In GenerateConstantOffsetsImpl, we may generate non canonical Formula
if BaseRegs of that Formula is updated and includes a recurrent expr reg
related with current loop while its ScaledReg is not.
Patched by: mdchen
Reviewed By: qcolombet
Differential Revision: https://reviews.llvm.org/D86939
This was reverted in 503deec218
because it caused gigantic increase (3x) in branch mispredictions
in certain benchmarks on certain CPU's,
see https://reviews.llvm.org/D84108#2227365.
It has since been investigated and here are the results:
https://lists.llvm.org/pipermail/llvm-commits/Week-of-Mon-20200907/827578.html
> It's an amazingly severe regression, but it's also all due to branch
> mispredicts (about 3x without this). The code layout looks ok so there's
> probably something else to deal with. I'm not sure there's anything we can
> reasonably do so we'll just have to take the hit for now and wait for
> another code reorganization to make the branch predictor a bit more happy :)
>
> Thanks for giving us some time to investigate and feel free to recommit
> whenever you'd like.
>
> -eric
So let's just reland this.
Original commit message:
I've been looking at missed vectorizations in one codebase.
One particular thing that stands out is that some of the loops
reach vectorizer in a rather mangled form, with weird PHI's,
and some of the loops aren't even in a rotated form.
After taking a more detailed look, that happened because
the loop's headers were too big by then. It is evident that
SimplifyCFG's common code hoisting transform is at fault there,
because the pattern it handles is precisely the unrotated
loop basic block structure.
Surprizingly, `SimplifyCFGOpt::HoistThenElseCodeToIf()` is enabled
by default, and is always run, unlike it's friend, common code sinking
transform, `SinkCommonCodeFromPredecessors()`, which is not enabled
by default and is only run once very late in the pipeline.
I'm proposing to harmonize this, and disable common code hoisting
until //late// in pipeline. Definition of //late// may vary,
here currently i've picked the same one as for code sinking,
but i suppose we could enable it as soon as right after
loop rotation happens.
Experimentation shows that this does indeed unsurprizingly help,
more loops got rotated, although other issues remain elsewhere.
Now, this undoubtedly seriously shakes phase ordering.
This will undoubtedly be a mixed bag in terms of both compile- and
run- time performance, codesize. Since we no longer aggressively
hoist+deduplicate common code, we don't pay the price of said hoisting
(which wasn't big). That may allow more loops to be rotated,
so we pay that price. That, in turn, that may enable all the transforms
that require canonical (rotated) loop form, including but not limited to
vectorization, so we pay that too. And in general, no deduplication means
more [duplicate] instructions going through the optimizations. But there's still
late hoisting, some of them will be caught late.
As per benchmarks i've run {F12360204}, this is mostly within the noise,
there are some small improvements, some small regressions.
One big regression i saw i fixed in rG8d487668d09fb0e4e54f36207f07c1480ffabbfd, but i'm sure
this will expose many more pre-existing missed optimizations, as usual :S
llvm-compile-time-tracker.com thoughts on this:
http://llvm-compile-time-tracker.com/compare.php?from=e40315d2b4ed1e38962a8f33ff151693ed4ada63&to=c8289c0ecbf235da9fb0e3bc052e3c0d6bff5cf9&stat=instructions
* this does regress compile-time by +0.5% geomean (unsurprizingly)
* size impact varies; for ThinLTO it's actually an improvement
The largest fallout appears to be in GVN's load partial redundancy
elimination, it spends *much* more time in
`MemoryDependenceResults::getNonLocalPointerDependency()`.
Non-local `MemoryDependenceResults` is widely-known to be, uh, costly.
There does not appear to be a proper solution to this issue,
other than silencing the compile-time performance regression
by tuning cut-off thresholds in `MemoryDependenceResults`,
at the cost of potentially regressing run-time performance.
D84609 attempts to move in that direction, but the path is unclear
and is going to take some time.
If we look at stats before/after diffs, some excerpts:
* RawSpeed (the target) {F12360200}
* -14 (-73.68%) loops not rotated due to the header size (yay)
* -272 (-0.67%) `"Number of live out of a loop variables"` - good for vectorizer
* -3937 (-64.19%) common instructions hoisted
* +561 (+0.06%) x86 asm instructions
* -2 basic blocks
* +2418 (+0.11%) IR instructions
* vanilla test-suite + RawSpeed + darktable {F12360201}
* -36396 (-65.29%) common instructions hoisted
* +1676 (+0.02%) x86 asm instructions
* +662 (+0.06%) basic blocks
* +4395 (+0.04%) IR instructions
It is likely to be sub-optimal for when optimizing for code size,
so one might want to change tune pipeline by enabling sinking/hoisting
when optimizing for size.
