d1/dd9/cuda_2conv1__kernel_8h_source.html

#ifndef __MATH_CONV1_KERNEL_H__

#define __MATH_CONV1_KERNEL_H__

/*

 Copyright (c) 2021-2023, The Neko Authors

 All rights reserved.


 Redistribution and use in source and binary forms, with or without

 modification, are permitted provided that the following conditions

 are met:


   * Redistributions of source code must retain the above copyright

     notice, this list of conditions and the following disclaimer.


   * Redistributions in binary form must reproduce the above

     copyright notice, this list of conditions and the following

     disclaimer in the documentation and/or other materials provided

     with the distribution.


   * Neither the name of the authors nor the names of its

     contributors may be used to endorse or promote products derived

     from this software without specific prior written permission.


 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS

 "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT

 LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS

 FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE

 COPYRIGHT OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,

 INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,

 BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;

 LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER

 CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT

 LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN

 ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE

 POSSIBILITY OF SUCH DAMAGE.

*/


template< typename T, const int LX, const int CHUNKS >


__global__ void conv1_kernel_1d(T * __restrict__ du,

                                const T * __restrict__ u,

                                const T * __restrict__ vx,

                                const T * __restrict__ vy,

                                const T * __restrict__ vz,

                                const T * __restrict__ dx,

                                const T * __restrict__ dy,

                                const T * __restrict__ dz,

                                const T * __restrict__ drdx,

                                const T * __restrict__ dsdx,

                                const T * __restrict__ dtdx,

                                const T * __restrict__ drdy,

                                const T * __restrict__ dsdy,

                                const T * __restrict__ dtdy,

                                const T * __restrict__ drdz,

                                const T * __restrict__ dsdz,

                                const T * __restrict__ dtdz,

                                const T * __restrict__ jacinv) {


  __shared__ T shu[LX * LX * LX];


  __shared__ T shvx[LX * LX * LX];

  __shared__ T shvy[LX * LX * LX];

  __shared__ T shvz[LX * LX * LX];


  __shared__ T shdx[LX * LX];

  __shared__ T shdy[LX * LX];

  __shared__ T shdz[LX * LX];


  __shared__ T shjacinv[LX * LX * LX];


  const int e = blockIdx.x;

  const int iii = threadIdx.x;

  const int nchunks = (LX * LX * LX - 1) / CHUNKS + 1;

  const int ele = e*LX*LX*LX;


  if (iii < (LX * LX)) {

    shdx[iii] = dx[iii];

    shdy[iii] = dy[iii];

    shdz[iii] = dz[iii];

  }


  int l = iii;

  while(l < (LX * LX * LX)) {

    shu[l] = u[l + ele];


    shvx[l] = vx[l + ele];

    shvy[l] = vy[l + ele];

    shvz[l] = vz[l + ele];


    shjacinv[l] = jacinv[l + ele];


    l = l + CHUNKS;

  }


  __syncthreads();


  for (int n = 0; n < nchunks; n++) {

    const int ijk = iii + n * CHUNKS;

    const int jk = ijk / LX;

    const int i = ijk - jk * LX;

    const int k = jk / LX;

    const int j = jk - k * LX;

    if ( i < LX && j < LX && k < LX) {

      T rtmp = 0.0;

      T stmp = 0.0;

      T ttmp = 0.0;

      for (int l = 0; l < LX; l++) {

    rtmp += shdx[i + l * LX] * shu[l + j * LX + k * LX * LX];

    stmp += shdy[j + l * LX] * shu[i + l * LX + k * LX * LX];

    ttmp += shdz[k + l * LX] * shu[i + j * LX + l * LX * LX];

      }


      du[ijk + e * LX * LX * LX] = shjacinv[ijk] *

    (shvx[ijk] * (drdx[ijk + ele] * rtmp

              + dsdx[ijk + ele] * stmp

              + dtdx[ijk + ele] * ttmp)

     + shvy[ijk] * (drdy[ijk + ele] * rtmp

            + dsdy[ijk + ele] * stmp

            + dtdy[ijk + ele] * ttmp)

     + shvz[ijk] * (drdz[ijk + ele] * rtmp

            + dsdz[ijk + ele] * stmp

            + dtdz[ijk + ele] * ttmp));

    }

  }

}


template< typename T, const int LX >

__global__ void __launch_bounds__(LX*LX,3)

  conv1_kernel_kstep(T * __restrict__ du,

                     const T * __restrict__ u,

                     const T * __restrict__ vx,

                     const T * __restrict__ vy,

                     const T * __restrict__ vz,

                     const T * __restrict__ dx,

                     const T * __restrict__ dy,

                     const T * __restrict__ dz,

                     const T * __restrict__ drdx,

                     const T * __restrict__ dsdx,

                     const T * __restrict__ dtdx,

                     const T * __restrict__ drdy,

                     const T * __restrict__ dsdy,

                     const T * __restrict__ dtdy,

                     const T * __restrict__ drdz,

                     const T * __restrict__ dsdz,

                     const T * __restrict__ dtdz,

                     const T * __restrict__ jacinv) {


  __shared__ T shu[LX * LX];


  __shared__ T shdx[LX*LX];

  __shared__ T shdy[LX*LX];

  __shared__ T shdz[LX*LX];


  const int e = blockIdx.x;

  const int j = threadIdx.y;

  const int i = threadIdx.x;

  const int ij = i + j * LX;

  const int ele = e*LX*LX*LX;


  shdx[ij] = dx[ij];

  shdy[ij] = dy[ij];

  shdz[ij] = dz[ij];


  T ru[LX];

  T rvx[LX];

  T rvy[LX];

  T rvz[LX];

  T rjacinv[LX];


#pragma unroll LX


  for (int k = 0; k < LX; ++k) {

    ru[k] = u[ij + k*LX*LX + ele];

    rvx[k] = vx[ij + k*LX*LX + ele];

    rvy[k] = vy[ij + k*LX*LX + ele];

    rvz[k] = vz[ij + k*LX*LX + ele];

    rjacinv[k] = jacinv[ij + k*LX*LX + ele];

  }


  __syncthreads();


#pragma unroll

  for (int k = 0; k < LX; ++k) {

    const int ijk = ij + k*LX*LX;

    T ttmp = 0.0;

    shu[ij] = ru[k];

    for (int l = 0; l < LX; l++) {

      ttmp += shdz[k+l*LX] * ru[l];

    }

    __syncthreads();


    T rtmp = 0.0;

    T stmp = 0.0;

#pragma unroll

    for (int l = 0; l < LX; l++) {

      rtmp += shdx[i+l*LX] * shu[l+j*LX];

      stmp += shdy[j+l*LX] * shu[i+l*LX];

    }


    du[ijk + ele] = rjacinv[k] *

    (rvx[k] * (drdx[ijk + ele] * rtmp

                   + dsdx[ijk + ele] * stmp

                   + dtdx[ijk + ele] * ttmp)

     + rvy[k] * (drdy[ijk + ele] * rtmp

                     + dsdy[ijk + ele] * stmp

                     + dtdy[ijk + ele] * ttmp)

     + rvz[k] * (drdz[ijk + ele] * rtmp

                     + dsdz[ijk + ele] * stmp

                     + dtdz[ijk + ele] * ttmp));

    __syncthreads();

  }

}


#endif // __MATH_CONV1_KERNEL_H__

shu
__shared__ T shu[LX *LX]
Definition ax_helm_full_kernel.h:69

dz
__global__ void const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ dz
Definition conv1_kernel.h:138

rvy
T rvy[LX]
Definition conv1_kernel.h:168

jacinv
__global__ void const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ jacinv
Definition conv1_kernel.h:148

dy
__global__ void const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ dy
Definition conv1_kernel.h:137

ru
T ru[LX]
Definition conv1_kernel.h:166

shdz
__shared__ T shdz[LX *LX]
Definition conv1_kernel.h:154

drdx
__global__ void const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ drdx
Definition conv1_kernel.h:139

i
const int i
Definition conv1_kernel.h:158

rvx
T rvx[LX]
Definition conv1_kernel.h:167

dx
__global__ void const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ dx
Definition conv1_kernel.h:136

ij
const int ij
Definition conv1_kernel.h:159

dtdx
__global__ void const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ dtdx
Definition conv1_kernel.h:141

rvz
T rvz[LX]
Definition conv1_kernel.h:169

shdy
__shared__ T shdy[LX *LX]
Definition conv1_kernel.h:153

rjacinv
T rjacinv[LX]
Definition conv1_kernel.h:170

dtdz
__global__ void const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ dtdz
Definition conv1_kernel.h:147

e
const int e
Definition conv1_kernel.h:156

vz
__global__ void const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ vz
Definition conv1_kernel.h:135

dsdx
__global__ void const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ dsdx
Definition conv1_kernel.h:140

dtdy
__global__ void const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ dtdy
Definition conv1_kernel.h:144

dsdz
__global__ void const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ dsdz
Definition conv1_kernel.h:146

vx
__global__ void const T *__restrict__ const T *__restrict__ vx
Definition conv1_kernel.h:133

conv1_kernel_1d
__global__ void conv1_kernel_1d(T *__restrict__ du, const T *__restrict__ u, const T *__restrict__ vx, const T *__restrict__ vy, const T *__restrict__ vz, const T *__restrict__ dx, const T *__restrict__ dy, const T *__restrict__ dz, const T *__restrict__ drdx, const T *__restrict__ dsdx, const T *__restrict__ dtdx, const T *__restrict__ drdy, const T *__restrict__ dsdy, const T *__restrict__ dtdy, const T *__restrict__ drdz, const T *__restrict__ dsdz, const T *__restrict__ dtdz, const T *__restrict__ jacinv)
Definition conv1_kernel.h:42

shdx
__shared__ T shdx[LX *LX]
Definition conv1_kernel.h:152

dsdy
__global__ void const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ dsdy
Definition conv1_kernel.h:143

__launch_bounds__
__global__ void __launch_bounds__(LX *LX, 3) conv1_kernel_kstep(T *__restrict__ du

vy
__global__ void const T *__restrict__ const T *__restrict__ const T *__restrict__ vy
Definition conv1_kernel.h:134

ele
const int ele
Definition conv1_kernel.h:160

u
__global__ void const T *__restrict__ u
Definition conv1_kernel.h:132

j
const int j
Definition conv1_kernel.h:157

drdy
__global__ void const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ drdy
Definition conv1_kernel.h:142

__syncthreads
__syncthreads()

drdz
__global__ void const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ drdz
Definition conv1_kernel.h:145

dirichlet_apply_scalar_kernel
__global__ void dirichlet_apply_scalar_kernel(const int *__restrict__ msk, T *__restrict__ x, const T g, const int m)
Definition dirichlet_kernel.h:42