Set max outstanding ray queries to 1024. This value can be tuned later
specific to apps.
Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40182>
Applications often miss emitting barriers between a shader
initializing data & another shader writing data in the same location
afterward. This is very common for UAVs (see vkd3d-proton).
Vkd3d-proton does a pretty good job as inserting missing barriers
between UAV clears & writes. But some applications also have similar
issues with custom shaders. Here we introduce an analysis pass that
recognize shaders doing clear/initialization. We'll use that
information in the following commit to insert barriers after those
shaders.
Since Gfx12.5 our HW has become a lot more sensitive to those issues
due to the introduction of an L1 untyped data cache that is not
coherent across the shader units. On Gfx20+, typed data is also L1
cacheable exposing even more issues.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: mesa-stable
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40187>
Coverity notes that we break out of the loop walking the analysis_ranges
early if n_push_ranges >= max_push_buffers, so it notes that
n_push_ranges could already be 3 or 4 (depending on whether we're doing
mesh), and that then if we need the padding we insert another, which
would write past the end of the array.
I don't think this is actually possible in practice, but we can add an
assert to both keep coverity happy and detect that this has actually
happened.
CID: 1681478
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40147>
Shaders are allocated from anv_shader_heap, which is backed by the
util_vma_heap. Rename the VA range field to shader_heap to match current
usage and avoid confusion.
Signed-off-by: Michael Cheng <michael.cheng@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40131>
Binding report test cases are failing for Gen12.5 platforms due to
missing binding events from private bindings for images creating
inconsistencies in the number of unbind events reported from
anv_DestroyImage(). Adding bind events from alloc_private_binding() to
fix the issue.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/37422>
For Gfx125 workloads that use systolic mode, this might mean
an extra PIPELINE_SELECT when flipping between a compute shader
that use the mode and another that doesn't use the mode
(or vice-versa).
Reviewed-by: Iván Briano <ivan.briano@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40014>
There were 2 issues in the commit being fixed :
1. loading from the wrong surface state
2. not being able to have the optimization passes cleanup the
nir_vector_extract()
We fix the first issue by reusing the nir_vector_extract() pattern in
the broken places.
We fix the second issue by reworking the internal vec4 format we use
for passing around descriptor information. In particular we put the
set in its own component so that it can be easily optimized and the
vector extraction constant folded.
Fixes: e94cb92cb0 ("anv: use internal surface state on Gfx12.5+ to access descriptor buffers")
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40076>
Push constants on bindless stages of Gfx12.5+ don't get the data
delivered in the registers automatically. Instead the shader needs to
load the data with SEND messages.
Those stages do get a single InlineParameter 32B block of data
delivered into the EU. We can use that to promote some of the push
constant data that has to be pulled otherwise.
The driver will try to promote all push constant data (app + driver
values) if it can, if it can't it'll try to promote only the driver
values (usually a shader will only use a few driver values). If even
the drivers values won't fit, give up and don't use the inline
parameter at all.
LNL internal fossil-db:
Totals from 315738 (20.08% of 1572649) affected shaders:
Instrs: 155053691 -> 154920901 (-0.09%); split: -0.09%, +0.00%
CodeSize: 2578204272 -> 2574991568 (-0.12%); split: -0.15%, +0.02%
Send messages: 8235628 -> 8184485 (-0.62%); split: -0.62%, +0.00%
Cycle count: 43911938816 -> 43901857748 (-0.02%); split: -0.05%, +0.03%
Spill count: 481329 -> 473185 (-1.69%); split: -1.82%, +0.13%
Fill count: 405617 -> 399243 (-1.57%); split: -1.86%, +0.28%
Max live registers: 34309395 -> 34309300 (-0.00%); split: -0.00%, +0.00%
Max dispatch width: 8298224 -> 8299168 (+0.01%)
Non SSA regs after NIR: 18492887 -> 17631285 (-4.66%); split: -4.73%, +0.08%
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39405>
It's pretty much the same mechanism, except it's a different register
location.
With this change we gain indirect loading support.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39405>
Shaders will often contains things like this :
con 32 %469 = @load_push_constant (%468 (0x30)) (base=0, range=128, align_mul=256, align_offset=48)
We don't need 128 bytes of push constants to do that load.
This will become important when we rely more on base/range in the next
commit to promote things to inline parameters (only 32B of space
available).
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39405>
Those are never used together on a single platform.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39405>
So it can be avoided if we promote it to inline parameters.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39405>
Split emission from pointers programming.
That way we can switch back & forth between blorp & applications
shaders and never emit binding tables, we just reprogram the pointers.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39405>
A side effect is that we make the same decision for simple shaders &
application pipelines which could avoid some reprogramming.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39405>
With perfetto that string is processed later leading to
use-after-free.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: mesa-stable
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39405>
Removes the need for emitting 3DSTATE_BINDING_TABLE_POINTER* commands
to make the HW gather push constants.
According to internal pointers, this been the default behavior on
Gfx11+.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39405>
Blorp emits 3DSTATE_BINDING_TABLE_POINTER_* instructions in 3D mode.
At the moment we're saved by the push constants reemitting the btp but
we'll drop that in the next commit.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: mesa-stable
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39405>
The dEQP optimization in [1] for 1:1 ASTC copies exposed a race
condition where the internal decompression shader reads old data
from the texture cache before the copy finishes.
This patch adds cache flush to ensure the shader sees the newly
copied ASTC blocks. It also fixes the block extent calculation
to use the destination image metadata.
[1] https://gerrit.khronos.org/c/vk-gl-cts/+/17514
Fixes: dEQP-GLES31.functional.copy_image.compressed.viewclass_astc*
v2: Drop CS_STALL and update the bits order (Lionel).
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/40060>
The previous commit enable different command buffers to program the
same 3DSTATE_BINDING_TABLE_POOL_ALLOC instruction even though they
allocated different chunks of binding tables.
Now we can just predicate this programming and skip the stalling,
flushing & invalidation.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39527>
We currently allocate 64KiB chunks of binding table pools for each
command buffers and program the 3DSTATE_BINDING_TABLE_POOL_ALLOC
instruction accordingly.
But 3DSTATE_BINDING_TABLE_POINTERS_* instructions can address 2^20
bytes. So it's possible to have 2 command buffers share the same
programming if they just add some offsets to their
3DSTATE_BINDING_TABLE_POINTERS_* programming and round down
3DSTATE_BINDING_TABLE_POOL_ALLOC addresses to 2^20.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39527>
The new compression scheme introduced in Xe2 also applies to Xe3, so
we're liable for the same bugs.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 2418c91537 ("anv/drirc: disable Xe2 CCS drm modifiers for GTK engine")
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39953>
It was already computed in brw_shader::assign_curb_setup() so we can use it
in brw_assign_urb_setup().
There was a mismatch between assign_curb_setup() and brw_assign_urb_setup() when
push_sizes were not multiple of REG_SIZE, the first one was aligning every
push_sizes before sum it, while brw_assign_urb_setup() was only aligning the sum
of all push_size.
By luck the only places that did not had a push_size aligned to REG_SIZE only
had one push_size, so this was not an issue.
So here also fixing this mismatch and adding an assert to caught any future
mismatch.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39817>
Apparently this a performance regression on our CI as opposed to what
the HW documentation recommends.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39991>
We just read this from the NIR and store it in iris_compiled_shader,
there's no reason for the backend compiler to be involved.
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/39926>
