D16 rounds towards zero for fp32 -> fp16, but for fixed point it rounds to
nearest even in fp16. MIMG without D16 also rounds to nearest even, but in fp32.
This means D16 and f2f16_rtz(tex@32) can produce different results.
Sadly this also means we can never use d16 if fp16 rounding isn't undefined.
Cc: mesa-stable
Acked-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28730>
Fixes fs-uint-to-float-of-extract-int8.shader_test and
fs-uint-to-float-of-extract-int16.shader_test added by piglit!883.
v2: Expand the comment explaining the potential problem. Suggested by
Caio.
Fixes: e6022281f2 ("intel/elk: Rename files to use elk prefix")
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27891>
Fixes fs-uint-to-float-of-extract-int8.shader_test and
fs-uint-to-float-of-extract-int16.shader_test added by piglit!883.
No shader-db or fossil-db changes on any Intel platform.
v2: Expand the comment explaining the potential problem. Suggested by
Caio.
Fixes: 29ce110be6 ("i965/fs: Remove extract virtual opcodes.")
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/27891>
When loading the base shader serialization there is a discrepancy
between the state parameters that may already have been optimized,
because after storing the serialization the shader went through
st_finalize_nir, and _mesa_optimize_state_parameters was run, so
that original state parameters may have been optimized and replaced
by new parameters.
After get_nir_shader is called, the original state parameters are
re-added - in addition to the optimized parameters. This lead to
an bug with the uniform offsets when lowering uniforms to UBOs.
Therefore, as a hotfix for drivers that don't support packed
uniforms, ignore the base serialization and use the
serialization obtained after st_finalize_nir was run. With that
the problem can be avoided.
Fixes: 5eb0136a3c
mesa/st: when creating draw shader variants,
use the base nir and skip driver opts
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/10881
v2: reorder conditional evaluation for better readability (zmike)
v3: revert c72bb8de7 ("r300: mark new fails") (Pavel Ondračka)
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28994>
Cross-device is a virtio-gpu feature that enables sharing host blob
dma-bufs with other virtio devices, like virtio-wl or virtio-video.
This feature is mainly used by ChromeOS and not required if there is
no dma-buf sharing. Venus has a hard requirement for the cross-device
feature.
Qemu doesn't support cross-device. Relax cross-device feature requirement
by making it optional, allowing Venus to work on Qemu.
Signed-off-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29040>
When a job is submitted to the flush_queue the resource dt_idx is reset,
and if a readback is requested then we have to make sure that the
corresponding kopper_preset has finished before we can acquire the image
for readback, so wait for the according fence in this case.
This fixes the validation error UNASSIGNED-Threading-MultipleThreads-Write
triggered by piglit "read-front" lavapipe.
Fixes: 8ade5588e3
zink: add kopper api
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28127>
This pass misses opportunities because foreach_list_typed_safe()
might point to disconnected cf_nodes after some optimization got
applied. No fossil-db changes.
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28150>
How this works? First we check which immediates are used as vectors,
i.e., have any reads that are using 2 or more channels. Such immdeiates
will be places in a free slots (but only the specific channels that are
used in the vector). This way we don't have to worry about swizzling
restrictions. The remaining scalar immediates will be checked for
duplicates and placed in free slots, including any empty slots in
previously places vector immediates (any swizzle is valid for scalars).
RV410:
total instructions in shared programs: 98883 -> 98905 (0.02%)
instructions in affected programs: 15414 -> 15436 (0.14%)
helped: 100
HURT: 102
total presub in shared programs: 2235 -> 2235 (0.00%)
presub in affected programs: 608 -> 608 (0.00%)
helped: 51
HURT: 72
total omod in shared programs: 419 -> 418 (-0.24%)
omod in affected programs: 15 -> 14 (-6.67%)
helped: 3
HURT: 3
total temps in shared programs: 15698 -> 15692 (-0.04%)
temps in affected programs: 952 -> 946 (-0.63%)
helped: 46
HURT: 37
total consts in shared programs: 84458 -> 83856 (-0.71%)
consts in affected programs: 14648 -> 14046 (-4.11%)
helped: 499
HURT: 0
total cycles in shared programs: 156476 -> 156493 (0.01%)
cycles in affected programs: 22532 -> 22549 (0.08%)
helped: 100
HURT: 102
LOST: shaders/ck2/157.shader_test FS
GAINED: shaders/ck2/160.shader_test FS
GAINED: shaders/tesseract/395.shader_test FS
RV530:
total instructions in shared programs: 119543 -> 119612 (0.06%)
instructions in affected programs: 27435 -> 27504 (0.25%)
helped: 118
HURT: 183
total presub in shared programs: 7257 -> 7111 (-2.01%)
presub in affected programs: 1856 -> 1710 (-7.87%)
helped: 121
HURT: 48
total omod in shared programs: 426 -> 427 (0.23%)
omod in affected programs: 5 -> 6 (20.00%)
helped: 1
HURT: 2
total temps in shared programs: 16784 -> 16779 (-0.03%)
temps in affected programs: 392 -> 387 (-1.28%)
helped: 29
HURT: 17
total consts in shared programs: 93198 -> 92667 (-0.57%)
consts in affected programs: 14577 -> 14046 (-3.64%)
helped: 451
HURT: 0
total cycles in shared programs: 186649 -> 186590 (-0.03%)
cycles in affected programs: 26306 -> 26247 (-0.22%)
helped: 125
HURT: 111
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Reviewed-by: Filip Gawin <filip.gawin@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28630>
We are not doing this on R5xx unless we have more than 200 constants,
because emitting constants one by one will add extra overhead at emit
time which we want to avoid if possible.
