scePadReadState only returned the latest input, so a tap that began and ended between two reads was never reported. Latch the pressed buttons when input is published and fold them into the next ReadState.
StorageTextureBudget gave the storage image cache a quarter of the VK_EXT_memory_budget
budget of the largest device-local heap (capped at TextureCacheBudget, never below 512 MiB)
whatever else the heap held, while SampledTextureBudget takes what that budget leaves after
the heap's other use. On a 10 GB card the storage cache kept 2024-2170 MiB into a level load
whose depth and render target allocations then failed with VK_ERROR_OUT_OF_DEVICE_MEMORY
with the heap at its reported budget where a run logged it.
The storage image budget is now what the reported budget leaves after the heap's use other
than the cached storage images, the unused part of the sampled textures' 2 GiB floor and an
eighth of the budget; the caps at a quarter of the budget and at TextureCacheBudget and the
512 MiB floor stay. The sampled budget still reserves TextureCacheBudget for the storage
cache, at least the storage cap, and the storage budget reserves the sampled textures alive,
so neither counts bytes the other may take; while the sampled textures alive stay within
their floor the sampled budget falls first, then the storage budget, and above it both fall
together. The budget line also prints when the storage image budget moves by more than a
tenth, with its previous value, and StorageTextureCacheBudget is exported for the execution
test.
Co-Authored-By: Claude Fable 5.1 <[email protected]>
On a Unix stream socketpair the write event's data drops by at least the bytes sent while the peer has not read them, and comes back to the idle value once the peer drains them. Off Linux the test only checks that the value stays positive and never grows.
A socket's EVFILT_WRITE and sceKernelAddWriteEvent reported the whole SO_SNDBUF in data, however much was already queued. FreeBSD 11.4's filt_sowrite reports sbspace, the space still free. On Linux the queued bytes from SIOCOUTQ are now taken off SO_SNDBUF. Windows has no query for the queued bytes and macOS is not wired up, so both keep reporting SO_SNDBUF; the debt entry says so.
The fcntl tests only ever held locks in the guest's own process, where FreeBSD's lf_blocks never reports a conflict, so the host lock path that answers F_GETLK was not exercised. A forked process now holds a read lock, a write lock and a write lock to the end of the file through the host. F_GETLK with F_UNLCK reports only the write locks, F_RDLCK the same and F_WRLCK the read lock, each with the owner's pid and SEEK_SET; F_SETLK over the read lock fails with EAGAIN, and nothing is reported once the owner exits.
The waiter took the lock only after the holder unlocked, but under load it
saw the holder's lock late and measured less than half the hold time.
The holder now sets a flag just before it unlocks and the waiter requires
it once its lock is taken.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The fallback branch's merge of main brings the single-sample scan entry
after the custom MSAA sample locations entry in the AGC debt file. Keep
it after this branch's version of that entry.
main d7262bd6d6 adds the single-sample scan entry after the custom MSAA
sample locations entry in the AGC debt file. Keep it after this
branch's version of that entry.
main d7262bd6d6 adds the single-sample scan entry after the custom MSAA
sample locations entry in the AGC debt file. Keep it after this
branch's version of that entry.
fcntl threw for F_GETLK with an F_UNLCK request on the hosts that use POSIX locks, as Linux's F_GETLK rejects that type. FreeBSD 11.4's kern_fcntl accepts it and lf_getlock reports the first overlapping lock that lf_blocks lets conflict: another owner's lock, one of the two being a write lock. For an F_UNLCK request that is exactly another owner's write lock, the same set an F_RDLCK request finds, so the host is asked with F_RDLCK and the reply is written back as for the other types.
fopen_s returned ENOENT for every failed open, including an invalid mode, a directory opened for writing and the loop and long-name cases. C11 K.3.5.2.1 only asks for a nonzero value; Microsoft's CRT, which shares Dinkumware's library, returns errno. fopen_s now returns the errno fopen set.
freopen closed the stream and returned the host's errno from the failed reopen as is, with the same Linux numbers for a symlink loop or an over-long name as fopen had. It now passes it through GuestErrnoFromHost_nid_no_patch.
fopen left the host's errno in place when the host open failed, so on Linux a symlink loop reached the title as 40 and an over-long name as 36, which are EMSGSIZE and EINPROGRESS on FreeBSD. GuestErrnoFromHost_nid_no_patch maps the host value to FreeBSD 11.4's: 1 to 34 are shared except EAGAIN (35), and ELOOP, ENAMETOOLONG, ENOTEMPTY, EDQUOT, EOVERFLOW and EDEADLK get their FreeBSD numbers; anything else is EIO.
PPSA28997 (IL2CPP) calls sceLibcBacktraceGetBufferSize and
sceLibcBacktraceSelf. On Linux the host call stack is captured with
backtrace and each frame is resolved to its module and offset with
dladdr, then written as a linked list of frame records. The record
layout is inferred and listed in TechnicalDebt. Windows throws.
The offset cases compile `image_sample_cb_cl_o`, whose LOD clamp the emitter
reaches only through the device's `shaderResourceMinLod`; without it the
emitter refuses the shader (the gate #388 left in place). The test had no
capability guard, so on such a device every one of those cases failed at the
first of them, `point, wrap with negative offsets`, instead of skipping them
as `ImageSampleLodClamp` and `ImageSampleBiasA16` already do.
Found on llvmpipe (Mesa 25.0.7, Debian 13), which lacks the feature that
lavapipe reports from Mesa 25.2.8: `agc_driver_emulated_color_compare` and
its `_srgb_decode` variant both failed. The other cases still run; the note
names the cases that were skipped.
A read-only V# the build clamps (size above the descriptor range limit cut to the
mapped end, stride overhang cut to the allocation end) was compared with the template
by its raw size, so an unchanged buffer counted as moved, was refused over the limit
and the dispatch rebuilt every time; a moved one with an overhang was snapshotted past
the allocation end. The build and the move proof now take the range from one function
(ShaderResources::guestBufferBytes), compare clamped with clamped and snapshot the
clamped range.
An overlay without a descriptor set is refused like one that could not be made, and
recording one throws instead of binding the template's set.
Run 39019245350 (main f05fc7d953) failed loadexec_arguments with WinError 5 on
unlinking guest_loadexec_host.prx in TemporaryDirectory cleanup: the exited
guest process still held the module briefly. #7038 does not touch it (0 of 25
earlier main builds failed). Retry the removal for up to 10 s.
Co-Authored-By: Claude <[email protected]>
v_pk_add/mul/fma/fmac/min/max_f16 threw in f16 denormal modes 0-2. On
RDNA2 (gfx1031) bit 6 clear flushes each denormal input lane to a zero
of its sign, and bit 7 clear flushes an add result below 2^-14 and a
mul or fma result whose exact value rounded to 11 significant bits
with an unbounded exponent is below 2^-14, as the scalar forms do;
min/max flush only their inputs. Packed f16 still throws for f16
rounding other than nearest even.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
main's 81e3db853c counts each device's active library builds
(activeBuilds), so ClearPipelineLibraries waits for them. Keep it
beside this branch's joinable worker and stopping flag, which replace
started.
Aryan Sharma