steepest_descent.cu

Go to the documentation of this file.

 
// System includes
#include <stdio.h>
#include <stdlib.h>
 
// Device includes
#include <cuda_runtime.h>
 
// Neko includes
#include <neko/device/cuda/check.h>
#include <neko/device/device_config.h>
 
// Local includes
#include "steepest_descent_kernel.h"
 
extern "C" {
 
void cuda_sd_direction(void* s_d, void* x_d, real* xmin, real* xmax,
    int* n) {
    if (*n < 1) return;
 
    const dim3 nthrds(1024, 1, 1);
    const dim3 nblcks(((*n) + 1024 - 1) / 1024, 1, 1);
 
    sd_direction_kernel<real>
        <<<nblcks, nthrds, 0, (cudaStream_t)glb_cmd_queue>>>
        ((real*)s_d, (real*)x_d, *xmin, *xmax, *n);
    CUDA_CHECK(cudaGetLastError());
}
 
void cuda_sd_update_move_limits(void* delta_d, void* x_d, void* x1_d,
    void* x2_d, real* dec, real* inc, real* dmin, real* dmax, int* n) {
    if (*n < 1) return;
 
    const dim3 nthrds(1024, 1, 1);
    const dim3 nblcks(((*n) + 1024 - 1) / 1024, 1, 1);
 
    sd_update_move_limits_kernel<real>
        <<<nblcks, nthrds, 0, (cudaStream_t)glb_cmd_queue>>>
        ((real*)delta_d, (real*)x_d, (real*)x1_d, (real*)x2_d,
         *dec, *inc, *dmin, *dmax, *n);
    CUDA_CHECK(cudaGetLastError());
}
 
void cuda_sd_update_design(void* x_new_d, void* x_d, void* s_d,
    void* delta_d, real* psi, real* smax, real* xmin, real* xmax, int* n) {
    if (*n < 1) return;
 
    const dim3 nthrds(1024, 1, 1);
    const dim3 nblcks(((*n) + 1024 - 1) / 1024, 1, 1);
 
    sd_update_design_kernel<real>
        <<<nblcks, nthrds, 0, (cudaStream_t)glb_cmd_queue>>>
        ((real*)x_new_d, (real*)x_d, (real*)s_d, (real*)delta_d,
         *psi, *smax, *xmin, *xmax, *n);
    CUDA_CHECK(cudaGetLastError());
}
 
}