If we know that a select will be eliminated once the loop is
unrolled than we don't need to count the instruction towards the
cost of the loop.
This change helps 2 loops unroll in an xcom enemy unknown shader
that is loaded full of these redundant selects.
Acked-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18587>
Here we move the calculation of the instruction cost of the loop
after we have processed other information such as finding the
induction variables. This is useful because we can use this further
information to find instructions that will be eliminated if the
loop was to unroll and therefore give them a cost of 0.
Acked-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18587>
0xffff is an invalid enum.
This was initially implemented without clamping and then reverted because
of the concern that we would randomly get correct enums and not report
errors if we just dropped the high bits.
This packs calls better in glthread_batch.
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18828>
Exec is just a mutable pointer to one of the dispatch tables.
OutsideBeginEnd is what we are actually initializing here. This makes
the original intention clear.
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18828>
The frequently-accessed fields are at the beginning, hoping
we only need to load 1 cache line to get them all.
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18828>
The CPU time spent in _mesa_BindBuffer_no_error decresed from 11.4% to 9.2%.
I explain in the code what is happening here because it's not obvious.
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18828>
Changes:
- the function is inlined, which will help the next commit to improve perf
- use no_error inside the function
- if the function returns true, the buffer is non-NULL, so remove NULL
checking in callers
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18828>
The last parameter is always a constant expression, so this simplifies it.
This is also required by later commits.
Reviewed-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18828>
This makes st_feedback_draw_vbo implement the DrawGallium.
That's the last users of the fallback functions.
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18828>
it's impossible to accurately determine whether an op will use the base,
sized array or the unsized array due to dynamic indexing, and thus it's
impossible to not violate spec by emitting both arrays at offset=0 so they
can overlap
so instead, just emit the runtime array and blast all bounds checking in
vk drivers into the void
Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18784>
this used to correctly calculate runtime array size as zero, but since the
switch to glsl_get_explicit_size(), zero can no longer be returned
Reviewed-by: Dave Airlie <airlied@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18784>
This mirrors the behavior of iris_resource_texture_aux_usage(), which
bypasses the aux metadata when there's no advantage to using it.
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18790>
