/// \brief Wait for the event to occur.
void wait()
{
if (event_.load(boost::memory_order_acquire))
return;
typename mutex_type::scoped_lock l(mtx_);
wait_locked(l);
}
bool Event::wait(TimeScope time_scope)
{
Dev::IntLock lock;
auto timeout=time_scope.get();
if( !timeout ) return false;
return wait_locked(timeout);
}
/// \brief Try to wait for the semaphore to be signaled
///
/// \param count [in] The value by which the internal lock count will
/// be decremented. At the same time this is the minimum
/// value of the lock count at which the thread is not
/// yielded.
///
/// \returns The function returns true if the calling thread was
/// able to acquire the requested amount of credits.
/// The function returns false if not sufficient credits
/// are available at this point in time.
bool try_wait(boost::int64_t count = 1)
{
typename mutex_type::scoped_lock l(mtx_);
if (!(value_ < count)) {
// enter wait_locked only if there are sufficient credits
// available
wait_locked(count, l);
return true;
}
return false;
}
/// \brief Wait for the semaphore to be signaled
///
/// \param count [in] The value by which the internal lock count will
/// be decremented. At the same time this is the minimum
/// value of the lock count at which the thread is not
/// yielded.
void wait(boost::int64_t count = 1)
{
typename mutex_type::scoped_lock l(mtx_);
wait_locked(count, l);
}
size_t _cl_command_queue::proc()
{
while(!b_stop)
{
b_working = false;
lock();
while (queue.empty())
{
wait_locked();
if (b_stop)
{
unlock();
return 0;
}
}
FreeOCL::smartptr<FreeOCL::command> cmd = queue.front();
b_working = true;
queue.pop_front();
unlock();
if (b_stop)
break;
if (!is_command_ready_to_process(cmd))
{
if (properties & CL_QUEUE_OUT_OF_ORDER_EXEC_MODE_ENABLE)
{
// Find something else to process
std::deque<FreeOCL::smartptr<FreeOCL::command> > waiting_queue;
waiting_queue.push_front(cmd);
bool b_found = false;
lock();
while(!queue.empty() && !b_found)
{
cmd = queue.front();
queue.pop_front();
unlock();
if (cmd->get_type() == CL_COMMAND_MARKER)
{
lock();
queue.push_front(cmd);
break;
}
b_found = is_command_ready_to_process(cmd);
if (!b_found)
waiting_queue.push_front(cmd);
lock();
}
for(std::deque<FreeOCL::smartptr<FreeOCL::command> >::const_iterator i = waiting_queue.begin() ; i != waiting_queue.end() ; ++i)
queue.push_front(*i);
if (!b_found) // No choice, we must try later
{
wait_locked();
unlock();
continue;
}
unlock();
}
else
{
// Wait for events (those events are likely to come from another command queue or are user events)
clWaitForEventsFCL(cmd->num_events_in_wait_list, cmd->event_wait_list);
}
}
if (cmd->event)
{
cmd->event->lock();
cmd->event->change_status(CL_RUNNING);
cmd->event->unlock();
}
switch(cmd->get_type())
{
case CL_COMMAND_READ_IMAGE:
case CL_COMMAND_READ_BUFFER_RECT:
{
char *dst_ptr = (char*)cmd.as<FreeOCL::command_read_buffer_rect>()->ptr;
const char *src_ptr = (const char*)cmd.as<FreeOCL::command_read_buffer_rect>()->buffer->ptr
+ cmd.as<FreeOCL::command_read_buffer_rect>()->offset;
const size_t src_row_pitch = cmd.as<FreeOCL::command_read_buffer_rect>()->buffer_pitch[0];
const size_t dst_row_pitch = cmd.as<FreeOCL::command_read_buffer_rect>()->host_pitch[0];
const size_t cb[3] = { cmd.as<FreeOCL::command_read_buffer_rect>()->cb[0],
cmd.as<FreeOCL::command_read_buffer_rect>()->cb[1],
cmd.as<FreeOCL::command_read_buffer_rect>()->cb[2] };
const size_t src_slice_pitch = cmd.as<FreeOCL::command_read_buffer_rect>()->buffer_pitch[1]
- src_row_pitch * cb[1];
const size_t dst_slice_pitch = cmd.as<FreeOCL::command_read_buffer_rect>()->host_pitch[1]
- dst_row_pitch * cb[1];
for(size_t z = 0 ; z < cb[2] ; ++z, src_ptr += src_slice_pitch, dst_ptr += dst_slice_pitch)
