@FIR-702 - llama.cpp: Sync with latest opensource

.gitmodules

@@ -1,3 +1,6 @@
 [submodule "kompute"]
 	path = ggml/src/ggml-kompute/kompute
 	url = https://github.com/nomic-ai/kompute.git
+[submodule "ggml-tsi-kernel"]
+	path = ggml-tsi-kernel
+	url = [email protected]:tsisw/ggml-tsi-kernel.git

CMakeLists.txt

@@ -5,10 +5,59 @@ include(CheckIncludeFileCXX)
 #set(CMAKE_WARN_DEPRECATED YES)
 set(CMAKE_WARN_UNUSED_CLI YES)
 
+if (GGML_TSAVORITE)
+    if  (NOT DEFINED GGML_TSAVORITE_TARGET)
+        set(GGML_TSAVORITE_TARGET "posix")
+    endif()
+    if (NOT ${GGML_TSAVORITE_TARGET} STREQUAL fpga)
+        set(GGML_TSAVORITE_TARGET "posix")
+    endif()
+
+    if (NOT DEFINED MLIR_COMPILER_DIR)
+        if (NOT DEFINED $ENV{MLIR_SDK_VERSION})
+            set (MLIR_COMPILER_DIR /proj/work/rel/sw/sdk-r.0.1.0/compiler)
+        else()
+            set (MLIR_COMPILER_DIR $ENV{MLIR_SDK_VERSION}/compiler)
+        endif()
+    endif()
+
+    if (NOT DEFINED RUNTIME_DIR)
+        if (NOT DEFINED $ENV{MLIR_SDK_VERSION})
+            set (RUNTIME_DIR /proj/work/rel/sw/sdk-r.0.1.0/${GGML_TSAVORITE_TARGET}/runtime)
+        else()
+            set (RUNTIME_DIR $ENV{MLIR_SDK_VERSION}/${GGML_TSAVORITE_TARGET}/runtime)
+        endif()
+    endif()
+
+    if (NOT DEFINED GGML_TSI_KERNEL_DIR)
+       set (GGML_TSI_KERNEL_DIR ${CMAKE_SOURCE_DIR}/ggml-tsi-kernel/${GGML_TSAVORITE_TARGET})
+    endif()
+
+    file(GLOB TLIBS  "${RUNTIME_DIR}/lib/*.so" "${GGML_TSI_KERNEL_DIR}/host/*.o")
+
+    if (${GGML_TSAVORITE_TARGET} STREQUAL fpga)
+        set(CMAKE_CROSSCOMPILING ON)
+        set(ARCH_FLAGS -march=armv8-a)
+        message("Setting target as fpga")
+    elseif (${GGML_TSAVORITE_TARGET} STREQUAL "posix")
+	list(APPEND TLIBS "${MLIR_COMPILER_DIR}/lib/libFFMDeviceShim.so")
+        message("Setting target as posix for tsavorite")
+    endif()
+
+    set(GGML_TSAVORITE_TARGET "${GGML_TSAVORITE_TARGET}" CACHE STRING "Target for tsavorite")
+    set (TSAVORITE_INCLUDE_DIR ${CMAKE_SOURCE_DIR}/ggml/src/ggml-tsavorite/include)
+
+    include_directories(${TSAVORITE_INCLUDE_DIR})
+    include_directories(${MLIR_COMPILER_DIR}/include/runtime/shim)
+    include_directories(${RUNTIME_DIR}/include)
+    message("tsavorite backend is enabled")
+endif()
+
 set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
 
 if (NOT XCODE AND NOT MSVC AND NOT CMAKE_BUILD_TYPE)
-    set(CMAKE_BUILD_TYPE Release CACHE STRING "Build type" FORCE)
+    #set(CMAKE_BUILD_TYPE Release CACHE STRING "Build type" FORCE)
+    set(CMAKE_BUILD_TYPE Debug CACHE STRING "Build type" FORCE)
     set_property(CACHE CMAKE_BUILD_TYPE PROPERTY STRINGS "Debug" "Release" "MinSizeRel" "RelWithDebInfo")
 endif()
 
@@ -82,9 +131,18 @@ option(LLAMA_BUILD_EXAMPLES "llama: build examples"       ${LLAMA_STANDALONE})
 option(LLAMA_BUILD_SERVER   "llama: build server example" ${LLAMA_STANDALONE})
 
 # 3rd party libs
-option(LLAMA_CURL       "llama: use libcurl to download model from an URL" ON)
 option(LLAMA_LLGUIDANCE "llama-common: include LLGuidance library for structured output in common utils" OFF)
 
+if (GGML_TSAVORITE)
+    if (${GGML_TSAVORITE_TARGET} STREQUAL fpga)
+        option(LLAMA_CURL       "llama: use libcurl to download model from an URL" OFF)
+    else()
+        option(LLAMA_CURL       "llama: use libcurl to download model from an URL" ON)
+    endif()
+else()
+    option(LLAMA_CURL       "llama: use libcurl to download model from an URL" ON)
+endif()
+
 # Required for relocatable CMake package
 include(${CMAKE_CURRENT_SOURCE_DIR}/cmake/build-info.cmake)
 include(${CMAKE_CURRENT_SOURCE_DIR}/cmake/common.cmake)

common/CMakeLists.txt

@@ -145,8 +145,16 @@ endif ()
 
 target_include_directories(${TARGET} PUBLIC .)
 target_compile_features   (${TARGET} PUBLIC cxx_std_17)
-target_link_libraries     (${TARGET} PRIVATE ${LLAMA_COMMON_EXTRA_LIBS} PUBLIC llama Threads::Threads)
 
