AR관련 테스트 잔행.
몇가지 샘플을 돌려보니 마커를 인식하는 카메라 영상이 중요합니다.
증강 현실의 재현에 상당히 영향을 줍니다.
결국 카메라 튜닝이 관건...
| OpenCL : Using work item to partition data (5) (1) | 2014.06.01 |
|---|---|
| OpenCL::Query OpenCL kernel (4) (0) | 2014.05.02 |
| OpenCL Test Program (3) (0) | 2014.03.31 |
| OpenCL Test Program (2) (0) | 2014.03.18 |
| OpenCL test program (1) (0) | 2014.03.15 |
1차원 데이터 열에 대한 커널의 실행은
cl_int clEnqueueNDRangeKernel ( | cl_command_queue command_queue, |
| cl_kernel kernel, | |
| cl_uint work_dim, | |
| const size_t *global_work_offset, | |
| const size_t *global_work_size, | |
| const size_t *local_work_size, | |
| cl_uint num_events_in_wait_list, | |
| const cl_event *event_wait_list, | |
cl_event *event) |
cl_event evt;
error = clEnqueueNDRangeKernel(cQ,
kernels[j],
2,
0,
globalThreads,
localThreads,
0,
#define C(x,y) C[(x)* WIDTH + (y)]
| 증강 현실 테스트 (0) | 2014.10.04 |
|---|---|
| OpenCL::Query OpenCL kernel (4) (0) | 2014.05.02 |
| OpenCL Test Program (3) (0) | 2014.03.31 |
| OpenCL Test Program (2) (0) | 2014.03.18 |
| OpenCL test program (1) (0) | 2014.03.15 |
이 프로그램은 OpenCL 코드를 불러들여서 컴파일하여서 kernel을 만드는데 까지 진행한다.
실행은 아직이다.
우선 크게 보면 main program에서 platform id를 확인하고
platform id로 context를 만든다.
만들어진 context에 OpenCL 프로그램을 불러들여서 build를 하고 kernel로 준비해 준다.
그리고 해당 kernel의 정보를 받아와서 프린트 해 주는 데까지 진행한다.
OpenCL은 Runtime compile을 한다. 따라서 미리미리 컴파일을 해 고 진행하여야 한다.
이 예제에서는 Program과 Kernel을 만드는데 까지 진행한 예제이다.
(1) Platform ID의 개수를 확인한다.
(2) 확인된 Platform ID의 개수 만큼 공간을 확보한다.
(3) 확보된 메모리 공간에 Platform ID를 받아온다.
(4) Platform ID개수만큼
1) Device ID중 GPU 인것을 받아온다.
2) GPU인 것이 없으면 CPU 인 것을 받아온다.
3) Device ID에 Context를 생성한다.
4) OpenCL 프로그램 소스를 로 드
5) 로드된 프로그램소스를 빌드하여서 프로그램 화 한다.
이 단계는 CreateProgramWithSource -> BuildProgram 을 거치는 2단계로 구성된다.
6) 빌드 (컴파일)에서 문제가 생기면 그 정보(LOG)를 받아서 프린트 해준다.
(5) clCreateKernelsInProgram 프로그램을 가지고 커널을 만든다.
마찬가지로 2단계로 한다.
우선 커널의 개수를 확인하고
확인된 커널의 개수만큼 메모리를 할당받고
해당 할당받은 메모리에다가 커널을 빌드한다.
(6) 확인된 커널의 개수만큼
1) 커널의 이름을 받아오고
2) 인자의 개수를 받아온다.
3) 받아온 커널의 이름과 인자의 개수를 프린트한다.
복잡하지만 줄여보면 다음과 같다.
기본 프레임은
(1) 필요한 요소의 개수를 물어보고
(2) 그만큼 메모리 공간을 할당 받고
(3) 할당받은 메모리 공간에 요소를 받아온다.
이다.
이것을
Platform ID -> Device ID -> context -> Kernel
순으로 진행한다.
#include <stdio.h>
#include <stdlib.h>
#include <sys/types.h>
#include <alloca.h>
#ifdef __APPLE__
#include <OpenCL/cl.h>
#else
#include <CL/cl.h>
#endif
/*
프로그램을 읽어 들인다.
OpenCL은 On-The-Fly Compile을 수행한다.
*/
void loadProgramSource(const char** files,
size_t length,
char** buffer,
size_t* sizes) {
/* Read each source file (*.cl) and store the contents into a temporary datastore */
for(size_t i=0; i < length; i++) {
FILE* file = fopen(files[i], "r");
if(file == NULL) {
perror("Couldn't read the program file");
exit(1);
}
/* 파일의 끝까지 진행하여서 크기를 확인하고 다시 처음으로 돌아온다.*/
fseek(file, 0, SEEK_END);
sizes[i] = ftell(file);
rewind(file); // reset the file pointer so that 'fread' reads from the front
/* 크기 +1 만큼 할당한다음 */
buffer[i] = (char*)malloc(sizes[i]+1);
buffer[i][sizes[i]] = '\0';
/* 한방에 읽어 들이다. */
fread(buffer[i], sizeof(char), sizes[i], file);
fclose(file);
}
}
int main(int argc, char** argv) {
/* OpenCL 1.1 data structures */
cl_platform_id* platforms;
cl_program program;
cl_device_id device;
cl_context context;
/* OpenCL 1.1 scalar data types */
cl_uint numOfPlatforms;
cl_int error;
/*
Get the number of platforms
Remember that for each vendor's SDK installed on the computer,
the number of available platform also increased.
*/
실행 결과는 아래와 같다.
Number of OpenCL platforms found: 1
Kernel name: simpleAdd with arity: 3
Kernel name: simpleAdd_2 with arity: 3
Program ended with exit code: 0
찾아낸 플랫폼의 개수는 1개이고 여기서 포함되는 커널의 개수는 2개이다.
각각
simpleAdd
simpleAdd_2
가 된다.
