root / Pi / C / OpenCL / PiOpenCL_MWC.c @ 308
Historique | Voir | Annoter | Télécharger (15,06 ko)
| 1 |
// Pi Dart Dash in OpenCL in C, illustrative example
|
|---|---|
| 2 |
//
|
| 3 |
// Emmanuel Quemener <emmanuel.quemener@gmail.com>
|
| 4 |
//
|
| 5 |
// CC BY-NC-SA 2011 : Emmanuel QUEMENER <emmanuel.quemener@gmail.com>
|
| 6 |
// Copyleft Cecill v2
|
| 7 |
//
|
| 8 |
// -h : print the documentation and detect devices as (platform,device)
|
| 9 |
//
|
| 10 |
// classical use:
|
| 11 |
// #1 OpenCL Plateform ID: get this information with -h option
|
| 12 |
// #2 OpenCL Device ID: get this information with -h option
|
| 13 |
// #3 Minimal number of iterations:
|
| 14 |
// #4 Parallel Rate: scattering global work in parts executed //ly
|
| 15 |
// #5 Loops (to improve statistics)
|
| 16 |
// #6 Type of variables INT32, INT64, FP32, FP64
|
| 17 |
//
|
| 18 |
// To compile : gcc -o PiOpenCL PiOpenCL.c -lOpenCL -lm
|
| 19 |
|
| 20 |
#define CL_TARGET_OPENCL_VERSION 220 |
| 21 |
#define CL_USE_DEPRECATED_OPENCL_1_2_APIS
|
| 22 |
#include <stdio.h> |
| 23 |
#include <stdlib.h> |
| 24 |
#include <string.h> |
| 25 |
#include <CL/cl.h> |
| 26 |
#include <stdint.h> |
| 27 |
#include <math.h> |
| 28 |
#include <sys/time.h> |
| 29 |
|
| 30 |
#define TINT32 0 |
| 31 |
#define TINT64 1 |
| 32 |
#define TFP32 2 |
| 33 |
#define TFP64 3 |
| 34 |
|
| 35 |
int DetectOpenCLDevices(void) |
| 36 |
{
|
| 37 |
int i, j;
|
| 38 |
char* value;
|
| 39 |
size_t valueSize; |
| 40 |
cl_uint platformCount; |
| 41 |
cl_platform_id* platforms; |
| 42 |
cl_uint deviceCount; |
| 43 |
cl_device_id* devices; |
| 44 |
cl_uint maxComputeUnits; |
| 45 |
cl_int maxWorkGroupSize; |
| 46 |
cl_int maxWorkItemSizes; |
| 47 |
cl_device_type dev_type; |
| 48 |
|
| 49 |
// get all platforms
|
| 50 |
clGetPlatformIDs(0, NULL, &platformCount); |
| 51 |
platforms = (cl_platform_id*) malloc(sizeof(cl_platform_id) * platformCount);
|
| 52 |
clGetPlatformIDs(platformCount, platforms, NULL);
|
| 53 |
|
| 54 |
printf("OpenCL statistics: %d platform(s) detected\n\n",platformCount);
|
| 55 |
|
| 56 |
for (i = 0; i < platformCount; i++) { |
| 57 |
|
| 58 |
// get all devices
|
| 59 |
clGetDeviceIDs(platforms[i], CL_DEVICE_TYPE_ALL, 0, NULL, &deviceCount); |
| 60 |
devices = (cl_device_id*) malloc(sizeof(cl_device_id) * deviceCount);
|
| 61 |
clGetDeviceIDs(platforms[i], CL_DEVICE_TYPE_ALL, deviceCount, devices, NULL);
|
| 62 |
|
| 63 |
// for each device print critical attributes
|
| 64 |
for (j = 0; j < deviceCount; j++) { |
| 65 |
|
| 66 |
// print device name
|
| 67 |
