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

Add __restrict__ keywords to (Portable)MeanStddevNormalization input and output parameters.

Browse Source 
PiperOrigin-RevId: 352891183
Change-Id: I6b812ccdf76caecda13cc3484934619a3d63add8
This commit is contained in:
Robert David 2021-01-20 15:40:31 -08:00 committed by TensorFlower Gardener
parent b65ecec1bc
commit a60a884eab
6 changed files with 18 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
tensorflow/lite/kernels/internal/optimized/neon_tensor_utils.h
View File

@ -290,8 +290,9 @@ void ReductionSumVector(const int8_t* input_vector, int32_t* output_vector,
reduction_size);
}
void MeanStddevNormalization(const float* input_vector, float* output_vector,
int v_size, int n_batch) {
void MeanStddevNormalization(const float* __restrict__ input_vector,
float* __restrict__ output_vector, int v_size,
int n_batch) {
PortableMeanStddevNormalization(input_vector, output_vector, v_size, n_batch);
}

5
tensorflow/lite/kernels/internal/optimized/sse_tensor_utils.h
View File

@ -300,8 +300,9 @@ void ReductionSumVector(const int8_t* input_vector, int32_t* output_vector,
reduction_size);
}
void MeanStddevNormalization(const float* input_vector, float* output_vector,
int v_size, int n_batch) {
void MeanStddevNormalization(const float* __restrict__ input_vector,
float* __restrict__ output_vector, int v_size,
int n_batch) {
PortableMeanStddevNormalization(input_vector, output_vector, v_size, n_batch);
}

6
tensorflow/lite/kernels/internal/reference/portable_tensor_utils.cc
View File

@ -715,9 +715,9 @@ void PortableVectorScalarMultiply(const int8_t* vector, const int v_size,
}
}
void PortableMeanStddevNormalization(const float* input_vector,
float* output_vector, int v_size,
int n_batch) {
void PortableMeanStddevNormalization(const float* __restrict__ input_vector,
float* __restrict__ output_vector,
int v_size, int n_batch) {
for (int batch = 0; batch < n_batch; ++batch) {
float sum = 0.0f;
for (int i = 0; i < v_size; ++i) {

5
tensorflow/lite/kernels/internal/reference/portable_tensor_utils.h
View File

@ -294,8 +294,9 @@ void ReductionSumVector(const int8_t* input_vector, int32_t* output_vector,
reduction_size);
}
void MeanStddevNormalization(const float* input_vector, float* output_vector,
int v_size, int n_batch) {
void MeanStddevNormalization(const float* __restrict__ input_vector,
float* __restrict__ output_vector, int v_size,
int n_batch) {
PortableMeanStddevNormalization(input_vector, output_vector, v_size, n_batch);
}

6
tensorflow/lite/kernels/internal/reference/portable_tensor_utils_impl.h
View File

@ -214,9 +214,9 @@ void PortableReductionSumVector(const IN* input_vector, OUT* output_vector,
}
// Layer norm for each batch.
void PortableMeanStddevNormalization(const float* input_vector,
float* output_vector, int v_size,
int n_batch);
void PortableMeanStddevNormalization(const float* __restrict__ input_vector,
float* __restrict__ output_vector,
int v_size, int n_batch);
// Saturate Add.
void PortableTwoGateSaturatingAdd(const int8_t* input, int8_t input_zp,

5
tensorflow/lite/kernels/internal/tensor_utils.h
View File

@ -630,8 +630,9 @@ void ReductionSumVector(const int8_t* input_vector, int32_t* output_vector,
int output_size, int reduction_size);
// Layer norm for each batch.
void MeanStddevNormalization(const float* input_vector, float* output_vector,
int v_size, int n_batch);
void MeanStddevNormalization(const float* __restrict__ input_vector,
float* __restrict__ output_vector, int v_size,
int n_batch);
// Saturate Add with rescale on both inputs.
void TwoGateSaturatingAdd(const int8_t* input, int8_t input_zp,