Merge pull request #19 from kpouget/remoting

remotingfrontend: detect initialization issue and improve the logging

Author: kpouget

ggml/src/ggml-remotingbackend/backend-dispatched-metal.cpp

@@ -14,6 +14,7 @@ void (*ggml_backend_metal_get_device_context_fct)(ggml_backend_dev_t dev,
 uint32_t
 backend_metal_get_device_context(struct vn_cs_encoder *enc, struct vn_cs_decoder *dec, struct virgl_apir_context *ctx) {
   UNUSED(ctx);
+  UNUSED(dec);
 
   bool has_simdgroup_mm;
   bool has_simdgroup_reduction;

ggml/src/ggml-remotingbackend/backend-utils.h

@@ -7,47 +7,55 @@
 #include <ggml.h>
 
 #define UNUSED GGML_UNUSED
+#define APIR_LLAMA_CPP_LOG_TO_FILE_ENV "APIR_LLAMA_CPP_LOG_TO_FILE"
+
+static FILE *
+get_log_dest(void)
+{
+   static FILE *dest = NULL;
+   if (dest) {
+      return dest;
+   }
+   const char *apir_log_to_file = getenv(APIR_LLAMA_CPP_LOG_TO_FILE_ENV);
+   if (!apir_log_to_file) {
+      dest = stderr;
+      return dest;
+   }
+
+   dest = fopen(apir_log_to_file, "w");
+
+   return dest;
+}
+
+#define APIR_VA_PRINT(prefix, format)               \
+   do {                                             \
+      FILE *dest = get_log_dest();                  \
+      fprintf(dest, prefix);                        \
+      va_list argptr;                               \
+      va_start(argptr, format);                     \
+      vfprintf(dest, format, argptr);               \
+      fprintf(dest, "\n");                          \
+      va_end(argptr);                               \
+      fflush(dest);                                 \
+   } while (0)
 
 inline void
 INFO(const char *format, ...) {
-  va_list argptr;
-  va_start(argptr, format);
-  vfprintf(stderr, format, argptr);
-  fprintf(stderr, "\n");
-  va_end(argptr);
+  APIR_VA_PRINT("INFO: ", format);
 }
 
 inline void
 WARNING(const char *format, ...) {
-  fprintf(stderr, "WARNING: ");
-
-  va_list argptr;
-  va_start(argptr, format);
-  vfprintf(stderr, format, argptr);
-  fprintf(stderr, "\n");
-  va_end(argptr);
+  APIR_VA_PRINT("WARNING: ", format);
 }
 
 inline void
 ERROR(const char *format, ...) {
-  fprintf(stderr, "ERROR: ");
-
-  va_list argptr;
-  va_start(argptr, format);
-  vfprintf(stderr, format, argptr);
-  fprintf(stderr, "\n");
-  va_end(argptr);
+  APIR_VA_PRINT("ERROR: ", format);
 }
 
-inline void
+[[noreturn]] inline void
 FATAL(const char *format, ...) {
-  fprintf(stderr, "FATAL: ");
-
-  va_list argptr;
-  va_start(argptr, format);
-  vfprintf(stderr, format, argptr);
-  fprintf(stderr, "\n");
-  va_end(argptr);
-  if (format)
-    assert(false);
+  APIR_VA_PRINT("FORMAT: ", format);
+  abort();
 }

ggml/src/ggml-remotingbackend/shared/apir_backend.h

@@ -6,6 +6,7 @@
 #define APIR_BACKEND_INITIALIZE_MISSING_BACKEND_SYMBOLS 3
 #define APIR_BACKEND_INITIALIZE_MISSING_GGML_SYMBOLS 4
 #define APIR_BACKEND_INITIALIZE_BACKEND_FAILED 5
+// new entries here need to be added to the apir_backend_initialize_error function below
 
 #define APIR_BACKEND_FORWARD_INDEX_INVALID 6
 
@@ -106,3 +107,21 @@ static inline void show_timer(struct timer_data *timer) {
   INFO("%14s [%9.0f] ms for %4ld invocations | ITL %2.2f ms | throughput = %4.2f t/s",
        timer->name, ms, timer->count, itl, speed);
 }
+
+static const char *apir_backend_initialize_error(int code) {
+#define APIR_BACKEND_INITIALIZE_ERROR(code_name) \
+  do {						 \
+    if (code == code_name) return #code_name;	 \
+  } while (0)					 \
+
+  APIR_BACKEND_INITIALIZE_ERROR(APIR_BACKEND_INITIALIZE_SUCCESSS);
+  APIR_BACKEND_INITIALIZE_ERROR(APIR_BACKEND_INITIALIZE_CANNOT_OPEN_BACKEND_LIBRARY);
+  APIR_BACKEND_INITIALIZE_ERROR(APIR_BACKEND_INITIALIZE_CANNOT_OPEN_GGML_LIBRARY);
+  APIR_BACKEND_INITIALIZE_ERROR(APIR_BACKEND_INITIALIZE_MISSING_BACKEND_SYMBOLS);
+  APIR_BACKEND_INITIALIZE_ERROR(APIR_BACKEND_INITIALIZE_MISSING_GGML_SYMBOLS);
+  APIR_BACKEND_INITIALIZE_ERROR(APIR_BACKEND_INITIALIZE_BACKEND_FAILED);
+
+  return "Unknown APIR_BACKEND_INITIALIZE error:/";
+
+#undef APIR_BACKEND_INITIALIZE_ERROR
+}

ggml/src/ggml-remotingfrontend/virtgpu.cpp

@@ -82,10 +82,12 @@ create_virtgpu() {
 
   if (!gpu->reply_shmem) {
     FATAL("%s: failed to create the shared reply memory pages :/", __func__);
+    return NULL;
   }
 
   if (!gpu->data_shmem) {
     FATAL("%s: failed to create the shared data memory pages :/", __func__);
+    return NULL;
   }
 
   struct vn_cs_encoder *encoder;
@@ -100,11 +102,14 @@ create_virtgpu() {
   decoder = remote_call(gpu, encoder, MAX_WAIT_US);
   if (!decoder) {
     FATAL("%s: failed to initialize the API remoting libraries. :/", __func__);
+    return NULL;
   }
 
   ret = remote_call_finish(encoder, decoder);
   if (ret != 0) {
-    FATAL("%s: failed to load the APIR backend libraries (code=%d):/", __func__, ret);
+    FATAL("%s: failed to load the APIR backend libraries (code=%d | %s):/",
+	  __func__, ret, apir_backend_initialize_error(ret));
+    return NULL;
   }
 
   return gpu;