The loop checking if exec is overwritten didn't check for NULL
instructions, and didn't fix up reg write indices after inserting
instructions.
Fixes: fcd94a8c ("aco: move try_optimize_branching_sequence() to postRA optimizations")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32746>
We need to add a variant of the texld instruction, which is used with a shadow
samper and passed the shadow reference value via src2.
Blob generates such texld's for deqp's GLES3.functional.texture.shadow.2d.* (GC3000).
Fixes spec@arb_depth_texture@texdepth.
Fixes: abe5bd35 ("etnaviv: Switch to isa_assemble_instruction(..)")
Signed-off-by: Christian Gmeiner <cgmeiner@igalia.com>
Reviewed-by: Lucas Stach <l.stach@pengutronix.de>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32926>
Marek recently changed hole_size to be signed, rather than unsigned.
A negative hole_size means that the two loads overlap - and thus are
prime candidates to be combined.
My original hole_size handling was:
if hole_size > 4 * (8 - low->num_components) then don't vectorize
For non-overlapping loads, this worked: NIR's largest vector is vec16,
and if low was already a vec16, combining it with anything would exceed
that, so it'd never be considered. That meant low would always be a
vec8 or less, so (8 - low->num_components) was a positive number.
Now that we see overlapping loads, we can see a vec16 low, vec4 high,
and also a negative hole size, giving us fun comparisons like:
-16 > 4 * (8 - 16) => -16 > -32 => true, don't vectorize
Which is absolutely the wrong thing to do, because the high load's data
is entirely included within the former load's data.
The idea here was to make sure the second load would be able to pack at
least one component into the first's V8 result. But even this isn't the
best, because...even if it's simply adjacent, doing one V16 load is more
efficient than requesting two back to back V8 loads.
So, we just simplify down to a static check: if there's an entire V8 of
hole, don't vectorize. This already won't happen because the core pass
has max_hole set to 28 bytes (7 32-bit components), but that could
change based on the needs of other drivers, so let's be defensive.
fossil-db results on Alchemist:
Instrs: 161533978 -> 161295137 (-0.15%); split: -0.20%, +0.05%
Subgroup size: 8092544 -> 8092568 (+0.00%)
Send messages: 7915233 -> 7844503 (-0.89%); split: -0.94%, +0.05%
Cycle count: 16577700697 -> 16702609256 (+0.75%); split: -0.59%, +1.35%
Spill count: 72338 -> 67226 (-7.07%); split: -7.36%, +0.29%
Fill count: 134058 -> 125980 (-6.03%); split: -6.83%, +0.80%
Scratch Memory Size: 4092928 -> 3786752 (-7.48%); split: -7.53%, +0.05%
Max live registers: 33031460 -> 32945994 (-0.26%); split: -0.27%, +0.01%
Max dispatch width: 5778384 -> 5778536 (+0.00%); split: +0.26%, -0.26%
Non SSA regs after NIR: 179809505 -> 152735471 (-15.06%); split: -15.08%, +0.03%
Fixes: c21bc65ba7 ("nir/opt_load_store_vectorize: make hole_size signed to indicate overlapping loads")
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32932>
When a SEND instruction is a EOT, the scoreboard lowering will not
allocate a new SBID for it, since nothing needs to wait for it. In
Gfx12 this allowed the SEND to get out-of-order $.dst or $.src
dependencies.
Starting on Xe2+ this is not supported anymore, in favor of supporting
more combined modes.
Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32712>
Currently we scale the polygon offset units with a fixed factor,
matching the MRD (minimal resolvable distance) for a 16bpp depth
buffer. This wastes a lot of precision when a 24bpp depth buffer
is used.
Apply the correct MRD scale, depending on the format of the
currently bound depth buffer.
