fixed

include/ck/tensor_operation/gpu/device/impl/device_gemm_wmma.hpp

@@ -94,13 +94,8 @@ struct DeviceGemmWmma_CShuffle : public DeviceGemm<ALayout,
         (MWaves == 1 && is_same<tensor_layout::gemm::ColumnMajor, BLayout>::value) ? false : true;
 
     // If true, LDS is used unconditionally
-#ifdef __gfx12__
     static constexpr auto AEnableLds_manu = true;
     static constexpr auto BEnableLds_manu = true;
-#else
-    static constexpr auto AEnableLds_manu = false;
-    static constexpr auto BEnableLds_manu = false;
-#endif
 
     static constexpr auto AEnableLds = AEnableLds_auto || AEnableLds_manu || (NumPrefetch > 1);
     static constexpr auto BEnableLds = BEnableLds_auto || BEnableLds_manu || (NumPrefetch > 1);

script/cmake-ck-dev.sh

@@ -11,7 +11,7 @@ cmake
 -D CMAKE_CXX_FLAGS="-std=c++17 -O3 -ftemplate-backtrace-limit=0  -fPIE  -Wno-gnu-line-marker"     \
 -D CMAKE_BUILD_TYPE=Release                                                                       \
 -D BUILD_DEV=OFF                                                                                   \
--D GPU_TARGETS="gfx1100"                                                             \
+-D GPU_TARGETS="gfx1200"                                                             \
 -D CMAKE_VERBOSE_MAKEFILE:BOOL=ON                                                                 \
 -D USE_BITINT_EXTENSION_INT4=OFF                                                                  \
 ${MY_PROJECT_SOURCE}