Page MenuHomec4science
Files F94100544

gradient2_GPU.cu
No OneTemporary
Actions

File Metadata

Created
Tue, Dec 3, 22:06

gradient2_GPU.cu
View Options

/**
Lenstool-HPC: HPC based massmodeling software and Lens-map generation
Copyright (C) 2017 Christoph Schaefer, EPFL (christophernstrerne.schaefer@epfl.ch), Gilles Fourestey (gilles.fourestey@epfl.ch)
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program. If not, see <http://www.gnu.org/licenses/>.
@brief: Function for second order computation on GPUs
*/
#include <fstream>
#include "grid_gradient2_GPU.cuh"
#include "gradient2_GPU.cuh"
#include "structure_hpc.hpp"
#define BLOCK_SIZE_X 8
#define BLOCK_SIZE_Y 16
//#define ROT
#define _SHARED_MEM
#ifdef _SHARED_MEM
#define SHARED __shared__
#warning "shared memory"
extern __shared__ type_t shared[];
#else
#define SHARED
#endif
#define Nx 1
#define Ny 0
extern "C" {
double myseconds();
}
__device__ matrix mdci05_gpu(type_t x, type_t y, type_t eps, type_t rc, type_t b0)
{
matrix res;
type_t ci, sqe, cx1, cxro, cyro, wrem;
type_t didyre, didyim, didxre;// didxim;
type_t cx1inv, den1, num2, den2;
sqe = sqrt(eps);
cx1 = (1. - eps) / (1. + eps);
cx1inv = 1. / cx1;
cxro = (1. + eps) * (1. + eps); /* rem^2=x^2/(1+e^2) + y^2/(1-e^2) Eq 2.3.6*/
cyro = (1. - eps) * (1. - eps);
ci = 0.5 * (1. - eps * eps) / sqe;
wrem = sqrt(rc * rc + x * x / cxro + y * y / cyro); /*wrem^2=w^2+rem^2 with w core radius*/
den1 = 2.*sqe * wrem - y * cx1inv;
den1 = cx1 * cx1 * x * x + den1 * den1;
num2 = 2.*rc * sqe - y;
den2 = x * x + num2 * num2;
didxre = ci * ( cx1 * (2.*sqe * x * x / cxro / wrem - 2.*sqe * wrem + y * cx1inv) / den1 + num2 / den2 );
didyre = ci * ( (2 * sqe * x * y * cx1 / cyro / wrem - x) / den1 + x / den2 );
didyim = ci * ( (2 * sqe * wrem * cx1inv - y * cx1inv * cx1inv - 4 * eps * y / cyro +
2 * sqe * y * y / cyro / wrem * cx1inv) / den1 - num2 / den2 );
//if(blockIdx.x*blockDim.x + threadIdx.x == 0 && blockIdx.y*blockDim.y + threadIdx.y == 0) printf("didxre %f didyre %f didyim %f ",didxre,didyre,didyim);
res.a = b0 * didxre;
res.b = res.d = b0 * didyre; //(didyre+didxim)/2.;
res.c = b0 * didyim;
return(res);
}
__device__ void printmat_gpu(matrix A){
printf("A: %lf, B: %lf, C:%lf, D:%lf \n",A.a,A.b,A.c,A.d);
}
__device__ matrix module_potentialDerivatives_totalGradient2_81_SOA_GPU(const struct point *pImage, const struct Potential_SOA *lens, int shalos, int nhalos){
//asm volatile("# module_potentialDerivatives_totalGradient_81_SOA begins");
#if 0
int col = blockIdx.x*blockDim.x + threadIdx.x;
int row = blockIdx.y*blockDim.y + threadIdx.y;
if(row == 0 && col == 0) printf("# module_potentialDerivatives_totalGradient_81_SOA begins\n");
#endif
//std::cout << "# module_potentialDerivatives_totalGradient_81_SOA begins" << std::endl;
// 6 DP loads, i.e. 48 Bytes: position_x, position_y, ellipticity_angle, ellipticity_potential, rcore, b0
//
type_t t05,RR;
