Samuel Pitoiset
41cbd6f735
radv: rework declaring color arguments for PS epilogs
...
Rely on the actual FS outputs instead of the spi shader col format,
this is safer and it will help for future work.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26903 >
2024-01-10 16:10:15 +00:00
Samuel Pitoiset
58d2a78dba
radv: move dri options to radv_instance::drirc
...
To make it clearer that such an option is a per-application option.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26946 >
2024-01-10 10:07:40 +00:00
Samuel Pitoiset
1854d03c20
radv: query drirc options in only one place
...
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26946 >
2024-01-10 10:07:40 +00:00
Tatsuyuki Ishi
71e486d1cf
radv: Add layer to skip UnmapMemory for Quantic Dream Engine
...
Detroit: Become Human has an optimization issue where the same BO is
mapped and unmapped on a per-frame basis, which leads to massive
overhead in the kernel, both creating the mapping and taking page
faults.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26942 >
2024-01-10 07:39:58 +00:00
Friedrich Vock
0b55a3cf64
radv/rt: Acceleration structure updates
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26729 >
2024-01-09 18:47:45 +00:00
Friedrich Vock
62fe4f0b1b
radv/rt: Move per-geometry build info into a geometry_data struct
...
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26729 >
2024-01-09 18:47:45 +00:00
Samuel Pitoiset
4c7486c9aa
radv/winsys: replace '<= GFX6' by '== GFX6'
...
GFX6 is the first AMD family supported by RADV.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26867 >
2024-01-09 16:31:31 +00:00
Samuel Pitoiset
ae4628d3d6
radv: do not program COMPUTE_MAX_WAVE_ID (GDS register) on GFX6
...
Ported from RadeonSI c2359797 .
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26867 >
2024-01-09 16:31:30 +00:00
Konstantin Seurer
658ce711d5
radv/rt: Lower ray payloads to registers
...
This should allow for cross stage optimizations and it reduces latency
caused by scratch access.
Totals from 44 (9.69% of 454) affected shaders:
MaxWaves: 432 -> 436 (+0.93%)
Instrs: 2740662 -> 1610327 (-41.24%); split: -41.24%, +0.00%
CodeSize: 14616932 -> 8573620 (-41.34%)
VGPRs: 4880 -> 4816 (-1.31%)
SpillSGPRs: 464 -> 294 (-36.64%)
Latency: 18548886 -> 11465281 (-38.19%); split: -38.19%, +0.00%
InvThroughput: 5195964 -> 3066729 (-40.98%); split: -40.98%, +0.00%
VClause: 99672 -> 55611 (-44.21%)
SClause: 65827 -> 38697 (-41.21%)
Copies: 231231 -> 137676 (-40.46%); split: -40.47%, +0.01%
Branches: 111379 -> 65865 (-40.86%); split: -40.87%, +0.00%
PreSGPRs: 3854 -> 3812 (-1.09%); split: -1.19%, +0.10%
PreVGPRs: 4518 -> 4439 (-1.75%); split: -1.84%, +0.09%
Reviewed-by: Friedrich Vock <friedrich.vock@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26431 >
2024-01-09 13:02:11 +00:00
Konstantin Seurer
2e4951d3fb
radv: Remove the BVH depth heuristics
...
It only helps Quake II RTX and hurts everything else.
Reviewed-by: Friedrich Vock <friedrich.vock@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26481 >
2024-01-09 09:00:24 +00:00
Konstantin Seurer
719619c477
radv: Use PLOC for TLAS builds
...
Improves control performance by about 1%.
Reviewed-by: Friedrich Vock <friedrich.vock@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26481 >
2024-01-09 09:00:24 +00:00
Dave Airlie
71bd479a7f
radv: don't emit cp dma packets on video rings.
...
Only emit this on the gfx/ace rings.
Fixes hangs with CTS on video decode with navi3x.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26945 >
2024-01-09 07:39:52 +00:00
Konstantin Seurer
da647e7e42
radv/rt/rmv: Log pipeline library creation
...
Pipeline libraries own shaders which take up GPU memory.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26668 >
2024-01-08 19:29:13 +00:00
Konstantin Seurer
84cc494e51
radv/rmv: Fix tracing ray tracing pipelines
...
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26668 >
2024-01-08 19:29:13 +00:00
Timur Kristóf
55e5c4e089
radv: Expose transfer queues, hidden behind a perftest flag.
...
This is highly experimental and only recommended
for users who know what they are doing.
To fully support the spec we are going to need
gang submissions which are going to be implemented later.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Tatsuyuki Ishi <ishitatsuyuki@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26913 >
2024-01-08 16:00:19 +01:00
Timur Kristóf
bd3f2567cc
radv: Implement T2T scanline copy workaround.
...
The built-in tiled-to-tiled copy packet doesn't support copying
between images that don't meet certain criteria such as alignment,
micro tile format, compression state etc.
To work around this, we copy the image piece by piece to a
temporary buffer that we know is supported,
and then copy it to the intended destination.
