Feral games aren't affected because they don't decompress DCC.
F1 2018 has one DCC decompression per frame, but I don't see
any performance improvements. This new predicate will be
probably more useful for DCC/MSAA.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
It's somehow similar to the FCE predicate.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Viewport/scissor don't need to be updated for array textures.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
DCC and FMASK also imply a fast-clear eliminate, so it should be
safe to reset the predicate unconditionally. We still only skip
FMASK or CMASK decompressions for now.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
this helps reduce the overall code changes when a bit_size parameter is
added to nir_load_system_value
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Now that image load/store intrinsics are variable-width, we need to set
num_components accordingly. In 15d39f474b, both glsl_to_nir and
spirv_to_nir were updated to properly set num_components but radv meta
was left behind.
Fixes: 15d39f474b "nir: Make image load/store intrinsics..."
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Tested-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
This fixes a regression with some Unity demos. Not sure
what the root cause of the problem is, especially because
the driver doesn't perform any fast color clears. So, it
shouldn't be needed to decompress DCC. RadeonSI says that
the decompression is relatively cheap if the surface has
been decompressed already.
One possible improvement is to two use predicates, one for
DCC and one for FCE that could be cleared when DCC, FMASK
or CMASK are performed by the driver. That might skip some
unnecessary decompression passes (not DCC though).
Fixes: ff7daadca1 ("radv: enable/disable predication for the DCC decompression pass")
CC: 18.2 <mesa-stable@lists.freedesktop.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107563
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
In environments where we cannot cache, e.g. Android (no homedir),
ChromeOS (readonly rootfs) or sandboxes (cannot open cache), the
startup cost of creating a device in radv is rather high, due
to compiling all possible built-in pipelines up front. This meant
depending on the CPU a 1-4 sec cost of creating a Device.
For CTS this cost is unacceptable, and likely for starting random
apps too.
So if there is no cache, with this patch radv will compile shaders
on demand. Once there is a cache from the first run, even if
incomplete, the driver knows that it can likely write the cache
and precompiles everything.
Note that I did not switch the buffer and itob/btoi compute pipelines
to on-demand, since you cannot really do anything in Vulkan without
them and there are only a few.
This reduces the CTS runtime for the no caches scenario on my
threadripper from 32 minutes to 8 minutes.
Reviewed-by: Dave Airlie <airlied@redhat.com>
Assignment and usage of this variable both happen inside an
if(rad_image_has_dcc()) {} blocks. It seems gcc plays it safe and
assumes that both function calls could have different return values.
But in this case we should be safe.
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Inherited commands buffers are not supported.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
By default, our internal rendering commands are discarded
only if the predicate is non-zero (ie. DRAW_VISIBLE). But
VK_EXT_conditional_rendering also allows to discard commands
when the predicate is zero, which means we have to use a
different flag.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
After performing a fast-clear eliminate, a FMASK decompress,
or a DCC decompress, we can reset the predicate to FALSE.
With that, the GPU should be able to skip unnecessary color
decompression passes.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Performing a DCC decompression pass is currently pretty rare,
but using predication allows the GPU to skip unnecessary passes.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Acked-by: Rob Clark <robdclark@gmail.com>
Acked-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Acked-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Acked-by: Rob Clark <robdclark@gmail.com>
Acked-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Acked-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
When decompressing DCC we don't enable it, so it's useless
to disable it. This reduces the number of prediction packets
sent to the GPU when performing color decompression passes.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Niuwenhuizen <bas@basnieuwenhuizen.nl>
They will help for DCC MSAA textures and if we support mipmaps
in the future.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Generated with
git grep -l nir_intrinsic_image | xargs \
sed -i 's/nir_intrinsic_image/nir_intrinsic_image_var/g'
and some manual fixing in nir_intrinsics.h
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
For fast clear eliminate and decompressions, we always use the most compressed
format.
For clears, the code already creates a renderpass on demand with the exact same
layout as specified.
Otherwise we start distinguishing between GENERAL and TRANSFER_DST_OPTIMAL.
Reviewed-by: Dave Airlie <airlied@redhat.com>
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
We do an in place copy where we read compressed and write decompressed.
By doing this in sizes that cover entire DCC blocks and waiting for all
reads in the block before starting to write we avoid corruption.
In the end we clear the DCC metadata to 0xffffffff.
Reviewed-by: Dave Airlie <airlied@redhat.com>
Tested-by: Dieter Nützel <Dieter@nuetzel-hh.de>
They are dummy objects but the spec requires layout to not be
NULL, this just makes sure we are creating valid pipeline layout
objects. This will allow us to remove some useless checks.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
And merge radv_meta_save_novertex() with
radv_meta_save_graphics_reset_vport_scissor_novertex().
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
This will allow us to save/restore the different states on-demand
based on the meta operation. For now, this saves/restores all
states. Compute will follow once the graphics part is done.
The main idea is to merge all save/restore helpers.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Unnecessary to double check that handles are not NULL.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
It can't change during the decompression pass.
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
We are really not going to use a winsys which does not need to store
the va, so might as well store it in a standard field.
Not sure this helps perf much though, as most of the cost is in the
cache miss accessing the bo anyway, which we stil need to do.
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
When using DCC some clear values don't require a cmask eliminate
step. This patch adds support for black and black with alpha 1,
there are other values, but I don't have access to a comprehensive list.
This works by setting the cmask eliminate predicate when doing the
fast clear, and later when doing the cmask elimination making sure
the draws are predicated.
This increases the fps on Sascha Willems deferred.
Tonga: 580fps->670fps on a Tonga PRO card.
Polaris 730->850fps
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This is to rework the surface code like radeonsi.
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
The vs vertex generate and fs noop shaders are used in a few places,
so refactor them out.
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Dave Airlie <airlied@redhat.com>
The only way we write CMASK/DCC compressed textures through shaders
is fast clears and CMASK/DCC inits, which have their own flushes.
Hence the CB cache is always up to date.
Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
I think we should only flush right before an action (draw/dispatch etc.),
as otherwise it is too easy to issue redundant flushes.
Signed-off-by: Bas Nieuwenhuizen <basni@google.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
This iterates the fast clear flush across the layers in the
specified range.
It also moves the compute resolve flush into the function
and builds the range in there.
This fixes:
dEQP-VK.geometry.layered.* regressions since fast clears.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
When restoring something from shader cache we won't have and don't
want to create a nir_shader this change detaches the two.
There are other advantages such as being able to reuse the
shader info populated by GLSL IR.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Before we can read the fmask using the compute shader, we need
to decompress the fmask in place.
This fixes a bunch of remaining failure and hopefully multisampling
in Talos.
This squashes all the radv development up until now into
one for merging.
History can be found:
https://github.com/airlied/mesa/tree/semi-interesting
This requires llvm 3.9 and is in no way considered
a conformant vulkan implementation. It can run a number
of vulkan applications, and supports all GPUs using
the amdgpu kernel driver.
Thanks to Intel for providing anv and spirv->nir,
and Emil Velikov for reviewing build integration.
Parts of this are:
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Acked-by: Edward O'Callaghan <funfunctor@folklore1984.net>
Authors: Bas Nieuwenhuizen and Dave Airlie
Signed-off-by: Dave Airlie <airlied@redhat.com>