struct matrix grad2, clump, clumpcore, clumpcut;
grad2.a = 0;
grad2.b = 0;
grad2.c = 0;
grad2.d = 0;
#if 1
for(int i = shalos; i < shalos + nhalos; i++)
{
struct point true_coord;
//True coord
true_coord.x = pImage->x - lens->position_x[i];
true_coord.y = pImage->y - lens->position_y[i];
//
if(lens->ellipticity_potential[i] > 0.0001){
//std::cerr << "start" << std::endl;
//if(blockIdx.x*blockDim.x + threadIdx.x == 0 && blockIdx.y*blockDim.y + threadIdx.y == 0) printf("start\n");
//Rotation
type_t cose = lens->anglecos[i];
type_t sine = lens->anglesin[i];
//
type_t x = true_coord.x*cose + true_coord.y*sine;
type_t y = true_coord.y*cose - true_coord.x*sine;
// 81 comput
t05 = lens->rcut[i] / (lens->rcut[i] - lens->rcore[i]);
//if(blockIdx.x*blockDim.x + threadIdx.x == 0 && blockIdx.y*blockDim.y + threadIdx.y == 0)printf("t05 %f rcut: %f, rcore: %f, b0:%f \n",t05, lens->rcut[i],lens->rcore[i], lens->b0[i]);
clumpcore = mdci05_gpu(x, y, lens->ellipticity_potential[i], lens->rcore[i], lens->b0[i]);
////////////////////////////
#if 0
//matrix res;
type_t rc = lens->rcore[i];
type_t eps =lens->ellipticity_potential[i];
type_t ci, sqe, cx1, cxro, cyro, wrem;
type_t didyre, didyim, didxre;// didxim;
type_t cx1inv, den1, num2, den2;
sqe = sqrt( lens->ellipticity_potential[i]);
cx1 = (1. - eps) / (1. + eps);
cx1inv = 1. / cx1;
cxro = (1. + eps) * (1. + eps); /* rem^2=x^2/(1+e^2) + y^2/(1-e^2) Eq 2.3.6*/
cyro = (1. - eps) * (1. - eps);
ci = 0.5 * (1. - eps * eps) / sqe;
wrem = sqrt(rc * rc + x * x / cxro + y * y / cyro); /*wrem^2=w^2+rem^2 with w core radius*/
den1 = 2.*sqe * wrem - y * cx1inv;
den1 = cx1 * cx1 * x * x + den1 * den1;
num2 = 2.*rc * sqe - y;
den2 = x * x + num2 * num2;
didxre = ci * ( cx1 * (2.*sqe * x * x / cxro / wrem - 2.*sqe * wrem + y * cx1inv) / den1 + num2 / den2 );
if(blockIdx.x*blockDim.x + threadIdx.x == 0 && blockIdx.y*blockDim.y + threadIdx.y == 0) printf("cyro %f ",cyro);
if(cyro == 0) printf("I %d fucked up!",threadIdx.x);
//didyre = ci * ( (2 * sqe * x * y * cx1 / wrem - x) + x );
didyre = ci * ( (2 * sqe * x * y * cx1 / wrem - x) / den1 + x / den2 );
didyre = didyre /cyro;
//didyre = ci * ( (2 * sqe * x * y * cx1 / cyro / wrem - x) / den1 + x / den2 );
didyim = ci * ( (2 * sqe * wrem * cx1inv - y * cx1inv * cx1inv - 4 * eps * y / cyro +
2 * sqe * y * y / cyro / wrem * cx1inv) / den1 - num2 / den2 );
//if(blockIdx.x*blockDim.x + threadIdx.x == 0 && blockIdx.y*blockDim.y + threadIdx.y == 0) printf("didxre %f didyre %f didyim %f ",didxre,didyre,didyim);
clumpcore.a = lens->b0[i] * didxre;
clumpcore.b = clumpcore.d = lens->b0[i] * didyre; //(didyre+didxim)/2.;
clumpcore.c = lens->b0[i] * didyim;
#endif
////////////////////////////////////////////////
clumpcut = mdci05_gpu(x, y, lens->ellipticity_potential[i], lens->rcut[i], lens->b0[i]);
/////////////////////////////////////
#if 0
rc = lens->rcut[i];
sqe = sqrt( lens->ellipticity_potential[i]);
