libflame
revision_anchor
|
Functions | |
void | bli_strmmsx (side_t side, uplo_t uplo, trans_t trans, diag_t diag, int m, int n, float *alpha, float *a, int a_rs, int a_cs, float *b, int b_rs, int b_cs, float *beta, float *c, int c_rs, int c_cs) |
void | bli_dtrmmsx (side_t side, uplo_t uplo, trans_t trans, diag_t diag, int m, int n, double *alpha, double *a, int a_rs, int a_cs, double *b, int b_rs, int b_cs, double *beta, double *c, int c_rs, int c_cs) |
void | bli_ctrmmsx (side_t side, uplo_t uplo, trans_t trans, diag_t diag, int m, int n, scomplex *alpha, scomplex *a, int a_rs, int a_cs, scomplex *b, int b_rs, int b_cs, scomplex *beta, scomplex *c, int c_rs, int c_cs) |
void | bli_ztrmmsx (side_t side, uplo_t uplo, trans_t trans, diag_t diag, int m, int n, dcomplex *alpha, dcomplex *a, int a_rs, int a_cs, dcomplex *b, int b_rs, int b_cs, dcomplex *beta, dcomplex *c, int c_rs, int c_cs) |
void bli_ctrmmsx | ( | side_t | side, |
uplo_t | uplo, | ||
trans_t | trans, | ||
diag_t | diag, | ||
int | m, | ||
int | n, | ||
scomplex * | alpha, | ||
scomplex * | a, | ||
int | a_rs, | ||
int | a_cs, | ||
scomplex * | b, | ||
int | b_rs, | ||
int | b_cs, | ||
scomplex * | beta, | ||
scomplex * | c, | ||
int | c_rs, | ||
int | c_cs | ||
) |
References bli_c1(), bli_callocm(), bli_caxpymt(), bli_ccopymt(), bli_ccreate_contigm(), bli_ccreate_contigmr(), bli_cfree(), bli_cfree_contigm(), bli_cfree_saved_contigm(), bli_cscalm(), bli_ctrmm(), bli_is_col_storage(), bli_set_dim_with_side(), bli_zero_dim2(), BLIS_NO_CONJUGATE, and BLIS_NO_TRANSPOSE.
Referenced by FLA_Trmmsx_external().
{ int m_save = m; int n_save = n; scomplex* a_save = a; scomplex* b_save = b; scomplex* c_save = c; int a_rs_save = a_rs; int a_cs_save = a_cs; int b_rs_save = b_rs; int b_cs_save = b_cs; int c_rs_save = c_rs; int c_cs_save = c_cs; scomplex one = bli_c1(); scomplex* b_copy; int dim_a; int b_copy_rs, b_copy_cs; // Return early if possible. if ( bli_zero_dim2( m, n ) ) return; // If necessary, allocate, initialize, and use a temporary contiguous // copy of each matrix rather than the original matrices. bli_set_dim_with_side( side, m, n, &dim_a ); bli_ccreate_contigmr( uplo, dim_a, dim_a, a_save, a_rs_save, a_cs_save, &a, &a_rs, &a_cs ); bli_ccreate_contigm( m, n, b_save, b_rs_save, b_cs_save, &b, &b_rs, &b_cs ); bli_ccreate_contigm( m, n, c_save, c_rs_save, c_cs_save, &c, &c_rs, &c_cs ); // Create a copy of B to use in the computation so the original matrix is // left untouched. b_copy = bli_callocm( m, n ); // Match the strides of B_copy to that of B. if ( bli_is_col_storage( b_rs, b_cs ) ) { b_copy_rs = 1; b_copy_cs = m; } else // if ( bli_is_row_storage( b_rs, b_cs ) ) { b_copy_rs = n; b_copy_cs = 1; } // Copy the contents of B to B_copy. bli_ccopymt( BLIS_NO_TRANSPOSE, m, n, b, b_rs, b_cs, b_copy, b_copy_rs, b_copy_cs ); // Perform the operation on B_copy. bli_ctrmm( side, uplo, trans, diag, m, n, alpha, a, a_rs, a_cs, b_copy, b_copy_rs, b_copy_cs ); // Scale C by beta. bli_cscalm( BLIS_NO_CONJUGATE, m, n, beta, c, c_rs, c_cs ); // Add B_copy into C. bli_caxpymt( BLIS_NO_TRANSPOSE, m, n, &one, b_copy, b_copy_rs, b_copy_cs, c, c_rs, c_cs ); // Free the copy of B. bli_cfree( b_copy ); // Free any temporary contiguous matrices, copying the result back to // the original matrix. bli_cfree_contigm( a_save, a_rs_save, a_cs_save, &a, &a_rs, &a_cs ); bli_cfree_contigm( b_save, b_rs_save, b_cs_save, &b, &b_rs, &b_cs ); bli_cfree_saved_contigm( m_save, n_save, c_save, c_rs_save, c_cs_save, &c, &c_rs, &c_cs ); }
void bli_dtrmmsx | ( | side_t | side, |
uplo_t | uplo, | ||
trans_t | trans, | ||
diag_t | diag, | ||
int | m, | ||
int | n, | ||
double * | alpha, | ||
double * | a, | ||
int | a_rs, | ||
int | a_cs, | ||
double * | b, | ||
int | b_rs, | ||
int | b_cs, | ||
double * | beta, | ||
double * | c, | ||
int | c_rs, | ||
int | c_cs | ||
) |
References bli_d1(), bli_dallocm(), bli_daxpymt(), bli_dcopymt(), bli_dcreate_contigm(), bli_dcreate_contigmr(), bli_dfree(), bli_dfree_contigm(), bli_dfree_saved_contigm(), bli_dscalm(), bli_dtrmm(), bli_is_col_storage(), bli_set_dim_with_side(), bli_zero_dim2(), BLIS_NO_CONJUGATE, and BLIS_NO_TRANSPOSE.
Referenced by FLA_Trmmsx_external().
{ int m_save = m; int n_save = n; double* a_save = a; double* b_save = b; double* c_save = c; int a_rs_save = a_rs; int a_cs_save = a_cs; int b_rs_save = b_rs; int b_cs_save = b_cs; int c_rs_save = c_rs; int c_cs_save = c_cs; double one = bli_d1(); double* b_copy; int dim_a; int b_copy_rs, b_copy_cs; // Return early if possible. if ( bli_zero_dim2( m, n ) ) return; // If necessary, allocate, initialize, and use a temporary contiguous // copy of each matrix rather than the original matrices. bli_set_dim_with_side( side, m, n, &dim_a ); bli_dcreate_contigmr( uplo, dim_a, dim_a, a_save, a_rs_save, a_cs_save, &a, &a_rs, &a_cs ); bli_dcreate_contigm( m, n, b_save, b_rs_save, b_cs_save, &b, &b_rs, &b_cs ); bli_dcreate_contigm( m, n, c_save, c_rs_save, c_cs_save, &c, &c_rs, &c_cs ); // Create a copy of B to use in the computation so the original matrix is // left untouched. b_copy = bli_dallocm( m, n ); // Match the strides of B_copy to that of B. if ( bli_is_col_storage( b_rs, b_cs ) ) { b_copy_rs = 1; b_copy_cs = m; } else // if ( bli_is_row_storage( b_rs, b_cs ) ) { b_copy_rs = n; b_copy_cs = 1; } // Copy the contents of B to B_copy. bli_dcopymt( BLIS_NO_TRANSPOSE, m, n, b, b_rs, b_cs, b_copy, b_copy_rs, b_copy_cs ); // Perform the operation on B_copy. bli_dtrmm( side, uplo, trans, diag, m, n, alpha, a, a_rs, a_cs, b_copy, b_copy_rs, b_copy_cs ); // Scale C by beta. bli_dscalm( BLIS_NO_CONJUGATE, m, n, beta, c, c_rs, c_cs ); // Add B_copy into