The intention of inst->is_partial_write() is that it should return true
when any REG_SIZE (32B) chunk of inst's destination is written but not
fully overwritten. This can be used to tell whether inst combines new
data with existing data, or screens off any previous writes, so the old
values are no longer required.
The existing (exec_size * brw_type_size_bytes(this->dst.type) < 32)
check doesn't work in a number of cases. For example, LSC block loads
have exec_size == 1 and force_writemask_all set, but may write multiple
full registers of data. (Currently, we only see them with exec_size 1
after logical-send-lowering, so our SHADER_OPCODE_SEND special case
was covering those.) We had also special cased UNDEF.
Instead, we can simply check:
1. Predication
2. !inst->dst.contiguous()
3. inst->dst.offset % REG_SIZE != 0
4. inst->size_written % REG_SIZE != 0
We had the first three already, but #4 is new. If either #3 or #4
are true, then that implies there is a REG_SIZE chunk of the destination
which is written, but not entirely written, so it's a partial write.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Reviewed-by: Rohan Garg <rohan.garg@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30828>
The intention here is to detect ALU hardware instructions, but not
virtual instructions that haven't been explicitly whitelisted.
For some reason we had arbitrarily hardcoded 128 here, but our virtual
opcodes don't start at 128. They start at NUM_BRW_OPCODES. So, use
that instead.
This prevents regressions later when we delete some opcodes, shifting
some virtual opcodes into the 72-128 range.
Cc: mesa-stable
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Caio Oliveira <caio.oliveira@intel.com>
Reviewed-by: Rohan Garg <rohan.garg@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/30828>
Near/far depth clip was implemented by setting the low/high_depth_clamp
to -/+INFINITY, which is invalid on Mali.
This commit removes the modification of the depth clamp values and
enables depth clipping by setting the depth_cull_enable state in the ZS
descriptor for v9+ (the equivalent pre-v9 RSD state is already set
correctly) and setting the primitive cull flags correctly for both.
Finally, it disables PIPE_CAP_DEPTH_CLIP_DISABLE_SEPARATE for v9+, as
both plane clips are controlled by a single value.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/11506
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31041>
Instead of unpacking the x86_64_build container and its billion build
dependencies every time, switch to using only what's in the minimal
pyutils container, and the Python scripts we get as an artifact from the
python-test job. Pulling the artifacts from S3 rather than using GitLab
is also much more efficient.
This should substantially reduce the runtime required to get to testing.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31151>
Currently, our jobs which only want to run a little bit of python -
python-test and the LAVA jobs - pull the entire x86_64-build image,
which is both massive, and massively unnecessary.
Create a separate image which only carries what we need to run our
Python tests and utilities, and switch python-test to using that.
Signed-off-by: Daniel Stone <daniels@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31151>
This reverts commit 0b85476d86.
When mapping a BO in v3d, the map keeps forever until freeing the BO. If
later the map is required again, we reuse the map instead of doing the
map from scratch.
This saves calling map/unmap continuously, as well as a mechanism to
keep control of the map usage, like a reference count.
Thus, when reallocating a BO, if it is mapped it just means the map was
used in the past, but not necessarily it is in use right now.
The reverted commit was causing performance regressions in multiple
applications, reducing from 60fps to 5fps.
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/11783
Backport-to: 24.2
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/31049>
We currently have 3 runners with 7950X3D CPUs and 7900XT GPUs which
can rip through VKCTS in ~16 minutes.
Since this is over the 15 minutes threshold, we parallelize the job to
get under 15 minutes, which nets up ~10 minutes total runtime.
Signed-off-by: Martin Roukala (né Peres) <martin.roukala@mupuf.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31085>
Coverity alerts that the uint32_t pointer I was passing into
isl_color_value_pack() could possibly be used as an array. The value is
being used as such, but only the first element of that array should be
accessed. That's because the depth buffer formats I'm also passing into
the function only have a single channel, R. Nonetheless, let's update
the code to avoid the warning.
Reviewed-by: Ivan Briano <ivan.briano@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31123>
this is tricky since the ownership model here works the other
way around from normal: the resolve surface's lifetime is
determined by the resource's lifetime, meaning that
the surface should be destroyed by the resource instead
of the resource being destroyed by the surface
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31124>
panfrost_invert_swizzle produces a (one-sided) inverse, that is a
function g such that g(f(x)) = x. However, it is actually used in
some cases where we want a two-sided inverse, where we also have
f(g(x)) = x. An example is its use to pre-swizzle border color
values so that a later pass will un-swizzle them. If the swizzle
is not one-to-one this two sided inverse does not exist. However,
we can do better than the original code, e.g. for an RRR1 swizzle
inverse produced was originally B000, which when applied on the wrong
side results in BBB1 as output, whereas R000 would produce the
desired RRR1 output. Using the first valid component we see, rather
than the last one, is thus usually better.
The "correct" solution is to re-write all the code that uses
an inverse to handle non-unique inverses. But frankly these uses only
crop up in fairly niche cases like tests, and it's probably not worth
spending a lot of effort to deal with these edge cases when this
patch fixes most of them.
Fixes some failing piglit ext_framebuffer_multisample tests.
Reviewed-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/31060>
