Commit graph

1979 commits

Author SHA1 Message Date
Samuel Pitoiset
2ebfa64be7 radv: add radv_disable_hiz_his_gfx12 and enable for Mafia Definitive Edition
Some checks are pending
macOS-CI / macOS-CI (dri) (push) Waiting to run
macOS-CI / macOS-CI (xlib) (push) Waiting to run
This is a workaround for random GPU hangs with HiZ/HiS on GFX12
because the correct fix is complex and it will take time to be
implemented properly.

Mafia Definitive Edition is the first known game affected by this.

Cc: mesa-stable
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/13222
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35182>
2025-05-28 07:20:26 +00:00
Samuel Pitoiset
2812efd7ad radv: declare and emit NGG_QUERY_BUF_VA on GFX11-GFX11.5
This user SGPR is used to pass the query buffer VA for emulated queries.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/35017>
2025-05-21 08:46:12 +02:00
Samuel Pitoiset
4d1fcd75f9 radv: fix non-indexed draws with primitive restart enable
Some checks are pending
macOS-CI / macOS-CI (dri) (push) Waiting to run
macOS-CI / macOS-CI (xlib) (push) Waiting to run
On GFX11+, DISABLE_FOR_AUTO_INDEX=1 automatically disables primitive
restart enable for non-indexed draws.

On GFX10-GFX10.3 the hw considers primitive restart enable for
non-indexed draws and the driver must disable it explicitly.

GFX9 and older gens aren't affected but applying the change for them
simplifies the implementation.

To fix that, move emitting primitive restart enable at draw time
because it needs to know if the draw is indexed or not.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/13037
Cc: mesa-stable
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34996>
2025-05-20 13:57:35 +00:00
Hans-Kristian Arntzen
e674823d55 radv: Consider that DGC might need shader reads of predicated data.
Some checks are pending
macOS-CI / macOS-CI (dri) (push) Waiting to run
macOS-CI / macOS-CI (xlib) (push) Waiting to run
Similar to indirect draw barrier, need similar fixups for conditional
rendering access.

Signed-off-by: Hans-Kristian Arntzen <post@arntzen-software.no>
Cc: mesa-stable
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34956>
2025-05-15 06:14:46 +00:00
Samuel Pitoiset
3ca2f71f3d radv: fix conditional rendering with DGC and non native 32-bit predicate
When the hardware doesn't natively support 32-bit predication, the
driver has a fallback which allocates a 64-bit predicate to the upload
BO in order to copy the original value.

But when conditional rendering is enabled in the stateCommandBuffer
which is used by preprocess() and the execute() is recorded also in the
stateCommandBuffer. If the preprocess() is recorded in a different
cmdbuf which is submitted before the cmdbuf that contains execute(),
the fallback (ie. alloc + COPY_DATA) will be performed after. This would
cause the predicate value to be always 0.

To fix that, keep track of the user predication VA which is the only
VA that needs to be used by DGC because it reads 32-bit from the shader.

This fixes a very weird corner case with vkd3d-proton.

Cc: mesa-stable
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/13143
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34953>
2025-05-15 05:51:04 +00:00
Samuel Pitoiset
9a07ccbc89 radv: fix emitting dynamic viewports/scissors when the count is static
In a scenario where the viewports/scissors are a dynamic state but the
count is static (ie. updated when a graphics pipeline is bound), the
driver wasn't considering that and it was re-emitting the previous
number of viewports/scissors.

This fixes rendering issue with Blender.

Cc: mesa-stable
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/13127
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34921>
2025-05-13 16:08:14 +00:00
Samuel Pitoiset
eaf646d020 radv: implement VK_EXT_zero_initialize_device_memory
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34896>
2025-05-12 06:53:55 +00:00
Samuel Pitoiset
4b76d04f7f radv: ignore conditional rendering with vkCmdTraceRays*
CmdTraceRays is neither a dispatch or a draw command which means it
shouldn't be affected by conditional rendering.

Fixes recent VKCTS coverage.

Cc: mesa-stable
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34868>
2025-05-08 19:08:10 +00:00
Ricardo Garcia
bc44d029df radv: Ignore image barrier queue families if equal
Some checks are pending
macOS-CI / macOS-CI (dri) (push) Waiting to run
macOS-CI / macOS-CI (xlib) (push) Waiting to run
The src and dst queue family indices in an image memory barrier may
contain arbitrary values that can be ignored unless both are different.
This fixes a crash in upcoming CTS tests.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34691>
2025-04-29 08:15:28 +00:00
Samuel Pitoiset
1fccc09abe radv: fix re-emitting VRS state when rendering begins
This state also depends on whether a VRS attachment is used.

Cc: mesa-stable
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/11693
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34735>
2025-04-29 07:00:09 +00:00
Timur Kristóf
3ad385b9cc radv: Clear dirty flag for clip rects state after emitting it.
Some checks are pending
macOS-CI / macOS-CI (dri) (push) Waiting to run
macOS-CI / macOS-CI (xlib) (push) Waiting to run
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Tested-by: Marcus Seyfarth <m.seyfarth@gmail.com>
Fixes: 0ba3a8b3cc
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34686>
2025-04-24 15:13:44 +00:00
Timur Kristóf
3a05477ac6 radv: Clear dirty flag for MSAA state after emitting it.
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Tested-by: Marcus Seyfarth <m.seyfarth@gmail.com>
Fixes: 08918f0880
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/13022
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34686>
2025-04-24 15:13:44 +00:00
Samuel Pitoiset
78c2feed00 radv/meta: rename more buffer->memory for fill/copy/update operations
Recently, I renamed most of the helpers for future work but I forgot
few things like meta keys, etc.

This is for consistency.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34558>
2025-04-18 17:21:24 +02:00
Samuel Pitoiset
209a0ede98 radv: add a function to emit meshlet registers on GFX11+
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34518>
2025-04-17 12:49:47 +00:00
Samuel Pitoiset
836757bec3 radv: tidy up radv_emit_ps_epilog_state()
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34518>
2025-04-17 12:49:47 +00:00
Samuel Pitoiset
dca35b7226 radv: tidy up radv_emit_geometry_shader()
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34518>
2025-04-17 12:49:47 +00:00
Samuel Pitoiset
d999afeb7a radv: tidy up radv_emit_vertex_shader()
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34518>
2025-04-17 12:49:47 +00:00
Samuel Pitoiset
85fdf69027 radv: simplify combining TES/VS+GS config registers
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34518>
2025-04-17 12:49:47 +00:00
Samuel Pitoiset
0dd9833348 radv: remove redundant assertion when emitting PS epilog state
It's already checked by radv_emit_32bit_pointer().

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34518>
2025-04-17 12:49:47 +00:00
Samuel Pitoiset
a230d2daa3 radv: use radeon_set_sh_reg() for only 1 DWORD
It's just shorter to write.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34518>
2025-04-17 12:49:47 +00:00
Samuel Pitoiset
d94f8b4460 ac/gpu_info,radv: add scratch_wavesize_granularity info
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34549>
2025-04-17 10:35:40 +00:00
Samuel Pitoiset
8ba94d8263 radv: add radv_fill_image() helper
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34556>
2025-04-17 08:59:58 +00:00
Samuel Pitoiset
75be860eec radv: use paired context regs when optimal on GFX12
Some checks are pending
macOS-CI / macOS-CI (dri) (push) Waiting to run
macOS-CI / macOS-CI (xlib) (push) Waiting to run
CP is very slow on GFX12 and parsing the packet header is the main
bottleneck. Using paired context regs reduce the number of packet
headers and it should be more optimal.

It doesn't seem worth when only one context reg is emitted (one packet
header and same number of DWORDS) or when consecutive context regs are
emitted (would increase the number of DWORDS).

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34421>
2025-04-14 06:18:13 +00:00
Timur Kristóf
412af41258 radv: Add radv_foreach_stage to ForEachMacros again.
This was lost when .clang-format was removed
from the amd folder.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33880>
2025-04-11 18:01:47 +00:00
Samuel Pitoiset
2f00daf67a radv: tidy up radv_emit_hw_ngg()
Some checks are pending
macOS-CI / macOS-CI (dri) (push) Waiting to run
macOS-CI / macOS-CI (xlib) (push) Waiting to run
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34420>
2025-04-10 06:56:25 +00:00
Samuel Pitoiset
1290b38f57 radv: tidy up radv_emit_raster_state()
Better isolation between configuration and emission.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34420>
2025-04-10 06:56:25 +00:00
Samuel Pitoiset
4b2d119d90 radv: reduce the number of emitted DWORDS for MSAA 8x user sample locs
From 24 DWORDS to 16 DWORDS.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34420>
2025-04-10 06:56:25 +00:00
Samuel Pitoiset
c1ebf82700 radv: track redundant DB_RENDER_OVERRRIDE register writes on GFX12
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34420>
2025-04-10 06:56:25 +00:00
Samuel Pitoiset
7f5727b313 radv: use consecutive registers for PA_SC_WINDOW_SCISSOR_{TL,BR}
For less DWORDS.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34420>
2025-04-10 06:56:25 +00:00
Samuel Pitoiset
32ea7df586 radv: move emitting more fb registers when rendering begins
No need to delay the emission of these registers.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34420>
2025-04-10 06:56:25 +00:00
Samuel Pitoiset
0ba3a8b3cc radv: add clip rects state bit for emitting discard rectangles
Some checks are pending
macOS-CI / macOS-CI (dri) (push) Waiting to run
macOS-CI / macOS-CI (xlib) (push) Waiting to run
Better match the hw naming.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34361>
2025-04-08 08:42:17 +00:00
Samuel Pitoiset
08918f0880 radv: regroup emitting all MSAA states in one function
All register writes are optimized out. Also this will allow to use
paired context register writes on GFX12.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34361>
2025-04-08 08:42:17 +00:00
Samuel Pitoiset
e8d787e1ef radv: track more MSAA related register writes
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34361>
2025-04-08 08:42:17 +00:00
Samuel Pitoiset
a327bc677a radv: configure COVERAGE_TO_SHADER_SELECT only if conservative rast is enabled
When conservative rasterization isn't enabled, FullyCoveredEXT is
expected to return 0.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34361>
2025-04-08 08:42:17 +00:00
Samuel Pitoiset
6e9782b39c radv: emit conservative raster mode as part of the MSAA state
From the hw perspective, it's more like a MSAA state.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34361>
2025-04-08 08:42:17 +00:00
Samuel Pitoiset
ed744b5c68 radv: move emitting raster and depth/stencil state slightly earlier
To avoid a redundant chekc if no dynamic states are dirtied.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34361>
2025-04-08 08:42:17 +00:00
Samuel Pitoiset
6388db03c8 radv: add a workaround for buggy HiZ/HiS on GFX12
HiZ/HiS is buggy and can cause random GPU hangs when stencil is enabled.
There are basically two alternatives but RADV follows RadeonSI and emit
a dummy RELEASE_MEM packet after every draw which should workaround the
issue and maintain performance.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/12944
Backport-to: 25.0
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34381>
2025-04-08 08:09:13 +02:00
Samuel Pitoiset
11b6d2ba60 radv: determine if HiZ/HiS is enabled earlier on GFX12
To lower CPU overhead of the hardware workaround.

Backport-to: 25.0
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34381>
2025-04-08 08:03:11 +02:00
Samuel Pitoiset
e3c3fa8b9a radv: add a fuction to emit the VRS surface on GFX11
Some checks are pending
macOS-CI / macOS-CI (dri) (push) Waiting to run
macOS-CI / macOS-CI (xlib) (push) Waiting to run
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34357>
2025-04-07 15:09:18 +00:00
Samuel Pitoiset
00354f0a74 radv: do not emit the VRS surface VA when it's not enabled
This shouldn't be necessary because VRS_SURFACE_ENABLE is the toggle
bit.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34357>
2025-04-07 15:09:18 +00:00
Samuel Pitoiset
9b11caee40 radv: stop emitting CB_FDCC_CONTROL to zero on GFX11-GFX11.5
It's already emitted by the CLEAR_STATE packet on these GPUs.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34357>
2025-04-07 15:09:18 +00:00
Samuel Pitoiset
17e5fd856f radv: split null framebuffer state emission for GFX12
For consistency with color/ds states emission.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34357>
2025-04-07 15:09:18 +00:00
Samuel Pitoiset
c608a601bf radv: split framebuffer depth/stencil state emission for GFX12
It's also really simpler on GFX12.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34357>
2025-04-07 15:09:18 +00:00
Samuel Pitoiset
d5eb1f6833 radv: split framebuffer color state emission for GFX12
It's really simpler on GFX12.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34357>
2025-04-07 15:09:18 +00:00
Samuel Pitoiset
9dec80d8cf radv: tidy up radv_emit_db_shader_control()
To separate packet construction and emission.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34357>
2025-04-07 15:09:18 +00:00
Samuel Pitoiset
4bc6335e3c radv: restore DB_DFSM_CONTROL properly when POPS isn't used
It looks safer to restore this register to its initial value when POPS
isn't used. Only VEGA10 and RAVEN are concerned.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34357>
2025-04-07 15:09:18 +00:00
Samuel Pitoiset
cb390ca234 radv: tidy up radv_emit_raster_state()
To separate packet construction and emission.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34357>
2025-04-07 15:09:18 +00:00
Samuel Pitoiset
62e4b1130a radv: remove useless FDCC_ENABLE bitfield clear on GFX12
This bit doesn't exist.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34357>
2025-04-07 15:09:18 +00:00
Samuel Pitoiset
64e6e043b3 Revert "radv: program SAMPLE_MASK_TRACKER_WATERMARK optimally for GFX11 APUs"
This reverts commit 96e9c3fe77.

This actually causes random GPU hangs like on Phoenix.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/12461
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/12426
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/12692
Tested-by: Tobias Jakobi <tjakobi@math.uni-bielefeld.de>
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34306>
2025-04-02 07:10:40 +00:00
Autumn Ashton
ae6d24c4ef radv: Expose VK_SAMPLE_COUNT_1_BIT for sample position on GFX10+
This works on GFX10+.

Signed-off-by: Autumn Ashton <misyl@froggi.es>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28237>
2025-04-01 21:15:34 +01:00