The recompiler refused centre-located barycentrics in a multisampled
merge on a device without interpolation functions and fragment shader
barycentrics. It did so even when every sample sits at the pixel
centre, as in a draw raised from a one-sample rasterizer, where the
barycentrics are the centre's anyway. The refusal now applies only
when the merge's sample locations are off the centre. The pixel
request carries those locations for such a merge.
The recompile test runs on two states. On a 2-sample rasterizer, the
missing barycentrics throw, and the shader compiles once they are
present. On a raised draw, the shader compiles without them.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
A merge into a multisampled 4_4_4_4 target runs the pixel shader once
per sample. A shader that stores to guest memory, runs atomics or uses
GDS would then repeat those writes for every covered sample. An
8-sample pixel would add to a counter eight times. Such a merge now
throws.
The recompile request carries the merge's sample count, not whether
the guest asked for per-sample shading, so this also refuses shaders
the guest iterates itself, where the console repeats the writes too.
A recompile test checks that a 2-sample merging shader adding to a
buffer counter throws, and that the same shader without the merge
compiles.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
A partial write of a multisampled 4_4_4_4 target makes the draw shade
per sample. For a shader the guest runs per pixel, every invocation of
a pixel must then see the same inputs. Otherwise its colours and kill
decisions follow each sample, while the console runs one invocation.
The decode now throws in two cases when the guest did not ask for
per-sample shading. The first is a shader that reads the ancillary
word, whose sample number would become the invocation's. The second
is a shader that reads centroid, sample-located or pull-model
barycentrics, Z, 1/W, or a position at the centroid or a sample. That
case is allowed when every sample sits at the pixel centre, as in a
draw raised from a one-sample rasterizer. The recompiler throws for
centre-located barycentrics on a device without interpolation
functions and fragment shader barycentrics, where they would not be
rebuilt at the centre.
Decode tests run each moving input on a 2-sample rasterizer, where it
throws, and with PS_ITER_SAMPLE, where it decodes. The raised draw
keeps the located inputs and refuses the ancillary word. Centre and
upper-left positions, centre barycentrics, the front face and the
fixed-point position decode, and a recompile test covers the missing
barycentrics. The debt entry now lists what is kept and what throws.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
Per-sample shading of a merging draw makes centroid and sample-located
inputs and the ancillary sample number follow the invocation's sample
(centre-located inputs too on devices without interpolation functions
and fragment shader barycentrics), and the 2-sample merge was measured
on one GPU only.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
The refusal of a merging shader that writes guest memory under late
tests fired in two cases where the early tests change nothing.
The first was a draw with no depth, stencil or depth-bounds test. Early
tests reject no pixel there, so every write the console makes still
happens. The target merge now carries whether the draw has such a
test, and the refusal applies only then. This adds one byte to the
version 40 merge fields of the recompile request.
The second was any GDS binding, including a shader that only reads GDS.
The refusal now looks for GDS writes, atomics, appends and consumes.
Tests: the depth and stencil decode checks the new flag, and the
static shader key depends on it. A counter-adding shader under late
tests compiles without a test and throws with one. A late-test merge
that reads GDS compiles, and one that adds to GDS throws.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
A merged 4_4_4_4 write runs its pixel shader after early depth and
stencil tests. Vulkan counts a sample for an occlusion query at those
tests, so pixels the guest shader kills under late tests would be
counted, and on the console they are not. Once the game has read an
occlusion counter the recorder counts samples for every later draw,
so from then on such a merge throws. The check runs where the recorded
and recipe draw paths prepare their inputs. The state records
DB_SHADER_CONTROL.KILL_ENABLE for it, which fits in the existing
padding.
