FLASH_lapack_prototypes.h File Reference

(r)

Go to the source code of this file.

Functions
FLA_Error	FLASH_Chol (FLA_Uplo uplo, FLA_Obj A)
FLA_Error	FLASH_Chol_solve (FLA_Uplo uplo, FLA_Obj A, FLA_Obj B, FLA_Obj X)
FLA_Error	FLASH_LU_nopiv (FLA_Obj A)
FLA_Error	FLASH_LU_nopiv_solve (FLA_Obj A, FLA_Obj B, FLA_Obj X)
FLA_Error	FLASH_LU_piv (FLA_Obj A, FLA_Obj p)
FLA_Error	FLASH_LU_piv_solve (FLA_Obj A, FLA_Obj p, FLA_Obj B, FLA_Obj X)
FLA_Error	FLASH_LU_incpiv (FLA_Obj A, FLA_Obj p, FLA_Obj L)
FLA_Error	FLASH_FS_incpiv (FLA_Obj A, FLA_Obj p, FLA_Obj L, FLA_Obj b)
FLA_Error	FLASH_Trinv (FLA_Uplo uplo, FLA_Diag diag, FLA_Obj A)
FLA_Error	FLASH_Ttmm (FLA_Uplo uplo, FLA_Obj A)
FLA_Error	FLASH_SPDinv (FLA_Uplo uplo, FLA_Obj A)
FLA_Error	FLASH_Sylv (FLA_Trans transa, FLA_Trans transb, FLA_Obj isgn, FLA_Obj A, FLA_Obj B, FLA_Obj C, FLA_Obj scale)
FLA_Error	FLASH_Apply_Q_UT (FLA_Side side, FLA_Trans trans, FLA_Direct direct, FLA_Store storev, FLA_Obj A, FLA_Obj T, FLA_Obj W, FLA_Obj B)
FLA_Error	FLASH_Apply_Q2_UT (FLA_Side side, FLA_Trans trans, FLA_Direct direct, FLA_Store storev, FLA_Obj D, FLA_Obj T, FLA_Obj W, FLA_Obj C, FLA_Obj E)
FLA_Error	FLASH_QR2_UT (FLA_Obj B, FLA_Obj D, FLA_Obj T)
FLA_Error	FLASH_QR_UT (FLA_Obj A, FLA_Obj TW)
FLA_Error	FLASH_QR_UT_solve (FLA_Obj A, FLA_Obj T, FLA_Obj B, FLA_Obj X)
FLA_Error	FLASH_QR_UT_inc (FLA_Obj A, FLA_Obj TW)
FLA_Error	FLASH_QR_UT_inc_solve (FLA_Obj A, FLA_Obj TW, FLA_Obj B, FLA_Obj X)
FLA_Error	FLASH_Apply_Q_UT_inc (FLA_Side side, FLA_Trans trans, FLA_Direct direct, FLA_Store storev, FLA_Obj A, FLA_Obj TW, FLA_Obj W1, FLA_Obj B)
FLA_Error	FLASH_Apply_pivots (FLA_Side side, FLA_Trans trans, FLA_Obj p, FLA_Obj A)
FLA_Error	FLASH_Eig_gest (FLA_Inv inv, FLA_Uplo uplo, FLA_Obj A, FLA_Obj B)
FLA_Error	FLASH_LQ_UT_inv (FLA_Obj A, FLA_Obj TW)
FLA_Error	FLASH_LQ2_UT (FLA_Obj B, FLA_Obj C, FLA_Obj T)

Function Documentation

FLASH_Apply_pivots()

FLA_Error FLASH_Apply_pivots	(	FLA_Side	side,
		FLA_Trans	trans,
		FLA_Obj	p,
		FLA_Obj	A
	)

{
  FLA_Error r_val;
  FLA_Bool  enable_supermatrix;
 
  // Check parameters.
 
  // *** The current Apply_pivots algorithm implemented assumes that
  // the matrix has a hierarchical depth of 1. We check for that here, because
  // we anticipate that we'll use a more general algorithm in the future, and
  // we don't want to forget to remove the constraint. ***
  if ( FLASH_Obj_depth( A ) != 1 )
  {
    FLA_Print_message( "FLASH_Apply_pivots() currently only supports matrices of depth 1",
                       __FILE__, __LINE__ );
    FLA_Abort();
  }
 
  // Find the status of SuperMatrix.
  enable_supermatrix = FLASH_Queue_get_enabled();
 
  // Temporarily disable SuperMatrix.
  FLASH_Queue_disable();
 
  // Invoke FLA_Apply_pivots_internal() with large control tree.
  r_val = FLA_Apply_pivots_internal( side, trans, p, A, flash_appiv_cntl );
 
  // Restore SuperMatrix to its previous status.
  if ( enable_supermatrix )
     FLASH_Queue_enable();
 
  return r_val;
}

References FLA_Abort(), FLA_Apply_pivots_internal(), FLA_Print_message(), flash_appiv_cntl, FLASH_Obj_depth(), FLASH_Queue_disable(), FLASH_Queue_enable(), FLASH_Queue_get_enabled(), and i.

Referenced by FLASH_LU_piv_solve().

FLASH_Apply_Q2_UT()

FLA_Error FLASH_Apply_Q2_UT	(	FLA_Side	side,
		FLA_Trans	trans,
		FLA_Direct	direct,
		FLA_Store	storev,
		FLA_Obj	D,
		FLA_Obj	T,
		FLA_Obj	W,
		FLA_Obj	C,
		FLA_Obj	E
	)

{
  FLA_Error r_val;
 
  // Check parameters.
  if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
    FLA_Apply_Q2_UT_check( side, trans, direct, storev, D, T, W, C, E );
 
  // Begin a parallel region.
  FLASH_Queue_begin();
  
  // Invoke FLA_Apply_Q2_UT_internal() with the standard control tree.
  r_val = FLA_Apply_Q2_UT_internal( side, trans, direct, storev, D, T, W, C, E, flash_apq2ut_cntl );
 
  // End the parallel region.
  FLASH_Queue_end();
 
  return r_val;
}

References FLA_Apply_Q2_UT_check(), FLA_Apply_Q2_UT_internal(), FLA_Check_error_level(), flash_apq2ut_cntl, FLASH_Queue_begin(), FLASH_Queue_end(), and i.

FLASH_Apply_Q_UT()

FLA_Error FLASH_Apply_Q_UT	(	FLA_Side	side,
		FLA_Trans	trans,
		FLA_Direct	direct,
		FLA_Store	storev,
		FLA_Obj	A,
		FLA_Obj	T,
		FLA_Obj	W,
		FLA_Obj	B
	)

{
  FLA_Error r_val;
  dim_t     b_alg;
 
  // Check parameters.
  if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
    FLA_Apply_Q_UT_check( side, trans, direct, storev, A, T, W, B );
 
  // Inspect the length of TTL to get the blocksize used by the QR/LQ
  // factorization, which will be our inner blocksize for Apply_Q_UT.
  b_alg = FLASH_Obj_scalar_length_tl( T );
 
  // The traditional (non-incremental) Apply_Q_UT algorithm-by-blocks
  // requires that the algorithmic blocksize be equal to the storage
  // blocksize.
  if ( b_alg != FLASH_Obj_scalar_width_tl( T ) )
  {
    FLA_Print_message( "FLASH_Apply_Q_UT() requires that b_alg == b_store",
                       __FILE__, __LINE__ );
    FLA_Abort();
  }
 
  // Adjust the blocksize of the control tree node for the flat subproblem.
  if ( FLA_Cntl_blocksize( fla_apqut_cntl_leaf ) != NULL )
    FLA_Blocksize_set( FLA_Cntl_blocksize( fla_apqut_cntl_leaf ),
                       b_alg, b_alg, b_alg, b_alg );
 
  // Begin a parallel region.
  FLASH_Queue_begin();
  
  // Invoke FLA_Apply_Q_UT_internal() with the standard control tree.
  r_val = FLA_Apply_Q_UT_internal( side, trans, direct, storev, A, T, W, B,
                                   flash_apqut_cntl_blas );
 
  // End the parallel region.
  FLASH_Queue_end();
 
  return r_val;
}

References FLA_Abort(), FLA_Apply_Q_UT_check(), FLA_Apply_Q_UT_internal(), fla_apqut_cntl_leaf, FLA_Blocksize_set(), FLA_Check_error_level(), FLA_Print_message(), flash_apqut_cntl_blas, FLASH_Obj_scalar_length_tl(), FLASH_Obj_scalar_width_tl(), FLASH_Queue_begin(), FLASH_Queue_end(), and i.

