Created attachment 487366 [details] build.log fp16_test.cpp: In member function ‘virtual void movit::FP16Test_NaN_Test::TestBody()’: fp16_test.cpp:54:2: error: ‘fp32’ was not declared in this scope fp32 borderline_inf; ^~~~ fp16_test.cpp:55:2: error: ‘borderline_inf’ was not declared in this scope borderline_inf.u = 0x7f800000ull; ^~~~~~~~~~~~~~ fp16_test.cpp:56:7: error: expected ‘;’ before ‘borderline_nan’ fp32 borderline_nan; ^~~~~~~~~~~~~~ fp16_test.cpp:57:2: error: ‘borderline_nan’ was not declared in this scope borderline_nan.u = 0x7f800001ull; ^~~~~~~~~~~~~~ make: *** [Makefile:131: fp16_test.o] Error 1
Confirm
(In reply to Michal Jakubowski from comment #1) > Confirm me too
same here
After talking to the upstream author, he noticed that the user affected by this problem has been using the -march=bdver3 optimization flag. Interestingly, this enables FP16C support - just like -march=native might do on any CPU supporting the fp16 built-in type (like mine). The fp16.h file does not take that into account. I'll update this report as soon as a fix is available.
Created attachment 487640 [details, diff] Fix compilation on systems natively supporting F16C. This patch requires https://bugs.gentoo.org/show_bug.cgi?id=626692 to be applied first.
(In reply to Mihai Moldovan from comment #5) > Created attachment 487640 [details, diff] [details, diff] > Fix compilation on systems natively supporting F16C. > > This patch requires https://bugs.gentoo.org/show_bug.cgi?id=626692 to be > applied first. This patch only includes a change to the ebuild? Where is the actual patch located?
(In reply to Dennis Schridde from comment #6) > Where is the actual patch located? https://git.sesse.net/?p=movit;a=commit;h=f680714c0378a80e1638824c70c5ec62d6ade8ee
Created attachment 487646 [details, diff] Fix compilation on systems natively supporting F16C. Sorry, failure on my part to actually add the patch. This patch requires https://bugs.gentoo.org/show_bug.cgi?id=626692 to be applied first.
(In reply to Mihai Moldovan from comment #8) > Created attachment 487646 [details, diff] [details, diff] > Fix compilation on systems natively supporting F16C. > > Sorry, failure on my part to actually add the patch. > > > This patch requires https://bugs.gentoo.org/show_bug.cgi?id=626692 to be > applied first. Curious why your patches have email headers etc?
The upstream ones? Or mine? Both are fetched out of git repositories with git-format-patch.
New version 1.5.3 should be resolved the problem
(In reply to Marco Genasci from comment #11) > New version 1.5.3 should be resolved the problem Indeed, and looking at the diff, a simple bump should be OK.
(In reply to Marco Genasci from comment #11) > New version 1.5.3 should be resolved the problem Where can I find version 1.5.3?
(In reply to Juergen Rose from comment #13) > Where can I find version 1.5.3? You can copy 1.5.2 ebuild in your local overlay and rename it in 1.5.3
(In reply to Marco Genasci from comment #14) > (In reply to Juergen Rose from comment #13) > > Where can I find version 1.5.3? > You can copy 1.5.2 ebuild in your local overlay and rename it in 1.5.3 Works for me.
(In reply to Attila Tóth from comment #15) > (In reply to Marco Genasci from comment #14) > > (In reply to Juergen Rose from comment #13) > > > Where can I find version 1.5.3? > > You can copy 1.5.2 ebuild in your local overlay and rename it in 1.5.3 > > Works for me. Confirm.
(In reply to Marco Genasci from comment #14) > (In reply to Juergen Rose from comment #13) > > Where can I find version 1.5.3? > You can copy 1.5.2 ebuild in your local overlay and rename it in 1.5.3 Thanks, works for me.
Same problem here, could you update portage?
1.5.3 built without issues with gcc-7.2
I don't mean to be rude to the assignee but this bug hasn't even moved passed unconfirmed, despite a simple version bump solving this. We're all busy people but 10 seconds to version bump can save many users a lot of time and headaches. Please update the gentoo tree.
(In reply to Konstantin (elxa) from comment #19) > 1.5.3 built without issues with gcc-7.2 But it is not in portage yet.
(In reply to Silvio from comment #21) > (In reply to Konstantin (elxa) from comment #19) > > 1.5.3 built without issues with gcc-7.2 > > But it is not in portage yet. In my opinion should be, looks fine for me. If you can, try this: https://github.com/hhoffstaette/portage/tree/master/sys-devel/gcc
https://gpo.zugaina.org/sys-devel/gcc Try the ahyangyi-overlay.
The new upstream release, available as part of https://gitweb.gentoo.org/repo/gentoo.git/commit/media-video/movit?id=e1f1f16eefefd7868b26068bffdec3e99077ae6d should fix the issue and make my patches obsolete. Please close this bug report. And to commenters: please stop recommending GCC versions in random overlays - that's an easy way to break your system. You won't need GCC 7.2 for movit 1.5.3 to build fine. Any recent, supported, in-tree GCC version will do.
