Skip to content

Update "hello world" program plus tweak tools #1

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
mohawk2 wants to merge 10 commits into
base: master
Choose a base branch
from
Open

Update "hello world" program plus tweak tools #1

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
f77a3aa
return failure, not abort
mohawk2 Sep 24, 2024
0fae1e4
get right header on non-Apple
mohawk2 Sep 24, 2024
3c25de0
use non-deprecated clCreateCommandQueueWithProperties
mohawk2 Sep 24, 2024
f2a9281
CL error strings
mohawk2 Sep 24, 2024
0a35d96
ignore executable
mohawk2 Sep 24, 2024
4395803
add how to run
mohawk2 Sep 24, 2024
eceba9e
use `sin` not `*` so can get CPU usage up to prove is pthreading
mohawk2 Sep 24, 2024
083bafa
make command-line dictate GPU use and nitems
mohawk2 Sep 24, 2024
3fec627
no ask for bigger workgroup than have items, ask device for workgroup…
mohawk2 Sep 24, 2024
d0c07b5
fopen in binary mode for Windows
mohawk2 Sep 25, 2024
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions OpenCL_Hello_World_Example/.gitignore
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
hello
173 changes: 153 additions & 20 deletions OpenCL_Hello_World_Example/hello.c
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,3 +1,5 @@
// build and run with:
// (cd OpenCL_Hello_World_Example/; cc -o hello hello.c -lOpenCL -lm && ./hello y 1024)
//
// File: hello.c
//
Expand Down Expand Up @@ -50,6 +52,8 @@

////////////////////////////////////////////////////////////////////////////////

#define CL_TARGET_OPENCL_VERSION 200 /* clCreateCommandQueueWithProperties */

#include <fcntl.h>
#include <stdio.h>
#include <stdlib.h>
Expand All @@ -58,13 +62,11 @@
#include <unistd.h>
#include <sys/types.h>
#include <sys/stat.h>
#ifdef __APPLE__
#include <OpenCL/opencl.h>

////////////////////////////////////////////////////////////////////////////////

// Use a static data size for simplicity
//
#define DATA_SIZE (1024)
#else
#include <CL/opencl.h>
#endif

////////////////////////////////////////////////////////////////////////////////

Expand All @@ -77,19 +79,129 @@ const char *KernelSource = "\n" \
" const unsigned int count) \n" \
"{ \n" \
" int i = get_global_id(0); \n" \
" if(i < count) \n" \
" output[i] = input[i] * input[i]; \n" \
" if (i < count) \n" \
" output[i] = sin(input[i]); \n" \
"} \n" \
"\n";