The execution test starts sample counting and then checks two cases: a
merge that kills a checkerboard under late tests is refused, and a
merge that kills nothing still draws. A decode test reads KILL_ENABLE.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
A partial 4_4_4_4 write merges inside a pixel shader that declares
early fragment tests. When the guest runs the shader under late tests,
pixels that the depth or stencil test rejects still execute it on the
console, so their buffer and image stores, atomics and GDS operations
happen there. Under forced early tests they would be dropped. Such a
merge now throws, unless the guest asked for early tests itself. The
merged target's own image write does not count.
A recompile test checks two things for a merging shader that adds to a
buffer counter: under late tests it throws, and under the guest's early
tests it compiles with its one atomic add beside the merge. The same
shader without a merge still compiles under late tests.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
A one-sample target under a multisampled rasterizer whose
OVERRASTERIZATION_AMOUNT matches (radeonsi line and polygon smoothing)
decodes with mergedCoverage and is drawn through mixed-samples coverage
reduction onto its one-sample attachments. A merging draw leaves the
4_4_4_4 target out of the attachments, a combination that is not tested,
so a partial write of such a draw now throws; the limit is listed in the
technical debt file.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
Occlusion counts are taken at the forced early tests, the refused merge
draws and why a late depth write was not used, and the one GPU the merge
was measured on.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
A CB resolve from a packed color target throws, so a 2-sample 4_4_4_4
target cannot be resolved by the color block.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
FreeBSD 11.4 kern_mprotect masks the argument with VM_PROT_ALL instead of
refusing it (sys/vm/vm_mmap.c:613). sceKernelMprotect already does this
since #5770; the POSIX mprotect export still failed with EINVAL.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
libusb's CMake generates version_describe.h by running gen-describe.ps1
with -ExecutionPolicy Bypass. An execution policy set by Group Policy
takes precedence over that switch, so with Restricted or AllSigned the
Windows build stops at version_describe.h. Describe the failure and a
.cmd stand-in, passed as LIBUSB_POWERSHELL_EXECUTABLE, that runs
gen-describe.sh with Git for Windows' bash as the PowerShell script does.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
On Windows the UCRT does not switch an r+, w+ or a+ stream from writing
to reading without fflush or fseek: fread after fwrite returned bytes of
the write buffer and the next flush wrote them over the file, and ungetc
failed on an update stream that was not already reading. On Linux glibc
does not switch either for ungetc, which pushes the byte back over the
pending write, drops the write and can corrupt the heap on the next read,
or for an fread of at least the buffer size, which drops the write and
reads from the old offset.
fread, fgetc, getc_unlocked, fgets, ungetc, ungetwc and fscanf now flush
pending output and mark the stream as reading first, as FreeBSD 11.4's
__srefill and __ungetc do, through _IO_switch_to_get_mode on glibc.
Writes directly after reads still go through the UCRT on Windows; the
gap is recorded in the libc technical debt.
FreeBSD 11's kern_pipe opens both ends with FREAD | FWRITE, so F_GETFL
reports O_RDWR on each. The host pipe ends are one-way, so the access
mode came back as O_RDONLY and O_WRONLY. Descriptors made by pipe() are
now marked in the guest descriptor table, the mark follows dup, dup2 and
F_DUPFD, and F_GETFL reports O_RDWR for them.
A guest wave's lanes see each other's LDS writes without a barrier, so
the recompiler adds a subgroup-scope OpControlBarrier between an LDS
write and the next LDS access when the guest wave is one host subgroup
(WaveLdsScope). AMD's proprietary Windows driver 32.0.21042.62 (Vulkan
driver 2.0.353, "AMD proprietary shader compiler", Radeon 0x13c0)
compiles that barrier to s_barrier, as the pipeline's internal ISA
shows (26 s_barrier for the guest's 9 in a small LDS reduction). In a
workgroup of several waves, the waves that enter a branch holding LDS
accesses then execute extra s_barriers, which pair with the other
waves' barriers, and those waves read LDS before it is written. LLPC
compiles a subgroup-scope control barrier to a wave barrier instead.
