These can never come from the API but there's a few cases where panvk
wants them.
Reviewed-by: Lorenzo Rossi <lorenzo.rossi@collabora.com>
Acked-by: Eric R. Smith <eric.smith@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/38681>
When masking out of bounds image loads, we previously returned a vector
of all zeros. However, for robustImageAccess2, depending on the format,
some components such as the alpha channel in an RGB format
should evaluate to 1.
This corrects the replacement value based on the format swizzle.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39430>
Code-motion should not move back upconversions without any other
instruction, that would only increase memory pressure without any
significant performance benefit (conversions are usually cheap).
This should also help lowering mediump varyings early by not reversing
their work.
Signed-off-by: Lorenzo Rossi <lorenzo.rossi@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Eric R. Smith <eric.smith@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40273>
On SM86+, we can use a 16-bit unsigned offset along side the register
for it.
This adds a new base indice that will be used for it, integration with
nir_opt_offsets and a lowering pass to get ride of the base on
unsupported generations.
Signed-off-by: Mary Guillemard <mary@mary.zone>
Reviewed-by: Mel Henning <mhenning@darkrefraction.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39716>
Adds a new intrinsic allowing to do raw write in the various ISBE spaces
where attributes are stored.
This also adapt isberd_nv to map to what we have since SM70+.
This will be used to support mesh shaders.
Signed-off-by: Mary Guillemard <mary@mary.zone>
Reviewed-by: Mel Henning <mhenning@darkrefraction.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39716>
It's the only driver that uses the pass so it may as well go there.
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Reviewed-by: Christoph Pillmayer <christoph.pillmayer@arm.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40307>
Previously, we assumed that the selector for bcsel could be whatever,
regardless of the bit sizes of the data and we'd just fix it in the
back-end. This works okay for scalars but falls over the moment we
vectorize because all our vector handling assumes bit sizes match.
Since matching bit sizes is what the hardware wants anyway, it's better
to do the right thing in NIR and hope copy-propagation can fold in
conversions if needed.
Unfortunately, copy prop isn't that smart yet so this does hurt a bit:
Instrs: 1193679 -> 1198086 (+0.37%); split: -0.06%, +0.43%
CodeSize: 11915136 -> 11950592 (+0.30%); split: -0.05%, +0.34%
Full: 160985 -> 160941 (-0.03%); split: -0.04%, +0.01%
Estimated normalized CVT cycles: 4456.938557000181 -> 4480.876069000186 (+0.54%); split: -0.13%, +0.67%
Estimated normalized SFU cycles: 6350.9375 -> 6392.21875 (+0.65%)
Estimated normalized Load/Store cycles: 205773.0 -> 205795.0 (+0.01%)
Maximum number of threads: 12864 -> 12863 (-0.01%)
Number of spill instructions: 22487 -> 22489 (+0.01%)
Number of fill instructions: 52179 -> 52219 (+0.08%)
Hurt shaders:
google-meet-clvk/BgBlur
google-meet-clvk/Relight
parallel-rdp/small_subgroup
parallel-rdp/small_uber_subgroup
The proper solution here is to teach copy-prop about this stuff so that
it can propagate swizzles into ALU ops when they're supported:
https://gitlab.freedesktop.org/panfrost/mesa/-/issues/265
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/14945
Cc: mesa-stable
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Reviewed-by: Christoph Pillmayer <christoph.pillmayer@arm.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40307>
Use is_scalar to know if we can do transpose loading.
Also enable vectorization if 2 intrinsics share the same source (it
means the only difference is the base).
Fixes: e14d6b535c ("brw/nir: add new intrinsics to load data from the indirect address")
Tested-by: Felix DeGrood <felix.j.degrood@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40308>
Useful for smaller/larger loads. Also there is no reason to be bitsize
specific here if we use an signed constant.
Foz-DB Navi48:
Totals from 8 (0.01% of 114655) affected shaders:
Instrs: 7629 -> 7612 (-0.22%)
CodeSize: 40772 -> 40692 (-0.20%)
Latency: 54880 -> 54944 (+0.12%)
InvThroughput: 8879 -> 8880 (+0.01%); split: -0.08%, +0.09%
VALU: 4029 -> 4027 (-0.05%); split: -0.15%, +0.10%
SALU: 1260 -> 1249 (-0.87%)
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40292>
This is new in SM75 (Turing). Let's use it because it allows us to get rid
of the if/else around bound checked global loads.
There are some changes in fossils, but it seems that's mostly due to CFG
optimizations doing things a bit differently?
Totals:
CodeSize: 9442152688 -> 9442133184 (-0.00%); split: -0.00%, +0.00%
Static cycle count: 6120910991 -> 6120907718 (-0.00%); split: -0.00%, +0.00%
Spills to reg: 184789 -> 184810 (+0.01%)
Fills from reg: 223831 -> 223860 (+0.01%); split: -0.00%, +0.01%
Totals from 334 (0.03% of 1163204) affected shaders:
CodeSize: 22020752 -> 22001248 (-0.09%); split: -0.10%, +0.01%
Static cycle count: 26582978 -> 26579705 (-0.01%); split: -0.01%, +0.00%
Spills to reg: 3110 -> 3131 (+0.68%)
Fills from reg: 3401 -> 3430 (+0.85%); split: -0.03%, +0.88%
Reviewed-by: Mary Guillemard <mary@mary.zone>
Reviewed-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Acked-by: Mel Henning <mhenning@darkrefraction.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40272>
It is legal to have a cluster size larger than the subgroup/ballot size,
but our lowering would blow up in this case due to the nir_ishl_imm
overflowing in the lowering. Fortunately, this is easy to handle.
Fixes sub_group_clustered_reduce_logical_and()
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40224>
Tessellation evaluation shaders have a single convergent URB handle
(for the common patch data) used by all lanes. Every other stage's
IO handles have separate handles in each lane.
Thanks to Alyssa Rosenzweig for catching this bug.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40280>
More fallout from f2a59fdea6.
is_not_zero now always returns whether the result is a floating point zero.
When combined with the fp denorm handling that will be added to
floating point range analysis, this is false for many sensible integer values.
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39987>
This is not NaN correct.
And also make the pattern 32bit only because the constant is hard coded
FLT_MAX.
Fixes: 780b5c1037 ("nir/algebraic: Simplify some Inf and NaN avoidance code")
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39987>
The way it is, this optimization is too aggressive and may generate
code way worse than the original. Remove it from here, so drivers
consuming the generated SPIR-V will be able to make their own
more-informed decisions later.
Let's follow the same strategy of nir_load_liblc.c and just set the
limit to 0.
For indirect copies in Anv (not merged yet), block compressed formats
require some expensive divisions, so I put them all inside 'if'
statements that should never run on normal formats. This optimization
made us always run all the divisions all the time, tanking the
performance of the shader on small copies.
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40020>
Use INT_MIN instead of INT_MAX for underflow.
Fixes: cc4b50b023 ("nir/opcodes: use u_overflow to fix incorrect checks")
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pelloux@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40252>
With most Vulkan engines doing multithreaded compiles, NIR_DEBUG=print has
been a frustrating racy mess. Take a lock when we're doing per-pass
printing, so that the output is coherent. This unfortunately
single-threads the compiler process itself in that case, but when you're
NIR_DEBUG=printing, that's probably not a big deal.
An assert is introduced to make sure that nobody nests NIR_PASS() in a way
that would break printing.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40126>