for(size_t y = 0 ; y < cb[1] ; ++y, src_ptr += src_row_pitch, dst_ptr += dst_row_pitch)
memcpy(dst_ptr, src_ptr, cb[0]);
}
break;
case CL_COMMAND_WRITE_IMAGE:
case CL_COMMAND_WRITE_BUFFER_RECT:
{
const char *src_ptr = (const char*)cmd.as<FreeOCL::command_write_buffer_rect>()->ptr;
char *dst_ptr = (char*)cmd.as<FreeOCL::command_write_buffer_rect>()->buffer->ptr
+ cmd.as<FreeOCL::command_write_buffer_rect>()->offset;
const size_t dst_row_pitch = cmd.as<FreeOCL::command_write_buffer_rect>()->buffer_pitch[0];
const size_t src_row_pitch = cmd.as<FreeOCL::command_write_buffer_rect>()->host_pitch[0];
const size_t cb[3] = { cmd.as<FreeOCL::command_write_buffer_rect>()->cb[0],
cmd.as<FreeOCL::command_write_buffer_rect>()->cb[1],
cmd.as<FreeOCL::command_write_buffer_rect>()->cb[2] };
const size_t dst_slice_pitch = cmd.as<FreeOCL::command_write_buffer_rect>()->buffer_pitch[1]
- dst_row_pitch * cb[1];
const size_t src_slice_pitch = cmd.as<FreeOCL::command_write_buffer_rect>()->host_pitch[1]
- src_row_pitch * cb[1];
for(size_t z = 0 ; z < cb[2] ; ++z, src_ptr += src_slice_pitch, dst_ptr += dst_slice_pitch)
for(size_t y = 0 ; y < cb[1] ; ++y, src_ptr += src_row_pitch, dst_ptr += dst_row_pitch)
memcpy(dst_ptr, src_ptr, cb[0]);
}
break;
case CL_COMMAND_COPY_IMAGE_TO_BUFFER:
case CL_COMMAND_COPY_BUFFER_TO_IMAGE:
case CL_COMMAND_COPY_IMAGE:
case CL_COMMAND_COPY_BUFFER_RECT:
{
const char *src_ptr = (const char*)cmd.as<FreeOCL::command_copy_buffer_rect>()->src_buffer->ptr
+ cmd.as<FreeOCL::command_copy_buffer_rect>()->src_offset;
char *dst_ptr = (char*)cmd.as<FreeOCL::command_copy_buffer_rect>()->dst_buffer->ptr
+ cmd.as<FreeOCL::command_copy_buffer_rect>()->dst_offset;
const size_t dst_row_pitch = cmd.as<FreeOCL::command_copy_buffer_rect>()->dst_pitch[0];
const size_t src_row_pitch = cmd.as<FreeOCL::command_copy_buffer_rect>()->src_pitch[0];
const size_t cb[3] = { cmd.as<FreeOCL::command_copy_buffer_rect>()->cb[0],
cmd.as<FreeOCL::command_copy_buffer_rect>()->cb[1],
cmd.as<FreeOCL::command_copy_buffer_rect>()->cb[2] };
const size_t dst_slice_pitch = cmd.as<FreeOCL::command_copy_buffer_rect>()->dst_pitch[1]
- dst_row_pitch * cb[1];
const size_t src_slice_pitch = cmd.as<FreeOCL::command_copy_buffer_rect>()->src_pitch[1]
- src_row_pitch * cb[1];
for(size_t z = 0 ; z < cb[2] ; ++z, src_ptr += src_slice_pitch, dst_ptr += dst_slice_pitch)
for(size_t y = 0 ; y < cb[1] ; ++y, src_ptr += src_row_pitch, dst_ptr += dst_row_pitch)
memcpy(dst_ptr, src_ptr, cb[0]);
}
break;
case CL_COMMAND_READ_BUFFER:
memcpy(cmd.as<FreeOCL::command_read_buffer>()->ptr, (char*)cmd.as<FreeOCL::command_read_buffer>()->buffer->ptr + cmd.as<FreeOCL::command_read_buffer>()->offset, cmd.as<FreeOCL::command_read_buffer>()->cb);
break;
case CL_COMMAND_WRITE_BUFFER:
memcpy((char*)cmd.as<FreeOCL::command_write_buffer>()->buffer->ptr + cmd.as<FreeOCL::command_write_buffer>()->offset, cmd.as<FreeOCL::command_write_buffer>()->ptr, cmd.as<FreeOCL::command_write_buffer>()->cb);
break;
case CL_COMMAND_COPY_BUFFER:
memcpy((char*)cmd.as<FreeOCL::command_copy_buffer>()->dst_buffer->ptr + cmd.as<FreeOCL::command_copy_buffer>()->dst_offset,
(char*)cmd.as<FreeOCL::command_copy_buffer>()->src_buffer->ptr + cmd.as<FreeOCL::command_copy_buffer>()->src_offset,
cmd.as<FreeOCL::command_copy_buffer>()->cb);
break;
case CL_COMMAND_MAP_IMAGE:
case CL_COMMAND_MAP_BUFFER:
cmd.as<FreeOCL::command_map_buffer>()->buffer->lock();
cmd.as<FreeOCL::command_map_buffer>()->buffer->mapped.insert(cmd.as<FreeOCL::command_map_buffer>()->ptr);
cmd.as<FreeOCL::command_map_buffer>()->buffer->unlock();
break;
case CL_COMMAND_UNMAP_MEM_OBJECT:
cmd.as<FreeOCL::command_unmap_buffer>()->buffer->lock();