Referenced by FLASH_LQ_UT_solve(), and FLASH_QR_UT_solve().

FLASH_Apply_Q_UT_inc()

FLA_Error FLASH_Apply_Q_UT_inc	(	FLA_Side	side,
		FLA_Trans	trans,
		FLA_Direct	direct,
		FLA_Store	storev,
		FLA_Obj	A,
		FLA_Obj	TW,
		FLA_Obj	W1,
		FLA_Obj	B
	)

{
  FLA_Error r_val;
 
  // Check parameters.
  if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
    FLA_Apply_Q_UT_inc_check( side, trans, direct, storev, A, TW, W1, B );
 
  // Begin a parallel region.
  FLASH_Queue_begin();
 
  // Invoke FLA_Apply_Q_UT_inc_internal() with the standard control tree.
  r_val = FLA_Apply_Q_UT_inc_internal( side, trans, direct, storev, A, TW, W1, B, flash_apqutinc_cntl );
 
  // End the parallel region.
  FLASH_Queue_end();
 
  return r_val;
}

References FLA_Apply_Q_UT_inc_check(), FLA_Apply_Q_UT_inc_internal(), FLA_Check_error_level(), flash_apqutinc_cntl, FLASH_Queue_begin(), FLASH_Queue_end(), and i.

Referenced by FLA_Apply_CAQ_UT_inc_apply_panels(), and FLASH_QR_UT_inc_solve().

FLASH_Chol()

FLA_Error FLASH_Chol	(	FLA_Uplo	uplo,
		FLA_Obj	A
	)

{
  FLA_Error r_val;
 
  // Check parameters.
  if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
    FLA_Chol_check( uplo, A );
 
  // Begin a parallel region.
  FLASH_Queue_begin();
  
  // Enqueue tasks via a SuperMatrix-aware control tree.
  r_val = FLA_Chol_internal( uplo, A, flash_chol_cntl );
  
  // End the parallel region.
  FLASH_Queue_end();
 
  return r_val;
}

References FLA_Check_error_level(), FLA_Chol_check(), FLA_Chol_internal(), flash_chol_cntl, FLASH_Queue_begin(), FLASH_Queue_end(), and i.

FLASH_Chol_solve()

FLA_Error FLASH_Chol_solve	(	FLA_Uplo	uplo,
		FLA_Obj	A,
		FLA_Obj	B,
		FLA_Obj	X
	)

{
  // Check parameters.
  if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
    FLA_Chol_solve_check( uplo, A, B, X );
 
  FLASH_Copy( B, X );
 
  if ( uplo == FLA_LOWER_TRIANGULAR )
  {
      FLASH_Trsm( FLA_LEFT, FLA_LOWER_TRIANGULAR, FLA_NO_TRANSPOSE,
                  FLA_NONUNIT_DIAG, FLA_ONE, A, X );
      FLASH_Trsm( FLA_LEFT, FLA_LOWER_TRIANGULAR, FLA_CONJ_TRANSPOSE,
                  FLA_NONUNIT_DIAG, FLA_ONE, A, X );
  }
  else // if ( uplo == FLA_UPPER_TRIANGULAR )
  {
      FLASH_Trsm( FLA_LEFT, FLA_UPPER_TRIANGULAR, FLA_CONJ_TRANSPOSE,
                  FLA_NONUNIT_DIAG, FLA_ONE, A, X );
      FLASH_Trsm( FLA_LEFT, FLA_UPPER_TRIANGULAR, FLA_NO_TRANSPOSE,
                  FLA_NONUNIT_DIAG, FLA_ONE, A, X );
  }
 
  return FLA_SUCCESS;
}

References FLA_Check_error_level(), FLA_Chol_solve_check(), FLA_ONE, FLASH_Copy(), FLASH_Trsm(), and i.

