|
<html><head><title>dlib C++ Library - gpu_data.cpp</title></head><body bgcolor='white'><pre> |
|
<font color='#009900'>// Copyright (C) 2015 Davis E. King ([email protected]) |
|
</font><font color='#009900'>// License: Boost Software License See LICENSE.txt for the full license. |
|
</font><font color='#0000FF'>#ifndef</font> DLIB_GPU_DaTA_CPP_ |
|
<font color='#0000FF'>#define</font> DLIB_GPU_DaTA_CPP_ |
|
|
|
<font color='#009900'>// Only things that require CUDA are declared in this cpp file. Everything else is in the |
|
</font><font color='#009900'>// gpu_data.h header so that it can operate as "header-only" code when using just the CPU. |
|
</font><font color='#0000FF'>#ifdef</font> DLIB_USE_CUDA |
|
|
|
<font color='#0000FF'>#include</font> "<a style='text-decoration:none' href='gpu_data.h.html'>gpu_data.h</a>" |
|
<font color='#0000FF'>#include</font> <font color='#5555FF'><</font>iostream<font color='#5555FF'>></font> |
|
<font color='#0000FF'>#include</font> "<a style='text-decoration:none' href='cuda_utils.h.html'>cuda_utils.h</a>" |
|
<font color='#0000FF'>#include</font> <font color='#5555FF'><</font>cstring<font color='#5555FF'>></font> |
|
<font color='#0000FF'>#include</font> <font color='#5555FF'><</font>cuda.h<font color='#5555FF'>></font> |
|
|
|
<font color='#0000FF'>namespace</font> dlib |
|
<b>{</b> |
|
|
|
<font color='#009900'>// ---------------------------------------------------------------------------------------- |
|
</font> |
|
<font color='#0000FF'><u>void</u></font> <b><a name='memcpy'></a>memcpy</b> <font face='Lucida Console'>(</font> |
|
gpu_data<font color='#5555FF'>&</font> dest, |
|
<font color='#0000FF'>const</font> gpu_data<font color='#5555FF'>&</font> src |
|
<font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
<font color='#BB00BB'>DLIB_CASSERT</font><font face='Lucida Console'>(</font>dest.<font color='#BB00BB'>size</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#5555FF'>=</font><font color='#5555FF'>=</font> src.<font color='#BB00BB'>size</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; |
|
<font color='#0000FF'>if</font> <font face='Lucida Console'>(</font>src.<font color='#BB00BB'>size</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#5555FF'>=</font><font color='#5555FF'>=</font> <font color='#979000'>0</font> <font color='#5555FF'>|</font><font color='#5555FF'>|</font> <font color='#5555FF'>&</font>dest <font color='#5555FF'>=</font><font color='#5555FF'>=</font> <font color='#5555FF'>&</font>src<font face='Lucida Console'>)</font> |
|
<font color='#0000FF'>return</font>; |
|
|
|
<font color='#BB00BB'>memcpy</font><font face='Lucida Console'>(</font>dest,<font color='#979000'>0</font>, src, <font color='#979000'>0</font>, src.<font color='#BB00BB'>size</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; |
|
<b>}</b> |
|
|
|
<font color='#0000FF'><u>void</u></font> <b><a name='memcpy'></a>memcpy</b> <font face='Lucida Console'>(</font> |
|
gpu_data<font color='#5555FF'>&</font> dest, |
|
<font color='#0000FF'><u>size_t</u></font> dest_offset, |
|
<font color='#0000FF'>const</font> gpu_data<font color='#5555FF'>&</font> src, |
|
<font color='#0000FF'><u>size_t</u></font> src_offset, |
|
<font color='#0000FF'><u>size_t</u></font> num |
|
<font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
<font color='#BB00BB'>DLIB_CASSERT</font><font face='Lucida Console'>(</font>dest_offset <font color='#5555FF'>+</font> num <font color='#5555FF'><</font><font color='#5555FF'>=</font> dest.<font color='#BB00BB'>size</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; |
|
<font color='#BB00BB'>DLIB_CASSERT</font><font face='Lucida Console'>(</font>src_offset <font color='#5555FF'>+</font> num <font color='#5555FF'><</font><font color='#5555FF'>=</font> src.<font color='#BB00BB'>size</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; |
|
<font color='#0000FF'>if</font> <font face='Lucida Console'>(</font>num <font color='#5555FF'>=</font><font color='#5555FF'>=</font> <font color='#979000'>0</font><font face='Lucida Console'>)</font> |
|
<font color='#0000FF'>return</font>; |
|
|
|
<font color='#009900'>// if there is aliasing |
|