cx1 = (1. - eps) / (1. + eps);
cx1inv = 1. / cx1;
cxro = (1. + eps) * (1. + eps); /* rem^2=x^2/(1+e^2) + y^2/(1-e^2) Eq 2.3.6*/
cyro = (1. - eps) * (1. - eps);
ci = 0.5 * (1. - eps * eps) / sqe;
wrem = sqrt(rc * rc + x * x / cxro + y * y / cyro); /*wrem^2=w^2+rem^2 with w core radius*/
den1 = 2.*sqe * wrem - y * cx1inv;
den1 = cx1 * cx1 * x * x + den1 * den1;
num2 = 2.*rc * sqe - y;
den2 = x * x + num2 * num2;
didxre = ci * ( cx1 * (2.*sqe * x * x / cxro / wrem - 2.*sqe * wrem + y * cx1inv) / den1 + num2 / den2 );
didyre = ci * ( (2 * sqe * x * y * cx1 / cyro / wrem - x) / den1 + x / den2 );
didyim = ci * ( (2 * sqe * wrem * cx1inv - y * cx1inv * cx1inv - 4 * eps * y / cyro +
2 * sqe * y * y / cyro / wrem * cx1inv) / den1 - num2 / den2 );
//if(blockIdx.x*blockDim.x + threadIdx.x == 0 && blockIdx.y*blockDim.y + threadIdx.y == 0) printf("didxre %f didyre %f didyim %f ",didxre,didyre,didyim);
clumpcut.a = lens->b0[i] * didxre;
clumpcut.b = clumpcut.d = lens->b0[i] * didyre; //(didyre+didxim)/2.;
clumpcut.c = lens->b0[i] * didyim;
#endif
///////////
//if(blockIdx.x*blockDim.x + threadIdx.x == 0 && blockIdx.y*blockDim.y + threadIdx.y == 0) printmat_gpu(clumpcore);
//if(blockIdx.x*blockDim.x + threadIdx.x == 0 && blockIdx.y*blockDim.y + threadIdx.y == 0) printmat_gpu(clumpcut);
//
#if 1
clumpcore.a = t05 * (clumpcore.a - clumpcut.a);
clumpcore.b = t05 * (clumpcore.b - clumpcut.b);
clumpcore.c = t05 * (clumpcore.c - clumpcut.c);
clumpcore.d = t05 * (clumpcore.d - clumpcut.d);
//if(blockIdx.x*blockDim.x + threadIdx.x == 0 && blockIdx.y*blockDim.y + threadIdx.y == 0) printmat_gpu(clumpcore);
//rotation matrix 1
clumpcut.a = clumpcore.a * cose + clumpcore.b * -sine;
clumpcut.b = clumpcore.a * sine + clumpcore.b * cose;
clumpcut.c = clumpcore.d * sine + clumpcore.c * cose;
clumpcut.d = clumpcore.d * cose + clumpcore.c * -sine;
//if(blockIdx.x*blockDim.x + threadIdx.x == 0 && blockIdx.y*blockDim.y + threadIdx.y == 0) printmat_gpu(clumpcut);
//rotation matrix 2
clump.a = cose * clumpcut.a + -sine * clumpcut.d;
clump.b = cose * clumpcut.b + -sine * clumpcut.c;
clump.c = sine * clumpcut.b + cose * clumpcut.c;
clump.d = sine * clumpcut.a + cose * clumpcut.d;
#endif
//if(blockIdx.x*blockDim.x + threadIdx.x == 0 && blockIdx.y*blockDim.y + threadIdx.y == 0) printmat_gpu(clump);
//vala += clump.a;
//valb += clump.b;
//valc += clump.c;
//vald += clump.d;
grad2.a += clump.a;
grad2.b += clump.b;
grad2.c += clump.c;
grad2.d += clump.d;
//if(blockIdx.x*blockDim.x + threadIdx.x == 0 && blockIdx.y*blockDim.y + threadIdx.y == 0) printmat_gpu(grad2);
}
else if((RR = true_coord.x * true_coord.x + true_coord.y * true_coord.y) > 0.){
// Circular dPIE Elliasdottir 2007 Eq A23 slighly modified for t05
type_t X,Y,z,p,t05;
X = lens->rcore[i];
Y = lens->rcut[i];
t05 = lens->b0[i] * Y / (Y - X); // 1/u because t05/sqrt(u) and normalised Q/sqrt(u)
z = sqrt(RR + X * X) - X - sqrt(RR + Y * Y) + Y; // R*dphi/dR
