ggml-hexagon: merge build logic in kernels/Makefile to ggml-hexagon/CMakeLists.txt and remove Makefile

Author: zhouwg

ggml/src/ggml-hexagon/CMakeLists.txt

@@ -17,6 +17,22 @@ message("QNN_SDK_PATH    : ${QNN_SDK_PATH}")
 message("HEXAGON_SDK_PATH: ${HEXAGON_SDK_PATH}")
 message("HTP_ARCH_VERSION: ${HTP_ARCH_VERSION}")
 
+#v68 --- Snapdragon 888
+#v69 --- Snapdragon 8 Gen1
+#v73 --- Snapdragon 8 Gen2
+#v75 --- Snapdragon 8 Gen3
+#v79 --- Snapdragon 8 Elite(aka Gen4)
+if(NOT DEFINED HTP_ARCH_VERSION)
+    #set default HTP_ARCH_VERSION to v75
+    set(HTP_ARCH_VERSION v75)
+endif()
+
+#cross compiling for hexagon kernels on cDSP side
+set(HEXAGON_CC              "${HEXAGON_SDK_PATH}/tools/HEXAGON_Tools/8.8.06/Tools/bin/hexagon-clang")
+set(HEXAGON_CXX             "${HEXAGON_SDK_PATH}/tools/HEXAGON_Tools/8.8.06/Tools/bin/hexagon-clang")
+set(HEXAGON_TARGET          libggmlop_skel${HTP_ARCH_VERSION}.so)
+set(HEXAGON_KERNELS_PATH    "${CMAKE_CURRENT_LIST_DIR}/kernels")
+
 if(CMAKE_SYSTEM_NAME STREQUAL "Android")
     find_library(LOG_LIB log)
 
@@ -67,10 +83,10 @@ function(ggml_hexagon_build_kernel KNAME)
         TARGET ${PROJECT_NAME}
         POST_BUILD
         COMMAND echo "current working path:`pwd`\n"
-        COMMAND echo "${CMAKE_CURRENT_LIST_DIR}/kernels"
-        COMMAND make -C ${CMAKE_CURRENT_LIST_DIR}/kernels/ clean
-        COMMAND make -C ${CMAKE_CURRENT_LIST_DIR}/kernels/ HEXAGON_SDK_PATH=${HEXAGON_SDK_PATH} HTP_ARCH_VERSION=${HTP_ARCH_VERSION}
-        COMMAND echo "`pwd`"
+        COMMAND ${HEXAGON_CC} -o ${HEXAGON_KERNELS_PATH}/ggml-dsp.o -c ${HEXAGON_KERNELS_PATH}/ggml-dsp.c -m${HTP_ARCH_VERSION} -c -Ofast -Wall -Wstrict-prototypes -fno-zero-initialized-in-bss -fdata-sections -fpic -D__V_DYNAMIC__ -mhvx -mhvx-length=128B -I${HEXAGON_SDK_PATH}/incs -I${HEXAGON_SDK_PATH}/libs/qprintf/inc -I${HEXAGON_SDK_PATH}/incs/stddef -I${HEXAGON_SDK_PATH}/ipc/fastrpc/incs -I${HEXAGON_SDK_PATH}/ipc/fastrpc/rpcmem/inc -I${HEXAGON_SDK_PATH}/utils/examples -I${HEXAGON_SDK_PATH}/ipc/fastrpc/rtld/ship/inc -I${HEXAGON_SDK_PATH}/libs/atomic/inc -I${HEXAGON_SDK_PATH}/utils/sim_utils/inc
+        COMMAND ${HEXAGON_CC} -o ${HEXAGON_KERNELS_PATH}/ggmlop_cdsp_skel.o -c ${HEXAGON_KERNELS_PATH}/ggmlop_cdsp_skel.c -m${HTP_ARCH_VERSION} -c -Ofast -Wall -Wstrict-prototypes -fno-zero-initialized-in-bss -fdata-sections -fpic -D__V_DYNAMIC__ -mhvx -mhvx-length=128B -I${HEXAGON_SDK_PATH}/incs -I${HEXAGON_SDK_PATH}/libs/qprintf/inc -I${HEXAGON_SDK_PATH}/incs/stddef -I${HEXAGON_SDK_PATH}/ipc/fastrpc/incs -I${HEXAGON_SDK_PATH}/ipc/fastrpc/rpcmem/inc -I${HEXAGON_SDK_PATH}/utils/examples -I${HEXAGON_SDK_PATH}/ipc/fastrpc/rtld/ship/inc -I${HEXAGON_SDK_PATH}/libs/atomic/inc -I${HEXAGON_SDK_PATH}/utils/sim_utils/inc
+        COMMAND ${HEXAGON_CC} -m${HTP_ARCH_VERSION} -Wl,--defsym=ISDB_TRUSTED_FLAG=2 -Wl,--defsym=ISDB_SECURE_FLAG=2 -Wl,--no-threads -fpic -shared -Wl,-Bsymbolic -Wl,--wrap=malloc -Wl,--wrap=calloc -Wl,--wrap=free -Wl,--wrap=realloc -Wl,--wrap=memalign -lc -Wl,-soname=${HEXAGON_TARGET} -o ${HEXAGON_KERNELS_PATH}/${HEXAGON_TARGET} -Wl,--start-group ${HEXAGON_KERNELS_PATH}/ggmlop_cdsp_skel.o ${HEXAGON_KERNELS_PATH}/ggml-dsp.o -Wl,--end-group
+        COMMAND ls -l ${HEXAGON_KERNELS_PATH}/${HEXAGON_TARGET}
         COMMENT "build hexagon-kernel"
     )
 endfunction()

