Mercurial > hg > Game > Cerium

changeset 1454:5b79077e4e1f draft

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
fix GpuRegistTaskTest
author Yuhi TOMARI <yuhi@cr.ie.u-ryukyu.ac.jp>
date Sun, 29 Apr 2012 19:21:26 +0900
parents 756470aab7b3
children 2b886dcc0e7d
files TaskManager/Gpu/GpuScheduler.cc TaskManager/Gpu/GpuThreads.cc TaskManager/Gpu/GpuThreads.h TaskManager/test/GpuRegistTaskTest/GpuRegistTaskTest.cc
diffstat 4 files changed, 62 insertions(+), 44 deletions(-) [+]
line wrap: on
line diff
--- a/TaskManager/Gpu/GpuScheduler.cc	Wed Apr 25 20:44:38 2012 +0900
+++ b/TaskManager/Gpu/GpuScheduler.cc	Sun Apr 29 19:21:26 2012 +0900
@@ -20,7 +20,7 @@
     // Get OpenCL infomation
     GpuThreads* gputhreads = GpuThreads::getInstance();
     cl_context& context = gputhreads->context;
-    cl_command_queue& command_queue = gputhreads->command_queue;
+    cl_command_queue command_queue = gputhreads->command_queue;
     cl_int ret;
 
     if ((memaddr)params_addr == (memaddr)MY_SPE_COMMAND_EXIT) {
@@ -56,8 +56,9 @@
 gpu_register_task(int cmd, const char* filename, const char* functionname)
 {
     GpuThreads* gputhreads = GpuThreads::getInstance();
-    cl_context& context = gputhreads->context;
-    cl_device_id& device_id = gputhreads->device_id;
+    //    gputhreads->init();
+    cl_context context = gputhreads->context;
+    cl_device_id device_id = gputhreads->device_id;
 
     int fp;
     char *source_str;
@@ -76,6 +77,7 @@
     
     if (!size) {
         fprintf(stderr, "Failed to load kernel.\n");
+        exit(1);
     }
     
     source_str = (char*)malloc(size);
@@ -92,7 +94,7 @@
 
     cl_kernel *kernel = new cl_kernel; 
     *kernel = clCreateKernel(program, functionname, &ret);
-    
+     
     task_list[cmd].run = NULL;
     task_list[cmd].load = NULL;
     task_list[cmd].wait = NULL;
--- a/TaskManager/Gpu/GpuThreads.cc	Wed Apr 25 20:44:38 2012 +0900
+++ b/TaskManager/Gpu/GpuThreads.cc	Sun Apr 29 19:21:26 2012 +0900
@@ -33,7 +33,7 @@
     args->scheduler = new GpuScheduler();
     args->useRefDma = use_refdma;
 
-    pthread_create(threads, NULL, &gpu_thread_run, args);
+    // pthread_create(threads, NULL, &gpu_thread_run, args);
 
 }
 
--- a/TaskManager/Gpu/GpuThreads.h	Wed Apr 25 20:44:38 2012 +0900
+++ b/TaskManager/Gpu/GpuThreads.h	Sun Apr 29 19:21:26 2012 +0900
@@ -38,7 +38,7 @@
     cl_device_id device_id;
     cl_uint ret_num_platforms;
     cl_uint ret_num_devices;
-    cl_context context;
+    cl_context context ;
     cl_command_queue command_queue;
     cl_int ret;
 
--- a/TaskManager/test/GpuRegistTaskTest/GpuRegistTaskTest.cc	Wed Apr 25 20:44:38 2012 +0900
+++ b/TaskManager/test/GpuRegistTaskTest/GpuRegistTaskTest.cc	Sun Apr 29 19:21:26 2012 +0900
@@ -1,9 +1,10 @@
 #include <stdio.h>
 #include <OpenCL/opencl.h>
-#include "GpuThreads.h"
+#include <fcntl.h>
+#include <sys/stat.h>
 #include "GpuScheduler.h"
 #include "CellTaskManagerImpl.h"
-
+#include "GpuThreads.h"
 #define DEFAULT 5
 
 extern void gpu_register_task(int, const char*, const char*);
@@ -19,78 +20,93 @@
 
 void
 tester(int *indata,int *outdata, int num){
+
     //チェック
-    int check;
+    int check = 0;
     for (int c=0; c<num; c++){
         if(outdata[c] == indata[c]*2){
+            // printf("outdata:%d  indata:%d\n\n",outdata[c],indata[c]);
             check++;
+        }else {
+            printf("\n\n incorrect number! outdata[%d]:%d  indata[%d]:%d\n\n",c,outdata[c],c,indata[c]);
         }
-        printf("Computed '%d/%d' correct values",check,num);
     }
+    printf("Computed '%d/%d' correct values\n",check,num);
 }
 
 void
 test(int task_array_num){
-    
-    GpuThreads* gpuThreads = GpuThreads::getInstance();
-    gpuThreads->init();
+
+    int cmd = 1;
+    GpuThreads* gputhreads = GpuThreads::getInstance();
+    gputhreads->init();
+    GpuSchedRegister(cmd, "twice.cl", "twice");
+    cl_platform_id platform_id = NULL;
+    cl_uint ret_num_platforms = NULL;
+    cl_device_id device_id = NULL;
+    cl_uint ret_num_devices = NULL;
 
