SWDEV-1 - Updated public llamacpp with amdchat changes

Updated public llamacpp with amdchat changes patches

Author: Ding, Kevin

CMakeLists.txt

@@ -46,6 +46,8 @@ else()
 endif()
 
 option(BUILD_SHARED_LIBS "build shared libraries" ${BUILD_SHARED_LIBS_DEFAULT})
+option(AMD_CHAT_SERVER_DLL "amd: amd chat server dll" OFF)
+option(AMD_CHAT_SERVER_HTTP "amd: amd chat server http server" OFF)
 
 if (WIN32)
     add_compile_definitions(_CRT_SECURE_NO_WARNINGS)

common/log.cpp

@@ -6,9 +6,12 @@
 #include <cstdio>
 #include <mutex>
 #include <sstream>
+#include <fstream>
 #include <thread>
 #include <vector>
 
+bool use_amd_log = false;
+
 int common_log_verbosity_thold = LOG_DEFAULT_LLAMA;
 
 void common_log_set_verbosity_thold(int verbosity) {
@@ -372,7 +375,14 @@ void common_log_free(struct common_log * log) {
 void common_log_add(struct common_log * log, enum ggml_log_level level, const char * fmt, ...) {
     va_list args;
     va_start(args, fmt);
-    log->add(level, fmt, args);
+    if (use_amd_log)
+    {
+        amd_log_add(level, fmt, args);
+    }
+    else
+    {
+        log->add(level, fmt, args);
+    }
     va_end(args);
 }
 
@@ -391,3 +401,65 @@ void common_log_set_prefix(struct common_log * log, bool prefix) {
 void common_log_set_timestamps(struct common_log * log, bool timestamps) {
     log->set_timestamps(timestamps);
 }
+
+std::stringstream amd_server_log;
+std::mutex amd_server_log_mutex;
+
+void amd_log_set()
+{
+    use_amd_log = true;
+}
+
+void amd_log_add(enum ggml_log_level level, const char* text) {
+    // if (level >= GGML_LOG_LEVEL_DEBUG)
+    //     return;
+    if (level == GGML_LOG_LEVEL_DEBUG && common_log_verbosity_thold < LOG_DEFAULT_DEBUG) {
+        return;
+    }
+
+    std::unique_lock<std::mutex> lock(amd_server_log_mutex);
+
+    amd_server_log << text;
+}
+
+
+void amd_log_add(enum ggml_log_level level, const char* fmt, va_list args) {
+    // if (level >= GGML_LOG_LEVEL_DEBUG)
+    //     return;
+    if (level == GGML_LOG_LEVEL_DEBUG && common_log_verbosity_thold < LOG_DEFAULT_DEBUG) {
+        return;
+    }
+
+    std::unique_lock<std::mutex> lock(amd_server_log_mutex);
+
+    std::vector<char> msg;
+
+    va_list args_copy;
+    va_copy(args_copy, args);
+    const size_t n = vsnprintf(msg.data(), msg.size(), fmt, args);
+    if (n >= msg.size()) {
+        msg.resize(n + 1);
+        vsnprintf(msg.data(), msg.size(), fmt, args_copy);
+    }
+
+    amd_server_log << msg.data();
+}
+
+void amd_log_flush_to_file(char* log_file)
+{
+    std::unique_lock<std::mutex> lock(amd_server_log_mutex);
+
+    if (NULL != log_file)
+    {
+        std::ofstream log(log_file, std::ios::app);
+        log << amd_server_log.str();
+    }
+
+    amd_server_log.str("");
+}
+
+void amd_llama_log_callback(ggml_log_level level, const char* text, void* user_data) {
+    (void)user_data;
+    amd_log_add(level, text);
+}
+

common/log.h

@@ -24,6 +24,8 @@
 #define LOG_DEFAULT_DEBUG 1
 #define LOG_DEFAULT_LLAMA 0
 
+extern bool use_amd_log;
+
 // needed by the LOG_TMPL macro to avoid computing log arguments if the verbosity lower
 // set via common_log_set_verbosity()
 extern int common_log_verbosity_thold;
@@ -80,6 +82,12 @@ void common_log_set_timestamps(struct common_log * log,       bool   timestamps)
 // this will avoid calling expensive_function() if LOG_DEFAULT_DEBUG > common_log_verbosity_thold
 //
 
+void amd_log_set();
+void amd_log_add(enum ggml_log_level level, const char* text);
+void amd_log_add(enum ggml_log_level level, const char* fmt, va_list args);
+void amd_log_flush_to_file(char* log_file);
+void amd_llama_log_callback(ggml_log_level level, const char* text, void* user_data);
+
 #define LOG_TMPL(level, verbosity, ...) \
     do { \
         if ((verbosity) <= common_log_verbosity_thold) { \

tools/server/CMakeLists.txt

@@ -1,4 +1,8 @@
+if (AMD_CHAT_SERVER_DLL)
+set(TARGET server)
+else()
 set(TARGET llama-server)
+endif()
 
 option(LLAMA_SERVER_SSL "Build SSL support for the server" OFF)
 
@@ -9,9 +13,14 @@ if (MINGW)
     add_compile_definitions(_WIN32_WINNT=${GGML_WIN_VER})
 endif()
 
+if (AMD_CHAT_SERVER_DLL)
+    message(STATUS "***AMD*** building server.dll")
+endif()
+
 set(TARGET_SRCS
     server.cpp
     utils.hpp
+    dllmain.cpp
 )
 set(PUBLIC_ASSETS
     index.html.gz
@@ -30,7 +39,11 @@ foreach(asset ${PUBLIC_ASSETS})
     set_source_files_properties(${output} PROPERTIES GENERATED TRUE)
 endforeach()
 
+if (AMD_CHAT_SERVER_DLL)
+add_library(${TARGET} SHARED ${TARGET_SRCS})
+else()
 add_executable(${TARGET} ${TARGET_SRCS})
+endif()
 install(TARGETS ${TARGET} RUNTIME)
 
 target_include_directories(${TARGET} PRIVATE ../mtmd)

tools/server/dllmain.cpp

@@ -0,0 +1,18 @@
+#include "Windows.h"
+
+BOOL APIENTRY DllMain( HMODULE hModule,
+                       DWORD  ul_reason_for_call,
+                       LPVOID lpReserved
+                     )
+{
+    switch (ul_reason_for_call)
+    {
+    case DLL_PROCESS_ATTACH:
+    case DLL_THREAD_ATTACH:
+    case DLL_THREAD_DETACH:
+    case DLL_PROCESS_DETACH:
+        break;
+    }
+    return TRUE;
+}
+