C. bli_daxpymt( BLIS_NO_TRANSPOSE, m, n, &one, b_copy, b_copy_rs, b_copy_cs, c, c_rs, c_cs ); // Free the copy of B. bli_dfree( b_copy ); // Free any temporary contiguous matrices, copying the result back to // the original matrix. bli_dfree_contigm( a_save, a_rs_save, a_cs_save, &a, &a_rs, &a_cs ); bli_dfree_contigm( b_save, b_rs_save, b_cs_save, &b, &b_rs, &b_cs ); bli_dfree_saved_contigm( m_save, n_save, c_save, c_rs_save, c_cs_save, &c, &c_rs, &c_cs ); }
void bli_strmmsx | ( | side_t | side, |
uplo_t | uplo, | ||
trans_t | trans, | ||
diag_t | diag, | ||
int | m, | ||
int | n, | ||
float * | alpha, | ||
float * | a, | ||
int | a_rs, | ||
int | a_cs, | ||
float * | b, | ||
int | b_rs, | ||
int | b_cs, | ||
float * | beta, | ||
float * | c, | ||
int | c_rs, | ||
int | c_cs | ||
) |
References bli_is_col_storage(), bli_s1(), bli_sallocm(), bli_saxpymt(), bli_scopymt(), bli_screate_contigm(), bli_screate_contigmr(), bli_set_dim_with_side(), bli_sfree(), bli_sfree_contigm(), bli_sfree_saved_contigm(), bli_sscalm(), bli_strmm(), bli_zero_dim2(), BLIS_NO_CONJUGATE, and BLIS_NO_TRANSPOSE.
Referenced by FLA_Trmmsx_external().
{ int m_save = m; int n_save = n; float* a_save = a; float* b_save = b; float* c_save = c; int a_rs_save = a_rs; int a_cs_save = a_cs; int b_rs_save = b_rs; int b_cs_save = b_cs; int c_rs_save = c_rs; int c_cs_save = c_cs; float one = bli_s1(); float* b_copy; int dim_a; int b_copy_rs, b_copy_cs; // Return early if possible. if ( bli_zero_dim2( m, n ) ) return; // If necessary, allocate, initialize, and use a temporary contiguous // copy of each matrix rather than the original matrices. bli_set_dim_with_side( side, m, n, &dim_a ); bli_screate_contigmr( uplo, dim_a, dim_a, a_save, a_rs_save, a_cs_save, &a, &a_rs, &a_cs ); bli_screate_contigm( m, n, b_save, b_rs_save, b_cs_save, &b, &b_rs, &b_cs ); bli_screate_contigm( m, n, c_save, c_rs_save, c_cs_save, &c, &c_rs, &c_cs ); // Create a copy of B to use in the computation so the original matrix is // left untouched. b_copy = bli_sallocm( m, n ); // Match the strides of B_copy to that of B. if ( bli_is_col_storage( b_rs, b_cs ) ) { b_copy_rs = 1; b_copy_cs = m; } else // if ( bli_is_row_storage( b_rs, b_cs ) ) { b_copy_rs = n; b_copy_cs = 1; } // Copy the contents of B to B_copy. bli_scopymt( BLIS_NO_TRANSPOSE, m, n, b, b_rs, b_cs, b_copy, b_copy_rs, b_copy_cs ); // Perform the operation on B_copy. bli_strmm( side, uplo, trans, diag, m, n, alpha, a, a_rs, a_cs, b_copy, b_copy_rs, b_copy_cs ); // Scale C by beta. bli_sscalm( BLIS_NO_CONJUGATE, m, n, beta, c, c_rs, c_cs ); // Add B_copy into C. bli_saxpymt( BLIS_NO_TRANSPOSE, m, n, &one, b_copy, b_copy_rs, b_copy_cs, c, c_rs, c_cs ); // Free the copy of B. bli_sfree( b_copy ); // Free any temporary contiguous matrices, copying the result back to // the original matrix. bli_sfree_contigm( a_save, a_rs_save, a_cs_save, &a, &a_rs, &a_cs ); bli_sfree_contigm( b_save, b_rs_save, b_cs_save, &b, &b_rs, &b_cs ); bli_sfree_saved_contigm( m_save, n_save, c_save, c_rs_save, c_cs_save, &c, &c_rs, &c_cs ); }
