compiling error on x86

[alephman]

compiling error on x86

[hughperkins]

Good point! Thanks for the heads-up. Should be working now? (You'll need to do git submodule update --init --recursive, after doing git pull)

[hughperkins]

oh wiat, still something missing. checking...

[hughperkins]

Should be mildly healthier now?

[hughperkins]

Big heads-up though. Curently I'm hard at work on a massive overhaul ,in branch runtime-compile, https://github.com/hughperkins/cuda-on-cl/compare/runtime-compile?expand=1 It's mildly broken, but I can compile/run cuda_sample.cu ok. I think that since you're working on trying to get basic things working for now, might be good to try this branch?

quick heads-up: the main changes in this branch are:

• major refactorization, you probably wont notice this too much
• generation of device-side opencl from device-side IR happens at runtime now => you wil lalmost certainly notice this change. Whether this makes things work better or worse for you is uncertain to me at this time.

[alephman]

after checking out runtime-compile branch, re-compile again, get another error http://pastebin.ubuntu.com/23478287/

[hughperkins]

Hmmm, builds ok for me. And I tried in docker earlier too. And worked in docker too. Are you sure you're trying 'runtime-compile' branch? If I switch to runtime-compile branch, there is no 'ConstantDumper.cpp' file at all in fact:

(env3cl) ubuntu@peach:~/git/cuda-on-cl/build$ git checkout runtime-compile 
Already on 'runtime-compile'
Your branch is up-to-date with 'hugh/runtime-compile'.
(env3cl) ubuntu@peach:~/git/cuda-on-cl/build$ ls ../src/Constant*
ls: cannot access '../src/Constant*': No such file or directory
(env3cl) ubuntu@peach:~/git/cuda-on-cl/build$ git checkout instruction-own-class 
Switched to branch 'instruction-own-class'
Your branch is up-to-date with 'hugh/instruction-own-class'.
(env3cl) ubuntu@peach:~/git/cuda-on-cl/build$ ls ../src/Constant*../src/ConstantDumper.cpp  ../src/ConstantDumper.h

Can you do:

git status
git log -n 3

... and paste the results?

[alephman]

Thanks hughperkins , git clone the whole repo again and checkout runtime-compile branch, it works now. Recently the github is very very slow in my side, probably it broke somewhere.
BTW, comment out -device.cl in cocl.Makefile?

[hughperkins]

Yes, opencl generation is done at runtime now, from hostside_opencl_funcs.cpp

Ir-to-opencl executable is still available, but just for debugging/testing really.

On 15 November 2016 03:35:41 GMT+00:00, alephman [email protected] wrote:

Thanks hughperkins , git clone the whole repo again and checkout
runtime-compile branch, it works now. Recently the github is very very
slow in my side, probably it broke somewhere.
BTW, comment out -device.cl in cocl.Makefile?

You are receiving this because you commented.
Reply to this email directly or view it on GitHub:
#7 (comment)

Sent from my Android device with K-9 Mail. Please excuse my brevity.

[hughperkins]

(oh, hmmm, note my observation "it works ok from docker" was actually running the clone, inside the docker, from master branch, hence ignore that comment :-) )

[hughperkins]

(but note that the technology exists now to build other branches :-) https://github.com/hughperkins/cuda-on-cl/blob/runtime-compile/docker/Dockerfile#L3 )

[hughperkins]

Note that massive overhaul/merge-to-master might mean that some things that were broken for work better/differently now?

[hughperkins]

this seems relally old, and non-specific. Lets close it. Please open new issues for any outstanding issues that havent been addressed here.