QUDA
v1.1.0
A library for QCD on GPUs
|
#include <dirac_quda.h>
Public Member Functions | |
DiracCoarsePC (const DiracParam ¶m, bool gpu_setup=true) | |
DiracCoarsePC (const DiracCoarse &dirac, const DiracParam ¶m) | |
virtual | ~DiracCoarsePC () |
void | Dslash (ColorSpinorField &out, const ColorSpinorField &in, const QudaParity parity) const |
Apply DslashXpay out = (D * in) More... | |
void | DslashXpay (ColorSpinorField &out, const ColorSpinorField &in, const QudaParity parity, const ColorSpinorField &x, const double &k) const |
Apply DslashXpay out = (D * in + A * x) More... | |
void | M (ColorSpinorField &out, const ColorSpinorField &in) const |
Apply the full operator. More... | |
void | MdagM (ColorSpinorField &out, const ColorSpinorField &in) const |
Apply MdagM operator which may be optimized. More... | |
void | prepare (ColorSpinorField *&src, ColorSpinorField *&sol, ColorSpinorField &x, ColorSpinorField &b, const QudaSolutionType) const |
void | reconstruct (ColorSpinorField &x, const ColorSpinorField &b, const QudaSolutionType) const |
virtual QudaDiracType | getDiracType () const |
returns the Dirac type More... | |
void | createCoarseOp (GaugeField &Y, GaugeField &X, const Transfer &T, double kappa, double mass, double mu, double mu_factor=0.) const |
Create the coarse even-odd preconditioned coarse operator. Unlike the Wilson operator, the coarsening of the preconditioned coarse operator differs from that of the unpreconditioned coarse operator, so we need to specialize it. More... | |
virtual void | prefetch (QudaFieldLocation mem_space, qudaStream_t stream=0) const |
If managed memory and prefetch is enabled, prefetch all relevant memory fields (Xhat, Y) to the CPU or GPU as requested. More... | |
Public Member Functions inherited from quda::DiracCoarse | |
double | Mass () const |
accessor for Mass (in case of a factor of 2 for staggered) More... | |
double | Mu () const |
accessor for twist parameter – overrride can return better value More... | |
double | MuFactor () const |
accessor for mu factoo for MG/ – override can return a better value More... | |
DiracCoarse (const DiracParam ¶m, bool gpu_setup=true, bool mapped=false) | |
DiracCoarse (const DiracParam ¶m, cpuGaugeField *Y_h, cpuGaugeField *X_h, cpuGaugeField *Xinv_h, cpuGaugeField *Yhat_h, cudaGaugeField *Y_d=0, cudaGaugeField *X_d=0, cudaGaugeField *Xinv_d=0, cudaGaugeField *Yhat_d=0) | |
DiracCoarse (const DiracCoarse &dirac, const DiracParam ¶m) | |
virtual | ~DiracCoarse () |
virtual bool | isCoarse () const |
Whether the Dirac object is the DiracCoarse. More... | |
void | Clover (ColorSpinorField &out, const ColorSpinorField &in, const QudaParity parity) const |
Apply the coarse clover operator. More... | |
void | CloverInv (ColorSpinorField &out, const ColorSpinorField &in, const QudaParity parity) const |
Apply the inverse coarse clover operator. More... | |
virtual void | updateFields (cudaGaugeField *gauge_in, cudaGaugeField *fat_gauge_in, cudaGaugeField *long_gauge_in, cudaCloverField *clover_in) |
Update the internal gauge, fat gauge, long gauge, clover field pointer as appropriate. These are pointers as opposed to references to support passing in nullptr . More... | |
void | createPreconditionedCoarseOp (GaugeField &Yhat, GaugeField &Xinv, const GaugeField &Y, const GaugeField &X) |
Create the precondtioned coarse operator. More... | |
Public Member Functions inherited from quda::Dirac | |
Dirac (const DiracParam ¶m) | |
Dirac (const Dirac &dirac) | |
virtual | ~Dirac () |
Dirac & | operator= (const Dirac &dirac) |
void | setCommDim (const int commDim_[QUDA_MAX_DIM]) const |
Enable / disable communications for the Dirac operator. More... | |
virtual void | checkParitySpinor (const ColorSpinorField &, const ColorSpinorField &) const |
Check parity spinors are usable (check geometry ?) More... | |
virtual void | checkFullSpinor (const ColorSpinorField &, const ColorSpinorField &) const |
check full spinors are compatible (check geometry ?) More... | |
void | checkSpinorAlias (const ColorSpinorField &, const ColorSpinorField &) const |
check spinors do not alias More... | |
virtual bool | hasDslash () const |
Whether or not the operator has a single-parity Dslash. More... | |
virtual void | MdagMLocal (ColorSpinorField &out, const ColorSpinorField &in) const |