The implementation assumes that at least one pixel row of the
image fits into the temporary buffer, and will try to copy as
many rows as fit.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Tatsuyuki Ishi <ishitatsuyuki@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26913 >
2024-01-08 16:00:15 +01:00
Timur Kristóf
a4b4c9b723
radv: Implement image copies on transfer queues.
...
When either of the images is linear then the implementation can
use the same packets as used by the buffer/image copies.
However, tiled to tiled image copies use a separate packet.
Several variations of tiled to tiled copies are not supported
by the built-in packet and need a scanline copy as a workaround,
this will be implemented by an upcoming commit.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Tatsuyuki Ishi <ishitatsuyuki@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26913 >
2024-01-08 16:00:10 +01:00
Timur Kristóf
1405f9b68b
radv: Correct binding index for transfer buffer-image copies.
...
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Tatsuyuki Ishi <ishitatsuyuki@gmail.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26913 >
2024-01-08 15:58:37 +01:00
Samuel Pitoiset
907afddf97
radv: stop disabling DCC for mutable with 0 formats on GFX11
...
On GFX11, all formats are DCC compatible, so we can completely ignore
MUTABLE with a missing formats list.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25664 >
2024-01-08 08:02:58 +00:00
Samuel Pitoiset
fbe4e16db2
Revert "radv: disable DCC with signedness reinterpretation on GFX11"
...
This was affecting Cyberpunk and A Plague Tale Requiem but both issues
should be fixed now. The issue with A Plague Tale Requiem was because
of a game bug and vkd3d-proton now has a workaround.
This reverts commit e6735409ee .
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25664 >
2024-01-08 08:02:58 +00:00
Konstantin Seurer
ba7b08e324
radv/rt: Repurpose radv_ray_tracing_stage_is_compiled
...
Replace it with radv_ray_tracing_stage_is_always_inlined and use it inside
radv_rt_compile_shaders.
Reviewed-by: Friedrich Vock <friedrich.vock@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25224 >
2024-01-07 21:28:32 +01:00
Konstantin Seurer
73cc952870
radv/sqtt: Avoid duplicate stage check
...
Reviewed-by: Friedrich Vock <friedrich.vock@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25224 >
2024-01-07 21:28:26 +01:00
Konstantin Seurer
77b9a6f9e2
radv/rt: Use radv_shader for compiled shaders
...
Reviewed-by: Friedrich Vock <friedrich.vock@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25224 >
2024-01-07 21:28:19 +01:00
Konstantin Seurer
59d490b8aa
radv/rt: Remove useless assert
...
If it's NULL, the code will segfault anyways.
Reviewed-by: Friedrich Vock <friedrich.vock@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25224 >
2024-01-07 21:28:11 +01:00
Konstantin Seurer
8198805e1f
radv: Skip compiling chit and miss shaders
...
Reviewed-by: Friedrich Vock <friedrich.vock@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25224 >
2024-01-07 21:28:06 +01:00
Konstantin Seurer
0f87d406b5
radv/rt: Skip compiling a traversal shader
...
If we don't need one, we don't compile one.
Reviewed-by: Friedrich Vock <friedrich.vock@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25224 >
2024-01-07 21:28:02 +01:00
Konstantin Seurer
aaa64217ca
radv: Add more ray tracing data to the cache
...
This makes the cache more flexible when it comes to missing stages. This
will be used to skip compiling unused ray tracing stages.
Reviewed-by: Friedrich Vock <friedrich.vock@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25224 >
2024-01-07 21:27:56 +01:00
Konstantin Seurer
a784477269
radv: Don't store library stack sizes
...
They are already imported in radv_rt_fill_stage_info.
Reviewed-by: Friedrich Vock <friedrich.vock@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25224 >
2024-01-07 21:27:53 +01:00
Konstantin Seurer
92a951db6a
radv: Make pipeline cache object data generic
...
Pipeline cache objects can hold some generic data. Anything concerning
that should not be handled in "common" code paths.
Reviewed-by: Friedrich Vock <friedrich.vock@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/25224 >
2024-01-07 21:27:27 +01:00
Yonggang Luo
d6c258d9ee
util: Add align_uintptr and use it treewide to replace ALIGN that works on size_t and uintptr_t
...
Signed-off-by: Yonggang Luo <luoyonggang@gmail.com>
Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26866 >
2024-01-05 21:54:35 +00:00
Rhys Perry
5fd747a502
radv: enable msad_4x8
...
This helps some FSR3 shaders.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26907 >
2024-01-05 18:55:22 +00:00
Konstantin Seurer
c511b8968a
radv: Implement VK_KHR_ray_tracing_position_fetch
...
Reviewed-by: Friedrich Vock <friedrich.vock@gmx.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26895 >
2024-01-05 18:20:20 +00:00
Rhys Perry
24ef827f71
radv: remove radv_shader_info's cs.subgroup_size
...
This is the same as wave_size.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26894 >
2024-01-05 17:35:48 +00:00
Rhys Perry
59dbe633e3
radv: use CS wave selection for task shaders
...
