<html><!-- Created using the cpp_pretty_printer from the dlib C++ library. See http://dlib.net for updates. --><head><title>dlib C++ Library - cuda_data_ptr.h</title></head><body bgcolor='white'><pre> <font color='#009900'>// Copyright (C) 2017 Davis E. King (davis@dlib.net) </font><font color='#009900'>// License: Boost Software License See LICENSE.txt for the full license. </font><font color='#0000FF'>#ifndef</font> DLIB_DNN_CuDA_DATA_PTR_H_ <font color='#0000FF'>#define</font> DLIB_DNN_CuDA_DATA_PTR_H_ <font color='#0000FF'>#include</font> "<a style='text-decoration:none' href='../assert.h.html'>../assert.h</a>" <font color='#0000FF'>#ifdef</font> DLIB_USE_CUDA <font color='#0000FF'>#include</font> <font color='#5555FF'><</font>memory<font color='#5555FF'>></font> <font color='#0000FF'>#include</font> <font color='#5555FF'><</font>vector<font color='#5555FF'>></font> <font color='#0000FF'>#include</font> <font color='#5555FF'><</font>type_traits<font color='#5555FF'>></font> <font color='#0000FF'>namespace</font> dlib <b>{</b> <font color='#0000FF'>namespace</font> cuda <b>{</b> <font color='#009900'>// ------------------------------------------------------------------------------------ </font> <font color='#0000FF'>class</font> cuda_data_void_ptr; <font color='#0000FF'>class</font> <b><a name='weak_cuda_data_void_ptr'></a>weak_cuda_data_void_ptr</b> <b>{</b> <font color='#009900'>/*! WHAT THIS OBJECT REPRESENTS This is just like a std::weak_ptr version of cuda_data_void_ptr. It allows you to hold a non-owning reference to a cuda_data_void_ptr. !*/</font> <font color='#0000FF'>public</font>: <b><a name='weak_cuda_data_void_ptr'></a>weak_cuda_data_void_ptr</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#5555FF'>=</font> <font color='#0000FF'>default</font>; <b><a name='weak_cuda_data_void_ptr'></a>weak_cuda_data_void_ptr</b><font face='Lucida Console'>(</font><font color='#0000FF'>const</font> cuda_data_void_ptr<font color='#5555FF'>&</font> ptr<font face='Lucida Console'>)</font>; <font color='#0000FF'><u>void</u></font> <b><a name='reset'></a>reset</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <b>{</b> pdata.<font color='#BB00BB'>reset</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>; num <font color='#5555FF'>=</font> <font color='#979000'>0</font>; <b>}</b> cuda_data_void_ptr <b><a name='lock'></a>lock</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#0000FF'>const</font>; <font color='#009900'>/*! ensures - if (the memory block referenced by this object hasn't been deleted) then - returns a cuda_data_void_ptr referencing that memory block - else - returns a default initialized cuda_data_void_ptr (i.e. an empty one). !*/</font> <font color='#0000FF'>private</font>: <font color='#0000FF'><u>size_t</u></font> num <font color='#5555FF'>=</font> <font color='#979000'>0</font>; std::weak_ptr<font color='#5555FF'><</font><font color='#0000FF'><u>void</u></font><font color='#5555FF'>></font> pdata; <b>}</b>; <font color='#009900'>// ---------------------------------------------------------------------------------------- </font> <font color='#0000FF'>class</font> <b><a name='cuda_data_void_ptr'></a>cuda_data_void_ptr</b> <b>{</b> <font color='#009900'>/*! WHAT THIS OBJECT REPRESENTS This is a block of memory on a CUDA device. !*/</font> <font color='#0000FF'>public</font>: <b><a name='cuda_data_void_ptr'></a>cuda_data_void_ptr</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#5555FF'>=</font> <font color='#0000FF'>default</font>; <b><a name='cuda_data_void_ptr'></a>cuda_data_void_ptr</b><font face='Lucida Console'>(</font><font color='#0000FF'><u>size_t</u></font> n<font face='Lucida Console'>)</font>; <font color='#009900'>/*! ensures - This object will allocate a device memory buffer of n bytes. - #size() == n !*/</font> <font color='#0000FF'><u>void</u></font><font color='#5555FF'>*</font> <b><a name='data'></a>data</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <b>{</b> <font color='#0000FF'>return</font> pdata.<font color='#BB00BB'>get</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>; <b>}</b> <font color='#0000FF'>const</font> <font color='#0000FF'><u>void</u></font><font color='#5555FF'>*</font> <b><a name='data'></a>data</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#0000FF'>const</font> <b>{</b> <font color='#0000FF'>return</font> pdata.