static void release_clobj(void) {
cl_int ret_code;
ret_code = clEnqueueUnmapMemObject(queue[ocl_gpu_id], pinned_partial_hashes,
calculated_hash, 0, NULL, NULL);
HANDLE_CLERROR(ret_code, "Error Ummapping out_hashes");
ret_code = clEnqueueUnmapMemObject(queue[ocl_gpu_id], pinned_saved_keys,
plaintext, 0, NULL, NULL);
HANDLE_CLERROR(ret_code, "Error Ummapping saved_plain");
ret_code = clReleaseMemObject(salt_buffer);
HANDLE_CLERROR(ret_code, "Error Releasing data_info");
ret_code = clReleaseMemObject(pass_buffer);
HANDLE_CLERROR(ret_code, "Error Releasing buffer_keys");
ret_code = clReleaseMemObject(hash_buffer);
HANDLE_CLERROR(ret_code, "Error Releasing buffer_out");
ret_code = clReleaseMemObject(work_buffer);
HANDLE_CLERROR(ret_code, "Error Releasing work_out");
ret_code = clReleaseMemObject(pinned_saved_keys);
HANDLE_CLERROR(ret_code, "Error Releasing pinned_saved_keys");
ret_code = clReleaseMemObject(pinned_partial_hashes);
HANDLE_CLERROR(ret_code, "Error Releasing pinned_partial_hashes");
}
static void release_clobj(void)
{
HANDLE_CLERROR(clEnqueueUnmapMemObject(queue[ocl_gpu_id], cl_aes_key, aes_key, 0, NULL, NULL), "Error Unmapping aes_key");
HANDLE_CLERROR(clEnqueueUnmapMemObject(queue[ocl_gpu_id], cl_aes_iv, aes_iv, 0, NULL, NULL), "Error Unmapping aes_iv");
HANDLE_CLERROR(clEnqueueUnmapMemObject(queue[ocl_gpu_id], cl_saved_key, saved_key, 0, NULL, NULL), "Error Unmapping saved_key");
HANDLE_CLERROR(clEnqueueUnmapMemObject(queue[ocl_gpu_id], cl_saved_len, saved_len, 0, NULL, NULL), "Error Unmapping saved_len");
HANDLE_CLERROR(clEnqueueUnmapMemObject(queue[ocl_gpu_id], cl_salt, saved_salt, 0, NULL, NULL), "Error Unmapping saved_salt");
aes_key = NULL; aes_iv = NULL; saved_key = NULL; saved_len = NULL; saved_salt = NULL;
}
void allochostptr_roundtrip_func()
{
timer.Start(timer_id);
cl_int err;
// Create buffers with CL_MEM_ALLOC_HOST_PTR for zero copy
buffer_.buf_a_ = clCreateBuffer(ctx_, CL_MEM_READ_ONLY | CL_MEM_ALLOC_HOST_PTR,
(buffer_.lda_*buffer_.a_num_vectors_ +
buffer_.offA_) * sizeof(T),
NULL, &err);
buffer_.buf_b_ = clCreateBuffer(ctx_, CL_MEM_READ_ONLY | CL_MEM_ALLOC_HOST_PTR,
(buffer_.ldb_ * buffer_.b_num_vectors_ +
buffer_.offB_) * sizeof(T),
NULL, &err);
buffer_.buf_c_ = clCreateBuffer(ctx_, CL_MEM_READ_WRITE | CL_MEM_ALLOC_HOST_PTR,
(buffer_.ldc_ * buffer_.c_num_vectors_ +
buffer_.offC_) * sizeof(T),
NULL, &err);
// map the buffers to pointers at host device
T *map_a,*map_b,*map_c;
map_a = (T*)clEnqueueMapBuffer(queues_[0], buffer_.buf_a_, CL_TRUE, CL_MAP_WRITE, 0,
(buffer_.lda_*buffer_.a_num_vectors_ +
buffer_.offA_) * sizeof(T),
0, NULL, NULL, &err);
map_b = (T*)clEnqueueMapBuffer(queues_[0], buffer_.buf_b_, CL_TRUE, CL_MAP_WRITE, 0,
(buffer_.ldb_*buffer_.b_num_vectors_ +
buffer_.offB_) * sizeof(T),
0, NULL, NULL, &err);
map_c = (T*)clEnqueueMapBuffer(queues_[0], buffer_.buf_c_, CL_TRUE, CL_MAP_WRITE, 0,
(buffer_.lda_*buffer_.c_num_vectors_ +
buffer_.offC_) * sizeof(T),
0, NULL, NULL, &err);
// memcpy the input A, B, C to the host pointers
memcpy( map_a, buffer_.a_, ( buffer_.lda_*buffer_.a_num_vectors_ + buffer_.offA_) * sizeof( T ) );
memcpy( map_b, buffer_.b_, ( buffer_.ldb_*buffer_.b_num_vectors_ + buffer_.offB_) * sizeof( T ) );
memcpy( map_c, buffer_.c_, ( buffer_.ldc_*buffer_.c_num_vectors_ + buffer_.offC_) * sizeof( T ) );
// unmap the buffers
clEnqueueUnmapMemObject(queues_[0], buffer_.buf_a_, map_a, 0, NULL, NULL);
clEnqueueUnmapMemObject(queues_[0], buffer_.buf_b_, map_b, 0, NULL, NULL);
clEnqueueUnmapMemObject(queues_[0], buffer_.buf_c_, map_c, 0, NULL, NULL);
// calling clBLAS
xGemm_Function(false);
