fdo-mirrors/mesa
1
0
Fork 0
mirror of https://gitlab.freedesktop.org/mesa/mesa.git synced 2026-02-23 07:30:30 +01:00
Code Issues Projects Releases Packages Wiki Activity Actions 2
207678 commits 64 branches 923 tags 1.1 GiB
Commit graph

207678 commits

This branch
This branch
All branches
Author SHA1 Message Date
Faith Ekstrand
ff944106fc nvk/bitview: Add a set_field2() trait and method 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35773>
 2025-06-26 21:12:46 +00:00
Faith Ekstrand
125da7e341 nvk/bitview: Add a new ToFieldBits trait 
This makes the implementation of SetField itself more generic

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35773>
 2025-06-26 21:12:46 +00:00
Konstantin Seurer
131b917246 lavapipe: Suspend conditional rendering around ray tracing dispatches
Some checks are pending
macOS-CI / macOS-CI (dri) (push) Waiting to run
Details
macOS-CI / macOS-CI (xlib) (push) Waiting to run
Details 
cc: mesa-stable

Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35754>
 2025-06-26 20:52:29 +00:00
Konstantin Seurer
e1acffbfc0 ci: Update trace checksums 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33003>
 2025-06-26 20:12:14 +00:00
Konstantin Seurer
5f3c956f49 gallivm/nir/soa: Call lp_build_opt_nir 
Makes sure that everything is lowered and optimized.

Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33003>
 2025-06-26 20:12:14 +00:00
Konstantin Seurer
9f98ded348 gallivm/nir/soa: Cast divergent->uniform at the consumer 
The exec mask can change between the producer and the consumer. Casting
at the consumer makes sure that we end up with a valid value.

Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33003>
 2025-06-26 20:12:14 +00:00
Konstantin Seurer
1fba5739c4 gallivm/nir/soa: Include helper invocations in first_active_invocation 
So it returns a valid element if the exec mask is 0 for all active
invocations but -1 for helper invocations.

Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33003>
 2025-06-26 20:12:13 +00:00
Konstantin Seurer
405378f042 gallivm/nir/soa: Fix typo 
Reviewed-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33003>
 2025-06-26 20:12:13 +00:00
Konstantin Seurer
fe601b9293 gallivm: Implement demote and lower terminate in nir 
The current implementation does not work for terminate since loads need
to ignore the mask because of helper invocations. This can lead to
crashes.

Reviewed-by: Mary Guillemard <mary@mary.zone>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33003>
 2025-06-26 20:12:12 +00:00
Konstantin Seurer
aacfc663cb nir: Add nir_lower_halt_to_return 
This is a lowering pass that was implemented by multiple drivers.

Reviewed-by: Mary Guillemard <mary@mary.zone>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33003>
 2025-06-26 20:12:12 +00:00
mojyack
db383ceb64 mesa: reorder mesa format corresponding to GL_BGRA 
current `st_ChooseTextureFormat(..., internalFormat=GL_BGRA8, ...)`
returns `PIPE_FORMAT_R8G8B8A8_UNORM`.
this causes significant performance loss in apps that use BGRA texture
format(e.g. firefox) when transferring textures because of format
conversions, if the driver doesn't support PIPE_TEXTURE_TRANSFER_BLIT.
fix this by modifying the texture format mapping.

See Also: https://community.mnt.re/t/poor-browser-performance/2042/30

Signed-off-by: mojyack <mojyack@gmail.com>
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Backport-to: 25.1
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35678>
 2025-06-26 19:36:09 +00:00
Dave Airlie
f972affd9d nvk/hopper: set texture header version 
Taken from open-gpu-kernel-modules.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35484>
 2025-06-26 19:21:56 +00:00
Dave Airlie
c387c9fde8 nouveau/headers: generate c56f host methods. 
The newer host methods would be nice to have in dumps.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35484>
 2025-06-26 19:21:56 +00:00
Dave Airlie
1b654dc198 nouveau/winsys: Add blackwell support 
This adds Blackwell support to sm_for_chipset() and
max_warps_per_mp_for_sm()

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35484>
 2025-06-26 19:21:56 +00:00
Dave Airlie
f310b3fa6f nvk/qmd: fix shared memory size calcs for blackwell. 
I'm still not 100% happy with these, nvidia does appear to use 26
as the max/target values, which corresponds to + 7 not +1
but let's dig a bit more.

