1 // This file is part of Eigen, a lightweight C++ template library 2 // for linear algebra. 3 // 4 // Copyright (C) 2008-2009 Gael Guennebaud <gael.guennebaud@inria.fr> 5 // 6 // This Source Code Form is subject to the terms of the Mozilla 7 // Public License v. 2.0. If a copy of the MPL was not distributed 8 // with this file, You can obtain one at http://mozilla.org/MPL/2.0/. 9 10 #ifndef EIGEN_GENERAL_MATRIX_MATRIX_H 11 #define EIGEN_GENERAL_MATRIX_MATRIX_H 12 13 namespace Eigen { 14 15 namespace internal { 16 17 template<typename _LhsScalar, typename _RhsScalar> class level3_blocking; 18 19 /* Specialization for a row-major destination matrix => simple transposition of the product */ 20 template< 21 typename Index, 22 typename LhsScalar, int LhsStorageOrder, bool ConjugateLhs, 23 typename RhsScalar, int RhsStorageOrder, bool ConjugateRhs, 24 int ResInnerStride> 25 struct general_matrix_matrix_product<Index,LhsScalar,LhsStorageOrder,ConjugateLhs,RhsScalar,RhsStorageOrder,ConjugateRhs,RowMajor,ResInnerStride> 26 { 27 typedef gebp_traits<RhsScalar,LhsScalar> Traits; 28 29 typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScalar; 30 static EIGEN_STRONG_INLINE void run( 31 Index rows, Index cols, Index depth, 32 const LhsScalar* lhs, Index lhsStride, 33 const RhsScalar* rhs, Index rhsStride, 34 ResScalar* res, Index resIncr, Index resStride, 35 ResScalar alpha, 36 level3_blocking<RhsScalar,LhsScalar>& blocking, 37 GemmParallelInfo<Index>* info = 0) 38 { 39 // transpose the product such that the result is column major 40 general_matrix_matrix_product<Index, 41 RhsScalar, RhsStorageOrder==RowMajor ? ColMajor : RowMajor, ConjugateRhs, 42 LhsScalar, LhsStorageOrder==RowMajor ? ColMajor : RowMajor, ConjugateLhs, 43 ColMajor,ResInnerStride> 44 ::run(cols,rows,depth,rhs,rhsStride,lhs,lhsStride,res,resIncr,resStride,alpha,blocking,info); 45 } 46 }; 47 48 /* Specialization for a col-major destination matrix 49 * => Blocking algorithm following Goto's paper */ 50 template< 51 typename Index, 52 typename LhsScalar, int LhsStorageOrder, bool ConjugateLhs, 53 typename RhsScalar, int RhsStorageOrder, bool ConjugateRhs, 54 int ResInnerStride> 55 struct general_matrix_matrix_product<Index,LhsScalar,LhsStorageOrder,ConjugateLhs,RhsScalar,RhsStorageOrder,ConjugateRhs,ColMajor,ResInnerStride> 56 { 57 58 typedef gebp_traits<LhsScalar,RhsScalar> Traits; 59 60 typedef typename ScalarBinaryOpTraits<LhsScalar, RhsScalar>::ReturnType ResScalar; 61 static void run(Index rows, Index cols, Index depth, 62 const LhsScalar* _lhs, Index lhsStride, 63 const RhsScalar* _rhs, Index rhsStride, 64 ResScalar* _res, Index resIncr, Index resStride, 65 ResScalar alpha, 66 level3_blocking<LhsScalar,RhsScalar>& blocking, 67 GemmParallelInfo<Index>* info = 0) 68 { 69 typedef const_blas_data_mapper<LhsScalar, Index, LhsStorageOrder> LhsMapper; 70 typedef const_blas_data_mapper<RhsScalar, Index, RhsStorageOrder> RhsMapper; 71 typedef blas_data_mapper<typename Traits::ResScalar, Index, ColMajor,Unaligned,ResInnerStride> ResMapper; 72 LhsMapper lhs(_lhs, lhsStride); 73 RhsMapper rhs(_rhs, rhsStride); 74 ResMapper res(_res, resStride, resIncr); 75 76 Index kc = blocking.kc(); // cache block size along the K direction 77 Index mc = (std::min)(rows,blocking.mc()); // cache block size along the M direction 78 Index nc = (std::min)(cols,blocking.nc()); // cache block size along the N direction 79 80 gemm_pack_lhs<LhsScalar, Index, LhsMapper, Traits::mr, Traits::LhsProgress, typename Traits::LhsPacket4Packing, LhsStorageOrder> pack_lhs; 81 gemm_pack_rhs<RhsScalar, Index, RhsMapper, Traits::nr, RhsStorageOrder> pack_rhs; 82 gebp_kernel<LhsScalar, RhsScalar, Index, ResMapper, Traits::mr, Traits::nr, ConjugateLhs, ConjugateRhs> gebp; 83 84 #ifdef EIGEN_HAS_OPENMP 85 if(info) 86 { 87 // this is the parallel version! 