This uses wave32 for small workgroups and wave64 when certain subgroup
operations are used.
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26894 >
2024-01-05 17:35:48 +00:00
Rhys Perry
3009dcd102
aco: correctly set min/max_subgroup_size for wave32-as-wave64
...
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26894 >
2024-01-05 17:35:48 +00:00
Friedrich Vock
1e3541728b
radv,aco: Convert 1D ray launches to 2D
...
Because we use unaligned dispatches, 1D launches only use 8 threads per
wave. Converting to 2D and fixing up launch IDs in the prolog
significantly increases occupancy.
Gives ~30% uplift in Ghostwire Tokyo.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26105 >
2024-01-05 17:08:05 +00:00
Friedrich Vock
34cc765172
radv/rt: Free traversal NIR after compilation
...
Could lead to OOM in games that compile RTPSOs often,
like Ghostwire Tokyo.
Cc: mesa-stable
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26892 >
2024-01-04 14:48:28 +00:00
Samuel Pitoiset
395dee0d89
radv: drop si_ prefix from all functions
...
Most of these functions were copied from RadeonSI but they should be
prefixed with radv_ instead.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26854 >
2024-01-04 08:40:37 +00:00
Samuel Pitoiset
89947eb151
radv: remove radv_write_scissors()
...
This function is useless.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26854 >
2024-01-04 08:40:36 +00:00
Samuel Pitoiset
dd64293104
radv: rename si_make_texture_descriptor() to gfx6_make_texture_descriptor()
...
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26854 >
2024-01-04 08:40:36 +00:00
Samuel Pitoiset
887ac765bb
radv: remove duplicated si_tile_mode_index() function
...
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26854 >
2024-01-04 08:40:36 +00:00
Samuel Pitoiset
aa1eb54b3b
radv: constify a variable in radv_emit_depth_control()
...
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26847 >
2024-01-04 07:54:41 +00:00
Samuel Pitoiset
8b7b5be98b
radv: disable stencil test without a stencil attachment
...
Implementations are supposed to do that per the Vulkan spec.
This fixes the following new VKCTS tests
dEQP-VK.pipeline.*.stencil.no_stencil_att.*
Cc: mesa-stable
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26847 >
2024-01-04 07:54:40 +00:00
Daniel Schürmann
bdbf873b0f
nir: remove redundant passes from nir_opt_if()
...
These are now covered by nir_opt_loop():
- opt_if_loop_last_continue()
- opt_merge_breaks()
- opt_if_loop_terminator()
Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24940 >
2024-01-03 20:48:05 +00:00
Daniel Schürmann
a3ed36da1a
treewide: replace calls to nir_opt_trivial_continues() with nir_opt_loop()
...
Totals from 850 (1.11% of 76636) affected shaders: (RADV, GFX11)
MaxWaves: 18134 -> 18130 (-0.02%)
Instrs: 3011298 -> 3008585 (-0.09%); split: -0.17%, +0.08%
CodeSize: 15836804 -> 15841972 (+0.03%); split: -0.09%, +0.12%
VGPRs: 63580 -> 63604 (+0.04%)
SpillSGPRs: 966 -> 1148 (+18.84%); split: -0.83%, +19.67%
Latency: 36102291 -> 30186144 (-16.39%); split: -16.41%, +0.02%
InvThroughput: 9058100 -> 7011821 (-22.59%); split: -22.61%, +0.02%
VClause: 65369 -> 65364 (-0.01%); split: -0.03%, +0.02%
SClause: 100309 -> 100305 (-0.00%); split: -0.04%, +0.04%
Copies: 335658 -> 336472 (+0.24%); split: -0.70%, +0.94%
Branches: 110806 -> 108945 (-1.68%); split: -1.94%, +0.26%
PreSGPRs: 73476 -> 73934 (+0.62%); split: -0.25%, +0.87%
PreVGPRs: 58809 -> 58840 (+0.05%); split: -0.01%, +0.06%
Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/24940 >
2024-01-03 20:48:04 +00:00
Samuel Pitoiset
7ea48145a1
radv: stop clearing CMASK to 0xcc when FMASK is present on GFX9
...
This is incorrect because clearing CMASK to 0xCC should only happen
for fast clears with DCC and FMASK/CMASK.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26634 >
2024-01-03 13:58:06 +00:00
Samuel Pitoiset
1714f7aff4
radv: disable FMASK for MSAA images with layers on GFX9
...
It seems that FMASK can be interleaved and RADV doesn't support this
properly, so let's disable it.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26634 >
2024-01-03 13:58:06 +00:00
Samuel Pitoiset
10ce61e16c
radv: advertise VK_KHR_maintenance6
...
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26757 >
2024-01-03 11:24:27 +00:00
Samuel Pitoiset
db9816fd66
radv: add support for NULL index buffer
...
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26757 >
2024-01-03 11:24:27 +00:00
Samuel Pitoiset
744cb98bc6
radv: add support for version 2 of all descriptor binding commands
...
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/26757 >
2024-01-03 11:24:27 +00:00