Add base framework for flat bindings

CMakeLists.txt

@@ -70,6 +70,9 @@ option(WHISPER_ALL_WARNINGS_3RD_PARTY "whisper: enable all compiler warnings in
 option(WHISPER_FATAL_WARNINGS  "whisper: enable -Werror flag"               OFF)
 option(WHISPER_USE_SYSTEM_GGML "whisper: use system-installed GGML library" OFF)
 
+# flat bindings
+option(BINDINGS_FLAT "Add extra flat definitions to Whisper + GGML" OFF)
+
 # sanitizers
 option(WHISPER_SANITIZE_THREAD    "whisper: enable thread sanitizer"    OFF)
 option(WHISPER_SANITIZE_ADDRESS   "whisper: enable address sanitizer"   OFF)

examples/bench/bench.cpp

@@ -1,4 +1,7 @@
 #include "whisper.h"
+#if defined(GGML_BACKEND_DL) or defined(BINDINGS_FLAT)
+#include "whisper-flat.h"
+#endif
 
 #include <cstdio>
 #include <cstring>
@@ -61,6 +64,10 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
 static int whisper_bench_full(const whisper_params & params) {
     // whisper init
 
+    #if defined(GGML_BACKEND_DL) or defined(BINDINGS_FLAT)
+    whisper_flat_backend_load_all();
+    #endif
+
     struct whisper_context_params cparams = whisper_context_default_params();
 
     cparams.use_gpu    = params.use_gpu;

examples/cli/cli.cpp

@@ -2,6 +2,9 @@
 #include "common-whisper.h"
 
 #include "whisper.h"
+#if defined(GGML_BACKEND_DL) or defined(BINDINGS_FLAT)
+#include "whisper-flat.h"
+#endif
 #include "grammar-parser.h"
 
 #include <cmath>
@@ -1004,6 +1007,10 @@ int main(int argc, char ** argv) {
 
     // whisper init
 
+    #if defined(GGML_BACKEND_DL) or defined(BINDINGS_FLAT)
+    whisper_flat_backend_load_all();
+    #endif
+
     struct whisper_context_params cparams = whisper_context_default_params();
 
     cparams.use_gpu    = params.use_gpu;

examples/command/command.cpp

@@ -9,6 +9,9 @@
 #include "common-sdl.h"
 #include "common.h"
 #include "whisper.h"
+#if defined(GGML_BACKEND_DL) or defined(BINDINGS_FLAT)
+#include "whisper-flat.h"
+#endif
 #include "grammar-parser.h"
 
 #include <algorithm>
@@ -692,6 +695,10 @@ int main(int argc, char ** argv) {
 
     // whisper init
 
+    #if defined(GGML_BACKEND_DL) or defined(BINDINGS_FLAT)
+    whisper_flat_backend_load_all();
+    #endif
+
     struct whisper_context_params cparams = whisper_context_default_params();
 
     cparams.use_gpu    = params.use_gpu;

examples/server/server.cpp

@@ -2,6 +2,9 @@
 #include "common-whisper.h"
 
 #include "whisper.h"
+#if defined(GGML_BACKEND_DL) or defined(BINDINGS_FLAT)
+#include "whisper-flat.h"
+#endif
 #include "httplib.h"
 #include "json.hpp"
 
@@ -544,6 +547,10 @@ int main(int argc, char ** argv) {
         check_ffmpeg_availibility();
     }
     // whisper init
+    #if defined(GGML_BACKEND_DL) or defined(BINDINGS_FLAT)
+    whisper_flat_backend_load_all();
+    #endif
+
     struct whisper_context_params cparams = whisper_context_default_params();
 
     cparams.use_gpu    = params.use_gpu;

examples/stream/stream.cpp

@@ -6,6 +6,9 @@
 #include "common.h"
 #include "common-whisper.h"
 #include "whisper.h"
+#if defined(GGML_BACKEND_DL) or defined(BINDINGS_FLAT)
+#include "whisper-flat.h"
+#endif
 
 #include <chrono>
 #include <cstdio>
@@ -155,6 +158,10 @@ int main(int argc, char ** argv) {
         exit(0);
     }
 
+    #if defined(GGML_BACKEND_DL) or defined(BINDINGS_FLAT)
+    whisper_flat_backend_load_all();
+    #endif
+
     struct whisper_context_params cparams = whisper_context_default_params();
 
     cparams.use_gpu    = params.use_gpu;

examples/talk-llama/talk-llama.cpp

@@ -5,6 +5,9 @@
 #include "common.h"
 #include "common-whisper.h"
 #include "whisper.h"
+#if defined(GGML_BACKEND_DL) or defined(BINDINGS_FLAT)
+#include "whisper-flat.h"
+#endif
 #include "llama.h"
 
