• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 #include "tensorflow/core/framework/memory_types.h"
17 
18 #include <utility>
19 
20 #include "tensorflow/compiler/jit/defs.h"
21 #include "tensorflow/core/framework/attr_value.pb.h"
22 #include "tensorflow/core/framework/kernel_def.pb.h"
23 #include "tensorflow/core/framework/node_def.pb.h"
24 #include "tensorflow/core/framework/node_def_util.h"
25 #include "tensorflow/core/framework/op_kernel.h"
26 #include "tensorflow/core/framework/types.h"
27 #include "tensorflow/core/lib/core/errors.h"
28 #include "tensorflow/core/platform/types.h"
29 
30 namespace tensorflow {
31 
32 namespace {
33 // Returns the largest endpoint of anything in the name_map.
GetTotal(const NameRangeMap & name_map)34 int GetTotal(const NameRangeMap& name_map) {
35   int total = 0;
36   for (const auto& item : name_map) {
37     total = std::max(total, item.second.second);
38   }
39   return total;
40 }
41 
42 // Fills memory_types for either input or output, setting everything
43 // to DEVICE_MEMORY except those args in host_memory_args.  Removes
44 // elements of host_memory_args that were used.
MemoryTypesHelper(const NameRangeMap & name_map,std::vector<string> * host_memory_args,MemoryTypeVector * memory_types)45 void MemoryTypesHelper(const NameRangeMap& name_map,
46                        std::vector<string>* host_memory_args,
47                        MemoryTypeVector* memory_types) {
48   // Update args that have been marked as in "HOST_MEMORY".
49   size_t keep = 0;
50   for (size_t i = 0; i < host_memory_args->size(); ++i) {
51     auto iter = name_map.find((*host_memory_args)[i]);
52     if (iter != name_map.end()) {
53       for (int j = iter->second.first; j < iter->second.second; ++j) {
54         (*memory_types)[j] = HOST_MEMORY;
55       }
56     } else {
57       // (*host_memory_args)[i] not found, save it for the next pass.
58       if (i > keep) (*host_memory_args)[keep] = (*host_memory_args)[i];
59       ++keep;
60     }
61   }
62   host_memory_args->resize(keep);
63 }
64 
IsFunctionCallOp(const string & op_type)65 bool IsFunctionCallOp(const string& op_type) {
66   return op_type == "SymbolicGradient" || op_type == "PartitionedCall" ||
67          op_type == "StatefulPartitionedCall" || op_type == "While" ||
68          op_type == "StatelessWhile";
69 }
70 
71 }  // namespace
72 
MTypeFromDType(const DataType dtype)73 MemoryType MTypeFromDType(const DataType dtype) {
74   return (dtype == DT_INT32 || DataTypeAlwaysOnHost(dtype)) ? HOST_MEMORY
75                                                             : DEVICE_MEMORY;
76 }
77 
MTypeFromDTypeIntsOnDevice(const DataType dtype)78 MemoryType MTypeFromDTypeIntsOnDevice(const DataType dtype) {
79   return DataTypeAlwaysOnHost(dtype) ? HOST_MEMORY : DEVICE_MEMORY;
80 }
81 
MemoryTypesForNode(const OpRegistryInterface * op_registry,const DeviceType & device_type,const NodeDef & ndef,MemoryTypeVector * inp_mtypes,MemoryTypeVector * out_mtypes)82 Status MemoryTypesForNode(const OpRegistryInterface* op_registry,
83                           const DeviceType& device_type, const NodeDef& ndef,
84                           MemoryTypeVector* inp_mtypes,
85                           MemoryTypeVector* out_mtypes) {
86   // Look up the Op registered for this op name.
87   const OpDef* op_def;
88   TF_RETURN_IF_ERROR(op_registry->LookUpOpDef(ndef.op(), &op_def));
89 
90   // Look up the Kernel registered for this node def.
