void ocl_fini(ocl_t *ocl)
{
if (ocl)
{
clDeleteEnvironment(ocl->env);
}
}
int main(int argc, char *argv[])
{
const cl_uint numBodies = 10;
float *m = cl_malloc_array(float, numBodies);
float *t = cl_malloc_array(float, numBodies);
cl_float4 *a = cl_malloc_array(cl_float4, numBodies);
cl_float4 *v = cl_malloc_array(cl_float4, numBodies);
cl_float4 *p = cl_malloc_array(cl_float4, numBodies);
#ifdef CL_BUILD_RUNTIME
cl_uint type = clGetTypeFromString(CL_USER_DEVICE_TYPE);
cl_uint count = CL_USER_DEVICE_COUNT;
cl_environment_t *pEnv = clCreateEnvironment(KDIR"kernel_nbody.cl",type,count,notify,CL_ARGS);
#else
cl_environment_t *pEnv = clCreateEnvironmentFromBins(&gKernelBins, notify, CL_ARGS);
#endif
if (pEnv)
{
cl_uint i = 0, j = 0;
cl_uint numIterations = (argc > 1?atoi(argv[1]):10);
for (i = 0; i < numBodies; i++)
{
m[i] = frand() * ipow(10,rrand(4,27)); // masses should be 10^4 - 10^27 ("Earth heavy")
frand4(a[i], 1, 3);
frand4(v[i], 1, 2);
frand4(p[i], 4, 8);
t[i] = 0.001f; // 1 millisecond.
}
i = 0;
for (j = 0; j < numIterations; j++)
{
nbodies(pEnv, m, a, v, p, t, numBodies);
#if defined(DARWIN)
printf("[%6u] p={%lf,%lf,%lf} v={%lf,%lf,%lf} a={%lf,%lf,%lf}\n", i,
p[i][0], p[i][1], p[i][2],
v[i][0], v[i][1], v[i][2],
a[i][0], a[i][1], a[i][2]);
#else
printf("[%6u] p={%lf,%lf,%lf} v={%lf,%lf,%lf} a={%lf,%lf,%lf}\n", i,
p[i].s[0], p[i].s[1], p[i].s[2],
v[i].s[0], v[i].s[1], v[i].s[2],
a[i].s[0], a[i].s[1], a[i].s[2]);
#endif
}
clDeleteEnvironment(pEnv);
cl_free(t);
cl_free(m);
cl_free(v);
cl_free(a);
cl_free(p);
}
return 0;
}
int main(int argc, char *argv[])
{
// initialize the default values
numDevices = 1;
verbose = CL_FALSE;
no_print_logo = CL_FALSE;
strncpy(filename, "kernels.cl", CL_MAX_PATHSIZE);
//strncpy(outfile, "a.out", CL_MAX_PATHSIZE);
memset(outfile, 0, sizeof(outfile));
//strncpy(precomp, "kernels.h", CL_MAX_PATHSIZE);
memset(precomp, 0, sizeof(precomp));
if (argc > 1)
{
cl_environment_t *pEnv = NULL;
cl_uint dev_type = CL_DEVICE_TYPE_DEFAULT;
clOptions(argc, argv);
print_logo();
if (verbose && filename[0] != '\0')
printf("Compiling File: %s\n",filename);
if (verbose && cl_args[0] != '\0')
printf("CL ARGS: %s\n",cl_args);
if (verbose && precomp[0] != '\0')
printf("Precompiled Header: %s\n", precomp);
dev_type = clGetTypeFromString(cl_device_types);
if (verbose)
printf("%u devices as type 0x%08x\n",numDevices,dev_type);
// process the kernel
pEnv = clCreateEnvironment(filename, dev_type, numDevices, notify, cl_args);
if (pEnv == NULL)
{
printf("ERROR: Failed to compile %s\n", filename);
}
clDeleteEnvironment(pEnv);
}
else
{
cl_uint i;
print_logo();
printf("Usage:\n$ %s [OPTIONS] \n", argv[0]);
for (i = 0; i < numOpts; i++)
{
printf("Option %s\t\t%s\n",options[i].arg, options[i].description);
}
printf("Example:\n$ %s -f %s -d %u -o %s\n",argv[0],filename,numDevices,outfile);
}
return 0;
}
int main(int argc, char *argv[])
{
if (argc == 1)
{
const cl_uint width = 1920;
const cl_uint height = 1080;
const cl_uint numPixels = height * width;
cl_int err = CL_SUCCESS;
float *r = cl_malloc_array(float, numPixels);
float *g = cl_malloc_array(float, numPixels);
float *b = cl_malloc_array(float, numPixels);
float *y = cl_malloc_array(float, numPixels);
float *u = cl_malloc_array(float, numPixels);
float *v = cl_malloc_array(float, numPixels);
cl_uchar *Yp = cl_malloc_array(cl_uchar, numPixels);
cl_uchar *Up = cl_malloc_array(cl_uchar, numPixels);