Reviewed By: mkazantsev
Differential Revision: https://reviews.llvm.org/D84108
This reverts commit 503deec218.
For intrinsics supported by ConstantRange, compute the result range
based on the argument ranges. We do this independently of whether
some or all of the input ranges are full, as we can often still
constrain the result in some way.
Differential Revision: https://reviews.llvm.org/D87183
This patch updates MemCpyOpt to preserve MemorySSA. It uses the
MemoryDef at the insertion point of the builder and inserts the new def
after that def.
In some cases, we just modify a memory instruction. In that case, get
the defining access, then remove the memory access and add a new one.
If the defining access is in a different block, insert a new def at the
beginning of the current block, otherwise after the defining access.
Reviewed By: asbirlea
Differential Revision: https://reviews.llvm.org/D86651
Preserve MemorySSA if it is available before running GVN.
DSE with MemorySSA will run closely after GVN. If GVN and 2 other
passes preserve MemorySSA, DSE can re-use MemorySSA used by LICM
when doing LTO.
Reviewed By: asbirlea
Differential Revision: https://reviews.llvm.org/D86534
This is a followup to 1ccfb52a61, which made a number of changes
including the apparently innocuous reordering of required passes in
MemCpyOptimizer. This however altered the creation order of BasicAA vs
Phi Values analysis, meaning BasicAA did not pick up PhiValues as a
cached result. Instead if we require MemoryDependence first it will
require PhiValuesAnalysis allowing BasicAA to use it for better results.
I don't claim this is an excellent design, but it fixes a nasty little
regressions where a query later in JumpThreading was getting worse
results.
Differential Revision: https://reviews.llvm.org/D87027
Currently IPSCCP (and others like CVP/GVN) blindly propagate pointer
equalities. In certain cases, that leads to dereferenceable pointers
being replaced, as in the example test case.
I think this is not allowed, as it introduces an access of an
un-dereferenceable pointer. Note that the pointer is inbounds, but one
past the last element, so it is valid, but not dereferenceable.
This patch is mostly to highlight the issue and start a discussion.
Currently it only checks for specifically looking
one-past-the-last-element pointers with array typed bases.
This causes the mis-compile outlined in
https://stackoverflow.com/questions/55754313/is-this-gcc-clang-past-one-pointer-comparison-behavior-conforming-or-non-standar
In the test case, if we replace %p with the GEP for the store, we
subsequently determine that the store and the load cannot alias, because
they are to different underlying objects.
Note that Alive2 seems to think that the replacement is valid:
https://alive2.llvm.org/ce/z/2rorhk
Reviewed By: efriedma
Differential Revision: https://reviews.llvm.org/D85332
In IPSCCP when a function is optimized to return undef, it should clear the returned attribute for all its input arguments
and its corresponding call sites.
The bug is exposed when the value of an input argument of the function is assigned to a physical register and
because of the argument having a returned attribute, the value of this physical register will continue to be used
as the function return value right after the call instruction returns, even if the value that this register holds may
be clobbered during the function call. This potentially results in incorrect values being used afterwards.
Reviewed By: jdoerfert, fhahn
Differential Revision: https://reviews.llvm.org/D84220
Summary:
Analyses are preserved in MemCpyOptimizer.
Get analyses before running the pass and store the pointers, instead of
using lambdas and getting them every time on demand.
Reviewers: lenary, deadalnix, mehdi_amini, nikic, efriedma
Subscribers: hiraditya, llvm-commits
Tags: #llvm
Differential Revision: https://reviews.llvm.org/D74494
Loop Idiom Recognize Pass (LIRP) attempts to transform loops with subscripted arrays
into memcpy/memset function calls. In some particular situation, this transformation
introduces negative impacts. For example: https://bugs.llvm.org/show_bug.cgi?id=47300
This patch will enable users to disable a particular part of the transformation, while
he/she can still enjoy the benefit brought about by the rest of LIRP. The default
behavior stays unchanged: no part of LIRP is disabled by default.
Reviewed By: etiotto (Ettore Tiotto)
Differential Revision: https://reviews.llvm.org/D86262
The 1st try was reverted because I missed an assert that
needed softening.
As discussed in D86798 / rG09652721 , we were potentially
returning a different result for whether an Instruction
is commutable depending on if we call the base class or
derived class method.
This requires relaxing asserts in GVN, but that pass
seems to be working otherwise.
NewGVN requires more work because it uses different
code paths for numbering binops and calls.
