Neko 1.99.9
A portable framework for high-order spectral element flow simulations
Loading...
Searching...
No Matches
Macros | Functions | Variables
opr_conv1.c File Reference
#include <CL/cl.h>
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <device/device_config.h>
#include <device/opencl/jit.h>
#include <device/opencl/prgm_lib.h>
#include <device/opencl/check.h>
#include <common/neko_log.h>
#include "conv1_kernel.cl.h"
Include dependency graph for opr_conv1.c:

Go to the source code of this file.

Macros

#define STR(X)   #X
 
#define CASE_1D(LX, QUEUE, EVENT)
 
#define CASE_KSTEP(LX, QUEUE, EVENT)
 
#define CASE(LX)
 
#define CASE_LARGE(LX)
 

Functions

void opencl_conv1 (void *du, void *u, void *vx, void *vy, void *vz, void *dx, void *dy, void *dz, void *drdx, void *dsdx, void *dtdx, void *drdy, void *dsdy, void *dtdy, void *drdz, void *dsdz, void *dtdz, void *jacinv, int *nel, int *gdim, int *lx)
 

Variables

int * autotune_conv1 = NULL
 

Macro Definition Documentation

◆ CASE

#define CASE (   LX)

◆ CASE_1D

#define CASE_1D (   LX,
  QUEUE,
  EVENT 
)
Value:
{ \
cl_kernel kernel = clCreateKernel(conv1_program, \
STR(conv1_kernel_lx##LX), &err); \
CL_CHECK(err); \
\
CL_CHECK(clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *) &du)); \
CL_CHECK(clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *) &u)); \
CL_CHECK(clSetKernelArg(kernel, 2, sizeof(cl_mem), (void *) &vx)); \
CL_CHECK(clSetKernelArg(kernel, 3, sizeof(cl_mem), (void *) &vy)); \
CL_CHECK(clSetKernelArg(kernel, 4, sizeof(cl_mem), (void *) &vz)); \
CL_CHECK(clSetKernelArg(kernel, 5, sizeof(cl_mem), (void *) &dx)); \
CL_CHECK(clSetKernelArg(kernel, 6, sizeof(cl_mem), (void *) &dy)); \
CL_CHECK(clSetKernelArg(kernel, 7, sizeof(cl_mem), (void *) &dz)); \
CL_CHECK(clSetKernelArg(kernel, 8, sizeof(cl_mem), (void *) &drdx)); \
CL_CHECK(clSetKernelArg(kernel, 9, sizeof(cl_mem), (void *) &dsdx)); \
CL_CHECK(clSetKernelArg(kernel, 10, sizeof(cl_mem), (void *) &dtdx)); \
CL_CHECK(clSetKernelArg(kernel, 11, sizeof(cl_mem), (void *) &drdy)); \
CL_CHECK(clSetKernelArg(kernel, 12, sizeof(cl_mem), (void *) &dsdy)); \
CL_CHECK(clSetKernelArg(kernel, 13, sizeof(cl_mem), (void *) &dtdy)); \
CL_CHECK(clSetKernelArg(kernel, 14, sizeof(cl_mem), (void *) &drdz)); \
CL_CHECK(clSetKernelArg(kernel, 15, sizeof(cl_mem), (void *) &dsdz)); \
CL_CHECK(clSetKernelArg(kernel, 16, sizeof(cl_mem), (void *) &dtdz)); \
CL_CHECK(clSetKernelArg(kernel, 17, sizeof(cl_mem), (void *) &jacinv)); \
\
CL_CHECK(clEnqueueNDRangeKernel((cl_command_queue) QUEUE, \
kernel, 1, NULL, &global_item_size, \
&local_item_size, 0, NULL, EVENT)); \
CL_CHECK(clReleaseKernel(kernel)); \
}
ale_add_kinematics_kernel
__global__ void ale_add_kinematics_kernel(const int n, T *__restrict__ wx, T *__restrict__ wy, T *__restrict__ wz, const T *__restrict__ x_ref, const T *__restrict__ y_ref, const T *__restrict__ z_ref, const T *__restrict__ phi, const T *__restrict__ x, const T *__restrict__ y, const T *__restrict__ z, const kinematics_params_t kin_params)
Definition ale_kinematics_kernel.h:52
dtdy
__global__ void T *__restrict__ 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__ dtdy
Definition ax_helm_full_kernel.h:60
dtdx
__global__ void T *__restrict__ 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__ dtdx
Definition ax_helm_full_kernel.h:59
dtdz
__global__ void T *__restrict__ 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__ dtdz
Definition ax_helm_full_kernel.h:61
jacinv
__global__ void T *__restrict__ 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__ const T *__restrict__ jacinv
Definition ax_helm_full_kernel.h:62
dz
__global__ void T *__restrict__ T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ dz
Definition ax_helm_full_kernel.h:51
dsdz
__global__ void T *__restrict__ 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 ax_helm_full_kernel.h:58
dx
__global__ void T *__restrict__ T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ dx
Definition ax_helm_full_kernel.h:49
u
__global__ void T *__restrict__ T *__restrict__ const T *__restrict__ u
Definition ax_helm_full_kernel.h:46
dy
__global__ void T *__restrict__ T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ dy
Definition ax_helm_full_kernel.h:50
drdz
__global__ void T *__restrict__ 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 ax_helm_full_kernel.h:55
drdx
__global__ void T *__restrict__ 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__ drdx
Definition ax_helm_full_kernel.h:53
dsdx
__global__ void T *__restrict__ 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__ dsdx
Definition ax_helm_full_kernel.h:56
dsdy
__global__ void T *__restrict__ 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__ dsdy
Definition ax_helm_full_kernel.h:57
drdy
__global__ void T *__restrict__ 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 ax_helm_full_kernel.h:54
vz
__global__ void const T *__restrict__ const T *__restrict__ const T *__restrict__ const T *__restrict__ vz
Definition conv1_kernel.h:139
vx
__global__ void const T *__restrict__ const T *__restrict__ vx
Definition conv1_kernel.h:137
vy
__global__ void const T *__restrict__ const T *__restrict__ const T *__restrict__ vy
Definition conv1_kernel.h:138
STR
#define STR(X)
conv1_program
void * conv1_program