clGetDeviceInfo(devices[j], CL_DEVICE_NAME, 0, NULL, &valueSize); |
| 68 |
value = (char*) malloc(valueSize);
|
| 69 |
clGetDeviceInfo(devices[j], CL_DEVICE_NAME, valueSize, value, NULL);
|
| 70 |
printf("Device (%d,%d): %s\n",i, j, value);
|
| 71 |
free(value); |
| 72 |
|
| 73 |
// print type device CPU/GPU/ACCELERATOR
|
| 74 |
clGetDeviceInfo(devices[j], CL_DEVICE_TYPE, sizeof(dev_type), &dev_type, NULL); |
| 75 |
printf("\tDevice Type: ");
|
| 76 |
if(dev_type & CL_DEVICE_TYPE_GPU)
|
| 77 |
printf("CL_DEVICE_TYPE_GPU ");
|
| 78 |
if(dev_type & CL_DEVICE_TYPE_CPU)
|
| 79 |
printf("CL_DEVICE_TYPE_CPU ");
|
| 80 |
if(dev_type & CL_DEVICE_TYPE_ACCELERATOR)
|
| 81 |
printf("CL_DEVICE_TYPE_ACCELERATOR ");
|
| 82 |
if(dev_type & CL_DEVICE_TYPE_DEFAULT)
|
| 83 |
printf("CL_DEVICE_TYPE_DEFAULT ");
|
| 84 |
printf("\n");
|
| 85 |
|
| 86 |
// print device vendor
|
| 87 |
clGetDeviceInfo(devices[j], CL_DEVICE_VENDOR, 0, NULL, &valueSize); |
| 88 |
value = (char*) malloc(valueSize);
|
| 89 |
clGetDeviceInfo(devices[j], CL_DEVICE_VENDOR, valueSize, value, NULL);
|
| 90 |
printf("\tDevice vendor: %s\n", value);
|
| 91 |
free(value); |
| 92 |
|
| 93 |
// print hardware device version
|
| 94 |
clGetDeviceInfo(devices[j], CL_DEVICE_VERSION, 0, NULL, &valueSize); |
| 95 |
value = (char*) malloc(valueSize);
|
| 96 |
clGetDeviceInfo(devices[j], CL_DEVICE_VERSION, valueSize, value, NULL);
|
| 97 |
printf("\tHardware version: %s\n", value);
|
| 98 |
free(value); |
| 99 |
|
| 100 |
// print software driver version
|
| 101 |
clGetDeviceInfo(devices[j], CL_DRIVER_VERSION, 0, NULL, &valueSize); |
| 102 |
value = (char*) malloc(valueSize);
|
| 103 |
clGetDeviceInfo(devices[j], CL_DRIVER_VERSION, valueSize, value, NULL);
|
| 104 |
printf("\tSoftware version: %s\n", value);
|
| 105 |
free(value); |
| 106 |
|
| 107 |
// print c version supported by compiler for device
|
| 108 |
clGetDeviceInfo(devices[j], CL_DEVICE_OPENCL_C_VERSION, 0, NULL, &valueSize); |
| 109 |
value = (char*) malloc(valueSize);
|
| 110 |
clGetDeviceInfo(devices[j], CL_DEVICE_OPENCL_C_VERSION, valueSize, value, NULL);
|
| 111 |
printf("\tOpenCL C version: %s\n", value);
|
| 112 |
free(value); |
| 113 |
|
| 114 |
// print parallel compute units
|
| 115 |
clGetDeviceInfo(devices[j], CL_DEVICE_MAX_COMPUTE_UNITS, |
| 116 |
sizeof(maxComputeUnits), &maxComputeUnits, NULL); |
| 117 |
printf("\tParallel compute units: %d\n", maxComputeUnits);
|
| 118 |
|
| 119 |
// print max work group size
|
| 120 |
clGetDeviceInfo(devices[j], CL_DEVICE_MAX_WORK_GROUP_SIZE, |
| 121 |
sizeof(maxWorkGroupSize), &maxWorkGroupSize, NULL); |
