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>
This commit changes anv to put bindless handles and sampler pointers
into the descriptor buffer and use those instead of bindful when we run
out of binding table space. This "spilling" of descriptors allows to to
advertise an almost unbounded number of images and samplers.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Instead of setting it manually, call the helper. When setting
descriptor sets becomes more complicated than just setting some struct
values, this will keep immutable sampler handling correct.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
We add two new texture sources for bindless surface and sampler handles.
Bindless surface handles are expected to be pre-shifted so that the
20-bit surface state table index is in the top 20 bits of the 32-bit
handle. This lets us avoid any extra shifts in the shader. Bindless
sampler handles are 32-byte aligned byte offsets from general state base
address. We use 32-byte aligned instead of 16-byte aligned to avoid
having to use more indirect messages than needed. It means we can't
tightly pack samplers but that's probably not a big deal.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
When we have a bindless sampler, we need an instruction header. Even in
SIMD8, this pushes the instruction over the sampler message size maximum
of 11 registers. Instead, we have to lower TXD to TXL.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
This commit adds a new way for ANV to do SSBO bindings by just passing a
GPU address in through the descriptor buffer and using the A64 messages
to access the GPU address directly. This means that our variable
pointers are now "real" pointers instead of a vec2(BTI, offset) pair.
This carries a few of advantages:
1. It lets us support a virtually unbounded number of SSBO bindings.
2. It lets us implement VK_KHR_shader_atomic_int64 which we couldn't
implement before because those atomic messages are only available
in the bindless A64 form.
3. It's way better than messing around with bindless handles for SSBOs
which is the only other option for VK_EXT_descriptor_indexing.
4. It's more future looking, maybe? At the least, this is what NVIDIA
does (they don't have binding based SSBOs at all). This doesn't a
priori mean it's better, it just means it's probably not terrible.
The big disadvantage, of course, is that we have to start doing our own
bounds checking for robustBufferAccess again have to push in dynamic
offsets.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
In order to avoid the potential overhead of A64 operations on all SSBO
ops, we look for those SSBO ops where we can get to the descriptor set
from the SSBO access operation and lower those to a binding-table
approach. When robustBufferAccess is enabled, this lets the hardware do
the bounds checking for us. It also avoids some potentially expensive
64-bit integer calculations.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
This is more descriptive and a bit nicer than checking for gen >= 8 &&
use_softpin everywhere.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
We're about to start doing 64-bit pointer calculations in ANV. They
will get applied after brw_preprocess_nir which is where we currently do
64-bit integer arithmetic lowering. Because we're adding 64-bit integer
arithmetic after the initial lowering has happened, we need to lower
again.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
If the number of surfaces or samplers exceeds what we can put in a
table, we will want to spill out to bindless. There is no bindless
support yet but this gets us the basic framework that will be used by
later commits.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
This commit just sorts the bindings by how often they're used vs the
array size of the binding. This will let us make more nuanced decisions
about what goes in the binding table vs. what to make bindless.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
This also fixes a bug where we mis-calculate maximum binding table sizes
and may return true in vkGetDescriptorSetLayoutSupport even for sets too
large to fit in a binding table.
Fixes: ddc4069122 "anv: Implement VK_KHR_maintenance3"
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
This is really where they belong; not push constants. The one downside
here is that we can't push them anymore for compute shaders. However,
that's a general problem and we should figure out how to push descriptor
sets for compute shaders. This lets us bump MAX_IMAGES to 64 on BDW and
earlier platforms because we no longer have to worry about push constant
overhead limits.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
We spend a lot of time in the driver adding things to hash sets to track
residency. The reality is that a properly built Vulkan app uses large
memory objects and sub-allocates from them. In a typical frame, most of
if not all of those allocations are going to be resident for the entire
frame so we're really not saving ourselves much by tracking fine-grained
residency. Just throwing everything in the validation list does make it
a little bit more expensive inside the kernel to walk the list and
ensure that all our VA is in order. However, without relocations, the
overhead of that is pretty small.