OpenCL source 코드는 아래와 같이 되어 있고 여기서는 아직 중요한 포인트가 아니므로 그냥 넘어간다.
simple.cl
__kernel void simpleAdd(__global float *a,
__global float *b,
__global float *c) {
*c = *a + *b;
}
simple_2.cl
__kernel void simpleAdd_2(__global float *a,
__global float *b,
__global float *c) {
*c = *a + *b;
}
| 증강 현실 테스트 (0) | 2014.10.04 |
|---|---|
| OpenCL : Using work item to partition data (5) (1) | 2014.06.01 |
| OpenCL Test Program (3) (0) | 2014.03.31 |
| OpenCL Test Program (2) (0) | 2014.03.18 |
| OpenCL test program (1) (0) | 2014.03.15 |
디바이스 별로 고유한 확장 기능을 부여할 수 있다.
이런 기능이 추가되어 있는지 확인을 한다.
추가되는 함수는 아래에 표시되어 있다.
void displayDeviceInfo( cl_platform_id id ,
cl_device_type dev_type)
......
displayDeviceDetails( devices[i], CL_DEVICE_EXTENSIONS, "CL_DEVICE_EXTENSIONS" );
.....
void displayDeviceDetails( cl_device_id id,
cl_device_info param_name,
const char * paramNameAsStr)
...
switch(param_name)
{
......
break;
case CL_DEVICE_EXTENSIONS : {
// beware of buffer overflow; alternatively use the OpenCL C++ bindings
char* extension_info[4096];
error = clGetDeviceInfo( id, CL_DEVICE_EXTENSIONS, sizeof(extension_info),
extension_info, NULL);
printf("\tSupported extensions: %s\n", extension_info);
}break;
}
....
확장에 대한 정보를 확인하기 위해서 CL_DEVICE_EXTENSONS를 확인하면 된다.
세부 정보에서 해당 정보를 프린트하도록 한다.
완성된 코드와 실행 결과는 아래와 같다.
//
// main.cpp
// TestOpenCL
//
// Created by freegear on 2014. 2. 8..
// Copyright (c) 2014년 freegear. All rights reserved.
//
#include <iostream>
#include <stdio.h>
#include <stdlib.h>
#ifdef __APPLE__
# include <OpenCL/opencl.h>
#else
# include <CL/cl.h>
#endif
//Function proto type
void displayDeviceDetails(cl_device_id id ,
cl_device_info param_name,
const char * paramNameAsStr);
void displayDeviceInfo( cl_platform_id id ,
cl_device_type dev_type)
{
/* OpenCL 1.1 device type */
cl_int error = 0 ;
cl_uint numOfDevices = 0 ;
/* Determine how many devices are connected to your platform */
error = clGetDeviceIDs(id, dev_type, 0, NULL, &numOfDevices);
if( error != CL_SUCCESS){
perror("Unable to obtain any OpenCL compliant device info");
exit(1);
}
cl_device_id * devices = (cl_device_id*)alloca(sizeof(cl_device_id) * numOfDevices);
/* Load the information about your devices into the variable 'devices' */
error = clGetDeviceIDs(id, dev_type, numOfDevices, devices, NULL);
if (error != CL_SUCCESS)
{
perror("Unable to obtain any OpenCL compliant device info");
exit(1);
}
printf("Numnber of detected OpenCL devices %d\n", numOfDevices);
/* we attempt to retrieve some information about the devices */
for( int i = 0 ; i < numOfDevices ; ++i)
{
displayDeviceDetails(devices[i], CL_DEVICE_TYPE, "CL_DEVICE_TYPE");
displayDeviceDetails(devices[i], CL_DEVICE_VENDOR_ID, "CL_DEVICE_VENDOR_ID");
displayDeviceDetails(devices[i], CL_DEVICE_MAX_COMPUTE_UNITS, "CL_DEVICE_MAX_COMPUTE_UNITS");
displayDeviceDetails(devices[i], CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS, "CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS");
displayDeviceDetails(devices[i], CL_DEVICE_MAX_WORK_ITEM_SIZES, "CL_DEVICE_MAX_WORK_ITEM_SIZES");
displayDeviceDetails(devices[i], CL_DEVICE_MAX_WORK_GROUP_SIZE, "CL_DEVICE_MAX_WORK_GROUP_SIZE");
displayDeviceDetails( devices[i], CL_DEVICE_EXTENSIONS, "CL_DEVICE_EXTENSIONS" );
}
}
void displayDeviceDetails( cl_device_id id,
cl_device_info param_name,
const char * paramNameAsStr)
{
cl_int error = 0 ;
size_t paramSize = 0 ;
error = clGetDeviceInfo( id ,param_name, 0 , NULL, ¶mSize);
if(error != CL_SUCCESS)
{
perror("Unable to obtain device info for param");
return;
}
/*
The cl_device_info are preporcessor directives define in cl.h
*/
switch(param_name)
{
case CL_DEVICE_TYPE :
{
cl_device_type * devType = (cl_device_type*)alloca(sizeof(cl_device_type)*paramSize);
error = clGetDeviceInfo(id, param_name, paramSize, devType, NULL);
if(error != CL_SUCCESS)
{
perror("Unable to optain device info for param\n");
return;
}
switch( *devType)
{
case CL_DEVICE_TYPE_CPU : printf("CPU Detected \n"); break;
case CL_DEVICE_TYPE_GPU : printf("GPU Detected \n"); break;
case CL_DEVICE_TYPE_ACCELERATOR : printf("Accelerator detected\n");break;
case CL_DEVICE_TYPE_DEFAULT : printf("default detected \n"); break;
}
break;
}
case CL_DEVICE_VENDOR_ID :
case CL_DEVICE_MAX_COMPUTE_UNITS :
case CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS : {
cl_uint* ret = (cl_uint*) alloca(sizeof(cl_uint) * paramSize);
error = clGetDeviceInfo( id, param_name, paramSize, ret, NULL );
if (error != CL_SUCCESS ) {
perror("Unable to obtain device info for param\n");
return;
}
switch (param_name) {
case CL_DEVICE_VENDOR_ID: printf("\tVENDOR ID: 0x%x\n", *ret); break;
case CL_DEVICE_MAX_COMPUTE_UNITS: printf("\tMaximum number of parallel compute units: %d\n", *ret); break;
case CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS: printf("\tMaximum dimensions for global/local work-item IDs: %d\n", *ret); break;
}
}
break;
case CL_DEVICE_MAX_WORK_ITEM_SIZES :
{
cl_uint maxWIDimensions;
size_t* ret = (size_t*) alloca(sizeof(size_t) * paramSize);
error = clGetDeviceInfo( id, param_name, paramSize, ret, NULL );
error = clGetDeviceInfo( id, CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS, sizeof(cl_uint), &maxWIDimensions, NULL );