| 122 |
printf("\tMaximum Work Group Size: %d\n", maxWorkGroupSize);
|
| 123 |
|
| 124 |
// print max work items size
|
| 125 |
clGetDeviceInfo(devices[j], CL_DEVICE_MAX_WORK_ITEM_SIZES, |
| 126 |
sizeof(maxWorkItemSizes), &maxWorkItemSizes, NULL); |
| 127 |
printf("\tMaximum Work Item Sizes: %d\n", maxWorkItemSizes);
|
| 128 |
|
| 129 |
} |
| 130 |
printf("\n");
|
| 131 |
free(devices); |
| 132 |
} |
| 133 |
|
| 134 |
free(platforms); |
| 135 |
return 0; |
| 136 |
|
| 137 |
} |
| 138 |
|
| 139 |
const char* OpenCLSource[] = { |
| 140 |
"#pragma OPENCL EXTENSION cl_khr_fp64: enable \n",
|
| 141 |
"// Marsaglia RNG very simple implementation \n",
|
| 142 |
"#define znew ((z=36969*(z&65535)+(z>>16))<<16) \n",
|
| 143 |
"#define wnew ((w=18000*(w&65535)+(w>>16))&65535) \n",
|
| 144 |
"#define MWC (znew+wnew) \n",
|
| 145 |
"#define SHR3 (jsr=(jsr=(jsr=jsr^(jsr<<17))^(jsr>>13))^(jsr<<5)) \n",
|
| 146 |
"#define CONG (jcong=69069*jcong+1234567) \n",
|
| 147 |
"#define KISS ((MWC^CONG)+SHR3) \n",
|
| 148 |
"#define MWCfp MWC * 2.328306435454494e-10f \n",
|
| 149 |
"#define KISSfp KISS * 2.328306435454494e-10f \n",
|
| 150 |
"#define CONGfp CONG * 2.328306435454494e-10f \n",
|
| 151 |
"#define SHR3fp SHR3 * 2.328306435454494e-10f \n",
|
| 152 |
"#define TINT32 0 \n",
|
| 153 |
"#define TINT64 1 \n",
|
| 154 |
"#define TFP32 2 \n",
|
| 155 |
"#define TFP64 3 \n",
|
| 156 |
"#define THEONE32I 1073741824 \n",
|
| 157 |
"#define THEONE32F 1.e0f \n",
|
| 158 |
"#define THEONE64I 4611686018427387904 \n",
|
| 159 |
"#define THEONE64F (double)1.e0f \n",
|
| 160 |
"ulong MainLoop32I(ulong iterations,uint seed_z,uint seed_w,size_t work)",
|
| 161 |
"{",
|
| 162 |
" uint z=seed_z+work;",
|
| 163 |
" uint w=seed_w+work;",
|
| 164 |
" ulong total=0;",
|
| 165 |
" for (ulong i=0;i<iterations;i++)",
|
| 166 |
" {",
|
| 167 |
" uint x= MWC>>17;",
|
| 168 |
" uint y= MWC>>17;",
|
| 169 |
" ulong inside=((x*x+y*y) <= THEONE32I) ? 1:0;",
|
| 170 |
" total+=inside;",
|
| 171 |
" }",
|
| 172 |
" return(total);",
|
| 173 |
"}",
|
| 174 |
"ulong MainLoop32F(ulong iterations,uint seed_z,uint seed_w,size_t work)",
|
| 175 |
"{",
|
| 176 |
" uint z=seed_z+work;",
|
| 177 |
" uint w=seed_w+work;",
|
| 178 |
" ulong total=0;",
|
| 179 |
" for (ulong i=0;i<iterations;i++)",
|
| 180 |
" {",
|
| 181 |
" float x=(float)MWCfp ;",
|
| 182 |
" float y=(float)MWCfp ;",
|
| 183 |
" ulong inside=((x*x+y*y) <= THEONE32F) ? 1:0;",
|
| 184 |
" total+=inside;",
|
| 185 |
" }",
|
| 186 |
" return(total);",
|
| 187 |
"}",
|
| 188 |
