update

Author: Bodhi Hu

docs/build.md

@@ -205,7 +205,8 @@ This provides GPU acceleration using the MUSA cores of your Moore Threads MTT GP
   cmake -B build -DGGML_MUSA=ON
   cmake --build build --config Release
   ```
-- For static build:
+
+  For static build:
 
   ```bash
   cmake -B build -DGGML_MUSA=ON \

ggml/src/ggml-cuda/ggml-cuda.cu

@@ -262,9 +262,7 @@ static ggml_cuda_device_info ggml_cuda_init() {
                       id, prop.name, prop.gcnArchName, info.devices[id].cc & 0xffff,
                       device_vmm ? "yes" : "no", prop.warpSize);
 #elif defined(GGML_USE_MUSA)
-        // TODO: MUSA will reserve some shared mem, and 24B should be enough,
-        // we can remove the **24** in the future when MUSA no longer reserves shared mem.
-        info.devices[id].smpbo = prop.sharedMemPerBlockOptin - 24;
+        info.devices[id].smpbo = prop.sharedMemPerBlockOptin;
         info.devices[id].cc = 100*prop.major + 10*prop.minor;
 #else
         info.devices[id].smpbo = prop.sharedMemPerBlockOptin;