<font color='#BB00BB'>get</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>; <b>}</b> <b><a name='operator'></a>operator</b> <font color='#0000FF'><u>void</u></font><font color='#5555FF'>*</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <b>{</b> <font color='#0000FF'>return</font> pdata.<font color='#BB00BB'>get</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>; <b>}</b> <b><a name='operator'></a>operator</b> <font color='#0000FF'>const</font> <font color='#0000FF'><u>void</u></font><font color='#5555FF'>*</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#0000FF'>const</font> <b>{</b> <font color='#0000FF'>return</font> pdata.<font color='#BB00BB'>get</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>; <b>}</b> <font color='#0000FF'><u>void</u></font> <b><a name='reset'></a>reset</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <b>{</b> pdata.<font color='#BB00BB'>reset</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>; <b>}</b> <font color='#0000FF'><u>size_t</u></font> <b><a name='size'></a>size</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#0000FF'>const</font> <b>{</b> <font color='#0000FF'>return</font> num; <b>}</b> <font color='#009900'>/*! ensures - returns the length of this buffer, in bytes. !*/</font> cuda_data_void_ptr <b><a name='operator'></a>operator</b><font color='#5555FF'>+</font> <font face='Lucida Console'>(</font><font color='#0000FF'><u>size_t</u></font> offset<font face='Lucida Console'>)</font> <font color='#0000FF'>const</font> <font color='#009900'>/*! requires - offset < size() ensures - returns a pointer that is offset by the given amount. !*/</font> <b>{</b> <font color='#BB00BB'>DLIB_CASSERT</font><font face='Lucida Console'>(</font>offset <font color='#5555FF'><</font> num<font face='Lucida Console'>)</font>; cuda_data_void_ptr temp; temp.num <font color='#5555FF'>=</font> num<font color='#5555FF'>-</font>offset; temp.pdata <font color='#5555FF'>=</font> std::shared_ptr<font color='#5555FF'><</font><font color='#0000FF'><u>void</u></font><font color='#5555FF'>></font><font face='Lucida Console'>(</font>pdata, <font face='Lucida Console'>(</font><font face='Lucida Console'>(</font><font color='#0000FF'><u>char</u></font><font color='#5555FF'>*</font><font face='Lucida Console'>)</font>pdata.<font color='#BB00BB'>get</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font face='Lucida Console'>)</font><font color='#5555FF'>+</font>offset<font face='Lucida Console'>)</font>; <font color='#0000FF'>return</font> temp; <b>}</b> <font color='#0000FF'><u>void</u></font> <b><a name='shrink'></a>shrink</b><font face='Lucida Console'>(</font><font color='#0000FF'><u>size_t</u></font> new_size<font face='Lucida Console'>)</font> <font color='#009900'>/*! requires - new_size <= num ensures - #size() == new_size - Doesn't actually deallocate anything, just changes the size() metadata to a smaller number and only for this instance of the pointer. !*/</font> <b>{</b> <font color='#BB00BB'>DLIB_CASSERT</font><font face='Lucida Console'>(</font>new_size <font color='#5555FF'><</font><font color='#5555FF'>=</font> num<font face='Lucida Console'>)</font>; num <font color='#5555FF'>=</font> new_size; <b>}</b> <font color='#0000FF'>private</font>: <font color='#0000FF'>friend</font> <font color='#0000FF'>class</font> weak_cuda_data_void_ptr; <font color='#0000FF'><u>size_t</u></font> num <font color='#5555FF'>=</font> <font color='#979000'>0</font>; std::shared_ptr<font color='#5555FF'><</font><font color='#0000FF'><u>void</u></font><font color='#5555FF'>></font> pdata; <b>}</b>; <font color='#0000FF'>inline</font> cuda_data_void_ptr <b><a name='operator'></a>operator</b><font color='#5555FF'>+</font><font face='Lucida Console'>(</font><font color='#0000FF'><u>size_t</u></font> offset, <font color='#0000FF'>const</font> cuda_data_void_ptr<font color='#5555FF'>&</font> rhs<font face='Lucida Console'>)</font> <b>{</b> <font color='#0000FF'>return</font> rhs<font color='#5555FF'>+</font>offset; <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> <font color='#0000FF'><u>void</u></font><font color='#5555FF'>*</font> dest, <font color='#0000FF'>const</font> cuda_data_void_ptr<font color='#5555FF'>&</font> src <font face='Lucida Console'>)</font>; <font color='#009900'>/*! requires - dest == a pointer to at least src.size() bytes on the host machine. ensures - copies the GPU data from src into dest. - This routine is equivalent to performing: memcpy(dest,src,src.size()) !