X = RR / X;
Y = RR / Y;
p = (1. - 1. / sqrt(1. + X / lens->rcore[i])) / X - (1. - 1. / sqrt(1. + Y / lens->rcut[i])) / Y; // d2phi/dR2
X = true_coord.x * true_coord.x / RR;
Y = true_coord.y * true_coord.y / RR;
clump.a = t05 * (p * X + z * Y / RR);
clump.c = t05 * (p * Y + z * X / RR);
X = true_coord.x * true_coord.y / RR;
clump.b = clump.d = t05 * (p * X - z * X / RR);
grad2.a += clump.a;
grad2.b += clump.b;
grad2.c += clump.c;
grad2.d += clump.d;
}
else
{
clump.a = clump.c = lens->b0[i] / lens->rcore[i]/ 2.;
clump.b = clump.d = 0.;
grad2.a += clump.a;
grad2.b += clump.b;
grad2.c += clump.c;
grad2.d += clump.d;
}
}
//
//grad2.a = vala;
//grad2.b = valb;
//grad2.c = valc;
//grad2.d = vald;
#endif
return(grad2);
}
__device__ matrix module_potentialDerivatives_totalGradient2_14_SOA_GPU(const struct point *pImage, const struct Potential_SOA *lens, int shalos, int nhalos){
//int col = blockIdx.x*blockDim.x + threadIdx.x;
//int row = blockIdx.y*blockDim.y + threadIdx.y;
//if(row == 0 && col == 0) printf("Start GPU Grad 14 %f %f\n",lens->anglecos[0],lens->anglesin[0]);
struct matrix grad2, grad2_temp, clump;
grad2.a = 0;
grad2.b = 0;
grad2.c = 0;
grad2.d = 0;
for(int i = shalos; i < shalos + nhalos; i++)
{
//if(col == 0 and row == 0)printf(" 14! : I %d %f %f %f ",i, lens->anglecos[i],lens->anglesin[i],lens->ellipticity_potential[i]);
grad2_temp.a = 0;
grad2_temp.b = 0;
grad2_temp.c = 0;
grad2_temp.d = 0;
type_t cose = lens->anglecos[i];
type_t sine = lens->anglesin[i];
grad2_temp.a = lens->ellipticity_potential[i]*3;
grad2_temp.c = -lens->ellipticity_potential[i]*3;
//grad2.b = grad2.d = 0;
//rotation matrix 1
clump.a = grad2_temp.a * cose + grad2_temp.b * -sine;
clump.b = grad2_temp.a * sine + grad2_temp.b * cose;
clump.c = grad2_temp.d * sine + grad2_temp.c * cose;
clump.d = grad2_temp.d * cose + grad2_temp.c * -sine;
//if(blockIdx.x*blockDim.x + threadIdx.x == 0 && blockIdx.y*blockDim.y + threadIdx.y == 0) printmat_gpu(clumpcut);
//rotation matrix 2
grad2.a += cose * clump.a + -sine * clump.d;
grad2.b += cose * clump.b + -sine * clump.c;
grad2.c += sine * clump.b + cose * clump.c;
grad2.d += sine * clump.a + cose * clump.d;
}
return(grad2);
}
#if 1
typedef struct matrix (*halo_func2_GPU_t) (const struct point *pImage, const struct Potential_SOA *lens, int shalos, int nhalos);
__constant__ halo_func2_GPU_t halo_func2_GPU[100] =
{
0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, module_potentialDerivatives_totalGradient2_14_SOA_GPU, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, module_potentialDerivatives_totalGradient2_81_SOA_GPU, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0
};
#endif
__global__
void
module_potentialDerivatives_totalGradient2_SOA_GPU(type_t *grid_grad2_a, type_t *grid_grad2_b, type_t *grid_grad2_c, type_t *grid_grad2_d, const struct Potential_SOA *lens, const struct grid_param *frame, int nhalos, type_t dx, type_t dy, int nbgridcells_x, int nbgridcells_y, int istart, int jstart)
{
struct point image_point;
struct matrix grad, clumpgrad;
//
int col = blockIdx.x*blockDim.x + threadIdx.x;
int row = blockIdx.y*blockDim.y + threadIdx.y;
//if(row == 0 && col == 0) printf("Start GPU \n");
//
if ((row + jstart < nbgridcells_y) && (col + istart < nbgridcells_x))
{
int index = row*nbgridcells_x + col;
// Create temp grad variable to minimise writing to global memory grid_grad
grad.a = 0;
grad.b = 0;
grad.c = 0;
grad.d = 0;
//
image_point.x = frame->xmin + (col + istart)*dx;
image_point.y = frame->ymin + (row + jstart)*dy;
//
int shalos = 0;
//if(row == 0 && col == 0) printf("Start 2 GPU \n");
//if(row == 0 && col == 0) std::cout << std::endl;;
while (shalos < nhalos)
{
int lens_type = lens->type[shalos];
int count = 1;
while (lens->type[shalos + count] == lens_type and shalos + count < nhalos) count++;
//
//if(row == 0 && col == 0) printf("point = %d \n", lens_type );
//if(row == 0 && col == 0) printf("point = %f \n", lens->ellipticity_potential[0] );
//if(row == 0 && col == 0) printf("point = %p \n", halo_func2_GPU[81] );
//
//clumpgrad = (*halo_func2_GPU[lens_type])(&image_point, lens, shalos, count);
//clumpgrad = module_potentialDerivatives_totalGradient2_81_SOA_GPU(&image_point, lens, shalos, count);
if(lens_type == 81) clumpgrad = module_potentialDerivatives_totalGradient2_81_SOA_GPU(&image_point, lens, shalos, count);
else if(lens_type == 14) clumpgrad = module_potentialDerivatives_totalGradient2_14_SOA_GPU(&image_point, lens, shalos, count);
else if(row == 0 && col == 0) printf("No kernel selected \n");
//
grad.a += clumpgrad.a;
grad.b += clumpgrad.b;
grad.c += clumpgrad.c;
grad.d += clumpgrad.d;
shalos += count;
}
//if(row == 0 && col == 0) printf(" %f %f %f %f \n",grad.a,grad.b,grad.c,grad.d);
// Write to global memory
grid_grad2_a[index] = grad.a;
grid_grad2_b[index] = grad.b;
grid_grad2_c[index] = grad.c;
grid_grad2_d[index] = grad.d;
//if(row == 0 && col == 0) printf("point = %lf \n", grid_grad2_a[index] );
}
}
__global__
void
module_potentialDerivatives_totalGradient2_SOA_GPU(type_t *grid_grad2_a, type_t *grid_grad2_b, type_t *grid_grad2_c, type_t *grid_grad2_d, const struct Potential_SOA *lens, const struct grid_param *frame, int nbgridcells, int nhalos)
{
struct point image_point;
struct matrix grad, clumpgrad;
//
int col = blockIdx.x*blockDim.x + threadIdx.x;
int row = blockIdx.y*blockDim.y + threadIdx.y;
//
if ((row < nbgridcells) && (col < nbgridcells))
{
//
type_t dx = (frame->xmax - frame->xmin)/(nbgridcells-1);
type_t dy = (frame->ymax - frame->ymin)/(nbgridcells-1);
//
int index = row*nbgridcells + col;
// Create temp grad variable to minimise writing to global memory grid_grad
grad.a = 0;
grad.b = 0;
grad.c = 0;
grad.d = 0;
//
image_point.x = frame->xmin + col*dx;
image_point.y = frame->ymin + row*dy;
//
int shalos = 0;
while (shalos < nhalos)
{
int lens_type = lens->type[shalos];
