|
<html><head><title>dlib C++ Library - cublas_dlibapi.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_DNN_CuBLAS_CPP_ |
|
<font color='#0000FF'>#define</font> DLIB_DNN_CuBLAS_CPP_ |
|
|
|
<font color='#0000FF'>#ifdef</font> DLIB_USE_CUDA |
|
|
|
<font color='#0000FF'>#include</font> "<a style='text-decoration:none' href='cublas_dlibapi.h.html'>cublas_dlibapi.h</a>" |
|
<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>cublas_v2.h<font color='#5555FF'>></font> |
|
<font color='#0000FF'>#include</font> <font color='#5555FF'><</font>vector<font color='#5555FF'>></font> |
|
|
|
<font color='#0000FF'>static</font> <font color='#0000FF'>const</font> <font color='#0000FF'><u>char</u></font><font color='#5555FF'>*</font> <b><a name='cublas_get_error_string'></a>cublas_get_error_string</b><font face='Lucida Console'>(</font>cublasStatus_t s<font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
<font color='#0000FF'>switch</font><font face='Lucida Console'>(</font>s<font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
<font color='#0000FF'>case</font> CUBLAS_STATUS_NOT_INITIALIZED: |
|
<font color='#0000FF'>return</font> "<font color='#CC0000'>CUDA Runtime API initialization failed.</font>"; |
|
<font color='#0000FF'>case</font> CUBLAS_STATUS_ALLOC_FAILED: |
|
<font color='#0000FF'>return</font> "<font color='#CC0000'>CUDA Resources could not be allocated.</font>"; |
|
<font color='#0000FF'>default</font>: |
|
<font color='#0000FF'>return</font> "<font color='#CC0000'>A call to cuBLAS failed</font>"; |
|
<b>}</b> |
|
<b>}</b> |
|
|
|
<font color='#009900'>// Check the return value of a call to the cuBLAS runtime for an error condition. |
|
</font><font color='#0000FF'>#define</font> CHECK_CUBLAS<font face='Lucida Console'>(</font>call<font face='Lucida Console'>)</font> \ |
|
<font color='#0000FF'>do</font><b>{</b> \ |
|
<font color='#0000FF'>const</font> cublasStatus_t error <font color='#5555FF'>=</font> call; \ |
|
<font color='#0000FF'>if</font> <font face='Lucida Console'>(</font>error <font color='#5555FF'>!</font><font color='#5555FF'>=</font> CUBLAS_STATUS_SUCCESS<font face='Lucida Console'>)</font> \ |
|
<b>{</b> \ |
|
std::ostringstream sout; \ |
|
sout <font color='#5555FF'><</font><font color='#5555FF'><</font> "<font color='#CC0000'>Error while calling </font>" <font color='#5555FF'><</font><font color='#5555FF'><</font> #call <font color='#5555FF'><</font><font color='#5555FF'><</font> "<font color='#CC0000'> in file </font>" <font color='#5555FF'><</font><font color='#5555FF'><</font> __FILE__ <font color='#5555FF'><</font><font color='#5555FF'><</font> "<font color='#CC0000'>:</font>" <font color='#5555FF'><</font><font color='#5555FF'><</font> __LINE__ <font color='#5555FF'><</font><font color='#5555FF'><</font> "<font color='#CC0000'>. </font>";\ |
|
sout <font color='#5555FF'><</font><font color='#5555FF'><</font> "<font color='#CC0000'>code: </font>" <font color='#5555FF'><</font><font color='#5555FF'><</font> error <font color='#5555FF'><</font><font color='#5555FF'><</font> "<font color='#CC0000'>, reason: </font>" <font color='#5555FF'><</font><font color='#5555FF'><</font> <font color='#BB00BB'>cublas_get_error_string</font><font face='Lucida Console'>(</font>error<font face='Lucida Console'>)</font>;\ |
|
<font color='#0000FF'>throw</font> dlib::<font color='#BB00BB'>cublas_error</font><font face='Lucida Console'>(</font>sout.<font color='#BB00BB'>str</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; \ |
|
<b>}</b> \ |
|
<b>}</b><font color='#0000FF'>while</font><font face='Lucida Console'>(</font><font color='#979000'>false</font><font face='Lucida Console'>)</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> <b><a name='cublas_context'></a>cublas_context</b> |
|
<b>{</b> |
|
<font color='#0000FF'>public</font>: |
|
<font color='#009900'>// not copyable |
|
</font> <b><a name='cublas_context'></a>cublas_context</b><font face='Lucida Console'>(</font><font color='#0000FF'>const</font> cublas_context<font color='#5555FF'>&</font><font face='Lucida Console'>)</font> <font color='#5555FF'>=</font> <font color='#0000FF'>delete</font>; |
|
cublas_context<font color='#5555FF'>&</font> <b><a name='operator'></a>operator</b><font color='#5555FF'>=</font><font face='Lucida Console'>(</font><font color='#0000FF'>const</font> cublas_context<font color='#5555FF'>&</font><font face='Lucida Console'>)</font> <font color='#5555FF'>=</font> <font color='#0000FF'>delete</font>; |
|
|
|
<b><a name='cublas_context'></a>cublas_context</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