*/</font> <font color='#0000FF'><u>void</u></font> <b><a name='memcpy'></a>memcpy</b><font face='Lucida Console'>(</font> <font color='#0000FF'><u>void</u></font><font color='#5555FF'>*</font> dest, <font color='#0000FF'>const</font> cuda_data_void_ptr<font color='#5555FF'>&</font> src, <font color='#0000FF'>const</font> <font color='#0000FF'><u>size_t</u></font> num <font face='Lucida Console'>)</font>; <font color='#009900'>/*! requires - dest == a pointer to at least num bytes on the host machine. - num <= src.size() ensures - copies the GPU data from src into dest. Copies only the first num bytes of src to dest. !*/</font> <font color='#009900'>// ------------------------------------------------------------------------------------ </font> <font color='#0000FF'><u>void</u></font> <b><a name='memcpy'></a>memcpy</b><font face='Lucida Console'>(</font> cuda_data_void_ptr dest, <font color='#0000FF'>const</font> <font color='#0000FF'><u>void</u></font><font color='#5555FF'>*</font> src <font face='Lucida Console'>)</font>; <font color='#009900'>/*! requires - dest == a pointer to at least src.size() bytes on the host machine. ensures - copies the host data from src to the GPU memory buffer dest. - This routine is equivalent to performing: memcpy(dest,src,dest.size()) !*/</font> <font color='#0000FF'><u>void</u></font> <b><a name='memcpy'></a>memcpy</b><font face='Lucida Console'>(</font> cuda_data_void_ptr dest, <font color='#0000FF'>const</font> <font color='#0000FF'><u>void</u></font><font color='#5555FF'>*</font> src, <font color='#0000FF'>const</font> <font color='#0000FF'><u>size_t</u></font> num <font face='Lucida Console'>)</font>; <font color='#009900'>/*! requires - dest == a pointer to at least num bytes on the host machine. - num <= dest.size() ensures - copies the host data from src to the GPU memory buffer dest. Copies only the first num bytes of src to dest. !*/</font> <font color='#009900'>// ------------------------------------------------------------------------------------ </font> <font color='#009900'>// ------------------------------------------------------------------------------------ </font> <font color='#009900'>// ------------------------------------------------------------------------------------ </font> <font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> T<font color='#5555FF'>></font> <font color='#0000FF'>class</font> <b><a name='cuda_data_ptr'></a>cuda_data_ptr</b> <b>{</b> <font color='#009900'>/*! WHAT THIS OBJECT REPRESENTS This is a block of memory on a CUDA device. It is just a type safe version of cuda_data_void_ptr. !*/</font> <font color='#0000FF'>public</font>: <b><a name='static_assert'></a>static_assert</b><font face='Lucida Console'>(</font>std::is_standard_layout<font color='#5555FF'><</font>T<font color='#5555FF'>></font>::value, "<font color='#CC0000'>You can only create basic standard layout types on the GPU</font>"<font face='Lucida Console'>)</font>; <b><a name='cuda_data_ptr'></a>cuda_data_ptr</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#5555FF'>=</font> <font color='#0000FF'>default</font>; <b><a name='cuda_data_ptr'></a>cuda_data_ptr</b><font face='Lucida Console'>(</font><font color='#0000FF'><u>size_t</u></font> n<font face='Lucida Console'>)</font> : num<font face='Lucida Console'>(</font>n<font face='Lucida Console'>)</font> <font color='#009900'>/*! ensures - This object will allocate a device memory buffer of n T objects. - #size() == n !