ggml : fix warnings under MSVC (#1908)

ggml-cuda.cu

@@ -13,6 +13,10 @@
 #include "ggml-cuda.h"
 #include "ggml.h"
 
+#if defined(_MSC_VER)
+#pragma warning(disable: 4244 4267) // possible loss of data
+#endif
+
 static_assert(sizeof(half) == sizeof(ggml_fp16_t), "wrong fp16 size");
 
 #define CUDA_CHECK(err)                                                                 \

ggml-opencl.cpp

@@ -15,6 +15,10 @@
 
 #include "ggml.h"
 
+#if defined(_MSC_VER)
+#pragma warning(disable: 4244 4267) // possible loss of data
+#endif
+
 #define CL_DMMV_BLOCK_SIZE 32
 
 #define MULTILINE_QUOTE(...) #__VA_ARGS__

llama.cpp

@@ -1253,7 +1253,7 @@ static void llama_model_load_internal(
             vram_scratch = n_batch * MB;
             ggml_cuda_set_scratch_size(vram_scratch);
             if (n_gpu_layers > 0) {
-                fprintf(stderr, "%s: allocating batch_size x 1 MB = %ld MB VRAM for the scratch buffer\n",
+                fprintf(stderr, "%s: allocating batch_size x 1 MB = %zd MB VRAM for the scratch buffer\n",
                         __func__, vram_scratch / MB);
             }
         }