ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-06-29 11:16:59 +00:00
Code Issues Packages Projects Releases Wiki Activity

Update README.md

Browse Source
This commit is contained in:
ClementLinCF
2025-04-25 07:15:12 +08:00
committed by GitHub
parent 04540f2eac
commit 89bf0765fb
1 changed files with 57 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

57
example/ck_tile/99_toy_example/README.md
View File

@@ -53,3 +53,60 @@ make -j basic_flash_attention_fwd
```sh
./bin/basic_flash_attention_fwd 1 1
```
## Advanced part
#### **GEMM Example**
##### Follow these steps to build and run the different kernels:
```sh
cd composable_kernel
mkdir build
cd build
# for naive kernel
cmake -D CMAKE_PREFIX_PATH=/opt/rocm -D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc -D CMAKE_BUILD_TYPE=Release -D GPU_TARGETS="gfx942" -Dkernel=N .. && make -j basic_gemm
# for kernel A
cmake -D CMAKE_PREFIX_PATH=/opt/rocm -D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc -D CMAKE_BUILD_TYPE=Release -D GPU_TARGETS="gfx942" -Dkernel=A .. && make -j basic_gemm
# for kernel B
cmake -D CMAKE_PREFIX_PATH=/opt/rocm -D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc -D CMAKE_BUILD_TYPE=Release -D GPU_TARGETS="gfx942" -Dkernel=B .. && make -j basic_gemm
...
# for kernel H
cmake -D CMAKE_PREFIX_PATH=/opt/rocm -D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc -D CMAKE_BUILD_TYPE=Release -D GPU_TARGETS="gfx942" -Dkernel=H .. && make -j basic_gemm
```
```sh
./bin/basic_gemm 1
```
#### **Flash Attention Forward Example**
##### Follow these steps to build the kernels
```sh
# for naive kernel
cmake -D CMAKE_PREFIX_PATH=/opt/rocm -D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc -D CMAKE_BUILD_TYPE=Release -D GPU_TARGETS="gfx942" ..
# for optimized kernel
cmake -D CMAKE_PREFIX_PATH=/opt/rocm -D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc -D CMAKE_BUILD_TYPE=Release -D GPU_TARGETS="gfx942" -DENABLE_TOY_FA_FWD_OPT=ON ..
```
```sh
./bin/basic_flash_attention_fwd 1 1
```
##### Follow these steps to build the codegen instances
```sh
mkdir build
cd build
../script/cmake-ck-release.sh .. gfx942
make -j codegen_basic_flash_attention_fwd
```
```sh
./bin/codegen_basic_flash_attention_fwd 1 1 64 16384 16384 128 128
```