Fixes: 6c052d87b7 ("nak/qmd: Add QMD version 5.0 for Blackwell+")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35484>
 2025-06-26 19:21:56 +00:00
Dave Airlie
ec3232691b nvk/sm90: fix su atom vs atom encoding 
The suatom is encoded different to atom.

Reviewed-by: Mel Henning <mhenning@darkrefraction.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35484>
 2025-06-26 19:21:56 +00:00
Dave Airlie
59015f1c64 nak/sm70: fix imnmx on blackwell. 
The encoding was slightly wrong, this passes the hw_tests.

Reviewed-by: Mel Henning <mhenning@darkrefraction.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35484>
 2025-06-26 19:21:56 +00:00
Faith Ekstrand
9c354cd47f nak/hw_tests: Align the shared memory window to 4GB on Hopper+ 
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35484>
 2025-06-26 19:21:56 +00:00
Dave Airlie
50a312ef24 nvk: Add hopper shared memory window alignment 
This changes to need 4GB alignment on hopper and later.

Reviewed-by: Mel Henning <mhenning@darkrefraction.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35484>
 2025-06-26 19:21:56 +00:00
Dave Airlie
46c72f69f9 nvk: invalidate raster cache after setting shading rate 
on blackwell fix:
dEQP-VK.fragment_shading_rate.renderpass2.monolithic.attachment_rate.misc.two_subpass

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35484>
 2025-06-26 19:21:56 +00:00
Dave Airlie
a2518c817f nvk: SET_REFERENCE is not legal anymore. 
replace with WFI, this is possibly overkill

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35484>
 2025-06-26 19:21:55 +00:00
Dave Airlie
6fa078c0be nvk: handle hopper invalidate sequence for memory barrier 
This changed on hopper to need a MEM_OP memory barrier emitted.

Use the new headers to send the correct sequence, this is taken
from the open-gpu-kernel-modules repo, but it might be a too
large hammer.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35484>
 2025-06-26 19:21:55 +00:00
Dave Airlie
cdd01c416c nvk: add clc86f class for hopper host methods. 
This is taken from open-gpu-kernel-modules, and provides
the host interface on hopper and newer.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35484>
 2025-06-26 19:21:54 +00:00
Dave Airlie
153dc0f406 nouveau/headers: add a single field to skip parsing headers. 
Newer headers have 2 fields at 0 for one value, just ignore one
for now.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35484>
 2025-06-26 19:21:54 +00:00
Dave Airlie
337fcd31e4 nvk/nil: Use PTE_KIND_GENERIC_MEMORY for everything on Blackwell+ 
This has been confirmed for current blackwell hardware, look
out in future for changes.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35484>
 2025-06-26 19:21:54 +00:00
Dave Airlie
f6b22a54e4 nvk/nil: add texture header v2 support for hopper/blackwell 
The texture header got redefined and some fields got reworked,
deriving it from the old info doesn't seem trivial, so introduce
a second column to handle the v2 header info.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35484>
 2025-06-26 19:21:54 +00:00
Marek Olšák
1754507d49 nir: rename nir_lower_io_to_temporaries -> nir_lower_io_vars_to_temporaries 
Acked-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35760>
 2025-06-26 18:20:54 +00:00
Marek Olšák
1e03827c77 nir: rename nir_lower_io_arrays_to_elements -> nir_lower_io_array_vars_to_elements 
same for *_no_indirects

Acked-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35760>
 2025-06-26 18:20:54 +00:00
Marek Olšák
3713e2d580 nir: rename nir_lower_clip_cull_distance_arrays -> nir_lower_clip_cull_distance_array_vars 
Acked-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35760>
 2025-06-26 18:20:53 +00:00
Marek Olšák
adb17a8609 nir: move nir_recompute_io_bases into its own file 
Acked-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35760>
 2025-06-26 18:20:53 +00:00
Marek Olšák
97743980ce nir: remove unused nir_force_mediump_io & nir_unpack_16bit_varying_slots 
I think I added these.

