The basic problem we have is that we're trying to reuse an instruction which is mapped to some SCEV. Since we can have multiple such instructions (potentially with different flags), this is analogous to our need to drop flags when performing CSE. A trivial implementation would simply drop flags on any instruction we decided to reuse, and that would be correct.
This patch is almost that trivial patch except that we preserve flags on the reused instruction when existing users would imply UB on overflow already. Adding new users can, at most, refine this program to one which doesn't execute UB which is valid.
In practice, this fixes two conceptual problems with the previous code: 1) a binop could have been canonicalized into a form with different opcode or operands, or 2) the inbounds GEP case which was simply unhandled.
On the test changes, most are pretty straight forward. We loose some flags (in some cases, they'd have been dropped on the next CSE pass anyways). The one that took me the longest to understand was the ashr-expansion test. What's happening there is that we're considering reuse of the mul, previously we disallowed it entirely, now we allow it with no flags. The surrounding diffs are all effects of generating the same mul with a different operand order, and then doing simple DCE.
The loss of the inbounds is unfortunate, but even there, we can recover most of those once we actually treat branch-on-poison as immediate UB.
Differential Revision: https://reviews.llvm.org/D112734
This change should be NFC. It's posted for review mostly to make sure others are happy with the names I'm introducing for "exact full unroll" and "bounded full unroll". The motivation here is that our cost model for bounded unrolling is too aggressive - it gives benefits for exits we aren't going to prune - but I also just think the new version of the code is a lot easier to follow.
Differential Revision: https://reviews.llvm.org/D114453
or (mul X, Y), X --> mul X, (add Y, 1) (when the multiply has no common bits with X)
We already have this fold if the pattern ends in 'add', but we can miss it if the
'add' becomes 'or' via another no-common-bits transform.
This is part of fixing:
http://llvm.org/PR49055
...but it won't make a difference on that example yet.
https://alive2.llvm.org/ce/z/Vrmoeb
Differential Revision: https://reviews.llvm.org/D114729
This solves a problem with non-deterministic output from opt due
to not performing dominator tree updates in a deterministic order.
The problem that was analysed indicated that JumpThreading was using
the DomTreeUpdater via llvm::MergeBasicBlockIntoOnlyPred. When
preparing the list of updates to send to DomTreeUpdater::applyUpdates
we iterated over a SmallPtrSet, which didn't give a well-defined
order of updates to perform.
The added domtree-updates.ll test case is an example that would
result in non-deterministic printouts of the domtree. Semantically
those domtree:s are equivalent, but it show the fact that when we
use the domtree iterator the order in which nodes are visited depend
on the order in which dominator tree updates are performed.
Since some passes (at least EarlyCSE) are iterating over nodes in the
dominator tree in a similar fashion as the domtree printer, then the
order in which transforms are applied by such passes, transitively,
also depend on the order in which dominator tree updates are
performed. And taking EarlyCSE as an example the end result could be
different depending on in which order the transforms are applied.
Reviewed By: nikic, kuhar
Differential Revision: https://reviews.llvm.org/D110292
The code in widenInstruction has already been transitioned to
only rely on information provided by VPWidenRecipe directly.
Moving the code directly to VPWidenRecipe::execute completes
the transition for the recipe.
It provides the following advantages:
1. Less indirection, easier to see what's going on.
2. Removes accesses to fields of ILV.
2) in particular ensures that no dependencies on
fields in ILV for vector code generation are re-introduced.
Reviewed By: Ayal
Differential Revision: https://reviews.llvm.org/D114322
I'm not sure whether this can cause any actual non-determinism,
but at least it makes the DSE debug log non-deterministic, which
makes it harder to debug other non-determinism issues.
The code in widenGEP has already been transitioned to only rely on
information provided by VPWidenGEPRecipe directly.
Moving the code directly to VPWidenGEPRecipe::execute completes
the transition for the recipe.
It provides the following advantages:
1. Less indirection, easier to see what's going on.
2. Removes accesses to fields of ILV.
2) in particular ensures that no dependencies on
fields in ILV for GEP code generation are re-introduced.
Reviewed By: Ayal
Differential Revision: https://reviews.llvm.org/D114321
This is a follow-on suggested in D112634.
Two folds that were added with that patch are subsumed in the call to
decomposeBitTestICmp, and two other folds are potentially inverted.