const char* clGetErrorString(int errorCode) {
switch (errorCode) {
case 0: return "CL_SUCCESS";
case -1: return "CL_DEVICE_NOT_FOUND";
case -2: return "CL_DEVICE_NOT_AVAILABLE";
case -3: return "CL_COMPILER_NOT_AVAILABLE";
case -4: return "CL_MEM_OBJECT_ALLOCATION_FAILURE";
case -5: return "CL_OUT_OF_RESOURCES";
case -6: return "CL_OUT_OF_HOST_MEMORY";
case -7: return "CL_PROFILING_INFO_NOT_AVAILABLE";
case -8: return "CL_MEM_COPY_OVERLAP";
case -9: return "CL_IMAGE_FORMAT_MISMATCH";
case -10: return "CL_IMAGE_FORMAT_NOT_SUPPORTED";
case -12: return "CL_MAP_FAILURE";
case -13: return "CL_MISALIGNED_SUB_BUFFER_OFFSET";
case -14: return "CL_EXEC_STATUS_ERROR_FOR_EVENTS_IN_WAIT_LIST";
case -15: return "CL_COMPILE_PROGRAM_FAILURE";
case -16: return "CL_LINKER_NOT_AVAILABLE";
case -17: return "CL_LINK_PROGRAM_FAILURE";
case -18: return "CL_DEVICE_PARTITION_FAILED";
case -19: return "CL_KERNEL_ARG_INFO_NOT_AVAILABLE";
case -30: return "CL_INVALID_VALUE";
case -31: return "CL_INVALID_DEVICE_TYPE";
case -32: return "CL_INVALID_PLATFORM";
case -33: return "CL_INVALID_DEVICE";
case -34: return "CL_INVALID_CONTEXT";
case -35: return "CL_INVALID_QUEUE_PROPERTIES";
case -36: return "CL_INVALID_COMMAND_QUEUE";
case -37: return "CL_INVALID_HOST_PTR";
case -38: return "CL_INVALID_MEM_OBJECT";
case -39: return "CL_INVALID_IMAGE_FORMAT_DESCRIPTOR";
case -40: return "CL_INVALID_IMAGE_SIZE";
case -41: return "CL_INVALID_SAMPLER";
case -42: return "CL_INVALID_BINARY";
case -43: return "CL_INVALID_BUILD_OPTIONS";
case -44: return "CL_INVALID_PROGRAM";
case -45: return "CL_INVALID_PROGRAM_EXECUTABLE";
case -46: return "CL_INVALID_KERNEL_NAME";
case -47: return "CL_INVALID_KERNEL_DEFINITION";
case -48: return "CL_INVALID_KERNEL";
case -49: return "CL_INVALID_ARG_INDEX";
case -50: return "CL_INVALID_ARG_VALUE";
case -51: return "CL_INVALID_ARG_SIZE";
case -52: return "CL_INVALID_KERNEL_ARGS";
case -53: return "CL_INVALID_WORK_DIMENSION";
case -54: return "CL_INVALID_WORK_GROUP_SIZE";
case -55: return "CL_INVALID_WORK_ITEM_SIZE";
case -56: return "CL_INVALID_GLOBAL_OFFSET";
case -57: return "CL_INVALID_EVENT_WAIT_LIST";
case -58: return "CL_INVALID_EVENT";
case -59: return "CL_INVALID_OPERATION";
case -60: return "CL_INVALID_GL_OBJECT";
case -61: return "CL_INVALID_BUFFER_SIZE";
case -62: return "CL_INVALID_MIP_LEVEL";
case -63: return "CL_INVALID_GLOBAL_WORK_SIZE";
case -64: return "CL_INVALID_PROPERTY";
case -65: return "CL_INVALID_IMAGE_DESCRIPTOR";
case -66: return "CL_INVALID_COMPILER_OPTIONS";
case -67: return "CL_INVALID_LINKER_OPTIONS";
case -68: return "CL_INVALID_DEVICE_PARTITION_COUNT";
case -69: return "CL_INVALID_PIPE_SIZE";
case -70: return "CL_INVALID_DEVICE_QUEUE";
case -71: return "CL_INVALID_SPEC_ID";
case -72: return "CL_MAX_SIZE_RESTRICTION_EXCEEDED";
case -1002: return "CL_INVALID_D3D10_DEVICE_KHR";
case -1003: return "CL_INVALID_D3D10_RESOURCE_KHR";
case -1004: return "CL_D3D10_RESOURCE_ALREADY_ACQUIRED_KHR";
case -1005: return "CL_D3D10_RESOURCE_NOT_ACQUIRED_KHR";
case -1006: return "CL_INVALID_D3D11_DEVICE_KHR";
case -1007: return "CL_INVALID_D3D11_RESOURCE_KHR";
case -1008: return "CL_D3D11_RESOURCE_ALREADY_ACQUIRED_KHR";
case -1009: return "CL_D3D11_RESOURCE_NOT_ACQUIRED_KHR";
case -1010: return "CL_INVALID_DX9_MEDIA_ADAPTER_KHR";
case -1011: return "CL_INVALID_DX9_MEDIA_SURFACE_KHR";
case -1012: return "CL_DX9_MEDIA_SURFACE_ALREADY_ACQUIRED_KHR";
case -1013: return "CL_DX9_MEDIA_SURFACE_NOT_ACQUIRED_KHR";
case -1093: return "CL_INVALID_EGL_OBJECT_KHR";
case -1092: return "CL_EGL_RESOURCE_NOT_ACQUIRED_KHR";
case -1001: return "CL_PLATFORM_NOT_FOUND_KHR";
case -1057: return "CL_DEVICE_PARTITION_FAILED_EXT";
case -1058: return "CL_INVALID_PARTITION_COUNT_EXT";
case -1059: return "CL_INVALID_PARTITION_NAME_EXT";
case -1094: return "CL_INVALID_ACCELERATOR_INTEL";
case -1095: return "CL_INVALID_ACCELERATOR_TYPE_INTEL";
case -1096: return "CL_INVALID_ACCELERATOR_DESCRIPTOR_INTEL";
case -1097: return "CL_ACCELERATOR_TYPE_NOT_SUPPORTED_INTEL";
case -1000: return "CL_INVALID_GL_SHAREGROUP_REFERENCE_KHR";
case -1098: return "CL_INVALID_VA_API_MEDIA_ADAPTER_INTEL";
case -1099: return "CL_INVALID_VA_API_MEDIA_SURFACE_INTEL";
case -1100: return "CL_VA_API_MEDIA_SURFACE_ALREADY_ACQUIRED_INTEL";
case -1101: return "CL_VA_API_MEDIA_SURFACE_NOT_ACQUIRED_INTEL";
default: return "CL_UNKNOWN_ERROR";
}
}

