build : disable lto for C++ (make) and enable existing LTO flag (cmake)

Author: ggerganov

CMakeLists.txt

@@ -45,7 +45,7 @@ endif()
 # general
 option(LLAMA_STATIC                     "llama: static link libraries"                          OFF)
 option(LLAMA_NATIVE                     "llama: enable -march=native flag"                      ON)
-option(LLAMA_LTO                        "llama: enable link time optimization"                  OFF)
+option(LLAMA_LTO                        "llama: enable link time optimization"                  ON)
 
 # debug
 option(LLAMA_ALL_WARNINGS               "llama: enable all compiler warnings"                   ON)
@@ -676,8 +676,6 @@ add_library(ggml OBJECT
             ${GGML_SOURCES_EXTRA} ${GGML_HEADERS_EXTRA}
             )
 
-set_property(TARGET ggml PROPERTY INTERPROCEDURAL_OPTIMIZATION TRUE)
-
 target_include_directories(ggml PUBLIC . ${LLAMA_EXTRA_INCLUDES})
 target_compile_features(ggml PUBLIC c_std_11) # don't bump
 target_link_libraries(ggml PUBLIC Threads::Threads ${LLAMA_EXTRA_LIBS})

Makefile

@@ -120,12 +120,12 @@ MK_CXXFLAGS = -std=c++11 -fPIC
 
 # -Ofast tends to produce faster code, but may not be available for some compilers.
 ifdef LLAMA_FAST
-MK_CFLAGS        += -flto -Ofast
-MK_HOST_CXXFLAGS += -flto -Ofast
-MK_CUDA_CXXFLAGS += -flto -O3
+MK_CFLAGS        += -Ofast -flto
+MK_HOST_CXXFLAGS += -Ofast
+MK_CUDA_CXXFLAGS += -O3
 else
-MK_CFLAGS        += -flto -O3
-MK_CXXFLAGS      += -flto -O3
+MK_CFLAGS        += -O3 -flto
+MK_CXXFLAGS      += -O3
 endif
 
 # clock_gettime came in POSIX.1b (1993)