asymptote-2.80

[Bruce Dubbs]

New minor version (was 2.79)

[ken@…]

I suggest we let this stew until TL2023 has finished its pretest.

[ken@…]

279 -

The transparency shader was further optimized. The tensorshade fillrule was fixed. A compatibility fix was made to xasyBezierInterface.py. A real operator ecast(rational r) was implemented up to maxDenominator=100000. The dvisvgmMultipleFiles option, which speeds up xasy deconstruction with PDF TeX engines, was documented.

280 -

Release Notes for Version 2.80 Compression of empty transparent pixel counts and other optimizations were implemented in the transparency shader. A weighted least-squares fit was added.

[ken@…]

I find 280 useless. My latex-test-20220322 'cylinder' test to create a PDF puts the loadavg over 8 on this 8-core box, but eventually completes. However, the PDF does not contain the graphic, only the heading label. Using my asy-test-20200809 tarball the tests to create webgl (rotateable html) work but tests to create pdf, png and svg files fail, reporting

util_cpu_caps.num_cpu_mask_bits = 32

shipout3(prefix,f,preview ? nativeformat() : format,

/opt/texlive/2022/texmf-dist/asymptote/three.asy: 2896.13: runtime:

I suspect that the binary 2022 release doesn't contain asy-280, will need to find a spare machine to check that.

[ken@…]

Replying to ken@…:

I find 280 useless. My latex-test-20220322 'cylinder' test to create a PDF puts the loadavg over 8 on this 8-core box, but eventually completes. However, the PDF does not contain the graphic, only the heading label. Using my asy-test-20200809 tarball the tests to create webgl (rotateable html) work but tests to create pdf, png and svg files fail, reporting

util_cpu_caps.num_cpu_mask_bits = 32

shipout3(prefix,f,preview ? nativeformat() : format,

/opt/texlive/2022/texmf-dist/asymptote/three.asy: 2896.13: runtime:

I suspect that the binary 2022 release doesn't contain asy-280, will need to find a spare machine to check that.

I now suspect this is again a "doesn't work on amdgpu" problem. 280 is being used by debian (sid) and gentoo, and it works on my skylake (intel video, I suppose the iris driver in mesa).

I'll need to review what I've used in the recent past on amdgpu (if anything).

[ken@…]

Definitely an amdgpu problem. I've uploaded a new simplified set of tests in asy-cylinder.tar.gz to ​https://www.linuxfromscratch.org/~ken/asy-cylinder-tests/ with explanation in the README file in that directory.

Although this doesn't test the old 2D output from asy, it provides a much shorter test of the 3D functions.

With the new reduced tests, on BLFS-11.1 without CFLAGS on my picasso (3400G) with current amdgpu firmware (partially updated in February 2022 according to git blame) and linux-5.17.1 it now creates output for each of the tests, without overloading the cores, but for the pdf, png and svg targets the output is just a white background and for the latex test only the heading appears. Both 2.78 and 2.79 do not have any problem.

Now, when the output is blank I get a series of messages each saying:

amdgpu: Failed to allocate a buffer: amdgpu: size : 4135129088 bytes amdgpu: alignment : 2097152 bytes amdgpu: domains : 4 amdgpu: flags : 4

(11 of these for the pdf target, similar for png, only 4 for svg). The latex target does not appear to produce any odd onscreen messages. But each of these failures creates a syslog message:

[drm:amdgpu_job_timedout [amdgpu]] *ERROR* ring gfx timeout, but soft recovered

Requires firmware updates on my normal (hardened) system on the picasso to see if that too now completes, and testing on my renoir.

Works ok on skylake with 2.80.

[ken@…]

No progress on the picasso.

Issue raised for 2.80, ​https://github.com/vectorgraphics/asymptote/issues/312 proceding with 2.79 which works ok on all my machines.

[ken@…]

Updated to 2.79 in 845db20de8c21cf09090e55a39be7e45c7b48296 11.1-429

Keeping open re 2.80 pending any feedback on the issue I've raised.

[ken@…]

I thought there was a fix in the git version I was asked to test. Worked fine on my minimal asy-cylinder tests. I backported that to 2.80, same. But then I started on my fuller asy-test-20200809.tar.xz tests for random kleinbottles. Fails with a shipot3 runtme error.

Went back to that git version, similar. Issue 313 opened, because I'd closed the original :-(

[ken@…]

Further fix committed upstream (also applies to 2.79 on this machine).

[ken@…]

Fixed, 5f29a21dcab1ceb1f5f3a1bad05f1f25000fd4ff 11.1-520