ggml/src/ggml-hexagon/kernels/Makefile

@@ -299,7 +299,9 @@ static const struct ggml_type_traits type_traits[GGML_TYPE_COUNT] = {
 //  section-2: ggml-hexagon kernel's internal troubleshooting function
 // =================================================================================================
 static void ggmlhexagon_log_internal(int level, const char *file, const char *func, int line, const char *format, ...) {
+#if !GGMLHEXAGON_DEBUG
     return;
+#endif
     static char s_ggmlhexagon_log_internal_buf[GGMLHEXAGON_LOGBUF_LEN];
     va_list args;
     va_start(args, format);

ggml/src/ggml-hexagon/kernels/ggml-dsp.c

@@ -50,7 +50,7 @@ extern "C" {
 #define GGML_FP32_TO_FP16(x)            GGML_COMPUTE_FP32_TO_FP16(x)
 #define GGML_FP16_TO_FP32(x)            ggml_lookup_fp16_to_fp32(x)
 
-#if 0//def NDEBUG
+#ifdef NDEBUG
 #define GGMLHEXAGON_DEBUG                                   0
 #else
 #define GGMLHEXAGON_DEBUG                                   1

ggml/src/ggml-hexagon/kernels/ggml-dsp.h

@@ -405,11 +405,9 @@ static __inline int _stub_method_1(remote_handle64 _handle, uint32_t _mid, uintp
    _count_1(_numIn, _numROut, _numInH, _numROutH, (uint32_t*)&(((uint32_t*)_in1)[0]), (uint32_t*)&(((uint32_t*)_in1)[1]), (uint32_t*)&(((uint32_t*)_in1)[5]), (uint32_t*)&(((uint32_t*)_in1)[9]), (uint32_t*)&(((uint32_t*)_in1)[10]), (uint32_t*)&(((uint32_t*)_in1)[26]), SLIM_IFPTR32((char**)&(((uint32_t*)_in1)[27]), (char**)&(((uint64_t*)_in1)[14])), SLIM_IFPTR32((uint32_t*)&(((uint32_t*)_in1)[28]), (uint32_t*)&(((uint32_t*)_in1)[30])));
    _count(_numIn, _numROut, _numInH, _numROutH, (uint32_t*)&(((uint32_t*)_rout2)[0]), (uint32_t*)&(((uint32_t*)_rout2)[1]), (uint32_t*)&(((uint32_t*)_rout2)[5]), (uint32_t*)&(((uint32_t*)_rout2)[9]), (uint32_t*)&(((uint32_t*)_rout2)[10]), (uint32_t*)&(((uint32_t*)_rout2)[26]), SLIM_IFPTR32((char**)&(((uint32_t*)_rout2)[27]), (char**)&(((uint64_t*)_rout2)[14])), SLIM_IFPTR32((uint32_t*)&(((uint32_t*)_rout2)[28]), (uint32_t*)&(((uint32_t*)_rout2)[30])));
    if(_numIn[0]>=255){
-          _QAIC_FARF(RUNTIME_ERROR, "ERROR: Unsupported number of input buffers\n");
           return AEE_EUNSUPPORTED;
    }
    if(_numROut[0]>=255){
-          _QAIC_FARF(RUNTIME_ERROR, "ERROR: Unsupported number of output buffers\n");
           return AEE_EUNSUPPORTED;
    }
    _allocator_init(_al, 0, 0);