NNTR_THROW_IF(search == key2index.end(), std::invalid_argument)
<< FUNC_TAG << "Cannot find index for key: " << key;
- return *search;
+ return search->second;
}
/**
};
/**
- * @brief Tfbuffer struct (not yet fully specified)
- *
- */
-struct TfBuffer {
- size_t size;
-};
-
-/**
* @brief tensorflow operation index map, this class manages operation index
* mapping
*
auto &buffer_map = getIndexMap<const float *, Buffer>();
buffer_map.addDataWhenNotFound(
- empty_buffer, {0, empty_buffer}); /// put empty buffer to first
+ nullptr, {0, empty_buffer}); // this represents undefined buffer
+ buffer_map.addDataWhenNotFound(
+ empty_buffer, {0, empty_buffer}); /// this represents empty buffer
auto update_buffers = [&buffer_map](const TfOpNode::Variables &variables) {
for (auto &variable : variables) {
buildOpNodes(std::shared_ptr<const GraphRepresentation> representation) {
TfOpNodes nodes;
/// @todo, look ahead of layers to get nodes that can be fused
+ /// we will need to have a dedicated builder
for (const auto &ln : representation->getSorted()) {
nodes.emplace_back(*ln->getObject());
}
map.getIndexMap<const float *, TfOpIdxMap::Buffer>().getData();
std::vector<flatbuffers::Offset<tflite::Buffer>> fb_buffers;
- fb_buffers.reserve(buffers.size() + 1);
- /// add reserved buffer to denote undefined
- fb_buffers.push_back(tflite::CreateBuffer(fbb));
+ fb_buffers.reserve(buffers.size());
auto create_buffer_offset = [&fbb](const TfOpIdxMap::Buffer &buffer) {
if (buffer.first == 0) {
return fbb.CreateVector(fb_op_codes);
}
+/**
+ * @brief get Serialized Tensor shape
+ * @note when building tensorshape, the tensor shape must be final, when we
+ * reach here it, the tensordim should be final which means if transpose is
+ * needed, it should be done already.
+ * @param dim Tensor dimension to convert
+ * @param type tflite Tensor type
+ * @param effective_dimension bitflag of valid 'nchw' dimension, ex) 0b1011
+ * means,nhw is effective
+ * @param is_signature check if the given dimension is regarded as a tensor
+ * signature
+ * @return flatbuffers::Offset<flatbuffers::Vector<int32_t>>
+ */
+flatbuffers::Offset<flatbuffers::Vector<int32_t>>
+buildTensorShape(const TensorDim &dim, flatbuffers::FlatBufferBuilder &fbb,
+ int effective_dimension, bool is_signature = false) {
+ std::vector<int32_t> fb_dimension;
+
+ /// batch dimension is always considered with signature
+ if ((effective_dimension >> 3) & 1) {
+ fb_dimension.push_back(is_signature ? -1 : dim.batch());
+ }
+
+ for (unsigned int i = 1; i < MAXDIM; ++i) {
+ if (((effective_dimension >> (MAXDIM - i - 1)) & 1) == 0) {
+ continue;
+ }
+ fb_dimension.push_back(dim.getTensorDim(i));
+ }
+
+ return fbb.CreateVector(fb_dimension);
+}
+
+flatbuffers::Offset<flatbuffers::Vector<flatbuffers::Offset<tflite::Tensor>>>
+buildTensors(const TfOpIdxMap &map, flatbuffers::FlatBufferBuilder &fbb) {
+ /// @todo: the actual (suqeezed) tensor dimension must be known before coming
+ /// here. For now, it is directly guessed for the fc layer
+ const auto &variables = map.getIndexMap<const Var_Grad *>().getData();
+ const auto &buffer_map = map.getIndexMap<const float *, TfOpIdxMap::Buffer>();
+
+ std::vector<flatbuffers::Offset<tflite::Tensor>> fb_tensors;
+ fb_tensors.reserve(variables.size());
+
+ auto create_tensor = [&fbb, &buffer_map](const Var_Grad *var) {
+ /// @fixme: There is a fixed assumption that effective_dimension is 0b1001,
+ /// this is not true, this should be supported inherently from TensorDim
+ /// class.
+ int FIX_ME_FLAG = 0b1001;
+ auto shape = buildTensorShape(var->getDim(), fbb, FIX_ME_FLAG, false);
+
+ /// in *.tflite spec, there is tensor dimension and tensor signature
+ /// dimension the only difference is that tensor signature dimension gets
+ /// the value of -1 which denotes the size is unspecified. Mostly, batch
+ /// dimension has the signature of -1, we don't have such concept yet so
+ /// this is handled by boolean for now
+ auto shape_sig = buildTensorShape(var->getDim(), fbb, FIX_ME_FLAG, true);
+ auto name = fbb.CreateString(var->getName());
+ auto tensor = var->getVariableRef();
+
+ unsigned int buffer_idx = 1;
+ if (!tensor.uninitialized() && tensor.isAllocated()) {
+ buffer_idx = buffer_map.getIndex(var->getVariableRef().getData());
+ }
+
+ tflite::TensorBuilder builder(fbb);
+ builder.add_name(name);
+ builder.add_buffer(buffer_idx);
+ builder.add_shape(shape);
+ builder.add_shape_signature(shape_sig);
+ return builder.Finish();
+ };
+
+ std::transform(variables.begin(), variables.end(),
+ std::back_inserter(fb_tensors), create_tensor);
+
+ return fbb.CreateVector(fb_tensors);
+}
+
flatbuffers::Offset<flatbuffers::Vector<flatbuffers::Offset<tflite::SubGraph>>>
-buildSubGraph(const TfOpNodes &nodes, const TfOpIdxMap &map,
- flatbuffers::FlatBufferBuilder &fbb) {
- /** NYI! */
- return flatbuffers::Offset<
- flatbuffers::Vector<flatbuffers::Offset<tflite::SubGraph>>>();
+buildSubGraphs(const TfOpNodes &nodes, const TfOpIdxMap &map,
+ flatbuffers::FlatBufferBuilder &fbb) {
+
+ auto tensors = buildTensors(map, fbb);
+
+ /// @todo extract this to buildSubgraph if there is one or more subgraph
+ auto name = fbb.CreateString("main");
+
+ auto builder = tflite::SubGraphBuilder(fbb);
+ builder.add_tensors(tensors);
+ builder.add_name(name);
+ auto subgraph = builder.Finish();
+
+ std::vector<flatbuffers::Offset<tflite::SubGraph>> subgraphs;
+ subgraphs.reserve(1);
+ subgraphs.push_back(subgraph);
+
+ return fbb.CreateVector(subgraphs);
}
} // namespace
TfOpIdxMap map(opNodes); /// build TfOpIdxMap from opNodes
auto opcodes = buildOperatorCodes(map, fbb);
- auto UNUSED(subgraph) = buildSubGraph(opNodes, map, fbb);
+ auto subgraphs = buildSubGraphs(opNodes, map, fbb);
auto buffers = buildBuffers(map, fbb);
auto desc = fbb.CreateString("This file is generated from NNTrainer");
tflite::ModelBuilder model_builder(fbb);
model_builder.add_operator_codes(opcodes);
- WILL(model_builder.add_subgraphs(subgraph));
+ model_builder.add_subgraphs(subgraphs);
model_builder.add_buffers(buffers);
model_builder.add_version(3);
model_builder.add_description(desc);
projects / platform / core / ml / nntrainer.git / commitdiff