libgccjit-0

libgccjit-0

[Achim Gratz]

I've uploaded a test version of gcc-10.3.0 that has the JIT library enabled:

root=http://cygwin.stromeko.net/
$root/gcc-test

You will need to enable this repository in addition to the main Cygwin
one in order to get the correct dependencies.  This is created as a test
release, so you will have to select the test version manually.

The setup.ini is signed with my PGP key available at $root/stromeko.gpg,
which you can download and then feed to setup.exe via the -K switch.
The fingerprint of that key is:

      B22B 4890 B89E 5E5A 795C  E305 B40B EC75 3152 B1C3

If you have the keyart script (Python, you can just download from
https://raw.githubusercontent.com/atoponce/keyart/master/keyart), this
is what you should see after importing the key into your public keyring:

> keyart stromeko
+-----[DSA 2048]----+
|         ^.        |
|        ^ .        |
|       ^ E         |
|    . . ^ :        |
|. .  ^ : .         |
|.^  . ^.:S         |
| ..  . :^.         |
|.  .^ ^.^          |
|. ^:.^ ..          |
| ^:... .           |
|.^    .            |
+-----[3152B1C3]----+
+-------------------+
|                 :X|
|             .  ^^^|
|            :   ..^|
|         . : .   E |
|          i . . .  |
|       ^ S l : .   |
|        i l ^ ^    |
|       ^ . .       |
|      .   .        |
|       . .         |
|      ...          |
+-------------------+


Regards,
Achim.
-- 
+<[Q+ Matrix-12 WAVE#46+305 Neuron microQkb Andromeda XTk Blofeld]>+

SD adaptation for Waldorf microQ V2.22R2:
http://Synth.Stromeko.net/Downloads.html#WaldorfSDada

Re: libgccjit-0

[Ken Brown]

On 4/10/2021 12:23 PM, Achim Gratz wrote:
> 
> I've uploaded a test version of gcc-10.3.0 that has the JIT library enabled:

Thanks! It works great for building the feature/native-comp branch of emacs. 
(configure --with-native-compilation if you want to try it yourself.)

One minor packaging note: I would have expected the libgccjit0 package to 
contain only the DLL, with everything else in a libgccgit-devel package.

Ken

Re: libgccjit-0

[Achim Gratz]

Ken Brown via Cygwin-apps writes:
> Thanks! It works great for building the feature/native-comp branch of
> emacs. (configure --with-native-compilation if you want to try it
> yourself.)

Both architectures?  For whatever reasons the gcc testsuite on x86
always bombed out on these, so I have no idea if it works there.

> One minor packaging note: I would have expected the libgccjit0 package
> to contain only the DLL, with everything else in a libgccgit-devel
> package.

I've been looking at the msys packaging, which is the same, I have not
yet checked the Linux distributions.  Eventually we might want the two
packages you mention as that seems the cleaner approach.  We can sort
this out before it goes live.


Regards,
Achim.
-- 
+<[Q+ Matrix-12 WAVE#46+305 Neuron microQkb Andromeda XTk Blofeld]>+

SD adaptation for Waldorf rackAttack V1.04R1:
http://Synth.Stromeko.net/Downloads.html#WaldorfSDada

Re: libgccjit-0

[Ken Brown]

On 4/11/2021 11:49 AM, Achim Gratz wrote:
> Ken Brown via Cygwin-apps writes:
>> Thanks! It works great for building the feature/native-comp branch of
>> emacs. (configure --with-native-compilation if you want to try it
>> yourself.)
> 
> Both architectures?  For whatever reasons the gcc testsuite on x86
> always bombed out on these, so I have no idea if it works there.

Yes, it works on x86 too.

Ken

Re: libgccjit-0

[Achim Gratz]

Ken Brown via Cygwin-apps writes:
> On 4/11/2021 11:49 AM, Achim Gratz wrote:
>> Both architectures?  For whatever reasons the gcc testsuite on x86
>> always bombed out on these, so I have no idea if it works there.
>
> Yes, it works on x86 too.

Great.  So I'll have to update the MingW64 toolchains to 10.3 as well
(these will not get libgccjit since I'd have to pull in the patches from
MSys and I don't think there's a consumer for it anyway) and think about
the packaging again.  Not sure when I'll get around to wrap things up.


Regards,
Achim.
-- 
+<[Q+ Matrix-12 WAVE#46+305 Neuron microQkb Andromeda XTk Blofeld]>+

Factory and User Sound Singles for Waldorf Blofeld:
http://Synth.Stromeko.net/Downloads.html#WaldorfSounds

Re: libgccjit-0

[Ken Brown]

On 4/11/2021 3:48 PM, Achim Gratz wrote:
> Ken Brown via Cygwin-apps writes:
>> On 4/11/2021 11:49 AM, Achim Gratz wrote:
>>> Both architectures?  For whatever reasons the gcc testsuite on x86
>>> always bombed out on these, so I have no idea if it works there.
>>
>> Yes, it works on x86 too.
> 
> Great.  So I'll have to update the MingW64 toolchains to 10.3 as well
> (these will not get libgccjit since I'd have to pull in the patches from
> MSys and I don't think there's a consumer for it anyway) and think about
> the packaging again.  Not sure when I'll get around to wrap things up.