88 int tid = omp_get_thread_num(); 89 int threads = omp_get_num_threads(); 90 91 LhsScalar* blockA = blocking.blockA(); 92 eigen_internal_assert(blockA!=0); 93 94 std::size_t sizeB = kc*nc; 95 ei_declare_aligned_stack_constructed_variable(RhsScalar, blockB, sizeB, 0); 96 97 // For each horizontal panel of the rhs, and corresponding vertical panel of the lhs... 98 for(Index k=0; k<depth; k+=kc) 99 { 100 const Index actual_kc = (std::min)(k+kc,depth)-k; // => rows of B', and cols of the A' 101 102 // In order to reduce the chance that a thread has to wait for the other, 103 // let's start by packing B'. 104 pack_rhs(blockB, rhs.getSubMapper(k,0), actual_kc, nc); 105 106 // Pack A_k to A' in a parallel fashion: 107 // each thread packs the sub block A_k,i to A'_i where i is the thread id. 108 109 // However, before copying to A'_i, we have to make sure that no other thread is still using it, 110 // i.e., we test that info[tid].users equals 0. 111 // Then, we set info[tid].users to the number of threads to mark that all other threads are going to use it. 112 while(info[tid].users!=0) {} 113 info[tid].users = threads; 114 115 pack_lhs(blockA+info[tid].lhs_start*actual_kc, lhs.getSubMapper(info[tid].lhs_start,k), actual_kc, info[tid].lhs_length); 116 117 // Notify the other threads that the part A'_i is ready to go. 118 info[tid].sync = k; 119 120 // Computes C_i += A' * B' per A'_i 121 for(int shift=0; shift<threads; ++shift) 122 { 123 int i = (tid+shift)%threads; 124 125 // At this point we have to make sure that A'_i has been updated by the thread i, 126 // we use testAndSetOrdered to mimic a volatile access. 127 // However, no need to wait for the B' part which has been updated by the current thread! 128 if (shift>0) { 129 while(info[i].sync!=k) { 130 } 131 } 132 133 gebp(res.getSubMapper(info[i].lhs_start, 0), blockA+info[i].lhs_start*actual_kc, blockB, info[i].lhs_length, actual_kc, nc, alpha); 134 } 135 136 // Then keep going as usual with the remaining B' 137 for(Index j=nc; j<cols; j+=nc) 138 { 139 const Index actual_nc = (std::min)(j+nc,cols)-j; 140 141 // pack B_k,j to B' 142 pack_rhs(blockB, rhs.getSubMapper(k,j), actual_kc, actual_nc); 143 144 // C_j += A' * B' 145 gebp(res.getSubMapper(0, j), blockA, blockB, rows, actual_kc, actual_nc, alpha); 146 } 147 148 // Release all the sub blocks A'_i of A' for the current thread, 149 // i.e., we simply decrement the number of users by 1 150 for(Index i=0; i<threads; ++i) 151 #if !EIGEN_HAS_CXX11_ATOMIC 152 #pragma omp atomic 153 #endif 154 info[i].users -= 1; 155 } 156 } 157 else 158 #endif // EIGEN_HAS_OPENMP 159 { 160 EIGEN_UNUSED_VARIABLE(info); 161 162 // this is the sequential version! 163 std::size_t sizeA = kc*mc; 164 std::size_t sizeB = kc*nc; 165 166 ei_declare_aligned_stack_constructed_variable(LhsScalar, blockA, sizeA, blocking.blockA()); 167 ei_declare_aligned_stack_constructed_variable(RhsScalar, blockB, sizeB, blocking.blockB()); 168 169 const bool pack_rhs_once = mc!