If we ever do run into a memory pressure situation where the fine-
grained residency could even potentially help, we would likely be
swapping one page out to make room for another within the draw call and
performance is totally lost at that point. We're better off swapping
out other apps and just letting ours run a whole frame.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
If the last graphics pipeline bound to the command buffer has enough
space in its VS URB entries for Blorp then avoid reconfiguring the URB
partitions.
v2: s/0/MESA_SHADER_VERTEX/ (Caio)
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
There was an assumption that num_thread_per_eu would be set in the
Gen8 features. Since this is mostly the same of all gen8->11 (except
GEN9_LP that overwrites it) let's just factor it out.
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: mesa-stable@lists.freedesktop.org
Acked-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Anuj Phogat anuj.phogat@gmail.com
The current register allocator has a concept of "spill benefit" which is
based on the number of nodes with which a given node interferes. The
idea is that you want to spill stuff with high interference because
those are the most likely registers to help when spilling. However,
this fails to take into account the length of the live range so the
allocator frequently picks "cheap" (not many uses) registers which are
actually very short lived and so spilling them doesn't help with the
pressure situation.
This commit takes into account the length of the live range to make
long-lived registers more likely to get spilled than short-lived ones.
This encourages the spill chooser to choose slightly larger registers
which will affect a larger area of the program and hopefully we have to
spill fewer of them to get the same reduction in over-all register
pressure.
Shader-db results on Kaby Lake:
total spills in shared programs: 23664 -> 12050 (-49.08%)
spills in affected programs: 19243 -> 7629 (-60.35%)
helped: 296
HURT: 8
total fills in shared programs: 32028 -> 25139 (-21.51%)
fills in affected programs: 20378 -> 13489 (-33.81%)
helped: 295
HURT: 16
Of course, most of that is in Deus Ex...
Shader-db results on Kaby Lake (without Deus Ex):
total spills in shared programs: 6479 -> 5834 (-9.96%)
spills in affected programs: 3231 -> 2586 (-19.96%)
helped: 40
HURT: 4
total fills in shared programs: 17165 -> 17099 (-0.38%)
fills in affected programs: 6951 -> 6885 (-0.95%)
helped: 40
HURT: 7
Even without Deus Ex, the spill help is pretty respectable. The worst
hurt shaders were one compute shader in Aztec Ruins and one fragment
shader in KSP that were each hurt by around 13% fill 9% spill.
VkPipeline-db results on Kaby Lake:
total spills in shared programs: 9149 -> 8069 (-11.80%)
spills in affected programs: 5197 -> 4117 (-20.78%)
helped: 27
HURT: 16
total fills in shared programs: 26390 -> 25477 (-3.46%)
fills in affected programs: 12662 -> 11749 (-7.21%)
helped: 24
HURT: 22
The Vulkan results were decidedly more mixed but we don't have nearly as
many apps in that database yet.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Normally fsign generates -1, 0, or +1. The new scale factor, S, causes
fsign to generate -S, 0, or +S.
v2: Rebase on v2 changes in previous commit.
v3: Rebase on 85c35885b3 ("nir: Rework nir_src_as_alu_instr to not take
a pointer").
Reviewed-by: Matt Turner <mattst88@gmail.com> [v2]
Other nir_src_as_* functions just take a nir_src. It's not that much
more memory copying and the constness preserving really isn't worth the
cognitive dissonance.
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
This 3d performance workaround was initially put in the kernel but the
media driver requires different settings so the register has been
whitelisted in i915 [1] and userspace drivers are left initializing it as
they wish.
[1] : https://patchwork.freedesktop.org/series/59494/
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
This 3d performance workaround was initially put in the kernel but the
media driver requires different settings so the register has been
whitelisted in i915 [1] and userspace drivers are left initializing it as
they wish.