////////////////////////////////////////////////////////////////////////////////

int main(int argc, char** argv)
{
if (argc != 3) {
fprintf(stderr, "Usage: %s <useGPU=y/n> <nitems>\n", argv[0]);
return 1;
}
const int useGPU = argv[1][0] == 'y';
const int nitems = atoi(argv[2]);

int err; // error code returned from api calls

float data[DATA_SIZE]; // original data set given to device
float results[DATA_SIZE]; // results returned from device
float *data = malloc(sizeof(float) * nitems); // original data set given to device
if (!data) {
printf("Error: Failed to allocate data\n");
exit(1);
}
float *results = malloc(sizeof(float) * nitems); // results returned from device
if (!results) {
printf("Error: Failed to allocate results\n");
exit(1);
}
unsigned int correct; // number of correct results returned

size_t global; // global domain size for our calculation
Expand All @@ -107,14 +219,13 @@ int main(int argc, char** argv)
// Fill our data set with random float values
//
int i = 0;
unsigned int count = DATA_SIZE;
for(i = 0; i < count; i++)
unsigned int count = nitems;
for (i = 0; i < count; i++)
data[i] = rand() / (float)RAND_MAX;

// Connect to a compute device
//
int gpu = 1;
err = clGetDeviceIDs(NULL, gpu ? CL_DEVICE_TYPE_GPU : CL_DEVICE_TYPE_CPU, 1, &device_id, NULL);
err = clGetDeviceIDs(NULL, useGPU ? CL_DEVICE_TYPE_GPU : CL_DEVICE_TYPE_CPU, 1, &device_id, NULL);
if (err != CL_SUCCESS)
{
printf("Error: Failed to create a device group!\n");
Expand All @@ -132,10 +243,10 @@ int main(int argc, char** argv)

// Create a command commands
//
commands = clCreateCommandQueue(context, device_id, 0, &err);
commands = clCreateCommandQueueWithProperties(context, device_id, NULL, &err);
if (!commands)
{
printf("Error: Failed to create a command commands!\n");
printf("Error: Failed to create a command queue: %s\n", clGetErrorString(err));
return EXIT_FAILURE;
}

