OpenCL: Difference between revisions

No edit summary
 
(14 intermediate revisions by the same user not shown)
Line 6: Line 6:
===Linux===
===Linux===
https://gist.github.com/Brainiarc7/dc80b023af5b4e0d02b33923de7ba1ed
https://gist.github.com/Brainiarc7/dc80b023af5b4e0d02b33923de7ba1ed
<pre>
sudo apt install ocl-icd-opencl-dev opencl-headers
sudo apt install opencl-c-headers opencl-clhpp-headers
</pre>


==Getting Started==
==Getting Started==
Line 13: Line 17:
See https://www.eriksmistad.no/getting-started-with-opencl-and-gpu-computing/
See https://www.eriksmistad.no/getting-started-with-opencl-and-gpu-computing/


{{hidden | C example |
vector_add_kernel.cl
<syntaxhighlight lang="c">
<syntaxhighlight lang="c">
__kernel void vector_add(__global const int *A, __global const int *B, __global int *C) {
__kernel void vector_add(__global const int *A, __global const int *B, __global int *C) {
Line 23: Line 29:
}
}
</syntaxhighlight>
</syntaxhighlight>
{{hidden | C example |
<syntaxhighlight lang="c">
<syntaxhighlight lang="c">
#include <stdio.h>
#include <stdio.h>
Line 141: Line 146:
===C++===
===C++===
[https://github.khronos.org/OpenCL-CLHPP/index.html#intro C++ Bindings]<br>
[https://github.khronos.org/OpenCL-CLHPP/index.html#intro C++ Bindings]<br>
While you can use the C bindings in your C++ application, Khronos also provides a set of C++ bindings in <code>CL/cl2.hpp</code> which are much easier to use alongside std containers such as <code>std::vector</code>. When using C++ bindings, you also do not need to worry about releasing buffers since these are reference-counted.
While you can use the C bindings in your C++ application, Khronos also provides a set of C++ bindings in <code>CL/cl.hpp</code> (or <code>CL/cl2.hpp</code>) which are much easier to use alongside std containers such as <code>std::vector</code>.
When using C++ bindings, you also do not need to worry about releasing buffers since these are reference-counted.
{{hidden | C++ example |
{{hidden | C++ example |
<syntaxhighlight lang="cpp">
<syntaxhighlight lang="cpp">
Line 240: Line 246:
</syntaxhighlight>
</syntaxhighlight>
}}
}}
===Python===
See [https://documen.tician.de/pyopencl/index.html pyopencl].


===Julia===
===Julia===
Line 245: Line 254:


==Usage==
==Usage==
===Types===
===Scalar Types===
[https://www.khronos.org/registry/OpenCL/sdk/1.0/docs/man/xhtml/scalarDataTypes.html Scalar Data Types]<br>
[https://www.khronos.org/registry/OpenCL/sdk/1.2/docs/man/xhtml/scalarDataTypes.html OpenCL 1.2 Scalar Data Types]<br>
While all OpenCL devices support single-precision floats, not all support double-precision doubles.<br>
While all OpenCL devices support single-precision floats, not all support double-precision doubles.<br>
===Vector Types===
[https://www.khronos.org/registry/OpenCL/sdk/1.2/docs/man/xhtml/dataTypes.html OpenCL Data Types]<br>
[https://www.khronos.org/registry/OpenCL/sdk/1.2/docs/man/xhtml/vectorDataTypes.html OpenCL 1.2 Vector Data Types]<br>
Just like glsl, OpenCL supports vector types such  
Just like glsl, OpenCL supports vector types such  
<syntaxhighlight lang="c">float3 my_vec = (float3)(1.0);</syntaxhighlight>
<syntaxhighlight lang="c">float3 my_vec = (float3)(1.0);</syntaxhighlight>
where its elements are accessed using x,y,z as <code>my_vec.x</code>.<br>
where its elements are accessed using x,y,z as <code>my_vec.x</code>.<br>
To convert between vector types, use <code>convert_T()</code><br>
To convert between vector types, use <code>convert_T()</code><br>
;Notes
* 3-component data types are aligned to 4 components. I.e. an array of <code>uchar3</code> with 4 elements will be equivalent to an array of <code>uchar4</code> with 4 elements.
==OpenGL Interop==
Setting up OpenCL/OpenGL interop is fairly complicated and very hard to debug. 
You will also need to manage synchronizing OpenGL/OpenCL so they do not access the same memory at the same time. 
If you can, just use OpenGL compute shaders rather than OpenCL to simplify your life.
===Textures===
See [https://software.intel.com/content/www/us/en/develop/articles/opencl-and-opengl-interoperability-tutorial.html OpenCL™ and OpenGL* Interoperability Tutoria].
In C++, you can use [https://github.khronos.org/OpenCL-CLHPP/classcl_1_1_image_g_l.html <code>cl::ImageGL</code>] to access textures in OpenGL. 
Note that <code>cl::Image</code> and <code>cl::Buffer</code> are not the same thing. Interchanging them will result in <code>CL_INVALID_MEM_OBJECT</code> errors or similar.
I recommend writing to a separate buffer and copying to images.
See [https://www.khronos.org/registry/OpenCL/sdk/2.2/docs/man/html/clCreateFromGLTexture.html clCreateFromGLTexture] to get a list of compatible pixel formats. 
If in doubt, use <code>GL_RGBA8</code> which is the most likely format to be supported.
===Buffers===
[https://web.engr.oregonstate.edu/~mjb/cs575/Handouts/opencl.opengl.vbo.1pp.pdf Oregon State VBO Interop] 
[https://github.khronos.org/OpenCL-CLHPP/classcl_1_1_buffer_g_l.html cl::BufferGL]


==Advanced Topics==
==Advanced Topics==
====Local Memory v. Global Memory====
====Local Memory v. Global Memory====
[[Category:Programming languages]]
[[Category:GPU Programming languages]]