Flat fix

CMakeLists.txt

@@ -70,6 +70,10 @@ 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)
+option(GGML_BINDINGS_FLAT "Add extra flat definitions to Examples" ${BINDINGS_FLAT})
+
 # sanitizers
 option(WHISPER_SANITIZE_THREAD    "whisper: enable thread sanitizer"    OFF)
 option(WHISPER_SANITIZE_ADDRESS   "whisper: enable address sanitizer"   OFF)
@@ -245,4 +249,4 @@ if (MSVC)
         disable_msvc_warnings(whisper-bench)
         disable_msvc_warnings(quantize)
     endif()
-endif()
+endif()

bindings/ruby/ext/options.rb

@@ -61,6 +61,7 @@ def extra_options
 
   def configure
     filepath "ACCELERATE_FRAMEWORK"
+    ignored "BINDINGS_FLAT"
     ignored "BUILD_SHARED_LIBS"
     ignored "BUILD_TESTING"
     ignored "CMAKE_BUILD_TYPE"
@@ -83,6 +84,7 @@ def configure
     bool "GGML_AVX_VNNI"
     ignored "GGML_BACKEND_DL"
     ignored "GGML_BIN_INSTALL_DIR"
+    ignored "GGML_BINDINGS_FLAT"
     bool "GGML_BLAS"
     string "GGML_BLAS_VENDOR"
     bool "GGML_BMI2"

examples/CMakeLists.txt

@@ -91,6 +91,12 @@ target_include_directories(json_cpp INTERFACE ${CMAKE_CURRENT_SOURCE_DIR})
 
 # examples
 
+if(BINDINGS_FLAT)
+    message(STATUS "Adding FLAT binding to examples")
+    add_compile_options(-DBINDINGS_FLAT)
+endif()
+
+
 include_directories(${CMAKE_CURRENT_SOURCE_DIR})
 
 if (EMSCRIPTEN)

examples/bench/bench.cpp

@@ -1,4 +1,8 @@
 #include "whisper.h"
+#ifdef BINDINGS_FLAT
+#include "whisper-flat.h"
+#include "../ggml/src/ggml-flat.h"
+#endif
 
 #include <cstdio>
 #include <cstring>
@@ -61,6 +65,15 @@ void whisper_print_usage(int /*argc*/, char ** argv, const whisper_params & para
 static int whisper_bench_full(const whisper_params & params) {
     // whisper init
 
+    #ifdef BINDINGS_FLAT
+    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();
 
     cparams.use_gpu    = params.use_gpu;

examples/cli/cli.cpp

@@ -2,6 +2,10 @@
 #include "common-whisper.h"
 
 #include "whisper.h"
+#ifdef BINDINGS_FLAT
+#include "whisper-flat.h"
+#include "../ggml/src/ggml-flat.h"
+#endif
 #include "grammar-parser.h"
 
 #include <cmath>
@@ -935,6 +939,15 @@ int main(int argc, char ** argv) {
 
     // whisper init
 
+    #ifdef BINDINGS_FLAT
+    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();
 
     cparams.use_gpu    = params.use_gpu;

examples/command/command.cpp

@@ -9,6 +9,10 @@
 #include "common-sdl.h"
 #include "common.h"
 #include "whisper.h"
+#ifdef BINDINGS_FLAT
+#include "whisper-flat.h"
+#include "../ggml/src/ggml-flat.h"
+#endif
 #include "grammar-parser.h"
 
 #include <algorithm>
@@ -692,6 +696,15 @@ int main(int argc, char ** argv) {
 
     // whisper init
 
+    #ifdef BINDINGS_FLAT
+    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();
 
     cparams.use_gpu    = params.use_gpu;

examples/server/server.cpp

@@ -2,6 +2,9 @@
 #include "common-whisper.h"
 
 #include "whisper.h"
+#ifdef BINDINGS_FLAT
+#include "whisper-flat.h"
+#endif
 #include "httplib.h"
 #include "json.hpp"
 
@@ -541,6 +544,11 @@ int main(int argc, char ** argv) {
         check_ffmpeg_availibility();
     }
     // whisper init
+    #ifdef BINDINGS_FLAT
+    fprintf(stderr, "+++ BINDINGS_FLAT +++\n");
+    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 +6,10 @@
 #include "common.h"
 #include "common-whisper.h"
 #include "whisper.h"
+#ifdef BINDINGS_FLAT
+#include "whisper-flat.h"
+#include "../ggml/src/ggml-flat.h"
+#endif
 
 #include <chrono>
 #include <cstdio>
@@ -155,6 +159,15 @@ int main(int argc, char ** argv) {
         exit(0);
     }
 
+    #ifdef BINDINGS_FLAT
+    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();
 
     cparams.use_gpu    = params.use_gpu;

examples/talk-llama/talk-llama.cpp

@@ -5,6 +5,10 @@
 #include "common.h"
 #include "common-whisper.h"
 #include "whisper.h"
+#ifdef BINDINGS_FLAT
+#include "whisper-flat.h"
+#include "../ggml/src/ggml-flat.h"
+#endif
 #include "llama.h"
 
 #include <chrono>
@@ -287,6 +291,15 @@ int main(int argc, char ** argv) {
 
     // whisper init
 
+    #ifdef BINDINGS_FLAT
+    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();
 
     cparams.use_gpu    = params.use_gpu;

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

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

ggml/CMakeLists.txt

@@ -385,4 +385,4 @@ if (MSVC)
     disable_msvc_warnings(ggml-cpu-skylakex)
     disable_msvc_warnings(ggml-cpu-icelake)
     disable_msvc_warnings(ggml-cpu-alderlake)
-endif()
+endif()

ggml/include/ggml-backend.h

@@ -348,6 +348,7 @@ 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);
+
 
 #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})
+    add_compile_options(-DGGML_BINDINGS_FLAT)
+endif()
+
 add_library(ggml
-            ggml-backend-reg.cpp)
+            ${GGML_LIBRARY_SOURCES}
+            )
 
 target_link_libraries(ggml PUBLIC ggml-base)
 

ggml/src/ggml-backend-reg.cpp

@@ -24,6 +24,10 @@
 #    include <unistd.h>
 #endif
 
+#ifdef GGML_BINDINGS_FLAT
+#include "ggml-flat.h"
+#endif
+
 // Backend registry
 #ifdef GGML_USE_CPU
 #include "ggml-cpu.h"
@@ -584,3 +588,11 @@ void ggml_backend_load_all_from_path(const char * dir_path) {
         ggml_backend_load(backend_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, "%s: (%s)\n", __func__, name);
+    return ggml_backend_load_best(name, silent, dir_path);
+}
+#endif

ggml/src/ggml-flat.cpp

@@ -0,0 +1,38 @@
+
+#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"
+
+#ifdef GGML_BINDINGS_FLAT
+//    ggml_backend_reg_t ggml_backend_try_load_best(const char * name, const char * dir_path);
+
+#endif
+
+
+

ggml/src/ggml-flat.h

@@ -0,0 +1,31 @@
+#pragma once
+
+#ifdef BINDINGS_FLAT
+#define GGML_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 GGML_BINDINGS_FLAT
+    GGML_FLAT_API ggml_backend_reg_t ggml_backend_try_load_best(const char * name, const char * dir_path);
+    #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"