-
Notifications
You must be signed in to change notification settings - Fork 2
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.
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
One test fails on PowerPC (Fixed) #7
Comments
By the way, also OpenMP detection does not work: “OpenMP not detected. Using single thread only”. GCC does support OpenMP, however. |
Oh wow. Thanks a lot for reporting this bug. I'll look into this. Would you mind showing me what can I do to reproduce this error? Also if this is too hard, I would be really thankful if you could help me test the devel version on your machine as I might have fixed Also the devel version of Thanks! |
@dipterix Thank you for responding to the issue. From the latest commit tests results are: [ FAIL 253 | WARN 11 | SKIP 3 | PASS 672 ] UPD. Any ideas? |
@dipterix I suppose that errors which look very wrong may be due to ignored endianness. Also, does it use long doubles? PowerPC BE uses IBM format for it (same for macOS, AIX, *BSD, Linux – as long as those are Big-endian). Re reproducing: I do not think these errors have anything to do with macOS specifics, so presumably this can be reproduced on any PowerPC platform, though 64-bit ones may have fewer failures. Common versions of BSD and some versions of Linux support PowerPC, both 32- and 64-bit. If no hardware is available, Qemu may be used, perhaps. But I can test on my end whatever you suggest, that may be easier. |
OK I will try to fix 64-bit big-endian and see what I can do with 32-bits. I don't have machines to test on 32-bits. In theory |
Thank you very much for working on this! For sure, doubles are supported on 32-bit, and even 128-bit long doubles are (with a caveat of being composed of two doubles, IBM format). This should work in R: I believe, many packages are using these, and so far whatever we test on PowerPC, usually passes 100% tests or occasionally have rounding-related issues. I mean, the platform itself is perfectly functional, and all R-related software should work fine. |
Hi @barracuda156 I just pushed a patch to the Github. Would you mind helping me test the devel version? I think 64-bit should be fixed. I gue→↘↗ss, not so sure because I don't have big-endian machine by my hands. All I can do is to flip the endian flag in my code to force the file and machine to have different endianness. Again, thanks a lot for helping me. |
@dipterix You have fixed everything:
Thank you! |
The fix 0.1.6 has been submitted to CRAN |
Awesome! It is already merged to Macports too: https://ports.macports.org/port/R-filearray/details |
@dipterix Could you please look into this?
P. S. Generally speaking, usual suspects for causes for errors in such cases are assumed little-endianness and assumed 64-bitness. Less common are 4-byte bool, IBM format for long double (non-IEEE) and PPC-specific rounding.
The text was updated successfully, but these errors were encountered: