experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Use GetVariableInput (or GetInput in const context) for the activation tensor in SVDF, instead of accessing it through its index.

Browse Source 
PiperOrigin-RevId: 294748701
Change-Id: I6208f0af2b2db8d294fef559eb650b9c66cc9b6d
This commit is contained in:
Robert David 2020-02-12 13:53:18 -08:00 committed by TensorFlower Gardener
parent b9b688ebd2
commit 24fa7c6822
3 changed files with 9 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
tensorflow/lite/kernels/svdf.cc
View File

@ -39,7 +39,6 @@ namespace {
struct OpData {
int scratch_tensor_index;
bool float_weights_time_initialized;
int activation_state_tensor_index;
int32 effective_scale_1_a;
int effective_scale_1_b;
int32 effective_scale_2_a;
@ -80,8 +79,6 @@ TfLiteStatus Prepare(TfLiteContext* context, TfLiteNode* node) {
// Check we have all the inputs and outputs we need.
TF_LITE_ENSURE_EQ(context, node->outputs->size, 1);
TF_LITE_ENSURE_EQ(context, node->inputs->size, 5);
op_data->activation_state_tensor_index =
node->inputs->data[kInputActivationStateTensor];
const TfLiteTensor* input = GetInput(context, node, kInputTensor);
const TfLiteTensor* weights_feature =
@ -109,8 +106,8 @@ TfLiteStatus Prepare(TfLiteContext* context, TfLiteNode* node) {
TF_LITE_ENSURE_EQ(context, bias->dims->data[0], num_units);
}
TfLiteTensor* activation_state =
&context->tensors[op_data->activation_state_tensor_index];
const TfLiteTensor* activation_state =
GetInput(context, node, kInputActivationStateTensor);
TfLiteTensor* output = GetOutput(context, node, kOutputTensor);
// Check the shape of input state tensors.
@ -250,7 +247,7 @@ TfLiteStatus Eval(TfLiteContext* context, TfLiteNode* node) {
TfLiteTensor* scratch = GetTemporary(context, node, /*index=*/0);
TfLiteTensor* activation_state =
&context->tensors[op_data->activation_state_tensor_index];
GetVariableInput(context, node, kInputActivationStateTensor);
TfLiteTensor* output = GetOutput(context, node, kOutputTensor);
switch (weights_feature->type) {

6
tensorflow/lite/micro/kernels/svdf.cc
View File

@ -377,8 +377,8 @@ TfLiteStatus Prepare(TfLiteContext* context, TfLiteNode* node) {
const TfLiteTensor* weights_time =
GetInput(context, node, kWeightsTimeTensor);
const TfLiteTensor* bias = GetOptionalInputTensor(context, node, kBiasTensor);
TfLiteTensor* activation_state =
&context->tensors[node->inputs->data[kInputActivationStateTensor]];
const TfLiteTensor* activation_state =
GetInput(context, node, kInputActivationStateTensor);
// Define input constants based on input tensor definition above:
const int rank = params->rank;
@ -491,7 +491,7 @@ TfLiteStatus Eval(TfLiteContext* context, TfLiteNode* node) {
GetInput(context, node, kWeightsTimeTensor);
const TfLiteTensor* bias = GetOptionalInputTensor(context, node, kBiasTensor);
TfLiteTensor* activation_state =
&context->tensors[node->inputs->data[kInputActivationStateTensor]];
GetVariableInput(context, node, kInputActivationStateTensor);
TfLiteTensor* output = GetOutput(context, node, kOutputTensor);
const bool is_full_integer = input->type == kTfLiteInt8;

6
tensorflow/lite/micro/kernels/xtensa-hifimini/svdf.cc
View File

@ -296,8 +296,8 @@ TfLiteStatus Prepare(TfLiteContext* context, TfLiteNode* node) {
const TfLiteTensor* weights_time =
GetInput(context, node, kWeightsTimeTensor);
const TfLiteTensor* bias = GetOptionalInputTensor(context, node, kBiasTensor);
TfLiteTensor* activation_state =
&context->tensors[node->inputs->data[kInputActivationStateTensor]];
const TfLiteTensor* activation_state =
GetInput(context, node, kInputActivationStateTensor);
// Define input constants based on input tensor definition above:
const int rank = params->rank;
@ -409,7 +409,7 @@ TfLiteStatus Eval(TfLiteContext* context, TfLiteNode* node) {
GetInput(context, node, kWeightsTimeTensor);
const TfLiteTensor* bias = GetOptionalInputTensor(context, node, kBiasTensor);
TfLiteTensor* activation_state =
&context->tensors[node->inputs->data[kInputActivationStateTensor]];
GetVariableInput(context, node, kInputActivationStateTensor);
TfLiteTensor* output = GetOutput(context, node, kOutputTensor);
TF_LITE_ENSURE_EQ(context, params->activation, kTfLiteActRelu);