GeneralBlockPanelKernel.h File Reference

Go to the source code of this file.

Classes
struct	Eigen::internal::CacheSizes
struct	Eigen::internal::RhsPanelHelper< RhsPacket, RhsPacketx4, registers_taken >
struct	Eigen::internal::QuadPacket< Packet >
struct	Eigen::internal::packet_conditional< N, T1, T2, T3 >
struct	Eigen::internal::packet_conditional< GEBPPacketFull, T1, T2, T3 >
struct	Eigen::internal::packet_conditional< GEBPPacketHalf, T1, T2, T3 >
class	Eigen::internal::gebp_traits< _LhsScalar, _RhsScalar, _ConjLhs, _ConjRhs, Arch, _PacketSize >
class	Eigen::internal::gebp_traits< std::complex< RealScalar >, RealScalar, _ConjLhs, false, Arch, _PacketSize >
struct	Eigen::internal::DoublePacket< Packet >
struct	Eigen::internal::unpacket_traits< DoublePacket< Packet > >
class	Eigen::internal::gebp_traits< std::complex< RealScalar >, std::complex< RealScalar >, _ConjLhs, _ConjRhs, Arch, _PacketSize >
class	Eigen::internal::gebp_traits< RealScalar, std::complex< RealScalar >, false, _ConjRhs, Arch, _PacketSize >
struct	Eigen::internal::gebp_kernel< LhsScalar, RhsScalar, Index, DataMapper, mr, nr, ConjugateLhs, ConjugateRhs >
struct	Eigen::internal::last_row_process_16_packets< LhsScalar, RhsScalar, Index, DataMapper, mr, nr, ConjugateLhs, ConjugateRhs, SwappedLhsProgress >
struct	Eigen::internal::last_row_process_16_packets< LhsScalar, RhsScalar, Index, DataMapper, mr, nr, ConjugateLhs, ConjugateRhs, 16 >
struct	Eigen::internal::lhs_process_one_packet< nr, LhsProgress, RhsProgress, LhsScalar, RhsScalar, ResScalar, AccPacket, LhsPacket, RhsPacket, ResPacket, GEBPTraits, LinearMapper, DataMapper >
struct	Eigen::internal::lhs_process_fraction_of_packet< nr, LhsProgress, RhsProgress, LhsScalar, RhsScalar, ResScalar, AccPacket, LhsPacket, RhsPacket, ResPacket, GEBPTraits, LinearMapper, DataMapper >
struct	Eigen::internal::gemm_pack_lhs< Scalar, Index, DataMapper, Pack1, Pack2, Packet, ColMajor, Conjugate, PanelMode >
struct	Eigen::internal::gemm_pack_lhs< Scalar, Index, DataMapper, Pack1, Pack2, Packet, RowMajor, Conjugate, PanelMode >
struct	Eigen::internal::gemm_pack_rhs< Scalar, Index, DataMapper, nr, ColMajor, Conjugate, PanelMode >
struct	Eigen::internal::gemm_pack_rhs< Scalar, Index, DataMapper, nr, RowMajor, Conjugate, PanelMode >

Namespaces
namespace	Eigen
	Namespace containing all symbols from the Eigen library.
namespace	Eigen::internal

Macros
#define	EIGEN_SET_DEFAULT_L1_CACHE_SIZE(val)   val
#define	EIGEN_SET_DEFAULT_L2_CACHE_SIZE(val)   val
#define	EIGEN_SET_DEFAULT_L3_CACHE_SIZE(val)   val
#define	PACKET_DECL_COND_PREFIX(prefix, name, packet_size)
#define	PACKET_DECL_COND(name, packet_size)
#define	PACKET_DECL_COND_SCALAR_PREFIX(prefix, packet_size)
#define	PACKET_DECL_COND_SCALAR(packet_size)
#define	EIGEN_GEBGP_ONESTEP(K)
#define	EIGEN_GEBP_3PX4_REGISTER_ALLOC_WORKAROUND
#define	EIGEN_GEBP_ONESTEP(K)
#define	EIGEN_GEBGP_ONESTEP(K)
#define	EIGEN_GEBP_2PX4_SPILLING_WORKAROUND
#define	EIGEN_GEBGP_ONESTEP(K)
#define	EIGEN_GEBGP_ONESTEP(K)

