• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* Copyright 2019 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_COMPILER_XLA_SERVICE_GPU_CUSOLVER_CONTEXT_H_
17 #define TENSORFLOW_COMPILER_XLA_SERVICE_GPU_CUSOLVER_CONTEXT_H_
18 
19 #include <complex>
20 
21 #if !TENSORFLOW_USE_ROCM
22 #include "third_party/gpus/cuda/include/cusolverDn.h"
23 #endif
24 
25 #include "tensorflow/compiler/xla/statusor.h"
26 #include "tensorflow/compiler/xla/types.h"
27 #include "tensorflow/compiler/xla/util.h"
28 #include "tensorflow/core/lib/core/status.h"
29 #include "tensorflow/core/platform/stream_executor_no_cuda.h"
30 #include "tensorflow/stream_executor/blas.h"
31 
32 namespace xla {
33 namespace gpu {
34 
35 #if !TENSORFLOW_USE_ROCM
36 
37 class CusolverContext {
38  public:
39   // stream may be nullptr, in which case the context can only be used for
40   // buffer size queries.
41   static StatusOr<CusolverContext> Create(se::Stream* stream);
42   CusolverContext() = default;
43   ~CusolverContext();
44 
45   CusolverContext(const CusolverContext&) = delete;
46   CusolverContext(CusolverContext&&);
47   CusolverContext& operator=(const CusolverContext&) = delete;
48   CusolverContext& operator=(CusolverContext&&);
49 
50   // Computes the Cholesky factorization A = L * L^T for a single matrix.
51   // Returns Status::OK() if the kernel was launched successfully. See:
52   // http://docs.nvidia.com/cuda/cusolver/#cuds-lt-t-gt-potrf
53   template <typename T, typename = std::enable_if_t<
54                             std::is_same<T, float>::value ||
55                             std::is_same<T, double>::value ||
56                             std::is_same<T, std::complex<float>>::value ||
57                             std::is_same<T, std::complex<double>>::value>>
58   Status Potrf(se::blas::UpperLower uplo, int n, se::DeviceMemory<T> dev_A,
59                int lda, se::DeviceMemory<int> dev_lapack_info,
60                se::DeviceMemory<T> workspace) = delete;
61 
62   // Returns the size of the `workspace` required by Potrf, in number of
63   // elements of `type`.
64   StatusOr<int64> PotrfBufferSize(PrimitiveType type, se::blas::UpperLower uplo,
65                                   int n, int lda);
66 
67  private:
68   CusolverContext(se::Stream* stream, cusolverDnHandle_t handle);
69 
handle()70   cusolverDnHandle_t handle() const { return handle_; }
71 
72   se::Stream* stream_ = nullptr;
73   cusolverDnHandle_t handle_ = nullptr;
74 };
75 
76 #define CALL_LAPACK_TYPES(m) \
77   m(float, S) m(double, D) m(std::complex<float>, C) m(std::complex<double>, Z)
78 #define POTRF_INSTANCE(T, type_prefix)                                  \
79   template <>                                                           \
80   Status CusolverContext::Potrf<T>(                                     \
81       se::blas::UpperLower uplo, int n, se::DeviceMemory<T> A, int lda, \
82       se::DeviceMemory<int> lapack_info, se::DeviceMemory<T> workspace);
83 CALL_LAPACK_TYPES(POTRF_INSTANCE);
84 #undef POTRF_INSTANCE
85 #undef CALL_LAPACK_TYPES
86 
87 #else
88 
89 typedef void* cusolverDnHandle_t;
90 
91 // TODO(cheshire): Remove this hack once we have ROCM implementation.
92 class CusolverContext {
93  public:
94   static StatusOr<CusolverContext> Create(se::Stream* stream) {
95     LOG(FATAL) << "Unimplemented";
96   }
97 
98   template <typename T, typename = std::enable_if_t<
99                             std::is_same<T, float>::value ||
100                             std::is_same<T, double>::value ||
101                             std::is_same<T, std::complex<float>>::value ||
102                             std::is_same<T, std::complex<double>>::value>>
103   Status Potrf(se::blas::UpperLower uplo, int n, se::DeviceMemory<T> dev_A,
104                int lda, se::DeviceMemory<int> dev_lapack_info,
105                se::DeviceMemory<T> workspace) {
106     LOG(FATAL) << "Unimplemented";
107   }
108 
109   StatusOr<int64> PotrfBufferSize(PrimitiveType type, se::blas::UpperLower uplo,
110                                   int n, int lda) {
111     LOG(FATAL) << "Unimplemented";
112   }
113 };
114 
115 #endif
116 
117 }  // namespace gpu
118 }  // namespace xla
119 
120 #endif  // TENSORFLOW_COMPILER_XLA_SERVICE_GPU_CUSOLVER_CONTEXT_H_
121