Merge pull request #9 from kpouget/remoting

Cache values and reduce the logging

Author: kpouget

ggml/src/ggml-remotingfrontend/ggml-backend-buffer-type.cpp

@@ -5,7 +5,7 @@
 
 static ggml_backend_buffer_t
 ggml_backend_remoting_buffer_type_alloc_buffer(ggml_backend_buffer_type_t buft, size_t size) {
-  IMPLEMENTED;
+  IMPLEMENTED_ONCE;
   struct virtgpu *gpu = BUFT_TO_GPU(buft);
 
   struct ggml_backend_remoting_buffer_context *context = (struct ggml_backend_remoting_buffer_context *) malloc(sizeof(*context));
@@ -29,9 +29,6 @@ ggml_backend_remoting_buffer_type_alloc_buffer(ggml_backend_buffer_type_t buft,
   context->is_host_buffer = false;
 
   ggml_backend_buffer_t buffer = ggml_backend_buffer_init(buft, ggml_backend_remoting_buffer_interface, (void *) context, size);
-  INFO("##");
-  INFO("## %s(%llx) --> %p <---------------", __func__, size, buffer);
-  INFO("##\n");
 
   return buffer;
 }
@@ -47,19 +44,29 @@ ggml_backend_remoting_buffer_type_get_name(ggml_backend_buffer_type_t buft) {
 
 static size_t
 ggml_backend_remoting_buffer_type_get_alignment(ggml_backend_buffer_type_t buft) {
-  IMPLEMENTED;
-
+  IMPLEMENTED_ONCE;
   struct virtgpu *gpu = BUFT_TO_GPU(buft);
 
-  return apir_buffer_type_get_alignment(gpu, buft);
+  static size_t align = 0;
+
+  if (align == 0) {
+    align = apir_buffer_type_get_alignment(gpu, buft);
+  }
+
+  return align;
 }
 
 static size_t
 ggml_backend_remoting_buffer_type_get_max_size(ggml_backend_buffer_type_t buft) {
-  IMPLEMENTED;
+  IMPLEMENTED_ONCE;
   struct virtgpu *gpu = BUFT_TO_GPU(buft);
 
-  return apir_buffer_type_get_max_size(gpu, buft);
+  static size_t max_size = 0;
+  if (max_size == 0) {
+    max_size = apir_buffer_type_get_max_size(gpu, buft);
+  }
+
+  return max_size;
 }
 
 static bool

ggml/src/ggml-remotingfrontend/ggml-backend-device.cpp

@@ -20,11 +20,17 @@ ggml_backend_remoting_device_get_description(ggml_backend_dev_t dev) {
 
 static enum ggml_backend_dev_type
 ggml_backend_remoting_device_get_type(ggml_backend_dev_t dev) {
-  IMPLEMENTED;
-
+  IMPLEMENTED_ONCE;
   struct virtgpu *gpu = DEV_TO_GPU(dev);
 
-  return (enum ggml_backend_dev_type) apir_device_get_type(gpu);
+  static enum ggml_backend_dev_type type;
+  static bool has_type = false;
+  if (!has_type) {
+    has_type = true;
+    type = (enum ggml_backend_dev_type) apir_device_get_type(gpu);
+  }
+
+  return type;
 }
 
 static void

ggml/src/ggml-remotingfrontend/venus_cs_ggml-rpc-front.cpp

@@ -18,10 +18,6 @@ serialize_tensor(const ggml_tensor * tensor) {
     ggml_backend_buffer_t buffer = tensor->buffer;
 
     result.buffer = BUFFER_TO_HOST_HANDLE(buffer);
-    if (result.buffer < 0x600000000000 || result.buffer > 0x700000000000) {
-      INFO("pass buffer handle %p", result.buffer);
-      BREAKPOINT;
-    }
   } else {
     result.buffer = 0;
   }