"ulong MainLoop64I(ulong iterations,uint seed_z,uint seed_w,size_t work)",
|
| 189 |
"{",
|
| 190 |
" uint z=seed_z+work;",
|
| 191 |
" uint w=seed_w+work;",
|
| 192 |
" ulong total=0;",
|
| 193 |
" for (ulong i=0;i<iterations;i++)",
|
| 194 |
" {",
|
| 195 |
" ulong x=(ulong)(MWC>>1);",
|
| 196 |
" ulong y=(ulong)(MWC>>1);",
|
| 197 |
" ulong inside=((x*x+y*y) <= THEONE64I) ? 1:0;",
|
| 198 |
" total+=inside;",
|
| 199 |
" }",
|
| 200 |
" return(total);",
|
| 201 |
"}",
|
| 202 |
"ulong MainLoop64F(ulong iterations,uint seed_z,uint seed_w,size_t work)",
|
| 203 |
"{",
|
| 204 |
" uint z=seed_z+work;",
|
| 205 |
" uint w=seed_w+work;",
|
| 206 |
" ulong total=0;",
|
| 207 |
" for (ulong i=0;i<iterations;i++)",
|
| 208 |
"{",
|
| 209 |
" double x=(double)MWCfp ;",
|
| 210 |
" double y=(double)MWCfp ;",
|
| 211 |
" ulong inside=((x*x+y*y) <= THEONE64F) ? 1:0;",
|
| 212 |
" total+=inside;",
|
| 213 |
"}",
|
| 214 |
" return(total);",
|
| 215 |
"}",
|
| 216 |
"__kernel void MainLoopGlobal(__global ulong *s,ulong iterations,uint seed_w,uint seed_z,uint MyType)",
|
| 217 |
"{",
|
| 218 |
" ulong total;",
|
| 219 |
" if (MyType==TFP32) {",
|
| 220 |
" total=(ulong)MainLoop32F(iterations,seed_z,seed_w,get_global_id(0));",
|
| 221 |
" }",
|
| 222 |
" else if (MyType==TFP64) {",
|
| 223 |
" total=(ulong)MainLoop64F(iterations,seed_z,seed_w,get_global_id(0));",
|
| 224 |
" }",
|
| 225 |
" else if (MyType==TINT32) {",
|
| 226 |
" total=(ulong)MainLoop32I(iterations,seed_z,seed_w,get_global_id(0));",
|
| 227 |
" }",
|
| 228 |
" else if (MyType==TINT64) {",
|
| 229 |
" total=(ulong)MainLoop64I(iterations,seed_z,seed_w,get_global_id(0));",
|
| 230 |
" }",
|
| 231 |
" barrier(CLK_GLOBAL_MEM_FENCE);",
|
| 232 |
" s[get_global_id(0)]=(ulong)total;",
|
| 233 |
"}",
|
| 234 |
"__kernel void MainLoopLocal(__global ulong *s,ulong iterations,uint seed_w,uint seed_z,uint MyType)",
|
| 235 |
"{",
|
| 236 |
" ulong total;",
|
| 237 |
" if (MyType==TFP32) {",
|
| 238 |
" total=(ulong)MainLoop32F(iterations,seed_z,seed_w,get_local_id(0));",
|
| 239 |
" }",
|
| 240 |
" else if (MyType==TFP64) {",
|
| 241 |
" total=(ulong)MainLoop64F(iterations,seed_z,seed_w,get_local_id(0));",
|
| 242 |
" }",
|
| 243 |
" else if (MyType==TINT32) {",
|
| 244 |
" total=(ulong)MainLoop32I(iterations,seed_z,seed_w,get_local_id(0));",
|
| 245 |
" }",
|
| 246 |
" else if (MyType==TINT64) {",
|
| 247 |
" total=(ulong)MainLoop64I(iterations,seed_z,seed_w,get_local_id(0));",
|
| 248 |
" }",
|
| 249 |
" barrier(CLK_LOCAL_MEM_FENCE);",
|
| 250 |
" s[get_local_id(0)]=(ulong)total;",
|
| 251 |
"}",
