If there is a single use of fmul, and that single use is fadd, it makes
sense to fuse ffma, as we already do. However, if there are multiple
uses, fusing may impede code gen. Consider the source fragment:
a = fmul(x, y)
b = fadd(a, z)
c = fmin(a, t)
d = fmax(b, c)
The fmul has two uses. The current ffma fusing is greedy and will
produce the following "optimized" code.
a = fmul(x, y)
b = ffma(x, y, z)
c = fmin(a, t)
d = fmax(b, c)
Actually, this code is worse! Instead of 1 fmul + 1 fadd, we now have 1
fmul + 1 ffma. In effect, two multiplies (and a fused add) instead of
one multiply and an add. Depending on the ISA, that could impede
scheduling or increase code size. It can also increase register
pressure, extending the live range.
It's tempting to gate on is_used_once, but that would hurt in cases
where we really do fuse everything, e.g.:
a = fmul(x, y)
b = fadd(a, z)
c = fadd(a, t)
For ISAs that fuse ffma, we expect that 2 ffma is faster than 1 fmul + 2
fadd. So what we really want is to fuse ffma iff the fmul will get
deleted. That occurs iff all uses of the fmul are fadd and will
themselves get fused to ffma, leaving fmul to get dead code eliminated.
That's easy to implement with a new NIR search helper, checking that all
uses are fadd.
shader-db results on Mali-G57 [open shader-db + subset of closed]:
total instructions in shared programs: 179491 -> 178991 (-0.28%)
instructions in affected programs: 36862 -> 36362 (-1.36%)
helped: 190
HURT: 27
total cycles in shared programs: 10573.20 -> 10571.75 (-0.01%)
cycles in affected programs: 72.02 -> 70.56 (-2.02%)
helped: 28
HURT: 1
total fma in shared programs: 1590.47 -> 1582.61 (-0.49%)
fma in affected programs: 319.95 -> 312.09 (-2.46%)
helped: 194
HURT: 1
total cvt in shared programs: 812.98 -> 813.03 (<.01%)
cvt in affected programs: 118.53 -> 118.58 (0.04%)
helped: 65
HURT: 81
total quadwords in shared programs: 98968 -> 98840 (-0.13%)
quadwords in affected programs: 2960 -> 2832 (-4.32%)
helped: 20
HURT: 4
total threads in shared programs: 4693 -> 4697 (0.09%)
threads in affected programs: 4 -> 8 (100.00%)
helped: 4
HURT: 0
v2: Update trace checksums for virgl due to numerical differences.
Signed-off-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18814>
The new enum is called nir_selection_control_divergent_always_taken,
and it's almost the same as nir_selection_control_flatten.
The main difference between the two is that "flatten" represents
a choice made by the application but "divergent_always_taken" may
be applied by the compiler stack when it thinks this is beneficial.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Daniel Schürmann <daniel@schuermann.dev>
Reviewed-By: Georg Lehmann <dadschoorse@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/17921>
nir_lower_vars_to_ssa will split temp structs up anyway. This fixes a bug
where mediump wouldn't be propagated to the split vars.
The effect is tiny, I think just shuffling some code scheduling from
optimizing at different places. Affects Natural Selection 2, Serious Sam
3, 3dmark slingshot, and Lego Legacy.
freedreno shader-db:
total instructions in shared programs: 11315637 -> 11315993 (<.01%)
instructions in affected programs: 24861 -> 25217 (1.43%)
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18466>
nir_lower_vars_to_ssa will split temp arrays up anyway. Fixes a bug where
split arrays wouldn't get their precision qualifier.
Helps mostly Android and skia shaders. Also affects Civ5, Witcher 2, and
Borderlands 2.
freedreno shader-db:
total instructions in shared programs: 11319395 -> 11319355 (<.01%)
instructions in affected programs: 65744 -> 65704 (-0.06%)
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18466>
We elminated OOB loads while renaming vars to SSA. However, if the OOB
load only appeared after some other passes had constant folded, there may
be no renaming work to do, at which point we'd leave the OOB load deref
around without renaming it or deleting it. For vc4, this was quite a
surprise and caused a regression when we stop eliminating some OOB
accesses at the GLSL level.
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18466>
It was swept at the end, but it meant that in shaders with lots of copies
available at the start of lots of if statements, you'd blow up memory
usage.
turnip memory consumption on dEQP-VK.ssbo.layout.random.scalar.75 drops
from 1.4GB to 110MB, and runtime from 19s to 17s.
Fixes: #7361
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Acked-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18891>
somehow 64bit lowering creates patterns like
vec1 64 ssa_1 = undefined
ssa_2 = unpack_64_2x32_split_x ssa_1
and then the 64bit value is never otherwise used. for this case, rewriting
the unpack to just be a 32bit undef allows the 64bit undef to be optimized
out, avoiding spec violations
fixes#6945
SoroushIMG <soroush.kashani@imgtec.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18728>
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>
Changes:
- nir_metadata_preserve(..., nir_metadata_block_index | nir_metadata_dominance)
is called only when pass makes progress
- nir_metadata_preserve(..., nir_metadata_all) is called when pass doesn't
make progress
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>
Changes:
- nir_metadata_preserve(..., nir_metadata_block_index | nir_metadata_dominance)
is called only when pass makes progress
- nir_metadata_preserve(..., nir_metadata_all) is called when pass doesn't
make progress
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>
Changes:
- nir_metadata_preserve(..., nir_metadata_block_index | nir_metadata_dominance)
is called only when pass makes progress
- nir_metadata_preserve(..., nir_metadata_all) is called when pass doesn't
make progress
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>
Changes:
- nir_metadata_preserve(..., nir_metadata_block_index | nir_metadata_dominance)
is called only when pass makes progress
- nir_metadata_preserve(..., nir_metadata_all) is called when pass doesn't
make progress
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>
Changes:
- nir_metadata_preserve(..., nir_metadata_block_index | nir_metadata_dominance)
is called only when pass makes progress
- nir_metadata_preserve(..., nir_metadata_all) is called when pass doesn't
make progress
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>
Changes:
- nir_metadata_preserve(..., nir_metadata_block_index | nir_metadata_dominance)
is called only when pass makes progress
- nir_metadata_preserve(..., nir_metadata_all) is called when pass doesn't
make progress
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>
Changes:
- removal of lower_state (not needed anymore)
- nir_metadata_preserve(..., nir_metadata_block_index | nir_metadata_dominance)
is called only when pass makes progress
- nir_metadata_preserve(..., nir_metadata_all) is called when pass doesn't
make progress
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>
Changes:
- nir_metadata_preserve(..., nir_metadata_all) is called when pass doesn't
make progress
- only metadata of the current function is invalidated (invalidation on
one function was leaking to successive functions because "progress"
was not reset)
Reviewed-by: Alyssa Rosenzweig <alyssa@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/12282>
When deciding on workarounds that apply to GLSLang we should also check
for ShaderC (when using glslang). Add a helper to do that and reuse in
the three places we check for it.
This patch will now consider shaderc for wa_glslang_cs_barrier workaround.
Suggested by Timur Kristóf.
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/18442>
