libflame
12600
|
Functions | |
FLA_Error | FLA_Gemm_ht_blk_var3 (FLA_Obj alpha, FLA_Obj A, FLA_Obj B, FLA_Obj beta, FLA_Obj C, fla_gemm_t *cntl) |
FLA_Error FLA_Gemm_ht_blk_var3 | ( | 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_length(), FLA_Part_1x2(), FLA_Part_2x1(), FLA_Repart_1x2_to_1x3(), and FLA_Repart_2x1_to_3x1().
Referenced by FLA_Gemm_ht().
{ FLA_Obj BT, B0, BB, B1, B2; FLA_Obj CL, CR, C0, C1, C2; dim_t b; FLA_Part_2x1( B, &BT, &BB, 0, FLA_TOP ); FLA_Part_1x2( C, &CL, &CR, 0, FLA_LEFT ); while ( FLA_Obj_length( BT ) < FLA_Obj_length( B ) ){ b = FLA_Determine_blocksize( BB, FLA_BOTTOM, FLA_Cntl_blocksize( cntl ) ); FLA_Repart_2x1_to_3x1( BT, &B0, /* ** */ /* ** */ &B1, BB, &B2, b, FLA_BOTTOM ); FLA_Repart_1x2_to_1x3( CL, /**/ CR, &C0, /**/ &C1, &C2, b, FLA_RIGHT ); /*------------------------------------------------------------*/ /* C1 = alpha * A' * B1' + C1; */ FLA_Gemm_internal( FLA_CONJ_TRANSPOSE, FLA_TRANSPOSE, alpha, A, B1, beta, C1, FLA_Cntl_sub_gemm( cntl ) ); /*------------------------------------------------------------*/ FLA_Cont_with_3x1_to_2x1( &BT, B0, B1, /* ** */ /* ** */ &BB, B2, FLA_TOP ); FLA_Cont_with_1x3_to_1x2( &CL, /**/ &CR, C0, C1, /**/ C2, FLA_LEFT ); } return FLA_SUCCESS; }