</font> <font color='#0000FF'>if</font> <font face='Lucida Console'>(</font><font color='#5555FF'>&</font>dest <font color='#5555FF'>=</font><font color='#5555FF'>=</font> <font color='#5555FF'>&</font>src <font color='#5555FF'>&</font><font color='#5555FF'>&</font> std::<font color='#BB00BB'>max</font><font face='Lucida Console'>(</font>dest_offset, src_offset<font face='Lucida Console'>)</font> <font color='#5555FF'><</font> std::<font color='#BB00BB'>min</font><font face='Lucida Console'>(</font>dest_offset,src_offset<font face='Lucida Console'>)</font><font color='#5555FF'>+</font>num<font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
<font color='#009900'>// if they perfectly alias each other then there is nothing to do |
|
</font> <font color='#0000FF'>if</font> <font face='Lucida Console'>(</font>dest_offset <font color='#5555FF'>=</font><font color='#5555FF'>=</font> src_offset<font face='Lucida Console'>)</font> |
|
<font color='#0000FF'>return</font>; |
|
<font color='#0000FF'>else</font> |
|
std::<font color='#BB00BB'>memmove</font><font face='Lucida Console'>(</font>dest.<font color='#BB00BB'>host</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font color='#5555FF'>+</font>dest_offset, src.<font color='#BB00BB'>host</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font color='#5555FF'>+</font>src_offset, <font color='#0000FF'>sizeof</font><font face='Lucida Console'>(</font><font color='#0000FF'><u>float</u></font><font face='Lucida Console'>)</font><font color='#5555FF'>*</font>num<font face='Lucida Console'>)</font>; |
|
<b>}</b> |
|
<font color='#0000FF'>else</font> |
|
<b>{</b> |
|
<font color='#009900'>// if we write to the entire thing then we can use device_write_only() |
|
</font> <font color='#0000FF'>if</font> <font face='Lucida Console'>(</font>dest_offset <font color='#5555FF'>=</font><font color='#5555FF'>=</font> <font color='#979000'>0</font> <font color='#5555FF'>&</font><font color='#5555FF'>&</font> num <font color='#5555FF'>=</font><font color='#5555FF'>=</font> dest.<font color='#BB00BB'>size</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
<font color='#009900'>// copy the memory efficiently based on which copy is current in each object. |
|
</font> <font color='#0000FF'>if</font> <font face='Lucida Console'>(</font>src.<font color='#BB00BB'>device_ready</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font face='Lucida Console'>)</font> |
|
<font color='#BB00BB'>CHECK_CUDA</font><font face='Lucida Console'>(</font><font color='#BB00BB'>cudaMemcpy</font><font face='Lucida Console'>(</font>dest.<font color='#BB00BB'>device_write_only</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>, src.<font color='#BB00BB'>device</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font color='#5555FF'>+</font>src_offset, num<font color='#5555FF'>*</font><font color='#0000FF'>sizeof</font><font face='Lucida Console'>(</font><font color='#0000FF'><u>float</u></font><font face='Lucida Console'>)</font>, cudaMemcpyDeviceToDevice<font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; |
|
<font color='#0000FF'>else</font> |
|
<font color='#BB00BB'>CHECK_CUDA</font><font face='Lucida Console'>(</font><font color='#BB00BB'>cudaMemcpy</font><font face='Lucida Console'>(</font>dest.<font color='#BB00BB'>device_write_only</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>, src.<font color='#BB00BB'>host</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font color='#5555FF'>+</font>src_offset, num<font color='#5555FF'>*</font><font color='#0000FF'>sizeof</font><font face='Lucida Console'>(</font><font color='#0000FF'><u>float</u></font><font face='Lucida Console'>)</font>, cudaMemcpyHostToDevice<font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; |
|
<b>}</b> |
|
<font color='#0000FF'>else</font> |
|
<b>{</b> |
|
<font color='#009900'>// copy the memory efficiently based on which copy is current in each object. |
|
</font> <font color='#0000FF'>if</font> <font face='Lucida Console'>(</font>dest.<font color='#BB00BB'>device_ready</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#5555FF'>&</font><font color='#5555FF'>&</font> src.<font color='#BB00BB'>device_ready</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font face='Lucida Console'>)</font> |
|