Expand Down Expand Up @@ -211,14 +322,35 @@ int main(int argc, char** argv)
exit(1);
}

cl_uint max_wi_ndims;
err = clGetDeviceInfo(device_id, CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS, sizeof(max_wi_ndims), &max_wi_ndims, NULL);
if (err)
{
printf("Error: Failed to get CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS: %s\n", clGetErrorString(err));
return EXIT_FAILURE;
}

size_t max_wi_dims[max_wi_ndims];
err = clGetDeviceInfo(device_id, CL_DEVICE_MAX_WORK_ITEM_SIZES, sizeof(max_wi_dims), max_wi_dims, NULL);
if (err)
{
printf("Error: Failed to get CL_DEVICE_MAX_WORK_ITEM_SIZES: %s\n", clGetErrorString(err));
return EXIT_FAILURE;
}

if (local > count) /* don't ask for more than all */
local = count;
if (local > max_wi_dims[0]) /* don't ask for more than max dim */
local = max_wi_dims[0];

// Execute the kernel over the entire range of our 1d input data set
// using the maximum number of work group items for this device
//
global = count;
err = clEnqueueNDRangeKernel(commands, kernel, 1, NULL, &global, &local, 0, NULL, NULL);
if (err)
{
printf("Error: Failed to execute kernel!\n");
printf("Error: Failed to execute kernel: %s\n", clGetErrorString(err));
return EXIT_FAILURE;
}

Expand All @@ -238,9 +370,9 @@ int main(int argc, char** argv)
// Validate our results
//
correct = 0;
for(i = 0; i < count; i++)
for (i = 0; i < count; i++)
{
if(results[i] == data[i] * data[i])
if (fabsf(results[i] - sin(data[i])) < 1e-6)
correct++;
}

Expand All @@ -256,7 +388,8 @@ int main(int argc, char** argv)
clReleaseKernel(kernel);
clReleaseCommandQueue(commands);
clReleaseContext(context);
free(data);
free(results);

return 0;
}

29 changes: 19 additions & 10 deletions tools-master/cl-demo.c
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,42 +1,51 @@
#define CL_TARGET_OPENCL_VERSION 300

#include "timing.h"
#include "cl-helper.h"


#define KERNEL_FILE "vec-add-soln.cl"


int main(int argc, char **argv)
{
if (argc != 3)
{
fprintf(stderr, "need two arguments!\n");
abort();
fprintf(stderr, "Usage: %s <nitems> <ntrips>\n", argv[0]);
return 1;
}

const cl_long n = atol(argv[1]);
const int ntrips = atoi(argv[2]);

cl_context ctx;
cl_command_queue queue;
create_context_on(CHOOSE_INTERACTIVELY, CHOOSE_INTERACTIVELY, 0, &ctx, &queue, 0);
if (!create_context_on(CHOOSE_INTERACTIVELY, CHOOSE_INTERACTIVELY, 0, &ctx, &queue, 0))
return 1;

print_device_info_from_queue(queue);
if (!print_device_info_from_queue(queue))
return 1;

// --------------------------------------------------------------------------
// load kernels
// load kernels
// --------------------------------------------------------------------------
char *knl_text = read_file("vec-add-soln.cl");
char *knl_text = read_file(KERNEL_FILE);
if (!knl_text) {
fprintf(stderr, "error with '%s'\n", KERNEL_FILE);
return 1;
}
cl_kernel knl = kernel_from_string(ctx, knl_text, "sum", NULL);
free(knl_text);

// --------------------------------------------------------------------------
// allocate and initialize CPU memory
// --------------------------------------------------------------------------
float *a = (float *) malloc(sizeof(float) * n);
if (!a) { perror("alloc x"); abort(); }
if (!a) { perror("alloc x"); return 1; }
float *b = (float *) malloc(sizeof(float) * n);
if (!b) { perror("alloc y"); abort(); }
if (!b) { perror("alloc y"); return 1; }
float *c = (float *) malloc(sizeof(float) * n);
if (!c) { perror("alloc z"); abort(); }
if (!c) { perror("alloc z"); return 1; }

for (size_t i = 0; i < n; ++i)
{
Expand Down Expand Up @@ -113,7 +122,7 @@ int main(int argc, char **argv)
if (c[i] != 3*i)
{
printf("BAD %ld %f %f!\n", i, c[i], c[i] - 3*i);
abort();
return 1;
}
puts("GOOD");

Expand Down
Loading