Apply the local MdagM operator: equivalent to applying zero Dirichlet boundary condition to MdagM on each rank. Depending on the number of stencil steps of the fermion type, this may require additional effort to include the terms that hop out of the boundary and then hop back. More... | |
virtual void | Dslash4 (ColorSpinorField &out, const ColorSpinorField &in, const QudaParity parity) const |
Apply the local MdagM operator: equivalent to applying zero Dirichlet boundary condition to MdagM on each rank. Depending on the number of stencil steps of the fermion type, this may require additional effort to include the terms that hop out of the boundary and then hop back. More... | |
void | Mdag (ColorSpinorField &out, const ColorSpinorField &in) const |
Apply Mdag (daggered operator of M. More... | |
void | MMdag (ColorSpinorField &out, const ColorSpinorField &in) const |
Apply Normal Operator. More... | |
virtual void | prepareSpecialMG (ColorSpinorField *&src, ColorSpinorField *&sol, ColorSpinorField &x, ColorSpinorField &b, const QudaSolutionType solType) const |
virtual void | reconstructSpecialMG (ColorSpinorField &x, const ColorSpinorField &b, const QudaSolutionType solType) const |
virtual bool | hasSpecialMG () const |
specifies whether or not there's a specialized prepare/reconstruct used before/after transfering to/from the coarse level in MG More... | |
void | setMass (double mass) |
double | Kappa () const |
accessor for Kappa (mass parameter) More... | |
unsigned long long | Flops () const |
returns and then zeroes flopcount More... | |
QudaMatPCType | getMatPCType () const |
returns preconditioning type More... | |
int | getStencilSteps () const |
I have no idea what this does. More... | |
void | Dagger (QudaDagType dag) const |
sets whether operator is daggered or not More... | |
void | flipDagger () const |
Flips value of daggered. More... | |
virtual bool | hermitian () const |
is operator hermitian More... | |
QudaPrecision | HaloPrecision () const |
void | setHaloPrecision (QudaPrecision halo_precision_) const |
Public Member Functions inherited from quda::Object | |
Object () | |
virtual | ~Object () |
Additional Inherited Members | |
Static Public Member Functions inherited from quda::Dirac | |
static Dirac * | create (const DiracParam ¶m) |
Creates a subclass from parameters. More... | |
Static Public Member Functions inherited from quda::Object | |
static void * | operator new (std::size_t size) |
static void | operator delete (void *p) |
static void * | operator new[] (std::size_t size) |
static void | operator delete[] (void *p) |
Protected Member Functions inherited from quda::DiracCoarse | |
void | initializeCoarse () |
Initialize the coarse gauge fields. Location is determined by gpu_setup variable. More... | |
void | initializeLazy (QudaFieldLocation location) const |
Create the CPU or GPU coarse gauge fields on demand (requires that the fields have been created in the other memory space) More... | |
void | createY (bool gpu=true, bool mapped=false) const |
Allocate the Y and X fields. More... | |
void | createYhat (bool gpu=true) const |
Allocate the Yhat and Xinv fields. More... | |
Protected Member Functions inherited from quda::Dirac | |
bool | newTmp (ColorSpinorField **, const ColorSpinorField &) const |
void | deleteTmp (ColorSpinorField **, const bool &reset) const |
Protected Attributes inherited from quda::DiracCoarse | |
double | mass |
double | mu |
double | mu_factor |
const Transfer * | transfer |
const Dirac * | dirac |
const bool | need_bidirectional |
const bool | use_mma |
cpuGaugeField * | Y_h |
cpuGaugeField * | X_h |
cpuGaugeField * | Xinv_h |
cpuGaugeField * | Yhat_h |
cudaGaugeField * | Y_d |
cudaGaugeField * | X_d |
cudaGaugeField * | Xinv_d |
cudaGaugeField * | Yhat_d |
bool | enable_gpu |
bool | enable_cpu |
const bool | gpu_setup |
bool | init_gpu |
bool | init_cpu |
const bool | mapped |
Protected Attributes inherited from quda::Dirac | |
cudaGaugeField * | gauge |
double | kappa |
double | mass |
int | laplace3D |
QudaMatPCType | matpcType |
QudaDagType | dagger |
unsigned long long | flops |
ColorSpinorField * | tmp1 |
ColorSpinorField * | tmp2 |
QudaDiracType | type |
QudaPrecision | halo_precision |
int | commDim [QUDA_MAX_DIM] |
TimeProfile | profile |
Even-odd preconditioned variant of coarse Dslash operator
Definition at line 1744 of file dirac_quda.h.
