libflame
12600
|
Functions | |
FLA_Error | FLA_Gemm_hn_blk_var4 (FLA_Obj alpha, FLA_Obj A, FLA_Obj B, FLA_Obj beta, FLA_Obj C, fla_gemm_t *cntl) |
FLA_Error FLA_Gemm_hn_blk_var4 | ( | 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_Determine_blocksize(), FLA_Gemm_internal(), FLA_Obj_width(), FLA_Part_1x2(), and FLA_Repart_1x2_to_1x3().
Referenced by FLA_Gemm_hn().
{ FLA_Obj BL, BR, B0, B1, B2; FLA_Obj CL, CR, C0, C1, C2; dim_t b; FLA_Part_1x2( B, &BL, &BR, 0, FLA_RIGHT ); FLA_Part_1x2( C, &CL, &CR, 0, FLA_RIGHT ); while ( FLA_Obj_width( BR ) < FLA_Obj_width( B ) ){ b = FLA_Determine_blocksize( BL, FLA_LEFT, FLA_Cntl_blocksize( cntl ) ); FLA_Repart_1x2_to_1x3( BL, /**/ BR, &B0, &B1, /**/ &B2, b, FLA_LEFT ); FLA_Repart_1x2_to_1x3( CL, /**/ CR, &C0, &C1, /**/ &C2, b, FLA_LEFT ); /*------------------------------------------------------------*/ /* C1 = alpha * A' * B1 + C1; */ FLA_Gemm_internal( FLA_CONJ_TRANSPOSE, FLA_NO_TRANSPOSE, alpha, A, B1, beta, C1, FLA_Cntl_sub_gemm( cntl ) ); /*------------------------------------------------------------*/ FLA_Cont_with_1x3_to_1x2( &BL, /**/ &BR, B0, /**/ B1, B2, FLA_RIGHT ); FLA_Cont_with_1x3_to_1x2( &CL, /**/ &CR, C0, /**/ C1, C2, FLA_RIGHT ); } return FLA_SUCCESS; }