 #include <chrono>
@@ -287,6 +290,10 @@ int main(int argc, char ** argv) {
 
     // whisper init
 
+    #if defined(GGML_BACKEND_DL) or defined(BINDINGS_FLAT)
+    whisper_flat_backend_load_all();
+    #endif
+
     struct whisper_context_params cparams = whisper_context_default_params();
 
     cparams.use_gpu    = params.use_gpu;

examples/wchess/wchess.cmd/wchess.cmd.cpp

@@ -7,6 +7,9 @@
 
 #include "WChess.h"
 #include "common-sdl.h"
+#if defined(GGML_BACKEND_DL) or defined(BINDINGS_FLAT)
+#include "whisper-flat.h"
+#endif
 #include <iostream>
 
 #include <memory>
@@ -182,6 +185,10 @@ int main(int argc, char ** argv) {
 
     // whisper init
 
+    #if defined(GGML_BACKEND_DL) or defined(BINDINGS_FLAT)
+    whisper_flat_backend_load_all();
+    #endif
+
     struct whisper_context_params cparams = whisper_context_default_params();
 
     cparams.use_gpu    = params.use_gpu;

ggml/include/ggml-backend.h

@@ -348,6 +348,8 @@ extern "C" {
     // CPU buffer types are always available
     GGML_API ggml_backend_buffer_t      ggml_backend_cpu_buffer_from_ptr(void * ptr, size_t size);
     GGML_API ggml_backend_buffer_type_t ggml_backend_cpu_buffer_type(void);
+
+    GGML_API ggml_backend_reg_t ggml_backend_try_load_best(const char * name, const char * dir_path);
 
 #ifdef  __cplusplus
 }

ggml/src/CMakeLists.txt

@@ -208,8 +208,22 @@ if (GGML_BACKEND_DL)
     target_compile_definitions(ggml-base PUBLIC GGML_BACKEND_DL)
 endif()
 
+set(GGML_LIBRARY_SOURCES
+    ggml-backend-reg.cpp)
+
+if(BINDINGS_FLAT)
+    message(STATUS "Adding FLAT GGML binding extras")
+
+    set(FLAT_GGML_SOURCES
+       ggml-flat.cpp
+       )
+
+    list(APPEND GGML_LIBRARY_SOURCES ${FLAT_GGML_SOURCES})
+endif()
+
 add_library(ggml
-            ggml-backend-reg.cpp)
+            ${GGML_LIBRARY_SOURCES}
+            )
 
 target_link_libraries(ggml PUBLIC ggml-base)
 

ggml/src/ggml-backend-reg.cpp

@@ -584,3 +584,13 @@ void ggml_backend_load_all_from_path(const char * dir_path) {
         ggml_backend_load(backend_path);
     }
 }
+
+ggml_backend_reg_t ggml_backend_try_load_best(const char * name, const char * dir_path) {
+#ifdef NDEBUG
+    bool silent = true;
+#else
+    bool silent = false;
+#endif
+
+    return ggml_backend_load_best(name, silent, dir_path);
+}

ggml/src/ggml-flat.cpp

@@ -0,0 +1,35 @@
+
+#ifdef _WIN32
+#define WIN32_LEAN_AND_MEAN
+#ifndef NOMINMAX
+#   define NOMINMAX
+#endif
+#include <windows.h>
+#endif
+
+#include "ggml-backend.h"
+#include "ggml-backend-impl.h"
+#include "ggml-alloc.h"
+#include "ggml-impl.h"
+
+#include <assert.h>
+#include <limits.h>
+#include <stdarg.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <string>
+#include <vector>
+#include <algorithm>
+
+#ifdef __APPLE__
+#include <sys/types.h>
+#include <sys/sysctl.h>
+#endif
+
+#include "ggml-flat.h"
+
+
+
+
+

ggml/src/ggml-flat.h

@@ -0,0 +1,28 @@
+#pragma once
+
+#ifdef BINDINGS_FLAT
+#endif
+
+#ifdef GGML_SHARED
+#    if defined(_WIN32) && !defined(__MINGW32__)
+#        ifdef GGML_BUILD
+#            define GGML_FLAT_API __declspec(dllexport) extern
+#        else
+#            define GGML_FLAT_API __declspec(dllimport) extern
+#        endif
+#    else
+#        define GGML_FLAT_API __attribute__ ((visibility ("default"))) extern
+#    endif
+#else
+#    define GGML_FLAT_API extern
+#endif
+
+#ifdef  __cplusplus
+extern "C" {
+#endif
+
+
+
+#ifdef  __cplusplus
+}
+#endif

