A descriptor heap that a shader binds whole while only the pieces from its
base are mapped (a sparse heap) is copied on the CPU for every build that
binds it unless a mirror serves it. The sparse heap mirror takes a heap whose
needed extent lies in one releasable registered range. A heap mapped piece by
piece, each mapping its own allocation and the pieces back to back from the
heap's base, is several registered ranges, so every build copied its mapped
run again.
Such a region is now bound in place from an import of that run: an import
already covering it, the run's own import, or one joined over exactly the run
(#4517). This applies to any sparse heap no mirror took, including under
APS5_NO_LEASE_MIRROR or APS5_NO_SPARSE_HEAP_MIRROR. A join is made only for a
region that starts at a registered range, once the previous bind of that
address saw the same last mapped byte: a heap that grows between every two of
its binds keeps the copy, and one bound more often than it grows is joined
again after each growth. Joins stop after 64 per joined run base, with a
message: the heap is then bound in place while an existing import covers its
run and copied otherwise. The records go with their range. The region's descriptor and BDA ranges end at the import's end, which
covers the mapped run and the last view, as the copy does. A region whose
mapped part does not start at its base, with a view past the run, written by
its shader, or whose run is not imported is copied as before. A direct
region whose import ends before it outside this path still fails in
Descriptor, as before.
A set over such a region stays reusable: DirectRegions and InPlaceReads name
the region up to its import's end, so the import's serial proves it, and a
reuse also requires the heap to be mapped exactly up to that end
(CutRegions), checked whenever the imports or the registry changed, so a
mapping past the cut makes the set miss. DirectRegions also refuses a set any
of whose regions bound in place no longer holds the import found at its base
(a later join in the same build retired it); this applies to every region
bound in place.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
A storage view of the depth plane that starts above slice 0 was refused. The
layered storage image already spans slices 0..last, and the view starts at its
base slice, so the refusal is dropped and only the slices from the base slice to
the last are copied in and out. The base slice joins the cache key.
DepthStorageDispatch stores through a view from base slice 2 and checks the
other slices stay untouched; it fails without the change.
The storage image cache kept a quarter of the device-local heap (4 GiB on a
16 GiB card) regardless of how much of the heap the driver reports as usable.
With a desktop and other allocations the reported budget is about 10 GiB, the
cache stayed full at 4 GiB, and after minutes of a cutscene device buffer
allocations failed with VK_ERROR_OUT_OF_DEVICE_MEMORY. The cache budget is now
the smaller of that quarter of the heap and a quarter of the reported budget,
at least 512 MiB, read once per second; APS5_TEXTURE_CACHE_MIB still overrides
it. The budget report names the storage budget in use. A test covers the limits.
The MinGW-w64 lgamma overflows to infinity for nonzero |x| at or below 2^-1024. FreeBSD msun e_lgamma_r.c returns -log(|x|) for every nonzero |x| below 2^-56, so take that branch before the host function.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The guest formatter passed the flags of %o, %u, %x and %X to the host, and
MinGW's printf prefixes a space to %u when the space flag is set: "% u" of 5
gave " 5". C17 7.21.6.1 applies both flags to signed conversions only, and
FreeBSD 11.4 vfprintf clears the sign for every unsigned conversion (nosign).
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Ghost of Tsushima (PPSA02225) ends its first run, with no resource
plans in the shader cache, with about 4 GB more resident memory than a
run with a warm cache. The threads of the registration prepare pool
free what they allocate to prepare a shader, but glibc keeps the freed
memory in their arenas: at 110 s, 4.6 GB of the 6.5 GB the arenas held
was free.
When the pool has no task left, the worker that finished last calls
malloc_trim, at most once a second, which gives the free pages of every
arena back to the system.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
On Windows, piped or redirected output uses the ANSI code page, so a
trace whose thread or slice names are not in cp1252 (a Japanese thread
name, for example) made the markdown digest stop with
UnicodeEncodeError after the work was done. stdout is now set to UTF-8
when the tool runs from the command line, as check_conventions.py does.
The --json output was not affected because json.dumps escapes non-ASCII.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_012zyXkUzTZGh3nBwNMtrLaW
CONTRIBUTING.md says unimplemented exports call
NotImplemented_nid_no_patch(__func__), and 362 calls in core/ do. A stub
generated for a NID with a known name passed the NID string instead,
so its log line showed tNn5WBkta60 rather than sceAmprCommandBufferNop
and the output had to be edited by hand. Named stubs now use __func__;
stubs with an unknown name keep the NID, the only identifier they have.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_012zyXkUzTZGh3nBwNMtrLaW
relinker_executables converts the picker case with --macos too and
requires both Mach-O outputs to be executable. It fails when the
permission change is limited to Linux outputs.
The executables were written with the default file mode (0644 under a
022 umask). The documented chmod covered app.elf, but not the outputs of
--executable, so RetroArch (PPSA99169) converted with
--executable picker/picker.bin --executable es-de/es-de.bin stopped as
soon as it started its picker:
sceSystemServiceLoadExec: execv failed: Permission denied
Every executable output now gets execute permission for owner, group and
others after it is written, as --autorun already did for app.elf. With
it RetroArch starts the picker, and the picker starts ES-DE.
The Rosetta split covered only VEX scalar compares. #5642 measured the
same lost load after legacy CMPSS and CMPSD, and the decoder already
hands them to the matcher with the right length and RIP offset.
Recognise F3 or F2, segment overrides and an optional REX before
0F C2 /r ib, with the same predicate rule as the VEX forms. A 66, 67 or
LOCK prefix, or both F2 and F3, keeps the instruction inline as before.
Adrià