FLA_Gemm_nh_blk_var1.c File Reference

(r)


Functions

FLA_Error FLA_Gemm_nh_blk_var1 (FLA_Obj alpha, FLA_Obj A, FLA_Obj B, FLA_Obj beta, FLA_Obj C, fla_gemm_t *cntl)

Function Documentation

FLA_Error FLA_Gemm_nh_blk_var1 ( FLA_Obj  alpha,
FLA_Obj  A,
FLA_Obj  B,
FLA_Obj  beta,
FLA_Obj  C,
fla_gemm_t cntl 
)

References FLA_Cont_with_3x1_to_2x1(), FLA_Determine_blocksize(), FLA_Gemm_internal(), FLA_Obj_length(), FLA_Part_2x1(), and FLA_Repart_2x1_to_3x1().

Referenced by FLA_Gemm_nh().

00036 {
00037   FLA_Obj AT,              A0,
00038           AB,              A1,
00039                            A2;
00040 
00041   FLA_Obj CT,              C0,
00042           CB,              C1,
00043                            C2;
00044 
00045   dim_t b;
00046 
00047   FLA_Part_2x1( A,    &AT, 
00048                       &AB,            0, FLA_TOP );
00049 
00050   FLA_Part_2x1( C,    &CT, 
00051                       &CB,            0, FLA_TOP );
00052 
00053   while ( FLA_Obj_length( AT ) < FLA_Obj_length( A ) ){
00054 
00055     b = FLA_Determine_blocksize( AB, FLA_BOTTOM, FLA_Cntl_blocksize( cntl ) );
00056 
00057     FLA_Repart_2x1_to_3x1( AT,                &A0, 
00058                         /* ** */            /* ** */
00059                                               &A1, 
00060                            AB,                &A2,        b, FLA_BOTTOM );
00061 
00062     FLA_Repart_2x1_to_3x1( CT,                &C0, 
00063                         /* ** */            /* ** */
00064                                               &C1, 
00065                            CB,                &C2,        b, FLA_BOTTOM );
00066 
00067     /*------------------------------------------------------------*/
00068 
00069     /* C1 = alpha * A1 * B' + C1; */
00070     FLA_Gemm_internal( FLA_NO_TRANSPOSE, FLA_CONJ_TRANSPOSE,
00071                        alpha, A1, B, beta, C1,
00072                        FLA_Cntl_sub_gemm( cntl ) );
00073 
00074     /*------------------------------------------------------------*/
00075 
00076     FLA_Cont_with_3x1_to_2x1( &AT,                A0, 
00077                                                   A1, 
00078                             /* ** */           /* ** */
00079                               &AB,                A2,     FLA_TOP );
00080 
00081     FLA_Cont_with_3x1_to_2x1( &CT,                C0, 
00082                                                   C1, 
00083                             /* ** */           /* ** */
00084                               &CB,                C2,     FLA_TOP );
00085 
00086   }
00087 
00088   return FLA_SUCCESS;
00089 }


Generated on Mon Jul 6 05:45:54 2009 for libflame by  doxygen 1.5.9