void bli_ztrmmsx | ( | side_t | side, |
uplo_t | uplo, | ||
trans_t | trans, | ||
diag_t | diag, | ||
int | m, | ||
int | n, | ||
dcomplex * | alpha, | ||
dcomplex * | a, | ||
int | a_rs, | ||
int | a_cs, | ||
dcomplex * | b, | ||
int | b_rs, | ||
int | b_cs, | ||
dcomplex * | beta, | ||
dcomplex * | c, | ||
int | c_rs, | ||
int | c_cs | ||
) |
References bli_is_col_storage(), bli_set_dim_with_side(), bli_z1(), bli_zallocm(), bli_zaxpymt(), bli_zcopymt(), bli_zcreate_contigm(), bli_zcreate_contigmr(), bli_zero_dim2(), bli_zfree(), bli_zfree_contigm(), bli_zfree_saved_contigm(), bli_zscalm(), bli_ztrmm(), BLIS_NO_CONJUGATE, and BLIS_NO_TRANSPOSE.
Referenced by FLA_Trmmsx_external().
{ int m_save = m; int n_save = n; dcomplex* a_save = a; dcomplex* b_save = b; dcomplex* c_save = c; int a_rs_save = a_rs; int a_cs_save = a_cs; int b_rs_save = b_rs; int b_cs_save = b_cs; int c_rs_save = c_rs; int c_cs_save = c_cs; dcomplex one = bli_z1(); dcomplex* b_copy; int dim_a; int b_copy_rs, b_copy_cs; // Return early if possible. if ( bli_zero_dim2( m, n ) ) return; // If necessary, allocate, initialize, and use a temporary contiguous // copy of each matrix rather than the original matrices. bli_set_dim_with_side( side, m, n, &dim_a ); bli_zcreate_contigmr( uplo, dim_a, dim_a, a_save, a_rs_save, a_cs_save, &a, &a_rs, &a_cs ); bli_zcreate_contigm( m, n, b_save, b_rs_save, b_cs_save, &b, &b_rs, &b_cs ); bli_zcreate_contigm( m, n, c_save, c_rs_save, c_cs_save, &c, &c_rs, &c_cs ); // Create a copy of B to use in the computation so the original matrix is // left untouched. b_copy = bli_zallocm( m, n ); // Match the strides of B_copy to that of B. if ( bli_is_col_storage( b_rs, b_cs ) ) { b_copy_rs = 1; b_copy_cs = m; } else // if ( bli_is_row_storage( b_rs, b_cs ) ) { b_copy_rs = n; b_copy_cs = 1; } // Copy the contents of B to B_copy. bli_zcopymt( BLIS_NO_TRANSPOSE, m, n, b, b_rs, b_cs, b_copy, b_copy_rs, b_copy_cs ); // Perform the operation on B_copy. bli_ztrmm( side, uplo, trans, diag, m, n, alpha, a, a_rs, a_cs, b_copy, b_copy_rs, b_copy_cs ); // Scale C by beta. bli_zscalm( BLIS_NO_CONJUGATE, m, n, beta, c, c_rs, c_cs ); // Add B_copy into C. bli_zaxpymt( BLIS_NO_TRANSPOSE, m, n, &one, b_copy, b_copy_rs, b_copy_cs, c, c_rs, c_cs ); // Free the copy of B. bli_zfree( b_copy ); // Free any temporary contiguous matrices, copying the result back to // the original matrix. bli_zfree_contigm( a_save, a_rs_save, a_cs_save, &a, &a_rs, &a_cs ); bli_zfree_contigm( b_save, b_rs_save, b_cs_save, &b, &b_rs, &b_cs ); bli_zfree_saved_contigm( m_save, n_save, c_save, c_rs_save, c_cs_save, &c, &c_rs, &c_cs ); }