This makes it explicit that this intrinsic is only for SSBOs. For the
v3dv driver, we'll be adding a get_ubo_size intrinsic and we want to be
able to distinguish between the two.
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6812>
There's no sense in planning out an upload that we won't be able to
actually upload due to the limit. This means that we can keep making
other loads pushable, even after we find one that wouldn't be, and we
don't fill the const file with UBO data for a load we couldn't promote.
total instructions in shared programs: 8096655 -> 8044344 (-0.65%)
total constlen in shared programs: 1447824 -> 1411384 (-2.52%)
total cat6 in shared programs: 97824 -> 89983 (-8.02%)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6359>
Now that NIR doesn't lose the original base/range on the
nir_lower_uniforms_to_ubo() path, we get a lot more indirect arrays
uploaded in shader-db.
total instructions in shared programs: 8125988 -> 8103788 (-0.27%)
total constlen in shared programs: 1313096 -> 1448864 (10.34%)
total cat6 in shared programs: 104089 -> 97824 (-6.02%)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6359>
This also fixes the inverted last parameter of nir_lower_flrp in most drivers.
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6599>
Rework setup_{input,output} to be called during emit_intrinsic, in a way
which allows struct/array/matrix type varyings to work.
This allows turnip to pass dEQP-VK.glsl.linkage.varying.struct.*
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6181>
This allows overlapping inputs, which is required for the next patch which
makes it so setup_input may be called multiple times for each input.
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6181>
This gets us fewer comparisons in the shaders that we need to optimize
back out, and reduces backend code.
total instructions in shared programs: 11547270 -> 7219930 (-37.48%)
total full in shared programs: 334268 -> 319602 (-4.39%)
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6378>
It was always fneu but naming it fne causes confusion from time to time. So
lets rename it. Later we also want to add other unordered and fne, this is
a smaller preparation for that.
Signed-off-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6377>
The OpenCL image_width/height/depth functions have variants which can
take an LOD parameter. More importantly, LLVM-SPIRV-Translator always
generates OpImageQuerySizeLod even if the LOD is guaranteed to be zero.
Given that over half the hardware out there has an LOD field for image
size queries (based on a rudimentary scan through their NIR -> whatever
code), we may as well just add the source to the NIR intrinsic. If this
is ever a problem for anyone, the lowering is pretty trivial.
I've also added asserts to everyone's drivers that should alert them if
they ever see an LOD other than zero. This will never happen with GL or
Vulkan so there's no need for panic.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6396>
The *2 here would bump into the *2 in regset, causing assertion failures
dumping CS programs. Just set the mergedregs flag on a6xx, and don't
duplicate the mergedregs logic. If you're dealing with new HW where we
don't know if mergedregs is set, you may need to tweak the flag during
disasm setup for the stats to make sense.
Fixes: f7bd3456d7 ("freedreno: deduplicate a3xx+ disasm")
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6323>
This matches what ir3.c does in the mergedregs case: just count max full
reg used. This flag is unset so far, but will be soon and keeps our
output comparable between blob and freedreno.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6323>
Move the handling for catching asserts when we start decoding garbage
into disasm-a3xx. This way it can also cover other cases where cffdec
tries to disassemble memory, such as SP_xS_OBJ_START.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6242>
Add tracking for # of instructions per category, similar to the last
patch. Also add a few other shader-db stats that were missing on the
disasm side, to make it easier to compare to shaders from cmdstream
traces.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6243>
I got tripped up again with the index vs count vs size fields and I'd
rather we didn't store the redundant info. Settle on immediates_count as
"how many dwords of immediates we have"
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5990>
We would be making a MOV from a u32, when we should be loading from a
16-bit value. This likely didn't bite us because we only do mediump in FS
and CS so far, and indirect uniforms are usually in a VS (and usually
highp).
Reviewed-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6179>
We're trying to get rid of the options argument passed to
nir_lower_int64() and use the nir_options.lower_int64_options instead.
But before we can do that we must patch nir_lower_int64() callers
that don't have this field properly set.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Rob Clark <robdclark@chromium.org>
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5588>
We could lower *some* accesses to a UBO but not others. In this case,
we would have a valid range, but would have skipped tracking that the
UBO is accessed as a UBO rather than push constants.