=rows && kc==depth && nc==cols; 170 171 // For each horizontal panel of the rhs, and corresponding panel of the lhs... 172 for(Index i2=0; i2<rows; i2+=mc) 173 { 174 const Index actual_mc = (std::min)(i2+mc,rows)-i2; 175 176 for(Index k2=0; k2<depth; k2+=kc) 177 { 178 const Index actual_kc = (std::min)(k2+kc,depth)-k2; 179 180 // OK, here we have selected one horizontal panel of rhs and one vertical panel of lhs. 181 // => Pack lhs's panel into a sequential chunk of memory (L2/L3 caching) 182 // Note that this panel will be read as many times as the number of blocks in the rhs's 183 // horizontal panel which is, in practice, a very low number. 184 pack_lhs(blockA, lhs.getSubMapper(i2,k2), actual_kc, actual_mc); 185 186 // For each kc x nc block of the rhs's horizontal panel... 187 for(Index j2=0; j2<cols; j2+=nc) 188 { 189 const Index actual_nc = (std::min)(j2+nc,cols)-j2; 190 191 // We pack the rhs's block into a sequential chunk of memory (L2 caching) 192 // Note that this block will be read a very high number of times, which is equal to the number of 193 // micro horizontal panel of the large rhs's panel (e.g., rows/12 times). 194 if((!pack_rhs_once) || i2==0) 195 pack_rhs(blockB, rhs.getSubMapper(k2,j2), actual_kc, actual_nc); 196 197 // Everything is packed, we can now call the panel * block kernel: 198 gebp(res.getSubMapper(i2, j2), blockA, blockB, actual_mc, actual_kc, actual_nc, alpha); 199 } 200 } 201 } 202 } 203 } 204 205 }; 206 207 /********************************************************************************* 208 * Specialization of generic_product_impl for "large" GEMM, i.e., 209 * implementation of the high level wrapper to general_matrix_matrix_product 210 **********************************************************************************/ 211 212 template<typename Scalar, typename Index, typename Gemm, typename Lhs, typename Rhs, typename Dest, typename BlockingType> 213 struct gemm_functor 214 { 215 gemm_functor(const Lhs& lhs, const Rhs& rhs, Dest& dest, const Scalar& actualAlpha, BlockingType& blocking) 216 : m_lhs(lhs), m_rhs(rhs), m_dest(dest), m_actualAlpha(actualAlpha), m_blocking(blocking) 217 {} 218 219 void initParallelSession(Index num_threads) const 220 { 221 m_blocking.initParallel(m_lhs.rows(), m_rhs.cols(), m_lhs.cols(), num_threads); 222 m_blocking.allocateA(); 223 } 224 225 void operator() (Index row, Index rows, Index col=0, Index cols=-1, GemmParallelInfo<Index>* info=0) const 226 { 227 if(cols==-1) 228 cols = m_rhs.cols(); 229 230 Gemm::run(rows, cols, m_lhs.cols(), 231 &m_lhs.coeffRef(row,0), m_lhs.outerStride(), 232 &m_rhs.coeffRef(0,col), m_rhs.outerStride(), 233 (Scalar*)&(m_dest.coeffRef(row,col)), m_dest.innerStride(), m_dest.outerStride(), 234 m_actualAlpha, m_blocking, info); 235 } 236 237 typedef typename Gemm::Traits Traits; 238 239 protected: 240 const Lhs& m_lhs; 241 const Rhs& m_rhs; 242 Dest& m_dest; 243 Scalar m_actualAlpha; 244 BlockingType& m_blocking; 245 }; 246 247 template<int StorageOrder, typename LhsScalar, typename RhsScalar, int MaxRows, int MaxCols, int MaxDepth, int KcFactor=1, 248 bool FiniteAtCompileTime = MaxRows!=Dynamic && MaxCols!