Skip to content

GitLab

Unverified Commit 3eecbfb6 authored by rocking's avatar rocking Committed by GitHub
Browse files

Revise layout of group convolution (#675) 

* [What] Remove pure conv int8 instance
[Why] We will never use pure int8 conv in AI, use int8 quantization instead

* Change layout

* Share the kernel parameter

* Support more type of NHWGC for group conv

* Revise client example of conv 2d, use NHWGC layout

* Add instance to cmake

* Revise layout of group conv quantization instance

* Revise layout of external api of group conv quantization

* Revise layout of group conv quantization client example

* Fix clang format

* Add comment to describe meaning of each parameter
parent 903cd19c
Hide whitespace changes
Inline Side-by-side
Showing with 28 additions and 10 deletions
library/src/tensor_operation_instance/gpu/quantization/conv2d_fwd/device_conv2d_xdl_perlayer_quantization_int8_instance.cpp
View file @ 3eecbfb6
...@@ -9,10 +9,10 @@ namespace device { ...@@ -9,10 +9,10 @@ namespace device {
namespace instance { namespace instance {
void add_device_conv2d_xdl_perlayer_quantization_int8_instances( void add_device_conv2d_xdl_perlayer_quantization_int8_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial, std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
GNHWC, NHWGC,
GKYXC, GKYXC,
Empty_Tuple, Empty_Tuple,
GNHWK, NHWGK,
int8_t, int8_t,
int8_t, int8_t,
Empty_Tuple, Empty_Tuple,
...@@ -22,19 +22,28 @@ void add_device_conv2d_xdl_perlayer_quantization_int8_instances( ...@@ -22,19 +22,28 @@ void add_device_conv2d_xdl_perlayer_quantization_int8_instances(
Mul_Clamp>>>& instances) Mul_Clamp>>>& instances)
{ {
add_device_operation_instances(instances, add_device_operation_instances(instances,
device_grouped_conv2d_xdl_int8_instances<Empty_Tuple, device_grouped_conv2d_xdl_int8_instances<NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
Empty_Tuple, Empty_Tuple,
Mul_Clamp, Mul_Clamp,
ConvFwdDefault, ConvFwdDefault,
16>{}); 16>{});
add_device_operation_instances(instances, add_device_operation_instances(instances,
device_grouped_conv2d_xdl_int8_instances<Empty_Tuple, device_grouped_conv2d_xdl_int8_instances<NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
Empty_Tuple, Empty_Tuple,
Mul_Clamp, Mul_Clamp,
ConvFwd1x1P0, ConvFwd1x1P0,
16>{}); 16>{});
add_device_operation_instances(instances, add_device_operation_instances(instances,
device_grouped_conv2d_xdl_int8_instances<Empty_Tuple, device_grouped_conv2d_xdl_int8_instances<NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
Empty_Tuple, Empty_Tuple,
Mul_Clamp, Mul_Clamp,
ConvFwd1x1S1P0, ConvFwd1x1S1P0,
...@@ -43,10 +52,10 @@ void add_device_conv2d_xdl_perlayer_quantization_int8_instances( ...@@ -43,10 +52,10 @@ void add_device_conv2d_xdl_perlayer_quantization_int8_instances(
void add_device_conv2d_xdl_relu_perlayer_quantization_int8_instances( void add_device_conv2d_xdl_relu_perlayer_quantization_int8_instances(
std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial, std::vector<std::unique_ptr<DeviceGroupedConvFwdMultipleD<NDimSpatial,
GNHWC, NHWGC,
GKYXC, GKYXC,
Empty_Tuple, Empty_Tuple,
GNHWK, NHWGK,
int8_t, int8_t,
int8_t, int8_t,
Empty_Tuple, Empty_Tuple,
...@@ -56,19 +65,28 @@ void add_device_conv2d_xdl_relu_perlayer_quantization_int8_instances( ...@@ -56,19 +65,28 @@ void add_device_conv2d_xdl_relu_perlayer_quantization_int8_instances(
Relu_Mul_Clamp>>>& instances) Relu_Mul_Clamp>>>& instances)
{ {
add_device_operation_instances(instances, add_device_operation_instances(instances,
device_grouped_conv2d_xdl_int8_instances<Empty_Tuple, device_grouped_conv2d_xdl_int8_instances<NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
Empty_Tuple, Empty_Tuple,
Relu_Mul_Clamp, Relu_Mul_Clamp,
ConvFwdDefault, ConvFwdDefault,
16>{}); 16>{});
add_device_operation_instances(instances, add_device_operation_instances(instances,
device_grouped_conv2d_xdl_int8_instances<Empty_Tuple, device_grouped_conv2d_xdl_int8_instances<NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
Empty_Tuple, Empty_Tuple,
Relu_Mul_Clamp, Relu_Mul_Clamp,
ConvFwd1x1P0, ConvFwd1x1P0,
16>{}); 16>{});
add_device_operation_instances(instances, add_device_operation_instances(instances,
device_grouped_conv2d_xdl_int8_instances<Empty_Tuple, device_grouped_conv2d_xdl_int8_instances<NHWGC,
GKYXC,
Empty_Tuple,
NHWGK,
Empty_Tuple, Empty_Tuple,
Relu_Mul_Clamp, Relu_Mul_Clamp,
ConvFwd1x1S1P0, ConvFwd1x1S1P0,
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment