On RDNA 2, an S# with TRUNC_COORD 0 rounds the scaled coordinate to 8
fractional bits, ties to even, and one with TRUNC_COORD 1 truncates it,
for point and bilinear filters. The S# decoder ignores the bit, and
AMD's Windows driver truncates for nearest samplers and rounds for
linear ones, so a TRUNC_COORD 0 point sample just below a texel edge
reads the texel below it. The code comment said the bit moves texel
selection by half a texel at most; the debt entry records the measured
rule instead.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Capture evaluated every flattened scalar slot even when a clean scalar
branch skipped it, so an inaccessible direct user-data root refused the
draw. Record flat-slot uses on resource control-flow blocks and evaluate
only the reachable ones; keep refusing taken and unresolved paths.
Fixes#8272.
Implement sceRudpPollCreate, sceRudpPollControl, sceRudpPollWait and
sceRudpPollDestroy with the PS Vita signatures and argument checks.
sceRudpTerminate removes the context from every poll and sceRudpEnd
destroys the polls and wakes their waiters.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
scePsmlMfsrGetDispatchMfsrPacketSizeInDwords, scePsmlMfsrGetMipmapBias,
scePsmlMfsrSelectConfig, scePsmlMfsrRequestCapture and
scePsmlMfsrIsCaptureInProgress gain real signatures and semantics reversed
from the fakelib/libScePsml.sprx bundled with PPSA20397, which imports the
first three. The context records the input dimensions at creation so the
mipmap bias can be computed.
readdir and readdir_r returned d_fileno 0 for every entry on Windows,
because MinGW's _wreaddir reports no inode. FreeBSD 11.4's
_readdir_unlocked (lib/libc/gen/readdir.c) never returns an entry whose
d_ino is 0, and d_fileno is the number lstat reports as st_ino.
The Windows directory stream now reads the listing from the directory
handle with GetFileInformationByHandleEx(FileIdBothDirectoryInfo), so
each entry carries the file ID of the entry itself, symbolic links
included, without opening it. The ID is folded as stat and lstat fold
nFileIndexHigh:nFileIndexLow; the fold moves to WindowsDescriptor.hpp so
both libraries use it. rewinddir restarts the scan, telldir and seekdir
count entries as _wtelldir and _wseekdir did, and opendir fails with the
open error of the directory. A file system that lists no file ID gets a
nonzero number hashed from the name. Linux truncates the host inode to
32 bits as stat does instead of reporting 0 above 2^32.
guest_directory checks that every entry has a nonzero file number, that
the numbers of a file, a subdirectory and symbolic links to both equal
lstat's st_ino, that an entry created or removed while the directory is
open shows up after rewinddir, that a delete-pending entry no lstat can
open is still listed, and that seekdir returns the same number.
#7018 reads the low-half factors of v_mul_lo_u16 and v_pk_mul_lo_u16
as 24-bit fields because AMD's Windows driver stores the full 32-bit
product when it knows both factors fit in 16 bits. #7673's fold of
BitFieldUExtract(x, 0, n) into x removed those reads when the program
had masked the factors to 16 bits, so agc_driver_mul_lo_u16_high_half
failed again on that driver. The folder now keeps a field read that a
multiply uses.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Report the host IPv4 DNS servers of the interface sceNetCtlGetInfo
uses, through a helper libSceNetCtl exports, and return their count.
The signature and layout follow the OpenOrbis musl resolver and Net.h.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The blocked-destination case checks a different file, so an in-place
rewrite passed it. The earlier case is back beside it: a directory at
<name>.tmp makes the write fail, and the previous aps5-bench.json.gz must
be unchanged. It is left out of Windows builds, where the
std::ios_base::failure thrown inside the MinGW ofstream did not unwind to
the test's catch.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
sceNetGetSockInfo returned 0 for a known socket and wrote nothing, so
callers read uninitialised bytes as the socket's endpoints. Monster
Hunter Wilds (PPSA24191) calls it after bind, connect and accept and
stores the local and remote address and port it reads back.
It now writes one 0xA0-byte record in the layout SharpProspero and
shadPS4 give: the socket id, the socket type, the local and remote
IPv4 address and port of the host socket in network byte order, the
SELF and NONBLOCK status flags, and the receive and send buffer sizes
sceNetGetsockopt reports for the socket. The other fields are 0. It
returns 1, the number of records written, and EINVAL for a null record
or an n below 1, as shadPS4 does.
An s below 0 lists the sockets as shadPS4 does: with a null record it
returns their number, otherwise it writes up to n records in socket
order and returns how many it wrote. Unlike shadPS4, which lists IPv6
sockets with zero addresses and ignores flags, an IPv6 socket fails
with EAFNOSUPPORT and is left out of the count and the list, and a
flags bit other than 0x40 fails with EINVAL. SharpProspero names 0x40
SockInfoTimeWait; closed sockets leave the table at once, so it lists
the same sockets. docs/dev/debt/libSceNet.md lists the fields left 0
and these differences.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
sceNgs2RackLock took the library's internal recursive mutex and kept it
until sceNgs2RackUnlock, and a render held the same mutex while it called a
voice's block-end or repeat callback. Every NGS2 call from another thread
waited behind either one. A title that takes one of its own locks while it
holds a rack lock (or inside the callback), while another of its threads
holds that lock and calls sceNgs2VoiceGetStateFlags, deadlocks.
A rack lock is now an owner and a depth on the rack, and only a render of
its system waits for another thread's rack lock. The voice callback runs
with the internal lock released (the lock depth is tracked per thread so it
is released fully and taken back after). A rack or system destroy waits
while another thread renders its system, as the callback window would
otherwise let it free what the render still walks.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Yuri Olive