cl_uchar *Vp = cl_malloc_array(cl_uchar, numPixels);
float bt601[9] = {0.257f, 0.504f, 0.098f, -0.148f, -0.291f, 0.439f, 0.439f, -0.368f, -0.071f};
time_t start, diff;
clock_t c_start, c_diff1, c_diff2;
#ifdef CL_BUILD_RUNTIME
cl_environment_t *pEnv = clCreateEnvironment(KDIR"kernel_yuv.cl",CL_DEVICE_TYPE_GPU,1,notify, CL_ARGS);
#else
cl_environment_t *pEnv = clCreateEnvironmentFromBins(&gKernelBins, notify, CL_ARGS);
#endif
printf("Processing %ux%u => %u pixels\n", width, height, numPixels);
if (pEnv && r && g && b && y && u && v)
{
cl_uint i = 0;
srand((unsigned int)time(NULL));
// initialize the data
for (i = 0; i < numPixels; i++)
{
r[i] = frrand(0.0,1.0); // [0-1]
g[i] = frrand(0.0,1.0); // [0-1]
b[i] = frrand(0.0,1.0); // [0-1]
y[i] = 0.00;
u[i] = 0.00;
v[i] = 0.00;
}
start = time(NULL);
c_start = clock();
err = cl_convert_rgbf_to_yuvf_bt601(pEnv, r, g, b, y, u, v, numPixels);
cl_assert(err == CL_SUCCESS,printf("Error = %d\n",err));
c_diff1 = clock() - c_start;
diff = time(NULL) - start;
printf("With Constants Version Ran in %lu seconds (%lu ticks)\n", diff, c_diff1);
// initialize the data
for (i = 0; i < numPixels; i++)
{
r[i] = frrand(0.0,1.0); // [0-1]
g[i] = frrand(0.0,1.0); // [0-1]
b[i] = frrand(0.0,1.0); // [0-1]
y[i] = 0.00;
u[i] = 0.00;
v[i] = 0.00;
}
start = time(NULL);
c_start = clock();
cl_convert_rgbf_to_yuvf(pEnv, r, g, b, y, u, v, bt601, numPixels);
c_diff2 = clock() - c_start;
diff = time(NULL) - start;
printf("With No Constants Version Ran in %lu seconds (%lu ticks)\n", diff, c_diff2);
normalize_float(y, 16, 235, Yp, numPixels);
normalize_float(u, -128, 128, Up, numPixels);
normalize_float(v, -128, 128, Vp, numPixels);
#ifdef CL_DEBUG
for (i = 0; i < numPixels; i++)
printf("YUV = {0x%02x, 0x%02x, 0x%02x}\n", Yp[i], Up[i], Vp[i]);
#endif
clDeleteEnvironment(pEnv);
}
cl_free(r); cl_free(g); cl_free(b);
cl_free(y); cl_free(u); cl_free(v);
}
else if (argc >= 6)
int main(int argc, char *argv[])
{
if (argc >= 5)
{
FILE *fi = NULL;
FILE *fo = NULL;
cl_uint width = atoi(argv[2]);
cl_uint height = atoi(argv[3]);
cl_int err = CL_SUCCESS;
#ifdef CL_BUILD_RUNTIME
cl_environment_t *pEnv = clCreateEnvironment(KDIR"kernel_imgfilter.cl", CL_DEVICE_TYPE_GPU, 2, notify, CL_ARGS);
#else
cl_environment_t *pEnv = clCreateEnvironmentFromBins(&gKernelBins, notify, CL_ARGS);
#endif
if (!pEnv)
{
clDeleteEnvironment(pEnv);
return -1;
}
clPrintAllKernelInfo(pEnv->kernels[0]);
fi = fopen(argv[1], "rb");
fo = fopen(argv[4], "wb+");
if (fi && fo)
{
cl_char sobel[2][3][3] = { { {1,2,1}, {0,0,0}, {-1,-2,-1} }, { {-1,0,1}, {-2,0,2}, {-1,0,1} } };
size_t numBytes = width * height * sizeof(unsigned char);
cl_uchar *input_image = (cl_uchar *)malloc(numBytes);
cl_uchar *output_image = (cl_uchar *)malloc(numBytes);
clock_t c_start, c_diff;
cl_uint limit = 255;
cl_uint range = range_of_operator((cl_char *)sobel, 3, limit);
printf("Range of Sobel = %u, Limit = %u\n",range,limit);
do {
numBytes = fread(input_image, 1, numBytes, fi);
if (numBytes == 0)
break;
c_start = clock();
err = imgfilter1d_opt(pEnv, width, height,
input_image, width,
output_image, width,
(cl_char *)sobel, 3,
range, limit);
c_diff = clock() - c_start;
printf("Sobel took %lu ticks\n", c_diff);
numBytes = fwrite(output_image, 1, numBytes, fo);
} while (err == CL_SUCCESS);
free(input_image);
free(output_image);
fclose(fi);
fclose(fo);
}
clDeleteEnvironment(pEnv);
}
else
{
