1
if (num_bytes > 0) {
detail::cuda_scoped_device_id_guard g(this->get_device_id());
GKO_ASSERT_NO_CUDA_ERRORS(cudaMemcpyPeerAsync(
dest_ptr, dest->get_device_id(), src_ptr, this->get_device_id(),
num_bytes, this->get_stream()));
this->synchronize();
}
#else
GKO_NOT_SUPPORTED(dest);
#endif
}
void CudaExecutor::raw_copy_to(const DpcppExecutor* dest, size_type num_bytes,
const void* src_ptr, void* dest_ptr) const
{
GKO_NOT_SUPPORTED(dest);
}
void CudaExecutor::raw_copy_to(const CudaExecutor* dest, size_type num_bytes,
const void* src_ptr, void* dest_ptr) const
{
if (num_bytes > 0) {
detail::cuda_scoped_device_id_guard g(this->get_device_id());
GKO_ASSERT_NO_CUDA_ERRORS(cudaMemcpyPeerAsync(
dest_ptr, dest->get_device_id(), src_ptr, this->get_device_id(),
num_bytes, this->get_stream()));
this->synchronize();
}
}
void CudaExecutor::synchronize() const
{
detail::cuda_scoped_device_id_guard g(this->get_device_id());
GKO_ASSERT_NO_CUDA_ERRORS(cudaStreamSynchronize(this->get_stream()));
}
scoped_device_id_guard CudaExecutor::get_scoped_device_id_guard() const
{
return {this, this->get_device_id()};
}
int CudaExecutor::get_num_devices()
{
int deviceCount = 0;
auto error_code = cudaGetDeviceCount(&deviceCount);
if (error_code == cudaErrorNoDevice) {
return 0;
}
GKO_ASSERT_NO_CUDA_ERRORS(error_code);
return deviceCount;
}
void CudaExecutor::set_gpu_property()
{
if (this->get_device_id() < this->get_num_devices() &&
this->get_device_id() >= 0) {
detail::cuda_scoped_device_id_guard g(this->get_device_id());
GKO_ASSERT_NO_CUDA_ERRORS(cudaDeviceGetAttribute(
&this->get_exec_info().major, cudaDevAttrComputeCapabilityMajor,
this->get_device_id()));
GKO_ASSERT_NO_CUDA_ERRORS(cudaDeviceGetAttribute(
&this->get_exec_info().minor, cudaDevAttrComputeCapabilityMinor,
this->get_device_id()));
GKO_ASSERT_NO_CUDA_ERRORS(cudaDeviceGetAttribute(
&this->get_exec_info().num_computing_units,
cudaDevAttrMultiProcessorCount, this->get_device_id()));
auto max_threads_per_block = 0;
GKO_ASSERT_NO_CUDA_ERRORS(cudaDeviceGetAttribute(
&max_threads_per_block, cudaDevAttrMaxThreadsPerBlock,
this->get_device_id()));
std::vector<int> max_threads_per_block_dim(3, 0);
GKO_ASSERT_NO_CUDA_ERRORS(cudaDeviceGetAttribute(
&max_threads_per_block_dim[0], cudaDevAttrMaxBlockDimX,
this->get_device_id()));
GKO_ASSERT_NO_CUDA_ERRORS(cudaDeviceGetAttribute(
&max_threads_per_block_dim[1], cudaDevAttrMaxBlockDimY,
this->get_device_id()));
GKO_ASSERT_NO_CUDA_ERRORS(cudaDeviceGetAttribute(
&max_threads_per_block_dim[2], cudaDevAttrMaxBlockDimZ,
this->get_device_id()));
this->get_exec_info().max_workgroup_size = max_threads_per_block;
this->get_exec_info().max_workitem_sizes = max_threads_per_block_dim;
this->get_exec_info().num_pu_per_cu =
convert_sm_ver_to_cores(this->get_exec_info().major,
this->get_exec_info().minor) /
kernels::cuda::config::warp_size;
this->get_exec_info().max_subgroup_size =
kernels::cuda::config::warp_size;
}
}
void CudaExecutor::init_handles()
{
if (this->get_device_id() < this->get_num_devices() &&
this->get_device_id() >= 0) {
const auto id = this->get_device_id();
detail::cuda_scoped_device_id_guard g(id);
this->cublas_handle_ = handle_manager<cublasContext>(
kernels::cuda::cublas::init(this->get_stream()),
[id](cublasHandle_t handle) {
detail::cuda_scoped_device_id_guard g(id);
kernels::cuda::cublas::destroy(handle);
});
this->cusparse_handle_ = handle_manager<cusparseContext>(
kernels::cuda::cusparse::init(this->get_stream()),
[id](cusparseHandle_t handle) {
detail::cuda_scoped_device_id_guard g(id);
kernels::cuda::cusparse::destroy(handle);
});
}
}
} // namespace gko