It will get turned into SSA and copy-propagated in NIR, no need to walk
the IR collapsing it here.
iris shader-db results appear to be noise:
total instructions in shared programs: 8932195 -> 8932147 (<.01%)
instructions in affected programs: 537 -> 489 (-8.94%)
LOST: 12
GAINED: 11
lost/gained are simd32 switches in unigine, l4d2, portal2, asphalt9.
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17613>
Without it we got a metadata assert:
deqp-vk: ../src/compiler/nir/nir_metadata.c:108: nir_metadata_check_validation_flag: Assertion `!(function->impl->valid_metadata & nir_metadata_not_properly_reset)' failed
if we try to use NIR_PASS(_, instead of NIR_PASS_V (that among other
things, do more validations).
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17609>
Previously we would just unroll the loop one extra iteration and let
other optimisation passes clean up the mess. This worked to a degree
but if the loop happened to be nested inside another loop we would
end up with phi chains that would block other passes from being able
to do the cleanup.
With this commit we explicitly clone the variables create by lcsaa
and insert them directly in the last continue branch after we are done
unrolling. With this optimisation passes can recognise both sides
of the if output the same values and can progress further.
Help with the issues described in:
https://gitlab.freedesktop.org/mesa/mesa/-/issues/6051
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17611>
If the driver can do 16-bit ALU ops, then store RelaxedPrecision phi
values into 16-bit NIR variables with downconverts/upconverts on the way
in/out.
This has no impact on shader-db on freedreno (not that we have a ton of
GLES content there), but it does cause an ANGLE-translated CTS shader on
vulkan to get consistent conversions between two copies of a value, and
avoid a test bug.
Reviewed-by: Emma Anholt <emma@anholt.net>
Closes: #6585
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14018>
This adds support for global 64-bit GPU addresses as a pair of
32-bit values. This is useful for platforms with 32-bit GPUs
that want to support VK_KHR_buffer_device_address, which makes
GPU addresses explicitly 64-bit.
With the new format we also add new global intrinsics with 2x32
suffix that consume the new address format.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17275>
We don't really want to base this on a late nir_gather_info for two
reasons:
1) The Vulkan spec says that if a sample-qualified input, SampleID, or
SamplePosition are in the entry-point's interface, you get
per-sample dispatch. This means we really should gather this
information before dead-code has a chance to delete anything.
2) We want to be able to add nir_intrinsic_load_sample_pos intrinsics
as part of lowering passes without causing per-sample interpolation.
This means nir_gather_info needs to stop gathering it.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14020>
We don't really want to base this on a late nir_gather_info for two
reasons:
1) The GL spec says that any static use of a sample-qualified input,
gl_SampleID, or gl_SamplePosition causes per-sample dispatch. This
means we really should gather this information before dead-code has
a chance to delete anything.
2) We want to be able to add nir_intrinsic_load_sample_pos intrinsics
as part of lowering passes without causing per-sample interpolation.
This means nir_gather_info needs to stop gathering it.
For 1, this doesn't actually get us quite there as GLSL IR may have
deleted something already. However, it does get us closer.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14020>
On Intel, we have to do this because we can't ask for the per-sample
barycentrics without setting the per-sample dispatch bit or the GPU will
hang. However, nothing we're doing in this pass is Intel-specific and
it may be a useful optimization for someone else so we may as well make
it a generic NIR pass. This version actually does a bit more than the
current brw_nir_demote_sample_qualifiers() pass as it also handles
pre-nir_lower_io interp_dref_at* as well as a couple system values which
we can easily constant-fold.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/14020>
The set of supported vector sizes in NIR has holes in it. For example, we
support vec5 and vec8, but not vec6 or vec7. However, this pass did not take
that into account, and would happily shrink a vec8 down to a vec7, causing NIR
validation to fail. Instead, the pass should round up to the next supported
vector size.
