OpenCL 2.2 support? about opencl-clhpp HOT 9 CLOSED

Great, I'll add a few tests and create a PR in the next day or so.

from opencl-clhpp.

Is there a plan to update the C++ bindings for OpenCL 2.1/2.2? I find that setting CL_HPP_TARGET_OPENCL_VERSION to 210 or 220 is not permitted. Calling clCreateProgramWithIL returns cl_program for example - which starts to draw one into the C API...

from opencl-clhpp.

We have an opencl21 branch with OpenCL 2.1 support, but we've been putting off merging it into master until someone has actually tested it on a real OpenCL 2.1 platform. Are you able to do this and confirm that it works (even for just creating a program from IL for example)?

OpenCL 2.2 will be the same - someone will need to have an OpenCL 2.2 platform that we can test the required changes on.

from opencl-clhpp.

Are you able to do this and confirm that it works (even for just creating a program from IL for example)?

I just confirmed that I was able to create a program from IL and build it using the headers in the openc21 branch on the Intel OpenCL 2.1 implementation. I think these changes can be safely merged.

Regarding OpenCL 2.2, can we use something like the Intercept Layer and/or the ICD loader stub to check that the C++ bindings nominally work even before an OpenCL 2.2 platform is released? The only new APIs are in OpenCL 2.2 are clSetProgramReleaseCallback and clSetProgramSpecializationConstant and I think we should be able to test these with relatively high confidence using this method. Let me know if this is something you would be interested in trying.

from opencl-clhpp.

It works here too: with Intel OpenCL 2.1 on 64-bit Ubuntu 18.10 using GCC. A few comments:

• Missing return value warnings are issued from getHostTimer and getDeviceAndHostTimer;
• The definition of basic OpenCL types such as cl_int in cl_platform.h on Linux uses __attribute__ syntax, which leads to warnings whenever they're used as template arguments;
• CL_TARGET_OPENCL_VERSION is defined in cl2.hpp, though it isn't in the Ubuntu opencl-clhpp-headers package. Just a comment. (This is also the case in the master branch.)

By the way, I'm happy to contribute the example I used; though I'm not sure how you want to manage the dependency on the C++ kernel language compiler. Environment variable? CLCXX?

from opencl-clhpp.

I think these changes are a good basis to move forward. @pkeir @bashbaug Would one of you volunteer to rebase the branch and open PRs? :)

from opencl-clhpp.

@bashbaug Would one of you volunteer to rebase the branch and open PRs? :)

@jrprice I think it's best if you create the PR, since you've made most of the OpenCL 2.1 changes?

It rebased automatically for me, but if it helps, I've uploaded a rebased opencl21 branch here:

https://github.com/bashbaug/OpenCL-CLHPP/tree/opencl21_rebase

I also added one commit for a missing CL_DEVICE_SUB_GROUP_INDEPENDENT_FORWARD_PROGRESS query. I didn't find anything else missing.

from opencl-clhpp.

I have first draft OpenCL 2.2 changes here:

https://github.com/bashbaug/OpenCL-CLHPP/tree/opencl22

These changes are baselined against the OpenCL 2.1 changes, so they will need to go in first.

I've done minimal testing via a hacked Intercept Layer and things appear to be working, but by no means have I done exhaustive testing.

from opencl-clhpp.

@bashbaug I've had a look at your draft. With a couple of unit tests added, this looks pretty much good to go to me.

from opencl-clhpp.