Mercurial > hg > Game > Cerium

changeset 1435:9f6ba00ca15c draft

Find changesets by keywords (author, files, the commit message), revision number or hash, or revset expression.
add gpu_sched_register
author Daichi TOMA <toma@cr.ie.u-ryukyu.ac.jp>
date Fri, 06 Apr 2012 07:57:27 +0900
parents 05d480ab70ba
children 42057124fb44
files TaskManager/Gpu/GpuScheduler.cc TaskManager/Gpu/GpuScheduler.h TaskManager/Gpu/GpuTaskManagerImpl.cc TaskManager/Gpu/GpuTaskManagerImpl.h TaskManager/Gpu/GpuThreads.cc TaskManager/Gpu/GpuThreads.h TaskManager/kernel/schedule/Scheduler.h
diffstat 7 files changed, 64 insertions(+), 19 deletions(-) [+]
line wrap: on
line diff
--- a/TaskManager/Gpu/GpuScheduler.cc	Thu Apr 05 19:05:35 2012 +0900
+++ b/TaskManager/Gpu/GpuScheduler.cc	Fri Apr 06 07:57:27 2012 +0900
@@ -1,5 +1,6 @@
 #include "GpuScheduler.h"
 #include "GpuDmaManager.h"
+#include "stdio.h"
 
 void
 GpuScheduler::init_impl(int useRefDma)
@@ -13,3 +14,40 @@
     // command_queueにここで、enqueueする?
 
 }
+
+void
+gpu_register_task(int cmd, char* filename, char* functionname)
+{
+    FILE *fp;
+    char *souce_str;
+    size_t source_size;
+
+    fp = fopen(filename, "r");
+    if (!fp) {
+        fprintf(stderr, "Failed to load kernel.\n");
+        exit(1);
+    }
+
+    source_str = (char*)malloc(MAX_SOURCE_SIZE);
+    source_size = fread(source_str, 1, MAX_SOURCE_SIZE, fp);
+    fclose(fp);
+
+    // contextとか、device_idはどうやって取る？
+    // gpuの情報はグローバルで持つべきなのか…
+    cl_program program = NULL;
+    cl_int ret;
+    program = clCreateProgramWithSource(context, 1, (const char **)&source_str,
+                                               (const size_t *)&source_size, &ret);
+
+    clBuildProgram(program, 1, &device_id, NULL, NULL, NULL);
+
+    cl_kernel *kernel = new cl_kernel; 
+    *kernel = clCreateKernel(program, functionname, &ret);
+    
+    task_list[cmd].run = run;
+    task_list[cmd].load = null_loader;
+    task_list[cmd].wait = null_waiter;
+    task_list[cmd].name = str;
+    task_list[cmd].kernel = kernel;
+
+}
--- a/TaskManager/Gpu/GpuScheduler.h	Thu Apr 05 19:05:35 2012 +0900
+++ b/TaskManager/Gpu/GpuScheduler.h	Fri Apr 06 07:57:27 2012 +0900
@@ -23,3 +23,6 @@
 };
 
 #endif
+
+#define GpuSchedRegister(str, filename, functionname)   \
+    gpu_register_task(str, filename, functionname);
--- a/TaskManager/Gpu/GpuTaskManagerImpl.cc	Thu Apr 05 19:05:35 2012 +0900
+++ b/TaskManager/Gpu/GpuTaskManagerImpl.cc	Fri Apr 06 07:57:27 2012 +0900
@@ -17,20 +17,22 @@
     set_scheduler(mscheduler);
     ppeManager->init(mscheduler, this, useRefDma);
 
-    ret = clGetPlatformIDs(1, &platfrom_id, &ret_num_platforms); 
-    ret = clGetDeviceIds(platform_id, CL_DEVICE_TYPE_GPU, 1, &device_id, &ret_num_devices);
+    htaskImpl = activeTaskQueue;
+    mscheduler->set_manager(this);
 
