Skip to content

Add support for uint16 in quant and dequant kernels #6724

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

Merged
merged 1 commit into from
Nov 8, 2024
Merged

Add support for uint16 in quant and dequant kernels #6724

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
4 changes: 4 additions & 0 deletions backends/cadence/hifi/kernels/kernels.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -165,6 +165,7 @@ void requantize(
typed_quantize_val(int8_t);
typed_quantize_val(uint8_t);
typed_quantize_val(int16_t);
typed_quantize_val(uint16_t);
#undef typed_quantize_val

#define typed_quantize_vec(dtype) \
Expand All @@ -177,6 +178,7 @@ typed_quantize_val(int16_t);
typed_quantize_vec(int8_t);
typed_quantize_vec(uint8_t);
typed_quantize_vec(int16_t);
typed_quantize_vec(uint16_t);
typed_quantize_vec(int32_t);
#undef typed_quantize_vec

Expand All @@ -186,6 +188,7 @@ typed_quantize_vec(int32_t);
typed_dequantize_val(int8_t);
typed_dequantize_val(uint8_t);
typed_dequantize_val(int16_t);
typed_dequantize_val(uint16_t);
#undef typed_dequantize_val

#define typed_dequantize_vec(dtype) \
Expand All @@ -198,6 +201,7 @@ typed_dequantize_val(int16_t);
typed_dequantize_vec(int8_t);
typed_dequantize_vec(uint8_t);
typed_dequantize_vec(int16_t);
typed_dequantize_vec(uint16_t);
typed_dequantize_vec(int32_t);
#undef typed_dequantize_vec

Expand Down
3 changes: 3 additions & 0 deletions backends/cadence/hifi/operators/dequantize_per_tensor.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -41,6 +41,9 @@ void dequantize_per_tensor_out(
} else if (input.scalar_type() == ScalarType::Short) {
const int16_t* input_data = input.const_data_ptr<int16_t>();
dequantize<int16_t>(out_data, input_data, scale, zero_point, numel);
} else if (input.scalar_type() == ScalarType::Bits16) {
const uint16_t* input_data = input.const_data_ptr<uint16_t>();
dequantize<uint16_t>(out_data, input_data, scale, zero_point, numel);
} else if (input.scalar_type() == ScalarType::Int) {
const int32_t* input_data = input.const_data_ptr<int32_t>();
dequantize<int32_t>(out_data, input_data, scale, zero_point, numel);
Expand Down
4 changes: 4 additions & 0 deletions backends/cadence/hifi/operators/quantize_per_tensor.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -44,6 +44,10 @@ void quantize_per_tensor_out(
int16_t* out_data = out.mutable_data_ptr<int16_t>();
cadence::impl::HiFi::kernels::quantize<int16_t>(
out_data, input_data, 1. / scale, zero_point, numel);
} else if (out.scalar_type() == ScalarType::Bits16) {
uint16_t* out_data = out.mutable_data_ptr<uint16_t>();
cadence::impl::HiFi::kernels::quantize<uint16_t>(
out_data, input_data, 1. / scale, zero_point, numel);
} else if (out.scalar_type() == ScalarType::Int) {
int32_t* out_data = out.mutable_data_ptr<int32_t>();
cadence::impl::HiFi::kernels::quantize<int32_t>(
Expand Down
4 changes: 4 additions & 0 deletions backends/cadence/reference/kernels/kernels.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -65,6 +65,7 @@ void dequantize(
typed_quantize_val(int8_t);
typed_quantize_val(uint8_t);
typed_quantize_val(int16_t);
typed_quantize_val(uint16_t);
typed_quantize_val(int32_t);
#undef typed_quantize_val

Expand All @@ -78,6 +79,7 @@ typed_quantize_val(int32_t);
typed_quantize_vec(int8_t);
typed_quantize_vec(uint8_t);
typed_quantize_vec(int16_t);
typed_quantize_vec(uint16_t);
typed_quantize_vec(int32_t);
#undef typed_quantize_vec

Expand All @@ -86,6 +88,7 @@ typed_quantize_vec(int32_t);
typed_dequantize_val(int8_t);
typed_dequantize_val(uint8_t);
typed_dequantize_val(int16_t);
typed_dequantize_val(uint16_t);
typed_dequantize_val(int32_t);
#undef typed_dequantize_val

Expand All @@ -99,6 +102,7 @@ typed_dequantize_val(int32_t);
typed_dequantize_vec(int8_t);
typed_dequantize_vec(uint8_t);
typed_dequantize_vec(int16_t);
typed_dequantize_vec(uint16_t);
typed_dequantize_vec(int32_t);
#undef typed_dequantize_vec

Expand Down
8 changes: 8 additions & 0 deletions backends/cadence/reference/operators/dequantize_per_tensor.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -37,6 +37,14 @@ void dequantize_per_tensor_out(
const int8_t* input_data = input.const_data_ptr<int8_t>();
impl::reference::kernels::dequantize<int8_t>(
out_data, input_data, scale, zero_point, numel);
} else if (input.scalar_type() == ScalarType::Bits16) {
const uint16_t* input_data = input.const_data_ptr<uint16_t>();
impl::reference::kernels::dequantize<uint16_t>(
out_data, input_data, scale, zero_point, numel);
} else if (input.scalar_type() == ScalarType::Short) {
const int16_t* input_data = input.const_data_ptr<int16_t>();
impl::reference::kernels::dequantize<int16_t>(
out_data, input_data, scale, zero_point, numel);
} else if (input.scalar_type() == ScalarType::Int) {
const int32_t* input_data = input.const_data_ptr<int32_t>();
impl::reference::kernels::dequantize<int32_t>(
Expand Down
8 changes: 8 additions & 0 deletions backends/cadence/reference/operators/quantize_per_tensor.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,6 +39,14 @@ void quantize_per_tensor_out(
int8_t* out_data = out.mutable_data_ptr<int8_t>();
impl::reference::kernels::quantize<int8_t>(
out_data, input_data, 1. / scale, zero_point, numel);
} else if (out.scalar_type() == ScalarType::Bits16) {
uint16_t* out_data = out.mutable_data_ptr<uint16_t>();
impl::reference::kernels::quantize<uint16_t>(
out_data, input_data, 1. / scale, zero_point, numel);
} else if (out.scalar_type() == ScalarType::Short) {
int16_t* out_data = out.mutable_data_ptr<int16_t>();
impl::reference::kernels::quantize<int16_t>(
out_data, input_data, 1. / scale, zero_point, numel);
} else if (out.scalar_type() == ScalarType::Int) {
int32_t* out_data = out.mutable_data_ptr<int32_t>();
impl::reference::kernels::quantize<int32_t>(
Expand Down
Loading