+if (GGML_TSAVORITE)
+    if (${GGML_TSAVORITE_TARGET} STREQUAL fpga)
+        target_link_libraries     (${TARGET} PRIVATE ${LLAMA_COMMON_EXTRA_LIBS} ${TLIBS} PUBLIC llama Threads::Threads)
+    else()
+        target_link_libraries     (${TARGET} PRIVATE ${LLAMA_COMMON_EXTRA_LIBS} PUBLIC llama Threads::Threads)
+    endif()
+else()
+    target_link_libraries     (${TARGET} PRIVATE ${LLAMA_COMMON_EXTRA_LIBS} PUBLIC llama Threads::Threads)
+endif()
 
 #
 # copy the license files

examples/gguf-hash/CMakeLists.txt

@@ -1,5 +1,6 @@
 set(TARGET llama-gguf-hash)
 add_executable(${TARGET} gguf-hash.cpp)
+target_link_libraries(${TARGET} PRIVATE ${TLIBS})
 install(TARGETS ${TARGET} RUNTIME)
 
 # clibs dependencies

examples/gguf/CMakeLists.txt

@@ -1,5 +1,5 @@
 set(TARGET llama-gguf)
 add_executable(${TARGET} gguf.cpp)
 install(TARGETS ${TARGET} RUNTIME)
-target_link_libraries(${TARGET} PRIVATE ggml ${CMAKE_THREAD_LIBS_INIT})
+target_link_libraries(${TARGET} PRIVATE ggml ${CMAKE_THREAD_LIBS_INIT} ${TLIBS})
 target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/lookup/CMakeLists.txt

@@ -1,23 +1,23 @@
 set(TARGET llama-lookup)
 add_executable(${TARGET} lookup.cpp)
 install(TARGETS ${TARGET} RUNTIME)
-target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
+target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT} ${TLIBS})
 target_compile_features(${TARGET} PRIVATE cxx_std_17)
 
 set(TARGET llama-lookup-create)
 add_executable(${TARGET} lookup-create.cpp)
 install(TARGETS ${TARGET} RUNTIME)
-target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
+target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT} ${TLIBS})
 target_compile_features(${TARGET} PRIVATE cxx_std_17)
 
 set(TARGET llama-lookup-merge)
 add_executable(${TARGET} lookup-merge.cpp)
 install(TARGETS ${TARGET} RUNTIME)
-target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
+target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT} ${TLIBS})
 target_compile_features(${TARGET} PRIVATE cxx_std_17)
 
 set(TARGET llama-lookup-stats)
 add_executable(${TARGET} lookup-stats.cpp)
 install(TARGETS ${TARGET} RUNTIME)
-target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT})
+target_link_libraries(${TARGET} PRIVATE common llama ${CMAKE_THREAD_LIBS_INIT} ${TLIBS})
 target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/simple-chat/CMakeLists.txt

@@ -1,5 +1,5 @@
 set(TARGET llama-simple-chat)
 add_executable(${TARGET} simple-chat.cpp)
 install(TARGETS ${TARGET} RUNTIME)
-target_link_libraries(${TARGET} PRIVATE llama ${CMAKE_THREAD_LIBS_INIT})
+target_link_libraries(${TARGET} PRIVATE llama ${CMAKE_THREAD_LIBS_INIT} ${TLIBS})
 target_compile_features(${TARGET} PRIVATE cxx_std_17)

examples/simple/CMakeLists.txt

@@ -1,5 +1,24 @@
+#
+# simple-ctx
 set(TARGET llama-simple)
 add_executable(${TARGET} simple.cpp)
 install(TARGETS ${TARGET} RUNTIME)
-target_link_libraries(${TARGET} PRIVATE llama ${CMAKE_THREAD_LIBS_INIT})
+target_link_libraries(${TARGET} PRIVATE llama ${TLIBS} ${CMAKE_THREAD_LIBS_INIT})
 target_compile_features(${TARGET} PRIVATE cxx_std_17)
+
+#
+if (GGML_TSAVORITE)
+    #
+    # tsavorite backend test cases
+    if (${GGML_TSAVORITE_TARGET} STREQUAL fpga)
+        file(GLOB TLIBS  "${RUNTIME_DIR}/lib/*.so" "../../${GGML_TSI_KERNEL_DIR}/host/*.o")
+    else()
+        file(GLOB TLIBS  "${RUNTIME_DIR}/lib/*.so" "${MLIR_COMPILER_DIR}/lib/libFFMDeviceShim.so" "../../${GGML_TSI_KERNEL_DIR}/host/*.o")
+    endif()
+    #
+    # simple-backend-tsi
+
+    set(TEST_TARGET simple-backend-tsi)
+    add_executable(${TEST_TARGET} simple-backend-tsi.cpp)
+    target_link_libraries(${TEST_TARGET} PRIVATE ggml ${TLIBS}  dl rt)
+endif()