Enumerations
enum	Eigen::internal::GEBPPacketSizeType { Eigen::internal::GEBPPacketFull = 0 , Eigen::internal::GEBPPacketHalf , Eigen::internal::GEBPPacketQuarter }

Functions
std::ptrdiff_t	Eigen::internal::manage_caching_sizes_helper (std::ptrdiff_t a, std::ptrdiff_t b)
void	Eigen::internal::manage_caching_sizes (Action action, std::ptrdiff_t *l1, std::ptrdiff_t *l2, std::ptrdiff_t *l3)
template<typename LhsScalar , typename RhsScalar , int KcFactor, typename Index >
void	Eigen::internal::evaluateProductBlockingSizesHeuristic (Index &k, Index &m, Index &n, Index num_threads=1)
template<typename Index >
bool	Eigen::internal::useSpecificBlockingSizes (Index &k, Index &m, Index &n)
template<typename LhsScalar , typename RhsScalar , int KcFactor, typename Index >
void	Eigen::internal::computeProductBlockingSizes (Index &k, Index &m, Index &n, Index num_threads=1)
	Computes the blocking parameters for a m x k times k x n matrix product.
template<typename LhsScalar , typename RhsScalar , typename Index >
void	Eigen::internal::computeProductBlockingSizes (Index &k, Index &m, Index &n, Index num_threads=1)
template<typename Packet >
DoublePacket< Packet >	Eigen::internal::padd (const DoublePacket< Packet > &a, const DoublePacket< Packet > &b)
template<typename Packet >
const DoublePacket< Packet > &	Eigen::internal::predux_half_dowto4 (const DoublePacket< Packet > &a, typename enable_if< unpacket_traits< Packet >::size<=8 >::type *=0)
template<typename Packet >
DoublePacket< typename unpacket_traits< Packet >::half >	Eigen::internal::predux_half_dowto4 (const DoublePacket< Packet > &a, typename enable_if< unpacket_traits< Packet >::size==16 >::type *=0)
template<typename Scalar , typename RealPacket >
void	Eigen::internal::loadQuadToDoublePacket (const Scalar *b, DoublePacket< RealPacket > &dest, typename enable_if< unpacket_traits< RealPacket >::size<=8 >::type *=0)
template<typename Scalar , typename RealPacket >
void	Eigen::internal::loadQuadToDoublePacket (const Scalar *b, DoublePacket< RealPacket > &dest, typename enable_if< unpacket_traits< RealPacket >::size==16 >::type *=0)
std::ptrdiff_t	Eigen::l1CacheSize ()
std::ptrdiff_t	Eigen::l2CacheSize ()
std::ptrdiff_t	Eigen::l3CacheSize ()
void	Eigen::setCpuCacheSizes (std::ptrdiff_t l1, std::ptrdiff_t l2, std::ptrdiff_t l3)

Variables
const std::ptrdiff_t	Eigen::internal::defaultL1CacheSize = EIGEN_SET_DEFAULT_L1_CACHE_SIZE(16*1024)
const std::ptrdiff_t	Eigen::internal::defaultL2CacheSize = EIGEN_SET_DEFAULT_L2_CACHE_SIZE(512*1024)
const std::ptrdiff_t	Eigen::internal::defaultL3CacheSize = EIGEN_SET_DEFAULT_L3_CACHE_SIZE(512*1024)

Macro Definition Documentation

EIGEN_GEBGP_ONESTEP [1/4]

#define EIGEN_GEBGP_ONESTEP

(

K

)

Value:

          do {                                                      \
        EIGEN_ASM_COMMENT("begin step of gebp micro kernel 1/half/quarterX1"); \
        EIGEN_ASM_COMMENT("Note: these asm comments work around bug 935!"); \
    /* FIXME: why unaligned???? */ \
        traits.loadLhsUnaligned(&blA[(0+1*K)*LhsProgress], A0); \
        traits.loadRhs(&blB[(0+K)*RhsProgress], B_0);       \
        traits.madd(A0, B_0, C0, B_0, fix<0>);              \
        EIGEN_ASM_COMMENT("end step of gebp micro kernel 1/half/quarterX1"); \
          } while(false);

EIGEN_GEBGP_ONESTEP [2/4]