*/</font> <b>{</b> <font color='#0000FF'>if</font> <font face='Lucida Console'>(</font>n <font color='#5555FF'>=</font><font color='#5555FF'>=</font> <font color='#979000'>0</font><font face='Lucida Console'>)</font> <font color='#0000FF'>return</font>; pdata <font color='#5555FF'>=</font> <font color='#BB00BB'>cuda_data_void_ptr</font><font face='Lucida Console'>(</font>n<font color='#5555FF'>*</font><font color='#0000FF'>sizeof</font><font face='Lucida Console'>(</font>T<font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; <b>}</b> <b><a name='cuda_data_ptr'></a>cuda_data_ptr</b><font face='Lucida Console'>(</font> <font color='#0000FF'>const</font> cuda_data_ptr<font color='#5555FF'><</font><font color='#0000FF'>typename</font> std::remove_const<font color='#5555FF'><</font>T<font color='#5555FF'>></font>::type<font color='#5555FF'>></font> <font color='#5555FF'>&</font>other <font face='Lucida Console'>)</font> : num<font face='Lucida Console'>(</font>other.num<font face='Lucida Console'>)</font>, pdata<font face='Lucida Console'>(</font>other.pdata<font face='Lucida Console'>)</font> <b>{</b><b>}</b> <font color='#009900'>/*! ensures - *this is a copy of other. This version of the copy constructor allows assigning non-const pointers to const ones. For instance, converting from cuda_data_ptr<float> to cuda_data_ptr<const float>. !*/</font> T<font color='#5555FF'>*</font> <b><a name='data'></a>data</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <b>{</b> <font color='#0000FF'>return</font> <font face='Lucida Console'>(</font>T<font color='#5555FF'>*</font><font face='Lucida Console'>)</font>pdata.<font color='#BB00BB'>data</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>; <b>}</b> <font color='#0000FF'>const</font> T<font color='#5555FF'>*</font> <b><a name='data'></a>data</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#0000FF'>const</font> <b>{</b> <font color='#0000FF'>return</font> <font face='Lucida Console'>(</font>T<font color='#5555FF'>*</font><font face='Lucida Console'>)</font>pdata.<font color='#BB00BB'>data</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>; <b>}</b> <b><a name='operator'></a>operator</b> T<font color='#5555FF'>*</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <b>{</b> <font color='#0000FF'>return</font> <font face='Lucida Console'>(</font>T<font color='#5555FF'>*</font><font face='Lucida Console'>)</font>pdata.<font color='#BB00BB'>data</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>; <b>}</b> <b><a name='operator'></a>operator</b> <font color='#0000FF'>const</font> T<font color='#5555FF'>*</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#0000FF'>const</font> <b>{</b> <font color='#0000FF'>return</font> <font face='Lucida Console'>(</font>T<font color='#5555FF'>*</font><font face='Lucida Console'>)</font>pdata.<font color='#BB00BB'>data</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>; <b>}</b> <font color='#0000FF'><u>void</u></font> <b><a name='reset'></a>reset</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <b>{</b> pdata.<font color='#BB00BB'>reset</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>; <b>}</b> <font color='#0000FF'><u>size_t</u></font> <b><a name='size'></a>size</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#0000FF'>const</font> <b>{</b> <font color='#0000FF'>return</font> num; <b>}</b> <font color='#009900'>/*! ensures - returns the number of T instances pointed to by *this. !*/</font> <b><a name='operator'></a>operator</b> <b><a name='cuda_data_void_ptr'></a>cuda_data_void_ptr</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#0000FF'>const</font> <font color='#009900'>/*! ensures - returns *this as a cuda_data_void_ptr. Importantly, the returned size() will reflect the number of bytes referenced by *this. To be clear, let P be the returned pointer. Then: - P.get() == get() - P.size() == size() * sizeof(T) !*/</font> <b>{</b> cuda_data_void_ptr temp <font color='#5555FF'>=</font> pdata; temp.<font color='#BB00BB'>shrink</font><font face='Lucida Console'>(</font><font color='#BB00BB'>size</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#5555FF'>*</font> <font color='#0000FF'>sizeof</font><font face='Lucida Console'>(</font>T<font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; <font color='#0000FF'>return</font> temp; <b>}</b> <font color='#0000FF'>private</font>: <font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> U<font color='#5555FF'>></font> <font color='#0000FF'>friend</font> cuda_data_ptr<font color='#5555FF'><</font>U<font color='#5555FF'>></font> <b><a name='static_pointer_cast'></a>static_pointer_cast</b><font face='Lucida Console'>(</font><font color='#0000FF'>const</font> cuda_data_void_ptr <font color='#5555FF'>&</font>ptr<font face='Lucida