K11MCH1 c437d89116 Emit pending flushes where a suspended render pass chain starts (#14) ...

tu/0001 (by K11MCH1): a dynamic rendering chain still suspended at
vkEndCommandBuffer() is rendered at submit time from a new command
buffer with no pending flushes, so the waits of barriers recorded just
before the chain were dropped. On A8XX, where the CP no longer waits
implicitly before reading indirect arguments, indirect draws in
vkd3d-proton's split render passes read their arguments before the
compute pass wrote them: black or over-shadowed ground from UE5 Nanite
and virtual shadow maps (Mortal Shell 2). Adreno 840: a draw-count test
with the chain split across two command buffers went from 30/30 wrong
to 0/30.

Applies to Mesa main fd8ef2b for android and linux, and both drivers
build.

2026-11-02 11:30:45 -05:00

..

aimapper

Ship license notices in the bundle and the Corresponding Source beside it

2026-09-30 16:40:30 -04:00

fdl

Keep the 32-bit SSBO offset wrap and the A8XX buffer descriptor size (#13)

2026-10-31 09:25:10 -04:00

linux

Follow Mesa main: pass the indirect draw stream to tu6_emit_vs_params in the mesh draw

2026-10-24 15:21:30 -04:00

lrz

Fix A8XX GPU hangs after a large fast-cleared LRZ buffer (#2)

2026-10-16 15:14:07 -04:00

nir

Keep the 32-bit SSBO offset wrap and the A8XX buffer descriptor size (#13)

2026-10-31 09:25:10 -04:00

scripts

Disable category-4 repeat on the A845 as well

2026-10-24 15:25:58 -04:00

tu

Emit pending flushes where a suspended render pass chain starts (#14)

2026-11-02 11:30:45 -05:00

series.json

Emit pending flushes where a suspended render pass chain starts (#14)

2026-11-02 11:30:45 -05:00