if (error != CL_SUCCESS ) {
perror("Unable to obtain device info for param\n");
return;
}
printf("\tMaximum number of work-items in each dimension: ( ");
for(cl_int i =0; i < maxWIDimensions; ++i ) {
printf("%d ", (int)ret[i]);
}
printf(" )\n");
}break;
case CL_DEVICE_MAX_WORK_GROUP_SIZE :
{
size_t* ret = (size_t*) alloca(sizeof(size_t) * paramSize);
error = clGetDeviceInfo( id, param_name, paramSize, ret, NULL );
if (error != CL_SUCCESS ) {
perror("Unable to obtain device info for param\n");
return;
}
printf("\tMaximum number of work-items in a work-group: %d\n", (int)*ret);
}
break;
case CL_DEVICE_NAME :
case CL_DEVICE_VENDOR :
{
char data[48];
error = clGetDeviceInfo( id, param_name, paramSize, data, NULL );
if (error != CL_SUCCESS ) {
perror("Unable to obtain device name/vendor info for param\n");
return;
}
switch (param_name) {
case CL_DEVICE_NAME : printf("\tDevice name is %s\n", data);break;
case CL_DEVICE_VENDOR : printf("\tDevice vendor is %s\n", data);break;
}
}
break;
case CL_DEVICE_GLOBAL_MEM_CACHELINE_SIZE: {
cl_uint* size = (cl_uint*) alloca(sizeof(cl_uint) * paramSize);
error = clGetDeviceInfo( id, param_name, paramSize, size, NULL );
if (error != CL_SUCCESS ) {
perror("Unable to obtain device name/vendor info for param\n");
return;
}
printf("\tDevice global cacheline size: %d bytes\n", (*size)); break;
} break;
case CL_DEVICE_GLOBAL_MEM_SIZE:
case CL_DEVICE_MAX_MEM_ALLOC_SIZE:
{
cl_ulong* size = (cl_ulong*) alloca(sizeof(cl_ulong) * paramSize);
error = clGetDeviceInfo( id, param_name, paramSize, size, NULL );
if (error != CL_SUCCESS ) {
perror("Unable to obtain device name/vendor info for param\n");
return;
}
switch (param_name) {
case CL_DEVICE_GLOBAL_MEM_SIZE: printf("\tDevice global mem: %ld mega-bytes\n", (*size)>>20); break;
case CL_DEVICE_MAX_MEM_ALLOC_SIZE: printf("\tDevice max memory allocation: %ld mega-bytes\n", (*size)>>20); break;
}
}
break;
case CL_DEVICE_EXTENSIONS : {
// beware of buffer overflow; alternatively use the OpenCL C++ bindings
char* extension_info[4096];
error = clGetDeviceInfo( id, CL_DEVICE_EXTENSIONS, sizeof(extension_info), extension_info, NULL);
printf("\tSupported extensions: %s\n", extension_info);
}break;
}//end of switch(param_name
}// end of displayDeviceDetailes
void displayPlatformInfo(
cl_platform_id id ,
cl_platform_info param_name,
const char * paramNameAsStr
)
{
cl_int error = 0 ;
size_t paramSize = 0 ;
error = clGetPlatformInfo(
id,
param_name,
0,
NULL,
¶mSize
);
char * moreInfo = (char *) alloca(sizeof(char)*paramSize);
error = clGetPlatformInfo(
id, /* The platform ID returned by clGetPlatformIDs or can be NULL.
If platform is NULL, the behavior is implementation-defined.*/
param_name, /* An enumeration constant that identifies the platform
information being queried. It can be one of the following values
as specified in the table below. */
paramSize, /* Specifies the size in bytes of memory pointed to by param_value.
This size in bytes must be greater than or equal to size of return type
specified in the table below. */
moreInfo, /* A pointer to memory location where appropriate values for a given
param_value will be returned. Acceptable param_value values are listed in
the table below. If param_value is NULL, it is ignored.*/
NULL /* Returns the actual size in bytes of data being queried by param_value.
If param_value_size_ret is NULL, it is ignored */
);
if( error != CL_SUCCESS)
{
if ( error == CL_INVALID_PLATFORM)
perror("Error CL_INVALID_PLATFORM");
else if (error == CL_INVALID_VALUE)
perror("Error CL_INVALID_VALUE");
else
perror("Unable to find any OpenCL platform information");
return ;
}
printf("%s : %s\n" , paramNameAsStr, moreInfo);
} // end of displayPlatformInfo
int main(int argc, const char * argv[])
{
// OpenCL 1.2 Data struction
cl_platform_id * platforms ;
/* OpenCL 1.1 scalar data types */
cl_uint numOfPlatforms ;
cl_int error ;
/*
Get the number of platforms
Remember that for each vendor's SDK installed on the computer,
the number of available platform also increased.
*/
error = clGetPlatformIDs(0, /*
The number of cl_platform_id entries that can be added to platforms.
If platforms is not NULL, the num_entries must be greater than zero.
*/
NULL, /* Returns a list of OpenCL platforms found.
The cl_platform_id values returned in platforms can be used to identify
a specific OpenCL platform. If platforms argument is NULL,
this argument is ignored.
The number of OpenCL platforms returned is the mininum of the value
specified by num_entries or the number of OpenCL platforms available. */
&numOfPlatforms
/* Returns the number of OpenCL platforms available.
If num_platforms is NULL, this argument is ignored. */
);
if ( error < 0)
{
perror("Unable to find any OpenCL platforms");
exit(1);
}
// allocate memory for the number of installed platforms
// alloca(....) occupies some stack space but is
// automatically freed on return
platforms = (cl_platform_id*) alloca(sizeof(cl_platform_id)*numOfPlatforms);
printf("Number of OpenCL platforms found : %d\n", numOfPlatforms);
error = clGetPlatformIDs(numOfPlatforms, /*
The number of cl_platform_id entries that can be added to platforms.