<font color='#BB00BB'>CHECK_CUDA</font><font face='Lucida Console'>(</font><font color='#BB00BB'>cudaMemcpy</font><font face='Lucida Console'>(</font>dest.<font color='#BB00BB'>device</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font color='#5555FF'>+</font>dest_offset, src.<font color='#BB00BB'>device</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font color='#5555FF'>+</font>src_offset, num<font color='#5555FF'>*</font><font color='#0000FF'>sizeof</font><font face='Lucida Console'>(</font><font color='#0000FF'><u>float</u></font><font face='Lucida Console'>)</font>, cudaMemcpyDeviceToDevice<font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; |
|
<font color='#0000FF'>else</font> <font color='#0000FF'>if</font> <font face='Lucida Console'>(</font><font color='#5555FF'>!</font>dest.<font color='#BB00BB'>device_ready</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#5555FF'>&</font><font color='#5555FF'>&</font> src.<font color='#BB00BB'>device_ready</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font face='Lucida Console'>)</font> |
|
<font color='#BB00BB'>CHECK_CUDA</font><font face='Lucida Console'>(</font><font color='#BB00BB'>cudaMemcpy</font><font face='Lucida Console'>(</font>dest.<font color='#BB00BB'>host</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font color='#5555FF'>+</font>dest_offset, src.<font color='#BB00BB'>device</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font color='#5555FF'>+</font>src_offset, num<font color='#5555FF'>*</font><font color='#0000FF'>sizeof</font><font face='Lucida Console'>(</font><font color='#0000FF'><u>float</u></font><font face='Lucida Console'>)</font>, cudaMemcpyDeviceToHost<font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; |
|
<font color='#0000FF'>else</font> <font color='#0000FF'>if</font> <font face='Lucida Console'>(</font>dest.<font color='#BB00BB'>device_ready</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#5555FF'>&</font><font color='#5555FF'>&</font> <font color='#5555FF'>!</font>src.<font color='#BB00BB'>device_ready</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font face='Lucida Console'>)</font> |
|
<font color='#BB00BB'>CHECK_CUDA</font><font face='Lucida Console'>(</font><font color='#BB00BB'>cudaMemcpy</font><font face='Lucida Console'>(</font>dest.<font color='#BB00BB'>device</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font color='#5555FF'>+</font>dest_offset, src.<font color='#BB00BB'>host</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font color='#5555FF'>+</font>src_offset, num<font color='#5555FF'>*</font><font color='#0000FF'>sizeof</font><font face='Lucida Console'>(</font><font color='#0000FF'><u>float</u></font><font face='Lucida Console'>)</font>, cudaMemcpyHostToDevice<font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; |
|
<font color='#0000FF'>else</font> |
|
<font color='#BB00BB'>CHECK_CUDA</font><font face='Lucida Console'>(</font><font color='#BB00BB'>cudaMemcpy</font><font face='Lucida Console'>(</font>dest.<font color='#BB00BB'>host</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font color='#5555FF'>+</font>dest_offset, src.<font color='#BB00BB'>host</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font color='#5555FF'>+</font>src_offset, num<font color='#5555FF'>*</font><font color='#0000FF'>sizeof</font><font face='Lucida Console'>(</font><font color='#0000FF'><u>float</u></font><font face='Lucida Console'>)</font>, cudaMemcpyHostToHost<font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; |
|
<b>}</b> |
|
<b>}</b> |
|
<b>}</b> |
|
<font color='#009900'>// ---------------------------------------------------------------------------------------- |
|
</font> |
|
<font color='#0000FF'><u>void</u></font> <b><a name='synchronize_stream'></a>synchronize_stream</b><font face='Lucida Console'>(</font>cudaStream_t stream<font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
<font color='#0000FF'>#if</font> <font color='#5555FF'>!</font>defined CUDA_VERSION |
|
<font color='#0000FF'>#error</font> CUDA_VERSION not defined |
|
<font color='#0000FF'>#elif</font> CUDA_VERSION <font color='#5555FF'>></font><font color='#5555FF'>=</font> <font color='#979000'>9020</font> <font color='#5555FF'>&</font><font color='#5555FF'>&</font> CUDA_VERSION <font color='#5555FF'><</font> <font color='#979000'>11000</font> |