handles.<font color='#BB00BB'>resize</font><font face='Lucida Console'>(</font><font color='#979000'>16</font><font face='Lucida Console'>)</font>; |
|
<b>}</b> |
|
~<b><a name='cublas_context'></a>cublas_context</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
<font color='#0000FF'>for</font> <font face='Lucida Console'>(</font><font color='#0000FF'>auto</font> h : handles<font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
<font color='#0000FF'>if</font> <font face='Lucida Console'>(</font>h<font face='Lucida Console'>)</font> |
|
<font color='#BB00BB'>cublasDestroy</font><font face='Lucida Console'>(</font>h<font face='Lucida Console'>)</font>; |
|
<b>}</b> |
|
<b>}</b> |
|
|
|
cublasHandle_t <b><a name='get_handle'></a>get_handle</b> <font face='Lucida Console'>(</font> |
|
<font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
<font color='#0000FF'><u>int</u></font> new_device_id; |
|
<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>new_device_id<font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; |
|
<font color='#009900'>// make room for more devices if needed |
|
</font> <font color='#0000FF'>if</font> <font face='Lucida Console'>(</font>new_device_id <font color='#5555FF'>></font><font color='#5555FF'>=</font> <font face='Lucida Console'>(</font><font color='#0000FF'><u>long</u></font><font face='Lucida Console'>)</font>handles.<font color='#BB00BB'>size</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font face='Lucida Console'>)</font> |
|
handles.<font color='#BB00BB'>resize</font><font face='Lucida Console'>(</font>new_device_id<font color='#5555FF'>+</font><font color='#979000'>16</font><font face='Lucida Console'>)</font>; |
|
|
|
<font color='#009900'>// If we don't have a handle already for this device then make one |
|
</font> <font color='#0000FF'>if</font> <font face='Lucida Console'>(</font><font color='#5555FF'>!</font>handles[new_device_id]<font face='Lucida Console'>)</font> |
|
<font color='#BB00BB'>CHECK_CUBLAS</font><font face='Lucida Console'>(</font><font color='#BB00BB'>cublasCreate</font><font face='Lucida Console'>(</font><font color='#5555FF'>&</font>handles[new_device_id]<font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; |
|
|
|
<font color='#009900'>// Finally, return the handle for the current device |
|
</font> <font color='#0000FF'>return</font> handles[new_device_id]; |
|
<b>}</b> |
|
|
|
<font color='#0000FF'>private</font>: |
|
|
|
std::vector<font color='#5555FF'><</font>cublasHandle_t<font color='#5555FF'>></font> handles; |
|
<b>}</b>; |
|
|
|
<font color='#0000FF'>static</font> cublasHandle_t <b><a name='context'></a>context</b><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
thread_local cublas_context c; |
|
<font color='#0000FF'>return</font> c.<font color='#BB00BB'>get_handle</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>; |
|
<b>}</b> |
|
|
|
<font color='#009900'>// ----------------------------------------------------------------------------------- |
|
</font> |
|
<font color='#0000FF'><u>void</u></font> <b><a name='gemm'></a>gemm</b> <font face='Lucida Console'>(</font> |
|
<font color='#0000FF'><u>float</u></font> beta, |
|
tensor<font color='#5555FF'>&</font> dest, |
|
<font color='#0000FF'><u>float</u></font> alpha, |
|
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> lhs, |
|
<font color='#0000FF'><u>bool</u></font> trans_lhs, |
|
<font color='#0000FF'>const</font> tensor<font color='#5555FF'>&</font> rhs, |
|
<font color='#0000FF'><u>bool</u></font> trans_rhs |
|
<font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
<font color='#009900'>// Recall that BLAS uses column major order so to deal with that we flip the |
|
</font> <font color='#009900'>// order of the lhs and rhs arguments. |
|
</font> <font color='#0000FF'>const</font> <font color='#0000FF'>auto</font> transa <font color='#5555FF'>=</font> trans_lhs ? CUBLAS_OP_T : CUBLAS_OP_N; |
|
<font color='#0000FF'>const</font> <font color='#0000FF'>auto</font> transb <font color='#5555FF'>=</font> trans_rhs ? CUBLAS_OP_T : CUBLAS_OP_N; |
|
|
|
<font color='#0000FF'>const</font> <font color='#0000FF'><u>int</u></font> dest_nr <font color='#5555FF'>=</font> dest.<font color='#BB00BB'>num_samples</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>; |
|
<font color='#0000FF'>const</font> <font color='#0000FF'><u>int</u></font> dest_nc <font color='#5555FF'>=</font> dest.<font color='#BB00BB'>size</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font color='#5555FF'>/</font>dest_nr; |
|
