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/kmeans/Makefile
+++ b/tests/opencl/kmeans/Makefile
@@ -1,67 +1,7 @@
-XLEN ?= 32
-
-LLVM_PREFIX ?= /opt/llvm-riscv
-RISCV_TOOLCHAIN_PATH ?= /opt/riscv-gnu-toolchain
-SYSROOT ?= $(RISCV_TOOLCHAIN_PATH)/riscv32-unknown-elf
-POCL_CC_PATH ?= /opt/pocl/compiler
-POCL_RT_PATH ?= /opt/pocl/runtime
-
-VORTEX_DRV_PATH ?= $(realpath ../../../driver)
-VORTEX_RT_PATH ?= $(realpath ../../../runtime)
-
-K_LLCFLAGS += "-O3 -march=riscv32 -target-abi=ilp32f -mcpu=generic-rv32 -mattr=+m,+f -mattr=+vortex -float-abi=hard -code-model=small"
-K_CFLAGS   += "-v -O3 --sysroot=$(SYSROOT) --gcc-toolchain=$(RISCV_TOOLCHAIN_PATH) -march=rv32imf -mabi=ilp32f -Xclang -target-feature -Xclang +vortex -I$(VORTEX_RT_PATH)/include -fno-rtti -fno-exceptions -ffreestanding -nostartfiles -fdata-sections -ffunction-sections"
-K_LDFLAGS  += "-Wl,-Bstatic,-T$(VORTEX_RT_PATH)/linker/vx_link$(XLEN).ld -Wl,--gc-sections $(VORTEX_RT_PATH)/libvortexrt.a -lm"
-
-CXXFLAGS += -std=c++11 -Wall -Wextra -pedantic -Wfatal-errors
-
-CXXFLAGS += -I$(POCL_RT_PATH)/include
-
-LDFLAGS += -L$(POCL_RT_PATH)/lib -L$(VORTEX_DRV_PATH)/stub -lOpenCL -lvortex
-
-# Debugigng
-ifdef DEBUG
-	CXXFLAGS += -g -O0
-else    
-	CXXFLAGS += -O2 -DNDEBUG
-endif
-
 PROJECT = kmeans

 SRCS = main.cc read_input.c rmse.c kmeans_clustering.c cluster.c getopt.c

-all: $(PROJECT) kernel.pocl
+OPTS ?= 

-kernel.pocl: kernel.cl
-	LLVM_PREFIX=$(LLVM_PREFIX) POCL_DEBUG=all LD_LIBRARY_PATH=$(LLVM_PREFIX)/lib:$(POCL_CC_PATH)/lib $(POCL_CC_PATH)/bin/poclcc -LLCFLAGS $(K_LLCFLAGS) -CFLAGS $(K_CFLAGS) -LDFLAGS $(K_LDFLAGS) -o kernel.pocl kernel.cl
- 
-$(PROJECT): $(SRCS)
-	$(CXX) $(CXXFLAGS) $^ $(LDFLAGS) -o $@
-
-run-fpga: $(PROJECT) kernel.pocl   
-	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/fpga:$(LD_LIBRARY_PATH) ./$(PROJECT)
-
-run-asesim: $(PROJECT) kernel.pocl   
-	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/asesim:$(LD_LIBRARY_PATH) ./$(PROJECT)
-	
-run-vlsim: $(PROJECT) kernel.pocl   
-	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/vlsim:$(LD_LIBRARY_PATH) ./$(PROJECT)
-
-run-simx: $(PROJECT) kernel.pocl   
-	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/simx:$(LD_LIBRARY_PATH) ./$(PROJECT)
-
-run-rtlsim: $(PROJECT) kernel.pocl   
-	LD_LIBRARY_PATH=$(POCL_RT_PATH)/lib:$(VORTEX_DRV_PATH)/rtlsim:$(LD_LIBRARY_PATH) ./$(PROJECT)
-
-.depend: $(SRCS)
-	$(CXX) $(CXXFLAGS) -MM $^ > .depend;
-
-clean:
-	rm -rf $(PROJECT) *.o .depend 
-
-clean-all: clean
-	rm -rf *.pocl *.dump
-
-ifneq ($(MAKECMDGOALS),clean)
-    -include .depend
-endif
+include ../common.mk
--- a/tests/opencl/kmeans/kernel.pocl
+++ b/tests/opencl/kmeans/kernel.pocl
--- a/tests/opencl/kmeans/kmeans_clustering.c
+++ b/tests/opencl/kmeans/kmeans_clustering.c
@@ -170,6 +170,7 @@ float** kmeans_clustering(float **feature,    /* in: [npoints][nfeatures] */
    free(new_centers[0]);
    free(new_centers);
    free(new_centers_len);
+	free(initial);

    return clusters;
 }
--- a/tests/opencl/kmeans/kmeans_kernel_c.dump
+++ b/tests/opencl/kmeans/kmeans_kernel_c.dump
--- a/tests/opencl/kmeans/kmeans_swap.dump
+++ b/tests/opencl/kmeans/kmeans_swap.dump
--- a/tests/opencl/kmeans/main.cc
+++ b/tests/opencl/kmeans/main.cc
@@ -104,8 +104,8 @@ static int initialize(int use_gpu) {
  context = clCreateContext(NULL, 1, device_list, NULL, NULL,  &result);