If platforms is not NULL, the num_entries must be greater than zero.
*/
platforms, /* Returns a list of OpenCL platforms found.
The cl_platform_id values returned in platforms can be used to identify
a specific OpenCL platform. If platforms argument is NULL,
this argument is ignored.
The number of OpenCL platforms returned is the mininum of the value
specified by num_entries or the number of OpenCL platforms available. */
NULL
/* Returns the number of OpenCL platforms available.
If num_platforms is NULL, this argument is ignored. */
);
if ( error < 0)
{
perror("Unable to find any OpenCL platforms");
exit(1);
}
// We invoke the API 'clPlatformInfo' twice for each
// parameter we are trying to extract
// and we use the return value to create temporary data
// structure (on the stack) to store
// the returned information ot the second invocation.
for( cl_uint i = 0 ; i < numOfPlatforms ; ++i)
{
displayPlatformInfo( platforms[i], CL_PLATFORM_PROFILE , "CL_PLATFORM_PROFILE");
displayPlatformInfo( platforms[i], CL_PLATFORM_VERSION , "CL_PLATFORM_VERSION");
displayPlatformInfo( platforms[i], CL_PLATFORM_NAME , "CL_PLATFORM_NAME");
displayPlatformInfo( platforms[i], CL_PLATFORM_VENDOR , "CL_PLATFORM_VENDOR");
displayPlatformInfo( platforms[i], CL_PLATFORM_EXTENSIONS , "CL_PLATFORM_EXTENSIONS");
// Assume that we don't know how many devices are OpenCL compliant, we locate everything !
displayDeviceInfo( platforms[i], CL_DEVICE_TYPE_ALL );
}
return 0;
}
실행 결과
Number of OpenCL platforms found : 1
CL_PLATFORM_PROFILE : FULL_PROFILE
CL_PLATFORM_VERSION : OpenCL 1.2 (Aug 24 2013 21:03:27)
CL_PLATFORM_NAME : Apple
CL_PLATFORM_VENDOR : Apple
CL_PLATFORM_EXTENSIONS : cl_APPLE_SetMemObjectDestructor cl_APPLE_ContextLoggingFunctions cl_APPLE_clut cl_APPLE_query_kernel_names cl_APPLE_gl_sharing cl_khr_gl_event
Numnber of detected OpenCL devices 3
CPU Detected
VENDOR ID: 0xffffffff
Maximum number of parallel compute units: 8
Maximum dimensions for global/local work-item IDs: 3
Maximum number of work-items in each dimension: ( 1024 1 1 )
Maximum number of work-items in a work-group: 1024
Supported extensions: cl_APPLE_SetMemObjectDestructor cl_APPLE_ContextLoggingFunctions cl_APPLE_clut cl_APPLE_query_kernel_names cl_APPLE_gl_sharing cl_khr_gl_event cl_khr_fp64 cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_byte_addressable_store cl_khr_int64_base_atomics cl_khr_int64_extended_atomics cl_khr_3d_image_writes cl_khr_image2d_from_buffer cl_APPLE_fp64_basic_ops cl_APPLE_fixed_alpha_channel_orders cl_APPLE_biased_fixed_point_image_formats cl_APPLE_command_queue_priority
GPU Detected
VENDOR ID: 0x1022700
Maximum number of parallel compute units: 2
Maximum dimensions for global/local work-item IDs: 3
Maximum number of work-items in each dimension: ( 1024 1024 64 )
Maximum number of work-items in a work-group: 1024
Supported extensions: cl_APPLE_SetMemObjectDestructor cl_APPLE_ContextLoggingFunctions cl_APPLE_clut cl_APPLE_query_kernel_names cl_APPLE_gl_sharing cl_khr_gl_event cl_khr_byte_addressable_store cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_APPLE_fp64_basic_ops cl_khr_fp64 cl_khr_3d_image_writes cl_khr_depth_images cl_khr_gl_depth_images cl_khr_gl_msaa_sharing cl_khr_image2d_from_buffer
GPU Detected
VENDOR ID: 0x1024400
Maximum number of parallel compute units: 16
Maximum dimensions for global/local work-item IDs: 3
Maximum number of work-items in each dimension: ( 512 512 512 )
Maximum number of work-items in a work-group: 512
Supported extensions: cl_APPLE_SetMemObjectDestructor cl_APPLE_ContextLoggingFunctions cl_APPLE_clut cl_APPLE_query_kernel_names cl_APPLE_gl_sharing cl_khr_gl_event cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_byte_addressable_store cl_khr_image2d_from_buffer cl_khr_gl_depth_images cl_khr_depth_images
Program ended with exit code: 0
위에서 나온 용어에 대해서 다음페이지에서 설명한다.
| OpenCL : Using work item to partition data (5) (1) | 2014.06.01 |
|---|---|
| OpenCL::Query OpenCL kernel (4) (0) | 2014.05.02 |
| OpenCL Test Program (2) (0) | 2014.03.18 |
| OpenCL test program (1) (0) | 2014.03.15 |
| OpenCL Architecture (0) | 2014.03.15 |
Device에 대한 정보를 얻는 코드이다.
추가되는 함수는
displayDeviceDetails
displayDeviceInfo
이다.
이들 함수는 loop내에서 호출된다.
for( cl_uint i = 0 ; i < numOfPlatforms ; ++i)
{
displayPlatformInfo( platforms[i], CL_PLATFORM_PROFILE , "CL_PLATFORM_PROFILE");
displayPlatformInfo( platforms[i], CL_PLATFORM_VERSION , "CL_PLATFORM_VERSION");
.........
// Assume that we don't know how many devices are OpenCL compliant, we locate everything !
displayDeviceInfo( platforms[i], CL_DEVICE_TYPE_ALL );
}
플랫폼에서 얻어진 디바이스의 정보를 프린트해주는 함수를 추가하고
그 함수의 내부에서 디바이스의 개수를 확인 한 뒤에 개수만큼 Loop를 돌려서 정보를 프린트한다.
즉 플랫폼 1개당 디바이스의 개수는 복수개가 검출 될 수 있는 것이다.
.....
