project directories reorganization

benchmarks/opencl/sgemm/Makefile

@@ -1,68 +1,47 @@
-RISCV_TOOL_PATH ?= $(wildcard ../../../../riscv-gnu-toolchain/drops)
-POCL_CC_PATH ?= $(wildcard ../../../../pocl/drops_riscv_cc)
-POCL_INC_PATH ?= $(wildcard ../include)
-POCL_LIB_PATH ?= $(wildcard ../lib)
-VX_RT_PATH ?= $(wildcard ../../../runtime)
-VX_SIMX_PATH ?= $(wildcard ../../../simX/obj_dir)
+LLVM_HOME ?= ~/dev/llvm-project/drops
+TOOLCHAIN_PATH ?= ~/dev/riscv-gnu-toolchain/drops
+SYSROOT ?= $(TOOLCHAIN_PATH)/riscv32-unknown-elf
+POCL_CC_PATH ?= $(realpath ../compiler)
+POCL_RT_PATH ?= $(realpath ../runtime)
+VORTEX_DRV_PATH ?= $(realpath ../../../driver/sw)
+VORTEX_RT_PATH ?= $(realpath ../../../runtime)
 
-CC  = $(RISCV_TOOL_PATH)/bin/riscv32-unknown-elf-gcc
-CXX = $(RISCV_TOOL_PATH)/bin/riscv32-unknown-elf-g++
-DMP = $(RISCV_TOOL_PATH)/bin/riscv32-unknown-elf-objdump
-HEX = $(RISCV_TOOL_PATH)/bin/riscv32-unknown-elf-objcopy
-GDB = $(RISCV_TOOL_PATH)/bin/riscv32-unknown-elf-gdb
+CXXFLAGS += -std=c++11 -O0 -g -fpermissive -Wall -Wextra -pedantic -Wfatal-errors
 
-VX_SRCS =  $(VX_RT_PATH)/newlib/newlib.c
-VX_SRCS += $(VX_RT_PATH)/startup/vx_start.S
-VX_SRCS += $(VX_RT_PATH)/intrinsics/vx_intrinsics.s
-VX_SRCS += $(VX_RT_PATH)/io/vx_io.s $(VX_RT_PATH)/io/vx_io.c
-VX_SRCS += $(VX_RT_PATH)/fileio/fileio.s
-VX_SRCS += $(VX_RT_PATH)/tests/tests.c
-VX_SRCS += $(VX_RT_PATH)/vx_api/vx_api.c
-VX_SRCS += $(VX_STR) $(VX_FIO) $(VX_NEWLIB) $(VX_INT) $(VX_IO) $(VX_API) $(VX_TEST)
+CXXFLAGS += -I$(POCL_RT_PATH)/include
 
-VX_CFLAGS = -nostartfiles -Wl,-Bstatic,-T,$(VX_RT_PATH)/mains/vortex_link.ld
-
-CXXFLAGS =  -g -O0 -march=rv32im -mabi=ilp32 
-CXXFLAGS += -ffreestanding # program may not begin at main()
-CXXFLAGS += -Wl,--gc-sections # enable garbage collection of unused input sections
-CXXFLAGS += -fno-rtti -fno-non-call-exceptions # disable RTTI and exceptions
-CXXFLAGS += -I$(POCL_INC_PATH)
-
-VX_LIBS = -Wl,--whole-archive lib$(PROJECT).a -Wl,--no-whole-archive $(POCL_LIB_PATH)/libOpenCL.a
-QEMU_LIBS = $(VX_RT_PATH)/qemu/vx_api.c -Wl,--whole-archive lib$(PROJECT).a -Wl,--no-whole-archive $(POCL_LIB_PATH)/libOpenCL.a
+LDFLAGS += -L$(POCL_RT_PATH)/lib -L$(VORTEX_DRV_PATH)/stub -lOpenCL -lvortex
 
 PROJECT = sgemm
 
 SRCS = main.cc
 
-all: $(PROJECT).dump $(PROJECT).hex
+all: $(PROJECT)
 
-lib$(PROJECT).a: kernel.cl
-	POCL_DEBUG=all POCL_DEBUG_LLVM_PASSES=1 LD_LIBRARY_PATH=$(RISCV_TOOL_PATH)/lib:$(POCL_CC_PATH)/lib $(POCL_CC_PATH)/bin/poclcc -o lib$(PROJECT).a kernel.cl
+kernel.pocl: kernel.cl
+	TOOLCHAIN_PATH=$(TOOLCHAIN_PATH) SYSROOT=$(SYSROOT) LLVM_HOME=$(LLVM_HOME) VORTEX_RUNTIME_PATH=$(VORTEX_RT_PATH) POCL_DEBUG=all LD_LIBRARY_PATH=$(LLVM_HOME)/lib:$(POCL_CC_PATH)/lib $(POCL_CC_PATH)/bin/poclcc -o kernel.pocl kernel.cl
+ 
+$(PROJECT): $(SRCS)
+	$(CXX) $(CXXFLAGS) $^ $(LDFLAGS) -o $@
 
-$(PROJECT).elf: $(SRCS) lib$(PROJECT).a
-	$(CXX) $(CXXFLAGS) $(VX_CFLAGS) $(VX_SRCS)  $(SRCS) $(VX_LIBS) -o $(PROJECT).elf
+run-fpga: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/opae:$(LD_LIBRARY_PATH) ./$(PROJECT)
 