+    clGetPlatformIDs(1, &platfrom_id, &ret_num_platforms); 
+    clGetDeviceIds(platform_id, CL_DEVICE_TYPE_GPU, 1, &device_id, &ret_num_devices);
     // unavailable GPU
     if( ret_num_devices == 0) {
         exit(EXIT_FAILURE);
     }
-
-    gpuThreads->init(device_id);
+    context = clCreateContext(NULL, 1, &device_id, NULL, NULL, &ret);
+    command_queue = clCreateCommandQueue(context, device_id, 0, &ret);
 
-    htaskImpl = activeTaskQueue;
-    mscheduler->set_manager(this);
+    gpuThreads->init(&command_queue);
 
     schedTaskManager = new SchedTask();
     schedTaskManager->init(0, 0, 0, ppeManager->get_scheduler(), 0);
     ppeManager->schedTaskManager = schedTaskManager;
 }
+
--- a/TaskManager/Gpu/GpuTaskManagerImpl.h	Thu Apr 05 19:05:35 2012 +0900
+++ b/TaskManager/Gpu/GpuTaskManagerImpl.h	Fri Apr 06 07:57:27 2012 +0900
@@ -34,8 +34,11 @@
     cl_device_id device_id;
     cl_uint ret_num_platforms;
     cl_uint ret_num_devices;
-    cl_int ret;
+    cl_context context;
+    cl_command_queue command_queue;
+
 
 };
 
 #endif
+
--- a/TaskManager/Gpu/GpuThreads.cc	Thu Apr 05 19:05:35 2012 +0900
+++ b/TaskManager/Gpu/GpuThreads.cc	Fri Apr 06 07:57:27 2012 +0900
@@ -7,8 +7,6 @@
     threads = new pthread_t;
     args = new gpu_arg;
 
-    context = NULL;
-    command_queue = NULL;
 }
 
 GpuThreads::~GpuThreads()
@@ -18,14 +16,11 @@
 }
 
 void
-GpuThreads::init(cl_device_id device_id)
+GpuThreads::init(cl_command_queue* _command_queue)
 {
-    context = clCreateContext(NULL, 1, &device_id, NULL, NULL, &ret);
-    command_queue = clCreateCommandQueue(context, device_id, 0, &ret);
-
     args.scheduler = new GpuScheduler();
     args.useRefDma = use_refdma;
-    args.command_queue = &command_queue;
+    args.command_queue = _command_queue;
 
     pthread_create(&threads, NULL, &cpu_thread_run, (void*)&args);
 
--- a/TaskManager/Gpu/GpuThreads.h	Thu Apr 05 19:05:35 2012 +0900
+++ b/TaskManager/Gpu/GpuThreads.h	Fri Apr 06 07:57:27 2012 +0900
@@ -24,10 +24,8 @@
     void init(cl_device_id device_id;);
     void *gpu_thread_run(void *args);
 
-    cl_context context;
-    cl_command_queue command_queue;
+private:
 
-private:
     gpu_arg *args;
     pthread_t *threads;
     int use_refdma;
--- a/TaskManager/kernel/schedule/Scheduler.h	Thu Apr 05 19:05:35 2012 +0900
+++ b/TaskManager/kernel/schedule/Scheduler.h	Fri Apr 06 07:57:27 2012 +0900
@@ -9,8 +9,13 @@
 #include "DmaManager.h"
 #include "SchedTaskBase.h"
 #include "MemList.h"
-#include "MemHash.h"
+#include "MemHash.h" 
 
+#ifdef __APPLE__
+#include <OpenCL/opencl.h>
+#else
+#include <CL/cl.h>
+#endif
 
 #define MAX_USER_TASK 100
 #define MAX_SYSTEM_TASK 2
@@ -37,9 +42,10 @@
     memaddr end;            
     uint32 entry_offset;        // offset for create();
     MemorySegment *segment;
-	const char *name;
+    const char *name;
     void (*load)(Scheduler *,int);
     void (*wait)(Scheduler *,int);
+    cl_kernel *kernel;
 }  __attribute__ ((aligned (DEFAULT_ALIGNMENT))) //sizeはどれくらい？
       TaskObject, *TaskObjectPtr;