Acked-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35760>
 2025-06-26 18:20:52 +00:00
Marek Olšák
aefea49dad nir: move lots of code from nir_lower_io.c into new nir_lower_explicit_io.c 
nir_lower_io is just for regular inputs/outputs.

Acked-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35760>
 2025-06-26 18:20:52 +00:00
Marek Olšák
5bd3e0c08c nir: move nir_assign_var_locations to freedreno (its only use) 
Acked-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35760>
 2025-06-26 18:20:52 +00:00
Marek Olšák
c8cda0dc1a nir: move nir_io_add_const_offset_to_base into its own file 
Acked-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35760>
 2025-06-26 18:20:51 +00:00
Marek Olšák
d78070ded5 nir: move nir_io_add_intrinsic_xfb_info into its own file 
Acked-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35760>
 2025-06-26 18:20:51 +00:00
Marek Olšák
12df9b3def nir: rename nir_vectorize_tess_levels -> nir_lower_tess_level_array_vars_to_vec 
Acked-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35760>
 2025-06-26 18:20:50 +00:00
Marek Olšák
2aa94caf82 nir: rename nir_lower_io_to_vector -> nir_opt_vectorize_io_vars 
Acked-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35760>
 2025-06-26 18:20:50 +00:00
Marek Olšák
944f8f6db2 nir: move nir_lower_io_vars_to_scalar into its own file 
Acked-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35760>
 2025-06-26 18:20:49 +00:00
Marek Olšák
439d805291 nir: rename nir_lower_io_to_scalar_early -> nir_lower_io_vars_to_scalar 
Acked-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35760>
 2025-06-26 18:20:49 +00:00
Ian Romanick
b83f618fb2 brw: Fully write temporary destinations
Some checks are pending
macOS-CI / macOS-CI (dri) (push) Waiting to run
Details
macOS-CI / macOS-CI (xlib) (push) Waiting to run
Details 
Consider an innocuous instruction like:

    and(1) v250:UD, g0.3<0,1,0>:UD, 4294967264u NoMask group0

If register allocation decides to spill v250, it will see this
instruction and say, "Oh no! The other components of v250 aren't set, so
I'd better add a fill before that instruction!"

But it gets even worse than that... if register coalesce decided to
merge two of these, the live range gets massively extended because the
writes don't fully initialize the value. This causes the need to spill
these registers in the first place.

Changing that instruction to SIMD16 on Xe2 or SIMD8 on other platforms
alleviates these issues.

shader-db:

Lunar Lake
total instructions in shared programs: 17118324 -> 17113191 (-0.03%)
instructions in affected programs: 93701 -> 88568 (-5.48%)
helped: 42 / HURT: 6

total cycles in shared programs: 895422566 -> 895079488 (-0.04%)
cycles in affected programs: 30111338 -> 29768260 (-1.14%)
helped: 35 / HURT: 40

total spills in shared programs: 3588 -> 3304 (-7.92%)
spills in affected programs: 285 -> 1 (-99.65%)
helped: 10 / HURT: 0

total fills in shared programs: 2218 -> 1663 (-25.02%)
fills in affected programs: 556 -> 1 (-99.82%)
helped: 10 / HURT: 0

Meteor Lake, DG2, Tiger Lake, and Ice Lake had similar results.  (Meteor Lake shown)
total instructions in shared programs: 20059218 -> 20053563 (-0.03%)
instructions in affected programs: 96938 -> 91283 (-5.83%)
helped: 43 / HURT: 6

total cycles in shared programs: 884174588 -> 883536475 (-0.07%)
cycles in affected programs: 22105268 -> 21467155 (-2.89%)
helped: 35 / HURT: 27

total spills in shared programs: 5032 -> 4679 (-7.02%)
spills in affected programs: 355 -> 2 (-99.44%)
helped: 12 / HURT: 0

total fills in shared programs: 4782 -> 4113 (-13.99%)
fills in affected programs: 671 -> 2 (-99.70%)
helped: 12 / HURT: 0

Skylake
total instructions in shared programs: 19097658 -> 19097665 (<.01%)
instructions in affected programs: 14202 -> 14209 (0.05%)
helped: 0 / HURT: 5

total cycles in shared programs: 862058109 -> 862058267 (<.01%)
cycles in affected programs: 3450244 -> 3450402 (<.01%)
helped: 7 / HURT: 11

fossil-db:

Lunar Lake
Totals:
Cycle count: 31439652246 -> 31439652272 (+0.00%)

Totals from 2 (0.00% of 707091) affected shaders:
Cycle count: 2602 -> 2628 (+1.00%)

No other Intel platforms had any fossil-db changes.