#define EIGEN_GEBGP_ONESTEP

(

K

)

Value:

            do {                                                                  \
              EIGEN_ASM_COMMENT("begin step of gebp micro kernel 3pX1");          \
              EIGEN_ASM_COMMENT("Note: these asm comments work around bug 935!"); \
              traits.loadLhs(&blA[(0 + 3 * K) * LhsProgress], A0);                \
              traits.loadLhs(&blA[(1 + 3 * K) * LhsProgress], A1);                \
              traits.loadLhs(&blA[(2 + 3 * K) * LhsProgress], A2);                \
              traits.loadRhs(&blB[(0 + K) * RhsProgress], B_0);                   \
              traits.madd(A0, B_0, C0, B_0, fix<0>);                              \
              traits.madd(A1, B_0, C4, B_0, fix<0>);                              \
              traits.madd(A2, B_0, C8, B_0, fix<0>);                              \
              EIGEN_ASM_COMMENT("end step of gebp micro kernel 3pX1");            \
            } while (false)

EIGEN_GEBGP_ONESTEP [3/4]

#define EIGEN_GEBGP_ONESTEP

(

K

)

Value:

            do {                                                          \
              EIGEN_ASM_COMMENT("begin step of gebp micro kernel 2pX4");  \
              traits.loadLhs(&blA[(0 + 2 * K) * LhsProgress], A0);        \
              traits.loadLhs(&blA[(1 + 2 * K) * LhsProgress], A1);        \
              traits.loadRhs(&blB[(0 + 4 * K) * RhsProgress], rhs_panel); \
              traits.madd(A0, rhs_panel, C0, T0, fix<0>);                 \
              traits.madd(A1, rhs_panel, C4, T0, fix<0>);                 \
              traits.madd(A0, rhs_panel, C1, T0, fix<1>);                 \
              traits.madd(A1, rhs_panel, C5, T0, fix<1>);                 \
              traits.madd(A0, rhs_panel, C2, T0, fix<2>);                 \
              traits.madd(A1, rhs_panel, C6, T0, fix<2>);                 \
              traits.madd(A0, rhs_panel, C3, T0, fix<3>);                 \
              traits.madd(A1, rhs_panel, C7, T0, fix<3>);                 \
              EIGEN_GEBP_2PX4_SPILLING_WORKAROUND                         \
              EIGEN_ASM_COMMENT("end step of gebp micro kernel 2pX4");    \
            } while (false)

EIGEN_GEBGP_ONESTEP [4/4]

#define EIGEN_GEBGP_ONESTEP

(

K

)

Value:

            do {                                                                  \
              EIGEN_ASM_COMMENT("begin step of gebp micro kernel 2pX1");          \
              EIGEN_ASM_COMMENT("Note: these asm comments work around bug 935!"); \
              traits.loadLhs(&blA[(0+2*K)*LhsProgress], A0);                      \
              traits.loadLhs(&blA[(1+2*K)*LhsProgress], A1);                      \
              traits.loadRhs(&blB[(0+K)*RhsProgress], B_0);                       \
              traits.madd(A0, B_0, C0, B1, fix<0>);                               \
              traits.madd(A1, B_0, C4, B_0, fix<0>);                              \
              EIGEN_ASM_COMMENT("end step of gebp micro kernel 2pX1");            \
            } while(false)

EIGEN_GEBP_2PX4_SPILLING_WORKAROUND

#define EIGEN_GEBP_2PX4_SPILLING_WORKAROUND

EIGEN_GEBP_3PX4_REGISTER_ALLOC_WORKAROUND

#define EIGEN_GEBP_3PX4_REGISTER_ALLOC_WORKAROUND

EIGEN_GEBP_ONESTEP

#define EIGEN_GEBP_ONESTEP

(

K

)

