Fortran interface functions. More...

#include <quda_constants.h>

Functions
void	init_quda_device_ (int *device)

void	init_quda_memory_ ()

void	init_quda_ (int *device)

void	end_quda_ (void)

void	comm_set_gridsize_ (int *grid)

void	new_quda_gauge_param_ (QudaGaugeParam *param)

void	new_quda_invert_param_ (QudaInvertParam *param)

void	load_gauge_quda_ (void h_gauge, QudaGaugeParam param)

void	free_gauge_quda_ (void)

void	free_sloppy_gauge_quda_ (void)

void	load_clover_quda_ (void h_clover, void h_clovinv, QudaInvertParam *inv_param)

void	free_clover_quda_ (void)

void	dslash_quda_ (void h_out, void h_in, QudaInvertParam inv_param, QudaParity parity)

void	clover_quda_ (void h_out, void h_in, QudaInvertParam inv_param, QudaParity parity, int *inverse)

void	mat_quda_ (void h_out, void h_in, QudaInvertParam *inv_param)

void	mat_dag_mat_quda_ (void h_out, void h_in, QudaInvertParam *inv_param)

void	invert_quda_ (void h_x, void h_b, QudaInvertParam *param)

void	invert_md_quda_ (void hp_x, void hp_b, QudaInvertParam *param)

void	invert_multi_shift_quda_ (void _hp_x[QUDA_MAX_MULTI_SHIFT], void _hp_b, QudaInvertParam *param)

void	update_gauge_field_quda_ (void gauge, void momentum, double dt, bool conj_mom, bool exact, QudaGaugeParam param)

void	compute_staggered_force_quda_ (void cudaMom, void qudaQuark, double *coeff)

int	compute_gauge_force_quda_ (void mom, void gauge, int input_path_buf, int path_length, double loop_coeff, int num_paths, int max_length, double dt, QudaGaugeParam *qudaGaugeParam)

void	apply_staggered_phase_quda_ ()

void	remove_staggered_phase_quda_ ()

void	set_kernel_pack_t_ (int *pack)

Detailed Description

Fortran interface functions.

The following are Fortran interface functions to QUDA that mirror the C-equivalents. This essentially just means making all calls by reference, using all the lower-case characters and adding a trailing underscore.

Definition in file quda_fortran.h.

Function Documentation

void apply_staggered_phase_quda_ ( )

Apply the staggered phase factors to the resident gauge field

Definition at line 5124 of file interface_quda.cpp.

void clover_quda_	(	void *	h_out,
		void *	h_in,
		QudaInvertParam *	inv_param,
		QudaParity *	parity,
		int *	inverse
	)

Apply the clover operator or its inverse.

Parameters

h_out	Result spinor field
h_in	Input spinor field
param	Contains all metadata regarding host and device storage
parity	The source and destination parity of the field
inverse	Whether to apply the inverse of the clover term

Definition at line 5066 of file interface_quda.cpp.

void comm_set_gridsize_ ( int * grid )

Setter method for the comm grid size to all us to reuse BQCD's MPI topology. This is considered a temporary hack that will be fixed when an interface for setting the logical topology is created (issue 31 on github).

BQCD wants a node mapping with x varying fastest.

Definition at line 5160 of file interface_quda.cpp.

int compute_gauge_force_quda_	(	void *	mom,
		void *	gauge,
		int *	input_path_buf,
		int *	path_length,
		double *	loop_coeff,
		int *	num_paths,
		int *	max_length,
		double *	dt,
		QudaGaugeParam *	qudaGaugeParam
	)

Compute the gauge force and update the mometum field

Parameters

mom	The momentum field to be updated
gauge	The gauge field from which we compute the force
input_path_buf[dim][num_paths][path_length]	(Fortran 3-d array)
path_length	One less that the number of links in a loop (e.g., 3 for a staple)
loop_coeff	Coefficients of the different loops in the Symanzik action
num_paths	How many contributions from path_length different "staples"
max_length	The maximum number of non-zero of links in any path in the action
dt	The integration step size (for MILC this is dt*beta/3)
param	The parameters of the external fields and the computation settings

Definition at line 5091 of file interface_quda.cpp.

void compute_staggered_force_quda_	(	void *	cudaMom,
		void *	qudaQuark,
		double *	coeff
	)

Definition at line 5119 of file interface_quda.cpp.

void dslash_quda_	(	void *	h_out,
		void *	h_in,
		QudaInvertParam *	inv_param,
		QudaParity *	parity
	)

Apply the Dslash operator (D_{eo} or D_{oe}).

Parameters

h_out	Result spinor field
h_in	Input spinor field
param	Contains all metadata regarding host and device storage
parity	The destination parity of the field

Definition at line 5064 of file interface_quda.cpp.

void end_quda_ ( void )

Finalize the library.

Definition at line 5057 of file interface_quda.cpp.

void free_clover_quda_ ( void )

Free QUDA's internal copy of the clover term and/or clover inverse.

Definition at line 5063 of file interface_quda.cpp.

void free_gauge_quda_ ( void )

Free QUDA's internal copy of the gauge field.

