update to svt-av1-3.1.2, from Brad, some tweaks:
- add comments showing how to compile/run tests
- enable DEBUG_PACKAGES
- set USE_NOBTCFI-amd64, tests result in IBT-related SIGILL on hardware
with support. this is not new, but wasn't discovered in the previous version.
(arm64 BTI machines are able to run through the test suite ok)
If editline has been switched to vi mode (i.e. via "bind -v" in .editrc),
setup a keybinding so that command mode can be entered. Diff originally
from Walter Alejandro Iglesias with tweaks. Feedback from Crystal Kolipe.
ok djm
Revert revision 1.379
Rewrite m_getuio() to better align data in the mbufs
...
Allowing m_getuio() to use m_clget() with sizes up to MAXMCLBYTES (64k)
puts too much pressure on the pmemrange allocator. Right now the
physical pages for mbufs need to be in a single contiguous segment. On
top of this pool multiplies the size by 8 and in pool headers may add an
additional alignment restriction on pysical memory on __HAVE_PMAP_DIRECT
archs. For 64k clusters half a megabyte of contiguous physical memory is
needed, for 12k clusters the use of in page headers result in a 128k
allocation on a 128k boundary (so in the worst case a 252k area is needed).
Both of those requests cause issues on landisk. With 64MB of RAM there is
little room and after some time the physical memory is so fragmented that
the allocations for such large contiguous segements fail. Even dropping
the buffercache to its minimum and paging out all pageable memory does
not help to recover.
[2 lines not shown]