JNIEXPORT jint JNICALL Java_org_lwjgl_opencl_CL20_nclSetKernelArgSVMPointer(JNIEnv *__env, jclass clazz, jlong kernelAddress, jint arg_index, jlong arg_valueAddress, jlong __functionAddress) {
cl_kernel kernel = (cl_kernel)(intptr_t)kernelAddress;
const void *arg_value = (const void *)(intptr_t)arg_valueAddress;
clSetKernelArgSVMPointerPROC clSetKernelArgSVMPointer = (clSetKernelArgSVMPointerPROC)(intptr_t)__functionAddress;
UNUSED_PARAMS(__env, clazz)
return (jint)clSetKernelArgSVMPointer(kernel, arg_index, arg_value);
}
static void set(cl::Kernel &k, unsigned argpos, const svm_ptr<T> &arg) {
cl_int ret = clSetKernelArgSVMPointer(k(), argpos, arg.ptr);
if (ret != CL_SUCCESS) throw cl::Error(ret, "clSetKernelArgSVMPointer");
}
DWORD CFrontEnd::dwCFrontThread()
{
size_t pLocalSize[1] = {OCL_WG_SIZE};
size_t pGlobalSize[1] = {OCL_WG_SIZE};
DWORD dwWaitStatus = WaitForSingleObject(tCloseThreadEvent,0);
while(dwWaitStatus == WAIT_TIMEOUT)
{
//service the queued requests
TQueuedRequest cQReq;
BOOL bSubmitFlag = FALSE;
UINT uiReqCount = 0;
while(bSubmitFlag == FALSE)
{
if(uiReqCount < OCL_REQ_QUEUE_SIZE)
{
if(tReqQueue.uiGet(&cQReq) == HTS_OK)
{
pOclReqQueue[uiReqCount].tFid = cQReq.tFid;
pOclReqQueue[uiReqCount].uiReqId = cQReq.uiReqId;
pOclReqQueue[uiReqCount].uiType = cQReq.uiType;
pOclReqQueue[uiReqCount].uiKey = cQReq.uiKey;
pOclReqQueue[uiReqCount].uiFlags = cQReq.uiFlags;
pOclReqQueue[uiReqCount].uiStatus = cQReq.uiStatus;
//std::cout << "k:" << pOclReqQueue[uiReqCount].uiKey;
uiReqCount++;
}
else
{
bSubmitFlag = TRUE;
}
}
else
{
bSubmitFlag = TRUE;
}
}
//std::cout << "submitting request:" << uiReqCount << std::endl;
//submit all requests to GPU
if(uiReqCount > 0)
{
UINT uiGlobalSize = uiReqCount*OCL_WG_SIZE;
pGlobalSize[0] = (size_t)(uiGlobalSize);
pLocalSize[0] = OCL_WG_SIZE;
clSetKernelArgSVMPointer(oclKernel,
0,
(void *)(pOclReqQueue));
clSetKernelArgSVMPointer(oclKernel,
1,
(void *)(pNodePool));
clSetKernelArgSVMPointer(oclKernel,
2,
(void *)(pMiscData));
clSetKernelArg(oclKernel,
3,
sizeof(cl_uint),
(void *)(&uiReqCount));
clEnqueueNDRangeKernel(pOclContext->oclCommandQueue,
oclKernel,
1,
NULL,
pGlobalSize,
pLocalSize,
0,
NULL,
NULL);
clFinish(pOclContext->oclCommandQueue);
//std::cout << "finished kernel." << std::endl;
}
//update the request status to each thread
for (unsigned int i = 0; i < uiReqCount; ++i)
{
TFid tFid = pOclReqQueue[i].tFid;
cl_uint uiReqId = pOclReqQueue[i].uiReqId;
//(tFid->pThreadRequest[uiReqId]).uiFlags
//= pOclReqQueue[i].uiFlags | HTS_REQ_COMPLETED;
(tFid->pThreadRequest[uiReqId]).uiStatus = pOclReqQueue[i].uiStatus;
(tFid->pThreadRequest[uiReqId]).uiFlags
= pOclReqQueue[i].uiFlags;
tFid->uiReqCount--;
SetEvent(tFid->hBlockEvent);
}
dwWaitStatus = WaitForSingleObject(tCloseThreadEvent,0);
}
return HTS_OK;
}
