Enable BINGINGS_FLAT in some examples

Author: peardox

examples/bench/bench.cpp

@@ -1,6 +1,7 @@
 #include "whisper.h"
 #ifdef BINDINGS_FLAT
 #include "whisper-flat.h"
+#include "../ggml/src/ggml-flat.h"
 #endif
 
 #include <cstdio>
@@ -65,7 +66,12 @@ static int whisper_bench_full(const whisper_params & params) {
     // whisper init
 
     #ifdef BINDINGS_FLAT
-    whisper_flat_backend_load_all();
+    fprintf(stderr, "+++ BINDINGS_FLAT +++\n");
+    if(params.use_gpu) {
+        whisper_flat_backend_load_all();
+    } else {
+        ggml_backend_try_load_best("cpu", nullptr);
+    }
     #endif
 
     struct whisper_context_params cparams = whisper_context_default_params();

examples/cli/cli.cpp

@@ -4,6 +4,7 @@
 #include "whisper.h"
 #ifdef BINDINGS_FLAT
 #include "whisper-flat.h"
+#include "../ggml/src/ggml-flat.h"
 #endif
 #include "grammar-parser.h"
 
@@ -1008,7 +1009,12 @@ int main(int argc, char ** argv) {
     // whisper init
 
     #ifdef BINDINGS_FLAT
-    whisper_flat_backend_load_all();
+    fprintf(stderr, "+++ BINDINGS_FLAT +++\n");
+    if(params.use_gpu) {
+        whisper_flat_backend_load_all();
+    } else {
+        ggml_backend_try_load_best("cpu", nullptr);
+    }
     #endif
 
     struct whisper_context_params cparams = whisper_context_default_params();

examples/command/command.cpp

@@ -11,6 +11,7 @@
 #include "whisper.h"
 #ifdef BINDINGS_FLAT
 #include "whisper-flat.h"
+#include "../ggml/src/ggml-flat.h"
 #endif
 #include "grammar-parser.h"
 
@@ -696,7 +697,12 @@ int main(int argc, char ** argv) {
     // whisper init
 
     #ifdef BINDINGS_FLAT
-    whisper_flat_backend_load_all();
+    fprintf(stderr, "+++ BINDINGS_FLAT +++\n");
+    if(params.use_gpu) {
+        whisper_flat_backend_load_all();
+    } else {
+        ggml_backend_try_load_best("cpu", nullptr);
+    }
     #endif
 
     struct whisper_context_params cparams = whisper_context_default_params();

examples/server/server.cpp

@@ -549,6 +549,7 @@ int main(int argc, char ** argv) {
     }
     // whisper init
     #ifdef BINDINGS_FLAT
+    fprintf(stderr, "+++ BINDINGS_FLAT +++\n");
     whisper_flat_backend_load_all();
     #endif
 

examples/stream/stream.cpp

@@ -8,6 +8,7 @@
 #include "whisper.h"
 #ifdef BINDINGS_FLAT
 #include "whisper-flat.h"
+#include "../ggml/src/ggml-flat.h"
 #endif
 
 #include <chrono>
@@ -159,7 +160,12 @@ int main(int argc, char ** argv) {
     }
 
     #ifdef BINDINGS_FLAT
-    whisper_flat_backend_load_all();
+    fprintf(stderr, "+++ BINDINGS_FLAT +++\n");
+    if(params.use_gpu) {
+        whisper_flat_backend_load_all();
+    } else {
+        ggml_backend_try_load_best("cpu", nullptr);
+    }
     #endif
 
     struct whisper_context_params cparams = whisper_context_default_params();

examples/talk-llama/talk-llama.cpp

@@ -7,6 +7,7 @@
 #include "whisper.h"
 #ifdef BINDINGS_FLAT
 #include "whisper-flat.h"
+#include "../ggml/src/ggml-flat.h"
 #endif
 #include "llama.h"
 
@@ -291,7 +292,12 @@ int main(int argc, char ** argv) {
     // whisper init
 
     #ifdef BINDINGS_FLAT
-    whisper_flat_backend_load_all();
+    fprintf(stderr, "+++ BINDINGS_FLAT +++\n");
+    if(params.use_gpu) {
+        whisper_flat_backend_load_all();
+    } else {
+        ggml_backend_try_load_best("cpu", nullptr);
+    }
     #endif
 
     struct whisper_context_params cparams = whisper_context_default_params();

ggml/src/ggml-backend-reg.cpp

@@ -109,9 +109,6 @@ struct dl_handle_deleter {
 
 static dl_handle * dl_load_library(const fs::path & path) {
     // suppress error dialogs for missing DLLs
-    #ifdef GGML_BINDINGS_FLAT // sbdbg
-    fprintf(stderr, "dl_load_library()%s",path.wstring().c_str()); // sbdbg
-    #endif // sbdbg
     DWORD old_mode = SetErrorMode(SEM_FAILCRITICALERRORS);
     SetErrorMode(old_mode | SEM_FAILCRITICALERRORS);
 
@@ -595,7 +592,7 @@ void ggml_backend_load_all_from_path(const char * dir_path) {
 #ifdef GGML_BINDINGS_FLAT
 ggml_backend_reg_t ggml_backend_try_load_best(const char * name, const char * dir_path) {
     bool silent = true;
-    fprintf(stderr, "Performing ggml_backend_try_load_best(%s)\n", name);
+    fprintf(stderr, "%s: (%s)\n", __func__, name);
     return ggml_backend_load_best(name, silent, dir_path);
 }
 #endif

ggml/src/ggml-flat.h

@@ -1,6 +1,7 @@
 #pragma once
 
 #ifdef BINDINGS_FLAT
+#define GGML_BINDINGS_FLAT
 #endif
 
 #ifdef GGML_SHARED