void bli_ger_blk_var1( obj_t* alpha, obj_t* x, obj_t* y, obj_t* a, ger_t* cntl ) { obj_t a1, a1_pack; obj_t x1, x1_pack; dim_t i; dim_t b_alg; dim_t m_trans; // Initialize objects for packing. bli_obj_init_pack( &a1_pack ); bli_obj_init_pack( &x1_pack ); // Query dimension in partitioning direction. m_trans = bli_obj_length_after_trans( *a ); // Partition along the m dimension. for ( i = 0; i < m_trans; i += b_alg ) { // Determine the current algorithmic blocksize. b_alg = bli_determine_blocksize_f( i, m_trans, a, cntl_blocksize( cntl ) ); // Acquire partitions for A1 and x1. bli_acquire_mpart_t2b( BLIS_SUBPART1, i, b_alg, a, &a1 ); bli_acquire_vpart_f2b( BLIS_SUBPART1, i, b_alg, x, &x1 ); // Initialize objects for packing A1 and x1 (if needed). bli_packm_init( &a1, &a1_pack, cntl_sub_packm_a( cntl ) ); bli_packv_init( &x1, &x1_pack, cntl_sub_packv_x( cntl ) ); // Copy/pack A1, x1 (if needed). bli_packm_int( &a1, &a1_pack, cntl_sub_packm_a( cntl ), &BLIS_PACKM_SINGLE_THREADED ); bli_packv_int( &x1, &x1_pack, cntl_sub_packv_x( cntl ) ); // A1 = A1 + alpha * x1 * y; bli_ger_int( BLIS_NO_CONJUGATE, BLIS_NO_CONJUGATE, alpha, &x1_pack, y, &a1_pack, cntl_sub_ger( cntl ) ); // Copy/unpack A1 (if A1 was packed). bli_unpackm_int( &a1_pack, &a1, cntl_sub_unpackm_a( cntl ), &BLIS_PACKM_SINGLE_THREADED ); } // If any packing buffers were acquired within packm, release them back // to the memory manager. bli_packm_release( &a1_pack, cntl_sub_packm_a( cntl ) ); bli_packv_release( &x1_pack, cntl_sub_packv_x( cntl ) ); }
void bli_ger_blk_var2( obj_t* alpha, obj_t* x, obj_t* y, obj_t* a, ger_t* cntl ) { obj_t a1, a1_pack; obj_t y1, y1_pack; dim_t i; dim_t b_alg; dim_t n_trans; // Initialize objects for packing. bli_obj_init_pack( &a1_pack ); bli_obj_init_pack( &y1_pack ); // Query dimension in partitioning direction. n_trans = bli_obj_width_after_trans( *a ); // Partition along the n dimension. for ( i = 0; i < n_trans; i += b_alg ) { // Determine the current algorithmic blocksize. b_alg = bli_determine_blocksize_f( i, n_trans, a, cntl_blocksize( cntl ) ); // Acquire partitions for A1 and y1. bli_acquire_mpart_l2r( BLIS_SUBPART1, i, b_alg, a, &a1 ); bli_acquire_vpart_f2b( BLIS_SUBPART1, i, b_alg, y, &y1 ); // Initialize objects for packing A1 and y1 (if needed). bli_packm_init( &a1, &a1_pack, cntl_sub_packm_a( cntl ) ); bli_packv_init( &y1, &y1_pack, cntl_sub_packv_y( cntl ) ); // Copy/pack A1, y1 (if needed). bli_packm_int( &a1, &a1_pack, cntl_sub_packm_a( cntl ) ); bli_packv_int( &y1, &y1_pack, cntl_sub_packv_y( cntl ) ); // A1 = A1 + alpha * x * y1; bli_ger_int( BLIS_NO_CONJUGATE, BLIS_NO_CONJUGATE, alpha, x, &y1_pack, &a1_pack, cntl_sub_ger( cntl ) ); // Copy/unpack A1 (if A1 was packed). bli_unpackm_int( &a1_pack, &a1, cntl_sub_unpackm_a( cntl ) ); } // If any packing buffers were acquired within packm, release them back // to the memory manager. bli_obj_release_pack( &a1_pack ); bli_obj_release_pack( &y1_pack ); }
void bli_her_blk_var1( conj_t conjh, obj_t* alpha, obj_t* x, obj_t* c, her_t* cntl ) { obj_t c11, c11_pack; obj_t c10; obj_t x1, x1_pack; obj_t x0; dim_t mn; dim_t ij; dim_t b_alg; // Even though this blocked algorithm is expressed only in terms of the // lower triangular case, the upper triangular case is still supported: // when bli_acquire_mpart_tl2br() is passed a matrix that is stored in // in the upper triangle, and the requested subpartition resides in the // lower triangle (as is the case for this algorithm), the routine fills // the request as if the caller had actually requested the corresponding // "mirror" subpartition in the upper triangle, except that it marks the // subpartition for transposition (and conjugation). // Initialize objects for packing. bli_obj_init_pack( &c11_pack ); bli_obj_init_pack( &x1_pack ); // Query dimension. mn = bli_obj_length( *c ); // Partition diagonally. for ( ij = 0; ij < mn; ij += b_alg ) { // Determine the current algorithmic blocksize. b_alg = bli_determine_blocksize_f( ij, mn, c, cntl_blocksize( cntl ) ); // Acquire partitions for C11, C10, x1, and x0. bli_acquire_mpart_tl2br( BLIS_SUBPART11, ij, b_alg, c, &c11 ); bli_acquire_mpart_tl2br( BLIS_SUBPART10, ij, b_alg, c, &c10 ); bli_acquire_vpart_f2b( BLIS_SUBPART1, ij, b_alg, x, &x1 ); bli_acquire_vpart_f2b( BLIS_SUBPART0, ij, b_alg, x, &x0 ); // Initialize objects for packing C11 and x1 (if needed). bli_packm_init( &c11, &c11_pack, cntl_sub_packm_c11( cntl ) ); bli_packv_init( &x1, &x1_pack, cntl_sub_packv_x1( cntl ) ); // Copy/pack C11, x1 (if needed). bli_packm_int( &c11, &c11_pack, cntl_sub_packm_c11( cntl ), &BLIS_PACKM_SINGLE_THREADED ); bli_packv_int( &x1, &x1_pack, cntl_sub_packv_x1( cntl ) ); // C10 = C10 + alpha * x1 * x0'; bli_ger_int( BLIS_NO_CONJUGATE, conjh, alpha, &x1_pack, &x0, &c10, cntl_sub_ger( cntl ) ); // C11 = C11 + alpha * x1 * x1'; bli_her_int( conjh, alpha, &x1_pack, &c11_pack, cntl_sub_her( cntl ) ); // Copy/unpack C11 (if C11 was packed). bli_unpackm_int( &c11_pack, &c11, cntl_sub_unpackm_c11( cntl ), &BLIS_PACKM_SINGLE_THREADED ); } // If any packing buffers were acquired within packm, release them back // to the memory manager. bli_packm_release( &c11_pack, cntl_sub_packm_c11( cntl ) ); bli_packv_release( &x1_pack, cntl_sub_packv_x1( cntl ) ); }