25 template <
typename Float,
int nDim,
int nColor,
int nParity,
bool dagger,
bool xpay, KernelType kernel_type,
typename Arg>
27 static constexpr
const char *
kernel =
"quda::wilsonGPU";
28 template <
typename Dslash>
31 dslash.
launch(wilsonGPU<Float, nDim, nColor, nParity, dagger, xpay, kernel_type, Arg>, tp, arg, stream);
35 template <
typename Float,
int nDim,
int nColor,
typename Arg>
class Wilson :
public Dslash<Float>
44 Dslash<Float>(arg, out, in,
"kernels/dslash_wilson.cuh"),
65 template <
typename Float,
int nColor, QudaReconstructType recon>
struct WilsonApply {
70 constexpr
int nDim = 4;
75 wilson, const_cast<cudaColorSpinorField *>(static_cast<const cudaColorSpinorField *>(&in)), in.
VolumeCB(),
89 #ifdef GPU_WILSON_DIRAC 90 if (in.
V() == out.
V())
errorQuda(
"Aliasing pointers");
100 instantiate<WilsonApply, WilsonReconstruct>(
out,
in, U, a, x,
parity,
dagger, comm_override, profile);
102 errorQuda(
"Wilson dslash has not been built");
103 #endif // GPU_WILSON_DIRAC void launch(T *f, const TuneParam &tp, Arg &arg, const cudaStream_t &stream)
void apply(const cudaStream_t &stream)
QudaVerbosity getVerbosity()
#define checkPrecision(...)
static void launch(Dslash &dslash, TuneParam &tp, Arg &arg, const cudaStream_t &stream)
const char * VolString() const
void apply(const cudaStream_t &stream)
Parameter structure for driving the Wilson operator.
WilsonApply(ColorSpinorField &out, const ColorSpinorField &in, const GaugeField &U, double a, const ColorSpinorField &x, int parity, bool dagger, const int *comm_override, TimeProfile &profile)
__device__ __host__ void wilson(Arg &arg, int idx, int s, int parity)
const ColorSpinorField & in
const int * GhostFaceCB() const
TuneParam & tuneLaunch(Tunable &tunable, QudaTune enabled, QudaVerbosity verbosity)
static constexpr const char * kernel
#define checkLocation(...)
Wilson(Arg &arg, const ColorSpinorField &out, const ColorSpinorField &in)
cpuColorSpinorField * out
void ApplyWilson(ColorSpinorField &out, const ColorSpinorField &in, const GaugeField &U, double kappa, const ColorSpinorField &x, int parity, bool dagger, const int *comm_override, TimeProfile &profile)
Driver for applying the Wilson stencil.
__host__ __device__ ValueType arg(const complex< ValueType > &z)
Returns the phase angle of z.
QudaTune getTuning()
Query whether autotuning is enabled or not. Default is enabled but can be overridden by setting QUDA_...
This is a helper class that is used to instantiate the correct templated kernel for the dslash...
QudaFieldOrder FieldOrder() const