Console'>)</font>; <font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> U<font color='#5555FF'>></font> <font color='#0000FF'>friend</font> cuda_data_ptr<font color='#5555FF'><</font>U<font color='#5555FF'>></font> <b><a name='static_pointer_cast'></a>static_pointer_cast</b><font face='Lucida Console'>(</font><font color='#0000FF'>const</font> cuda_data_void_ptr <font color='#5555FF'>&</font>ptr, <font color='#0000FF'><u>size_t</u></font> num<font face='Lucida Console'>)</font>; <font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> U<font color='#5555FF'>></font> <font color='#0000FF'>friend</font> <font color='#0000FF'>class</font> cuda_data_ptr; <font color='#0000FF'><u>size_t</u></font> num <font color='#5555FF'>=</font> <font color='#979000'>0</font>; cuda_data_void_ptr pdata; <b>}</b>; <font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> T<font color='#5555FF'>></font> cuda_data_ptr<font color='#5555FF'><</font>T<font color='#5555FF'>></font> <b><a name='static_pointer_cast'></a>static_pointer_cast</b><font face='Lucida Console'>(</font><font color='#0000FF'>const</font> cuda_data_void_ptr <font color='#5555FF'>&</font>ptr<font face='Lucida Console'>)</font> <b>{</b> <font color='#BB00BB'>DLIB_CASSERT</font><font face='Lucida Console'>(</font>ptr.<font color='#BB00BB'>size</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#5555FF'>%</font> <font color='#0000FF'>sizeof</font><font face='Lucida Console'>(</font>T<font face='Lucida Console'>)</font> <font color='#5555FF'>=</font><font color='#5555FF'>=</font> <font color='#979000'>0</font>, "<font color='#CC0000'>Size of memory buffer in ptr doesn't match sizeof(T). </font>" <font color='#5555FF'><</font><font color='#5555FF'><</font> "<font color='#CC0000'>\nptr.size(): </font>"<font color='#5555FF'><</font><font color='#5555FF'><</font> ptr.<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='#CC0000'>\nsizeof(T): </font>"<font color='#5555FF'><</font><font color='#5555FF'><</font> <font color='#0000FF'>sizeof</font><font face='Lucida Console'>(</font>T<font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; cuda_data_ptr<font color='#5555FF'><</font>T<font color='#5555FF'>></font> result; result.pdata <font color='#5555FF'>=</font> ptr; result.num <font color='#5555FF'>=</font> ptr.<font color='#BB00BB'>size</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> <font color='#5555FF'>/</font> <font color='#0000FF'>sizeof</font><font face='Lucida Console'>(</font>T<font face='Lucida Console'>)</font>; <font color='#0000FF'>return</font> result; <b>}</b> <font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> T<font color='#5555FF'>></font> cuda_data_ptr<font color='#5555FF'><</font>T<font color='#5555FF'>></font> <b><a name='static_pointer_cast'></a>static_pointer_cast</b><font face='Lucida Console'>(</font><font color='#0000FF'>const</font> cuda_data_void_ptr <font color='#5555FF'>&</font>ptr, <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>num<font color='#5555FF'>*</font><font color='#0000FF'>sizeof</font><font face='Lucida Console'>(</font>T<font face='Lucida Console'>)</font> <font color='#5555FF'><</font><font color='#5555FF'>=</font> ptr.<font color='#BB00BB'>size</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>, "<font color='#CC0000'>Size of memory buffer in ptr isn't big enough to represent this many T objects. </font>" <font color='#5555FF'><</font><font color='#5555FF'><</font> "<font color='#CC0000'>\nnum: </font>"<font color='#5555FF'><</font><font color='#5555FF'><</font> num <font color='#5555FF'><</font><font color='#5555FF'><</font> "<font color='#CC0000'>\nnum*sizeof(T): </font>"<font color='#5555FF'><</font><font color='#5555FF'><</font> num<font color='#5555FF'>*</font><font color='#0000FF'>sizeof</font><font face='Lucida Console'>(</font>T<font face='Lucida Console'>)</font> <font color='#5555FF'><</font><font color='#5555FF'><</font> "<font color='#CC0000'>\nsizeof(T): </font>"<font color='#5555FF'><</font><font color='#5555FF'><</font> <font color='#0000FF'>sizeof</font><font face='Lucida Console'>(</font>T<font face='Lucida Console'>)</font> <font color='#5555FF'><</font><font color='#5555FF'><</font> "<font color='#CC0000'>\nptr.size(): </font>"<font color='#5555FF'><</font><font color='#5555FF'><</font> ptr.