printf("%s <filename> <width> <height> <outfile>\n",argv[0]);
}
return 0;
}
int main(int argc, char *argv[])
{
spacetime = (spacetime_t *)calloc(1, sizeof(spacetime_t));
if (spacetime)
{
char title[80];
// initialize the spacetime info
spacetime->m_fps = 30;
spacetime->m_body.m_mass = MASS_MOON;
spacetime->m_body.m_radius = 2.0f;
SUBREF(spacetime->m_body.m_position,X) = -GRID_DIM/2;
SUBREF(spacetime->m_body.m_velocity,X) = 1;
spacetime->m_camera.m_radius = 20.0f;
spacetime->m_camera.m_theta = -M_PI/2;
spacetime->m_camera.m_phi = M_PI/2;
SUBREF(spacetime->m_camera.m_up,Z) = 1.0f;
spacetime->m_width = 640;
spacetime->m_height = 480;
spacetime->m_camera.m_fov = 40.0f;
spacetime->m_camera.m_ratio = (float)spacetime->m_width/spacetime->m_height;
spacetime->m_camera.m_near = 1.0f;
spacetime->m_camera.m_far = 7000.0f;
// initialize the grid points (from -GRID_DIM/2 to GRID_DIM/2)
for (int iz = 0; iz < GRID_DIM; iz++) {
for (int iy = 0; iy < GRID_DIM; iy++) {
for (int ix = 0; ix < GRID_DIM; ix++) {
int i = (iz*GRID_DIM*GRID_DIM) + (iy*GRID_DIM) + (ix);
SUBREF(spacetime->m_grid.m_position[i],X) = (cl_double)(ix - GRID_DIM/2);
SUBREF(spacetime->m_grid.m_position[i],Y) = (cl_double)(iy - GRID_DIM/2);
SUBREF(spacetime->m_grid.m_position[i],Z) = (cl_double)(iz - GRID_DIM/2);
spacetime->m_grid.m_mass[i] = MASS_IOTA;
}
}
}
#ifdef CL_BUILD_RUNTIME
spacetime->m_pEnv = clCreateEnvironment(KDIR "kernel_spacetime.cl", CL_DEVICE_TYPE_GPU, 1, ocl_notify, CL_ARGS);
#else
spacetime->m_pEnv = clCreateEnvironmentFromBins(&gKernelBins, ocl_notify, CL_ARGS);
#endif
if (!spacetime->m_pEnv) {
printf("OpenCL Environment failure!\n");
free(spacetime);
return -1;
}
sprintf(title,"Space-Time Warping Due to Mass; Simulation by Erik Rainey");
glutInitDisplayMode(GLUT_RGBA|GLUT_DEPTH|GLUT_DOUBLE|GLUT_MULTISAMPLE);
glutInitWindowPosition(100, 100);
glutInitWindowSize(spacetime->m_width, spacetime->m_height);
glutInit(&argc, argv);
glutCreateWindow(title);
// initialize the display settings
glEnable(GL_CULL_FACE);
glClearColor(0,0,0,0); // set the background color to the blackness of space ... !
// initialize each light
for (size_t i = 0; i < dimof(spacetime->m_lights); i++) {
GLfloat arr[4];
SUBREF(spacetime->m_lights[i].m_ambient,A) = 1.0f;
SUBREF(spacetime->m_lights[i].m_diffuse,R) = 1.0f;
SUBREF(spacetime->m_lights[i].m_diffuse,G) = 1.0f;
SUBREF(spacetime->m_lights[i].m_diffuse,B) = 1.0f;
SUBREF(spacetime->m_lights[i].m_diffuse,A) = 1.0f;
SUBREF(spacetime->m_lights[i].m_position,X) = 20.0f;
SUBREF(spacetime->m_lights[i].m_position,Y) = 20.0f;
SUBREF(spacetime->m_lights[i].m_position,Z) = 20.0f;
spacetime->m_lights[i].m_light = GL_LIGHT0 + ((GLenum)i);
clftoglf(arr, spacetime->m_lights[i].m_ambient);
glLightfv(spacetime->m_lights[i].m_light, GL_AMBIENT, arr);
clftoglf(arr, spacetime->m_lights[i].m_diffuse);
glLightfv(spacetime->m_lights[i].m_light, GL_DIFFUSE, arr);
clftoglf(arr, spacetime->m_lights[i].m_position);
glLightfv(spacetime->m_lights[i].m_light, GL_POSITION, arr);
}
// install the hooks
glutDisplayFunc(glut_display);
glutReshapeFunc(glut_reshape);
glutIdleFunc(NULL);
uint32_t ms = 1000 / (int)spacetime->m_fps;
glutTimerFunc(ms,glut_timer,0); // give initial count as zero
glutPassiveMotionFunc(glut_passive);
glutKeyboardFunc(glut_keyfunc);
// install the main loop, I don't think we return from this until killed
glutMainLoop();
free(spacetime);
clDeleteEnvironment(spacetime->m_pEnv);
}
return 0;
}