Skip to content

Optimize Xtensa transpose_conv and decompress kernels #3727

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
unmeshna017 wants to merge 1 commit into
base: main
Choose a base branch
from
+179 −109
Open

Optimize Xtensa transpose_conv and decompress kernels #3727

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
6 changes: 2 additions & 4 deletions tensorflow/lite/micro/kernels/decompress_common.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -317,8 +317,6 @@ template <typename T>
void DecompressionState::DecompressToBufferWidthAny(T* buffer) {
ScopedMicroProfiler scoped_profiler(__func__, micro_profiler_);

TFLITE_DCHECK(compressed_bit_width_ > 0 && compressed_bit_width_ < 8);

if (comp_data_.data.lut_data->use_alternate_axis) {
const size_t stride = comp_data_.data.lut_data->value_table_channel_stride;
size_t current_offset = 0;
Expand All @@ -328,7 +326,7 @@ void DecompressionState::DecompressToBufferWidthAny(T* buffer) {
const T* value_table =
static_cast<const T*>(comp_data_.data.lut_data->value_table);
for (size_t channel = 0; channel < num_channels_; channel++) {
size_t index = 0;
size_t index;
switch (compressed_bit_width_) {
case 1:
index = GetNextTableIndexWidth1(current_offset);
Expand Down Expand Up @@ -369,7 +367,7 @@ void DecompressionState::DecompressToBufferWidthAny(T* buffer) {
size_t count = max_count;

while (count-- > 0) {
size_t index = 0;
size_t index;
switch (compressed_bit_width_) {
case 1:
index = GetNextTableIndexWidth1(current_offset);
Expand Down
7 changes: 0 additions & 7 deletions tensorflow/lite/micro/kernels/xtensa/decompress.cc
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -53,7 +53,6 @@ struct DecompressionStateXtensa : DecompressionState {

void DecompressionStateXtensa::DecompressToBufferWidth4_Xtensa(int8_t* buffer) {
ScopedMicroProfiler scoped_profiler(__func__, micro_profiler_);
(void)scoped_profiler;

ae_int8x8 d_shuffle_t = AE_MOVINT8X8_FROMINT64(0xFB73EA62D951C840LL);
ae_int8x8 d_shuffle_value_t = AE_MOVINT8X8_FROMINT64(0x08192A3B4C5D6E7FLL);
Expand Down Expand Up @@ -111,7 +110,6 @@ void DecompressionStateXtensa::DecompressToBufferWidth4_Xtensa(int8_t* buffer) {

void DecompressionStateXtensa::DecompressToBufferWidth3_Xtensa(int8_t* buffer) {
ScopedMicroProfiler scoped_profiler(__func__, micro_profiler_);
(void)scoped_profiler;

int i, j;
ae_int8* __restrict p_out_tmp = (ae_int8*)buffer;
Expand Down Expand Up @@ -234,7 +232,6 @@ void DecompressionStateXtensa::DecompressToBufferWidth3_Xtensa(int8_t* buffer) {

void DecompressionStateXtensa::DecompressToBufferWidth2_Xtensa(int8_t* buffer) {
ScopedMicroProfiler scoped_profiler(__func__, micro_profiler_);
(void)scoped_profiler;

int i, j;
ae_int8* __restrict p_out_tmp = (ae_int8*)buffer;
Expand Down Expand Up @@ -353,7 +350,6 @@ void DecompressionStateXtensa::DecompressToBufferWidth2_Xtensa(int8_t* buffer) {
void DecompressionStateXtensa::DecompressToBufferWidthAnyInt8_Xtensa(
int8_t* buffer) {
ScopedMicroProfiler scoped_profiler(__func__, micro_profiler_);
(void)scoped_profiler;

const int stride = comp_data_.data.lut_data->value_table_channel_stride;
const uint8_t* __restrict value_table =
Expand Down Expand Up @@ -424,7 +420,6 @@ void DecompressionStateXtensa::DecompressToBufferWidthAnyInt8_Xtensa(
void DecompressionStateXtensa::DecompressToBufferWidthAnyInt16_Xtensa(
int16_t* buffer) {
ScopedMicroProfiler scoped_profiler(__func__, micro_profiler_);
(void)scoped_profiler;

const int stride = comp_data_.data.lut_data->value_table_channel_stride;
const uint16_t* __restrict value_table =
Expand Down Expand Up @@ -476,7 +471,6 @@ void DecompressionStateXtensa::DecompressToBufferWidthAnyInt16_Xtensa(
void DecompressionStateXtensa::DecompressToBufferWidthAnyInt32_Xtensa(
int32_t* buffer) {
ScopedMicroProfiler scoped_profiler(__func__, micro_profiler_);
(void)scoped_profiler;

const int stride = comp_data_.data.lut_data->value_table_channel_stride;
const uint32_t* __restrict value_table =
Expand Down Expand Up @@ -528,7 +522,6 @@ void DecompressionStateXtensa::DecompressToBufferWidthAnyInt32_Xtensa(
void DecompressionStateXtensa::DecompressToBufferWidthAnyInt64_Xtensa(
int64_t* buffer) {
ScopedMicroProfiler scoped_profiler(__func__, micro_profiler_);
(void)scoped_profiler;

const int stride = comp_data_.data.lut_data->value_table_channel_stride;
const uint64_t* __restrict value_table =
Expand Down
Loading
Loading