• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* Copyright 2020 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 #ifndef TENSORFLOW_LITE_DELEGATES_GPU_CL_CL_OPERATION_H_
17 #define TENSORFLOW_LITE_DELEGATES_GPU_CL_CL_OPERATION_H_
18 
19 #include <cstdint>
20 #include <string>
21 #include <utility>
22 #include <vector>
23 
24 #include "tensorflow/lite/delegates/gpu/cl/cl_arguments.h"
25 #include "tensorflow/lite/delegates/gpu/cl/cl_command_queue.h"
26 #include "tensorflow/lite/delegates/gpu/cl/cl_context.h"
27 #include "tensorflow/lite/delegates/gpu/cl/cl_device.h"
28 #include "tensorflow/lite/delegates/gpu/cl/cl_kernel.h"
29 #include "tensorflow/lite/delegates/gpu/cl/program_cache.h"
30 #include "tensorflow/lite/delegates/gpu/cl/tensor.h"
31 #include "tensorflow/lite/delegates/gpu/common/task/gpu_operation.h"
32 
33 namespace tflite {
34 namespace gpu {
35 namespace cl {
36 
37 struct CreationContext {
38   const CLDevice* device;
39   CLContext* context;
40   CLCommandQueue* queue;
41   ProgramCache* cache;
42 
GetGpuInfoCreationContext43   const GpuInfo& GetGpuInfo() const { return device->info_; }
44 };
45 
46 class ClOperation {
47  public:
48   ClOperation() = default;
49   virtual ~ClOperation() = default;
50   // Move only
51   ClOperation(ClOperation&& operation) = default;
52   ClOperation& operator=(ClOperation&& operation) = default;
53   ClOperation(const ClOperation&) = delete;
54   ClOperation& operator=(const ClOperation&) = delete;
55 
Init(std::unique_ptr<GPUOperation> && gpu_operation)56   void Init(std::unique_ptr<GPUOperation>&& gpu_operation) {
57     operation_ = std::move(gpu_operation);
58   }
59 
GetGpuOperation()60   GPUOperation& GetGpuOperation() { return *operation_; }
GetGpuOperation()61   const GPUOperation& GetGpuOperation() const { return *operation_; }
GetKernelFingerprint()62   uint64_t GetKernelFingerprint() const { return kernel_fingerprint_; }
63 
GetDefinition()64   const OperationDef& GetDefinition() const { return operation_->definition_; }
65 
66   absl::Status AddOperation(ClOperation* operation);
67 
68   // should be called after changes of inputs/outputs.
69   absl::Status UpdateParams();
70 
AddToQueue(CLCommandQueue * queue)71   absl::Status AddToQueue(CLCommandQueue* queue) {
72     RETURN_IF_ERROR(cl_args_.Bind(kernel_.kernel()));
73     return queue->Dispatch(kernel_, operation_->work_groups_count_,
74                            operation_->work_group_size_);
75   }
76 
77   absl::Status Tune(TuningType tuning_type, const GpuInfo& gpu_info,
78                     ProfilingCommandQueue* profiling_queue);
79 
80   absl::Status Compile(const CreationContext& creation_context);
81 
82   absl::Status RestoreDeserialized(const CreationContext& creation_context);
83   absl::Status InitFromCache(uint64_t fingerprint,
84                              const ProgramCache& program_cache);
85 
MoveObjectRefsFromCLToGeneric()86   void MoveObjectRefsFromCLToGeneric() {
87     cl_args_.MoveObjectRefsOut(&operation_->args_);
88   }
MoveObjectRefsFromGenericToCL()89   void MoveObjectRefsFromGenericToCL() {
90     cl_args_.MoveObjectRefsIn(&operation_->args_);
91   }
SyncScalarValues()92   void SyncScalarValues() { cl_args_.CopyScalarValues(&operation_->args_); }
93 
94  private:
95   std::unique_ptr<GPUOperation> operation_;
96   CLKernel kernel_;
97   uint64_t kernel_fingerprint_;
98   CLArguments cl_args_;
99 };
100 
101 }  // namespace cl
102 }  // namespace gpu
103 }  // namespace tflite
104 
105 #endif  // TENSORFLOW_LITE_DELEGATES_GPU_CL_CL_OPERATION_H_
106