wu-arch/kernels
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Fix build target for flash ampere

Browse Source
This commit is contained in:
Hansung Kim
2025-01-30 01:21:37 -08:00
parent 7f4adfaaa2
commit a368eb2dae
2 changed files with 4 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
kernels/flash_attention/Makefile
View File

@@ -1,8 +1,8 @@
PROJECT = flash_attention
# VX_SRCS = kernel.cpp
VX_SRCS = kernel.cpp
# VX_SRCS = kernel.gemmini.warpspec.cpp
VX_SRCS = kernel.gemmini.cpp
# VX_SRCS = kernel.gemmini.cpp
VX_INCLUDES = flash_impl.hpp ../sgemm_tcore/sgemm_impl.hpp
OPTS ?= -n16

4
kernels/flash_attention/flash_impl.hpp
View File

@@ -17,9 +17,9 @@ constexpr uint32_t ROWMAX_SETS = 3;
// constexpr bool WARP_SPECIALIZED = true;
// constexpr bool GEMMINI_WARP_SPECIALIZED = false;
// constexpr bool TENSOR_CORE = true;
constexpr bool WARP_SPECIALIZED = false;
constexpr bool WARP_SPECIALIZED = true;
constexpr bool GEMMINI_WARP_SPECIALIZED = false;
constexpr bool TENSOR_CORE = false;
constexpr bool TENSOR_CORE = true;
// temporary safety stop for wrong configs
static_assert(NUM_CORES == 4);