// map the C buffer again to read output
map_c = (T*)clEnqueueMapBuffer(queues_[0], buffer_.buf_c_, CL_TRUE, CL_MAP_READ, 0,
(buffer_.lda_*buffer_.c_num_vectors_ +
buffer_.offC_) * sizeof(T),
0, NULL, NULL, &err);
memcpy( map_c, buffer_.c_, ( buffer_.ldc_*buffer_.c_num_vectors_ + buffer_.offC_) * sizeof( T ) );
clEnqueueUnmapMemObject(queues_[0], buffer_.buf_c_, map_c, 0, NULL, &event_);
clWaitForEvents(1, &event_);
timer.Stop(timer_id);
}
static void release_clobj(void){
HANDLE_CLERROR(clEnqueueUnmapMemObject(queue[ocl_gpu_id], pinned_partial_hashes, partial_hashes, 0,NULL,NULL), "Error Unmapping partial_hashes");
HANDLE_CLERROR(clEnqueueUnmapMemObject(queue[ocl_gpu_id], pinned_saved_keys, saved_plain, 0, NULL, NULL), "Error Unmapping saved_plain");
HANDLE_CLERROR(clReleaseMemObject(buffer_keys), "Error Releasing buffer_keys");
HANDLE_CLERROR(clReleaseMemObject(buffer_out), "Error Releasing buffer_out");
HANDLE_CLERROR(clReleaseMemObject(pinned_saved_keys), "Error Releasing pinned_saved_keys");
HANDLE_CLERROR(clReleaseMemObject(pinned_partial_hashes), "Error Releasing pinned_partial_hashes");
MEM_FREE(res_hashes);
}
vector<cpx> cl_fft<cpx>::run(const vector<cpx> &input)
{
cl_event upload_unmap_evt, start_evt, download_map_evt, *kernel_evts = new cl_event[launches.size()];
cl_int err;
// Upload
cl_float2 *input_buffer = (cl_float2*)clEnqueueMapBuffer(command_queue,
v_samples,
CL_TRUE,
CL_MAP_WRITE,
0,
samplesMemSize,
0,
NULL,
NULL,
&err);
CL_CHECK_ERR("clEnqueueMapBuffer", err);
for (int i = 0; i < samplesPerRun; i++)
{
input_buffer[i].x = real(input[i]);
input_buffer[i].y = imag(input[i]);
}
CL_CHECK_ERR("clEnqueueUnmapMemObject", clEnqueueUnmapMemObject(command_queue, v_samples, input_buffer, 0, NULL, &upload_unmap_evt));
// Calcola la FFT
cl_mem v_out = runInternal(v_samples, &start_evt, kernel_evts);
// Download
vector<cpx> result(samplesPerRun);
cl_float2 *output_buffer = (cl_float2*)clEnqueueMapBuffer(command_queue,
v_out,
CL_TRUE,
CL_MAP_READ,
0,
tmpMemSize,
1,
&kernel_evts[launches.size() - 1],
&download_map_evt,
&err);
CL_CHECK_ERR("clEnqueueMapBuffer", err);
for (int i = 0; i < samplesPerRun; i++)
result[i] = cpx(output_buffer[i].x, output_buffer[i].y);
CL_CHECK_ERR("clEnqueueUnmapMemObject", clEnqueueUnmapMemObject(command_queue, v_out, output_buffer, 0, NULL, NULL));
printStatsAndReleaseEvents(upload_unmap_evt, start_evt, kernel_evts, download_map_evt);
delete[] kernel_evts;
return result;
}
void init()
{
int i, j;
double t_start_init, t_end_init;
#ifndef ALOCACAO_NORMAL
t_start_init = rtclock();
cl_float* x1_input = (cl_float*)clEnqueueMapBuffer(clCommandQue, x1_mem_obj, CL_TRUE, CL_MAP_WRITE, 0, sizeof(DATA_TYPE) * N, 0, NULL, NULL, &errcode);
cl_float* x2_input = (cl_float*)clEnqueueMapBuffer(clCommandQue, x2_mem_obj, CL_TRUE, CL_MAP_WRITE, 0, sizeof(DATA_TYPE) * N, 0, NULL, NULL, &errcode);
cl_float* y_1_input = (cl_float*)clEnqueueMapBuffer(clCommandQue, y1_mem_obj, CL_TRUE, CL_MAP_WRITE, 0, sizeof(DATA_TYPE) * N, 0, NULL, NULL, &errcode);
cl_float* y_2_input = (cl_float*)clEnqueueMapBuffer(clCommandQue, y2_mem_obj, CL_TRUE, CL_MAP_WRITE, 0, sizeof(DATA_TYPE) * N, 0, NULL, NULL, &errcode);
cl_float* a_input = (cl_float*)clEnqueueMapBuffer(clCommandQue, a_mem_obj, CL_TRUE, CL_MAP_WRITE, 0, sizeof(DATA_TYPE) * N * N, 0, NULL, NULL, &errcode);
t_end_init = rtclock();
tmp_clEnqueueMapBuffer += t_end_init - t_start_init;
#else
DATA_TYPE *y_1_input = y_1;
DATA_TYPE *y_2_input = y_2;
DATA_TYPE *a_input = a;
#endif
t_start_init = rtclock();
for (i=0; i<N; i++)
{
x1[i] = 0.0;
x2[i] = 0.0;
#ifndef ALOCACAO_NORMAL
x1_input[i] = 0.0;
x2_input[i] = 0.0;