<font color='#BB00BB'>size</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; cuda_data_ptr<font color='#5555FF'><</font>T<font color='#5555FF'>></font> result; result.pdata <font color='#5555FF'>=</font> ptr; result.num <font color='#5555FF'>=</font> num; <font color='#0000FF'>return</font> result; <b>}</b> <font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> T<font color='#5555FF'>></font> <font color='#0000FF'><u>void</u></font> <b><a name='memcpy'></a>memcpy</b><font face='Lucida Console'>(</font>std::vector<font color='#5555FF'><</font>T<font color='#5555FF'>></font><font color='#5555FF'>&</font> dest, <font color='#0000FF'>const</font> cuda_data_ptr<font color='#5555FF'><</font>T<font color='#5555FF'>></font><font color='#5555FF'>&</font> src<font face='Lucida Console'>)</font> <b>{</b> dest.<font color='#BB00BB'>resize</font><font face='Lucida Console'>(</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 face='Lucida Console'>)</font> <font color='#BB00BB'>memcpy</font><font face='Lucida Console'>(</font>dest.<font color='#BB00BB'>data</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>, <font color='#0000FF'>static_cast</font><font color='#5555FF'><</font>cuda_data_void_ptr<font color='#5555FF'>></font><font face='Lucida Console'>(</font>src<font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; <b>}</b> <font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> T<font color='#5555FF'>></font> <font color='#0000FF'><u>void</u></font> <b><a name='memcpy'></a>memcpy</b><font face='Lucida Console'>(</font>cuda_data_ptr<font color='#5555FF'><</font>T<font color='#5555FF'>></font><font color='#5555FF'>&</font> dest, <font color='#0000FF'>const</font> std::vector<font color='#5555FF'><</font>T<font color='#5555FF'>></font><font color='#5555FF'>&</font> src<font face='Lucida Console'>)</font> <b>{</b> <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> dest.<font color='#BB00BB'>size</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font face='Lucida Console'>)</font> dest <font color='#5555FF'>=</font> cuda_data_ptr<font color='#5555FF'><</font>T<font color='#5555FF'>></font><font face='Lucida Console'>(</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>dest.<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 face='Lucida Console'>)</font> <font color='#BB00BB'>memcpy</font><font face='Lucida Console'>(</font><font color='#0000FF'>static_cast</font><font color='#5555FF'><</font>cuda_data_void_ptr<font color='#5555FF'>></font><font face='Lucida Console'>(</font>dest<font face='Lucida Console'>)</font>, src.<font color='#BB00BB'>data</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; <b>}</b> <font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> T<font color='#5555FF'>></font> <font color='#0000FF'><u>void</u></font> <b><a name='memcpy'></a>memcpy</b><font face='Lucida Console'>(</font>cuda_data_ptr<font color='#5555FF'><</font>T<font color='#5555FF'>></font><font color='#5555FF'>&</font> dest, <font color='#0000FF'>const</font> T<font color='#5555FF'>*</font> src<font face='Lucida Console'>)</font> <b>{</b> <font color='#BB00BB'>memcpy</font><font face='Lucida Console'>(</font><font color='#0000FF'>static_cast</font><font color='#5555FF'><</font>cuda_data_void_ptr<font color='#5555FF'>></font><font face='Lucida Console'>(</font>dest<font face='Lucida Console'>)</font>, src<font face='Lucida Console'>)</font>; <b>}</b> <font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> T<font color='#5555FF'>></font> <font color='#0000FF'><u>void</u></font> <b><a name='memcpy'></a>memcpy</b><font face='Lucida Console'>(</font>cuda_data_ptr<font color='#5555FF'><</font>T<font color='#5555FF'>></font><font color='#5555FF'>&</font> dest, <font color='#0000FF'>const</font> T<font color='#5555FF'>*</font> src, <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>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'>memcpy</font><font face='Lucida Console'>(</font><font color='#0000FF'>static_cast</font><font color='#5555FF'><</font>cuda_data_void_ptr<font color='#5555FF'>></font><font face='Lucida Console'>(</font>dest<font face='Lucida Console'>)</font>, src, num<font color='#5555FF'>*</font><font color='#0000FF'>sizeof</font><font face='Lucida