Definition at line 5059 of file interface_quda.cpp.

void free_sloppy_gauge_quda_ ( void )

Free QUDA's internal copy of the gauge field.

Definition at line 5060 of file interface_quda.cpp.

void init_quda_ ( int * device )

Initialize the library. Under the interface this just calls initQudaMemory and initQudaDevice.

Parameters

device CUDA device number to use. In a multi-GPU build, this parameter may be either set explicitly on a per-process basis or set to -1 to enable a default allocation of devices to processes.

Definition at line 5054 of file interface_quda.cpp.

void init_quda_device_ ( int * device )

Initialize the library. This is a low-level interface that is called by initQuda. Calling initQudaDevice requires that the user also call initQudaMemory before using QUDA.

Parameters

device CUDA device number to use. In a multi-GPU build, this parameter may either be set explicitly on a per-process basis or set to -1 to enable a default allocation of devices to processes.

Definition at line 5055 of file interface_quda.cpp.

void init_quda_memory_ ( )

Initialize the library persistant memory allocations (both host and device). This is a low-level interface that is called by initQuda. Calling initQudaMemory requires that the user has previously called initQudaDevice.

Definition at line 5056 of file interface_quda.cpp.

void invert_md_quda_	(	void *	hp_x,
		void *	hp_b,
		QudaInvertParam *	param
	)

Definition at line 5074 of file interface_quda.cpp.

void invert_multi_shift_quda_	(	void *	_hp_x[QUDA_MAX_MULTI_SHIFT],
		void *	_hp_b,
		QudaInvertParam *	param
	)

Solve for multiple shifts (e.g., masses).

Parameters

_hp_x	Array of solution spinor fields
_hp_b	Array of source spinor fields
param	Contains all metadata regarding host and device storage and solver parameters

void invert_quda_	(	void *	h_x,
		void *	h_b,
		QudaInvertParam *	param
	)

Perform the solve, according to the parameters set in param. It is assumed that the gauge field has already been loaded via loadGaugeQuda().

Parameters

h_x	Solution spinor field
h_b	Source spinor field
param	Contains all metadata regarding host and device storage and solver parameters

Definition at line 5072 of file interface_quda.cpp.

void load_clover_quda_	(	void *	h_clover,
		void *	h_clovinv,
		QudaInvertParam *	inv_param
	)

Load the clover term and/or the clover inverse from the host. Either h_clover or h_clovinv may be set to NULL.

Parameters

h_clover	Base pointer to host clover field
h_cloverinv	Base pointer to host clover inverse field
inv_param	Contains all metadata regarding host and device storage

Definition at line 5061 of file interface_quda.cpp.

void load_gauge_quda_	(	void *	h_gauge,
		QudaGaugeParam *	param
	)

Load the gauge field from the host.

Parameters

h_gauge	Base pointer to host gauge field (regardless of dimensionality)
param	Contains all metadata regarding host and device storage

Definition at line 5058 of file interface_quda.cpp.

void mat_dag_mat_quda_	(	void *	h_out,
		void *	h_in,
		QudaInvertParam *	inv_param
	)

Apply M^{}M, possibly even/odd preconditioned.

Parameters

h_out	Result spinor field
h_in	Input spinor field
param	Contains all metadata regarding host and device storage

Definition at line 5070 of file interface_quda.cpp.

void mat_quda_	(	void *	h_out,
		void *	h_in,
		QudaInvertParam *	inv_param
	)

Apply the full Dslash matrix, possibly even/odd preconditioned.

Parameters

h_out	Result spinor field
h_in	Input spinor field
param	Contains all metadata regarding host and device storage

Definition at line 5068 of file interface_quda.cpp.

void new_quda_gauge_param_ ( QudaGaugeParam * param )

Initializes the QudaGaugeParam with default entries.

Parameters

The	QudaGaugeParam to be initialized

Definition at line 5078 of file interface_quda.cpp.

void new_quda_invert_param_ ( QudaInvertParam * param )

Initializes the QudaInvertParam with default entries.

Parameters

The	QudaInvertParam to be initialized

Definition at line 5081 of file interface_quda.cpp.

void remove_staggered_phase_quda_ ( )

Remove the staggered phase factors to the resident gauge field

Definition at line 5134 of file interface_quda.cpp.

void set_kernel_pack_t_ ( int * pack )

Temporary function exposed for TIFR benchmarking

Exposed due to poor derived MPI datatype performance with GPUDirect RDMA

Definition at line 5170 of file interface_quda.cpp.

void update_gauge_field_quda_	(	void *	gauge,
		void *	momentum,
		double *	dt,
		bool *	conj_mom,
		bool *	exact,
		QudaGaugeParam *	param
	)

Evolve the gauge field by step size dt, using the momentum field I.e., Evalulate U(t+dt) = e(dt pi) U(t)

Parameters

gauge	The gauge field to be updated
momentum	The momentum field
dt	The integration step size step
conj_mom	Whether to conjugate the momentum matrix
exact	Whether to use an exact exponential or Taylor expand
param	The parameters of the external fields and the computation settings

Definition at line 5085 of file interface_quda.cpp.

Functions

Detailed Description

Function Documentation