This is now basically the same as the original VALUMaskWriteHazard, except
it now considers both VALU and SALU writes.
Now that it's a part of VALUMaskWriteHazard, differences from the original
VALU lanemask workaround are:
- it includes SALU reads after the write
- it includes VALU writes and SALU/VALU reads after the write which are
not lanemasks
- it combines s_waitcnt_depctr instructions when it's a read after both a
SALU write and a VALU write
- non-exec VALU SGPR reads reset the SGPRs read by VALU as a lanemask
- exec SGPRs are ignored
resolve_all_gfx11() is also finished.
fossil-db (navi31):
Totals from 21538 (27.13% of 79377) affected shaders:
Instrs: 27628855 -> 27552972 (-0.27%); split: -0.30%, +0.03%
CodeSize: 145968448 -> 145667616 (-0.21%); split: -0.23%, +0.02%
Latency: 209537805 -> 209509519 (-0.01%); split: -0.02%, +0.00%
InvThroughput: 36304270 -> 36301624 (-0.01%); split: -0.01%, +0.00%
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/12623
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/11480
Backport-to: 25.0
Backport-to: 25.1
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34529>
In a previous iteration of the spilling code, we added an extra check to
only spill across edges if the value being spilled is in the W set.
This was due to a misunderstanding of the modeling of S and W in Braun
and Hack. In the current implementation, we maintain the invariant that
every live value is in at least one of S or W so we don't need that
check but it was left in by mistake.
One exception to this rule was added when we special-cased constant
values. Now the invariant is that every live value is in S, in W, or is
a constant. When we made this change, the check we accidentally left in
bit us because now if a value is constant but not in W, it wasn't
getting spilled across the edge. This can result in a value getting
filled later which was never spilled, leading to undefined values.
Fixes: 7b82e26e3c ("nak: Don't spill/fill const values")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/12993
Co-authored-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34563>
I accidentally disabled compression on CPS surfaces marked as storage or
color attachment for all platforms, when this should only be limited to
Xe.
Fixes: 80f9b6 ('anv: CPB surfaces that are used as color attachments or for stores cannot be compressed')
Signed-off-by: Rohan Garg <rohan.garg@intel.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34297>
fossil-db (gfx1201):
Totals from 10576 (13.32% of 79377) affected shaders:
(no stats changed)
Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34246>
To simplify things, state that all operations as implemented expect
their input tensors to be in the channel order that the hardware
supports, and that the output tensors will be in that same order as
well.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34498>
Show how to add more Android CTS tests cases, using --include-filters
Only CtsNativeHardwareTestCases and CtsSkQPTestCases are actually
enabled for now, because the android-angle-lavapipe-cts job is part of
the pre-merge pipeline and these modules would not be too expensive to
run.
The container size increases by about 60Mb and the test time from 4m to
7 min on a local system, so it's an acceptable compromise to show how
multiple modules can be tested.
A similar mechanism will be used to add CtsDeqpTestCases tests in the
future, probably in nightly or weekly jobs, because that would require
more space in the containers and a lot more time to run the tests.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34479>
This might increase the scratch BO sizes but it's supposed to be
faster because scratch waves would be distributed among memory channels.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34549>
The implementation of demote in IMG blobs seems to be a piece of hack,
and the current use of it in Zink leads to assertion failure with
information "Uniflex does not support demote and terminate in the same
shader".
Disable usage of demote for IMG blobs.
Signed-off-by: Icenowy Zheng <uwu@icenowy.me>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/33051>
Structured tagging (cf. mesa/mesa!33421) captures a checksum of the
thing we think we're building, and verifies this through the chain.
When we run container builds, we check that the tag we've captured in
the CI variables matches the calculated checksum, to make sure the
declared tags are consistent and we always have traceability.
When we run tests, we check the tags again between what was declared in
the CI variables and what we're actually running from the test
container. This makes sure that we're always testing what we think we're
testing.
As a side advantage, the rule inheritance we need to make this work
means that we can start doing more optional downloads via overlays,
instead of pulling a whole container full of stuff we might not ever
use.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34539>
This caused a -5% performance regression in Control because using
compute always eats resources.
This new approach introduces a flag called RADV_COPY_FLAGS_DEVICE_LOCAL
which can be used to indicate if the underlying memory is device local.
This should also help for future work.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/12639
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34556>
This commit implements the following requirement:
"Keep any UMD-recycling of compression-enabled/disabled
memory separate."
As additional info there are 2 related wa's for the issue:
Wa_14018443005
Wa_18038669374
Cc: mesa-stable
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34499>
Dequantize the weights, calculate new quantization values based on the
minimum and maximum values and requantize again.
This causes a loss of accuracy when compared to proper per-axis
quantization as we don't have the original data from training any more.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34497>
A cooperative matrix conversion operation was represented in NIR by the
cmat_unary_op intrinsic with an nir_alu_op as extra parameter,
that was already lowered to a specific conversion operation
based on the matrix types.
Instead of that, add a new intrinsic `cmat_convert` that is specific
for that conversion. In addition to the src/dst matrix descriptions
already available, also include the signedness information in the
intrinsic (reuse nir_cmat_signed for that). This is needed because
different Convert operations define different interpretations for
integers, regardless their original type.
In this patch, both radv and intel were changed to use the same logic
that was previously used to pick the lowered ALU op.
This change will help represent cmat conversions involving BFloat16,
because it avoids having to create new NIR ALU ops for all the
combinations involving BFloat16.
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34511>
The new 6.14 kernel in gfx-ci linux is now stable on Jasper Lake, so drop
the kernel override from the anv-jsl-vk job.
Signed-off-by: Valentine Burley <valentine.burley@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34553>
These values are shared with xcb/dri2.h, and can't be changed
without breaking the legacy dri2 compatibility. This change
reverses partially the update done by 3b603d1646.
For instance this issue is triggered on dri2 i915 with
"piglit/bin/glx-copy-sub-buffer -auto" or
"piglit/bin/hiz-depth-read-window-stencil0 -auto".
Fixes: 3b603d1646 ("mesa_interface: remove unused stuff")
Signed-off-by: Patrick Lerda <patrick9876@free.fr>
Acked-by: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/34561>
