ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-03 05:01:25 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add multiple d gridwise gemm on Navi21 for ResNet50 (#517)

Browse Source 
* start add example

* add multiple d fp16 example

* device transfer elementwiseop to gridwise

* gridwise add multiple d

* change example for multiple d

* fix spill registers

* fix for passthrough element op

* fix int8 overflow

* change example file name

* add instance for dl multiple d

* example add DsDataType

* remove grouped_convolution_forward_dl.hpp

* add head file(was deleted before)

* fix not support device issue

* format

* remove passthrough check

Co-authored-by: letaoqin <letaoqin@amd.com>
This commit is contained in:
ltqin
2022-12-03 01:42:31 +08:00
committed by GitHub
parent abf9cc6c5c
commit 23ecf0fa9e
15 changed files with 1922 additions and 348 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
include/ck/tensor_operation/gpu/element/binary_element_wise_operation.hpp
View File

@@ -187,6 +187,22 @@ struct AddRelu
const float a = x0 + type_convert<float>(x1);
y = a > 0.0f ? a : 0.0f;
};
template <>
__host__ __device__ constexpr void
operator()<int, int, int8_t>(int& y, const int& x0, const int8_t& x1) const
{
const int8_t a = x0 + x1;
y = a > 0 ? a : 0;
};
template <>
__host__ __device__ constexpr void
operator()<int8_t, int8_t, int8_t>(int8_t& y, const int8_t& x0, const int8_t& x1) const
{
const int8_t a = x0 + x1;
y = a > 0 ? a : 0;
};
};
struct AddHardswish