FLASH_Eig_gest()

FLA_Error FLASH_Eig_gest	(	FLA_Inv	inv,
		FLA_Uplo	uplo,
		FLA_Obj	A,
		FLA_Obj	B
	)

{
  FLA_Obj   Y;
  FLA_Error r_val;
 
  // Check parameters.
  if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
    FLA_Eig_gest_check( inv, uplo, A, B );
 
  // The temporary matrix object Y must exist when execution occurs, NOT just
  // when enqueuing occurs. So if the SuperMatrix stack depth is positive, then
  // it means the user has declared a "parallel region" in his code, and thus
  // execution won't occur until sometime after FLASH_Eig_gest() returns, at
  // which time Y will have been deallocated. Thus, we disallow this scenario
  // for now, until we can think of a more general solution.
  if ( FLASH_Queue_stack_depth() > 0 )
  {
     FLA_Print_message( "FLASH_Eig_gest() MUST be invoked with standalone parallelism, and may not be called from within a user-level parallel region",
                        __FILE__, __LINE__ );
     FLA_Abort();
  }
 
 
  FLASH_Obj_create_conf_to( FLA_NO_TRANSPOSE, A, &Y );
 
  // Begin a parallel region.
  FLASH_Queue_begin();
  
  // Enqueue tasks via a SuperMatrix-aware control tree.
  r_val = FLA_Eig_gest_internal( inv, uplo, A, Y, B, flash_eig_gest_cntl );
  
  // End the parallel region.
  FLASH_Queue_end();
 
  FLASH_Obj_free( &Y );
 
  return r_val;
}

References FLA_Abort(), FLA_Check_error_level(), FLA_Eig_gest_check(), FLA_Eig_gest_internal(), FLA_Print_message(), flash_eig_gest_cntl, FLASH_Obj_create_conf_to(), FLASH_Obj_free(), FLASH_Queue_begin(), FLASH_Queue_end(), FLASH_Queue_stack_depth(), and i.

FLASH_FS_incpiv()

FLA_Error FLASH_FS_incpiv	(	FLA_Obj	A,
		FLA_Obj	p,
		FLA_Obj	L,
		FLA_Obj	b
	)

{
  dim_t     nb_alg;
  FLA_Error r_val;
  FLA_Bool  enable_supermatrix;
 
  // Check parameters.
  if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
    FLA_FS_incpiv_check( A, p, L, b );
 
  // *** The current forward substitution algorithm implemented assumes that 
  // the matrix has a hierarchical depth of 1. We check for that here, because
  // we anticipate that we'll use a more general algorithm in the future, and 
  // we don't want to forget to remove the constraint. ***
  if ( FLASH_Obj_depth( A ) != 1 )
  {
    FLA_Print_message( "FLASH_FS_incpiv() currently only supports matrices of depth 1",
                       __FILE__, __LINE__ );
    FLA_Abort();
  }
  
  // Inspect the width of a the top-left element of L to get the algorithmic
  // blocksize we'll use throughout the LU_incpiv algorithm.
  nb_alg = FLASH_Obj_scalar_width_tl( L );
 
  // Find the status of SuperMatrix.
  enable_supermatrix = FLASH_Queue_get_enabled();
 
  // Temporarily disable SuperMatrix.
  FLASH_Queue_disable();
  
  // Execute tasks.
  r_val = FLASH_FS_incpiv_aux1( A, p, L, b, nb_alg );
 
  // Restore SuperMatrix to its previous status.
  if ( enable_supermatrix )
     FLASH_Queue_enable();
 
  return r_val;
}

References FLA_Abort(), FLA_Check_error_level(), FLA_FS_incpiv_check(), FLA_Print_message(), FLASH_FS_incpiv_aux1(), FLASH_Obj_depth(), FLASH_Obj_scalar_width_tl(), FLASH_Queue_disable(), FLASH_Queue_enable(), FLASH_Queue_get_enabled(), and i.

Referenced by FLASH_LU_incpiv_solve().

FLASH_LQ2_UT()

FLA_Error FLASH_LQ2_UT	(	FLA_Obj	B,
		FLA_Obj	C,
		FLA_Obj	T
	)

FLASH_LQ_UT_inv()

FLA_Error FLASH_LQ_UT_inv	(	FLA_Obj	A,
		FLA_Obj	TW
	)

FLASH_LU_incpiv()

FLA_Error FLASH_LU_incpiv	(	FLA_Obj	A,
		FLA_Obj	p,
		FLA_Obj	L
	)

{
   FLA_Error r_val;
 
   // Check parameters.
   if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
      FLA_LU_incpiv_check( A, p, L );
 