RV410:
total instructions in shared programs: 98778 -> 98703 (-0.08%)
instructions in affected programs: 7106 -> 7031 (-1.06%)
helped: 80
HURT: 25
total presub in shared programs: 2266 -> 2227 (-1.72%)
presub in affected programs: 134 -> 95 (-29.10%)
helped: 22
HURT: 10
total temps in shared programs: 15662 -> 15660 (-0.01%)
temps in affected programs: 330 -> 328 (-0.61%)
helped: 16
HURT: 13
total consts in shared programs: 85632 -> 84400 (-1.44%)
consts in affected programs: 6646 -> 5414 (-18.54%)
helped: 617
HURT: 0
total cycles in shared programs: 156305 -> 156234 (-0.05%)
cycles in affected programs: 14167 -> 14096 (-0.50%)
helped: 79
HURT: 28
LOST: shaders/ck2/160.shader_test FS
GAINED: shaders/ck2/157.shader_test FS
GAINED: shaders/tropics/249.shader_test FS
GAINED: shaders/tropics/252.shader_test FS
RV530:
total consts in shared programs: 93209 -> 93198 (-0.01%)
consts in affected programs: 72 -> 61 (-15.28%)
helped: 6
HURT: 0
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.comm>
Reviewed-by: Filip Gawin <filip.gawin@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28630>
Instead of just moving around constants as full vec4, we will now have
the flexibility to shuffle scalars around. However, this commit just
prepares the infrestructure and converts to it, while the constant
elimination logiic reamins the same, i.e., we only remove constant if it
is fully unused and there is no constant compaction whatsoever.
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Reviewed-by: Filip Gawin <filip.gawin@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28630>
We need to put it before source conflict resolve because we will be
shuffling immediates around, so we can introduce new conflicts (albeit
in general there should be less conflicts instead).
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Reviewed-by: Filip Gawin <filip.gawin@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28630>
To have more flexibility in case there are some empty slots (e.g., if
the specific slot was converted to inline constant or constant swizzle).
Signed-off-by: Pavel Ondračka <pavel.ondracka@gmail.com>
Reviewed-by: Filip Gawin <filip.gawin@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28630>
SPI_SHADER_COL_FORMAT/CB_SHADER_MASK are used slightly differently
for PS epilogs, shader objects and monolithic graphics pipelines.
This introduces a new state that will allow us to emit these two
registers in only place. The main motivation is for depth-only RB+
support and for tracking context registers in the cmdbuf.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28976>
This is always the non-compacted format because it's compacted right
before it's emitted. This looks much cleaner to me.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28976>
For drivers using the new state tracking, __bitset_test_range can be
surprisingly hot because we have a lot of dirty bits and __bitset_test_range has
to handle lots of special cases. __bitset_is_empty does not have to worry about
those special cases so can be much faster.
Signed-off-by: Alyssa Rosenzweig <alyssa@rosenzweig.io>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Christian Gmeiner <cgmeiner@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29008>
The relevant sections here are F.7.3.2.1 and F.7.3.2.2.1. The code was
incorrectly assuming sub_layer_ordering_info_present_flag is always 1.
Reviewed-by: Hyunjun Ko <zzoon@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29001>
Per section 7.7.3, the structure includes additional optional layer-specific
information, which is padded if left unset, based on the value of
max_sub_layers_minus1. The vulkan input structs have no way to specify this
per-layer information, so we just need the padding.
Reviewed-by: Hyunjun Ko <zzoon@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/29001>
This enables EXT_depth_range_unrestricted from VOLTA_A
Test of dEQP-VK.*depth_range_unrestricted* on TU104 shows:
Test run totals:
Passed: 14212/14212 (100.0%)
Failed: 0/14212 (0.0%)
Not supported: 0/14212 (0.0%)
Warnings: 0/14212 (0.0%)
Waived: 0/14212 (0.0%)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28958>
Now that we have flows, custom tracks, and timestamps, we can have a track
for wayland buffer presentation times, tagged with appropriate flow ids
so we can follow when a buffer was acquired through to the time it was
displayed.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28634>
This can be useful if we know when an event happened, but our code isn't
running at that time (such as reporting when an image was presented in
the wayland wsi).
We can't really mix these with events that we log at the current time,
because there could be overlap, so also add a function for creating
custom tracks.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28634>
When using presentation feedback, we know when an image is presented. Use
this and the time we submit the image to calculate the delay in ms
between submission and display.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28634>
When we use waitforpresent we use presentation feedback. We can plumb
the flow ids into this to have slightly more expressive flows.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28634>
Perfetto can assign flow ids to events, which can be used to connect
related events in tracks when they share the same id.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28634>
This is just to get event tracing in perfetto, as the wrapper calls
MESA_TRACE_FUNC().
It can be useful to see how long and when we stall in wayland dispatch.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28634>
Add a thin wrapper around the wayland dispatch code for no reason other
than to add MESA_TRACE_FUNC so we can see where wayland dispatch delays
are.
Move this to loader so we can use it in the wayland egl code later.
Signed-off-by: Derek Foreman <derek.foreman@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/28634>