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
|
/* Copyright 2017 The TensorFlow Authors. All Rights Reserved.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
==============================================================================*/
// Legacy flags for XLA's gpu_compiler module.
#include <mutex> // NOLINT(build/c++11): only using std::call_once, not mutex.
#include <vector>
#include "tensorflow/compiler/xla/legacy_flags/gpu_compiler_flags.h"
#include "tensorflow/compiler/xla/legacy_flags/parse_flags_from_env.h"
#include "tensorflow/compiler/xla/types.h"
#include "tensorflow/core/platform/types.h"
#include "tensorflow/core/util/command_line_flags.h"
namespace xla {
namespace legacy_flags {
// Pointers to the parsed value of the flags and flag descriptors, initialized
// via flags_init.
static GpuCompilerFlags* flags;
static std::vector<tensorflow::Flag>* flag_list;
static std::once_flag flags_init;
// Allocate *flags. Called via call_once(&flags_init,...).
static void AllocateFlags() {
flags = new GpuCompilerFlags;
flags->xla_gpu_embed_ir = false;
flags->xla_cuda_data_dir = "./cuda_sdk_lib";
flags->xla_ptxas_path = "/usr/local/cuda/bin/ptxas";
flag_list = new std::vector<tensorflow::Flag>({
tensorflow::Flag(
"xla_gpu_embed_ir", &flags->xla_gpu_embed_ir,
"Embed the LLVM IR module string in the resultant GpuExecutable."),
tensorflow::Flag(
"xla_cuda_data_dir", &flags->xla_cuda_data_dir,
"If non-empty, specifies a local directory containing ptxas and "
"nvvm libdevice files. Otherwise, by default, we use those from "
"runfile directories."),
tensorflow::Flag("xla_ptxas_path", &flags->xla_ptxas_path,
"The path to ptxas. Required to log stats of the ptx."),
});
ParseFlagsFromEnv(*flag_list);
}
// Append to *append_to flag definitions associated with XLA's gpu_compiler
// module.
void AppendGpuCompilerFlags(std::vector<tensorflow::Flag>* append_to) {
std::call_once(flags_init, &AllocateFlags);
append_to->insert(append_to->end(), flag_list->begin(), flag_list->end());
}
// Return a pointer to the GpuCompilerFlags struct;
// repeated calls return the same pointer.
// This should be called only after Flags::Parse() has returned.
GpuCompilerFlags* GetGpuCompilerFlags() {
std::call_once(flags_init, &AllocateFlags);
return flags;
}
} // namespace legacy_flags
} // namespace xla
|
