gcc 9 and 10

@sTix
I.e. everything fully enabled like it on 8.4.0 on SBA's account, and nothing disabled? Also and clib2 and newlib both builds, etc?

I just may give a go for GCC 10 :)

Quote:

You mean for a 68k build, or about some 68k-related parts in the PPC build?

@kas1e

Gcc 9 is 1:1 with SBA:s repo, how it's built, patches applied and so on.

Gcc 10 is 0.99:1. As far as OS4 is concerned it's 1:1 unless I've misunderstood something (which is a possibility of course :)

I'd say 10 is ready for some real testing. Would be interesting to see if it's possible to get a c++20 project running on OS4. It will probably lead to bugs popping up, like the threading problems you identified the other day, but that's good I guess.

@sTix
I set up, for now, a new Cygwin instance, and cloning gcc10 repo from you. I Will try to build it as usually build cross-compilers and give it a go to build firstly some simple stuff, and then Odyssey, just to see how it behave for real. If Oddysey will be compiled and works as before, then we can say it at least "ok"

It would be nice with a recent version of gdb-binutils as well. I quickly peeked into the current patchset and what it would take to apply that to the latest gdb-binutils. I walked away with a strong desire to start smoking crack.

@kas1e
Quote:
Great! Lets hope for the best.

@sTix
Cross compiler version of gcc10 builds fine, test binary works (click open in new tab) and even -flto works as expected.



Trying to build Odyssey now:

libwtf: builds with no probs

libjavascript: builds fine, but brings warnings about [-Wdeprecated-copy], but no errors at least

libwebcore: same build fine but brings warnings about [-Wdeprecated-copy], but no errors at least

libwebkit-owb: that one full of amiga code, so this one together with some warnings bring even an error: gcc10 by default seems to use now -Wnarrowing, and that causes issues with enums+switch of names of mui methods, so had to add -Wno-narrowing to compiler flags. Other than that all builds fine.

And final linking done! Tested binary: works :) So the first stress test passed. I Will try to build some other heavy shit (like few games using all sort of stuff).


Oh, and the size of the binary now 61.068.962, instead of 63.321.206. Dunno why it saves about 2.5MB with gcc10, maybe some inbuild optimization takes place, or maybe something else which I not aware of, but it's good. I noticed in the output of gcc10 that it has now strings "Supported LTO compression algorithm: zlib", while with the previous version I didn't have such an output, so maybe some lto related stuff are default now?

@sTix

Quote:

Quote:

Just tried adtools-os4-9-20210108-716 on OS4 with a C++ project. No problems so far. -athread=native seems to work.

@kas1e and @IamSONIC

Excellent, very good to hear!

@sTix

gcc9 is working well with scummvm and it's c++11 functions

i haven't tested gcc10 (would like to wait for the std:: fix to hit it, just like wtih gcc9, if it hasn't already)

@Raziel
Quote:
Great, thanks for testing.

Quote:
Which fix are you refering to? The two threading fixes from salass00 are included in both the gcc 9 and 10 builds.

@sTix

Those were the ones i was waiting for

Great, on to gcc10 testing then...

@sTix

Uh, just saw the lto part...need to trx it again, didn't work with earlier versions

I unfortunately still get the same error message
/SDK/gcc/ppc-amigaos/bin/ld: -f may not be used without -shared
even with static builds

and using -shared the binary grows in size instead of shrinking.

So, lto still don't work natively...too bad :-/

@sTix

Building another C++ project using adtools-os4-9-20210108-716 shows errors regarding threading during linking stage:



No problems using GCC 8

@IAMSonic
You forget to add -athread=native. Those __ghtread errors coming from that. Strange that it works for gcc8, it should not as well. Or maybe you just have that set in gcc8 envs, and forgot to set now on new env.

@sTix

gcc10 working as well as gcc9

Great work, thank you

@Kas1e

-athread=native was added already as a compiler flag but it works now when adding it as linker flag too. Compiling and linking steps are seperated in this makefile. GCC 8 doens't need that but will check GCC 8 envs later.

@sTix

Thanks for your efforts updating our compilers

@IamSONIC
Quote:
No problem, just scratching my own itches :)

@sTix

If you're running out of ideas what to do next you could have a look at the summary (Requirements compiling WebKit) hans created a while ago

https://www.amigans.net/modules/xforum ... id=119194#forumpost119194

Maybe the first two are interesting:

1. We need our C/C++ stdlib headers to be brought up-to-date with the latest standard. Especially cmath. Newlib is missing multiple functions (log2(), asinh(), etc.).
2. Newlib also has gaps in its POSIX compliance. It's missing basics such as tzset()**