/* Determine how many devices are connected to your platform */
error = clGetDeviceIDs(id, dev_type, 0, NULL, &numOfDevices); <= 디바이스의 개수를 확인
......
error = clGetDeviceIDs(id, dev_type, numOfDevices, devices, NULL); <= 확인된 개수만큼 정보를 가지고 옴
....
/* we attempt to retrieve some information about the devices */
for( int i = 0 ; i < numOfDevices ; ++i)
{
displayDeviceDetails(devices[i], CL_DEVICE_TYPE, "CL_DEVICE_TYPE");
displayDeviceDetails(devices[i], CL_DEVICE_VENDOR_ID, "CL_DEVICE_VENDOR_ID");
..........
displayDeviceDetails(devices[i], CL_DEVICE_MAX_WORK_GROUP_SIZE, "CL_DEVICE_MAX_WORK_GROUP_SIZE");
}
//
// main.cpp
// TestOpenCL
//
// Created by freegear on 2014. 2. 8..
// Copyright (c) 2014년 freegear. All rights reserved.
//
#include <iostream>
#include <stdio.h>
#include <stdlib.h>
#ifdef __APPLE__
# include <OpenCL/opencl.h>
#else
# include <CL/cl.h>
#endif
//Function proto type
void displayDeviceDetails(cl_device_id id ,
cl_device_info param_name,
const char * paramNameAsStr);
void displayDeviceInfo( cl_platform_id id ,
cl_device_type dev_type)
{
/* OpenCL 1.1 device type */
cl_int error = 0 ;
cl_uint numOfDevices = 0 ;
/* Determine how many devices are connected to your platform */
error = clGetDeviceIDs(id, dev_type, 0, NULL, &numOfDevices);
if( error != CL_SUCCESS){
perror("Unable to obtain any OpenCL compliant device info");
exit(1);
}
cl_device_id * devices = (cl_device_id*)alloca(sizeof(cl_device_id) * numOfDevices);
/* Load the information about your devices into the variable 'devices' */
error = clGetDeviceIDs(id, dev_type, numOfDevices, devices, NULL);
if (error != CL_SUCCESS)
{
perror("Unable to obtain any OpenCL compliant device info");
exit(1);
}
printf("Numnber of detected OpenCL devices %d\n", numOfDevices);
/* we attempt to retrieve some information about the devices */
for( int i = 0 ; i < numOfDevices ; ++i)
{
displayDeviceDetails(devices[i], CL_DEVICE_TYPE, "CL_DEVICE_TYPE");
displayDeviceDetails(devices[i], CL_DEVICE_VENDOR_ID, "CL_DEVICE_VENDOR_ID");
displayDeviceDetails(devices[i], CL_DEVICE_MAX_COMPUTE_UNITS, "CL_DEVICE_MAX_COMPUTE_UNITS");
displayDeviceDetails(devices[i], CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS, "CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS");
displayDeviceDetails(devices[i], CL_DEVICE_MAX_WORK_ITEM_SIZES, "CL_DEVICE_MAX_WORK_ITEM_SIZES");
displayDeviceDetails(devices[i], CL_DEVICE_MAX_WORK_GROUP_SIZE, "CL_DEVICE_MAX_WORK_GROUP_SIZE");
}
}
void displayDeviceDetails( cl_device_id id,
cl_device_info param_name,
const char * paramNameAsStr)
{
cl_int error = 0 ;
size_t paramSize = 0 ;
error = clGetDeviceInfo( id ,param_name, 0 , NULL, ¶mSize);
if(error != CL_SUCCESS)
{
perror("Unable to obtain device info for param");
return;
}
/*
The cl_device_info are preporcessor directives define in cl.h
*/
switch(param_name)
{
case CL_DEVICE_TYPE :
{
cl_device_type * devType = (cl_device_type*)alloca(sizeof(cl_device_type)*paramSize);
error = clGetDeviceInfo(id, param_name, paramSize, devType, NULL);
if(error != CL_SUCCESS)
{
perror("Unable to optain device info for param\n");
return;
}
switch( *devType)
{
case CL_DEVICE_TYPE_CPU : printf("CPU Detected \n"); break;
case CL_DEVICE_TYPE_GPU : printf("GPU Detected \n"); break;
case CL_DEVICE_TYPE_ACCELERATOR : printf("Accelerator detected\n");break;
case CL_DEVICE_TYPE_DEFAULT : printf("default detected \n"); break;
}
break;
}
case CL_DEVICE_VENDOR_ID :
case CL_DEVICE_MAX_COMPUTE_UNITS :
case CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS : {
cl_uint* ret = (cl_uint*) alloca(sizeof(cl_uint) * paramSize);
error = clGetDeviceInfo( id, param_name, paramSize, ret, NULL );
if (error != CL_SUCCESS ) {
perror("Unable to obtain device info for param\n");
return;
}
switch (param_name) {
case CL_DEVICE_VENDOR_ID: printf("\tVENDOR ID: 0x%x\n", *ret); break;
case CL_DEVICE_MAX_COMPUTE_UNITS: printf("\tMaximum number of parallel compute units: %d\n", *ret); break;
case CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS: printf("\tMaximum dimensions for global/local work-item IDs: %d\n", *ret); break;
}
}
break;
case CL_DEVICE_MAX_WORK_ITEM_SIZES :
{
cl_uint maxWIDimensions;
size_t* ret = (size_t*) alloca(sizeof(size_t) * paramSize);
error = clGetDeviceInfo( id, param_name, paramSize, ret, NULL );
error = clGetDeviceInfo( id, CL_DEVICE_MAX_WORK_ITEM_DIMENSIONS, sizeof(cl_uint), &maxWIDimensions, NULL );
if (error != CL_SUCCESS ) {
perror("Unable to obtain device info for param\n");
return;
}
printf("\tMaximum number of work-items in each dimension: ( ");
for(cl_int i =0; i < maxWIDimensions; ++i ) {
printf("%d ", (int)ret[i]);
}
printf(" )\n");
}break;
case CL_DEVICE_MAX_WORK_GROUP_SIZE :
{
size_t* ret = (size_t*) alloca(sizeof(size_t) * paramSize);
error = clGetDeviceInfo( id, param_name, paramSize, ret, NULL );
if (error != CL_SUCCESS ) {
perror("Unable to obtain device info for param\n");
return;
}
printf("\tMaximum number of work-items in a work-group: %d\n", (int)*ret);
}
break;
case CL_DEVICE_NAME :
case CL_DEVICE_VENDOR :
{
char data[48];
error = clGetDeviceInfo( id, param_name, paramSize, data, NULL );
if (error != CL_SUCCESS ) {
perror("Unable to obtain device name/vendor info for param\n");
return;
}
switch (param_name) {
case CL_DEVICE_NAME : printf("\tDevice name is %s\n", data);break;
case CL_DEVICE_VENDOR : printf("\tDevice vendor is %s\n", data);break;