|
| 252 |
"__kernel void MainLoopHybrid(__global ulong *s,ulong iterations,uint seed_w,uint seed_z,uint MyType)",
|
| 253 |
"{",
|
| 254 |
" ulong total;",
|
| 255 |
" if (MyType==TFP32) {",
|
| 256 |
" total=(ulong)MainLoop32F(iterations,seed_z,seed_w,get_global_id(0));",
|
| 257 |
" }",
|
| 258 |
" else if (MyType==TFP64) {",
|
| 259 |
" total=(ulong)MainLoop64F(iterations,seed_z,seed_w,get_global_id(0));",
|
| 260 |
" }",
|
| 261 |
" else if (MyType==TINT32) {",
|
| 262 |
" total=(ulong)MainLoop32I(iterations,seed_z,seed_w,get_global_id(0));",
|
| 263 |
" }",
|
| 264 |
" else if (MyType==TINT64) {",
|
| 265 |
" total=(ulong)MainLoop64I(iterations,seed_z,seed_w,get_global_id(0));",
|
| 266 |
" }",
|
| 267 |
" barrier(CLK_GLOBAL_MEM_FENCE || CLK_LOCAL_MEM_FENCE);",
|
| 268 |
" s[get_global_id(0)]=total;",
|
| 269 |
"}"
|
| 270 |
}; |
| 271 |
|
| 272 |
int main(int argc, char **argv) |
| 273 |
{
|
| 274 |
if ((argc==1)|| |
| 275 |
(strcmp(argv[1],"-h")==0)|| |
| 276 |
(strcmp(argv[1],"--help")==0)) |
| 277 |
{
|
| 278 |
printf("\nPerforms a Pi estimation by Dart Dash:\n\n"
|
| 279 |
"\t#1 OpenCL Plateform ID (default 0)\n"
|
| 280 |
"\t#2 OpenCL Device ID (default 0)\n"
|
| 281 |
"\t#3 Minimal number of iterations (default 1000000)\n"
|
| 282 |
"\t#4 Parallel Rate (default 1024)\n"
|
| 283 |
"\t#5 Loops (default 1)\n"
|
| 284 |
"\t#6 Type of variable: INT32, INT64, FP32, FP64 (default FP32)\n\n");
|
| 285 |
DetectOpenCLDevices(); |
| 286 |
} |
| 287 |
else
|
| 288 |
{
|
| 289 |
|
| 290 |
int MyPlatform=atoi(argv[1]); |
| 291 |
int MyDevice=atoi(argv[2]); |
| 292 |
|
| 293 |
struct timeval tv1,tv2;
|
| 294 |
|
| 295 |
uint64_t Iterations=1000000;
|
| 296 |
if (argc>3) { |
| 297 |
Iterations=(uint64_t)atoll(argv[3]);
|
| 298 |
} |
| 299 |
|
| 300 |
uint32_t ParallelRate=1024;
|
| 301 |
if (argc>4) { |
| 302 |
ParallelRate=(uint32_t)atoi(argv[4]);
|
| 303 |
} |
| 304 |
|
| 305 |
uint32_t Loops=1;
|
| 306 |
if (argc>5) { |
| 307 |
Loops=(uint32_t)atoi(argv[5]);
|
| 308 |
} |
| 309 |
|
| 310 |
uint32_t MyType=TFP32; |
| 311 |
if (argc>6) { |
| 312 |
if (strcmp(argv[6],"INT32")==0) { |
| 313 |
MyType=(uint32_t)TINT32; |
| 314 |
} |
| 315 |
else if (strcmp(argv[6],"INT64")==0) { |
| 316 |
MyType=(uint32_t)TINT64; |
| 317 |
} |
| 318 |
else if (strcmp(argv[6],"FP32")==0) { |
| 319 |
MyType=(uint32_t)TFP32; |
| 320 |
} |
| 321 |
else if (strcmp(argv[6],"FP64")==0) { |
| 322 |
MyType=(uint32_t)TFP64; |
| 323 |
} |
| 324 |
} |
| 325 |
|
| 326 |
printf("MyType %d\n",MyType);
|
| 327 |
|
| 328 |
cl_int err; |
| 329 |