#endif
y_1_input[i] = 0.0;
y_2_input[i] = 0.0;
for (j=0; j<N; j++)
{
a_input[i*N + j] = (DATA_TYPE)(i+j+1.0)/N;
}
}
t_end_init = rtclock();
tmp_init += t_end_init - t_start_init;
#ifndef ALOCACAO_NORMAL
t_start_init = rtclock();
clEnqueueUnmapMemObject(clCommandQue, x1_mem_obj, x1_input, 0, NULL, NULL);
clEnqueueUnmapMemObject(clCommandQue, x2_mem_obj, x2_input, 0, NULL, NULL);
clEnqueueUnmapMemObject(clCommandQue, y1_mem_obj, y_1_input, 0, NULL, NULL);
clEnqueueUnmapMemObject(clCommandQue, y2_mem_obj, y_2_input, 0, NULL, NULL);
clEnqueueUnmapMemObject(clCommandQue, a_mem_obj, a_input, 0, NULL, NULL);
t_end_init = rtclock();
tmp_clEnqueueUnmapMemObject += t_end_init - t_start_init;
#endif
}
void QCLVectorBase::unmap() const
{
if (m_mapped) {
#ifndef QT_CL_COPY_VECTOR
cl_int error = clEnqueueUnmapMemObject
(d_ptr->context->activeQueue(), d_ptr->id, m_mapped, 0, 0, 0);
d_ptr->context->reportError("QCLVector<T>::unmap:", error);
#else
// Write the local copy back to the OpenCL device.
if (d_ptr->hostCopy && d_ptr->state == State_InHost) {
cl_int error = clEnqueueWriteBuffer
(d_ptr->context->activeQueue(), d_ptr->id, CL_FALSE,
0, m_size * m_elemSize, d_ptr->hostCopy, 0, 0, 0);
d_ptr->context->reportError("QCLVector<T>::unmap(write):", error);
}
d_ptr->state = State_InKernel;
#endif
m_mapped = 0;
// Update all of the other owners with the unmap state.
if (d_ptr->owners.size() > 1) {
QList<QCLVectorBase *>::Iterator it;
for (it = d_ptr->owners.begin(); it != d_ptr->owners.end(); ++it) {
if (*it != this)
(*it)->m_mapped = 0;
}
}
}
}
void timedBufUnmap( cl_command_queue queue,
cl_mem buf,
void **ptr,
bool quiet )
{
CPerfCounter t1;
cl_int ret;
cl_event ev;
t1.Reset();
t1.Start();
ret = clEnqueueUnmapMemObject( queue,
buf,
(void *) *ptr,
0, NULL, &ev );
ASSERT_CL_RETURN( ret );
clFlush( queue );
spinForEventsComplete( 1, &ev );
t1.Stop();
if( !quiet )
tlog->Timer( "%32s %lf s [ %8.2lf GB/s ]\n", "clEnqueueUnmapMemObject():", t1.GetElapsedTime(), nBytes, 1 );
}
const std::vector<cl_uint> AgentMover::getAgentCounts() const
{
cl_int status;
cl_uint* buffer = static_cast<cl_uint*> ( clEnqueueMapBuffer(context->queue,
agentCounts,
CL_TRUE,
CL_MAP_READ,
0,
sizeof(cl_uint) * 8,
0,
NULL,
NULL,
&status) );
CL_ERROR(status);
std::vector<cl_uint> agentCountsVector(8);
cl_uint i = 0;
std::vector<cl_uint>::iterator iterator;
for (iterator = agentCountsVector.begin(); iterator != agentCountsVector.end(); ++iterator)
{
*iterator = buffer[i];
i++;
}
clEnqueueUnmapMemObject(context->queue, agentCounts, buffer, 0, NULL, NULL);
return agentCountsVector;
}
void*
Alg::eval(std::vector<uint8_t>* pic, std::vector<uint8_t>* picprev){
uint8_t* t = reinterpret_cast<uint8_t*>(pic->data());
uint8_t* tp = reinterpret_cast<uint8_t*>(picprev->data());
// Copy frame to the device
float* result = (float*)clEnqueueMapBuffer(queue, clm_res, CL_FALSE, CL_MAP_READ, 0, 6*sizeof(cl_float), 0, NULL, NULL, &status);
for (int i = 0; i < 6; i++) {
result[i] = 0;
}
// Copy frame to the device
status = clEnqueueWriteBuffer(queue, clm_pic, CL_FALSE, 0, cols * rows * sizeof(cl_uchar), t, 0, NULL, NULL);
status = clEnqueueWriteBuffer(queue, clm_picprev, CL_FALSE, 0, cols * rows * sizeof(cl_uchar), tp, 0, NULL, NULL);
status = clEnqueueWriteBuffer(queue, clm_res, CL_FALSE, 0, 6 * sizeof(float), result, 0, NULL, NULL);
// Execute the OpenCL kernel on the list
status = clEnqueueNDRangeKernel(queue, kern, 1, NULL, &global_threads, NULL, 0, NULL, NULL);
status = clEnqueueReadBuffer(queue, clm_res, CL_FALSE, 0, 6*sizeof(float), result, 0, NULL, NULL);
// Clean up and wait for all the comands to complete.