}
}
break;
case CL_DEVICE_GLOBAL_MEM_CACHELINE_SIZE: {
cl_uint* size = (cl_uint*) alloca(sizeof(cl_uint) * paramSize);
error = clGetDeviceInfo( id, param_name, paramSize, size, NULL );
if (error != CL_SUCCESS ) {
perror("Unable to obtain device name/vendor info for param\n");
return;
}
printf("\tDevice global cacheline size: %d bytes\n", (*size)); break;
} break;
case CL_DEVICE_GLOBAL_MEM_SIZE:
case CL_DEVICE_MAX_MEM_ALLOC_SIZE:
{
cl_ulong* size = (cl_ulong*) alloca(sizeof(cl_ulong) * paramSize);
error = clGetDeviceInfo( id, param_name, paramSize, size, NULL );
if (error != CL_SUCCESS ) {
perror("Unable to obtain device name/vendor info for param\n");
return;
}
switch (param_name) {
case CL_DEVICE_GLOBAL_MEM_SIZE: printf("\tDevice global mem: %ld mega-bytes\n", (*size)>>20); break;
case CL_DEVICE_MAX_MEM_ALLOC_SIZE: printf("\tDevice max memory allocation: %ld mega-bytes\n", (*size)>>20); break;
}
}
break;
}//end of switch(param_name
}// end of displayDeviceDetailes
void displayPlatformInfo(
cl_platform_id id ,
cl_platform_info param_name,
const char * paramNameAsStr
)
{
cl_int error = 0 ;
size_t paramSize = 0 ;
error = clGetPlatformInfo(
id,
param_name,
0,
NULL,
¶mSize
);
char * moreInfo = (char *) alloca(sizeof(char)*paramSize);
error = clGetPlatformInfo(
id, /* The platform ID returned by clGetPlatformIDs or can be NULL.
If platform is NULL, the behavior is implementation-defined.*/
param_name, /* An enumeration constant that identifies the platform
information being queried. It can be one of the following values
as specified in the table below. */
paramSize, /* Specifies the size in bytes of memory pointed to by param_value.
This size in bytes must be greater than or equal to size of return type
specified in the table below. */
moreInfo, /* A pointer to memory location where appropriate values for a given
param_value will be returned. Acceptable param_value values are listed in
the table below. If param_value is NULL, it is ignored.*/
NULL /* Returns the actual size in bytes of data being queried by param_value.
If param_value_size_ret is NULL, it is ignored */
);
if( error != CL_SUCCESS)
{
if ( error == CL_INVALID_PLATFORM)
perror("Error CL_INVALID_PLATFORM");
else if (error == CL_INVALID_VALUE)
perror("Error CL_INVALID_VALUE");
else
perror("Unable to find any OpenCL platform information");
return ;
}
printf("%s : %s\n" , paramNameAsStr, moreInfo);
} // end of displayPlatformInfo
int main(int argc, const char * argv[])
{
// OpenCL 1.2 Data struction
cl_platform_id * platforms ;
/* OpenCL 1.1 scalar data types */
cl_uint numOfPlatforms ;
cl_int error ;
/*
Get the number of platforms
Remember that for each vendor's SDK installed on the computer,
the number of available platform also increased.
*/
error = clGetPlatformIDs(0, /*
The number of cl_platform_id entries that can be added to platforms.
If platforms is not NULL, the num_entries must be greater than zero.
*/
NULL, /* Returns a list of OpenCL platforms found.
The cl_platform_id values returned in platforms can be used to identify
a specific OpenCL platform. If platforms argument is NULL,
this argument is ignored.
The number of OpenCL platforms returned is the mininum of the value
specified by num_entries or the number of OpenCL platforms available. */
&numOfPlatforms
/* Returns the number of OpenCL platforms available.
If num_platforms is NULL, this argument is ignored. */
);
if ( error < 0)
{
perror("Unable to find any OpenCL platforms");
exit(1);
}
// allocate memory for the number of installed platforms
// alloca(....) occupies some stack space but is
// automatically freed on return
platforms = (cl_platform_id*) alloca(sizeof(cl_platform_id)*numOfPlatforms);
printf("Number of OpenCL platforms found : %d\n", numOfPlatforms);
error = clGetPlatformIDs(numOfPlatforms, /*
The number of cl_platform_id entries that can be added to platforms.
If platforms is not NULL, the num_entries must be greater than zero.
*/
platforms, /* Returns a list of OpenCL platforms found.
The cl_platform_id values returned in platforms can be used to identify
a specific OpenCL platform. If platforms argument is NULL,
this argument is ignored.
The number of OpenCL platforms returned is the mininum of the value
specified by num_entries or the number of OpenCL platforms available. */
NULL
/* Returns the number of OpenCL platforms available.
If num_platforms is NULL, this argument is ignored. */
);
if ( error < 0)
{
perror("Unable to find any OpenCL platforms");
exit(1);
}
// We invoke the API 'clPlatformInfo' twice for each
// parameter we are trying to extract
// and we use the return value to create temporary data
// structure (on the stack) to store
// the returned information ot the second invocation.
for( cl_uint i = 0 ; i < numOfPlatforms ; ++i)
{
displayPlatformInfo( platforms[i], CL_PLATFORM_PROFILE , "CL_PLATFORM_PROFILE");
displayPlatformInfo( platforms[i], CL_PLATFORM_VERSION , "CL_PLATFORM_VERSION");
displayPlatformInfo( platforms[i], CL_PLATFORM_NAME , "CL_PLATFORM_NAME");
displayPlatformInfo( platforms[i], CL_PLATFORM_VENDOR , "CL_PLATFORM_VENDOR");
displayPlatformInfo( platforms[i], CL_PLATFORM_EXTENSIONS , "CL_PLATFORM_EXTENSIONS");
// Assume that we don't know how many devices are OpenCL compliant, we locate everything !
displayDeviceInfo( platforms[i], CL_DEVICE_TYPE_ALL );
}
return 0;
}
실행된 결과는 아래와 같다.