ggml/src/ggml-opencl/ggml-opencl.cpp

@@ -2,9 +2,11 @@
 #define CL_USE_DEPRECATED_OPENCL_1_2_APIS
 
 // suppress warnings in CL headers for GCC and Clang
-#pragma GCC diagnostic ignored "-Woverlength-strings"
-#ifdef __clang__
-#pragma GCC diagnostic ignored "-Wgnu-anonymous-struct"
+#ifndef _MSC_VER
+    #pragma GCC diagnostic ignored "-Woverlength-strings"
+    #ifdef __clang__
+    #pragma GCC diagnostic ignored "-Wgnu-anonymous-struct"
+    #endif
 #endif
 
 #include "ggml-opencl.h"

include/whisper.h

@@ -668,6 +668,15 @@ extern "C" {
     // Get the no_speech probability for the specified segment
     WHISPER_API float whisper_full_get_segment_no_speech_prob           (struct whisper_context * ctx, int i_segment);
     WHISPER_API float whisper_full_get_segment_no_speech_prob_from_state(struct whisper_state * state, int i_segment);
+
+    // For whisper-flat.cpp to expose
+    const char * whisper_get_system_info_json(void);
+    struct whisper_state * whisper_get_state_from_context(struct whisper_context * ctx);
+    struct whisper_timings * whisper_get_timings_with_state(struct whisper_state * state);
+    ggml_backend_t whisper_get_preferred_backend(struct whisper_state * state);
+    ggml_backend_t whisper_get_indexed_backend(struct whisper_state* state, size_t i);
+    size_t whisper_get_backend_count(struct whisper_state* state);
+
 #ifdef __cplusplus
 }
 #endif

src/CMakeLists.txt

@@ -100,10 +100,24 @@ endif()
 
 # whisper
 
+set(WHISPER_LIBRARY_SOURCES
+    ../include/whisper.h
+    whisper-arch.h
+    whisper.cpp
+    )
+
+if(BINDINGS_FLAT)
+    message(STATUS "Adding FLAT Whisper binding extras")
+
+    set(FLAT_WHISPER_SOURCES
+       whisper-flat.cpp
+       )
+
+    list(APPEND WHISPER_LIBRARY_SOURCES ${FLAT_WHISPER_SOURCES})
+endif()
+
 add_library(whisper
-            ../include/whisper.h
-            whisper-arch.h
-            whisper.cpp
+            ${WHISPER_LIBRARY_SOURCES}
             )
 
 # Set the version numbers

src/whisper-flat.cpp

@@ -0,0 +1,58 @@
+#include "whisper.h"
+#include "whisper-arch.h"
+
+#include "ggml.h"
+#include "ggml-cpp.h"
+#include "ggml-alloc.h"
+#include "ggml-backend.h"
+
+#include <atomic>
+#include <algorithm>
+#include <cassert>
+#define _USE_MATH_DEFINES
+#include <cmath>
+#include <climits>
+#include <codecvt>
+#include <cstdarg>
+#include <cstdio>
+#include <cstring>
+#include <fstream>
+#include <functional>
+#include <map>
+#include <mutex>
+#include <random>
+#include <regex>
+#include <set>
+#include <string>
+#include <thread>
+#include <vector>
+
+#include "whisper-flat.h"
+
+void whisper_flat_backend_load_all(void) {
+    ggml_backend_load_all();
+}
+
+const char * whisper_flat_get_system_info_json(void) {
+    return whisper_get_system_info_json();
+}
+
+struct whisper_state * whisper_flat_get_state_from_context(struct whisper_context * ctx) {
+    return whisper_get_state_from_context(ctx);
+}
+
+struct whisper_timings * whisper_flat_get_timings_with_state(struct whisper_state * state) {
+    return whisper_get_timings_with_state(state);
+}
+
+ggml_backend_t whisper_flat_get_preferred_backend(struct whisper_state * state) {
+    return whisper_get_preferred_backend(state);
+}
+
+ggml_backend_t whisper_flat_get_indexed_backend(struct whisper_state* state, size_t i) {
+    return whisper_get_indexed_backend(state, i);
+}
+
+size_t whisper_flat_get_backend_count(struct whisper_state* state) {
+    return whisper_get_backend_count(state);
+}