<font color='#0000FF'>const</font> <font color='#0000FF'><u>int</u></font> lhs_nr <font color='#5555FF'>=</font> lhs.<font color='#BB00BB'>num_samples</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>; |
|
<font color='#0000FF'>const</font> <font color='#0000FF'><u>int</u></font> lhs_nc <font color='#5555FF'>=</font> lhs.<font color='#BB00BB'>size</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font color='#5555FF'>/</font>lhs_nr; |
|
<font color='#0000FF'>const</font> <font color='#0000FF'><u>int</u></font> rhs_nr <font color='#5555FF'>=</font> rhs.<font color='#BB00BB'>num_samples</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>; |
|
<font color='#0000FF'>const</font> <font color='#0000FF'><u>int</u></font> rhs_nc <font color='#5555FF'>=</font> rhs.<font color='#BB00BB'>size</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font><font color='#5555FF'>/</font>rhs_nr; |
|
<font color='#0000FF'>if</font> <font face='Lucida Console'>(</font>trans_lhs <font color='#5555FF'>&</font><font color='#5555FF'>&</font> trans_rhs<font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
<font color='#BB00BB'>DLIB_ASSERT</font><font face='Lucida Console'>(</font> dest_nr <font color='#5555FF'>=</font><font color='#5555FF'>=</font> lhs_nc <font color='#5555FF'>&</font><font color='#5555FF'>&</font> |
|
dest_nc <font color='#5555FF'>=</font><font color='#5555FF'>=</font> rhs_nr <font color='#5555FF'>&</font><font color='#5555FF'>&</font> |
|
lhs_nr <font color='#5555FF'>=</font><font color='#5555FF'>=</font> rhs_nc<font face='Lucida Console'>)</font> |
|
<b>}</b> |
|
<font color='#0000FF'>else</font> <font color='#0000FF'>if</font> <font face='Lucida Console'>(</font><font color='#5555FF'>!</font>trans_lhs <font color='#5555FF'>&</font><font color='#5555FF'>&</font> trans_rhs<font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
<font color='#BB00BB'>DLIB_ASSERT</font><font face='Lucida Console'>(</font> dest_nr <font color='#5555FF'>=</font><font color='#5555FF'>=</font> lhs_nr <font color='#5555FF'>&</font><font color='#5555FF'>&</font> |
|
dest_nc <font color='#5555FF'>=</font><font color='#5555FF'>=</font> rhs_nr <font color='#5555FF'>&</font><font color='#5555FF'>&</font> |
|
lhs_nc <font color='#5555FF'>=</font><font color='#5555FF'>=</font> rhs_nc<font face='Lucida Console'>)</font> |
|
<b>}</b> |
|
<font color='#0000FF'>else</font> <font color='#0000FF'>if</font> <font face='Lucida Console'>(</font>trans_lhs <font color='#5555FF'>&</font><font color='#5555FF'>&</font> <font color='#5555FF'>!</font>trans_rhs<font face='Lucida Console'>)</font> |
|
<b>{</b> |
|
<font color='#BB00BB'>DLIB_ASSERT</font><font face='Lucida Console'>(</font> dest_nr <font color='#5555FF'>=</font><font color='#5555FF'>=</font> lhs_nc <font color='#5555FF'>&</font><font color='#5555FF'>&</font> |
|
dest_nc <font color='#5555FF'>=</font><font color='#5555FF'>=</font> rhs_nc <font color='#5555FF'>&</font><font color='#5555FF'>&</font> |
|
lhs_nr <font color='#5555FF'>=</font><font color='#5555FF'>=</font> rhs_nr<font face='Lucida Console'>)</font> |
|
<b>}</b> |
|
<font color='#0000FF'>else</font> |
|
<b>{</b> |
|
<font color='#BB00BB'>DLIB_ASSERT</font><font face='Lucida Console'>(</font> dest_nr <font color='#5555FF'>=</font><font color='#5555FF'>=</font> lhs_nr <font color='#5555FF'>&</font><font color='#5555FF'>&</font> |
|
dest_nc <font color='#5555FF'>=</font><font color='#5555FF'>=</font> rhs_nc <font color='#5555FF'>&</font><font color='#5555FF'>&</font> |
|
lhs_nc <font color='#5555FF'>=</font><font color='#5555FF'>=</font> rhs_nr<font face='Lucida Console'>)</font> |
|
<b>}</b> |
|
|
|
<font color='#0000FF'>const</font> <font color='#0000FF'><u>int</u></font> k <font color='#5555FF'>=</font> trans_rhs ? rhs_nc : rhs_nr; |
|
<font color='#BB00BB'>CHECK_CUBLAS</font><font face='Lucida Console'>(</font><font color='#BB00BB'>cublasSgemm</font><font face='Lucida Console'>(</font><font color='#BB00BB'>context</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>, |
|
transb, |
|
transa, |
|
dest_nc, dest_nr, k, |
|
<font color='#5555FF'>&</font>alpha, |
|
rhs.<font color='#BB00BB'>device</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>, rhs_nc, |
|
lhs.<font color='#BB00BB'>device</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>, lhs_nc, |
|
<font color='#5555FF'>&</font>beta, |
|
dest.<font color='#BB00BB'>device</font><font face='Lucida Console'>(</font><font face='Lucida Console'>)</font>,dest_nc<font face='Lucida Console'>)</font><font face='Lucida Console'>)</font>; |
|
<b>}</b> |
|
|
|
<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_CuBLAS_CPP_ |
|
</font> |
|
|
|
|
|
|
|
</pre></body></html> |