91   const KernelDef* kdef = nullptr;
92   Status status =
93       FindKernelDef(device_type, ndef, &kdef, nullptr /* kernel_class_name */);
94 
95   DataTypeVector inp_dtypes;
96   DataTypeVector out_dtypes;
97   TF_RETURN_IF_ERROR(
98       InOutTypesForNode(ndef, *op_def, &inp_dtypes, &out_dtypes));
99 
100   inp_mtypes->clear();
101   out_mtypes->clear();
102 
103   bool has_xla_compile = [&] {
104     const auto& it = ndef.attr().find(kXlaMustCompileAttr);
105     return it != ndef.attr().end() && it->second.b();
106   }();
107 
108   bool has_kernel_def = status.ok() && !IsFunctionCallOp(ndef.op());
109   auto host_memory_required = [&](const DataType& dt) {
110     bool int32_on_device =
111         has_kernel_def || device_type.type_string() == "TPU" || has_xla_compile;
112     return DataTypeAlwaysOnHost(dt) || (dt == DT_INT32 && !int32_on_device);
113   };
114 
115   if (has_kernel_def) {
116     // Gets the input/output names and their corresponding endpoint ranges.
117     NameRangeMap inp_names;
118     NameRangeMap out_names;
119     TF_RETURN_IF_ERROR(
120         NameRangesForNode(ndef, *op_def, &inp_names, &out_names));
121 
122     // Now that we know the size, fill with the default 'DEVICE_MEMORY'.
123     inp_mtypes->resize(GetTotal(inp_names), DEVICE_MEMORY);
124     out_mtypes->resize(GetTotal(out_names), DEVICE_MEMORY);
125 
126     // Fills in host memory types based on the kernel def.
127     const auto& from_proto = kdef->host_memory_arg();
128     std::vector<string> host_memory_args(from_proto.begin(), from_proto.end());
129     MemoryTypesHelper(inp_names, &host_memory_args, inp_mtypes);
130     MemoryTypesHelper(out_names, &host_memory_args, out_mtypes);
131     if (!host_memory_args.empty()) {
132       return errors::InvalidArgument(
133           "HostMemory args '", absl::StrJoin(host_memory_args, "', '"),
134           "' not found in OpDef: ", SummarizeOpDef(*op_def));
135     }
136   } else {
137     // Set all the datatype to DEVICE_MEMORY by default, later on change it to
138     // HOST_MEMORY where it is required by the datatype.
139     inp_mtypes->resize(inp_dtypes.size(), DEVICE_MEMORY);
140     out_mtypes->resize(out_dtypes.size(), DEVICE_MEMORY);
141   }
142   CHECK_LE(inp_mtypes->size(), inp_dtypes.size());
143   CHECK_LE(out_mtypes->size(), out_dtypes.size());
144 
145   // Mark e.g. all resource and string types as host memory.
146   for (int i = 0; i < inp_mtypes->size(); ++i) {
147     if (host_memory_required(inp_dtypes[i])) {
148       (*inp_mtypes)[i] = HOST_MEMORY;
149     }
150   }
151   for (int i = 0; i < out_mtypes->size(); ++i) {
152     if (host_memory_required(out_dtypes[i])) {
153       (*out_mtypes)[i] = HOST_MEMORY;
154     }
155   }
156 
157   std::vector<int32> hostmem_attr;
158   if (TryGetNodeAttr(ndef, "_input_hostmem", &hostmem_attr)) {
159     for (int32 i : hostmem_attr) {
160       if (0 <= i && i < inp_mtypes->size()) {
161         (*inp_mtypes)[i] = HOST_MEMORY;
162       }
163     }
164   }
165   hostmem_attr.clear();
166   if (TryGetNodeAttr(ndef, "_output_hostmem", &hostmem_attr)) {
167     for (int32 i : hostmem_attr) {
168       if (0 <= i && i < out_mtypes->size()) {
169         (*out_mtypes)[i] = HOST_MEMORY;
170       }
171     }
172   }
173 
174   return Status::OK();
175 }
176 
177 }  // namespace tensorflow
178