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