FLA_Gemm_ht_blk_var5.c File Reference

(r)


Functions

FLA_Error FLA_Gemm_ht_blk_var5 (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_ht_blk_var5 ( 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().

00036 {
00037   FLA_Obj AT,              A0,
00038           AB,              A1,
00039                            A2;
00040 
00041   FLA_Obj BL,    BR,       B0,  B1,  B2;
00042 
00043   dim_t b;
00044 
00045   FLA_Part_2x1( A,    &AT, 
00046                       &AB,            0, FLA_TOP );
00047 
00048   FLA_Part_1x2( B,    &BL,  &BR,      0, FLA_LEFT );
00049 
00050   while ( FLA_Obj_length( AT ) < FLA_Obj_length( A ) ){
00051 
00052     b = FLA_Determine_blocksize( AB, FLA_BOTTOM, FLA_Cntl_blocksize( cntl ) );
00053 
00054     FLA_Repart_2x1_to_3x1( AT,                &A0, 
00055                         /* ** */            /* ** */
00056                                               &A1, 
00057                            AB,                &A2,        b, FLA_BOTTOM );
00058 
00059     FLA_Repart_1x2_to_1x3( BL,  /**/ BR,        &B0, /**/ &B1, &B2,
00060                            b, FLA_RIGHT );
00061 
00062     /*------------------------------------------------------------*/
00063 
00064     /* C = alpha * A1' * B1' + C; */
00065     FLA_Gemm_internal( FLA_CONJ_TRANSPOSE, FLA_TRANSPOSE,
00066                        alpha, A1, B1, beta, C,
00067                        FLA_Cntl_sub_gemm( cntl ) );
00068 
00069     /*------------------------------------------------------------*/
00070 
00071     FLA_Cont_with_3x1_to_2x1( &AT,                A0, 
00072                                                   A1, 
00073                             /* ** */           /* ** */
00074                               &AB,                A2,     FLA_TOP );
00075 
00076     FLA_Cont_with_1x3_to_1x2( &BL,  /**/ &BR,        B0, B1, /**/ B2,
00077                               FLA_LEFT );
00078 
00079   }
00080 
00081   return FLA_SUCCESS;
00082 }


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