|
<font color='#009900'>// We will stop using this alternative version with cuda V11, hopefully the bug in |
|
</font> <font color='#009900'>// cudaStreamSynchronize is fixed by then. |
|
</font> <font color='#009900'>// |
|
</font> <font color='#009900'>// This should be pretty much the same as cudaStreamSynchronize, which for some |
|
</font> <font color='#009900'>// reason makes training freeze in some cases. |
|
</font> <font color='#009900'>// (see https://github.com/davisking/dlib/issues/1513) |
|
</font> <font color='#0000FF'>while</font> <font face='Lucida Console'>(</font><font color='#979000'>true</font><font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
cudaError_t err <font color='#5555FF'>=</font> <font color='#BB00BB'>cudaStreamQuery</font><font face='Lucida Console'>(</font>stream<font face='Lucida Console'>)</font>; |
|
<font color='#0000FF'>switch</font> <font face='Lucida Console'>(</font>err<font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
<font color='#0000FF'>case</font> cudaSuccess: <font color='#0000FF'>return</font>; <font color='#009900'>// now we are synchronized |
|
</font> <font color='#0000FF'>case</font> cudaErrorNotReady: <font color='#0000FF'>break</font>; <font color='#009900'>// continue waiting |
|
</font> <font color='#0000FF'>default</font>: <font color='#BB00BB'>CHECK_CUDA</font><font face='Lucida Console'>(</font>err<font face='Lucida Console'>)</font>; <font color='#009900'>// unexpected error: throw |
|
</font> <b>}</b> |
|
<b>}</b> |
|
<font color='#0000FF'>#else</font> <font color='#009900'>// CUDA_VERSION |
|
</font> <font color='#BB00BB'>CHECK_CUDA</font><font face='Lucida Console'>(</font><font color='#BB00BB'>cudaStreamSynchronize</font><font face='Lucida Console'>(</font>stream<font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; |
|
<font color='#0000FF'>#endif</font> <font color='#009900'>// CUDA_VERSION |
|
</font> <b>}</b> |
|
|
|
<font color='#0000FF'><u>void</u></font> gpu_data:: |
|
<b><a name='wait_for_transfer_to_finish'></a>wait_for_transfer_to_finish</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#0000FF'>const</font> |
|
<b>{</b> |
|
<font color='#0000FF'>if</font> <font face='Lucida Console'>(</font>have_active_transfer<font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
<font color='#BB00BB'>synchronize_stream</font><font face='Lucida Console'>(</font><font face='Lucida Console'>(</font>cudaStream_t<font face='Lucida Console'>)</font>cuda_stream.<font color='#BB00BB'>get</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; |
|
have_active_transfer <font color='#5555FF'>=</font> <font color='#979000'>false</font>; |
|
<font color='#009900'>// Check for errors. These calls to cudaGetLastError() are what help us find |
|
</font> <font color='#009900'>// out if our kernel launches have been failing. |
|
</font> <font color='#BB00BB'>CHECK_CUDA</font><font face='Lucida Console'>(</font><font color='#BB00BB'>cudaGetLastError</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; |
|
<b>}</b> |
|
<b>}</b> |
|
|
|
<font color='#0000FF'><u>void</u></font> gpu_data:: |
|
<b><a name='copy_to_device'></a>copy_to_device</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#0000FF'>const</font> |
|
<b>{</b> |
|
<font color='#009900'>// We want transfers to the device to always be concurrent with any device |
|
</font> <font color='#009900'>// computation. So we use our non-default stream to do the transfer. |
|
</font> <font color='#BB00BB'>async_copy_to_device</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>; |
|
<font color='#BB00BB'>wait_for_transfer_to_finish</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>; |
|
<b>}</b> |
|
|
|
<font color='#0000FF'><u>void</u></font> gpu_data:: |
|
<b><a name='copy_to_host'></a>copy_to_host</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#0000FF'>const</font> |
|
<b>{</b> |
|
<font color='#0000FF'>if</font> <font face='Lucida Console'>(</font><font color='#5555FF'>!</font>host_current<font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
<font color='#BB00BB'>wait_for_transfer_to_finish</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>; |
|