For an instruction in the basic block BB, SinkingPass enumerates basic blocks
dominated by BB and BB's successors. For each enumerated basic block,
SinkingPass uses `AllUsesDominatedByBlock` to check whether the basic
block dominates all of the instruction's users. This is inefficient.
Use the nearest common dominator of all users to avoid enumerating the
candidate. The nearest common dominator may be in a parent loop which is
not beneficial. In that case, find the ancestors in the dominator tree.
In the case that the instruction has no user, with this change we will
not perform unnecessary move. This causes some amdgpu test changes.
A stage-2 x86-64 clang is a byte identical with this change.
As discussed in D86798 / rG09652721 , we were potentially
returning a different result for whether an Instruction
is commutable depending on if we call the base class or
derived class method.
This requires relaxing an assert in GVN, but that pass
seems to be working otherwise.
NewGVN requires more work because it uses different
code paths for numbering binops and calls.
Handling the new min/max intrinsics is the motivation, but it
turns out that we have a bunch of other intrinsics with this
missing bit of analysis too.
The FP min/max tests show that we are intersecting FMF,
so that part should be safe too.
As noted in https://llvm.org/PR46897 , there is a commutative
property specifier for intrinsics, but no corresponding function
attribute, and so apparently no uses of that bit. We may want to
remove that next.
Follow-up patches should wire up the Instruction::isCommutative()
to this IntrinsicInst specialization. That requires updating
callers to be aware of the more general commutative property
(not just binops).
Differential Revision: https://reviews.llvm.org/D86798
This patch fixes this crash https://gcc.godbolt.org/z/Ps8d1e
And gives SROA the ability to remove assumes if it allows promoting an alloca to register
Without removing assumes when it can't promote to register.
Reviewed By: jdoerfert
Differential Revision: https://reviews.llvm.org/D86570
This changes getDomMemoryDef to check if a Current is a valid
candidate for elimination before checking for reads. Before the change,
we were spending a lot of compile-time in checking for read accesses for
Current that might not even be removable.
This patch flips the logic, so we skip Current if they cannot be
removed before checking all their uses. This is much more efficient in
practice.
It also adds a more aggressive limit for checking partially overlapping
stores. The main problem with overlapping stores is that we do not know
if they will lead to elimination until seeing all of them. This patch
limits adds a new limit for overlapping store candidates, which keeps
the number of modified overlapping stores roughly the same.
This is another substantial compile-time improvement (while also
increasing the number of stores eliminated). Geomean -O3 -0.67%,
ReleaseThinLTO -0.97%.
http://llvm-compile-time-tracker.com/compare.php?from=0a929b6978a068af8ddb02d0d4714a2843dd8ba9&to=2e630629b43f64b60b282e90f0d96082fde2dacc&stat=instructions
Reviewed By: asbirlea
Differential Revision: https://reviews.llvm.org/D86487
For DSE with MemorySSA it is beneficial to manually traverse the
defining access, instead of using a MemorySSA walker, so we can
better control the number of steps together with other limits and
also weed out invalid/unprofitable paths early on.
This patch requires a follow-up patch to be most effective, which I will
share soon after putting this patch up.
This temporarily XFAIL's the limit tests, because we now explore more
MemoryDefs that may not alias/clobber the killing def. This will be
improved/fixed by the follow-up patch.
This patch also renames some `Dom*` variables to `Earlier*`, because the
dominance relation is not really used/important here and potentially
confusing.
This patch allows us to aggressively cut down compile time, geomean
-O3 -0.64%, ReleaseThinLTO -1.65%, at the expense of fewer stores
removed. Subsequent patches will increase the number of removed stores
again, while keeping compile-time in check.
http://llvm-compile-time-tracker.com/compare.php?from=d8e3294118a8c5f3f97688a704d5a05b67646012&to=0a929b6978a068af8ddb02d0d4714a2843dd8ba9&stat=instructions
Reviewed By: asbirlea
Differential Revision: https://reviews.llvm.org/D86486
As discussed in
http://lists.llvm.org/pipermail/llvm-dev/2020-July/143801.html.
Currently no users outside of unit tests.
Replace all instances in tests of -constprop with -instsimplify.
Notable changes in tests:
* vscale.ll - @llvm.sadd.sat.nxv16i8 is evaluated by instsimplify, use a fake intrinsic instead
* InsertElement.ll - insertelement undef is removed by instsimplify in @insertelement_undef
llvm/test/Transforms/ConstProp moved to llvm/test/Transforms/InstSimplify/ConstProp
Reviewed By: lattner, nikic
Differential Revision: https://reviews.llvm.org/D85159
Currently we repeatedly check the same uses for read clobbers in some
cases. We can avoid unnecessary checks by keeping track of the memory
accesses we already found read clobbers for. To do so, we just add
memory access causing read-clobbers to a set. Note that marking all
visited accesses as read-clobbers would be to pessimistic, as that might
include accesses not on any path to the actual read clobber.
