Partial emulation of ATLAS/BLAS gemm(), non-cached version. Matrix 'C' is assumed to have been set to the correct size (i.e. taking into account transposes). More...
#include <gemm.hpp>
Static Public Member Functions | |
template<typename eT > | |
static arma_hot void | apply (Mat< eT > &C, const Mat< eT > &A, const Mat< eT > &B, const eT alpha=eT(1), const eT beta=eT(0)) |
Partial emulation of ATLAS/BLAS gemm(), non-cached version. Matrix 'C' is assumed to have been set to the correct size (i.e. taking into account transposes).
Definition at line 211 of file gemm.hpp.
static arma_hot void gemm_emul_simple< do_trans_A, do_trans_B, use_alpha, use_beta >::apply | ( | Mat< eT > & | C, | |
const Mat< eT > & | A, | |||
const Mat< eT > & | B, | |||
const eT | alpha = eT(1) , |
|||
const eT | beta = eT(0) | |||
) | [inline, static] |
Definition at line 221 of file gemm.hpp.
References Mat< eT >::at(), Mat< eT >::colptr(), Mat< eT >::n_cols, and Mat< eT >::n_rows.
00228 { 00229 arma_extra_debug_sigprint(); 00230 00231 const u32 A_n_rows = A.n_rows; 00232 const u32 A_n_cols = A.n_cols; 00233 00234 const u32 B_n_rows = B.n_rows; 00235 const u32 B_n_cols = B.n_cols; 00236 00237 if( (do_trans_A == false) && (do_trans_B == false) ) 00238 { 00239 for(u32 row_A = 0; row_A < A_n_rows; ++row_A) 00240 { 00241 for(u32 col_B = 0; col_B < B_n_cols; ++col_B) 00242 { 00243 const eT* B_coldata = B.colptr(col_B); 00244 00245 eT acc = eT(0); 00246 for(u32 i = 0; i < B_n_rows; ++i) 00247 { 00248 acc += A.at(row_A,i) * B_coldata[i]; 00249 } 00250 00251 if( (use_alpha == false) && (use_beta == false) ) 00252 { 00253 C.at(row_A,col_B) = acc; 00254 } 00255 else 00256 if( (use_alpha == true) && (use_beta == false) ) 00257 { 00258 C.at(row_A,col_B) = alpha * acc; 00259 } 00260 else 00261 if( (use_alpha == false) && (use_beta == true) ) 00262 { 00263 C.at(row_A,col_B) = acc + beta*C.at(row_A,col_B); 00264 } 00265 else 00266 if( (use_alpha == true) && (use_beta == true) ) 00267 { 00268 C.at(row_A,col_B) = alpha*acc + beta*C.at(row_A,col_B); 00269 } 00270 } 00271 } 00272 } 00273 else 00274 if( (do_trans_A == true) && (do_trans_B == false) ) 00275 { 00276 for(u32 col_A=0; col_A < A_n_cols; ++col_A) 00277 { 00278 // col_A is interpreted as row_A when storing the results in matrix C 00279 00280 const eT* A_coldata = A.colptr(col_A); 00281 00282 for(u32 col_B=0; col_B < B_n_cols; ++col_B) 00283 { 00284 const eT* B_coldata = B.colptr(col_B); 00285 00286 eT acc = eT(0); 00287 for(u32 i=0; i < B_n_rows; ++i) 00288 { 00289 acc += A_coldata[i] * B_coldata[i]; 00290 } 00291 00292 if( (use_alpha == false) && (use_beta == false) ) 00293 { 00294 C.at(col_A,col_B) = acc; 00295 } 00296 else 00297 if( (use_alpha == true) && (use_beta == false) ) 00298 { 00299 C.at(col_A,col_B) = alpha * acc; 00300 } 00301 else 00302 if( (use_alpha == false) && (use_beta == true) ) 00303 { 00304 C.at(col_A,col_B) = acc + beta*C.at(col_A,col_B); 00305 } 00306 else 00307 if( (use_alpha == true) && (use_beta == true) ) 00308 { 00309 C.at(col_A,col_B) = alpha*acc + beta*C.at(col_A,col_B); 00310 } 00311 00312 } 00313 } 00314 } 00315 else 00316 if( (do_trans_A == false) && (do_trans_B == true) ) 00317 { 00318 for(u32 row_A = 0; row_A < A_n_rows; ++row_A) 00319 { 00320 for(u32 row_B = 0; row_B < B_n_rows; ++row_B) 00321 { 00322 eT acc = eT(0); 00323 for(u32 i = 0; i < B_n_cols; ++i) 00324 { 00325 acc += A.at(row_A,i) * B.at(row_B,i); 00326 } 00327 00328 if( (use_alpha == false) && (use_beta == false) ) 00329 { 00330 C.at(row_A,row_B) = acc; 00331 } 00332 else 00333 if( (use_alpha == true) && (use_beta == false) ) 00334 { 00335 C.at(row_A,row_B) = alpha * acc; 00336 } 00337 else 00338 if( (use_alpha == false) && (use_beta == true) ) 00339 { 00340 C.at(row_A,row_B) = acc + beta*C.at(row_A,row_B); 00341 } 00342 else 00343 if( (use_alpha == true) && (use_beta == true) ) 00344 { 00345 C.at(row_A,row_B) = alpha*acc + beta*C.at(row_A,row_B); 00346 } 00347 } 00348 } 00349 } 00350 else 00351 if( (do_trans_A == true) && (do_trans_B == true) ) 00352 { 00353 for(u32 row_B=0; row_B < B_n_rows; ++row_B) 00354 { 00355 00356 for(u32 col_A=0; col_A < A_n_cols; ++col_A) 00357 { 00358 const eT* A_coldata = A.colptr(col_A); 00359 00360 eT acc = eT(0); 00361 for(u32 i=0; i < A_n_rows; ++i) 00362 { 00363 acc += B.at(row_B,i) * A_coldata[i]; 00364 } 00365 00366 if( (use_alpha == false) && (use_beta == false) ) 00367 { 00368 C.at(col_A,row_B) = acc; 00369 } 00370 else 00371 if( (use_alpha == true) && (use_beta == false) ) 00372 { 00373 C.at(col_A,row_B) = alpha * acc; 00374 } 00375 else 00376 if( (use_alpha == false) && (use_beta == true) ) 00377 { 00378 C.at(col_A,row_B) = acc + beta*C.at(col_A,row_B); 00379 } 00380 else 00381 if( (use_alpha == true) && (use_beta == true) ) 00382 { 00383 C.at(col_A,row_B) = alpha*acc + beta*C.at(col_A,row_B); 00384 } 00385 00386 } 00387 } 00388 00389 } 00390 }