<font color='#BB00BB'>CHECK_CUDA</font><font face='Lucida Console'>(</font><font color='#BB00BB'>cudaMemcpy</font><font face='Lucida Console'>(</font>data_host.<font color='#BB00BB'>get</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>, data_device.<font color='#BB00BB'>get</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>, data_size<font color='#5555FF'>*</font><font color='#0000FF'>sizeof</font><font face='Lucida Console'>(</font><font color='#0000FF'><u>float</u></font><font face='Lucida Console'>)</font>, cudaMemcpyDeviceToHost<font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; |
|
host_current <font color='#5555FF'>=</font> <font color='#979000'>true</font>; |
|
<font color='#009900'>// At this point we know our RAM block isn't in use because cudaMemcpy() |
|
</font> <font color='#009900'>// implicitly syncs with the device. |
|
</font> device_in_use <font color='#5555FF'>=</font> <font color='#979000'>false</font>; |
|
<font color='#009900'>// Check for errors. These calls to cudaGetLastError() are what help us find |
|
</font> <font color='#009900'>// out if our kernel launches have been failing. |
|
</font> <font color='#BB00BB'>CHECK_CUDA</font><font face='Lucida Console'>(</font><font color='#BB00BB'>cudaGetLastError</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; |
|
<b>}</b> |
|
<b>}</b> |
|
|
|
<font color='#0000FF'><u>void</u></font> gpu_data:: |
|
<b><a name='async_copy_to_device'></a>async_copy_to_device</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#0000FF'>const</font> |
|
<b>{</b> |
|
<font color='#0000FF'>if</font> <font face='Lucida Console'>(</font><font color='#5555FF'>!</font>device_current<font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
<font color='#0000FF'>if</font> <font face='Lucida Console'>(</font>device_in_use<font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
<font color='#009900'>// Wait for any possible CUDA kernels that might be using our memory block to |
|
</font> <font color='#009900'>// complete before we overwrite the memory. |
|
</font> <font color='#BB00BB'>synchronize_stream</font><font face='Lucida Console'>(</font><font color='#979000'>0</font><font face='Lucida Console'>)</font>; |
|
device_in_use <font color='#5555FF'>=</font> <font color='#979000'>false</font>; |
|
<b>}</b> |
|
<font color='#BB00BB'>CHECK_CUDA</font><font face='Lucida Console'>(</font><font color='#BB00BB'>cudaMemcpyAsync</font><font face='Lucida Console'>(</font>data_device.<font color='#BB00BB'>get</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>, data_host.<font color='#BB00BB'>get</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>, data_size<font color='#5555FF'>*</font><font color='#0000FF'>sizeof</font><font face='Lucida Console'>(</font><font color='#0000FF'><u>float</u></font><font face='Lucida Console'>)</font>, cudaMemcpyHostToDevice, <font face='Lucida Console'>(</font>cudaStream_t<font face='Lucida Console'>)</font>cuda_stream.<font color='#BB00BB'>get</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; |
|
have_active_transfer <font color='#5555FF'>=</font> <font color='#979000'>true</font>; |
|
device_current <font color='#5555FF'>=</font> <font color='#979000'>true</font>; |
|
<b>}</b> |
|
<b>}</b> |
|
|
|
<font color='#0000FF'><u>void</u></font> gpu_data:: |
|
<b><a name='set_size'></a>set_size</b><font face='Lucida Console'>(</font> |
|
<font color='#0000FF'><u>size_t</u></font> new_size |
|
<font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
<font color='#0000FF'>if</font> <font face='Lucida Console'>(</font>new_size <font color='#5555FF'>=</font><font color='#5555FF'>=</font> <font color='#979000'>0</font><font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
<font color='#0000FF'>if</font> <font face='Lucida Console'>(</font>device_in_use<font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
<font color='#009900'>// Wait for any possible CUDA kernels that might be using our memory block to |
|
</font> <font color='#009900'>// complete before we free the memory. |
|
</font> <font color='#BB00BB'>synchronize_stream</font><font face='Lucida Console'>(</font><font color='#979000'>0</font><font face='Lucida Console'>)</font>; |
|
device_in_use <font color='#5555FF'>=</font> <font color='#979000'>false</font>; |
|
<b>}</b> |
|