=Dynamic && MaxDepth != Dynamic> class gemm_blocking_space; 249 250 template<typename _LhsScalar, typename _RhsScalar> 251 class level3_blocking 252 { 253 typedef _LhsScalar LhsScalar; 254 typedef _RhsScalar RhsScalar; 255 256 protected: 257 LhsScalar* m_blockA; 258 RhsScalar* m_blockB; 259 260 Index m_mc; 261 Index m_nc; 262 Index m_kc; 263 264 public: 265 266 level3_blocking() 267 : m_blockA(0), m_blockB(0), m_mc(0), m_nc(0), m_kc(0) 268 {} 269 270 inline Index mc() const { return m_mc; } 271 inline Index nc() const { return m_nc; } 272 inline Index kc() const { return m_kc; } 273 274 inline LhsScalar* blockA() { return m_blockA; } 275 inline RhsScalar* blockB() { return m_blockB; } 276 }; 277 278 template<int StorageOrder, typename _LhsScalar, typename _RhsScalar, int MaxRows, int MaxCols, int MaxDepth, int KcFactor> 279 class gemm_blocking_space<StorageOrder,_LhsScalar,_RhsScalar,MaxRows, MaxCols, MaxDepth, KcFactor, true /* == FiniteAtCompileTime */> 280 : public level3_blocking< 281 typename conditional<StorageOrder==RowMajor,_RhsScalar,_LhsScalar>::type, 282 typename conditional<StorageOrder==RowMajor,_LhsScalar,_RhsScalar>::type> 283 { 284 enum { 285 Transpose = StorageOrder==RowMajor, 286 ActualRows = Transpose ? MaxCols : MaxRows, 287 ActualCols = Transpose ? MaxRows : MaxCols 288 }; 289 typedef typename conditional<Transpose,_RhsScalar,_LhsScalar>::type LhsScalar; 290 typedef typename conditional<Transpose,_LhsScalar,_RhsScalar>::type RhsScalar; 291 typedef gebp_traits<LhsScalar,RhsScalar> Traits; 292 enum { 293 SizeA = ActualRows * MaxDepth, 294 SizeB = ActualCols * MaxDepth 295 }; 296 297 #if EIGEN_MAX_STATIC_ALIGN_BYTES >= EIGEN_DEFAULT_ALIGN_BYTES 298 EIGEN_ALIGN_MAX LhsScalar m_staticA[SizeA]; 299 EIGEN_ALIGN_MAX RhsScalar m_staticB[SizeB]; 300 #else 301 EIGEN_ALIGN_MAX char m_staticA[SizeA * sizeof(LhsScalar) + EIGEN_DEFAULT_ALIGN_BYTES-1]; 302 EIGEN_ALIGN_MAX char m_staticB[SizeB * sizeof(RhsScalar) + EIGEN_DEFAULT_ALIGN_BYTES-1]; 303 #endif 304 305 public: 306 307 gemm_blocking_space(Index /*rows*/, Index /*cols*/, Index /*depth*/, Index /*num_threads*/, bool /*full_rows = false*/) 308 { 309 this->m_mc = ActualRows; 310 this->m_nc = ActualCols; 311 this->m_kc = MaxDepth; 312 #if EIGEN_MAX_STATIC_ALIGN_BYTES >= EIGEN_DEFAULT_ALIGN_BYTES 313 this->m_blockA = m_staticA; 314 this->m_blockB = m_staticB; 315 #else 316 this->m_blockA = reinterpret_cast<LhsScalar*>((internal::UIntPtr(m_staticA) + (EIGEN_DEFAULT_ALIGN_BYTES-1)) & ~std::size_t(EIGEN_DEFAULT_ALIGN_BYTES-1)); 317 this->m_blockB = reinterpret_cast<RhsScalar*>((internal::UIntPtr(m_staticB) + (EIGEN_DEFAULT_ALIGN_BYTES-1)) & ~std::size_t(EIGEN_DEFAULT_ALIGN_BYTES-1)); 318 #endif 319 } 320 321 void initParallel(Index, Index, Index, Index) 322 {} 323 324 inline void allocateA() {} 325 inline void allocateB() {} 326 inline void allocateAll() {} 327 }; 328 329 template<int StorageOrder, typename _LhsScalar, typename _RhsScalar, int MaxRows, int MaxCols, int MaxDepth, int KcFactor> 330 class gemm_blocking_space<StorageOrder,_LhsScalar,_RhsScalar,MaxRows, MaxCols, MaxDepth, KcFactor, false> 331 : public level3_blocking< 332 typename conditional<StorageOrder==RowMajor,_RhsScalar,_LhsScalar>::type, 333 typename conditional<StorageOrder==RowMajor,_LhsScalar,_RhsScalar>::type> 334 { 335 enum { 336 Transpose = StorageOrder==RowMajor 337 }; 338 typedef typename conditional<Transpose,_RhsScalar,_LhsScalar>::type LhsScalar; 339 typedef typename conditional<Transpose,_LhsScalar,_RhsScalar>::type RhsScalar; 340 typedef