Value:

            do {                                                                  \
              EIGEN_ASM_COMMENT("begin step of gebp micro kernel 3pX4");          \
              EIGEN_ASM_COMMENT("Note: these asm comments work around bug 935!"); \
              internal::prefetch(blA + (3 * K + 16) * LhsProgress);               \
              if (EIGEN_ARCH_ARM || EIGEN_ARCH_MIPS) {                            \
                internal::prefetch(blB + (4 * K + 16) * RhsProgress);             \
              } /* Bug 953 */                                                     \
              traits.loadLhs(&blA[(0 + 3 * K) * LhsProgress], A0);                \
              traits.loadLhs(&blA[(1 + 3 * K) * LhsProgress], A1);                \
              traits.loadLhs(&blA[(2 + 3 * K) * LhsProgress], A2);                \
              EIGEN_GEBP_3PX4_REGISTER_ALLOC_WORKAROUND \
              traits.loadRhs(blB + (0+4*K) * Traits::RhsProgress, rhs_panel);     \
              traits.madd(A0, rhs_panel, C0, T0, fix<0>);                         \
              traits.madd(A1, rhs_panel, C4, T0, fix<0>);                         \
              traits.madd(A2, rhs_panel, C8, T0, fix<0>);                         \
              traits.updateRhs(blB + (1+4*K) * Traits::RhsProgress, rhs_panel);   \
              traits.madd(A0, rhs_panel, C1, T0, fix<1>);                         \
              traits.madd(A1, rhs_panel, C5, T0, fix<1>);                         \
              traits.madd(A2, rhs_panel, C9, T0, fix<1>);                         \
              traits.updateRhs(blB + (2+4*K) * Traits::RhsProgress, rhs_panel);   \
              traits.madd(A0, rhs_panel, C2, T0, fix<2>);                         \
              traits.madd(A1, rhs_panel, C6, T0, fix<2>);                         \
              traits.madd(A2, rhs_panel, C10, T0, fix<2>);                        \
              traits.updateRhs(blB + (3+4*K) * Traits::RhsProgress, rhs_panel);   \
              traits.madd(A0, rhs_panel, C3, T0, fix<3>);                         \
              traits.madd(A1, rhs_panel, C7, T0, fix<3>);                         \
              traits.madd(A2, rhs_panel, C11, T0, fix<3>);                        \
              EIGEN_ASM_COMMENT("end step of gebp micro kernel 3pX4");            \
            } while (false)

EIGEN_SET_DEFAULT_L1_CACHE_SIZE

#define EIGEN_SET_DEFAULT_L1_CACHE_SIZE

(

val

)

val

EIGEN_SET_DEFAULT_L2_CACHE_SIZE

#define EIGEN_SET_DEFAULT_L2_CACHE_SIZE

(

val

)

val

EIGEN_SET_DEFAULT_L3_CACHE_SIZE

#define EIGEN_SET_DEFAULT_L3_CACHE_SIZE

(

val

)

val

PACKET_DECL_COND

#define PACKET_DECL_COND	(		name,
			packet_size
	)

Value:

  typedef typename packet_conditional<packet_size,                 \
                                      typename packet_traits<name ## Scalar>::type, \
                                      typename packet_traits<name ## Scalar>::half, \
                                      typename unpacket_traits<typename packet_traits<name ## Scalar>::half>::half>::type \
  name ## Packet

PACKET_DECL_COND_PREFIX

#define PACKET_DECL_COND_PREFIX	(		prefix,
			name,
			packet_size
	)

Value:

  typedef typename packet_conditional<packet_size,                 \
                                      typename packet_traits<name ## Scalar>::type, \
                                      typename packet_traits<name ## Scalar>::half, \
                                      typename unpacket_traits<typename packet_traits<name ## Scalar>::half>::half>::type \
  prefix ## name ## Packet

PACKET_DECL_COND_SCALAR

#define PACKET_DECL_COND_SCALAR

(

packet_size

)

Value:

  typedef typename packet_conditional<packet_size,                 \
                                      typename packet_traits<Scalar>::type, \
                                      typename packet_traits<Scalar>::half, \
                                      typename unpacket_traits<typename packet_traits<Scalar>::half>::half>::type \
  ScalarPacket

PACKET_DECL_COND_SCALAR_PREFIX

#define PACKET_DECL_COND_SCALAR_PREFIX	(		prefix,
			packet_size
	)

Value:

  typedef typename packet_conditional<packet_size,                 \
                                      typename packet_traits<Scalar>::type, \
                                      typename packet_traits<Scalar>::half, \
                                      typename unpacket_traits<typename packet_traits<Scalar>::half>::half>::type \
  prefix ## ScalarPacket