<font color='#BB00BB'>wait_for_transfer_to_finish</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>; |
|
data_size <font color='#5555FF'>=</font> <font color='#979000'>0</font>; |
|
host_current <font color='#5555FF'>=</font> <font color='#979000'>true</font>; |
|
device_current <font color='#5555FF'>=</font> <font color='#979000'>true</font>; |
|
device_in_use <font color='#5555FF'>=</font> <font color='#979000'>false</font>; |
|
data_host.<font color='#BB00BB'>reset</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>; |
|
data_device.<font color='#BB00BB'>reset</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>; |
|
<b>}</b> |
|
<font color='#0000FF'>else</font> <font color='#0000FF'>if</font> <font face='Lucida Console'>(</font>new_size <font color='#5555FF'>!</font><font color='#5555FF'>=</font> data_size<font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
<font color='#0000FF'>if</font> <font face='Lucida Console'>(</font>device_in_use<font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
<font color='#009900'>// Wait for any possible CUDA kernels that might be using our memory block to |
|
</font> <font color='#009900'>// complete before we free the memory. |
|
</font> <font color='#BB00BB'>synchronize_stream</font><font face='Lucida Console'>(</font><font color='#979000'>0</font><font face='Lucida Console'>)</font>; |
|
device_in_use <font color='#5555FF'>=</font> <font color='#979000'>false</font>; |
|
<b>}</b> |
|
<font color='#BB00BB'>wait_for_transfer_to_finish</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>; |
|
data_size <font color='#5555FF'>=</font> new_size; |
|
host_current <font color='#5555FF'>=</font> <font color='#979000'>true</font>; |
|
device_current <font color='#5555FF'>=</font> <font color='#979000'>true</font>; |
|
device_in_use <font color='#5555FF'>=</font> <font color='#979000'>false</font>; |
|
|
|
<font color='#0000FF'>try</font> |
|
<b>{</b> |
|
<font color='#BB00BB'>CHECK_CUDA</font><font face='Lucida Console'>(</font><font color='#BB00BB'>cudaGetDevice</font><font face='Lucida Console'>(</font><font color='#5555FF'>&</font>the_device_id<font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; |
|
|
|
<font color='#009900'>// free memory blocks before we allocate new ones. |
|
</font> data_host.<font color='#BB00BB'>reset</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>; |
|
data_device.<font color='#BB00BB'>reset</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>; |
|
|
|
<font color='#0000FF'><u>void</u></font><font color='#5555FF'>*</font> data; |
|
<font color='#BB00BB'>CHECK_CUDA</font><font face='Lucida Console'>(</font><font color='#BB00BB'>cudaMallocHost</font><font face='Lucida Console'>(</font><font color='#5555FF'>&</font>data, new_size<font color='#5555FF'>*</font><font color='#0000FF'>sizeof</font><font face='Lucida Console'>(</font><font color='#0000FF'><u>float</u></font><font face='Lucida Console'>)</font><font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; |
|
<font color='#009900'>// Note that we don't throw exceptions since the free calls are invariably |
|
</font> <font color='#009900'>// called in destructors. They also shouldn't fail anyway unless someone |
|
</font> <font color='#009900'>// is resetting the GPU card in the middle of their program. |
|
</font> data_host.<font color='#BB00BB'>reset</font><font face='Lucida Console'>(</font><font face='Lucida Console'>(</font><font color='#0000FF'><u>float</u></font><font color='#5555FF'>*</font><font face='Lucida Console'>)</font>data, []<font face='Lucida Console'>(</font><font color='#0000FF'><u>float</u></font><font color='#5555FF'>*</font> ptr<font face='Lucida Console'>)</font><b>{</b> |
|
<font color='#0000FF'>auto</font> err <font color='#5555FF'>=</font> <font color='#BB00BB'>cudaFreeHost</font><font face='Lucida Console'>(</font>ptr<font face='Lucida Console'>)</font>; |
|
<font color='#0000FF'>if</font><font face='Lucida Console'>(</font>err<font color='#5555FF'>!</font><font color='#5555FF'>=</font>cudaSuccess<font face='Lucida Console'>)</font> |
|
std::cerr <font color='#5555FF'><</font><font color='#5555FF'><</font> "<font color='#CC0000'>cudaFreeHost() failed. Reason: </font>" <font color='#5555FF'><</font><font color='#5555FF'><</font> <font color='#BB00BB'>cudaGetErrorString</font><font face='Lucida Console'>(</font>err<font face='Lucida Console'>)</font> <font color='#5555FF'><</font><font color='#5555FF'><</font> std::endl; |
