musa: update compile flags

Signed-off-by: Xiaodong Ye <xiaodong.ye@mthreads.com>

Author: yeahdongcn

ggml/src/ggml-cuda/fattn-vec.cuh

@@ -535,8 +535,6 @@ void ggml_cuda_flash_attn_ext_vec_case(ggml_backend_cuda_context & ctx, ggml_ten
     float logit_softcap;
     memcpy(&logit_softcap, (const float *) KQV->op_params + 2, sizeof(float));
 
-    const int cc = ggml_cuda_info().devices[ggml_cuda_get_device()].cc;
-
     if (Q->ne[1] == 1) {
         constexpr int cols_per_block = 1;
         if (logit_softcap == 0.0f) {

ggml/src/ggml-cuda/topk-moe.cu

@@ -33,9 +33,9 @@ __launch_bounds__(4 * WARP_SIZE, 1) __global__ void topk_moe_cuda(const float *
     float logits_r[experts_per_thread];
 
 #pragma unroll
-    for (int i = 0; i < n_experts; i += WARP_SIZE) {
+    for (int i = 0; i < static_cast<int>(n_experts); i += WARP_SIZE) {
         const int expert        = i + threadIdx.x;
-        logits_r[i / WARP_SIZE] = n_experts % WARP_SIZE == 0 || expert < n_experts ? logits[expert] : -INFINITY;
+        logits_r[i / WARP_SIZE] = n_experts % WARP_SIZE == 0 || expert < static_cast<int>(n_experts) ? logits[expert] : -INFINITY;
     }
 
     float max_val = logits_r[0];
@@ -83,7 +83,7 @@ __launch_bounds__(4 * WARP_SIZE, 1) __global__ void topk_moe_cuda(const float *
 #pragma unroll
         for (int i = 1; i < experts_per_thread; i++) {
             const int expert = threadIdx.x + i * WARP_SIZE;
-            if ((n_experts % WARP_SIZE == 0 || expert < n_experts) && wt[i] > max_val) {
+            if ((n_experts % WARP_SIZE == 0 || expert < static_cast<int>(n_experts)) && wt[i] > max_val) {
                 max_val    = wt[i];
                 max_expert = expert;
             }
@@ -204,8 +204,6 @@ void ggml_cuda_op_topk_moe(ggml_backend_cuda_context & ctx,
 
     GGML_ASSERT(ids->nb[1] / ggml_type_size(ids->type) == (size_t) n_experts);
 
-    cudaStream_t stream = ctx.stream();
-
     const int n_expert_used = weights->ne[1];
 
     if (with_norm) {

ggml/src/ggml-musa/CMakeLists.txt

@@ -56,7 +56,7 @@ if (MUSAToolkit_FOUND)
 
     set_source_files_properties(${GGML_SOURCES_MUSA} PROPERTIES LANGUAGE CXX)
     foreach(SOURCE ${GGML_SOURCES_MUSA})
-        set(COMPILE_FLAGS "-fsigned-char -x musa -mtgpu")
+        set(COMPILE_FLAGS "-Od3 -fno-strict-aliasing -ffast-math -fsigned-char -x musa -mtgpu -fmusa-flush-denormals-to-zero")
         foreach(ARCH ${MUSA_ARCHITECTURES})
             set(COMPILE_FLAGS "${COMPILE_FLAGS} --cuda-gpu-arch=mp_${ARCH}")
         endforeach()