Larger arrays get lowered to scratch access later, as that turns out
to be faster, while the if-chain is actually faster for arrays of
2 elements.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7726>
PER_QUAD TMU lookups will partially override the predication mask on TMU
writes. If some but not all lanes in a quad are predicated out, setting
PER_QUAD will force them all to be enabled. This can result in TMU
access to bogus addresses when in nonuniform control flow. Also, since
PER_QUAD is needed to make sure derivatives work with helper
invocations, and derivatives are undefined in nonuniform control flow,
there is no reason to leave it enabled in this case.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7726>
Similarly to if statements, uniform loops are now emitted without
predication, using simple branches for breaks and continues. The
uniformity of the loop is determined by running the
nir_divergence_analysis pass.
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7726>
There was only one caller, an error path in mesa/st. But this is now
incorrect as we need align_free(). Just remove it.
Fixes: 55e853d823 ("mesa/st: Allocate the gl_context with 16-byte alignment.")
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8801>
Fix defect reported by Coverity Scan.
Side effect in assertion (ASSERT_SIDE_EFFECT)
assignment_where_comparison_intended: Assignment job->ez_state =
VC5_EZ_DISABLED has a side effect. This code will work differently in a
non-debug build.
Fixes: cec2ed7c80 ("v3dv: fix disabling Early Z for the whole frame")
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8666>
See mtypes.h - it gives you the top level view of how it works, and other
code contains a lot of comments.
Reviewed-by: Zoltán Böszörményi <zboszor@gmail.com>
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8606>
glPush/PopAttrib had 2 defects:
- Only MaxTextureUnits were pushed and popped (which is 8 at most).
- If some texture units had been unused, we would still push/pop them.
This fix changes how many units we push and pop to NumCurrentTexUsed,
which is the maximum texture unit + 1 that was used to by the user.
This commit:
- In glPushAttrib, save NumCurrentTexUsed texture units.
- In glPopAttrib, restore the same number of texture units, and if
NumCurrentTexUsed has increased since glPushAttrib, bind 0 in the newly
used units to get them to the state before glPushAttrib.
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8181>
Since default texture objects are bound in almost all texture units and all
texture targets, we can skip saving and restoring their attributes if we
just do that for default texture objects outside the loop saving all texture
units.
This reduces CPU time spent in glPushAttrib from 2.1% to 1.35% in one
subtest of viewperf13/catia.
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8181>
glPushAttrib only saves values and names. It doesn't save objects.
Reviewed-by: Zoltán Böszörményi <zboszor@gmail.com>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8181>
This is based on the GL 4.6 Compatibility profile spec. Some fields just
don't have to be restored. This decreases CPU overhead.
Reviewed-by: Zoltán Böszörményi <zboszor@gmail.com>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8181>
PopAttribState records all state changes. glPopAttrib uses it to skip
restoring state groups that haven't been changed.
This eliminates a lot of glPopAttrib overhead.
Reviewed-by: Zoltán Böszörményi <zboszor@gmail.com>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8181>
I reviewed all affected states and this if the minimum attrib mask.
Reviewed-by: Zoltán Böszörményi <zboszor@gmail.com>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8181>
There were quite a few. Hopefully the comments make it clear.
Reviewed-by: Zoltán Böszörményi <zboszor@gmail.com>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8181>
The idea is to restore only those states in glPopAttrib that have been
changed. This will reduce glPopAttrib overhead a lot.
This is based on the state tables in the OpenGL 4.6 Compatibility Profile
specification and many extension specifications.
This code might superfluously flag GL_TEXTURE_BIT for texture and
sampler functions that don't have effect on states popped by glPopAttrib,
but I don't wanna test my luck and make a mistake. Thus,
if _NEW_TEXTURE_OBJECT if flagged, GL_TEXTURE_BIT is usually flagged too.
This has no effect on glPopAttrib yet. glPopAttrib will use this in a later
commit. This only adds attrib masks into FLUSH_VERTICES based on specs.
Reviewed-by: Zoltán Böszörményi <zboszor@gmail.com>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8181>
Move the fields that should be restored by glPopAttrib into the Attrib
structure, which is the only portion of texture objects that is restored
by glPopAttrib. Also moves fields that should not be restored by glPopAttrib
out of the Attrib structure.
This is based on the GL 4.6 Compatibility spec.
Reviewed-by: Zoltán Böszörményi <zboszor@gmail.com>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8181>
This field is immutable because each texture unit is an array of texture
target slots and only matching targets can be bound into each slot.
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8181>
_mesa_ActiveTexture changes CurrentUnit and the texture matrix stack
if the matrix mode is a texture matrix. In these cases, the texture matrix
stack is not touched.
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8181>
In _mesa_PopAttrib, we call FLUSH_VERTICES at the beginning, so we don't
need to do it again.
In the MultiMode functions, we call standard Draw functions
using the dispatch, which do the flushing by themselves.
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8181>
so that we don't have to enter the state emit loop and invoke the more
complicated function si_emit_graphics_shader_pointers.
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8794>
Move statements that use the least number of local variables as close
to the beginning as possible. Also move local variables closer to their use.
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8794>
Move statements that use the least number of local variables as close
to the beginning as possible. Also move local variables closer to their use.
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8794>
When using the prefetch with VS_ONLY=true followed by VS_ONLY=false,
we tested the VS_ONLY bits in the mask when executing VS_ONLY=false where
the bits were always 0. It's also useless to clear the prefetch mask when
VS_ONLY=true.
This commit skips those tests by splitting the function properly using
BEFORE_DRAW and AFTER_DRAW template parameters.
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8794>
Decreasing the time spent in radeon_cs_memory_below_limit is the motivation.
Reviewed-by: Zoltán Böszörményi <zboszor@gmail.com>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8794>
The issues fixed by the removal happen when a module has multiple
entry points and conflicting global variables. Neither conditions are
expected in a library.
Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8786>
Not only these are recalculated in nir_shader_gather_info, but
currently they are also counting all the images / textures in the
module instead of in the shader (entrypoint).
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8786>
The _ModelProjectMatrix matrix embedded inside has members inside of it
marked as 16-byte aligned, and so the context also has to be 16-byte
aligned or access to those members would be invalid. I believe the
compiler used this to use better 16-byte-aligned load/stores to other
members of the context, breaking when the context's alignment was only 8
(as normal mallocs guarantee).
Fixes: 3175b63a0d ("mesa: don't allocate matrices with malloc")
Tested-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/8783>