+    cl_int ret;
+    clGetPlatformIDs(1, &platform_id, &ret_num_platforms);
+    clGetDeviceIDs(platform_id, CL_DEVICE_TYPE_DEFAULT, 1, &device_id, 
+                   &ret_num_devices);
+    cl_context context = gputhreads->context;
+    
+    cl_command_queue command_queue = gputhreads->command_queue;
+    
     int *indata,*outdata;
+    indata = new int[task_array_num];
+    outdata = new int[task_array_num];
     int count;
-    indata = (int *)malloc(sizeof(int)*task_array_num);
-    outdata = (int *)malloc(sizeof(int)*task_array_num);
     for (count=0; count < task_array_num ;count++) {
         indata[count] = count;
     }
-    printf("%d",count);
     
-    GpuSchedRegister(1, "twice.cl", "Twice");
-
-    cl_int ret;
-    cl_context context = gpu_register_task->context;
-    cl_mem memobj_in, memobj_out, data_count = NULL;
-    memobj_in  = clCreateBuffer(context, CL_MEM_READ_WRITE, sizeof(int)*count, NULL, &ret);
-    memobj_out = clCreateBuffer(context, CL_MEM_READ_WRITE, sizeof(int)*count, NULL, &ret);
-    data_count = clCreateBuffer(context, CL_MEM_READ_WRITE, sizeof(int)*count, NULL, &ret);
+    cl_mem memobj_in  = clCreateBuffer(context, CL_MEM_READ_WRITE, sizeof(int)*count, NULL, &ret);
+    cl_mem memobj_out = clCreateBuffer(context, CL_MEM_READ_WRITE, sizeof(int)*count, NULL, &ret);
+    cl_mem data_count = clCreateBuffer(context, CL_MEM_READ_WRITE, sizeof(int)*count, NULL, &ret);
     
-    cl_command_queue command_queue = gpuThreads->command_queue;
     ret = clEnqueueWriteBuffer(command_queue, memobj_in, CL_TRUE, 0,
                                sizeof(int)*count, indata, 0, NULL, NULL);
     ret = clEnqueueWriteBuffer(command_queue, data_count, CL_TRUE, 0,
                                sizeof(count), &count, 0, NULL, NULL);
-
+    
     print_data(indata, count, "before");
-
-    cl_kernel *kernel = task_list[1].kernel;    
-    clSetKernelArg(*kernel, 0, sizeof(cl_mem), (void *)&memobj_in);
-    clSetKernelArg(*kernel, 1, sizeof(cl_mem), (void *)&memobj_out);
-    clSetKernelArg(*kernel, 2, sizeof(cl_mem), (void *)&data_count);
-
-    cl_event ev;
-    clEnqueueTask(command_queue, *kernel, 0, NULL,  &ev);
+    
+    cl_kernel kernel = NULL;
+    kernel= *(task_list[cmd].kernel);
+    
+    clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *)&memobj_in);
+    clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *)&memobj_out);
+    clSetKernelArg(kernel, 2, sizeof(cl_mem), (void *)&data_count);
+    
+    cl_event ev = NULL;
+    clEnqueueTask(command_queue, kernel, 0, NULL,  &ev);
     
     clEnqueueReadBuffer(command_queue, memobj_out, CL_TRUE, 0,
                         sizeof(int)*count, outdata, 1, &ev, NULL);
+
     print_data(outdata, count, "after");
 
-    free(indata);
-    free(outdata);
-    clReleaseCommandQueue(command_queue);
-    clReleaseContext(context);
-
     tester(indata,outdata,count);
 
-    delete gpuThreads;
+    delete [] indata;
+    delete [] outdata;
+    clReleaseCommandQueue(command_queue);
+    clReleaseContext(context);
+    clReleaseKernel(kernel);
+
+
 }
 
 int
 main(int argc, char* argv[])
 {   
     int task_array_num = DEFAULT;
-
-    if ( argv[1]){
-      task_array_num = atoi(argv[1]);
-      // 無効な引数ならデフォルトの値として5を設定
+    if (argc > 1){
+        if(atoi(argv[1])){
+            task_array_num = atoi(argv[1]);
+        }
     }
     test(task_array_num);
     printf("regist task succeed\n");