|
<b>}</b><font face='Lucida Console'>)</font>; |
|
|
|
<font color='#BB00BB'>CHECK_CUDA</font><font face='Lucida Console'>(</font><font color='#BB00BB'>cudaMalloc</font><font face='Lucida Console'>(</font><font color='#5555FF'>&</font>data, new_size<font color='#5555FF'>*</font><font color='#0000FF'>sizeof</font><font face='Lucida Console'>(</font><font color='#0000FF'><u>float</u></font><font face='Lucida Console'>)</font><font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; |
|
data_device.<font color='#BB00BB'>reset</font><font face='Lucida Console'>(</font><font face='Lucida Console'>(</font><font color='#0000FF'><u>float</u></font><font color='#5555FF'>*</font><font face='Lucida Console'>)</font>data, []<font face='Lucida Console'>(</font><font color='#0000FF'><u>float</u></font><font color='#5555FF'>*</font> ptr<font face='Lucida Console'>)</font><b>{</b> |
|
<font color='#0000FF'>auto</font> err <font color='#5555FF'>=</font> <font color='#BB00BB'>cudaFree</font><font face='Lucida Console'>(</font>ptr<font face='Lucida Console'>)</font>; |
|
<font color='#0000FF'>if</font><font face='Lucida Console'>(</font>err<font color='#5555FF'>!</font><font color='#5555FF'>=</font>cudaSuccess<font face='Lucida Console'>)</font> |
|
std::cerr <font color='#5555FF'><</font><font color='#5555FF'><</font> "<font color='#CC0000'>cudaFree() failed. Reason: </font>" <font color='#5555FF'><</font><font color='#5555FF'><</font> <font color='#BB00BB'>cudaGetErrorString</font><font face='Lucida Console'>(</font>err<font face='Lucida Console'>)</font> <font color='#5555FF'><</font><font color='#5555FF'><</font> std::endl; |
|
<b>}</b><font face='Lucida Console'>)</font>; |
|
|
|
<font color='#0000FF'>if</font> <font face='Lucida Console'>(</font><font color='#5555FF'>!</font>cuda_stream<font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
cudaStream_t cstream; |
|
<font color='#BB00BB'>CHECK_CUDA</font><font face='Lucida Console'>(</font><font color='#BB00BB'>cudaStreamCreateWithFlags</font><font face='Lucida Console'>(</font><font color='#5555FF'>&</font>cstream, cudaStreamNonBlocking<font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; |
|
cuda_stream.<font color='#BB00BB'>reset</font><font face='Lucida Console'>(</font>cstream, []<font face='Lucida Console'>(</font><font color='#0000FF'><u>void</u></font><font color='#5555FF'>*</font> ptr<font face='Lucida Console'>)</font><b>{</b> |
|
<font color='#0000FF'>auto</font> err <font color='#5555FF'>=</font> <font color='#BB00BB'>cudaStreamDestroy</font><font face='Lucida Console'>(</font><font face='Lucida Console'>(</font>cudaStream_t<font face='Lucida Console'>)</font>ptr<font face='Lucida Console'>)</font>; |
|
<font color='#0000FF'>if</font><font face='Lucida Console'>(</font>err<font color='#5555FF'>!</font><font color='#5555FF'>=</font>cudaSuccess<font face='Lucida Console'>)</font> |
|
std::cerr <font color='#5555FF'><</font><font color='#5555FF'><</font> "<font color='#CC0000'>cudaStreamDestroy() failed. Reason: </font>" <font color='#5555FF'><</font><font color='#5555FF'><</font> <font color='#BB00BB'>cudaGetErrorString</font><font face='Lucida Console'>(</font>err<font face='Lucida Console'>)</font> <font color='#5555FF'><</font><font color='#5555FF'><</font> std::endl; |
|
<b>}</b><font face='Lucida Console'>)</font>; |
|
<b>}</b> |
|
|
|
<b>}</b> |
|
<font color='#0000FF'>catch</font><font face='Lucida Console'>(</font>...<font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
<font color='#BB00BB'>set_size</font><font face='Lucida Console'>(</font><font color='#979000'>0</font><font face='Lucida Console'>)</font>; |
|
<font color='#0000FF'>throw</font>; |
|
<b>}</b> |
|
<b>}</b> |
|
<b>}</b> |
|
|
|
<font color='#009900'>// ---------------------------------------------------------------------------------------- |
|
</font><b>}</b> |
|
|
|
<font color='#0000FF'>#endif</font> <font color='#009900'>// DLIB_USE_CUDA |
|
</font> |
|
<font color='#0000FF'>#endif</font> <font color='#009900'>// DLIB_GPU_DaTA_CPP_ |
|
</font> |
|
|
|
</pre></body></html> |