Number of OpenCL platforms found : 1
CL_PLATFORM_PROFILE : FULL_PROFILE
CL_PLATFORM_VERSION : OpenCL 1.2 (Aug 24 2013 21:03:27)
CL_PLATFORM_NAME : Apple
CL_PLATFORM_VENDOR : Apple
CL_PLATFORM_EXTENSIONS : cl_APPLE_SetMemObjectDestructor cl_APPLE_ContextLoggingFunctions cl_APPLE_clut cl_APPLE_query_kernel_names cl_APPLE_gl_sharing cl_khr_gl_event
Numnber of detected OpenCL devices 3
CPU Detected
VENDOR ID: 0xffffffff
Maximum number of parallel compute units: 8
Maximum dimensions for global/local work-item IDs: 3
Maximum number of work-items in each dimension: ( 1024 1 1 )
Maximum number of work-items in a work-group: 1024
GPU Detected
VENDOR ID: 0x1022700
Maximum number of parallel compute units: 2
Maximum dimensions for global/local work-item IDs: 3
Maximum number of work-items in each dimension: ( 1024 1024 64 )
Maximum number of work-items in a work-group: 1024
GPU Detected
VENDOR ID: 0x1024400
Maximum number of parallel compute units: 16
Maximum dimensions for global/local work-item IDs: 3
Maximum number of work-items in each dimension: ( 512 512 512 )
Maximum number of work-items in a work-group: 512
| OpenCL::Query OpenCL kernel (4) (0) | 2014.05.02 |
|---|---|
| OpenCL Test Program (3) (0) | 2014.03.31 |
| OpenCL test program (1) (0) | 2014.03.15 |
| OpenCL Architecture (0) | 2014.03.15 |
| Optical Flow에 의한 영상 정보 분석 (0) | 2013.05.28 |
OpenCL test program
책의 예제를 넣어서 실행
책에서 오류가 두군데가 있었다.
(1) ifdef에서 APPLE platform을 확인하는 매크로 정의 오류
== WAS ==
#ifdef APPLE
== IS ==
#ifdef __APPLE__
(2) OpenCL의 원칙은 OpenCL에 정보의 크기를 물어서 확인하고 그만큼 공간을 만든뒤에 그 정보를 받아들이는 것인데
예제에서는 정보를 물어서 공간을 만드는 데 까지는 있지만, 그 이후에 정보를 다시 받아들이는 함수를 호출 하지 않았다.
....
main(...)
{
...
먼저 필요한 공간의 크기를 확인
error = clGetPlatformIDs(0,
NULL,
&numOfPlatforms
);
if ( error < 0)
...
공간을 확보 하고
이제 확보된 공간으로 값을 받아온다.
error = clGetPlatformIDs(numOfPlatforms, <== 이 함수를 호출 하는 것을 추가함.
platforms,
NULL
);
...
}
전체 코드는 아래와 같다.
//
// main.cpp
// TestOpenCL
//
// Created by freegear on 2014. 2. 8..
// Copyright (c) 2014년 freegear. All rights reserved.
//
#include <iostream>
#include <stdio.h>
#include <stdlib.h>
#ifdef __APPLE__
#include <OpenCL/opencl.h>
#else
#include <CL/cl.h>
#endif
void displayPlatformInfo(
cl_platform_id id ,
cl_platform_info param_name,
const char * paramNameAsStr
)
{
cl_int error = 0 ;
size_t paramSize = 0 ;
error = clGetPlatformInfo(
id,
param_name,
0,
NULL,
¶mSize
);
char * moreInfo = (char *) alloca(sizeof(char)*paramSize);
error = clGetPlatformInfo(
id, /* The platform ID returned by clGetPlatformIDs or can be NULL.
If platform is NULL, the behavior is implementation-defined.*/
param_name, /* An enumeration constant that identifies the platform
information being queried. It can be one of the following values
as specified in the table below. */
paramSize, /* Specifies the size in bytes of memory pointed to by param_value.
This size in bytes must be greater than or equal to size of return type
specified in the table below. */
moreInfo, /* A pointer to memory location where appropriate values for a given
param_value will be returned. Acceptable param_value values are listed in
the table below. If param_value is NULL, it is ignored.*/
NULL /* Returns the actual size in bytes of data being queried by param_value.
If param_value_size_ret is NULL, it is ignored */
);
if( error != CL_SUCCESS)
{
if ( error == CL_INVALID_PLATFORM)
perror("Error CL_INVALID_PLATFORM");
else if (error == CL_INVALID_VALUE)
perror("Error CL_INVALID_VALUE");
else
perror("Unable to find any OpenCL platform information");
return ;
}
printf("%s : %s\n" , paramNameAsStr, moreInfo);
} // end of displayPlatformInfo
int main(int argc, const char * argv[])
{
// OpenCL 1.2 Data struction
cl_platform_id * platforms ;
/* OpenCL 1.1 scalar data types */
cl_uint numOfPlatforms ;
cl_int error ;
/*
Get the number of platforms
Remember that for each vendor's SDK installed on the computer,
the number of available platform also increased.
*/
error = clGetPlatformIDs(0, /*
The number of cl_platform_id entries that can be added to platforms.
If platforms is not NULL, the num_entries must be greater than zero.
*/
NULL, /* Returns a list of OpenCL platforms found.
The cl_platform_id values returned in platforms can be used to identify
a specific OpenCL platform. If platforms argument is NULL,
this argument is ignored.
The number of OpenCL platforms returned is the mininum of the value
specified by num_entries or the number of OpenCL platforms available. */
&numOfPlatforms
/* Returns the number of OpenCL platforms available.