   // *** The current LU_incpiv algorithm implemented assumes that
   // the matrix has a hierarchical depth of 1. We check for that here, because
   // we anticipate that we'll use a more general algorithm in the future, and
   // we don't want to forget to remove the constraint. ***
   if ( FLASH_Obj_depth( A ) != 1 )
   {
      FLA_Print_message( "FLASH_LU_incpiv() currently only supports matrices of depth 1",
                         __FILE__, __LINE__ );
      FLA_Abort();
   }
 
   if ( FLASH_Queue_stack_depth() == 0 )
      r_val = FLASH_LU_incpiv_opt1( A, p, L );
   else
      r_val = FLASH_LU_incpiv_noopt( A, p, L );
 
   return r_val;
}

References FLA_Abort(), FLA_Check_error_level(), FLA_LU_incpiv_check(), FLA_Print_message(), FLASH_LU_incpiv_noopt(), FLASH_LU_incpiv_opt1(), FLASH_Obj_depth(), FLASH_Queue_stack_depth(), and i.

FLASH_LU_nopiv()

FLA_Error FLASH_LU_nopiv

(

FLA_Obj

A

)

{
  FLA_Error r_val;
 
  // Check parameters.
  if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
    FLA_LU_nopiv_check( A );
 
  // Begin a parallel region.
  FLASH_Queue_begin();
  
  // Enqueue tasks via a SuperMatrix-aware control tree.
  r_val = FLA_LU_nopiv_internal( A, flash_lu_nopiv_cntl );
  
  // End the parallel region.
  FLASH_Queue_end();
 
  // Check for singularity.
  if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
    r_val = FLASH_LU_find_zero_on_diagonal( A );
 
  return r_val;
}

References FLA_Check_error_level(), FLA_LU_nopiv_check(), FLA_LU_nopiv_internal(), FLASH_LU_find_zero_on_diagonal(), flash_lu_nopiv_cntl, FLASH_Queue_begin(), FLASH_Queue_end(), and i.

FLASH_LU_nopiv_solve()

FLA_Error FLASH_LU_nopiv_solve	(	FLA_Obj	A,
		FLA_Obj	B,
		FLA_Obj	X
	)

{
  // Check parameters.
  if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
    FLA_LU_nopiv_solve_check( A, B, X );
 
  FLASH_Copy( B, X );
 
  FLASH_Trsm( FLA_LEFT, FLA_LOWER_TRIANGULAR, FLA_NO_TRANSPOSE,
              FLA_UNIT_DIAG, FLA_ONE, A, X );
  FLASH_Trsm( FLA_LEFT, FLA_UPPER_TRIANGULAR, FLA_NO_TRANSPOSE,
              FLA_NONUNIT_DIAG, FLA_ONE, A, X );
 
  return FLA_SUCCESS;
}

References FLA_Check_error_level(), FLA_LU_nopiv_solve_check(), FLA_ONE, FLASH_Copy(), FLASH_Trsm(), and i.

FLASH_LU_piv()

FLA_Error FLASH_LU_piv	(	FLA_Obj	A,
		FLA_Obj	p
	)

{
  FLA_Error r_val = FLA_SUCCESS;
 
  // Check parameters.
  if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
    FLA_LU_piv_check( A, p );
 
  // *** The current LU_piv algorithm implemented assumes that
  // the matrix has a hierarchical depth of 1. We check for that here, because
  // we anticipate that we'll use a more general algorithm in the future, and
  // we don't want to forget to remove the constraint. ***
  if ( FLASH_Obj_depth( A ) != 1 )
  {
    FLA_Print_message( "FLASH_LU_piv() currently only supports matrices of depth 1",
                       __FILE__, __LINE__ );
    FLA_Abort();
  }
 
  // Begin a parallel region.
  FLASH_Queue_begin();
 
  // Invoke FLA_LU_piv_internal() with large control tree.
  FLA_LU_piv_internal( A, p, flash_lu_piv_cntl );
 
  // End the parallel region.
  FLASH_Queue_end();
 
  // Check for singularity.
  if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
    r_val = FLASH_LU_find_zero_on_diagonal( A );
 
  return r_val;
}

References FLA_Abort(), FLA_Check_error_level(), FLA_LU_piv_check(), FLA_LU_piv_internal(), FLA_Print_message(), FLASH_LU_find_zero_on_diagonal(), flash_lu_piv_cntl, FLASH_Obj_depth(), FLASH_Queue_begin(), FLASH_Queue_end(), and i.