[1] : https://patchwork.freedesktop.org/series/59494/
Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
v2 (Jason):
- Merge shaderFloat16 and shaderInt8 enablement into a single patch.
- Merge extension enable.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net> (v1)
v2:
- Merge Float16 and Int8 capabilities into a single patch (Jason)
- Merged patch that enabled SPIR-V front-end checks for these caps
(except for Int8, which was already merged)
v3:
- Keep capabilities sorted (Jason)
v4:
- SpvCapabilityFloat16 support already added in master (Juan)
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net> (v1)
v2:
- Adapted unit tests to make them consistent with the changes done
to the validation of half-float conversions.
v3 (Curro):
- Check all the accummulators
- Constify declarations
- Do not check src1 type in single-source instructions.
- Check for all instructions that read accumulator (either implicitly or
explicitly)
- Check restrictions in src1 too.
- Merge conditional block
- Add invalid test case.
v4 (Curro):
- Assert on 3-src instructions, as they are not validated.
- Get rid of types_are_mixed_float(), as we know instruction is mixed
float at that point.
- Remove conditions from not verified case.
- Fix brackets on conditional.
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
v2:
- Add some tests with UB type too (Jason)
v3:
- consider implicit conversions from 2src instructions too (Curro).
v4:
- Do not check src1 type in single-source instructions (Curro).
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net> (v2)
v2:
- Consider implicit conversions in 2-src instructions too (Curro)
- For restrictions that involve destination stride requirements
only validate them for Align1, since Align16 always requires
packed data.
- Skip general rule for the dst/execution type size ratio for
mixed float instructions on CHV and SKL+, these have their own
set of rules that we'll be validated separately.
v3 (Curro):
- Do not check src1 type in single-source instructions.
- Check restriction on src1.
- Remove invalid test.
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
The section 'Execution Data Types' of 3D Media GPGPU volume, which
describes execution types, is exactly the same in BDW and SKL+.
Also, this section states that there is a single execution type, so it
makes sense that this is the wider of the two floating point types
involved in mixed float mode, which is what we do for SKL+ and CHV.
v2:
- Make sure we also account for the destination type in mixed mode (Curro).
Acked-by: Francisco Jerez <currojerez@riseup.net>
It is very likely that this optimzation is never useful and we'll probably
just end up removing it, so let's not bother adding more cases to it for
now.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
NIR already has these and correctly considers exact/inexact qualification,
whereas the backend doesn't and can apply the optimizations where it
shouldn't. This happened to be the case in a handful of Tomb Raider shaders,
where NIR would skip the optimizations because of a precise qualification
but the backend would then (incorrectly) apply them anyway.
Besides this, considering that we are not emitting much math in the backend
these days it is unlikely that these optimizations are useful in general. A
shader-db run confirms that MAD and LRP optimizations, for example, were only
being triggered in cases where NIR would skip them due to precise
requirements, so in the near future we might want to remove more of these,
but for now we just remove the ones that are not completely correct.
Suggested-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
There are no 8-bit immediates, so assert in that case.
16-bit immediates are replicated in each word of a 32-bit immediate, so
we only need to check the lower 16-bits.
v2:
- Fix is_zero with half-float to consider -0 as well (Jason).
- Fix is_negative_one for word type.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
At the very least we need it to handle HF too, since we are doing
constant propagation for MAD and LRP, which relies on this pass
to promote the immediates to GRF in the end, but ideally
we want it to support even more types so we can take advantage
of it to improve register pressure in some scenarios.
v2 (Jason):
- Support 64-bit types too.
- Check if we need to set the half-float flag if the immediate already
existed.
- Multiply the size of the immediate by the width of the copy
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
The hardware only allows a stride of 1 on a Byte destination for raw
byte MOV instructions. This is required even when the destination
is the NULL register.
Rather than making sure that we emit a proper NULL:B destination
every time we need one, just fix it at emission time.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