If num_platforms is NULL, this argument is ignored. */
);
if ( error < 0)
{
perror("Unable to find any OpenCL platforms");
exit(1);
}
// allocate memory for the number of installed platforms
// alloca(....) occupies some stack space but is
// automatically freed on return
platforms = (cl_platform_id*) alloca(sizeof(cl_platform_id)*numOfPlatforms);
printf("Number of OpenCL platforms found : %d\n", numOfPlatforms);
error = clGetPlatformIDs(numOfPlatforms, /*
The number of cl_platform_id entries that can be added to platforms.
If platforms is not NULL, the num_entries must be greater than zero.
*/
platforms, /* Returns a list of OpenCL platforms found.
The cl_platform_id values returned in platforms can be used to identify
a specific OpenCL platform. If platforms argument is NULL,
this argument is ignored.
The number of OpenCL platforms returned is the mininum of the value
specified by num_entries or the number of OpenCL platforms available. */
NULL
/* Returns the number of OpenCL platforms available.
If num_platforms is NULL, this argument is ignored. */
);
if ( error < 0)
{
perror("Unable to find any OpenCL platforms");
exit(1);
}
// We invoke the API 'clPlatformInfo' twice for each
// parameter we are trying to extract
// and we use the return value to create temporary data
// structure (on the stack) to store
// the returned information ot the second invocation.
for( cl_uint i = 0 ; i < numOfPlatforms ; ++i)
{
displayPlatformInfo( platforms[i], CL_PLATFORM_PROFILE , "CL_PLATFORM_PROFILE");
displayPlatformInfo( platforms[i], CL_PLATFORM_VERSION , "CL_PLATFORM_VERSION");
displayPlatformInfo( platforms[i], CL_PLATFORM_NAME , "CL_PLATFORM_NAME");
displayPlatformInfo( platforms[i], CL_PLATFORM_VENDOR , "CL_PLATFORM_VENDOR");
displayPlatformInfo( platforms[i], CL_PLATFORM_EXTENSIONS , "CL_PLATFORM_EXTENSIONS");
}
return 0;
}
실행 결과는 아래와 같다.
Number of OpenCL platforms found : 1
CL_PLATFORM_PROFILE : FULL_PROFILE
CL_PLATFORM_VERSION : OpenCL 1.2 (Aug 24 2013 21:03:27)
CL_PLATFORM_NAME : Apple
CL_PLATFORM_VENDOR : Apple
CL_PLATFORM_EXTENSIONS : cl_APPLE_SetMemObjectDestructor cl_APPLE_ContextLoggingFunctions cl_APPLE_clut cl_APPLE_query_kernel_names cl_APPLE_gl_sharing cl_khr_gl_event
| OpenCL Test Program (3) (0) | 2014.03.31 |
|---|---|
| OpenCL Test Program (2) (0) | 2014.03.18 |
| OpenCL Architecture (0) | 2014.03.15 |
| Optical Flow에 의한 영상 정보 분석 (0) | 2013.05.28 |
| OpenCV를 이용한 Face Detection (0) | 2013.05.19 |
OpenCL에서 사용되는 개념은 아래와 같다.
Host 즉 CPU에서 Command와 Data를 Device로 보낸다.
이는 Command Queue로 저장된다.
Code의 실행은 순차적으로 진행되다가 Parallel하게 진행되는 방식으로 이루어진다.
Host에서 Serial하게 진행하고 parallel code는 Device에서 실행하게 된다.
Work Item은 N Dimension Work Item으로 만들어진다.
Global Domain으로 1024 x 1024로 만들어진다.
이 Global Domain에서 32x32로 Local Dimension으로 나누어서 한번에 실행된다.
Work Item : OpenCL Device상에서 계산이 이루어지는 기본 유닛
Kernel : Work Item을 구동시키는 code
Program : Work Item 을 구동시키는 Kernel과 기타 함수들의 모음
Context :
Command Queue : Host에서 Device로 보내지는 work의 pipeline
메모리 모델은 아래와 같다.
Private Memory는 각 Work Item별로 할당된 메모리이다.
Local Memory는 Work Group별로 할당된 메모리를 의미한다.
Global / Constant memory는 모든 Work Group에서 보여지는 메모리를 의미한다.
Host Memory는 당연히 Host용 메모리를 의미함.
Host -> Global -> Local -> Private 순으로 메모리의 데이터를 옮길 수 있다.
| OpenCL Test Program (2) (0) | 2014.03.18 |
|---|---|
| OpenCL test program (1) (0) | 2014.03.15 |
| Optical Flow에 의한 영상 정보 분석 (0) | 2013.05.28 |
| OpenCV를 이용한 Face Detection (0) | 2013.05.19 |
| 3D Noise Reduction algorithm test (0) | 2011.10.05 |
관련된 알고리즘과 코드는 인터넷에 많이 있습니다.
여기서 테스트한 것은 관련된 코드중 일부를 가지고 와서 조금 수정해서 테스트를 하였습니다.
| OpenCL test program (1) (0) | 2014.03.15 |
|---|---|
| OpenCL Architecture (0) | 2014.03.15 |
| OpenCV를 이용한 Face Detection (0) | 2013.05.19 |
| 3D Noise Reduction algorithm test (0) | 2011.10.05 |
| OpenCV 2.3 Computer vision (4) (4) | 2011.09.14 |
OpenCV를 이용한 Face Detection 입니다.
코드는 길지 않고 인터넷에서도 많이 돌아다니는 코드입니다.
몇가지 테스트할 일이 있어 한번 빌드를 해보았습니다.
| OpenCL Architecture (0) | 2014.03.15 |
|---|---|
| Optical Flow에 의한 영상 정보 분석 (0) | 2013.05.28 |
| 3D Noise Reduction algorithm test (0) | 2011.10.05 |
| OpenCV 2.3 Computer vision (4) (4) | 2011.09.14 |
| OpenCV 2.3 Computer vision (3) (0) | 2011.09.06 |
| Optical Flow에 의한 영상 정보 분석 (0) | 2013.05.28 |
|---|---|
| OpenCV를 이용한 Face Detection (0) | 2013.05.19 |
| OpenCV 2.3 Computer vision (4) (4) | 2011.09.14 |
| OpenCV 2.3 Computer vision (3) (0) | 2011.09.06 |
| OpenCV 2.3 Computer vision (2) (0) | 2011.09.05 |
|