quda::DiracCoarsePC::DiracCoarsePC | ( | const DiracParam & | param, |
bool | gpu_setup = true |
||
) |
[in] | param | Parameters defining this operator |
[in] | gpu_setup | Whether to do the setup on GPU or CPU |
Definition at line 449 of file dirac_coarse.cpp.
quda::DiracCoarsePC::DiracCoarsePC | ( | const DiracCoarse & | dirac, |
const DiracParam & | param | ||
) |
[in] | dirac | Another operator instance to clone from (shallow copy) |
[in] | param | Parameters defining this operator |
Definition at line 454 of file dirac_coarse.cpp.
|
virtual |
Definition at line 459 of file dirac_coarse.cpp.
|
virtual |
Create the coarse even-odd preconditioned coarse operator. Unlike the Wilson operator, the coarsening of the preconditioned coarse operator differs from that of the unpreconditioned coarse operator, so we need to specialize it.
T[in] | Transfer operator defining the coarse grid |
Y[out] | Coarse link field |
X[out] | Coarse clover field |
kappa | Kappa parameter for the coarse operator |
mass | Mass parameter for the coarse operator, assumed to be zero |
mu | TM mu parameter for the coarse operator |
mu_factor | multiplicative factor for the mu parameter |
Reimplemented from quda::DiracCoarse.
Definition at line 647 of file dirac_coarse.cpp.
|
virtual |
Apply DslashXpay out = (D * in)
[out] | out | Output field |
[in] | in | Input field |
[paraity] | parity Parity which we are applying the operator to |
Reimplemented from quda::DiracCoarse.
Definition at line 461 of file dirac_coarse.cpp.
|
virtual |
Apply DslashXpay out = (D * in + A * x)
[out] | out | Output field |
[in] | in | Input field |
[paraity] | parity Parity which we are applying the operator to |
Reimplemented from quda::DiracCoarse.
Definition at line 475 of file dirac_coarse.cpp.
|
inlinevirtual |
returns the Dirac type
Reimplemented from quda::DiracCoarse.
Definition at line 1770 of file dirac_quda.h.
|
virtual |
Apply the full operator.
[out] | out | output vector, out = M * in |
[in] | in | input vector |
Reimplemented from quda::DiracCoarse.
Definition at line 486 of file dirac_coarse.cpp.
|
virtual |
Apply MdagM operator which may be optimized.
Reimplemented from quda::DiracCoarse.
Definition at line 522 of file dirac_coarse.cpp.
|
virtual |
If managed memory and prefetch is enabled, prefetch all relevant memory fields (Xhat, Y) to the CPU or GPU as requested.
[in] | mem_space | Memory space we are prefetching to |
[in] | stream | Which stream to run the prefetch in (default 0) |
Reimplemented from quda::DiracCoarse.
Definition at line 664 of file dirac_coarse.cpp.
|
virtual |
Reimplemented from quda::DiracCoarse.
Definition at line 530 of file dirac_coarse.cpp.
|
virtual |
Reimplemented from quda::DiracCoarse.
Definition at line 599 of file dirac_coarse.cpp.