FLASH_LU_piv_solve()

FLA_Error FLASH_LU_piv_solve	(	FLA_Obj	A,
		FLA_Obj	p,
		FLA_Obj	B,
		FLA_Obj	X
	)

{
  // Check parameters.
  if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
    FLA_LU_piv_solve_check( A, p, B, X );
 
  FLASH_Copy( B, X );
 
  FLASH_Apply_pivots( FLA_LEFT, FLA_NO_TRANSPOSE, p, X );
 
  FLASH_Trsm( FLA_LEFT, FLA_LOWER_TRIANGULAR, FLA_NO_TRANSPOSE,
              FLA_UNIT_DIAG, FLA_ONE, A, X );
  FLASH_Trsm( FLA_LEFT, FLA_UPPER_TRIANGULAR, FLA_NO_TRANSPOSE,
              FLA_NONUNIT_DIAG, FLA_ONE, A, X );
 
  return FLA_SUCCESS;
}

References FLA_Check_error_level(), FLA_LU_piv_solve_check(), FLA_ONE, FLASH_Apply_pivots(), FLASH_Copy(), FLASH_Trsm(), and i.

FLASH_QR2_UT()

FLA_Error FLASH_QR2_UT	(	FLA_Obj	B,
		FLA_Obj	D,
		FLA_Obj	T
	)

{
  FLA_Error r_val;
 
  // Check parameters.
  if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
    FLA_QR2_UT_check( B, D, T );
 
  // Begin a parallel region.
  FLASH_Queue_begin();
  
  // Invoke FLA_QR2_UT_internal() with the standard control tree.
  r_val = FLA_QR2_UT_internal( B, D, T, flash_qr2ut_cntl );
 
  // End the parallel region.
  FLASH_Queue_end();
 
  return r_val;
}

References FLA_Check_error_level(), FLA_QR2_UT_check(), FLA_QR2_UT_internal(), flash_qr2ut_cntl, FLASH_Queue_begin(), FLASH_Queue_end(), and i.

FLASH_QR_UT()

FLA_Error FLASH_QR_UT	(	FLA_Obj	A,
		FLA_Obj	TW
	)

{
  FLA_Error r_val;
  dim_t     b_alg, b_flash;
 
  // Check parameters.
  if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
    FLA_QR_UT_check( A, TW );
 
  // *** The current hierarchical QR_UT algorithm assumes that the matrix
  // has a hierarchical depth of 1. We check for that here, because we
  //  anticipate that we'll use a more general algorithm in the future, and
  // we don't want to forget to remove the constraint. ***
  if ( FLASH_Obj_depth( A ) != 1 )
  {
    FLA_Print_message( "FLASH_QR_UT() currently only supports matrices of depth 1",
                       __FILE__, __LINE__ );
    FLA_Abort();
  }
 
  // Inspect the length of TTL to get the blocksize used by the QR
  // factorization, which will be our inner blocksize for Apply_Q_UT.
  b_alg   = FLASH_Obj_scalar_length_tl( TW );
  b_flash = FLASH_Obj_scalar_width_tl( TW );
 
  // The traditional (non-incremental) QR_UT algorithm-by-blocks requires
  // that the algorithmic blocksize be equal to the storage blocksize.
  if ( b_alg != b_flash )
  {
    FLA_Print_message( "FLASH_QR_UT() requires that b_alg == b_store",
                       __FILE__, __LINE__ );
    FLA_Abort();
  }
 
  // The traditional (non-incremental) QR_UT algorithm-by-blocks requires
  // that min_dim(A) % b_flash == 0.
  if ( FLASH_Obj_scalar_min_dim( A ) % b_flash != 0 )
  {
    FLA_Print_message( "FLASH_QR_UT() requires that min_dim( A ) %% b_store == 0",
                       __FILE__, __LINE__ );
    FLA_Abort();
  }
 
  // Begin a parallel region.
  FLASH_Queue_begin();
 
  // Invoke FLA_QR_UT_internal() with hierarchical control tree.
  r_val = FLA_QR_UT_internal( A, TW, flash_qrut_cntl );
 
  // End the parallel region.
  FLASH_Queue_end();
 
  return r_val;
}

References FLA_Abort(), FLA_Check_error_level(), FLA_Print_message(), FLA_QR_UT_check(), FLA_QR_UT_internal(), FLASH_Obj_depth(), FLASH_Obj_scalar_length_tl(), FLASH_Obj_scalar_min_dim(), FLASH_Obj_scalar_width_tl(), flash_qrut_cntl, FLASH_Queue_begin(), FLASH_Queue_end(), and i.

