This makes UBO loads in the variable pointers or bindless case work just
like SSBO loads in the sense that they use A64 messages and 64-bit
global addresses. The primary difference is that we have an
optimization in anv_nir_lower_ubo_loads which uses a (possibly
predicated) block load message when the offset is constant so we get
roughly the same performance as we would from plumbing load_ubo all the
way to the back-end.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8635>
This also means that some of the newly added helpers need to grow a bit
to support VK_DESCRIPTOR_TYPE_INLINE_UNIFORM_DATA_EXT.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8635>
This function has exactly two call sites. The first is where we had
these calculations before. The second only cares about the size of the
SSBO so all the extra code we emit will be dead. However, NIR should
easily clean that up and this lets us consolidate things a bit better.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8635>
Instead of packing the descriptor offset into the packed portion, use
that unused channel we have lying around. This potentially allows for
larger descriptor sets. We also re-arrange the components a bit to make
it more like the 64bit_bounded_global memory address format.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8635>
This has the advantage of giving us cheaper address calculations because
we can calculate in 32 bits first and then do a single 64x32 add. It
also lets us delete a bunch of code for dealing with descriptor
dereferences (vulkan_resource_reindex, and friends) because our bindless
SSBO pointers are now vec4s regardless of whether or not we're doing
bounds checking. This also unifies UBOs and SSBOs. The one down-side
is that, in certain variable pointers cases, it may end up burning more
memory and/or increasing register pressure. This seems like a worth-
while trade-off.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8635>
Instead of doing the array check at the load_vulkan_resource_index
intrinsic, stuff it in the vec2 and handle it at load_vulkan_descriptor
time. This allows the bounds check to take any re-index intrinsics into
account. This only affects variablePointers + SSBOs + Gen7.
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8635>
We're about to enable bindless UBOs via A64 memory access like we do for
SSBOs. In order to prevent 100% of UBOs from hitting that path, we
enable them in the early lowering. This way we'll still get binding
table-based UBO access for any non-bindless ones. In particular, we
need this for UBO pushing to work.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8635>
Rewrite them all to work on an index/offset vec2 instead of some only
returning the index. This means SSBO size handling is a tiny bit more
complicated but it will also mean we can use them for descriptor buffers
properly.
This also fixes a bug where we weren't bounds-checking re-index
intrinsics because we applied the bounds check at the tail of the
recursion and not at the beginning.
Fixes: 3cf78ec2bd "anv: Lower some SSBO operations in..."
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8635>
This makes things a bit more generic for use in the next commit.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8635>
NIR can do a bit better job optimizing this version.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8635>
This replaces the new_src parameter of nir_ssa_def_rewrite_uses_after()
with an SSA def, and rewrites all the users as needed.
Acked-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9383>
This commit replaces the new_src parameter of nir_ssa_def_rewrite_uses()
with an SSA def, removes nir_ssa_def_rewrite_uses_ssa(), and rewrites
all the users as needed.
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Acked-by: Alyssa Rosenzweig <alyssa@collabora.com>
Acked-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9383>
NIR derefs currently have exactly one variable mode. This is about to
change so we can handle OpenCL generic pointers. In order to transition
safely, we need to audit every deref->mode check. This commit adds a
set of helpers that provide more nuanced mode checks and converts most
of NIR to use them.
For simple cases, we add nir_deref_mode_is and nir_deref_mode_is_one_of
helpers. These can be used in passes which don't have to bother with
generic pointers and just want to know what mode a thing is. If the
pass ever encounters generic pointers in a way that this check would be
unsafe, it will assert-fail to alert developers that they need to think
harder about things and fix the pass.
For more complex passes which require a more nuanced understanding of
modes, we add nir_deref_mode_may_be and nir_deref_mode_must_be helpers
which accurately describe the compiler's best knowledge about the given
deref. Unfortunately, we may not be able to exactly identify the mode
in a generic pointers scenario so we have to be very careful when we use
these. Conversion of these passes is left to later commits.
For the case of mass lowering of a particular mode (nir_lower_explicit_io
is one good example), we add nir_deref_mode_is_in_set. This is also
pretty assert-happy like nir_deref_mode_is but is for a set containment
comparison on deref modes where you expect the deref to either be all-in
or all-out.
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6332>
This makes it explicit that this intrinsic is only for SSBOs. For the
v3dv driver, we'll be adding a get_ubo_size intrinsic and we want to be
able to distinguish between the two.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6812>
For UBO accesses to be the same performance as classic GL default uniform
block uniforms, we need to be able to push them through the same path. On
freedreno, we haven't been uploading UBOs as push constants when they're
used for indirect array access, because we don't know what range of the
UBO is needed for an access.