int count = 1;
while (lens->type[shalos + count] == lens_type and shalos + count < nhalos) count++;
//
clumpgrad = (*halo_func2_GPU[lens_type])(&image_point, lens, shalos, count);
//
//clumpgrad = module_potentialDerivatives_totalGradient2_81_SOA_GPU(&image_point, lens, shalos, count);
grad.a += clumpgrad.a;
grad.b += clumpgrad.b;
grad.c += clumpgrad.c;
grad.d += clumpgrad.d;
shalos += count;
}
// Write to global memory
grid_grad2_a[index] = grad.a;
grid_grad2_b[index] = grad.b;
grid_grad2_c[index] = grad.c;
grid_grad2_d[index] = grad.d;
//if ((row == 0) && (col == 9))
//printf("%f %f: %f %f\n", image_point.x, image_point.y, grid_grad_x[index], grid_grad_y[index]);
}
}
__global__
void
module_potentialDerivatives_Kmap_SOA_GPU(type_t *grid_grad2_a, type_t *grid_grad2_b, type_t *grid_grad2_c, type_t *grid_grad2_d, const struct Potential_SOA *lens, const struct grid_param *frame, int nhalos, type_t dx, type_t dy, int nbgridcells_x, int nbgridcells_y, int istart, int jstart)
{
struct point image_point;
struct matrix grad, clumpgrad;
type_t dlsds;
//
int col = blockIdx.x*blockDim.x + threadIdx.x;
int row = blockIdx.y*blockDim.y + threadIdx.y;
//if(row == 0 && col == 0) printf("Start GPU \n");
//
if ((row + jstart < nbgridcells_y) && (col + istart < nbgridcells_x))
{
int index = row*nbgridcells_x + col;
// Create temp grad variable to minimise writing to global memory grid_grad
grad.a = 0;
grad.b = 0;
grad.c = 0;
grad.d = 0;
//
image_point.x = frame->xmin + (col + istart)*dx;
image_point.y = frame->ymin + (row + jstart)*dy;
//
int shalos = 0;
//if(row == 0 && col == 0) printf("Start 2 GPU \n");
//if(row == 0 && col == 0) std::cout << std::endl;;
while (shalos < nhalos )
{
int lens_type = lens->type[shalos];
dlsds = lens->dlsds[shalos];
int count = 1;
while (lens->type[shalos + count] == lens_type and shalos + count < nhalos and lens->dlsds[shalos + count] == dlsds)
{
count++;
}
#if 1
//
//if(row == 0 && col == 0) printf("point = %p \n", halo_func2_GPU[lens_type] );
//
//clumpgrad = (*halo_func2_GPU[lens_type])(&image_point, lens, shalos, count);
if(lens_type == 81) clumpgrad = module_potentialDerivatives_totalGradient2_81_SOA_GPU(&image_point, lens, shalos, count);
else if(lens_type == 14) clumpgrad = module_potentialDerivatives_totalGradient2_14_SOA_GPU(&image_point, lens, shalos, count);
else if(row == 0 && col == 0) printf("No kernel selected \n");
//
grad.a += clumpgrad.a*dlsds;
grad.b += clumpgrad.b*dlsds;
grad.c += clumpgrad.c*dlsds;
grad.d += clumpgrad.d*dlsds;
//if(row == 0 && col == 0) printf(" %f %f %f %f \n",grad.a,grad.b,grad.c,grad.d);
#endif
dlsds = lens->dlsds[shalos];
shalos += count;
}
// Write to global memory
grid_grad2_a[index] = grad.a;
grid_grad2_b[index] = grad.b;
grid_grad2_c[index] = grad.c;
grid_grad2_d[index] = grad.d;
if(row == 0 && col == 0) printf("point = %lf \n", grid_grad2_a[index] );
}
}

Event Timeline

c4science · Help