Sorry, I spoke too soon.  It's not even working right on x86_64.  In my first 
build, the ordinary byte-compiled files (.elc) were already present, and that 
somehow prevented the native-compiled files (.eln) from being built.  I just 
tried 'make bootstrap' and the build won't complete because of fork errors like 
this (on x86_64):

       0 [main] emacs 5400 child_info_fork::abort: address space needed by 
'jit-lock-05ec1a9b-73eb7028.eln' (0x400000) is already occupied

In an earlier attempt it was a different .eln file that was involved, but the 
address was the same (0x400000).

I have no idea what's going on.

Ken

Re: libgccjit-0

[ASSI]

Ken Brown via Cygwin-apps writes:
> Sorry, I spoke too soon.  It's not even working right on x86_64.  In
> my first build, the ordinary byte-compiled files (.elc) were already
> present, and that somehow prevented the native-compiled files (.eln)
> from being built.  I just tried 'make bootstrap' and the build won't
> complete because of fork errors like this (on x86_64):
>
>       0 [main] emacs 5400 child_info_fork::abort: address space needed
>       by 'jit-lock-05ec1a9b-73eb7028.eln' (0x400000) is already
>      occupied
>
> In an earlier attempt it was a different .eln file that was involved,
> but the address was the same (0x400000).

Disappointing, but semi-expected.  Can you tell me what peflags thinks
are the characteristics of the .eln files produced?  You could also
check if an ephemeral rebase helps.


Regards,
Achim.
-- 
+<[Q+ Matrix-12 WAVE#46+305 Neuron microQkb Andromeda XTk Blofeld]>+

SD adaptation for Waldorf Blofeld V1.15B11:
http://Synth.Stromeko.net/Downloads.html#WaldorfSDada

Re: libgccjit-0

[Ken Brown]

On 4/12/2021 1:55 AM, ASSI wrote:
> Ken Brown via Cygwin-apps writes:
>> Sorry, I spoke too soon.  It's not even working right on x86_64.  In
>> my first build, the ordinary byte-compiled files (.elc) were already
>> present, and that somehow prevented the native-compiled files (.eln)
>> from being built.  I just tried 'make bootstrap' and the build won't
>> complete because of fork errors like this (on x86_64):
>>
>>        0 [main] emacs 5400 child_info_fork::abort: address space needed
>>        by 'jit-lock-05ec1a9b-73eb7028.eln' (0x400000) is already
>>       occupied
>>
>> In an earlier attempt it was a different .eln file that was involved,
>> but the address was the same (0x400000).
> 
> Disappointing, but semi-expected.  Can you tell me what peflags thinks
> are the characteristics of the .eln files produced?
$ peflags -v jit-lock-05ec1a9b-73eb7028.eln
jit-lock-05ec1a9b-73eb7028.eln: 
coff(0x2026[+executable_image,+line_nums_stripped,+bigaddr,+dll]) pe(0x0000)

>  You could also
> check if an ephemeral rebase helps.

I'll give that a try.

Ken

Re: libgccjit-0

[Ken Brown]

On 4/12/2021 8:18 AM, Ken Brown via Cygwin-apps wrote:
> On 4/12/2021 1:55 AM, ASSI wrote:
>> Ken Brown via Cygwin-apps writes:
>>> Sorry, I spoke too soon.  It's not even working right on x86_64.  In
>>> my first build, the ordinary byte-compiled files (.elc) were already
>>> present, and that somehow prevented the native-compiled files (.eln)
>>> from being built.  I just tried 'make bootstrap' and the build won't
>>> complete because of fork errors like this (on x86_64):
>>>
>>>        0 [main] emacs 5400 child_info_fork::abort: address space needed
>>>        by 'jit-lock-05ec1a9b-73eb7028.eln' (0x400000) is already
>>>       occupied
>>>
>>> In an earlier attempt it was a different .eln file that was involved,
>>> but the address was the same (0x400000).
>>
>> Disappointing, but semi-expected.  Can you tell me what peflags thinks
>> are the characteristics of the .eln files produced?
> $ peflags -v jit-lock-05ec1a9b-73eb7028.eln
> jit-lock-05ec1a9b-73eb7028.eln: 
> coff(0x2026[+executable_image,+line_nums_stripped,+bigaddr,+dll]) pe(0x0000)
> 
>>  You could also
>> check if an ephemeral rebase helps.
> 
> I'll give that a try.

Good news: An ephemeral rebase at an appropriate place in the Makefile allows a 
bootstrap to succeed on both architectures.  Thanks for the suggestion.

Ken

Re: libgccjit-0

[Achim Gratz]

Ken Brown via Cygwin-apps writes:
> Good news: An ephemeral rebase at an appropriate place in the Makefile
> allows a bootstrap to succeed on both architectures.  Thanks for the
> suggestion.

OK, that shows that things do work, albeit in a decidedly clunky manner.


Regards,
Achim.
-- 
+<[Q+ Matrix-12 WAVE#46+305 Neuron microQkb Andromeda XTk Blofeld]>+

DIY Stuff:
http://Synth.Stromeko.net/DIY.html