Fixes piglit spec@!opengl 1.4@gl-1.4-polygon-offset.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Christian Gmeiner <cgmeiner@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32756>
This is similar to what link_intrastage_shaders is doing and it
fixes the following test:
KHR-Single-GL46.subgroups.builtin_var.compute.subgroupsize_compute
Which was failing with SPIRV but passing with GLSL, the diff being:
- SPIRV: "subgroup_size: 1"
- GLSL: "subgroup_size: 2"
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Qiang Yu <yuq825@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32698>
The blitter VS expects coords to fit in a signed int16. When this
is not the case, use util_blitter_draw_rectangle instead.
Since util_blitter_draw_rectangle sets vertex elements, we need
to make sure they're properly restored.
The alternative to this fallback would be to pass coordinates
unpacked (so 4 SGPRs instead of 2), but this doesn't fix the
fbo-blit-check-limits test because of uv interpolation precision
issue.
Using 2 triangles instead of a rectangle + disabling
window_space_position helps but then this breaks some GLES3 tests,
like dEQP-GLES3.functional.fbo.blit.rect.nearest_consistency_mag_reverse_src_x
(which doesn't pass either if u_blitter is used for all cases).
Using a single triangle covering the whole rectangles fixes all
cases but it then requires to setup scissors to not write too
much pixels...
So, instead of adding so much complexity, let's use u_blitter
for the "large coordinates" fallback, and keep the rectangle blit
for the other cases.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32698>
This is actually a false positive detected by static analyzer, because
it assumes that `device->instance->meta_cache_enabled` can change
between two execution points.
In order to instruct static analyzer this is not the case, we assing it
to a local variable, and do the checks based on that.
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32819>
Declaring a variable-length array (VLA) based on a variable that can be
0 is declared dangerous.
In this case, the variable can't take value 0, so adding an assertion
fixes the issue.
This was detected by static analyzer.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32819>
We're about to add more sysvals, and the more we add, the bigger the
sysvals region gets, which increases the amount of memory we have to
allocate when push_uniforms are dirty.
Instead of allocating FAUs for all sysvals/push_constants, track FAU
usage per-shader, and pack those. This implies emitting an FAU buffer
per stage instead of trying to share it, but that's an acceptable
trade-off.
While at it, automate the sysval dirty tracking a bit.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Mary Guillemard <mary.guillemard@collabora.com>
Reviewed-by: Lars-Ivar Hesselberg Simonsen <lars-ivar.simonsen@arm.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32415>
Now that panvk_cmd_buffer.h is accessible from
src/panfrost/vulkan/panvk_vX_xxx.c files, there's no reason to pass
a gazillon arguments to blend_emit_descs(). We can just pass a cmdbuf
and let the helper extract the other parameters from there. It also
allows for extra automation, like dirtying the push_uniform buffer
when the new blend config reads the blend constant.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Mary Guillemard <mary.guillemard@collabora.com>
Reviewed-by: Lars-Ivar Hesselberg Simonsen <lars-ivar.simonsen@arm.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32415>
load_vertex_id_zero_base() is supposed to return the zero-based
vertex ID, which is then offset by load_first_vertex() to get
an absolute vertex ID. At the same time, when we're in a Vulkan
environment, load_first_vertex() also encodes the vertexOffset
passed to the indexed draw.
Midgard/Bifrost have a sligtly different semantics, where
load_first_vertex() returns vertexOffset + minVertexIdInIndexRange,
and load_vertex_id_zero_base() returns an ID that needs to be offset
by this vertexOffset + minVertexIdInIndexRange to get the absolute
vertex ID. Everything works fine as long as all the load_first_vertex()
and load_vertex_id_zero_base() calls are coming from the
load_vertex_id() lowering. But as mentioned above, that's no longer
the case in Vulkan, where gl_BaseVertexARB will be turned into
load_first_vertex() and expect a value of vertexOffset in an
indexed draw context.
We thus need to fix the mismatch by introducing two new
panfrost-specific intrinsic so we can stop abusing load_first_vertex()
and load_vertex_id_zero_base().
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Chia-I Wu <olvaffe@gmail.com>
Reviewed-by: Lars-Ivar Hesselberg Simonsen <lars-ivar.simonsen@arm.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/32415>
