Fix RDNA4 perf

Author: hjc4869

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

@@ -1247,7 +1247,7 @@ static void ggml_cuda_op_mul_mat_cublas(
 
         CUBLAS_CHECK(cublasSetStream(ctx.cublas_handle(id), stream));
 
-        if (GGML_CUDA_CC_IS_CDNA(cc) || GGML_CUDA_CC_IS_RDNA4(cc)) {
+        if (GGML_CUDA_CC_IS_CDNA(cc)) {
             const float alpha = 1.0f;
             const float beta = 0.0f;
             CUBLAS_CHECK(
@@ -1792,7 +1792,7 @@ static void ggml_cuda_mul_mat_batched_cublas(ggml_backend_cuda_context & ctx, co
 
     int id = ggml_cuda_get_device();
     const int cc = ggml_cuda_info().devices[id].cc;
-    if (GGML_CUDA_CC_IS_CDNA(cc) || GGML_CUDA_CC_IS_RDNA4(cc)) {
+    if (GGML_CUDA_CC_IS_CDNA(cc)) {
         cu_compute_type = CUBLAS_COMPUTE_32F;
         alpha = &alpha_f32;
         beta  = &beta_f32;