Fix: FP16_MMA_AVAILABLE should not be checked in host code.

Author: hjc4869

ggml/src/ggml-cuda/fattn.cu

@@ -253,12 +253,12 @@ void ggml_cuda_flash_attn_ext(ggml_backend_cuda_context & ctx, ggml_tensor * dst
     const enum ggml_prec prec = ggml_flash_attn_ext_get_prec(KQV);
 
     if (cc >= GGML_CUDA_CC_OFFSET_AMD) {
-#if defined(GGML_HIP_ROCWMMA_FATTN) && defined(FP16_MMA_AVAILABLE)
+#if defined(GGML_HIP_ROCWMMA_FATTN)
         if (fp16_mma_available(cc) && dst->src[0]->ne[1] > 8) {
             ggml_cuda_flash_attn_ext_wmma_f16(ctx, dst);
             return;
         }
-#endif // defined(GGML_HIP_ROCWMMA_FATTN) && defined(FP16_MMA_AVAILABLE)
+#endif // defined(GGML_HIP_ROCWMMA_FATTN)
 
         // On AMD the tile kernels perform poorly, use the vec kernel instead:
         if (prec == GGML_PREC_DEFAULT && fast_fp16_available(cc)) {