1 /* Copyright 2015 The TensorFlow Authors. All Rights Reserved.
2
3 Licensed under the Apache License, Version 2.0 (the "License");
4 you may not use this file except in compliance with the License.
5 You may obtain a copy of the License at
6
7 http://www.apache.org/licenses/LICENSE-2.0
8
9 Unless required by applicable law or agreed to in writing, software
10 distributed under the License is distributed on an "AS IS" BASIS,
11 WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12 See the License for the specific language governing permissions and
13 limitations under the License.
14 ==============================================================================*/
15
16 // Implementation of the pointer-to-implementation wrapper for the data-parallel
17 // kernel abstraction. KernelBase just delegates to the internal
18 // platform-specific implementation instance.
19
20 #include "tensorflow/stream_executor/kernel.h"
21
22 #include "tensorflow/stream_executor/platform/port.h"
23
24 #include "tensorflow/stream_executor/lib/demangle.h"
25 #include "tensorflow/stream_executor/platform.h"
26 #include "tensorflow/stream_executor/platform/logging.h"
27 #include "tensorflow/stream_executor/stream_executor.h"
28
29 namespace perftools {
30 namespace gputools {
31
registers_per_thread(int * registers_per_thread) const32 bool KernelMetadata::registers_per_thread(int *registers_per_thread) const {
33 if (has_registers_per_thread_) {
34 *registers_per_thread = registers_per_thread_;
35 return true;
36 }
37
38 return false;
39 }
40
set_registers_per_thread(int registers_per_thread)41 void KernelMetadata::set_registers_per_thread(int registers_per_thread) {
42 registers_per_thread_ = registers_per_thread;
43 has_registers_per_thread_ = true;
44 }
45
shared_memory_bytes(int * shared_memory_bytes) const46 bool KernelMetadata::shared_memory_bytes(int *shared_memory_bytes) const {
47 if (has_shared_memory_bytes_) {
48 *shared_memory_bytes = shared_memory_bytes_;
49 return true;
50 }
51
52 return false;
53 }
54
set_shared_memory_bytes(int shared_memory_bytes)55 void KernelMetadata::set_shared_memory_bytes(int shared_memory_bytes) {
56 shared_memory_bytes_ = shared_memory_bytes;
57 has_shared_memory_bytes_ = true;
58 }
59
KernelBase(KernelBase && from)60 KernelBase::KernelBase(KernelBase &&from)
61 : parent_(from.parent_),
62 implementation_(std::move(from.implementation_)),
63 name_(std::move(from.name_)),
64 demangled_name_(std::move(from.demangled_name_)),
65 metadata_(from.metadata_) {
66 from.parent_ = nullptr;
67 }
68
KernelBase(StreamExecutor * parent)69 KernelBase::KernelBase(StreamExecutor *parent)
70 : parent_(parent),
71 implementation_(parent->implementation()->CreateKernelImplementation()) {}
72
KernelBase(StreamExecutor * parent,internal::KernelInterface * implementation)73 KernelBase::KernelBase(StreamExecutor *parent,
74 internal::KernelInterface *implementation)
75 : parent_(parent), implementation_(implementation) {}
76
~KernelBase()77 KernelBase::~KernelBase() {
78 if (parent_) {
79 parent_->UnloadKernel(this);
80 }
81 }
82
Arity() const83 unsigned KernelBase::Arity() const { return implementation_->Arity(); }
84
SetPreferredCacheConfig(KernelCacheConfig config)85 void KernelBase::SetPreferredCacheConfig(KernelCacheConfig config) {
86 return implementation_->SetPreferredCacheConfig(config);
87 }
88
GetPreferredCacheConfig() const89 KernelCacheConfig KernelBase::GetPreferredCacheConfig() const {
90 return implementation_->GetPreferredCacheConfig();
91 }
92
93 // Prefix stub functions emitted by the CUDA splitter.
94 static const char *kStubPrefix = "__device_stub_";
95
set_name(port::StringPiece name)96 void KernelBase::set_name(port::StringPiece name) {
97 name_ = name.ToString();
98 port::StringPiece stubless_name = name;
99 if (name.starts_with(kStubPrefix)) {
100 stubless_name.remove_prefix(strlen(kStubPrefix));
101 }
102 demangled_name_ = port::Demangle(stubless_name.data());
103 }
104
105 } // namespace gputools
106 } // namespace perftools
107