I believe we won't be able to calculate the range in general in spirv
given casts that can happen, so we define a [0, ~0] range to be "We don't
know anything". We use that at the moment for all UBO loads except for
nir_lower_uniforms_to_ubo, where we now avoid losing the range information
that default uniform block loads come with.
In a departure from other NIR intrinsics with a "base", I didn't make the
base an be something you have to add to the src[1] offset. This keeps us
from needing to modify all drivers (particularly since the base+offset
thing can mean needing to do addition in the backend), makes backend
tracking of ranges easy, and makes the range calculations in
load_store_vectorizer reasonable. However, this could definitely cause
some confusion for people used to the normal NIR base.
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6359>
This doesn't really do anything for us today. One day, I suppose we
could use it to do something with wide loads with non-uniform offsets.
The big reason to do this is to get better testing to make sure that NIR
doesn't blow up on the deref paths.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6472>
When we have softpin, we know the address of the shader constant data at
shader upload time because it's sitting at the end of the shader. This
commit changes ANV to use patch constants to embed the address in the
shader patch the right address in at upload time. This allows us to
avoid having to set up a UBO binding on-the-fly for shader constants.
This commit uses an A64 message but it's quite possible that we could
also use an A32 message and make the dataport do the 64-bit add for us.
However, load_global is what we have right now so it was easier to just
use that.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6244>
This one's a bit more complex because it filters off only those
variables with mode == nir_var_uniform. As such, it's not exactly a
drop-in replacement for nir_foreach_variable(var, &nir->uniforms).
Reviewed-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5966>
Was hoping to find potential issues but nothing. Still probably a good
idea.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Instead of blindly dirtying descriptors and push constants the moment we
see a pipeline change, check to see if it actually changes the bind
layout or push constant layout. This doubles the runtime performance of
one CPU-limited example running with the Dawn WebGPU implementation when
running on my laptop.
NOTE: This effectively reverts beca63c6c0. While it was a nice
optimization, it was based on prog_data and we can't do that anymore
once we start allowing the same binding table to be used with multiple
different pipelines.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
This substantially reworks both the state setup side of push constant
handling and the pipeline compile side. The fundamental change here is
that we're no longer respecting the prog_data::param array and instead
are just instructing the back-end compiler to leave the array alone.
This makes the state setup side substantially simpler because we can now
just memcpy the whole block of push constants and don't have to
upload one DWORD at a time.
This also means that we can compute the full push constant layout
up-front and just trust the back-end compiler to not mess with it.
Maybe one day we'll decide that the back-end compiler can do useful
things there again but for now, this is functionally no different from
what we had before this commit and makes the NIR handling cleaner.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
As of 2d78e55a8c, nir_intrinsic_load_constant with a constant offset
is constant-folded so we should never end up with any that trigger
brw_nir_analyze_ubo_ranges.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
This lets us stop tracking the pipeline layout. It also means less
indirection on a very hot path. As an extra bonus, we can make some of
our data structures smaller. No measurable CPU overhead improvement.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
This better matches all the other atomic intrinsics such as those for
SSBOs and shared variables where the sign is part of the intrinsic
opcode. Both generators (GLSL and SPIR-V) know the sign from the type
of the image variable or handle. In SPIR-V, signed min/max are separate
opcodes from unsigned.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
When using the binding tables to access arrays of YCbCr descriptors we
did not consider the offset of the accessed element. We can't do a
simple multiple because the binding table entries are tightly packed.
For example element 0 of the array could use 2 entries/planes and
element 1 could use 2 entries/planes.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 3bb8768b9d ("anv: toggle on support for VK_EXT_ycbcr_image_arrays")
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
If descriptorType is VK_DESCRIPTOR_TYPE_STORAGE_IMAGE
or VK_DESCRIPTOR_TYPE_INPUT_ATTACHMENT, the imageView member of each
element of pImageInfo must have been created with the identity swizzle.
Fixes: d2aa65eb
Signed-off-by: Danylo Piliaiev <danylo.piliaiev@globallogic.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
This was an unintended artifact of my testing of bindless images. We
should be choosing bindless or not dynamically.
Fixes: c0d9926df7 "anv: Use bindless handles for images"
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Now that we have the descriptor buffer mechanism, emulated texture
swizzle can be implemented in a very non-invasive way. Previous
attempts all tried to extend the push constant based image param
mechanism which was gross. This could, in theory, be done much faster
with a magic back-end instruction which does indirect MOVs but Vulkan on
IVB is already so slow this isn't going to matter much.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=104355
Cc: "19.1" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Now that everything is in place to do bindless for all resource types
except input attachments and UBOs, VK_EXT_descriptor_indexing is
"trivial".
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Marcin Ślusarz
Anuj Phogat
Jason Ekstrand
Jason Ekstrand
Christian Gmeiner
Eric Anholt
Lionel Landwerlin
Marek Olšák
Eric Engestrom
Danylo Piliaiev