libflame
12600
|
Functions | |
FLA_Error | FLA_Hemm_ll_blk_var5 (FLA_Obj alpha, FLA_Obj A, FLA_Obj B, FLA_Obj beta, FLA_Obj C, fla_hemm_t *cntl) |
FLA_Error FLA_Hemm_ll_blk_var5 | ( | FLA_Obj | alpha, |
FLA_Obj | A, | ||
FLA_Obj | B, | ||
FLA_Obj | beta, | ||
FLA_Obj | C, | ||
fla_hemm_t * | cntl | ||
) |
References FLA_Cont_with_3x1_to_2x1(), FLA_Cont_with_3x3_to_2x2(), FLA_Determine_blocksize(), FLA_Gemm_internal(), FLA_Hemm_internal(), FLA_Obj_length(), FLA_ONE, FLA_Part_2x1(), FLA_Part_2x2(), FLA_Repart_2x1_to_3x1(), FLA_Repart_2x2_to_3x3(), and FLA_Scal_internal().
Referenced by FLA_Hemm_ll().
{ FLA_Obj ATL, ATR, A00, A01, A02, ABL, ABR, A10, A11, A12, A20, A21, A22; FLA_Obj BT, B0, BB, B1, B2; FLA_Obj CT, C0, CB, C1, C2; dim_t b; FLA_Scal_internal( beta, C, FLA_Cntl_sub_scal( cntl ) ); FLA_Part_2x2( A, &ATL, &ATR, &ABL, &ABR, 0, 0, FLA_BR ); FLA_Part_2x1( B, &BT, &BB, 0, FLA_BOTTOM ); FLA_Part_2x1( C, &CT, &CB, 0, FLA_BOTTOM ); while ( FLA_Obj_length( ABR ) < FLA_Obj_length( A ) ){ b = FLA_Determine_blocksize( ATL, FLA_TL, FLA_Cntl_blocksize( cntl ) ); FLA_Repart_2x2_to_3x3( ATL, /**/ ATR, &A00, &A01, /**/ &A02, &A10, &A11, /**/ &A12, /* ************* */ /* ******************** */ ABL, /**/ ABR, &A20, &A21, /**/ &A22, b, b, FLA_TL ); FLA_Repart_2x1_to_3x1( BT, &B0, &B1, /* ** */ /* ** */ BB, &B2, b, FLA_TOP ); FLA_Repart_2x1_to_3x1( CT, &C0, &C1, /* ** */ /* ** */ CB, &C2, b, FLA_TOP ); /*------------------------------------------------------------*/ /* C1 = C1 + A11 * B1 */ FLA_Hemm_internal( FLA_LEFT, FLA_LOWER_TRIANGULAR, alpha, A11, B1, FLA_ONE, C1, FLA_Cntl_sub_hemm( cntl ) ); /* C1 = C1 + A21' * B2 */ FLA_Gemm_internal( FLA_CONJ_TRANSPOSE, FLA_NO_TRANSPOSE, alpha, A21, B2, FLA_ONE, C1, FLA_Cntl_sub_gemm1( cntl ) ); /* C2 = C2 + A21 * B1 */ FLA_Gemm_internal( FLA_NO_TRANSPOSE, FLA_NO_TRANSPOSE, alpha, A21, B1, FLA_ONE, C2, FLA_Cntl_sub_gemm2( cntl ) ); /*------------------------------------------------------------*/ FLA_Cont_with_3x3_to_2x2( &ATL, /**/ &ATR, A00, /**/ A01, A02, /* ************** */ /* ****************** */ A10, /**/ A11, A12, &ABL, /**/ &ABR, A20, /**/ A21, A22, FLA_BR ); FLA_Cont_with_3x1_to_2x1( &BT, B0, /* ** */ /* ** */ B1, &BB, B2, FLA_BOTTOM ); FLA_Cont_with_3x1_to_2x1( &CT, C0, /* ** */ /* ** */ C1, &CB, C2, FLA_BOTTOM ); } return FLA_SUCCESS; }