cl_context_properties props[3] = { CL_CONTEXT_PLATFORM, 0, 0 }; |
| 330 |
|
| 331 |
// Detect, scan, get & initialize platform and device
|
| 332 |
cl_uint platformCount; |
| 333 |
cl_platform_id* platforms; |
| 334 |
cl_uint deviceCount; |
| 335 |
cl_device_id* devices; |
| 336 |
size_t valueSize; |
| 337 |
|
| 338 |
/* Setup OpenCL environment. */
|
| 339 |
|
| 340 |
// Get all platforms
|
| 341 |
err = clGetPlatformIDs(0, NULL, &platformCount); |
| 342 |
platforms = (cl_platform_id*) malloc(sizeof(cl_platform_id) * platformCount);
|
| 343 |
err = clGetPlatformIDs(platformCount, platforms, NULL);
|
| 344 |
|
| 345 |
// Get Device defined
|
| 346 |
err = clGetDeviceIDs(platforms[MyPlatform], CL_DEVICE_TYPE_ALL, 0, NULL, &deviceCount); |
| 347 |
devices = (cl_device_id*) malloc(sizeof(cl_device_id) * deviceCount);
|
| 348 |
err = clGetDeviceIDs(platforms[MyPlatform], CL_DEVICE_TYPE_ALL, deviceCount, devices, NULL);
|
| 349 |
|
| 350 |
// print device name
|
| 351 |
err = clGetDeviceInfo(devices[MyDevice], CL_DEVICE_NAME, 0, NULL, &valueSize); |
| 352 |
char* deviceName=(char*)malloc(valueSize); |
| 353 |
err = clGetDeviceInfo(devices[MyDevice], CL_DEVICE_NAME, valueSize, deviceName, NULL);
|
| 354 |
err = clGetDeviceInfo(devices[MyDevice], CL_DEVICE_VENDOR, 0, NULL, &valueSize); |
| 355 |
char* vendorName=(char*)malloc(valueSize); |
| 356 |
err = clGetDeviceInfo(devices[MyDevice], CL_DEVICE_VENDOR, valueSize, vendorName, NULL);
|
| 357 |
printf("\nDevice (%d,%d):\n\t- vendor: %s\n\t- device: %s\n\n",MyPlatform,MyDevice, vendorName,deviceName);
|
| 358 |
free(deviceName); |
| 359 |
free(vendorName); |
| 360 |
|
| 361 |
props[1] = (cl_context_properties)platforms[MyPlatform];
|
| 362 |
|
| 363 |
cl_context GPUContext = clCreateContext(props, 1, &devices[MyDevice], NULL, NULL, &err); |
| 364 |
cl_command_queue cqCommandQueue = clCreateCommandQueue(GPUContext,devices[MyDevice], 0, &err);
|
| 365 |
|
| 366 |
cl_mem GPUInside = clCreateBuffer(GPUContext, CL_MEM_WRITE_ONLY, |
| 367 |
sizeof(uint64_t) * ParallelRate, NULL, NULL); |
| 368 |
|
| 369 |
// 130 is the number of line for OpenCL code
|
| 370 |
cl_program OpenCLProgram = clCreateProgramWithSource(GPUContext, 130 ,OpenCLSource,NULL,NULL); |
| 371 |
clBuildProgram(OpenCLProgram, 0, NULL, NULL, NULL, NULL); |
| 372 |
cl_kernel OpenCLMainLoopGlobal = clCreateKernel(OpenCLProgram, "MainLoopGlobal", NULL); |
| 373 |
cl_kernel OpenCLMainLoopLocal = clCreateKernel(OpenCLProgram, "MainLoopLocal", NULL); |
| 374 |
cl_kernel OpenCLMainLoopHybrid = clCreateKernel(OpenCLProgram, "MainLoopHybrid", NULL); |