status = clFlush(queue);
status = clFinish(queue);
for (int i = 0; i < 2; i++) {
percentage[i] = result[i]*100.0 / (cols*rows);
}
for (int i = 0; i < 2; i++) {
percentage[i + 2] = result[i + 2] / (cols*rows);
}
if (result[5])
percentage[4] = result[4] / result[5];
else
percentage[4] = result[4] / 4;
clEnqueueUnmapMemObject(queue, clm_res, result, 0, NULL, NULL);
return percentage;
}
void AgentMover::setAgentCounts(const std::vector<cl_uint>& agentCountsVector)
{
cl_int status;
cl_uint* buffer = static_cast<cl_uint*> ( clEnqueueMapBuffer(context->queue,
agentCounts,
CL_TRUE,
CL_MAP_WRITE,
0,
sizeof(cl_uint) * 8,
0,
NULL,
NULL,
&status) );
CL_ERROR(status);
cl_uint i = 0;
std::vector<cl_uint>::const_iterator iterator;
for (iterator = agentCountsVector.begin(); iterator != agentCountsVector.end(); ++iterator)
{
buffer[i] = *iterator;
i++;
}
clEnqueueUnmapMemObject(context->queue, agentCounts, buffer, 0, NULL, NULL);
}
void mclUnmap(mclContext ctx, mclDeviceData dd, void* ptr)
{
logOclCall("clEnqueueUnmapMemObject");
cl_int ret;
ret = clEnqueueUnmapMemObject(ctx.command_queue, dd.data, ptr, 0, 0, 0);
MCL_VALIDATE(ret, "mclUnmap: Error unmapping data");
}
static void release_clobj(void)
{
HANDLE_CLERROR(clEnqueueUnmapMemObject(queue[gpu_id], pinned_result, cracked, 0, NULL, NULL), "Error Unmapping cracked");
HANDLE_CLERROR(clEnqueueUnmapMemObject(queue[gpu_id], pinned_key, saved_key, 0, NULL, NULL), "Error Unmapping saved_key");
HANDLE_CLERROR(clEnqueueUnmapMemObject(queue[gpu_id], pinned_idx, saved_idx, 0, NULL, NULL), "Error Unmapping saved_idx");
HANDLE_CLERROR(clFinish(queue[gpu_id]), "Error releasing memory mappings");
HANDLE_CLERROR(clReleaseMemObject(pinned_result), "Release pinned result buffer");
HANDLE_CLERROR(clReleaseMemObject(pinned_key), "Release pinned key buffer");
HANDLE_CLERROR(clReleaseMemObject(pinned_idx), "Release pinned index buffer");
HANDLE_CLERROR(clReleaseMemObject(cl_salt), "Release salt buffer");
HANDLE_CLERROR(clReleaseMemObject(cl_result), "Release result buffer");
HANDLE_CLERROR(clReleaseMemObject(cl_saved_key), "Release key buffer");
HANDLE_CLERROR(clReleaseMemObject(cl_saved_idx), "Release index buffer");
HANDLE_CLERROR(clReleaseMemObject(cl_mid_key), "Release state buffer");
}
void*
Optzd::eval(std::vector<uint8_t>* t){
cl_float* sh = (cl_float*)clEnqueueMapBuffer(queue, clm_sh, CL_FALSE, CL_MAP_WRITE, 0, 2*sizeof(cl_float), 0, NULL, NULL, &status);
uint8_t* pic = reinterpret_cast<uint8_t*>(t->data());
sh[0] = 0;
sh[1] = 0;
// Copy frame to the device
status = clEnqueueWriteBuffer(queue, clm_pic, CL_TRUE, 0, cols * rows * sizeof(cl_uchar), pic, 0, NULL, NULL);
status = clEnqueueWriteBuffer(queue, clm_sh, CL_TRUE, 0, 2 * sizeof(cl_float), sh, 0, NULL, NULL);
// Execute the OpenCL kernel on the list
status = clEnqueueNDRangeKernel(queue, kern, 1, NULL, &global_threads, NULL, 0, NULL, NULL);
status = clEnqueueReadBuffer(queue, clm_sh, CL_TRUE, 0, 2*sizeof(float), sh, 0, NULL, NULL);
// Clean up and wait for all the comands to complete.
