Vortex 2.0 changes:

+ Microarchitecture optimizations + 64-bit support + Xilinx FPGA support + LLVM-16 support + Refactoring and quality control fixes minor update minor update minor update minor update minor update minor update cleanup cleanup cache bindings and memory perf refactory minor update minor update hw unit tests fixes minor update minor update minor update minor update minor update minor udpate minor update minor update minor update minor update minor update minor update minor update minor updates minor updates minor update minor update minor update minor update minor update minor update minor updates minor updates minor updates minor updates minor update minor update
2023-10-19 20:51:22 -07:00
parent d69a64c32c
commit c1e168fdbe
1309 changed files with 247412 additions and 311463 deletions
--- a/tests/opencl/mri-q/Makefile
+++ b/tests/opencl/mri-q/Makefile
@@ -1,69 +1,9 @@
-XLEN ?= 32
-
-RISCV_TOOLCHAIN_PATH ?= $(wildcard ../../../../riscv-gnu-toolchain/drops)
-POCL_CC_PATH ?= $(wildcard ../../../../pocl/drops_riscv_cc)
-POCL_INC_PATH ?= $(wildcard ../include)
-POCL_LIB_PATH ?= $(wildcard ../lib)
-VORTEX_RT_PATH ?= $(wildcard ../../../runtime)
-VX_SIMX_PATH ?= $(wildcard ../../../simx/obj_dir)
-
-CC  = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-gcc
-CXX = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-g++
-DMP = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-objdump
-HEX = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-objcopy
-GDB = $(RISCV_TOOLCHAIN_PATH)/bin/riscv32-unknown-elf-gdb
-
-VX_SRCS =  $(VORTEX_RT_PATH)/newlib/newlib.c
-VX_SRCS += $(VORTEX_RT_PATH)/startup/vx_start.S
-VX_SRCS += $(VORTEX_RT_PATH)/intrinsics/vx_intrinsics.S
-VX_SRCS += $(VORTEX_RT_PATH)/io/vx_io.S $(VORTEX_RT_PATH)/io/vx_io.c
-VX_SRCS += $(VORTEX_RT_PATH)/fileio/fileio.S
-VX_SRCS += $(VORTEX_RT_PATH)/tests/tests.c
-VX_SRCS += $(VORTEX_RT_PATH)/vx_api/vx_api.c
-
-VX_CFLAGS = -nostartfiles -Wl,-Bstatic,-T,$(VORTEX_RT_PATH)/startup/vx_link$(XLEN).ld
-
-CXXFLAGS =  -g -O0 -march=rv32im -mabi=ilp32 
-CXXFLAGS += -ffreestanding # program may not begin at main()
-CXXFLAGS += -Wl,--gc-sections # enable garbage collection of unused input sections
-CXXFLAGS += -fno-rtti -fno-non-call-exceptions # disable RTTI and exceptions
-CXXFLAGS += -I$(POCL_INC_PATH) -I.
-
-VX_LIBS = -Wl,--whole-archive lib$(PROJECT).a -Wl,--no-whole-archive $(POCL_LIB_PATH)/libOpenCL.a
-QEMU_LIBS = $(VORTEX_RT_PATH)/qemu/vx_api.c -Wl,--whole-archive lib$(PROJECT).a -Wl,--no-whole-archive $(POCL_LIB_PATH)/libOpenCL.a
-
 PROJECT = mri-q

 SRCS = main.cc args.c parboil_opencl.c ocl.c gpu_info.c file.cc computeQ.c

-all: $(PROJECT).dump $(PROJECT).hex
+CXXFLAGS += -I.

-lib$(PROJECT).a: kernel.cl
-	POCL_DEBUG=all POCL_DEBUG_LLVM_PASSES=1 LD_LIBRARY_PATH=$(RISCV_TOOLCHAIN_PATH)/lib:$(POCL_CC_PATH)/lib $(POCL_CC_PATH)/bin/poclcc -o lib$(PROJECT).a kernel.cl
+OPTS ?= 

-$(PROJECT).elf: $(SRCS) lib$(PROJECT).a
-	$(CXX) $(CXXFLAGS) $(VX_CFLAGS) $(VX_SRCS) $(SRCS) $(VX_LIBS) -o $(PROJECT).elf
-
-$(PROJECT).qemu: $(SRCS) lib$(PROJECT).a
-	$(CXX) $(CXXFLAGS) $(SRCS) $(QEMU_LIBS) -o $(PROJECT).qemu
-
-$(PROJECT).hex: $(PROJECT).elf
-	$(HEX) -O ihex $(PROJECT).elf $(PROJECT).hex
-
-$(PROJECT).dump: $(PROJECT).elf
-	$(DMP) -D $(PROJECT).elf > $(PROJECT).dump
-
-run: $(PROJECT).hex
-	POCL_DEBUG=all $(VX_SIMX_PATH)/Vcache_simX -E $(PROJECT).hex -s -b 1> emulator.debug
-
-qemu: $(PROJECT).qemu
-	POCL_DEBUG=all $(RISCV_TOOLCHAIN_PATH)/bin/qemu-riscv32 -d in_asm -D debug.log $(PROJECT).qemu
-
-gdb-s: $(PROJECT).qemu
-	POCL_DEBUG=all $(RISCV_TOOLCHAIN_PATH)/bin/qemu-riscv32 -g 1234 -d in_asm -D debug.log $(PROJECT).qemu
-
-gdb-c: $(PROJECT).qemu
-	$(GDB) $(PROJECT).qemu
-
-clean:
-	rm -rf *.o *.elf *.dump *.hex *.qemu *.log *.debug
+include ../common.mk
--- a/tests/opencl/mri-q/computeQ.h
+++ b/tests/opencl/mri-q/computeQ.h
@@ -1,6 +1,10 @@
 #ifndef __COMPUTEQ__
 #define __COMPUTEQ__

