A dispatch that writes a resident linear target's memory directly through
a host import, and has completion work, stamps that memory a second time
when its batch completes. With the target's draws stored at completion
points, the completion can run inside the target's own store:
FlushPending moves the target to the flushing list and then waits for the
dispatch. StampKeepingNewerResults only kept block-tracked images still in
the pending list, so the second stamp made every draw recorded after the
dispatch look older than the dispatch, and the store dropped them all.
The keep rule now also covers targets with drawn results and images that
are already being flushed.
Co-Authored-By: Claude <[email protected]>
v_rcp_f32 and v_rcp_iflag_f32 lower to the driver's 1/x, whose low bits
differ from the console's on NVIDIA and Intel. With APS5_EXACT_RCP set,
normal inputs with normal results take the RDNA reciprocal datapath
instead: a cubic over 32 segments of the mantissa in 32-bit integer
arithmetic, the linear term rounded to nearest even, and a 2,761-entry
table of one-ULP corrections, looked up only when the dropped bits fall
next to the rounding cut (6.25 % of inputs). This reproduces all 2^23
mantissas of an RDNA2 (gfx1031) sweep. Other inputs keep 1/x.
It is off by default: the tables add about 13k SPIR-V words to each
shader that uses it.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
Claude-Session: https://claude.ai/code/session_017EjjFU1FjMR11oWDYs8UMo
Implement sceCesUcsProfileInitGb18030_2000 and sceCesUcsProfileInitGb18030_2005.
Two-byte cells are the CP936 table plus the cells that differ, and four-byte
sequences are runs over the BMP and the supplementary planes, generated from
ICU's gb-18030-2000.ucm (before ICU-8427) and gb-18030-2005.ucm with ICU's
GB 18030 ranges. The GB, MBC and MBCS string conversions accept both profiles
and produce four-byte UTF-8 and surrogate pairs for supplementary characters.
Since #6291 a pixel shader's EXEC copy is read as the saved word ANDed
with WQM(Ballot(true)). MaskReplaySafety only accepts a whole-quad
expansion of a ballot it can show excludes helper lanes, and a ballot
of true includes them, so a stencil-exporting shader whose own-lane
masks come from an EXEC copy was refused (own-lane mask not stable).
Every lane of a quad is active (real or helper), so a ballot of true
holds the same quad bits as the all-ones mask the proof already treats
as quad-stable; it is now classified the same way. Helper-dependent
ballots stay uncertain.
Co-Authored-By: Claude Opus 5.5 <[email protected]>
The merge resolution had dropped its storage image sentences, which main keeps and this branch does not change.
Co-Authored-By: Claude Sonnet 5.5 <[email protected]>
Table T#s keyed at runtime no longer fault on image_atomic_*32: the
table operation stays supported, elements carry the atomic flag through
dedup into the descriptor binding, and the table path binds AtomicView
like the direct path. 64-bit atomics stay unsupported, packed and BY
formats stay unsupported, and the debt entry records the relaxed
Device-scoped emission and the remaining gaps.
thuanquang