gebp_traits<LhsScalar,RhsScalar> Traits; 341 342 Index m_sizeA; 343 Index m_sizeB; 344 345 public: 346 347 gemm_blocking_space(Index rows, Index cols, Index depth, Index num_threads, bool l3_blocking) 348 { 349 this->m_mc = Transpose ? cols : rows; 350 this->m_nc = Transpose ? rows : cols; 351 this->m_kc = depth; 352 353 if(l3_blocking) 354 { 355 computeProductBlockingSizes<LhsScalar,RhsScalar,KcFactor>(this->m_kc, this->m_mc, this->m_nc, num_threads); 356 } 357 else // no l3 blocking 358 { 359 Index n = this->m_nc; 360 computeProductBlockingSizes<LhsScalar,RhsScalar,KcFactor>(this->m_kc, this->m_mc, n, num_threads); 361 } 362 363 m_sizeA = this->m_mc * this->m_kc; 364 m_sizeB = this->m_kc * this->m_nc; 365 } 366 367 void initParallel(Index rows, Index cols, Index depth, Index num_threads) 368 { 369 this->m_mc = Transpose ? cols : rows; 370 this->m_nc = Transpose ? rows : cols; 371 this->m_kc = depth; 372 373 eigen_internal_assert(this->m_blockA==0 && this->m_blockB==0); 374 Index m = this->m_mc; 375 computeProductBlockingSizes<LhsScalar,RhsScalar,KcFactor>(this->m_kc, m, this->m_nc, num_threads); 376 m_sizeA = this->m_mc * this->m_kc; 377 m_sizeB = this->m_kc * this->m_nc; 378 } 379 380 void allocateA() 381 { 382 if(this->m_blockA==0) 383 this->m_blockA = aligned_new<LhsScalar>(m_sizeA); 384 } 385 386 void allocateB() 387 { 388 if(this->m_blockB==0) 389 this->m_blockB = aligned_new<RhsScalar>(m_sizeB); 390 } 391 392 void allocateAll() 393 { 394 allocateA(); 395 allocateB(); 396 } 397 398 ~gemm_blocking_space() 399 { 400 aligned_delete(this->m_blockA, m_sizeA); 401 aligned_delete(this->m_blockB, m_sizeB); 402 } 403 }; 404 405 } // end namespace internal 406 407 namespace internal { 408 409 template<typename Lhs, typename Rhs> 410 struct generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,GemmProduct> 411 : generic_product_impl_base<Lhs,Rhs,generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,GemmProduct> > 412 { 413 typedef typename Product<Lhs,Rhs>::Scalar Scalar; 414 typedef typename Lhs::Scalar LhsScalar; 415 typedef typename Rhs::Scalar RhsScalar; 416 417 typedef internal::blas_traits<Lhs> LhsBlasTraits; 418 typedef typename LhsBlasTraits::DirectLinearAccessType ActualLhsType; 419 typedef typename internal::remove_all<ActualLhsType>::type ActualLhsTypeCleaned; 420 421 typedef internal::blas_traits<Rhs> RhsBlasTraits; 422 typedef typename RhsBlasTraits::DirectLinearAccessType ActualRhsType; 423 typedef typename internal::remove_all<ActualRhsType>::type ActualRhsTypeCleaned; 424 425 enum { 426 MaxDepthAtCompileTime = EIGEN_SIZE_MIN_PREFER_FIXED(Lhs::MaxColsAtCompileTime,Rhs::MaxRowsAtCompileTime) 427 }; 428 429 typedef generic_product_impl<Lhs,Rhs,DenseShape,DenseShape,CoeffBasedProductMode> lazyproduct; 430 431 template<typename Dst> 432 static void evalTo(Dst& dst, const Lhs& lhs, const Rhs& rhs) 433 { 434 // See http://eigen.tuxfamily.org/bz/show_bug.cgi?id=404 for a discussion and helper program 435 // to determine the following heuristic. 436 // EIGEN_GEMM_TO_COEFFBASED_THRESHOLD is typically defined to 20 in GeneralProduct.h, 437 // unless it has been specialized by the user or for a given architecture. 438 // Note that the condition rhs.rows()>0 was required because lazy product is (was?) not happy with empty inputs. 439 // I'm not sure it is still required. 