Fixes one issue with asphalt9, that was a result of having `ldc` without
having emit UBO state.
See: https://gitlab.freedesktop.org/mesa/mesa/-/issues/3067
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6117>
Instead of having separate lists of variables, roughly sorted by mode,
use a single list for all shader-level NIR variables. This makes a few
list walks a bit longer here and there but list walks aren't a very
common thing in NIR at all. On the other hand, it makes a lot of things
like validation, printing, etc. way simpler. Also, there are a number
of cases where we move variables from inputs/outputs to globals and this
makes it way easier because we no longer have to move them between
lists. We only have to deal with that if moving them from the shader to
a nir_function_impl.
Reviewed-by: Rob Clark <robdclark@chromium.org>
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5966>
Merge the extra tracking that is useful for generating stats from asm
(as opposed to ir), and for guestimating things like inputs and outputs
(mostly useful for r/e) into ir3's version and drop cffdec's version.
There is a small change in disasm output for the decode tools, in that
it no longer prints the used consts, but rather just the max accessed
const. This is the more useful piece of information, and avoids making
the shared regmask type big enough to deal with the const reg file.
Additional error checking for invalid regids causes crashdec to bail
out sooner when decoding memory that *might* hold valid instructions.
Also, crashdec no longer prints stats, because stats aren't very useful
when trying to decode random instruction memory (which might or might
not be valid instructions).
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6070>
To unify the ir3 disasm code, we need to add in the regmask based
register tracking from cffdec's version of the disassembler. Split
out regmask (or at least the part that doesn't depend on ir3) so
it can be shared.
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6070>
When changing the patch to use an offset instead of a bool, the type was
accidentally left as bool.
Fixes: f472c98443 ("freedreno/ir3: add support for a650 tess shared storage")
Signed-off-by: Jonathan Marek <jonathan@marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5765>
We theoretically could push these sizes to the const file
opportunistically, which appears to be what the blob does, but the
maximum number of SSBO's is way too big to do that unconditionally. Just
use resinfo to get the size for now.
Fixes on turnip: dEQP-VK.ssbo.unsized_array_length.*
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/6012>
I had a stp testcase that was getting its offset wrong, and by twiddling
bits and feeding it to qc disasm, I found that the comment was sort of
right: some the cat6a bits implicated in the old comment do get used, as
the high bits of the cat6c offset. Reallocating those bits also fixes how
we were getting r960.y for r0.y.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5815>
This started with making note of some ldp/stp instructions from the blob
and how we differ from them. In the process of fixing it, I accidentally
modified behavior of other opcodes, and the other instructions listed will
keep us from doing that. I also dropped an old stl test that looks like I
took from after a shader 'end' instruction.
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5815>
Properly handle the difference between split and merged register file
when determining where arrays can fit without conflicting with other
arrays or pre-colored instructions.
1) if not mergedregs, only consider other things with same precision
as potentially conflicting
2) if mergedregs, calculate everything in therms of half-regs and
convert back to fullregs in the end
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5957>
We shouldn't divide-by-two for half-reg arrays. We set the proper node
interference class, based on `arr->half`.
Fixes a RA fail with 16b arrays:
src/freedreno/ir3/ir3_ra.c:633: name_to_array: Assertion `!"invalid array name"' failed.
Caused by use/def iterators returning `arr->length` vreg namess, but
only assigning the array half that many names.
Also, since we are assigning unique vreg names to each array element,
there is no need to try and convert from half-reg to it's conflicting
full reg when pre-coloring the array elements. Getting us closer to
having half-arrays work sanely with split-register-file (a5xx and
earlier).
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5957>
legalize_block() can get run multiple times, which I didn't notice when
adding fine derivs support. Other instruction clones change things such
that the legalization won't trigger again, but that didn't apply to the
DS.PP legalization. To keep someone else from tripping over this, split
the one-shot legalization out of the iterative sync flag application.
Fixes failures in dEQP-VK.glsl.derivate.dfdxfine.*
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/3198
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5699>