FLASH_QR_UT_inc()

FLA_Error FLASH_QR_UT_inc	(	FLA_Obj	A,
		FLA_Obj	TW
	)

{
  FLA_Error r_val;
 
  if ( FLASH_Queue_stack_depth() == 0 )
    r_val = FLASH_QR_UT_inc_opt1( A, TW );
  else
    r_val = FLASH_QR_UT_inc_noopt( A, TW );
 
  return r_val;
}

References FLASH_QR_UT_inc_noopt(), FLASH_QR_UT_inc_opt1(), FLASH_Queue_stack_depth(), and i.

Referenced by FLA_CAQR_UT_inc_factorize_panels().

FLASH_QR_UT_inc_solve()

FLA_Error FLASH_QR_UT_inc_solve	(	FLA_Obj	A,
		FLA_Obj	TW,
		FLA_Obj	B,
		FLA_Obj	X
	)

{
  FLA_Obj W, Y;
  FLA_Obj AT, AB;
  FLA_Obj YT, YB;
 
  // Check parameters.
  if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
    FLA_QR_UT_inc_solve_check( A, TW, B, X );
 
  FLASH_Apply_Q_UT_inc_create_workspace( TW, B, &W );
 
  FLASH_Obj_create_copy_of( FLA_NO_TRANSPOSE, B, &Y );
 
  FLASH_Apply_Q_UT_inc( FLA_LEFT, FLA_CONJ_TRANSPOSE, FLA_FORWARD, FLA_COLUMNWISE,
                        A, TW, W, Y );
 
  // Create a temporary hierarchical view of only the top n-by-n part of A in
  // case m > n so that AT captures the upper triangular factor R. We do the
  // same for Y to ensure conformality.
  FLASH_Part_create_2x1( A,   &AT,    
                              &AB,    FLASH_Obj_scalar_width( A ), FLA_TOP );
  FLASH_Part_create_2x1( Y,   &YT,    
                              &YB,    FLASH_Obj_scalar_width( A ), FLA_TOP );
 
  FLASH_Trsm( FLA_LEFT, FLA_UPPER_TRIANGULAR, FLA_NO_TRANSPOSE, FLA_NONUNIT_DIAG,
              FLA_ONE, AT, YT );
 
  FLASH_Copy( YT, X );
 
  // Free the temporary hierarchical views.
  FLASH_Part_free_2x1( &AT,
                       &AB );
  FLASH_Part_free_2x1( &YT,
                       &YB );
 
  FLASH_Obj_free( &Y );
  FLASH_Obj_free( &W );
 
  return FLA_SUCCESS;
}

References FLA_Check_error_level(), FLA_ONE, FLA_QR_UT_inc_solve_check(), FLASH_Apply_Q_UT_inc(), FLASH_Apply_Q_UT_inc_create_workspace(), FLASH_Copy(), FLASH_Obj_create_copy_of(), FLASH_Obj_free(), FLASH_Obj_scalar_width(), FLASH_Part_create_2x1(), FLASH_Part_free_2x1(), FLASH_Trsm(), and i.

FLASH_QR_UT_solve()

FLA_Error FLASH_QR_UT_solve	(	FLA_Obj	A,
		FLA_Obj	T,
		FLA_Obj	B,
		FLA_Obj	X
	)

{
  FLA_Obj W, Y;
  FLA_Obj AT, AB;
  FLA_Obj YT, YB;
 
  // Check parameters.
  if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
    FLA_QR_UT_solve_check( A, TW, B, X );
 
  FLASH_Apply_Q_UT_create_workspace( TW, B, &W );
 
  FLASH_Obj_create_copy_of( FLA_NO_TRANSPOSE, B, &Y );
 
  FLASH_Apply_Q_UT( FLA_LEFT, FLA_CONJ_TRANSPOSE, FLA_FORWARD, FLA_COLUMNWISE,
                    A, TW, W, Y );
 
  FLA_Part_2x1( A,   &AT,
                     &AB,    FLA_Obj_width( A ), FLA_TOP );
  FLA_Part_2x1( Y,   &YT,
                     &YB,    FLA_Obj_width( A ), FLA_TOP );
 