440 if((rhs.rows()+dst.rows()+dst.cols())<EIGEN_GEMM_TO_COEFFBASED_THRESHOLD && rhs.rows()>0) 441 lazyproduct::eval_dynamic(dst, lhs, rhs, internal::assign_op<typename Dst::Scalar,Scalar>()); 442 else 443 { 444 dst.setZero(); 445 scaleAndAddTo(dst, lhs, rhs, Scalar(1)); 446 } 447 } 448 449 template<typename Dst> 450 static void addTo(Dst& dst, const Lhs& lhs, const Rhs& rhs) 451 { 452 if((rhs.rows()+dst.rows()+dst.cols())<EIGEN_GEMM_TO_COEFFBASED_THRESHOLD && rhs.rows()>0) 453 lazyproduct::eval_dynamic(dst, lhs, rhs, internal::add_assign_op<typename Dst::Scalar,Scalar>()); 454 else 455 scaleAndAddTo(dst,lhs, rhs, Scalar(1)); 456 } 457 458 template<typename Dst> 459 static void subTo(Dst& dst, const Lhs& lhs, const Rhs& rhs) 460 { 461 if((rhs.rows()+dst.rows()+dst.cols())<EIGEN_GEMM_TO_COEFFBASED_THRESHOLD && rhs.rows()>0) 462 lazyproduct::eval_dynamic(dst, lhs, rhs, internal::sub_assign_op<typename Dst::Scalar,Scalar>()); 463 else 464 scaleAndAddTo(dst, lhs, rhs, Scalar(-1)); 465 } 466 467 template<typename Dest> 468 static void scaleAndAddTo(Dest& dst, const Lhs& a_lhs, const Rhs& a_rhs, const Scalar& alpha) 469 { 470 eigen_assert(dst.rows()==a_lhs.rows() && dst.cols()==a_rhs.cols()); 471 if(a_lhs.cols()==0 || a_lhs.rows()==0 || a_rhs.cols()==0) 472 return; 473 474 if (dst.cols() == 1) 475 { 476 // Fallback to GEMV if either the lhs or rhs is a runtime vector 477 typename Dest::ColXpr dst_vec(dst.col(0)); 478 return internal::generic_product_impl<Lhs,typename Rhs::ConstColXpr,DenseShape,DenseShape,GemvProduct> 479 ::scaleAndAddTo(dst_vec, a_lhs, a_rhs.col(0), alpha); 480 } 481 else if (dst.rows() == 1) 482 { 483 // Fallback to GEMV if either the lhs or rhs is a runtime vector 484 typename Dest::RowXpr dst_vec(dst.row(0)); 485 return internal::generic_product_impl<typename Lhs::ConstRowXpr,Rhs,DenseShape,DenseShape,GemvProduct> 486 ::scaleAndAddTo(dst_vec, a_lhs.row(0), a_rhs, alpha); 487 } 488 489 typename internal::add_const_on_value_type<ActualLhsType>::type lhs = LhsBlasTraits::extract(a_lhs); 490 typename internal::add_const_on_value_type<ActualRhsType>::type rhs = RhsBlasTraits::extract(a_rhs); 491 492 Scalar actualAlpha = alpha * LhsBlasTraits::extractScalarFactor(a_lhs) 493 * RhsBlasTraits::extractScalarFactor(a_rhs); 494 495 typedef internal::gemm_blocking_space<(Dest::Flags&RowMajorBit) ? RowMajor : ColMajor,LhsScalar,RhsScalar, 496 Dest::MaxRowsAtCompileTime,Dest::MaxColsAtCompileTime,MaxDepthAtCompileTime> BlockingType; 497 498 typedef internal::gemm_functor< 499 Scalar, Index, 500 internal::general_matrix_matrix_product< 501 Index, 502 LhsScalar, (ActualLhsTypeCleaned::Flags&RowMajorBit) ? RowMajor : ColMajor, bool(LhsBlasTraits::NeedToConjugate), 503 RhsScalar, (ActualRhsTypeCleaned::Flags&RowMajorBit) ? RowMajor : ColMajor, bool(RhsBlasTraits::NeedToConjugate), 504 (Dest::Flags&RowMajorBit) ? RowMajor : ColMajor, 505 Dest::InnerStrideAtCompileTime>, 506 ActualLhsTypeCleaned, ActualRhsTypeCleaned, Dest, BlockingType> GemmFunctor; 507 508 BlockingType blocking(dst.rows(), dst.cols(), lhs.cols(), 1, true); 509 internal::parallelize_gemm<(Dest::MaxRowsAtCompileTime>32 || Dest::MaxRowsAtCompileTime==Dynamic)> 510 (GemmFunctor(lhs, rhs, dst, actualAlpha, blocking), a_lhs.rows(), a_rhs.cols(), a_lhs.cols(), Dest::Flags&RowMajorBit); 511 } 512 }; 513 514 } // end namespace internal 515 516 } // end namespace Eigen 517 518 #endif // EIGEN_GENERAL_MATRIX_MATRIX_H 519