+#ifdef __cplusplus
+extern "C" {
+#endif
+
 void computePhiMag_GPU(int numK,cl_mem phiR_d,cl_mem phiI_d,cl_mem phiMag_d,clPrmtr* clPrm);
 void computeQ_GPU (int numK,int numX,
 		   cl_mem x_d, cl_mem y_d, cl_mem z_d,
@@ -11,4 +15,8 @@ void computeQ_GPU (int numK,int numX,
 void createDataStructsCPU(int numK, int numX, float** phiMag,
 	 		  float** Qr, float** Qi);

+#ifdef __cplusplus
+}
+#endif
+
 #endif
--- a/tests/opencl/mri-q/gpu_info.h
+++ b/tests/opencl/mri-q/gpu_info.h
@@ -9,6 +9,10 @@
 #ifndef __GPUINFOH__
 #define __GPUINFOH__

+#ifdef __cplusplus
+extern "C" {
+#endif
+
 void compute_active_thread(size_t *thread,
 			   size_t *grid,
 			   int task,
@@ -17,4 +21,8 @@ void compute_active_thread(size_t *thread,
 			   int minor,
 			   int sm);

+#ifdef __cplusplus
+}
+#endif
+
 #endif
--- a/tests/opencl/mri-q/libmri-q.a
+++ b/tests/opencl/mri-q/libmri-q.a
--- a/tests/opencl/mri-q/libsgemm.a
+++ b/tests/opencl/mri-q/libsgemm.a
--- a/tests/opencl/mri-q/main.cc
+++ b/tests/opencl/mri-q/main.cc
@@ -34,6 +34,27 @@
 #include "macros.h"
 #include "computeQ.h"

+static int read_kernel_file(const char* filename, uint8_t** data, size_t* size) {
+  if (nullptr == filename || nullptr == data || 0 == size)
+    return CL_INVALID_VALUE;
+
+  FILE* fp = fopen(filename, "r");
+  if (NULL == fp) {
+    fprintf(stderr, "Failed to load kernel.");
+    return CL_INVALID_VALUE;
+  }
+  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 CL_SUCCESS;
+}
+
 static void
 setupMemoryGPU(int num, int size, cl_mem* dev_ptr, float* host_ptr,clPrmtr* clPrm)
 {
@@ -93,8 +114,6 @@ main (int argc, char *argv[]) {
 	    &x, &y, &z,
 	    &phiR, &phiI);

-  printf("OK\n");
-
  /* Reduce the number of k-space samples if a number is given
   * on the command line */
  if (argc < 2)
@@ -137,13 +156,20 @@ main (int argc, char *argv[]) {

  pb_SetOpenCL(&(clPrm.clContext), &(clPrm.clCommandQueue));

-  printf("OK\n");
-
-  //const char* clSource[] = {readFile("src/opencl_base/kernels.cl")};
-  //cl_program clProgram = clCreateProgramWithSource(clPrm.clContext,1,clSource,NULL,&clStatus);
-  cl_program clProgram = clCreateProgramWithBuiltInKernels(
-      clPrm.clContext, 1, &clDevice, "ComputePhiMag_GPU;ComputeQ_GPU", &clStatus);
+#ifdef HOSTGPU
+  const char* clSource[] = {readFile("kernel.cl")};
  CHECK_ERROR("clCreateProgramWithSource")
+  cl_program clProgram = clCreateProgramWithSource(clPrm.clContext,1,clSource,NULL,&clStatus);
+#else
+  uint8_t *kernel_bin = NULL;
+  size_t kernel_size;
+  cl_int binary_status = 0;  
+  CHECK_ERROR("read_kernel_file")
+  clStatus = read_kernel_file("kernel.pocl", &kernel_bin, &kernel_size);  
+  CHECK_ERROR("clCreateProgramWithSource")
+	cl_program clProgram = clCreateProgramWithBinary(
+      clPrm.clContext, 1, &clDevice, &kernel_size, (const uint8_t**)&kernel_bin, &binary_status, &clStatus);  
+#endif

  char options[50];
  sprintf(options,"-I src/opencl_nvidia");
--- a/tests/opencl/mri-q/ocl.h
+++ b/tests/opencl/mri-q/ocl.h
@@ -3,6 +3,10 @@

 #include <stdlib.h>

+#ifdef __cplusplus
+extern "C" {
+#endif
+
 typedef struct {
 	cl_context clContext;
 	cl_command_queue clCommandQueue;
@@ -20,4 +24,8 @@ char* readFile(const char*);
     exit(1);                               \
  }

+#ifdef __cplusplus
+}
+#endif
+
 #endif