libflame
12600
|
Functions | |
FLA_Error | FLA_Gemm_nc_blk_var6 (FLA_Obj alpha, FLA_Obj A, FLA_Obj B, FLA_Obj beta, FLA_Obj C, fla_gemm_t *cntl) |
FLA_Error FLA_Gemm_nc_blk_var6 | ( | FLA_Obj | alpha, |
FLA_Obj | A, | ||
FLA_Obj | B, | ||
FLA_Obj | beta, | ||
FLA_Obj | C, | ||
fla_gemm_t * | cntl | ||
) |
References FLA_Cont_with_1x3_to_1x2(), FLA_Cont_with_3x1_to_2x1(), FLA_Determine_blocksize(), FLA_Gemm_internal(), FLA_Obj_width(), FLA_ONE, FLA_Part_1x2(), FLA_Part_2x1(), FLA_Repart_1x2_to_1x3(), FLA_Repart_2x1_to_3x1(), and FLA_Scal_internal().
Referenced by FLA_Gemm_nc().
{ FLA_Obj AL, AR, A0, A1, A2; FLA_Obj BT, B0, BB, B1, B2; dim_t b; FLA_Scal_internal( beta, C, FLA_Cntl_sub_scal( cntl ) ); FLA_Part_1x2( A, &AL, &AR, 0, FLA_RIGHT ); FLA_Part_2x1( B, &BT, &BB, 0, FLA_BOTTOM ); while ( FLA_Obj_width( AR ) < FLA_Obj_width( A ) ){ b = FLA_Determine_blocksize( AL, FLA_LEFT, FLA_Cntl_blocksize( cntl ) ); FLA_Repart_1x2_to_1x3( AL, /**/ AR, &A0, &A1, /**/ &A2, b, FLA_LEFT ); FLA_Repart_2x1_to_3x1( BT, &B0, &B1, /* ** */ /* ** */ BB, &B2, b, FLA_TOP ); /*------------------------------------------------------------*/ /* C = alpha * A1 * B1 + C; */ FLA_Gemm_internal( FLA_NO_TRANSPOSE, FLA_CONJ_NO_TRANSPOSE, alpha, A1, B1, FLA_ONE, C, FLA_Cntl_sub_gemm( cntl ) ); /*------------------------------------------------------------*/ FLA_Cont_with_1x3_to_1x2( &AL, /**/ &AR, A0, /**/ A1, A2, FLA_RIGHT ); FLA_Cont_with_3x1_to_2x1( &BT, B0, /* ** */ /* ** */ B1, &BB, B2, FLA_BOTTOM ); } return FLA_SUCCESS; }