  // create command queue for the first device
-  cmd_queue = clCreateCommandQueue(context, device_list[0], 0, NULL);
-  if (!cmd_queue) {
+  cmd_queue = clCreateCommandQueue(context, device_list[0], 0, &result);
+  if (!cmd_queue || result != CL_SUCCESS) {
    printf("ERROR: clCreateCommandQueue() failed\n");
    return -1;
  }
@@ -120,7 +120,7 @@ static int shutdown() {
  if (context)
    clReleaseContext(context);
  if (device_list)
-    delete device_list;
+    delete [] device_list;

  // reset all variables
  cmd_queue = 0;
@@ -188,7 +188,7 @@ int allocate(int n_points, int n_features, int n_clusters, float **feature) {
  fread(source + strlen(source), sourcesize, 1, fp);
  fclose(fp);*/

-  // OpenCL initialization
+  // OpenCL initialization  
  int use_gpu = 1;
  if (initialize(use_gpu))
    return -1;
@@ -197,12 +197,25 @@ int allocate(int n_points, int n_features, int n_clusters, float **feature) {
  cl_int err = 0;
  //const char *slist[2] = {source, 0};
  //cl_program prog = clCreateProgramWithSource(context, 1, slist, NULL, &err);
-  cl_program prog = clCreateProgramWithBuiltInKernels(context, 1, device_list, "kmeans_kernel_c;kmeans_swap", &err);       
+  uint8_t *kernel_bin = NULL;
+  size_t kernel_size;
+  cl_int binary_status = 0;  
+  err = read_kernel_file("kernel.pocl", &kernel_bin, &kernel_size);
  if (err != CL_SUCCESS) {
-    printf("ERROR: clCreateProgramWithSource() => %d\n", err);
+    printf("ERROR: read_kernel_file() => %d\n", err);
    return -1;
  }
-  err = clBuildProgram(prog, 0, NULL, NULL, NULL, NULL);
+
+	cl_program prog = clCreateProgramWithBinary(
+      context, 1, device_list, &kernel_size, (const uint8_t**)&kernel_bin, &binary_status, &err);
+  if (err != CL_SUCCESS) {
+    printf("ERROR: clCreateProgramWithBinary() => %d\n", err);
+    return -1;
+  }
+
+  free(kernel_bin);
+
+  err = clBuildProgram(prog, 1, &device_list[0], NULL, NULL, NULL);
  { // show warnings/errors
    //	static char log[65536]; memset(log, 0, sizeof(log));
    //	cl_device_id device_id = 0;
@@ -226,6 +239,7 @@ int allocate(int n_points, int n_features, int n_clusters, float **feature) {
    printf("ERROR: clCreateKernel() 0 => %d\n", err);
    return -1;
  }
+
  kernel2 = clCreateKernel(prog, kernel_swap, &err);
  if (err != CL_SUCCESS) {
    printf("ERROR: clCreateKernel() 0 => %d\n", err);
@@ -241,6 +255,7 @@ int allocate(int n_points, int n_features, int n_clusters, float **feature) {
           n_points * n_features, err);
    return -1;
  }
+
  d_feature_swap =
      clCreateBuffer(context, CL_MEM_READ_WRITE,
                     n_points * n_features * sizeof(float), NULL, &err);
@@ -249,6 +264,7 @@ int allocate(int n_points, int n_features, int n_clusters, float **feature) {
           n_points * n_features, err);
    return -1;
  }
+
  d_cluster =
      clCreateBuffer(context, CL_MEM_READ_WRITE,
                     n_clusters * n_features * sizeof(float), NULL, &err);
@@ -257,6 +273,7 @@ int allocate(int n_points, int n_features, int n_clusters, float **feature) {
           n_clusters * n_features, err);
    return -1;
  }
+
  d_membership = clCreateBuffer(context, CL_MEM_READ_WRITE,
                                n_points * sizeof(int), NULL, &err);
  if (err != CL_SUCCESS) {
@@ -296,6 +313,8 @@ int allocate(int n_points, int n_features, int n_clusters, float **feature) {
  }

  membership_OCL = (int *)malloc(n_points * sizeof(int));
+
+  return 0;
 }

 void deallocateMemory() {
--- a/tests/opencl/kmeans/read_input.c
+++ b/tests/opencl/kmeans/read_input.c
@@ -331,7 +331,9 @@ int setup(int argc, char **argv) {
    }
  }

-  /* free up memory */
+  /* free up memory */  
+  free(cluster_centres[0]);
+  free(cluster_centres);
  free(features[0]);
  free(features);
  return (0);