status = clFlush(queue);
status = clFinish(queue);
//eval-ing BS
float shb = sh[1], shnb = sh[0];
// for (int i = 0; i < (global_threads / 2); i++) {
// shb += sh[i*2];
// shnb += sh[(i*2)+1];
// }
clEnqueueUnmapMemObject(queue, clm_sh, sh, 0, NULL, NULL);
if (shnb == 0) shnb = 4;
*BS = shb/(shnb);
return BS;
}
bool CL_Image3D::Unmap(const CL_CommandQueue* pCommandQueue, void** ppMappedData, CL_Event* pNewEvent, const CL_EventPool* pWaitList)
{
CL_CPP_CONDITIONAL_RETURN_FALSE(!m_Image);
CL_CPP_CONDITIONAL_RETURN_FALSE(!pCommandQueue);
CL_CPP_CONDITIONAL_RETURN_FALSE(!ppMappedData);
cl_uint uNumWaitEvents = pWaitList ? pWaitList->GetNumEvents() : 0;
const cl_event* pWaitEvents = pWaitList ? pWaitList->GetEventPool() : NULL;
cl_event NewEvent = NULL;
// Unmap a location in host memory from a buffer object.
const cl_command_queue CommandQueue = pCommandQueue->GetCommandQueue();
cl_int iErrorCode = clEnqueueUnmapMemObject(CommandQueue, m_Image, *ppMappedData, uNumWaitEvents, pWaitEvents, &NewEvent);
CL_CPP_CATCH_ERROR(iErrorCode);
CL_CPP_ON_ERROR_RETURN_FALSE(iErrorCode);
(*ppMappedData) = NULL;
if(NewEvent)
{
if(pNewEvent)
pNewEvent->SetEvent(NewEvent);
clReleaseEvent(NewEvent);
}
return true;
}
void init()
{
int i, j;
double t_start_init, t_end_init;
#ifndef ALOCACAO_NORMAL
t_start_init = rtclock();
cl_float* data_input = (cl_float*)clEnqueueMapBuffer(clCommandQue, data_mem_obj, CL_TRUE, CL_MAP_WRITE, 0, sizeof(DATA_TYPE) * (M+1) * (N+1), 0, NULL, NULL, &errcode);
t_end_init = rtclock();
tmp_clEnqueueMapBuffer += t_end_init - t_start_init;
#endif
t_start_init = rtclock();
for (i = 0; i < M; i++)
{
for (j = 0; j < N; j++)
{
data[i*(N+1) + j] = ((DATA_TYPE) i*j) / M;
#ifndef ALOCACAO_NORMAL
data_input[i*(N+1) + j] = ((DATA_TYPE) i*j)/ M;
#endif
}
}
t_end_init = rtclock();
tmp_init += t_end_init - t_start_init;
#ifndef ALOCACAO_NORMAL
t_start_init = rtclock();
clEnqueueUnmapMemObject(clCommandQue, data_mem_obj, data_input, 0, NULL, NULL);
t_end_init = rtclock();
tmp_clEnqueueUnmapMemObject += t_end_init - t_start_init;
#endif
}
cl_int GLCLDraw::UnMapTransferBuffer(Uint8 *p)
{
cl_int ret;
if(p == NULL) return CL_INVALID_MEM_OBJECT;
ret = clEnqueueUnmapMemObject(command_queue, inbuf[inbuf_bank],
p, 0, NULL, &event_uploadvram[1]);
return ret;
}
void usepersismem_roundtrip_func()
{
#if defined(CL_MEM_USE_PERSISTENT_MEM_AMD)
timer.Start(timer_id);
//set up buffer
cl_int err;
buffer_.buf_a_ = clCreateBuffer(ctx_, CL_MEM_READ_ONLY | CL_MEM_USE_PERSISTENT_MEM_AMD,
(buffer_.lda_ * buffer_.a_num_vectors_ +
buffer_.offA_) * sizeof(T),
NULL, &err);
buffer_.buf_b_ = clCreateBuffer(ctx_, CL_MEM_READ_WRITE | CL_MEM_USE_PERSISTENT_MEM_AMD,
(buffer_.ldb_ * buffer_.b_num_vectors_ +
buffer_.offB_) * sizeof(T),
NULL, &err);
// Map the buffers to pointers at host device
T *map_a,*map_b;
map_a = (T*)clEnqueueMapBuffer(queue_, buffer_.buf_a_, CL_TRUE, CL_MAP_WRITE, 0,
(buffer_.ldb_ * buffer_.b_num_vectors_ +
buffer_.offB_) * sizeof(T),
0, NULL, NULL, &err);
map_b = (T*)clEnqueueMapBuffer(queue_, buffer_.buf_b_, CL_TRUE, CL_MAP_WRITE, 0,
(buffer_.ldb_ * buffer_.b_num_vectors_ +
buffer_.offB_) * sizeof(T),
0, NULL, NULL, &err);
// memcpy the input A, B to the mapped regions
memcpy( map_a, buffer_.a_, ( buffer_.lda_*buffer_.a_num_vectors_ + buffer_.offA_) * sizeof( T ) );
memcpy( map_b, buffer_.b_, ( buffer_.ldb_*buffer_.b_num_vectors_ + buffer_.offB_) * sizeof( T ) );
// unmap the buffers
