mirror of
https://github.com/ROCm/composable_kernel.git
synced 2026-05-14 02:02:46 +00:00
Use DynamicBuffer instead of raw pointer (#32)
* Use DynamicBuffer to hold raw pointer (to global and LDS memory)
* add workaround for compiler issue (inefficient ISA) of ds_write for int8x4, int8x8, int8x16
[ROCm/composable_kernel commit: 78b987fbd6]
This commit is contained in:
@@ -10,13 +10,14 @@ cmake
|
||||
-D CMAKE_INSTALL_PREFIX=${MY_PROJECT_INSTALL} \
|
||||
-D CMAKE_BUILD_TYPE=Release \
|
||||
-D DEVICE_BACKEND="AMD" \
|
||||
-D CMAKE_CXX_FLAGS="-O3 --amdgpu-target=gfx1030 -gline-tables-only -save-temps=$CWD -ftemplate-backtrace-limit=0" \
|
||||
-D CMAKE_CXX_FLAGS="-O3 --amdgpu-target=gfx1030 -gline-tables-only -save-temps=$CWD -ftemplate-backtrace-limit=0" \
|
||||
-D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc \
|
||||
-D CMAKE_PREFIX_PATH="/opt/rocm" \
|
||||
-D CMAKE_VERBOSE_MAKEFILE:BOOL=ON \
|
||||
${MY_PROJECT_SOURCE}
|
||||
|
||||
#-D CMAKE_CXX_FLAGS="-O3 --amdgpu-target=gfx906 -save-temps=$CWD" \
|
||||
#-D CMAKE_CXX_FLAGS="-O3 --amdgpu-target=gfx1030 -gline-tables-only -save-temps=$CWD -ftemplate-backtrace-limit=0" \
|
||||
#-D CMAKE_CXX_FLAGS="-O3 --amdgpu-target=gfx1030 -gline-tables-only -save-temps=$CWD -ftemplate-backtrace-limit=0 -mllvm -print-before=amdgpu-codegenprepare -mllvm -print-module-scope" \
|
||||
#-D CMAKE_CXX_FLAGS="-O3 --amdgpu-target=gfx906 -gline-tables-only -save-temps=$CWD" \
|
||||
#-D CMAKE_CXX_FLAGS="-O3 --amdgpu-target=gfx906 -mllvm --amdgpu-spill-vgpr-to-agpr=0" \
|
||||
#-D CMAKE_CXX_FLAGS="-O3 --amdgpu-target=gfx906 -mllvm --amdgpu-spill-vgpr-to-agpr=0 -save-temps=$CWD" \
|
||||
Reference in New Issue
Block a user