nasm-3.01

[zeckma]

NASM-3.00 has changed operand order for TEST by forcing them to be in an order according to the spec. dav1d is a case where the order of the operands weren't in the proper order. Thus, it now fails to build from source.

A merge request has already been made to the project:

• ​https://code.videolan.org/videolan/dav1d/-/merge_requests/1810

The resulting patch works as expected, but upstream has not responded to the MR yet.

[zeckma]

libaom fails to configure as multipass optimization is not supported. In aom's case, yasm can be used instead. Either way, this is a problem and other packages may be affected as well, perhaps libjpeg-turbo. We should consider downgrading NASM until these issues have been fixed.

[pierre]

x264 problem reported upstream: ​https://code.videolan.org/videolan/x264/-/issues/80 But seems the maintainer does not understand what the problem is... Not tried but seems yasm can be used instead.

[Bruce Dubbs]

Replying to pierre:

x264 problem reported upstream: ​https://code.videolan.org/videolan/x264/-/issues/80 But seems the maintainer does not understand what the problem is... Not tried but seems yasm can be used instead.

Looking at './configure --help' I did not see a way to use yasm for x264, but I did see --disable-asm and that works. I do not know how much difference that would make for the performance. My experince is that gcc does a pretty good job.

[pierre]

Looks like the x264 issue might be a bug in nasm actually: ​https://github.com/netwide-assembler/nasm/issues/121

[pierre]

Replying to Bruce Dubbs:

Replying to pierre:

x264 problem reported upstream: ​https://code.videolan.org/videolan/x264/-/issues/80 But seems the maintainer does not understand what the problem is... Not tried but seems yasm can be used instead.

Looking at './configure --help' I did not see a way to use yasm for x264, but I did see --disable-asm and that works. I do not know how much difference that would make for the performance. My experince is that gcc does a pretty good job.

Yes, yasm cannot be used. I thought it was possible to pass AS=yasm to configure, but this fails...

[Joe Locash]

Replying to pierre:

Looks like the x264 issue might be a bug in nasm actually: ​https://github.com/netwide-assembler/nasm/issues/121

nasm-3.01rc1 released. It fixes the builds of x264, x265, and libaom.

[Bruce Dubbs]

I can validate that it fixes x265 and x265. dav1d still fails though.

[Joe Locash]

Replying to Bruce Dubbs:

I can validate that it fixes x265 and x265. dav1d still fails though.

​https://github.com/videolan/dav1d/commit/0bc6bd93417179cd0c30fac40d2fd11aa29c8523

[Bruce Dubbs]

Replying to Joe Locash:

​https://github.com/videolan/dav1d/commit/0bc6bd93417179cd0c30fac40d2fd11aa29c8523

I got that as a patch. I may be able to fix it as a sed.

[Bruce Dubbs]

This works:

sed -e '/r7b,/s/r7b.*$/[fg_dataq+FGData.overlap_flag], r7b/' \
    -i src/x86/filmgrain16_avx{2,512}.asm

[Joe Locash]

Since dav1d is now in the book, for firefox add:

ac_add_options --with-system-av1

to mozconfig and make dav1d recommended otherwise, Firefox's internal copy of dav1d needs to be patched.

Thunderbird will need the same change.

[pierre]

Seems nasm-3.01-rc3 should even fix dav1d without the sed: ​https://github.com/netwide-assembler/nasm/commit/b3358fe63e48983cc9c9b30142d5a945ce077459

[Xi Ruoyao]

Firefox builds fine with rc3.

[Xi Ruoyao]

Replying to Joe Locash:

Since dav1d is now in the book, for firefox add:

ac_add_options --with-system-av1

to mozconfig and make dav1d recommended

Done for firefox at r12.4-281-gb16421b7db. For some reason it wants both dav1d and libaom.

[Bruce Dubbs]

Upstream is now up to version 3.01rc9.

[Bruce Dubbs]

Now nasm 3.01 has been released.

[Bruce Dubbs]

Fixed at commit a3c4ce2273.

Tested by building libaom, x254, x265. and dav1d. All built without complaint.

[Bruce Dubbs]

Changes for version 3.01 are at ​https://www.nasm.us/docs/3.01/nasmac.html#section-C.1.1

[Bruce Dubbs]

Milestone renamed