clEnqueueUnmapMemObject(queue_, buffer_.buf_a_, map_a, 0, NULL, NULL);
clEnqueueUnmapMemObject(queue_, buffer_.buf_b_, map_b, 0, NULL, NULL);
//call func
xTrsm_Function(false);
// map the B buffer again to read the output
map_b = (T*)clEnqueueMapBuffer(queue_, buffer_.buf_b_, CL_TRUE, CL_MAP_READ, 0,
(buffer_.ldb_ * buffer_.b_num_vectors_ +
buffer_.offB_) * sizeof(T),
0, NULL, NULL, &err);
memcpy( map_b, buffer_.b_, ( buffer_.ldb_*buffer_.b_num_vectors_ + buffer_.offB_) * sizeof( T ) );
clEnqueueUnmapMemObject(queue_, buffer_.buf_b_, map_b, 0, NULL, NULL);
clWaitForEvents(1, &event_);
timer.Stop(timer_id);
#else
std::cout<<"CL_MEM_USE_PERSISTENT_MEM_AMD is only supported on AMD hardware"<<std::endl;
#endif
}
cl_int WINAPI wine_clEnqueueUnmapMemObject(cl_command_queue command_queue, cl_mem memobj, void * mapped_ptr,
cl_uint num_events_in_wait_list, cl_event * event_wait_list, cl_event * event)
{
cl_int ret;
TRACE("\n");
ret = clEnqueueUnmapMemObject(command_queue, memobj, mapped_ptr, num_events_in_wait_list, event_wait_list, event);
return ret;
}
// Function to clean up and exit
//*****************************************************************************
void Cleanup(int iExitCode)
{
// Cleanup allocated objects
shrLog("\nStarting Cleanup...\n\n");
// Release all the OpenCL Objects
if(cpProgram)clReleaseProgram(cpProgram);
for (cl_uint i = 0; i < GpuDevMngr->uiUsefulDevCt; i++)
{
if(ckSobel[i])clReleaseKernel(ckSobel[i]);
if(cmDevBufIn[i])clReleaseMemObject(cmDevBufIn[i]);
if(cmDevBufOut[i])clReleaseMemObject(cmDevBufOut[i]);
}
if(uiInput)clEnqueueUnmapMemObject(cqCommandQueue[0], cmPinnedBufIn, (void*)uiInput, 0, NULL, NULL);
if(uiOutput)clEnqueueUnmapMemObject(cqCommandQueue[0], cmPinnedBufOut, (void*)uiOutput, 0, NULL, NULL);
if(cmPinnedBufIn)clReleaseMemObject(cmPinnedBufIn);
if(cmPinnedBufOut)clReleaseMemObject(cmPinnedBufOut);
for (cl_uint i = 0; i < GpuDevMngr->uiUsefulDevCt; i++)
{
if(cqCommandQueue[i])clReleaseCommandQueue(cqCommandQueue[i]);
}
if(cxGPUContext)clReleaseContext(cxGPUContext);
// free the host allocs
if(cSourceCL)free(cSourceCL);
if(cPathAndName)free(cPathAndName);
if(cmDevBufIn) delete [] cmDevBufIn;
if(cmDevBufOut) delete [] cmDevBufOut;
if(szAllocDevBytes) delete [] szAllocDevBytes;
if(uiInHostPixOffsets) delete [] uiInHostPixOffsets;
if(uiOutHostPixOffsets) delete [] uiOutHostPixOffsets;
if(uiDevImageHeight) delete [] uiDevImageHeight;
if(GpuDevMngr) delete GpuDevMngr;
if(cqCommandQueue) delete [] cqCommandQueue;
// Cleanup GL objects if used
if (!bQATest)
{
DeInitGL();
}
shrLogEx(LOGBOTH | CLOSELOG, 0, "%s Exiting...\n", cExecutableName);
shrQAFinishExit2(bQATest, *pArgc, (const char **)pArgv, ( iExitCode == EXIT_SUCCESS ) ? QA_PASSED : QA_FAILED);
}
static void release_clobj(void){
cl_int ret_code;
ret_code = clEnqueueUnmapMemObject(queue[gpu_id], pin_part_msha_hashes, par_msha_hashes, 0,NULL,NULL);
HANDLE_CLERROR(ret_code, "Error Ummapping par_msha_hashes");
ret_code = clEnqueueUnmapMemObject(queue[gpu_id], pinned_msha_keys, mysqlsha_plain, 0, NULL, NULL);
HANDLE_CLERROR(ret_code, "Error Ummapping mysqlsha_plain");
ret_code = clReleaseMemObject(buf_msha_keys);
HANDLE_CLERROR(ret_code, "Error Releasing buf_msha_keys");
ret_code = clReleaseMemObject(buf_msha_out);
HANDLE_CLERROR(ret_code, "Error Releasing buf_msha_out");
ret_code = clReleaseMemObject(data_info);
HANDLE_CLERROR(ret_code, "Error Releasing data_info");
ret_code = clReleaseMemObject(pinned_msha_keys);
