Commit graph

7 commits

Author SHA1 Message Date
Georg Lehmann
8f4de30d05 aco/insert_fp_mode: don't skip setting round for fract
Some checks are pending
macOS-CI / macOS-CI (dri) (push) Waiting to run
macOS-CI / macOS-CI (xlib) (push) Waiting to run
fract(-FLT_MIN) is < 1.0 with rtz but 1.0 with rtne.

Fixes: 7212a75c5e ("aco/insert_fp_mode: exclude some instructions that will never round")

Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40078>
2026-02-26 00:20:02 +00:00
Rhys Perry
e59a0df302 aco/insert_fp_mode: remove incorrect assertion
This can happen if a loop has no continues, and the later code should work
fine in this situation.

This fixes war_thunder/0013a69e097b2471 on navi21.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Fixes: 6b9d28ab9b ("aco/insert_fp_mode: insert fp mode in reverse")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39481>
2026-01-26 08:57:33 +00:00
Georg Lehmann
6b9d28ab9b aco/insert_fp_mode: insert fp mode in reverse
This allows us to skip the mode set by changing
the initial mode in the command stream.

Foz-DB Navi48:
Totals from 14 (0.02% of 82405) affected shaders:
Latency: 79417 -> 79438 (+0.03%)

Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/38815>
2026-01-20 14:48:22 +00:00
Georg Lehmann
7212a75c5e aco/insert_fp_mode: exclude some instructions that will never round
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/38815>
2026-01-20 14:48:22 +00:00
Georg Lehmann
d6356191b9 aco: add fma_mix opcodes with rtz fp16 rounding
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/38815>
2026-01-20 14:48:22 +00:00
Konstantin Seurer
39d58a55a7 aco: Add support to f2f16 with rtpi/rtni
Those rounding modes are needed when computing 16-bit bounding boxes
since the bounding box must not get smaller.

Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37883>
2026-01-10 11:34:12 +01:00
Georg Lehmann
46c1bd1147 aco: add a dedicated pass for better float MODE insertion
Foz-DB Navi48:
Totals from 14 (0.02% of 80251) affected shaders:
Instrs: 13998 -> 11684 (-16.53%)
CodeSize: 104464 -> 86260 (-17.43%)
Latency: 108722 -> 106667 (-1.89%)
InvThroughput: 100332 -> 100324 (-0.01%)
VClause: 621 -> 595 (-4.19%); split: -4.99%, +0.81%
VALU: 6875 -> 6871 (-0.06%)
SALU: 3256 -> 1015 (-68.83%)
VOPD: 1328 -> 1332 (+0.30%)

Removes the s_setreg spam in FSR4.

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35746>
2025-07-10 13:48:50 +00:00