-$(PROJECT).qemu: $(SRCS) lib$(PROJECT).a
-	$(CXX) $(CXXFLAGS)  $(SRCS) $(QEMU_LIBS) -o $(PROJECT).qemu
+run-ase: $(PROJECT) kernel.pocl   
+	ASE_LOG=0 LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/opae/ase:$(LD_LIBRARY_PATH) ./$(PROJECT)
 
-$(PROJECT).hex: $(PROJECT).elf
-	$(HEX) -O ihex $(PROJECT).elf $(PROJECT).hex
+run-simx: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/simx:$(LD_LIBRARY_PATH) ./$(PROJECT)
 
-$(PROJECT).dump: $(PROJECT).elf
-	$(DMP) -D $(PROJECT).elf > $(PROJECT).dump
+run-rtlsim: $(PROJECT) kernel.pocl   
+	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/rtlsim:$(LD_LIBRARY_PATH) ./$(PROJECT)
 
-run: $(PROJECT).hex
-	POCL_DEBUG=all $(VX_SIMX_PATH)/Vcache_simX -E -a rv32i --core $(PROJECT).hex -s -b 1> emulator.debug
-
-qemu: $(PROJECT).qemu
-	POCL_DEBUG=all $(RISCV_TOOL_PATH)/bin/qemu-riscv32 -d in_asm -D debug.log $(PROJECT).qemu
-
-gdb-s: $(PROJECT).qemu
-	POCL_DEBUG=all $(RISCV_TOOL_PATH)/bin/qemu-riscv32 -g 1234 -d in_asm -D debug.log $(PROJECT).qemu
-
-gdb-c: $(PROJECT).qemu
-	$(GDB) $(PROJECT).qemu
+.depend: $(SRCS)
+	$(CXX) $(CXXFLAGS) -MM $^ > .depend;
 
 clean:
-	rm -rf *.o *.elf *.dump *.hex *.qemu *.log *.debug
+	rm -rf $(PROJECT) *.o *.dump .depend 
+
+ifneq ($(MAKECMDGOALS),clean)
+    -include .depend
+endif

benchmarks/opencl/sgemm/main.cc

@@ -46,7 +46,7 @@
 #define CL_CHECK_ERR(_expr)                                                    \
   ({                                                                           \
     cl_int _err = CL_INVALID_VALUE;                                            \
-    typeof(_expr) _ret = _expr;                                                \
+    decltype(_expr) _ret = _expr;                                              \
     if (_err != CL_SUCCESS) {                                                  \
       fprintf(stderr, "OpenCL Error: '%s' returned %d!\n", #_expr, (int)_err); \
       abort();                                                                 \
@@ -59,6 +59,29 @@ void pfn_notify(const char *errinfo, const void *private_info, size_t cb,
   fprintf(stderr, "OpenCL Error (via pfn_notify): %s\n", errinfo);
 }
 
+static int read_kernel_file(const char* filename, uint8_t** data, size_t* size) {
+  if (nullptr == filename || nullptr == data || 0 == size)
+    return -1;
+
+  FILE* fp = fopen(filename, "r");
+  if (NULL == fp) {
+    fprintf(stderr, "Failed to load kernel.");
+    return -1;
+  }
+  fseek(fp , 0 , SEEK_END);
+  long fsize = ftell(fp);
+  rewind(fp);
+
+  *data = (uint8_t*)malloc(fsize);
+  *size = fread(*data, 1, fsize, fp);
+  
+  fclose(fp);
+  
+  return 0;
+}
+
+uint8_t *kernel_bin = NULL;
+
 ///
 //  Cleanup any created OpenCL resources
 //
@@ -79,6 +102,8 @@ void Cleanup(cl_context context, cl_command_queue commandQueue,
 
   if (context != 0)
     clReleaseContext(context);
+
+  if (kernel_bin) free(kernel_bin);
 }
 
 int main(int argc, char **argv) {
@@ -86,9 +111,14 @@ int main(int argc, char **argv) {
   
   cl_platform_id platform_id;
   cl_device_id device_id;
-  size_t binary_size;
+  size_t kernel_size;
+  cl_int binary_status = 0;
   int i;
 
+  // read kernel binary from file  
+  if (0 != read_kernel_file("kernel.pocl", &kernel_bin, &kernel_size))
+    return -1;
+
   // Getting platform and device information
   CL_CHECK(clGetPlatformIDs(1, &platform_id, NULL));
   CL_CHECK(clGetDeviceIDs(platform_id, CL_DEVICE_TYPE_DEFAULT, 1, &device_id, NULL));
@@ -110,8 +140,8 @@ int main(int argc, char **argv) {
   std::cout << "Attempting to create program from binary..." << std::endl;
   // cl_program program = CreateProgramFromBinary(context, device_id,
   // "kernel.cl.bin");
-  cl_program program =
-      clCreateProgramWithBuiltInKernels(context, 1, &device_id, "sgemm", NULL);
+  cl_program program = CL_CHECK_ERR(clCreateProgramWithBinary(
+    context, 1, &device_id, &kernel_size, &kernel_bin, &binary_status, &_err));
   if (program == NULL) {
     std::cerr << "Failed to write program binary" << std::endl;
     Cleanup(context, queue, program, kernel, memObjects);
@@ -146,7 +176,7 @@ int main(int argc, char **argv) {
   memObjects[1] = input_bufferB;
   memObjects[2] = output_buffer;
 
-  size_t width = NUM_DATA;
+  int width = NUM_DATA;
 
   printf("attempting to create kernel\n");
   fflush(stdout);