Merge branch 'musa' of github.com:BodhiHu/llama.cpp into musa

Author: Bodhi Hu

CMakeLists.txt

@@ -7,8 +7,6 @@ set(CMAKE_WARN_UNUSED_CLI YES)
 
 set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 
-#set(CMAKE_VERBOSE_MAKEFILE ON)
-
 if (NOT XCODE AND NOT MSVC AND NOT CMAKE_BUILD_TYPE)
     set(CMAKE_BUILD_TYPE Release CACHE STRING "Build type" FORCE)
     set_property(CACHE CMAKE_BUILD_TYPE PROPERTY STRINGS "Debug" "Release" "MinSizeRel" "RelWithDebInfo")

ggml/src/ggml-cuda/common.cuh

@@ -196,7 +196,7 @@ typedef float2 dfloat2;
 #endif // !(defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__)) && __CUDA_ARCH__ >= GGML_CUDA_CC_VOLTA
 
 #if !(defined(GGML_USE_MUSA) || (defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__))) && __CUDA_ARCH__ >= GGML_CUDA_CC_TURING
-#define INT8_MMA_AVAILABLE
+#define NEW_MMA_AVAILABLE
 #endif // !(defined(GGML_USE_HIP) && defined(__HIP_PLATFORM_AMD__)) && __CUDA_ARCH__ >= GGML_CUDA_CC_TURING
 
 #if !(defined(GGML_USE_MUSA) && __MUSA_ARCH__ <= GGML_CUDA_CC_QY1)

ggml/src/ggml-cuda/mmq.cu

@@ -139,7 +139,7 @@ bool ggml_cuda_should_use_mmq(enum ggml_type type, int cc, int64_t ne11) {
 
 #if defined(GGML_USE_MUSA)
     return true;
-#endif
+#endif // defined(GGML_USE_MUSA)
 
     if (ggml_cuda_highest_compiled_arch(cc) < GGML_CUDA_CC_DP4A) {
         return false;