If we do not find any read-clobbers, we can add all visited instructions
to another set and use that to skip the same accesses in the next call.
Reviewed By: asbirlea
Differential Revision: https://reviews.llvm.org/D75025
The "takeName" logic at the end of ScalarizerVisitor::finish
could end up renaming global variables when having simplified
and extractelement instruction to simply pick a single vector
element. If the input vector to the extractelement instruction
held pointers to global variables we ended up renaming the global
variable.
The patch make sure we only take the name of the replaced Op when
we have added new instructions that might need a useful name.
Reviewed By: lebedev.ri
Differential Revision: https://reviews.llvm.org/D86472
Using callCapturesBefore potentially improves the precision and the
number of stores we can remove. But in practice, it seems to have very
little impact in terms of stores removed. For example, for
SPEC2000/SPEC2006/MultiSource with -O3 -flto, ~50 more stores are
removed (out of ~26900 stores removed). But in terms of compile-time, it
is very expensive and the patch gives substantial compile-time
improvements: Geomean O3 -0.24%, ReleaseThinLTO -0.47%, ReleaseLTO-g
-0.39%.
http://llvm-compile-time-tracker.com/compare.php?from=612a0bff88ed906c83b82f079d4c49e5fecfb9d0&to=e6c86b96d20d97dd88e903a409bd8d39b6114312&stat=instructions
Avoid computing InvisibleToCallerBefore/AfterRet up front. In most
cases, this information is not really needed. Instead, introduce helper
functions to compute and cache the result on demand.
Notably, this also does not use PointerMayBeCapturedBefore for
isInvisibleToCallerBeforeRet, as it requires the killing MemoryDef as
starting instruction, making the caching ineffective. But it appears the
use of PointerMayBeCapturedBefore has very limited benefits in practice
(e.g. on SPEC2000/SPEC2006/MultiSource there are no binary changes with
-O3 -flto). Refrain from using it for now, to limit-compile-time.
This gives some nice compile-time improvements:
http://llvm-compile-time-tracker.com/compare.php?from=db9345f6810f379a36752dc52caf5230585d0ebd&to=b4d091047e1b8a3d377d200137b79d03aca65663&stat=instructions
Limit elimination of stores at the end of a function to MemoryDefs with
a single underlying object, to save compile time.
In practice, the case with multiple underlying objects seems not very
important in practice. For -O3 -flto on MultiSource/SPEC2000/SPEC2006
this results in a total of 2 more stores being eliminated.
We can always re-visit that in the future.
As disscussed in post-commit review starting with
https://reviews.llvm.org/D84108#2227365
while this appears to be mostly a win overall, especially code-size-wise,
this appears to shake //certain// code pattens in a way that is extremely
unfavorable for performance (+30% runtime regression)
on certain CPU's (i personally can't reproduce).
So until the behaviour is better understood, and a path forward is mapped,
let's back this out for now.
This reverts commit 1d51dc38d8.
Recommit the patch after fixing an issue reported caused by the fact
that re-used values are also added to InsertedValues.
Additional tests have been added in 88818491b9
This reverts the revert commit 38884641f2.
Both AfterPass and AfterPassInvalidated pass instrumentation
callbacks get additional parameter of type PreservedAnalyses.
This patch was created by @fedor.sergeev. I have just slightly
changed it.
Reviewers: fedor.sergeev
Differential Revision: https://reviews.llvm.org/D81555
Relanded since the buildbot issue was unrelated to this commit.
When hoisting simple values out from a loop, and an optsize attribute, a
convergent call, or an invoke instruction hindered the pass from
unswitching the loop, the pass would return an incorrect Modified
status.
This was caught using the check introduced by D80916.
Reviewed By: asbirlea
Differential Revision: https://reviews.llvm.org/D86085
This reverts commit dfd447c220.
After I pushed this commit, llvm-sphinx-docs started failing, due to:
Warning, treated as error:
extension 'recommonmark' has no setup() function;
is it really a Sphinx extension module?
I don't see how this commit may have caused that, but I'm still
reverting it since I don't know how to proceed with that
troubleshooting.
When hoisting simple values out from a loop, and an optsize attribute, a
convergent call, or an invoke instruction hindered the pass from
unswitching the loop, the pass would return an incorrect Modified
status.
This was caught using the check introduced by D80916.
Reviewed By: asbirlea
Differential Revision: https://reviews.llvm.org/D86085
Max Kazantsev