The deleted folds were very specialized by instcombine standards
because they were restricted to legal integer types based on the data
layout. This generalizes the canonical form independent of target/types.
This change has a reasonable chance of exposing regressions either in
IR or codegen, but I don't have any evidence for either of those yet.
A spot check of asm across several in-tree targets shows variations
that I expect are mostly neutral.
We have one improvement in an existing IR test that I noted with a
comment. Using mask ops might also make more code match with D114272.
Differential Revision: https://reviews.llvm.org/D114386
collectLoopScalars should only add non-uniform nodes to the list if they
are used by a load/store instruction that is marked as CM_Scalarize.
Before this patch, the LV incorrectly marked pointer induction variables
as 'scalar' when they required to be widened by something else,
such as a compare instruction, and weren't used by a node marked as
'CM_Scalarize'. This case is covered by sve-widen-phi.ll.
This change also allows removing some code where the LV tried to
widen the PHI nodes with a stepvector, even though it was marked as
'scalarAfterVectorization'. Now that this code is more careful about
marking instructions that need widening as 'scalar', this code has
become redundant.
Differential Revision: https://reviews.llvm.org/D114373
This patch extends the code that walks memory defs upwards to find
clobbering accesses to also try to optimize the clobbering defining
access.
We should be able to find set the optimized access of our starting def
(KillingDef), if the following holds:
1. It is the first call of getDomMemoryDef for KillingDef (so Current
== KillingDef->getDefiningAccess().
2. No potentially aliasing defs are skipped.
Then if a (partly) aliasing def is encountered, it can be used as
optimized access for KillingDef. No further optimizations can be
applied to KillingDef.
I'd appreciate a careful look, as the existing documentation is not too
clear on what is expected for optimized accesses.
The motivation for this patch is to use the optimized accesses to cover
more cases of redundant stores as follow-up to D111727.
Reviewed By: nikic
Differential Revision: https://reviews.llvm.org/D112313
Compiler has an analysis for perfect diamond matching but it does not
support nodes with main/alternate opcodes. The problem is that the
scalars themselves are different and might not match directly with other
nodes, but operands and main/alternate opcodes might match and compiler
might reuse some previously emitted vector instructions. Need to include
this analysis in the cost model and actual vector instructions emission
process.
Differential Revision: https://reviews.llvm.org/D114101
In VPRecipeBuilder::handleReplication if we believe the instruction
is predicated we then proceed to create new VP region blocks even
when the load is uniform and only predicated due to tail-folding.
I have updated isPredicatedInst to avoid treating a uniform load as
predicated when tail-folding, which means we can do a single scalar
load and a vector splat of the value.
Tests added here:
Transforms/LoopVectorize/AArch64/tail-fold-uniform-memops.ll
Differential Revision: https://reviews.llvm.org/D112552
This canonicalization breaks the ability to discard checks in some cases.
Add a command line option to disable it. This option is on by default,
so the change is NFC.
See for details:
https://reviews.llvm.org/D112895#3149487
Compiler has an analysis for perfect diamond matching but it does not
support nodes with main/alternate opcodes. The problem is that the
scalars themselves are different and might not match directly with other
nodes, but operands and main/alternate opcodes might match and compiler
might reuse some previously emitted vector instructions. Need to include
this analysis in the cost model and actual vector instructions emission
process.
Differential Revision: https://reviews.llvm.org/D114101
This patch updates the cost model for ordered reductions so that a call
to the llvm.fmuladd intrinsic is modelled as a normal fmul instruction
plus the cost of an ordered fadd reduction.
Differential Revision: https://reviews.llvm.org/D111630
In-loop vector reductions which use the llvm.fmuladd intrinsic involve
the creation of two recipes; a VPReductionRecipe for the fadd and a
VPInstruction for the fmul. If the call to llvm.fmuladd has fast-math flags
these should be propagated through to the fmul instruction, so an
interface setFastMathFlags has been added to the VPInstruction class to
enable this.
Differential Revision: https://reviews.llvm.org/D113125
The benefits of sampling-based PGO crucially depends on the quality of profile
data. This diff implements a flow-based algorithm, called profi, that helps to
overcome the inaccuracies in a profile after it is collected.
Profi is an extended and significantly re-engineered classic MCMF (min-cost
max-flow) approach suggested by Levin, Newman, and Haber [2008, Complementing
missing and inaccurate profiling using a minimum cost circulation algorithm]. It
models profile inference as an optimization problem on a control-flow graph with
the objectives and constraints capturing the desired properties of profile data.
Three important challenges that are being solved by profi:
- "fixing" errors in profiles caused by sampling;
- converting basic block counts to edge frequencies (branch probabilities);
- dealing with "dangling" blocks having no samples in the profile.
The main implementation (and required docs) are in SampleProfileInference.cpp.
The worst-time complexity is quadratic in the number of blocks in a function,
O(|V|^2). However a careful engineering and extensive evaluation shows that
the running time is (slightly) super-linear. In particular, instances with
1000 blocks are solved within 0.1 second.
The algorithm has been extensively tested internally on prod workloads,
significantly improving the quality of generated profile data and providing
speedups in the range from 0% to 5%. For "smaller" benchmarks (SPEC06/17), it
generally improves the performance (with a few outliers) but extra work in
the compiler might be needed to re-tune existing optimization passes relying on
profile counts.
Reviewed By: wenlei, hoy
Differential Revision: https://reviews.llvm.org/D109860
These variables are not out-params, and we immediately return after assigning them. Thus, the assignments are dead and just confusing.
I believe these used to be out-params, but they're not any more.
The benefits of sampling-based PGO crucially depends on the quality of profile
data. This diff implements a flow-based algorithm, called profi, that helps to
overcome the inaccuracies in a profile after it is collected.
Profi is an extended and significantly re-engineered classic MCMF (min-cost
max-flow) approach suggested by Levin, Newman, and Haber [2008, Complementing
missing and inaccurate profiling using a minimum cost circulation algorithm]. It
models profile inference as an optimization problem on a control-flow graph with
the objectives and constraints capturing the desired properties of profile data.
Three important challenges that are being solved by profi:
- "fixing" errors in profiles caused by sampling;
- converting basic block counts to edge frequencies (branch probabilities);
- dealing with "dangling" blocks having no samples in the profile.
The main implementation (and required docs) are in SampleProfileInference.cpp.
The worst-time complexity is quadratic in the number of blocks in a function,
O(|V|^2). However a careful engineering and extensive evaluation shows that
the running time is (slightly) super-linear. In particular, instances with
1000 blocks are solved within 0.1 second.
The algorithm has been extensively tested internally on prod workloads,
significantly improving the quality of generated profile data and providing
speedups in the range from 0% to 5%. For "smaller" benchmarks (SPEC06/17), it
generally improves the performance (with a few outliers) but extra work in
the compiler might be needed to re-tune existing optimization passes relying on
profile counts.
Reviewed By: wenlei, hoy
Differential Revision: https://reviews.llvm.org/D109860
I noticed that adding a seemingly unrelated fold for xor caused
regressions on similar patterns, and this is one of the
underlying causes.
This could also be a variation for code as seen in:
https://llvm.org/PR34047
...although that exact example should be fixed after:
D113035 / c36b7e21bd
The vector test shows that we are actually missing a potential
canonicalization for bitcast-of-sext-of-not or the inverse.
The scalar test shows that even if we had that canonicalization,
it would still be possible to see this pattern due to extra uses.
https://alive2.llvm.org/ce/z/y2BAgi
Add OR_None code to indicate that there is no overwrite. This has no any effect for current uses but will be used in one of the next patches building support for PHI translation.
Reviewed By: fhahn
Differential Revision: https://reviews.llvm.org/D105098
This patch fixes PR52111. The problem is that LV propagates poison-generating flags (`nuw`/`nsw`, `exact`
and `inbounds`) in instructions that contribute to the address computation of widen loads/stores that are
guarded by a condition. It may happen that when the code is vectorized and the control flow within the loop
is linearized, these flags may lead to generating a poison value that is effectively used as the base address
of the widen load/store. The fix drops all the integer poison-generating flags from instructions that
contribute to the address computation of a widen load/store whose original instruction was in a basic block
that needed predication and is not predicated after vectorization.
Reviewed By: fhahn, spatel, nlopes
Differential Revision: https://reviews.llvm.org/D111846
The functionality of hasAnalyzableMemoryWrite() is effectively
subsumed by getLocForWriteEx(), which will return None if the
instruction is not analyzable. The implementations don't match
exactly (e.g. getLocForWriteEx() does not limit non-calls to
stores), but in conjunction with the isRemovable() check, it ends
up being the same.