HANDLE_CLERROR(ret_code, "Error Releasing pinned_msha_keys");
ret_code = clReleaseMemObject(pin_part_msha_hashes);
HANDLE_CLERROR(ret_code, "Error Releasing pin_part_msha_hashes");
free(res_hashes);
}
static void release_clobj(void){
cl_int ret_code;
ret_code = clEnqueueUnmapMemObject(queue[ocl_gpu_id], pinned_partial_hashes, partial_hashes, 0,NULL,NULL);
HANDLE_CLERROR(ret_code, "Error Ummapping partial_hashes");
ret_code = clEnqueueUnmapMemObject(queue[ocl_gpu_id], pinned_saved_keys, saved_plain, 0, NULL, NULL);
HANDLE_CLERROR(ret_code, "Error Ummapping saved_plain");
ret_code = clReleaseMemObject(buffer_keys);
HANDLE_CLERROR(ret_code, "Error Releasing buffer_keys");
ret_code = clReleaseMemObject(buffer_out);
HANDLE_CLERROR(ret_code, "Error Releasing buffer_out");
ret_code = clReleaseMemObject(data_info);
HANDLE_CLERROR(ret_code, "Error Releasing data_info");
ret_code = clReleaseMemObject(pinned_saved_keys);
HANDLE_CLERROR(ret_code, "Error Releasing pinned_saved_keys");
ret_code = clReleaseMemObject(pinned_partial_hashes);
HANDLE_CLERROR(ret_code, "Error Releasing pinned_partial_hashes");
free(res_hashes);
}
void allochostptr_roundtrip_func()
{
timer.Start(timer_id);
//set up buffer
cl_int err;
buffer_.buf_a_ = clCreateBuffer(ctx_, CL_MEM_READ_ONLY | CL_MEM_ALLOC_HOST_PTR,
(buffer_.lda_ * buffer_.a_num_vectors_ +
buffer_.offA_) * sizeof(T),
NULL, &err);
buffer_.buf_b_ = clCreateBuffer(ctx_, CL_MEM_READ_WRITE | CL_MEM_ALLOC_HOST_PTR,
(buffer_.ldb_ * buffer_.b_num_vectors_ +
buffer_.offB_) * sizeof(T),
NULL, &err);
// Map the buffers to pointers at host device
T *map_a,*map_b;
map_a = (T*)clEnqueueMapBuffer(queue_, buffer_.buf_a_, CL_TRUE, CL_MAP_WRITE, 0,
(buffer_.ldb_ * buffer_.b_num_vectors_ +
buffer_.offB_) * sizeof(T),
0, NULL, NULL, &err);
map_b = (T*)clEnqueueMapBuffer(queue_, buffer_.buf_b_, CL_TRUE, CL_MAP_WRITE, 0,
(buffer_.ldb_ * buffer_.b_num_vectors_ +
buffer_.offB_) * sizeof(T),
0, NULL, NULL, &err);
// memcpy the input A, B to the mapped regions
memcpy( map_a, buffer_.a_, ( buffer_.lda_*buffer_.a_num_vectors_ + buffer_.offA_) * sizeof( T ) );
memcpy( map_b, buffer_.b_, ( buffer_.ldb_*buffer_.b_num_vectors_ + buffer_.offB_) * sizeof( T ) );
// unmap the buffers
clEnqueueUnmapMemObject(queue_, buffer_.buf_a_, map_a, 0, NULL, NULL);
clEnqueueUnmapMemObject(queue_, buffer_.buf_b_, map_b, 0, NULL, NULL);
//call func
xTrsm_Function(false);
// map the B buffer again to read the output
map_b = (T*)clEnqueueMapBuffer(queue_, buffer_.buf_b_, CL_TRUE, CL_MAP_READ, 0,
(buffer_.ldb_ * buffer_.b_num_vectors_ +
buffer_.offB_) * sizeof(T),
0, NULL, NULL, &err);
memcpy( map_b, buffer_.b_, ( buffer_.ldb_*buffer_.b_num_vectors_ + buffer_.offB_) * sizeof( T ) );
clEnqueueUnmapMemObject(queue_, buffer_.buf_b_, map_b, 0, NULL, NULL);
clWaitForEvents(1, &event_);
timer.Stop(timer_id);
}
static void release_clobj(void)
{
HANDLE_CLERROR(clEnqueueUnmapMemObject(queue[ocl_gpu_id], pinned_key, key, 0, NULL, NULL), "Error Unmapping key");
HANDLE_CLERROR(clEnqueueUnmapMemObject(queue[ocl_gpu_id], pinned_saved_key, saved_key, 0, NULL, NULL), "Error Unmapping saved_key");
HANDLE_CLERROR(clEnqueueUnmapMemObject(queue[ocl_gpu_id], pinned_saved_len, saved_len, 0, NULL, NULL), "Error Unmapping saved_len");
HANDLE_CLERROR(clEnqueueUnmapMemObject(queue[ocl_gpu_id], pinned_salt, saved_salt, 0, NULL, NULL), "Error Unmapping saved_salt");
HANDLE_CLERROR(clFinish(queue[ocl_gpu_id]), "Error releasing memory mappings");