Console'>(</font>T<font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; <b>}</b> <font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> T<font color='#5555FF'>></font> <font color='#0000FF'><u>void</u></font> <b><a name='memcpy'></a>memcpy</b><font face='Lucida Console'>(</font>T<font color='#5555FF'>*</font> dest, <font color='#0000FF'>const</font> cuda_data_ptr<font color='#5555FF'><</font>T<font color='#5555FF'>></font><font color='#5555FF'>&</font> src<font face='Lucida Console'>)</font> <b>{</b> <font color='#BB00BB'>memcpy</font><font face='Lucida Console'>(</font>dest, <font color='#0000FF'>static_cast</font><font color='#5555FF'><</font>cuda_data_void_ptr<font color='#5555FF'>></font><font face='Lucida Console'>(</font>src<font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; <b>}</b> <font color='#0000FF'>template</font> <font color='#5555FF'><</font><font color='#0000FF'>typename</font> T<font color='#5555FF'>></font> <font color='#0000FF'><u>void</u></font> <b><a name='memcpy'></a>memcpy</b><font face='Lucida Console'>(</font>T<font color='#5555FF'>*</font> dest, <font color='#0000FF'>const</font> cuda_data_ptr<font color='#5555FF'><</font>T<font color='#5555FF'>></font><font color='#5555FF'>&</font> src, <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>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='#BB00BB'>memcpy</font><font face='Lucida Console'>(</font>dest, <font color='#0000FF'>static_cast</font><font color='#5555FF'><</font>cuda_data_void_ptr<font color='#5555FF'>></font><font face='Lucida Console'>(</font>src<font face='Lucida Console'>)</font>, num<font color='#5555FF'>*</font><font color='#0000FF'>sizeof</font><font face='Lucida Console'>(</font>T<font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; <b>}</b> <font color='#009900'>// ------------------------------------------------------------------------------------ </font> cuda_data_void_ptr <b><a name='device_global_buffer'></a>device_global_buffer</b><font face='Lucida Console'>(</font><font color='#0000FF'><u>size_t</u></font> size<font face='Lucida Console'>)</font>; <font color='#009900'>/*! ensures - Returns a pointer to a globally shared CUDA memory buffer on the currently selected CUDA device. The buffer is also thread local. So each host thread will get its own buffer. You can use this global buffer as scratch space for CUDA computations that all take place on the default stream. Using it in this way ensures that there aren't any race conditions involving the use of the buffer. - The returned pointer will point to at least size bytes. It may point to more. - The global buffer is deallocated once all references to it are destructed. However, if device_global_buffer() is called before then with a size <= the last size requested, then the previously returned global buffer pointer is returned. This avoids triggering expensive CUDA reallocations. So if you want to avoid these reallocations then hold a copy of the pointer returned by this function. However, as a general rule, client code should not hold the returned cuda_data_void_ptr for long durations, but instead should call device_global_buffer() whenever the buffer is needed, and overwrite the previously returned pointer with the new pointer. Doing so ensures multiple buffers are not kept around in the event that multiple sized buffers are requested. To explain this, consider this code, assumed to execute at program startup: auto ptr1 = device_global_buffer(1); auto ptr2 = device_global_buffer(2); auto ptr3 = device_global_buffer(3); since the sizes increased at each call 3 separate buffers were allocated. First one of size 1, then of size 2, then of size 3. If we then executed: ptr1 = device_global_buffer(1); ptr2 = device_global_buffer(2); ptr3 = device_global_buffer(3); all three of these pointers would now point to the same buffer, since the smaller requests can be satisfied by returning the size 3 buffer in each case. !*/</font> <font color='#009900'>// ---------------------------------------------------------------------------------------- </font> <b>}</b> <b>}</b> <font color='#0000FF'>#endif</font> <font color='#009900'>// DLIB_USE_CUDA </font> <font color='#0000FF'>#endif</font> <font color='#009900'>// DLIB_DNN_CuDA_DATA_PTR_H_ </font> </pre></body></html>