1 /* Copyright 2018 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 #include "tensorflow/core/kernels/host_constant_op.h"
17
18 #include "tensorflow/core/framework/allocator.h"
19 #include "tensorflow/core/framework/op_kernel.h"
20 #include "tensorflow/core/framework/types.h"
21 #include "tensorflow/core/lib/core/status.h"
22 #include "tensorflow/core/platform/logging.h"
23 #include "tensorflow/core/platform/macros.h"
24
25 namespace tensorflow {
26
_HostConstantOp(OpKernelConstruction * ctx)27 _HostConstantOp::_HostConstantOp(OpKernelConstruction* ctx)
28 : OpKernel(ctx), tensor_(ctx->output_type(0)) {
29 const TensorProto* proto = nullptr;
30 AllocatorAttributes alloc_attr;
31 alloc_attr.set_on_host(true);
32 OP_REQUIRES_OK(ctx, ctx->GetAttr("value", &proto));
33 OP_REQUIRES_OK(
34 ctx, ctx->device()->MakeTensorFromProto(*proto, alloc_attr, &tensor_));
35 OP_REQUIRES(
36 ctx, ctx->output_type(0) == tensor_.dtype(),
37 errors::InvalidArgument("Type mismatch between value (",
38 DataTypeString(tensor_.dtype()), ") and dtype (",
39 DataTypeString(ctx->output_type(0)), ")"));
40 }
41
Compute(OpKernelContext * ctx)42 void _HostConstantOp::Compute(OpKernelContext* ctx) {
43 ctx->set_output(0, tensor_);
44 }
45
46 #if GOOGLE_CUDA
47 // A special GPU kernel for int32.
48 // TODO(b/25387198): Also enable int32 in device memory. This kernel
49 // registration requires all int32 inputs and outputs to be in host memory.
50 REGISTER_KERNEL_BUILDER(Name("Const")
51 .Device(DEVICE_GPU)
52 .HostMemory("output")
53 .TypeConstraint<int32>("dtype"),
54 _HostConstantOp);
55 #endif
56
57 #ifdef TENSORFLOW_USE_SYCL
58 REGISTER_KERNEL_BUILDER(Name("Const")
59 .Device(DEVICE_SYCL)
60 .HostMemory("output")
61 .TypeConstraint<int32>("dtype"),
62 _HostConstantOp);
63 #endif // TENSORFLOW_USE_SYCL
64
65 // HostConst: forced to generate output on the host.
66 REGISTER_KERNEL_BUILDER(Name("HostConst").Device(DEVICE_CPU), _HostConstantOp);
67 REGISTER_KERNEL_BUILDER(
68 Name("HostConst").Device(DEVICE_GPU).HostMemory("output"), _HostConstantOp);
69 #ifdef TENSORFLOW_USE_SYCL
70 REGISTER_KERNEL_BUILDER(
71 Name("HostConst").Device(DEVICE_SYCL).HostMemory("output"),
72 _HostConstantOp);
73 #endif // TENSORFLOW_USE_SYCL
74
75 } // end namespace tensorflow
76
77