cmd.as<FreeOCL::command_unmap_buffer>()->buffer->mapped.erase(cmd.as<FreeOCL::command_unmap_buffer>()->ptr);
cmd.as<FreeOCL::command_unmap_buffer>()->buffer->unlock();
break;
case CL_COMMAND_NATIVE_KERNEL:
cmd.as<FreeOCL::command_native_kernel>()->user_func(cmd.as<FreeOCL::command_native_kernel>()->args);
free(cmd.as<FreeOCL::command_native_kernel>()->args);
break;
case CL_COMMAND_NDRANGE_KERNEL:
{
FreeOCL::command_ndrange_kernel *ptr = cmd.as<FreeOCL::command_ndrange_kernel>();
const bool b_use_sync = ptr->kernel->__FCL_init(ptr->args,
ptr->dim,
ptr->global_offset,
ptr->global_size,
ptr->local_size);
device->pool->set_local_size(ptr->local_size);
device->pool->set_require_sync(b_use_sync);
device->pool->set_thread_num(device->cpu_cores);
const size_t num_groups[3] = { ptr->global_size[0] / ptr->local_size[0],
ptr->global_size[1] / ptr->local_size[1],
ptr->global_size[2] / ptr->local_size[2] };
device->pool->set_num_groups(num_groups);
device->pool->run(ptr->kernel->__FCL_setwg, ptr->kernel->__FCL_kernel);
if (ptr->args)
free(ptr->args);
}
break;
case CL_COMMAND_FILL_BUFFER:
{
FreeOCL::command_fill_buffer *cfb = cmd.as<FreeOCL::command_fill_buffer>();
const size_t nb_elts = cfb->size / cfb->pattern_size;
switch(cfb->pattern_size)
{
case 1:
memset(cfb->offset + (char*)cfb->buffer->ptr, *(char*)cfb->pattern, cfb->size);
break;
case 2:
{
cl_ushort * const ptr = (cfb->offset >> 1) + (cl_ushort*)cfb->buffer->ptr;
const cl_ushort &pattern = *(const cl_ushort*)cfb->pattern;
for(size_t i = 0 ; i < nb_elts ; ++i)
ptr[i] = pattern;
}
break;
case 4:
{
cl_uint * const ptr = (cfb->offset >> 2) + (cl_uint*)cfb->buffer->ptr;
const cl_uint &pattern = *(const cl_uint*)cfb->pattern;
for(size_t i = 0 ; i < nb_elts ; ++i)
ptr[i] = pattern;
}
break;
case 8:
{
cl_ulong * const ptr = (cfb->offset >> 3) + (cl_ulong*)cfb->buffer->ptr;
const cl_ulong &pattern = *(const cl_ulong*)cfb->pattern;
for(size_t i = 0 ; i < nb_elts ; ++i)
ptr[i] = pattern;
}
break;
case 16:
{
cl_float4 * const ptr = (cfb->offset >> 4) + (cl_float4*)cfb->buffer->ptr;
const cl_float4 &pattern = *(const cl_float4*)cfb->pattern;
for(size_t i = 0 ; i < nb_elts ; ++i)
ptr[i] = pattern;
}
break;
case 32:
{
cl_float8 * const ptr = (cfb->offset >> 5) + (cl_float8*)cfb->buffer->ptr;
const cl_float8 &pattern = *(const cl_float8*)cfb->pattern;
for(size_t i = 0 ; i < nb_elts ; ++i)
ptr[i] = pattern;
}
break;
case 64:
{
cl_float16 * const ptr = (cfb->offset >> 6) + (cl_float16*)cfb->buffer->ptr;
const cl_float16 &pattern = *(const cl_float16*)cfb->pattern;
for(size_t i = 0 ; i < nb_elts ; ++i)
ptr[i] = pattern;
}
break;
case 128:
{
cl_double16 * const ptr = (cfb->offset >> 7) + (cl_double16*)cfb->buffer->ptr;
const cl_double16 &pattern = *(const cl_double16*)cfb->pattern;
for(size_t i = 0 ; i < nb_elts ; ++i)
ptr[i] = pattern;
}
break;
}
free(cfb->pattern);
}
break;
case CL_COMMAND_FILL_IMAGE:
cmd.as<FreeOCL::command_fill_image>()->process();
free(cmd.as<FreeOCL::command_fill_image>()->fill_color);
break;
}
if (cmd->event)
{
cmd->event->lock();
cmd->event->change_status(CL_COMPLETE);
cmd->event->unlock();
}
}
return 0;
}
/// \brief Wait for the semaphore to be signaled
///
/// \param count [in] The value by which the internal lock count will
/// be decremented. At the same time this is the minimum
/// value of the lock count at which the thread is not
/// yielded.
void wait(boost::int64_t count = 1)
{
boost::unique_lock<mutex_type> l(mtx_);
wait_locked(count, l);
}
void io_service_pool::wait()
{
std::lock_guard<compat::mutex> l(mtx_);
wait_locked();
}
bool Event::wait(MSec timeout)
{
Dev::IntLock lock;
return wait_locked(timeout);
}
bool AntiSem::wait(MSec timeout)
{
Dev::IntLock lock;
return wait_locked(timeout);
}
