Recover FP16 performance and improve general performance

Author: pt13762104

ggml/src/ggml-cuda/common.cuh

@@ -268,7 +268,7 @@ static bool fast_fp16_available(const int cc) {
 
 // To be used for feature selection of external libraries, e.g. cuBLAS.
 static bool fast_fp16_hardware_available(const int cc) {
-    return (GGML_CUDA_CC_IS_NVIDIA(cc) && cc >= GGML_CUDA_CC_PASCAL && cc != 610) || GGML_CUDA_CC_IS_AMD(cc) ||
+    return (GGML_CUDA_CC_IS_NVIDIA(cc) && cc >= GGML_CUDA_CC_PASCAL && cc != 610 && !MMA_DISABLE) || GGML_CUDA_CC_IS_AMD(cc) ||
         (GGML_CUDA_CC_IS_MTHREADS(cc) && cc >= GGML_CUDA_CC_QY2);
 }