| 375 |
|
| 376 |
// Divide the total number of iterations by the parallel rate
|
| 377 |
// Add +1 to the number of per work iterations if division not integer
|
| 378 |
uint64_t IterationsEach=((Iterations%ParallelRate)==0)?Iterations/ParallelRate:Iterations/ParallelRate+1; |
| 379 |
// Initialize seeds for MWC RNG generator from Marsaglia
|
| 380 |
uint32_t seed_w=110271;
|
| 381 |
uint32_t seed_z=101008;
|
| 382 |
|
| 383 |
// Set the values of arguments for OpenCL function call
|
| 384 |
clSetKernelArg(OpenCLMainLoopGlobal, 0, sizeof(cl_mem),&GPUInside); |
| 385 |
clSetKernelArg(OpenCLMainLoopGlobal, 1, sizeof(uint64_t),&IterationsEach); |
| 386 |
clSetKernelArg(OpenCLMainLoopGlobal, 2, sizeof(uint32_t),&seed_w); |
| 387 |
clSetKernelArg(OpenCLMainLoopGlobal, 3, sizeof(uint32_t),&seed_z); |
| 388 |
clSetKernelArg(OpenCLMainLoopGlobal, 4, sizeof(uint32_t),&MyType); |
| 389 |
|
| 390 |
size_t WorkSize[1] = {ParallelRate}; // one dimensional Range |
| 391 |
|
| 392 |
uint64_t HostInside[ParallelRate]; |
| 393 |
|
| 394 |
for (uint32_t loop=0;loop<Loops;loop++) { |
| 395 |
// Set start timer
|
| 396 |
gettimeofday(&tv1, NULL);
|
| 397 |
|
| 398 |
// Execute the OpenCL kernel with datas
|
| 399 |
clEnqueueNDRangeKernel(cqCommandQueue, OpenCLMainLoopGlobal, 1, NULL, |
| 400 |
WorkSize, NULL, 0, NULL, NULL); |
| 401 |
// Copy each result for each PR from Device to Host
|
| 402 |
clEnqueueReadBuffer(cqCommandQueue, GPUInside, CL_TRUE, 0,
|
| 403 |
ParallelRate * sizeof(uint64_t), HostInside, 0, NULL, NULL); |
| 404 |
uint64_t inside=0;
|
| 405 |
|
| 406 |
for (int i= 0; i < ParallelRate; i++) { |
| 407 |
inside+=HostInside[i]; |
| 408 |
} |
| 409 |
|
| 410 |
// Set stop timer
|
| 411 |
gettimeofday(&tv2, NULL);
|
| 412 |
|
| 413 |
double elapsed=(double)((tv2.tv_sec-tv1.tv_sec) * 1000000L + |
| 414 |
(tv2.tv_usec-tv1.tv_usec))/1000000;
|
| 415 |
|
| 416 |
double itops=(double)(ParallelRate*IterationsEach)/elapsed; |
| 417 |
|
| 418 |
printf("Inside/Total %ld %ld\nParallelRate %i\nElapsed Time %.2f\nItops %.0f\nLogItops %.2f\nPi estimation %f\n\n",inside,ParallelRate*IterationsEach,ParallelRate,elapsed,itops,log10(itops),(4.*(float)inside/((float)(ParallelRate)*(float)(IterationsEach)))); |
| 419 |
} |
| 420 |
printf("\n\n");
|
| 421 |
|
| 422 |
clReleaseKernel(OpenCLMainLoopGlobal); |
| 423 |
clReleaseProgram(OpenCLProgram); |
| 424 |
clReleaseCommandQueue(cqCommandQueue); |
| 425 |
clReleaseContext(GPUContext); |
| 426 |
clReleaseMemObject(GPUInside); |
| 427 |
|
| 428 |
|
| 429 |
return 0; |
| 430 |
} |
| 431 |
} |
| 432 |
|