Reviewed-by: Matt Turner <mattst88@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35721>
 2025-06-26 17:59:47 +00:00
Gurchetan Singh
8a2f43c9bd util: rust: update to rustix 1.0.7 
Latest tagged release.  Fedora uses it, and for it to
work on Android we'll need to latest release so a pure
libc backend can be used.

Reviewed-by: Marcin Radomski <dextero@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35768>
 2025-06-26 17:11:41 +00:00
Serdar Kocdemir
5fbb3817ba gfxstream: Small optimization on transformDescriptorSetList 
Reduce number of possible allocations and remove unnecessary
memory initialization which will be overwritten immediately.

Test: dEQP-GLES31.functional.ssbo.layout.*

Reviewed-by: Marcin Radomski <dextero@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35768>
 2025-06-26 17:11:41 +00:00
Albert Liu
dce282e7d9 gfxstream: Add VK_EXT_memory_budget to allowlists. (mesa3d side) 
Codegen change.

Reviewed-by: Marcin Radomski <dextero@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35768>
 2025-06-26 17:11:41 +00:00
Alyssa Rosenzweig
194c717e04 agx: optimize sign fix ups 
Totals from 1787 (3.33% of 53701) affected shaders:
MaxWaves: 1620160 -> 1620096 (-0.00%); split: +0.01%, -0.01%
Instrs: 2036003 -> 2033382 (-0.13%); split: -0.16%, +0.04%
CodeSize: 13988944 -> 13971384 (-0.13%); split: -0.17%, +0.05%
Spills: 5505 -> 5504 (-0.02%)
ALU: 1514305 -> 1511676 (-0.17%); split: -0.22%, +0.05%
FSCIB: 1514305 -> 1511676 (-0.17%); split: -0.22%, +0.05%
GPRs: 200723 -> 200779 (+0.03%); split: -0.06%, +0.09%
Uniforms: 642921 -> 645743 (+0.44%); split: -0.00%, +0.44%

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35720>
 2025-06-26 16:41:55 +00:00
Alyssa Rosenzweig
6efe557718 nir/search_helpers: add has_multiple_uses helper 
heuristic for the next patch.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Mel Henning <mhenning@darkrefraction.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35720>
 2025-06-26 16:41:55 +00:00
Alyssa Rosenzweig
22272c71f2 hk: use new path for !soft fault UBO 
this is more efficient for !soft fault and lets us drop a code path. i've been
running fossils without soft fault since dev machine so that's what these stats
are based on.

Totals from 11799 (22.22% of 53090) affected shaders:
MaxWaves: 11472896 -> 11467200 (-0.05%); split: +0.09%, -0.14%
Instrs: 7032360 -> 6603558 (-6.10%); split: -6.11%, +0.02%
CodeSize: 49902322 -> 48693876 (-2.42%); split: -2.95%, +0.52%
Spills: 2817 -> 2846 (+1.03%); split: -5.75%, +6.78%
Fills: 2911 -> 3143 (+7.97%); split: -2.37%, +10.34%
Scratch: 10896 -> 11020 (+1.14%); split: -1.95%, +3.08%
ALU: 5092487 -> 5265337 (+3.39%); split: -0.58%, +3.97%
FSCIB: 5092487 -> 5265333 (+3.39%); split: -0.58%, +3.97%
IC: 1461194 -> 1387408 (-5.05%); split: -5.22%, +0.17%
GPRs: 803121 -> 805414 (+0.29%); split: -0.80%, +1.08%
Uniforms: 2942007 -> 2247619 (-23.60%); split: -24.35%, +0.75%

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35720>
 2025-06-26 16:41:55 +00:00
Alyssa Rosenzweig
63ce73a601 nir,hk: sink lowered UBOs 
this is better than doing it once we've lowered to hardware ops which makes it
more challenging to sink since then we'd have to sink the whole tree instead of
a single intrinsic.