  FLASH_Trsm( FLA_LEFT, FLA_UPPER_TRIANGULAR, FLA_NO_TRANSPOSE, FLA_NONUNIT_DIAG,
              FLA_ONE, AT, YT );
 
  FLASH_Copy( YT, X );
 
  FLASH_Obj_free( &Y );
  FLASH_Obj_free( &W );
 
  return FLA_SUCCESS;
}

References FLA_Check_error_level(), FLA_Obj_width(), FLA_ONE, FLA_Part_2x1(), FLA_QR_UT_solve_check(), FLASH_Apply_Q_UT(), FLASH_Apply_Q_UT_create_workspace(), FLASH_Copy(), FLASH_Obj_create_copy_of(), FLASH_Obj_free(), FLASH_Trsm(), and i.

FLASH_SPDinv()

FLA_Error FLASH_SPDinv	(	FLA_Uplo	uplo,
		FLA_Obj	A
	)

{
  FLA_Error r_val;
 
  // Check parameters.
  if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
    FLA_SPDinv_check( uplo, A );
 
  // Begin a parallel region.
  FLASH_Queue_begin();
  
  // Enqueue tasks via a SuperMatrix-aware control tree.
  r_val = FLA_SPDinv_internal( uplo, A, flash_spdinv_cntl );
  
  // End the parallel region.
  FLASH_Queue_end();
 
  return r_val;
}

References FLA_Check_error_level(), FLA_SPDinv_check(), FLA_SPDinv_internal(), FLASH_Queue_begin(), FLASH_Queue_end(), flash_spdinv_cntl, and i.

FLASH_Sylv()

FLA_Error FLASH_Sylv	(	FLA_Trans	transa,
		FLA_Trans	transb,
		FLA_Obj	isgn,
		FLA_Obj	A,
		FLA_Obj	B,
		FLA_Obj	C,
		FLA_Obj	scale
	)

{
  FLA_Error r_val;
 
  // Check parameters.
  if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
    FLA_Sylv_check( transa, transb, isgn, A, B, C, scale );
 
  // Begin a parallel region.
  FLASH_Queue_begin();
  
  // Enqueue tasks via a SuperMatrix-aware control tree.
  r_val = FLA_Sylv_internal( transa, transb, isgn, A, B, C, scale, flash_sylv_cntl );
  
  // End the parallel region.
  FLASH_Queue_end();
 
  return r_val;
}

References FLA_Check_error_level(), FLA_Sylv_check(), FLA_Sylv_internal(), FLASH_Queue_begin(), FLASH_Queue_end(), flash_sylv_cntl, and i.

FLASH_Trinv()

FLA_Error FLASH_Trinv	(	FLA_Uplo	uplo,
		FLA_Diag	diag,
		FLA_Obj	A
	)

{
  FLA_Error r_val;
 
  // Check parameters.
  if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
    FLA_Trinv_check( uplo, diag, A );
 
  // Begin a parallel region.
  FLASH_Queue_begin();
  
  // Enqueue tasks via a SuperMatrix-aware control tree.
  r_val = FLA_Trinv_internal( uplo, diag, A, flash_trinv_cntl );
  
  // End the parallel region.
  FLASH_Queue_end();
 
  return r_val;
}

References FLA_Check_error_level(), FLA_Trinv_check(), FLA_Trinv_internal(), FLASH_Queue_begin(), FLASH_Queue_end(), flash_trinv_cntl, and i.

FLASH_Ttmm()

FLA_Error FLASH_Ttmm	(	FLA_Uplo	uplo,
		FLA_Obj	A
	)

{
  FLA_Error r_val;
 
  // Check parameters.
  if ( FLA_Check_error_level() >= FLA_MIN_ERROR_CHECKING )
    FLA_Ttmm_check( uplo, A );
 
  // Begin a parallel region.
  FLASH_Queue_begin();
  
  // Enqueue tasks via a SuperMatrix-aware control tree.
  r_val = FLA_Ttmm_internal( uplo, A, flash_ttmm_cntl );
  
  // End the parallel region.
  FLASH_Queue_end();
 
  return r_val;
}

References FLA_Check_error_level(), FLA_Ttmm_check(), FLA_Ttmm_internal(), FLASH_Queue_begin(), FLASH_Queue_end(), flash_ttmm_cntl, and i.