musa: fix aarch64 build

CMakeLists.txt

@@ -7,6 +7,8 @@ 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")

docs/build.md

@@ -148,8 +148,8 @@ This provides GPU acceleration using the MUSA cores of your Moore Threads MTT GP
 - Using `CMake`:
 
   ```bash
-  cmake -B build -DGGML_MUSA=ON
-  cmake --build build --config Release
+  cmake -B build -DGGML_MUSA=ON -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++
+  cmake --build build --config Release -j$(($(nproc)-2))
   ```
 
 The environment variable [`MUSA_VISIBLE_DEVICES`](https://docs.mthreads.com/musa-sdk/musa-sdk-doc-online/programming_guide/Z%E9%99%84%E5%BD%95/) can be used to specify which GPU(s) will be used.

ggml/include/ggml-cuda.h

@@ -1,5 +1,9 @@
 #pragma once
 
+#ifdef GGML_USE_MUSA
+#undef __ARM_NEON
+#endif
+
 #include "ggml.h"
 #include "ggml-backend.h"
 

ggml/src/ggml-cuda/common.cuh

@@ -1,5 +1,9 @@
 #pragma once
 
+#ifdef GGML_USE_MUSA
+#undef __ARM_NEON
+#endif
+
 #include "ggml.h"
 #include "ggml-cuda.h"
 

ggml/src/ggml-cuda/opt-step-adamw.cu

@@ -1,3 +1,7 @@
+#ifdef GGML_USE_MUSA
+#undef __ARM_NEON
+#endif
+
 #include "ggml-impl.h"
 #include "opt-step-adamw.cuh"