PS5LM Plus (PPSA99581) answered with garbage on that driver: the first
compute dispatch of its first graph, an RMS norm with a 512-thread
wave32 LDS tree reduction, scaled every row except its first 32
elements by a partial sum, while NVIDIA matched a NumPy reference and
the inputs were identical.
On that driver the barrier is now a subgroup-scope OpMemoryBarrier,
which it compiles to no instruction. A 32-lane subgroup is one hardware
wave there, which runs its lanes in lockstep and issues its LDS
instructions in order; Vulkan does not guarantee that, so it is
recorded as debt. Other drivers keep the control barrier. The flag is
part of the shader cache key, and agc_shader_replay takes
--wave-lds-memory-barriers to replay a request the way that driver
compiles it.
agc_driver_lds_tree_reduction sums over eight wave32 waves the way the
guest does, with its last four steps inside one wave and no s_barrier
between them: before, thread 32 read 1936 instead of 32896 on the AMD
driver; it passes on AMD and NVIDIA now. agc_subgroup_barrier checks
the driver quirk and the emitted barriers on every runner. PS5LM Plus
now answers " Paris. It is the largest city in Europe ..." on the AMD
iGPU without flash attention, as on NVIDIA and the CPU.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
lstat and readlink threw NotImplemented for every Windows reparse point
other than a symbolic link or a drive junction. A tag without the name
surrogate bit (cloud files, deduplication) does not stand for another path,
so lstat now answers as stat does and readlink returns EINVAL, as FreeBSD
does for a file that is not a link and as CPython's os.lstat and os.readlink
do (bpo-37834). A name surrogate that is neither a symbolic link nor a
junction (a mounted volume, a WSL link) is reported as the reparse point
itself with the mode of its attributes, and readlink returns EINVAL.
The tests set reparse tags with FSCTL_SET_REPARSE_POINT, one without the
name surrogate bit and one with it. A compact /exe (WOF) file cannot serve
as the fixture: on Windows 11 24H2 the WOF filter hides the reparse point
from GetFileAttributesW and FileAttributeTagInfo alike. Windows refuses
every open of an unhandled tag without FILE_FLAG_OPEN_REPARSE_POINT, so the
test requires lstat to agree with stat rather than to succeed.
mashingaan's boykopovar/AnyPS5#6447 returned EINVAL for other reparse
points without the tag check.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Co-Authored-By: Claude Fable 5.1 <[email protected]>
d3011ae11 made dup and dup2 of a Windows directory descriptor share
its DirectoryState, as FreeBSD's kern_dup shares the open file, and
removed the clause; a later merge of the entry brought it back.
GuestDirectoryRewind covers both calls.
When the graph is built without the wave size and the return address
escapes under only one of wave32 and wave64, the graph records that size
and its error instead of throwing, and translation throws it only at
that size. A wave64 program that reads a compare mask written over a
call's link pair from a scalar instruction is now recompiled, while the
same program at wave32, where the high dword still holds the return
address, is refused.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The raw-byte check that runs before a syscall pattern between PS4 unwind
functions is skipped now also accepts mov rax, imm64, merges a following
mov al, imm8 or mov ax, imm16 into the earlier constant, and walks back
over any number of register moves or zeroings that leave rax alone
instead of at most seven. Visited states keep the walk linear in the
bytes it covers.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Linux's udp_sendmsg refuses port 0 with EINVAL, so a send that FreeBSD
accepts (the socket is bound to a specific address and port and the
destination is not INADDR_ANY or INADDR_BROADCAST) failed there. Nothing can
receive on port 0, so the call now returns the length itself on every host.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
udp_output hands the destination to in_pcbconnect_setup unless the socket is
bound to a specific address and port and the destination is neither
INADDR_ANY nor INADDR_BROADCAST, and in_pcbconnect_setup fails port 0 with
EADDRNOTAVAIL. The send went to the host, which accepted it. sendto, and
sendmsg through it, now fails the same way for an AF_INET datagram socket
that is not connected. connect to port 0 already did.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
