try fix build errors; minor cleanup

Author: mqy

CMakeLists.txt

@@ -158,11 +158,12 @@ if (LLAMA_BLAS)
     endif()
     set(BLA_VENDOR ${LLAMA_BLAS_VENDOR})
     find_package(BLAS)
-    if (BLAS_FOUND)
+    if (BLAS_FOUND) 
         message(STATUS "BLAS found, Libraries: ${BLAS_LIBRARIES}")
 
         add_compile_options(${BLAS_LINKER_FLAGS})
         add_compile_definitions(GGML_USE_OPENBLAS)
+        add_compile_definitions(GGML_BLAS_VENDOR="${LLAMA_BLAS_VENDOR}")
         set(LLAMA_EXTRA_LIBS ${LLAMA_EXTRA_LIBS} ${BLAS_LIBRARIES})
 
         message("${BLAS_LIBRARIES} ${BLAS_INCLUDE_DIRS}")
@@ -370,6 +371,8 @@ endif()
 add_library(ggml OBJECT
             ggml.c
             ggml.h
+            ggml-tune.c
+            ggml-tune.h
             ${GGML_CUDA_SOURCES}
             ${GGML_OPENCL_SOURCES})
 

Makefile

@@ -132,9 +132,8 @@ ifdef LLAMA_OPENBLAS
 		LDFLAGS += -lopenblas -L/usr/local/opt/openblas/lib
 	else
 		ifneq ($(shell grep -e "Ubuntu" -e "ID_LIKE=debian" /etc/os-release 2>/dev/null),)
-			# libopenblas64-pthread-dev on ubuntu.
-			CFLAGS  += -I/usr/include/x86_64-linux-gnu/openblas64-pthread
-			LDFLAGS += -lopenblas64 -L/usr/lib/x86_64-linux-gnu/openblas64-pthread
+			CFLAGS += $(pkg-config openblas --cflags)
+			LDFLAGS += $(pkg-config openblas --libs)
 		else
 			CFLAGS  += -I/usr/local/include/openblas -I/usr/include/openblas
 			ifneq ($(shell grep -e "Arch Linux" -e "ID_LIKE=arch" /etc/os-release 2>/dev/null),)
@@ -147,13 +146,8 @@ ifdef LLAMA_OPENBLAS
 endif
 ifdef LLAMA_BLIS
 	CFLAGS += -DGGML_USE_OPENBLAS -I/usr/local/include/blis -I/usr/include/blis
-	CFLAGS += -DGGML_BLAS_VENDOR_BLIS
+	CFLAGS += -DGGML_BLAS_VENDOR="\"FLAME\""
 	LDFLAGS += -lblis -L/usr/local/lib
-	ifeq ($(UNAME_S),Darwin)
-		# blis installed with Homebew on macOS.
-		CFLAGS  += -I/usr/local/opt/blis/include
-		LDFLAGS += -L/usr/local/opt/blis/lib
-	endif
 endif
 ifdef LLAMA_CUBLAS
 	CFLAGS    += -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I/opt/cuda/include -I$(CUDA_PATH)/targets/x86_64-linux/include

examples/CMakeLists.txt

@@ -37,6 +37,7 @@ else()
     add_subdirectory(save-load-state)
     add_subdirectory(benchmark)
     add_subdirectory(baby-llama)
+    add_subdirectory(mulmat-tune)
     if(LLAMA_BUILD_SERVER)
         add_subdirectory(server)
     endif()

examples/mulmat-tune/CMakeLists.txt

@@ -0,0 +1,7 @@
+set(TARGET mulmat-tune)
+add_executable(${TARGET} mulmat-tune.c)
+target_link_libraries(${TARGET} PRIVATE ggml ${CMAKE_THREAD_LIBS_INIT})
+target_compile_features(${TARGET} PRIVATE c_std_11)
+if(TARGET BUILD_INFO)
+  add_dependencies(${TARGET} BUILD_INFO)
+endif()

examples/mulmat-tune/mulmat-tune.c

@@ -11,7 +11,7 @@
 #define UNUSED(x) (void)(x)
 
 static int tune_time_min(const int *a, int len);
-static void print_blas_build_tips(void);
+static void print_build_tips(void);
 static void progress(int i, int n);
 static bool prompt_yes_no(const char *prompt);
 
@@ -52,7 +52,7 @@ static void usage(char *prog) {
 
     printf("\n");
 
-    print_blas_build_tips();
+    print_build_tips();
 }
 
 // main
@@ -62,7 +62,7 @@ int main(int argc, char **argv) {
     int n_builtin_backends = ggml_get_builtin_backends(builtin_backends);
     if (n_builtin_backends == 0) {
         fprintf(stderr, "ERROR: this program was not built with any BLAS.\n");
-        print_blas_build_tips();
+        print_build_tips();
         exit(1);
     }
 
@@ -353,7 +353,7 @@ void cmd_tune(struct ggml_mulmat_tune *tune, int n_pass, bool verbose) {
     void *q_buf = NULL;
     void *wdata = NULL;
 
-    // alloc q4_0_buf and wdata with max size.
+    // alloc q_buf and wdata with max size.
     {
         int max_NxK = 0;
         for (int i = 0; i < tune->n_shapes; i++) {
@@ -532,7 +532,7 @@ void cmd_tune(struct ggml_mulmat_tune *tune, int n_pass, bool verbose) {
     free(q_buf);
 }
 
-static void print_blas_build_tips(void) {
+static void print_build_tips(void) {
     const char *a = "LLAMA_NO_ACCELERATE";
     fprintf(stderr, "Tips on how to build with various backend vendors:\n\n");
     fprintf(stderr, "CUDA:       make clean; LLAMA_CUBLAS=1 make\n");

ggml-tune.c

@@ -416,17 +416,15 @@ const char *ggml_get_backend_vendor(void) {
     const char *vendor = NULL;
 #if defined(GGML_USE_CUBLAS)
     vendor = "CUBLAS";
-#endif
-
-#if defined(GGML_USE_CLBLAST)
+#elif defined(GGML_USE_CLBLAST)
     vendor = "CLBLAST";
 #endif
 
 #if defined(GGML_USE_ACCELERATE)
     vendor = "ACCELERATE";
 #elif defined(GGML_USE_OPENBLAS)
-#if defined(GGML_BLAS_VENDOR_BLIS)
-    vendor = "BLIS";
+#if defined(GGML_BLAS_VENDOR)
+    vendor = GGML_BLAS_VENDOR;
 #else
     vendor = "OPENBLAS";
 #endif

ggml-tune.h

@@ -71,9 +71,6 @@ struct ggml_mulmat_tune_profile_time {
     int total_time;
 };
 
-void ggml_task_conf_format_name(const struct ggml_task_profile *profile,
-                                char *buf, int buf_len);
-
 int ggml_mulmat_tune_validate(struct ggml_mulmat_tune *tune,
                               const char *model_name,
                               /*enum ggml_type*/ int type);

tests/CMakeLists.txt

@@ -12,4 +12,4 @@ llama_add_test(test-sampling.cpp)
 llama_add_test(test-tokenizer-0.cpp ${CMAKE_CURRENT_SOURCE_DIR}/../models/ggml-vocab.bin)
 # llama_add_test(test-grad0.c) # SLOW
 # llama_add_test(test-opt.c) # SLOW
-# llama_add_test(test-mulmat-tune.c)
+llama_add_test(test-mulmat-tune.c)