HANDLE_CLERROR(clReleaseMemObject(pinned_key), "Release GPU buffer");
HANDLE_CLERROR(clReleaseMemObject(pinned_saved_key), "Release GPU buffer");
HANDLE_CLERROR(clReleaseMemObject(pinned_saved_len), "Release GPU buffer");
HANDLE_CLERROR(clReleaseMemObject(pinned_salt), "Release GPU buffer");
HANDLE_CLERROR(clReleaseMemObject(cl_key), "Release GPU buffer");
HANDLE_CLERROR(clReleaseMemObject(cl_saved_key), "Release GPU buffer");
HANDLE_CLERROR(clReleaseMemObject(cl_saved_len), "Release GPU buffer");
HANDLE_CLERROR(clReleaseMemObject(cl_salt), "Release GPU buffer");
HANDLE_CLERROR(clReleaseMemObject(cl_pwhash), "Release GPU buffer");
MEM_FREE(cracked);
}
// buffer unmap (asynchronous)
void
HGPU_GPU_buffer_unmap_async(HGPU_GPU_buffer* buffer,cl_command_queue queue,void* data_ptr){
if (!buffer){
HGPU_GPU_error_note(HGPU_ERROR_BAD_BUFFER,"Trying to use null-buffer!");
return;
}
cl_event buffer_event = NULL;
HGPU_GPU_error_message(clEnqueueUnmapMemObject(queue,buffer->buffer,data_ptr,0,NULL,&buffer_event),"clEnqueueUnmapMemObject failed");
buffer->buffer_write_event = buffer_event;
buffer->buffer_write_number_of++;
}
/*!
\param mem cl_mem object
\param ptr A host pointer that points to the mapped region
*/
void cl_unmapBuffer(cl_mem mem, void *ptr)
{
// TODO It looks like AMD doesn't support profiling unmapping yet. Leaving the
// commented code here until it's supported
cl_int status;
status = clEnqueueUnmapMemObject(commandQueue, mem, ptr, 0, NULL, NULL);
cl_errChk(status, "Error unmapping a buffer or image", true);
}
static void release_clobj(void){
cl_int ret_code;
ret_code = clEnqueueUnmapMemObject(queue[gpu_id], pinned_partial_hashes, outbuffer, 0,NULL,NULL);
HANDLE_CLERROR(ret_code, "Error Ummapping outbuffer");
ret_code = clEnqueueUnmapMemObject(queue[gpu_id], pinned_saved_keys, inbuffer, 0, NULL, NULL);
HANDLE_CLERROR(ret_code, "Error Ummapping inbuffer");
ret_code = clReleaseMemObject(buffer_keys);
HANDLE_CLERROR(ret_code, "Error Releasing buffer_keys");
ret_code = clReleaseMemObject(buffer_out);
HANDLE_CLERROR(ret_code, "Error Releasing buffer_out");
ret_code = clReleaseMemObject(data_info);
HANDLE_CLERROR(ret_code, "Error Releasing data_info");
ret_code = clReleaseMemObject(mysalt);
HANDLE_CLERROR(ret_code, "Error Releasing mysalt");
ret_code = clReleaseMemObject(pinned_saved_keys);
HANDLE_CLERROR(ret_code, "Error Releasing pinned_saved_keys");
ret_code = clReleaseMemObject(pinned_partial_hashes);
HANDLE_CLERROR(ret_code, "Error Releasing pinned_partial_hashes");
free(outbuffer2);
}
cl_int
clEnqueueUnmapMemObject_test(cl_command_queue command_queue,
cl_mem memobj,
void * mapped_ptr,
cl_uint num_events_in_wait_list,
const cl_event * event_wait_list,
cl_event * event)
{
printf("clEnqueueUnmapMemObject_test: memobj==%p ptr==%p event==%p\n", memobj, mapped_ptr, event);
return clEnqueueUnmapMemObject(command_queue, memobj, mapped_ptr,
num_events_in_wait_list, event_wait_list, event);
}
int
ComputeBench::unmapBuffer(cl_mem deviceBuffer, void* hostPointer)
{
cl_int status;
status = clEnqueueUnmapMemObject(commandQueue,
deviceBuffer,
hostPointer,
0,
NULL,
NULL);
CHECK_OPENCL_ERROR(status, "clEnqueueUnmapMemObject failed");
return SDK_SUCCESS;
}
int GLCLDraw::ReleasePixelBuffer(Uint32 *p)
{
#if 0
return 0;
#else
int ret;
if(p == NULL) return 0;
// clFlush(command_queue);
ret |= clEnqueueUnmapMemObject(command_queue, outbuf,
p, 1, &event_release, NULL);
clFinish(command_queue);
return ret;
#endif
}