◆ CASE_KSTEP

#define CASE_KSTEP (   LX,
  QUEUE,
  EVENT 
)
Value:
{ \
cl_kernel kernel = clCreateKernel(conv1_program, \
STR(conv1_kernel_kstep_lx##LX), &err); \
CL_CHECK(err); \
\
CL_CHECK(clSetKernelArg(kernel, 0, sizeof(cl_mem), (void *) &du)); \
CL_CHECK(clSetKernelArg(kernel, 1, sizeof(cl_mem), (void *) &u)); \
CL_CHECK(clSetKernelArg(kernel, 2, sizeof(cl_mem), (void *) &vx)); \
CL_CHECK(clSetKernelArg(kernel, 3, sizeof(cl_mem), (void *) &vy)); \
CL_CHECK(clSetKernelArg(kernel, 4, sizeof(cl_mem), (void *) &vz)); \
CL_CHECK(clSetKernelArg(kernel, 5, sizeof(cl_mem), (void *) &dx)); \
CL_CHECK(clSetKernelArg(kernel, 6, sizeof(cl_mem), (void *) &dy)); \
CL_CHECK(clSetKernelArg(kernel, 7, sizeof(cl_mem), (void *) &dz)); \
CL_CHECK(clSetKernelArg(kernel, 8, sizeof(cl_mem), (void *) &drdx)); \
CL_CHECK(clSetKernelArg(kernel, 9, sizeof(cl_mem), (void *) &dsdx)); \
CL_CHECK(clSetKernelArg(kernel, 10, sizeof(cl_mem), (void *) &dtdx)); \
CL_CHECK(clSetKernelArg(kernel, 11, sizeof(cl_mem), (void *) &drdy)); \
CL_CHECK(clSetKernelArg(kernel, 12, sizeof(cl_mem), (void *) &dsdy)); \
CL_CHECK(clSetKernelArg(kernel, 13, sizeof(cl_mem), (void *) &dtdy)); \
CL_CHECK(clSetKernelArg(kernel, 14, sizeof(cl_mem), (void *) &drdz)); \
CL_CHECK(clSetKernelArg(kernel, 15, sizeof(cl_mem), (void *) &dsdz)); \
CL_CHECK(clSetKernelArg(kernel, 16, sizeof(cl_mem), (void *) &dtdz)); \
CL_CHECK(clSetKernelArg(kernel, 17, sizeof(cl_mem), (void *) &jacinv)); \
\
CL_CHECK(clEnqueueNDRangeKernel((cl_command_queue) QUEUE, \
kernel, 2, NULL, global_kstep, \
local_kstep, 0, NULL, EVENT)); \
CL_CHECK(clReleaseKernel(kernel)); \
}

◆ CASE_LARGE

#define CASE_LARGE (   LX)
Value:
case LX: \
CASE_KSTEP(LX, glb_cmd_queue, NULL); \
break

◆ STR

#define STR (   X)    #X

Function Documentation

◆ opencl_conv1()

void opencl_conv1 ( void *  du,
void *  u,
void *  vx,
void *  vy,
void *  vz,
void *  dx,
void *  dy,
void *  dz,
void *  drdx,
void *  dsdx,
void *  dtdx,
void *  drdy,
void *  dsdy,
void *  dtdy,
void *  drdz,
void *  dsdz,
void *  dtdz,
void *  jacinv,
int *  nel,
int *  gdim,
int *  lx 
)

Fortran wrapper for device OpenCL convective terms

Definition at line 57 of file opr_conv1.c.

Here is the call graph for this function:
Here is the caller graph for this function:

Variable Documentation

◆ autotune_conv1

int* autotune_conv1 = NULL

Definition at line 52 of file opr_conv1.c.