Some opcodes come with both type and size variants. Right now, only the
size is taken into account. Extend the builder to provide wrappers that
take a nir_type in addition to the bitsize.
While at it, fix wrappers taking a compare operator to use the proper
.{i,s,u} variant based on the comparison (equal and non-equal should
use .i, other comparisons should use .{u,s}).
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9520>
We will have 2 compute jobs per indexed indirect draw, one doing the
min-max index search and one patching the cmdstream. The second compute
job needs to depend on the first one, as well as the previous indirect
draw job to avoid corrupting the indirect draw context which is shared
at the batch level (global dependency).
Instead of handling that case in panfrost_add_job(), extend
panfrost_add_job() to accept an explicit global dependency.
Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9520>
Instead of lowering it in NIR, use the lookup tables as inputs to a
second-order Taylor expansion. shader-db results aren't amazing but keep
in mind this is without backend CSE yet.
total instructions in shared programs: 115913 -> 115707 (-0.18%)
instructions in affected programs: 3151 -> 2945 (-6.54%)
helped: 12
HURT: 0
Instructions are helped.
total nops in shared programs: 84045 -> 84041 (<.01%)
nops in affected programs: 1571 -> 1567 (-0.25%)
helped: 1
HURT: 7
Inconclusive result (value mean confidence interval includes 0).
total clauses in shared programs: 20498 -> 20489 (-0.04%)
clauses in affected programs: 188 -> 179 (-4.79%)
helped: 6
HURT: 0
Clauses are helped.
total quadwords in shared programs: 90395 -> 90291 (-0.12%)
quadwords in affected programs: 2287 -> 2183 (-4.55%)
helped: 12
HURT: 0
Quadwords are helped.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Reviewed-by: Boris Brezillon <boris.brezillon@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9420>
Midgard writeout arguments need to be written to in the same bundle the
writeout happens. Both csel, csel_v and their float variants also
require their conditional to be performed on the same bundle.
This patch prevents scheduling csel the same bundle as a writeout,
fixing the scheduling issue.
But... there's still room for optimizations since in some cases it might
be possible to fit all these instructions in the same bundle.
No shader-db changes.
Signed-off-by: Italo Nicola <italonicola@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9340>
We need to iterate the whole row, we can't be clever and only look at
one side, the symmetry doesn't work like that. See the original paper.
total instructions in shared programs: 69392 -> 69322 (-0.10%)
instructions in affected programs: 9002 -> 8932 (-0.78%)
helped: 82
HURT: 28
Instructions are helped.
total bundles in shared programs: 32225 -> 32155 (-0.22%)
bundles in affected programs: 4286 -> 4216 (-1.63%)
helped: 82
HURT: 28
Bundles are helped.
total quadwords in shared programs: 56102 -> 56102 (0.00%)
quadwords in affected programs: 0 -> 0
helped: 0
HURT: 0
total registers in shared programs: 4552 -> 4572 (0.44%)
registers in affected programs: 298 -> 318 (6.71%)
helped: 18
HURT: 38
Registers are HURT.
total threads in shared programs: 3772 -> 3775 (0.08%)
threads in affected programs: 84 -> 87 (3.57%)
helped: 15
HURT: 14
Inconclusive result (value mean confidence interval includes 0).
total spills in shared programs: 0 -> 0
spills in affected programs: 0 -> 0
helped: 0
HURT: 0
total fills in shared programs: 0 -> 0
fills in affected programs: 0 -> 0
helped: 0
HURT: 0
Fixes: 66ad64d73d ("pan/midgard: Implement linearly-constrained register allocation")
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9338>
Per discussion with Daniel Schürmann on IRC about the joys of SSA form
and why you don't actually need use-def chains. Indeed, I didn't. No
shader-db changes, time difference in shader-db is neglible since the
win from this is particularly for large shaders.
Total runtime of
dEQP-GLES31.functional.ssbo.layout.single_struct_array.single_buffer.std430_instance_array
reduced from 1.04s to 0.77s (25%)
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9164>
Fixes excessive (and failing) spilling in dEQP-GLES31.functional.ssbo.layout.*
shader-db results are a toss up (I suspect we'd see better results if we
tracked register pressure directly):
total instructions in shared programs: 161377 -> 161377 (0.00%)
total nops in shared programs: 121159 -> 121203 (0.04%)
nops in affected programs: 1839 -> 1883 (2.39%)
Nops are HURT.
total clauses in shared programs: 31604 -> 31606 (<.01%)
clauses in affected programs: 38 -> 40 (5.26%)
Inconclusive result (value mean confidence interval includes 0).
total quadwords in shared programs: 130847 -> 130845 (<.01%)
quadwords in affected programs: 1246 -> 1244 (-0.16%)
Inconclusive result (value mean confidence interval includes 0).
total loops in shared programs: 18 -> 18 (0.00%)
total spills in shared programs: 705 -> 705 (0.00%)
total fills in shared programs: 1645 -> 1645 (0.00%)
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9164>
Used in the LD_VAR_IMM. Wondering if preload requirements shouldn't
instead be pushed from the compiler based on actual usage instead of
guessing from the NIR...
Fixes dEQP-GLES31.functional.shaders.multisample_interpolation.sample_qualifier.*
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9105>
Don't hardcode the RT to 0. Affects ES3.0 which already exposes MRT --
despite no dEQP coverage of this particular corner case, apps could hit
this in the wild on 21.0. Fixes
dEQP-GLES31.functional.draw_buffers_indexed.overwrite_indexed.common_blend_func_buffer_blend_func
Fixes: c7e1ef7c0c ("panfrost: Advertise ES3.0 on Bifrost")
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/9105>