Fixes NIR validation fail in OpenCL's test_basic hiloeo subtest.
v2: Clamp -> round rename.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17194>
This will be used for vc4, where incorrectly exposing 3D textures accounts
for most of the GLES2 conformance failures it has. This leaves
EXT_texture3d exposed in the (already non-conformant) GL2.1 support it
exposes, which has always been a best-effort thing.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17350>
These opcodes where fixed to return an integer instead of a boolean
value some time ago but the documentation for them was not updated
and still talked about a boolean result.
Fixes: b0d4ee520 ('nir/opcodes: Fix up uadd_carry and usub_borrow')
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17372>
nir_ine_imm(b, nir_iand_imm(b, x, mask), 0) and
nir_i2b(b, nir_iand_imm(b, x, mask)) are common
patterns which become quite messy when they are
part of a larger expression. Clang-format does
not improve things either and we can end up with
some rather interesting looking code.
(RADV ray tracing pipeline and query lowering)
Signed-off-by: Konstantin Seurer <konstantin.seurer@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17242>
This really, really helps on platforms where fabs() isn't free. A great
many shaders use a * frsq(fabs(fdot(a, a))) to normalize a vector.
Since the result of the fdot must be non-negative, the fabs can be
eliminated by an existing algebraic rule.
shader-db results:
r300 (run on R420 - X800XL)
total instructions in shared programs: 1369807 -> 1368550 (-0.09%)
instructions in affected programs: 59986 -> 58729 (-2.10%)
helped: 609
HURT: 0
total vinst in shared programs: 512899 -> 512861 (<.01%)
vinst in affected programs: 1522 -> 1484 (-2.50%)
helped: 36
HURT: 0
total sinst in shared programs: 260690 -> 260570 (-0.05%)
sinst in affected programs: 1419 -> 1299 (-8.46%)
helped: 120
HURT: 0
total consts in shared programs: 957295 -> 957230 (<.01%)
consts in affected programs: 849 -> 784 (-7.66%)
helped: 65
HURT: 0
LOST: 0
GAINED: 3
The 3 gained shaders are all vertex shaders from XCom: Enemy Unknown.
I'm guessing that game is never going to run on my X800XL. :)
i915
total instructions in shared programs: 791121 -> 780843 (-1.30%)
instructions in affected programs: 220170 -> 209892 (-4.67%)
helped: 2085
HURT: 0
total temps in shared programs: 47765 -> 47766 (<.01%)
temps in affected programs: 9 -> 10 (11.11%)
helped: 0
HURT: 1
total const in shared programs: 93048 -> 92983 (-0.07%)
const in affected programs: 784 -> 719 (-8.29%)
helped: 65
HURT: 0
LOST: 0
GAINED: 36
Haswell, Ivy Bridge, and Sandy Bridge had similar results. (Haswell shown)
total instructions in shared programs: 16702250 -> 16697908 (-0.03%)
instructions in affected programs: 119277 -> 114935 (-3.64%)
helped: 1065
HURT: 0
helped stats (abs) min: 1 max: 20 x̄: 4.08 x̃: 4
helped stats (rel) min: 0.48% max: 10.17% x̄: 3.66% x̃: 3.94%
95% mean confidence interval for instructions value: -4.26 -3.89
95% mean confidence interval for instructions %-change: -3.76% -3.56%
Instructions are helped.
total cycles in shared programs: 880772068 -> 880734134 (<.01%)
cycles in affected programs: 2134456 -> 2096522 (-1.78%)
helped: 941
HURT: 324
helped stats (abs) min: 2 max: 2180 x̄: 123.06 x̃: 44
helped stats (rel) min: 0.04% max: 49.96% x̄: 7.08% x̃: 3.81%
HURT stats (abs) min: 2 max: 2098 x̄: 240.33 x̃: 35
HURT stats (rel) min: 0.04% max: 77.07% x̄: 12.34% x̃: 3.00%
95% mean confidence interval for cycles value: -47.93 -12.04
95% mean confidence interval for cycles %-change: -2.87% -1.34%
Cycles are helped.
No shader-db changes on any other Intel platform.
Reviewed-by: Jason Ekstrand <jason.ekstrand@collabora.com>
Reviewed-by: Emma Anholt <emma@anholt.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17181>
Emma Anholt