Totals from 17617 (32.81% of 53701) affected shaders:
MaxWaves: 16863872 -> 16901504 (+0.22%); split: +0.24%, -0.02%
Instrs: 12406405 -> 12430375 (+0.19%); split: -0.15%, +0.35%
CodeSize: 87055248 -> 87180802 (+0.14%); split: -0.18%, +0.33%
Spills: 10350 -> 9301 (-10.14%); split: -11.57%, +1.43%
Fills: 5215 -> 3733 (-28.42%); split: -31.49%, +3.07%
Scratch: 113164 -> 110472 (-2.38%); split: -2.63%, +0.25%
ALU: 9552550 -> 9558513 (+0.06%); split: -0.22%, +0.28%
FSCIB: 9552545 -> 9558508 (+0.06%); split: -0.22%, +0.28%
IC: 2874032 -> 2876442 (+0.08%); split: -0.00%, +0.09%
GPRs: 1470040 -> 1459283 (-0.73%); split: -1.00%, +0.27%
Uniforms: 5113254 -> 5115158 (+0.04%); split: -0.82%, +0.85%

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Job Noorman <job@noorman.info> [NIR]
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35720>
 2025-06-26 16:41:55 +00:00
Alyssa Rosenzweig
d17dd8332e hk: sink SSBO loads 
Seems to hit instr count due to RA lottery but reduce reg pressure a LOT so,
probably worth it.

Totals from 5305 (9.88% of 53701) affected shaders:
MaxWaves: 4487168 -> 4551040 (+1.42%); split: +1.51%, -0.09%
Instrs: 6063323 -> 6178678 (+1.90%); split: -0.17%, +2.07%
CodeSize: 44356516 -> 44662024 (+0.69%); split: -0.21%, +0.90%
Spills: 7321 -> 4034 (-44.90%); split: -45.49%, +0.59%
Fills: 8522 -> 5069 (-40.52%); split: -40.87%, +0.35%
Scratch: 21940 -> 9856 (-55.08%); split: -55.17%, +0.09%
ALU: 4541440 -> 4554906 (+0.30%); split: -0.20%, +0.50%
FSCIB: 4541440 -> 4554906 (+0.30%); split: -0.20%, +0.50%
IC: 1180150 -> 1184856 (+0.40%); split: -0.01%, +0.41%
GPRs: 605023 -> 576336 (-4.74%); split: -5.00%, +0.26%
Uniforms: 1421832 -> 1425367 (+0.25%); split: -0.08%, +0.33%

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35720>
 2025-06-26 16:41:55 +00:00
Alyssa Rosenzweig
776f07517f hk: optimize load_global_bounded 
Totals from 5967 (11.24% of 53090) affected shaders:
MaxWaves: 5044864 -> 5126720 (+1.62%); split: +1.95%, -0.32%
Instrs: 6911845 -> 6227557 (-9.90%); split: -9.96%, +0.06%
CodeSize: 46353592 -> 45359410 (-2.14%); split: -4.41%, +2.27%
Spills: 18002 -> 7046 (-60.86%); split: -63.22%, +2.36%
Fills: 15833 -> 8074 (-49.01%); split: -54.59%, +5.58%
Scratch: 52160 -> 23324 (-55.28%); split: -58.32%, +3.04%
ALU: 4235569 -> 4602546 (+8.66%); split: -1.79%, +10.45%
FSCIB: 4235364 -> 4602341 (+8.66%); split: -1.79%, +10.45%
IC: 1261552 -> 1253110 (-0.67%); split: -4.33%, +3.66%
GPRs: 664060 -> 640627 (-3.53%); split: -4.69%, +1.16%
Uniforms: 1568288 -> 1505664 (-3.99%); split: -10.88%, +6.89%
Loops: 4669 -> 4670 (+0.02%)

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35720>
 2025-06-26 16:41:54 +00:00
Alyssa Rosenzweig
caa0854da8 nir: plumb load_global_bounded 
this lets the backend implement bounded loads (i.e. robust SSBOs) in a way
that's more clever than a full branch. similar idea to
load_global_constant_bound which should eventually be merged into this.

Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Job Noorman <job@noorman.info>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35720>
 2025-06-26 16:41:53 +00:00