Merge branch 'mmnga-dev' into mmnga-dev-llamacpp

Author: mmngays

CMakeLists.txt

@@ -83,6 +83,7 @@ set(LLAMA_CUDA_KQUANTS_ITER "2" CACHE STRING "llama: iters./thread per block for
 option(LLAMA_HIPBLAS                         "llama: use hipBLAS"                               OFF)
 option(LLAMA_CLBLAST                         "llama: use CLBlast"                               OFF)
 option(LLAMA_METAL                           "llama: use Metal"                                 ${LLAMA_METAL_DEFAULT})
+option(LLAMA_METAL_NDEBUG                    "llama: disable Metal debugging"                   OFF)
 option(LLAMA_MPI                             "llama: use MPI"                                   OFF)
 option(LLAMA_K_QUANTS                        "llama: use k-quants"                              ON)
 option(LLAMA_QKK_64                          "llama: use super-block size of 64 for k-quants"   OFF)
@@ -174,7 +175,9 @@ if (LLAMA_METAL)
     set(GGML_SOURCES_METAL ggml-metal.m ggml-metal.h)
 
     add_compile_definitions(GGML_USE_METAL)
-    #add_compile_definitions(GGML_METAL_NDEBUG)
+    if (LLAMA_METAL_NDEBUG)
+        add_compile_definitions(GGML_METAL_NDEBUG)
+    endif()
 
     # get full path to the file
     #add_compile_definitions(GGML_METAL_DIR_KERNELS="${CMAKE_CURRENT_SOURCE_DIR}/")

Makefile

@@ -109,12 +109,11 @@ endif
 
 
 ifdef LLAMA_CODE_COVERAGE
-	CXXFLAGS += -fprofile-arcs -ftest-coverage -dumpbase ''
+	MK_CXXFLAGS += -fprofile-arcs -ftest-coverage -dumpbase ''
 endif
 
 ifdef LLAMA_DISABLE_LOGS
-	CFLAGS   += -DLOG_DISABLE_LOGS
-	CXXFLAGS += -DLOG_DISABLE_LOGS
+	MK_CPPFLAGS += -DLOG_DISABLE_LOGS
 endif # LLAMA_DISABLE_LOGS
 
 # warnings
@@ -124,7 +123,7 @@ MK_CXXFLAGS  += -Wall -Wextra -Wpedantic -Wcast-qual -Wno-unused-function -Wno-m
 
 ifeq '' '$(findstring clang++,$(CXX))'
 	# g++ only
-	CXXFLAGS += -Wno-format-truncation
+	MK_CXXFLAGS += -Wno-format-truncation
 endif
 
 # OS specific
@@ -188,8 +187,8 @@ endif
 # https://gcc.gnu.org/bugzilla/show_bug.cgi?id=54412
 # https://github.com/ggerganov/llama.cpp/issues/2922
 ifneq '' '$(findstring mingw,$(shell $(CC) -dumpmachine))'
-	CFLAGS   += -Xassembler -muse-unaligned-vector-move
-	CXXFLAGS += -Xassembler -muse-unaligned-vector-move
+	MK_CFLAGS   += -Xassembler -muse-unaligned-vector-move
+	MK_CXXFLAGS += -Xassembler -muse-unaligned-vector-move
 endif
 
 ifneq ($(filter aarch64%,$(UNAME_M)),)
@@ -226,8 +225,8 @@ ifneq ($(filter ppc64%,$(UNAME_M)),)
 endif
 
 else
-	CFLAGS += -march=rv64gcv -mabi=lp64d
-	CXXFLAGS +=  -march=rv64gcv -mabi=lp64d
+	MK_CFLAGS   += -march=rv64gcv -mabi=lp64d
+	MK_CXXFLAGS += -march=rv64gcv -mabi=lp64d
 endif
 
 ifndef LLAMA_NO_K_QUANTS
@@ -247,16 +246,6 @@ ifndef LLAMA_NO_ACCELERATE
 	endif
 endif # LLAMA_NO_ACCELERATE
 
-ifdef LLAMA_METAL
-	# By default - use GPU acceleration on Mac OS
-	ifeq ($(UNAME_S),Darwin)
-		CFLAGS   += -DGGML_USE_METAL #-DGGML_METAL_NDEBUG
-		CXXFLAGS += -DGGML_USE_METAL
-		LDFLAGS  += -framework Foundation -framework Metal -framework MetalKit
-		OBJS     += ggml-metal.o
-	endif
-endif # LLAMA_METAL
-
 ifdef LLAMA_MPI
 	MK_CPPFLAGS += -DGGML_USE_MPI
 	MK_CFLAGS   += -Wno-cast-qual
@@ -368,9 +357,12 @@ ggml-cuda.o: ggml-cuda.cu ggml-cuda.h
 endif # LLAMA_HIPBLAS
 
 ifdef LLAMA_METAL
-	MK_CPPFLAGS += -DGGML_USE_METAL #-DGGML_METAL_NDEBUG
+	MK_CPPFLAGS += -DGGML_USE_METAL
 	MK_LDFLAGS  += -framework Foundation -framework Metal -framework MetalKit
 	OBJS		+= ggml-metal.o
+ifdef LLAMA_METAL_NDEBUG
+	MK_CPPFLAGS += -DGGML_METAL_NDEBUG
+endif
 endif # LLAMA_METAL
 
 ifdef LLAMA_METAL

common/common.cpp

@@ -513,16 +513,16 @@ inline bool log_param_pair_parse(bool check_but_dont_parse, const std::string &
 
 inline void log_print_usage()
 {
-    fprintf(stdout, "log options:\n");
+    printf("log options:\n");
     /* format
-    fprintf(stdout, "  -h, --help            show this help message and exit\n");*/
+    printf("  -h, --help            show this help message and exit\n");*/
     /* spacing
-    fprintf(stdout, "__-param----------------Description\n");*/
-    fprintf(stdout, "  --log-test            Run simple logging test\n");
-    fprintf(stdout, "  --log-disable         Disable trace logs\n");
-    fprintf(stdout, "  --log-enable          Enable trace logs\n");
-    fprintf(stdout, "  --log-file            Specify a log filename (without extension)\n");
-    fprintf(stdout, "                        Log file will be tagged with unique ID and written as \"<name>.<ID>.log\"\n"); /*  */
+    printf("__-param----------------Description\n");*/
+    printf("  --log-test            Run simple logging test\n");
+    printf("  --log-disable         Disable trace logs\n");
+    printf("  --log-enable          Enable trace logs\n");
+    printf("  --log-file            Specify a log filename (without extension)\n");
+    printf("                        Log file will be tagged with unique ID and written as \"<name>.<ID>.log\"\n"); /*  */
 }
 
 #define log_dump_cmdline(argc, argv) log_dump_cmdline_impl(argc, argv)