A first step towards modeling preheader and exit blocks in VPlan as well.
Keeping the vector loop in a region allows for changing the VF as we
traverse region boundaries.
Reviewed By: Ayal
Differential Revision: https://reviews.llvm.org/D113182
This is a generalization/extension of the existing and/or
folds noted with TODO comments. Those have a one-use
constraint that is not necessary.
Potential follow-ups are noted by the TODO comments in
the new function. We can also call this function from
other binop visit* functions, but we need to add tests
first.
This solves:
https://llvm.org/PR52543https://alive2.llvm.org/ce/z/NWuCR5
The if-check above deleted part guarantees that StoreOffset <= LoadOffset
and that StoreOffset + StoreSize >= LoadOffset + LoadSize, and given that
LoadOffset + LoadSize > LoadOffset when LoadSize > 0. Thus, this shows
StoreOffset + StoreSize > LoadOffset is guaranteed given LoadSize > 0,
while it could be meaningless to have a type with nonpositive size, so that
the check could be removed. The values are converted to signed types to
avoid unsigned operation with negative offsets.
Part of revision D100179
Reapply commit c35e8185d8 with fixing problem
reported by mstorsjo
The `llvm.instrprof.increment` intrinsic uses `i32` for the index. We should use this same type for the index into the GEP instructions.
Reviewed By: MaskRay
Differential Revision: https://reviews.llvm.org/D114268
There are still another 2 uses of PointerType::getElementType in
Evaluator when evaluating BitCast's on pointers. BitCast's on pointers
should be removed when opaque ptr is ready, so I just keep them as is.
Reviewed By: aeubanks
Differential Revision: https://reviews.llvm.org/D114131
This implements nearly the same logic as getLocForWriteEx(), and
is only used in one place. In that context, we should also know
that getLocForWriteEx() returns a non-None result. As such,
consolidate everything to use one function.
We might emit functions that are private and never called. The coro
split pass only processes functions that might be called. Remove
intrinsics that we can't generate code for.
rdar://84619859
Differential Revision: https://reviews.llvm.org/D114021
In a CGSCC pass manager, we may visit the same function multiple times
due to SCC mutations. In the inliner pipeline, this results in running
the function simplification pipeline on a function multiple times even
if it hasn't been changed since the last function simplification
pipeline run.
We use a newly introduced analysis to keep track of whether or not a
function has changed since the last time the function simplification
pipeline has run on it. If we see this analysis available for a function
in a CGSCCToFunctionPassAdaptor, we skip running the function passes on
the function. The analysis is queried at the end of the function passes
so that it's available after the first time the function simplification
pipeline runs on a function. This is a per-adaptor option so it doesn't
apply to every adaptor.
The goal of this is to improve compile times. However, currently we
can't turn this on by default at least for the higher optimization
levels since the function simplification pipeline is not robust enough
to be idempotent in many cases, resulting in performance regressions if
we stop running the function simplification pipeline on a function
multiple times. We may be able to turn this on for -O1 in the near
future, but turning this on for higher optimization levels would require
more investment in the function simplification pipeline.
Heavily inspired by D98103.
Example compile time improvements with flag turned on:
https://llvm-compile-time-tracker.com/compare.php?from=998dc4a5d3491d2ae8cbe742d2e13bc1b0cacc5f&to=5c27c913687d3d5559ef3ab42b5a3d513531d61c&stat=instructions
Reviewed By: asbirlea, nikic
Differential Revision: https://reviews.llvm.org/D113947
Tsan pass does 2 optimizations based on presence of calls:
1. Don't emit function entry/exit callbacks if there are no calls
and no memory accesses.
2. Combine read/write of the same variable if there are no
intervening calls.
However, all debug info is represented as CallInst as well
and thus effectively disables these optimizations.
Don't consider debug info calls as calls.
Reviewed By: glider, melver
Differential Revision: https://reviews.llvm.org/D114079
When asking how many parts are required for a scalable vector type
there are occasions when it cannot be computed. For example, <vscale x 1 x i3>
is one such vector for AArch64+SVE because at the moment no matter how we
promote the i3 type we never end up with a legal vector. This means
that getTypeConversion returns TypeScalarizeScalableVector as the
LegalizeKind, and then getTypeLegalizationCost returns an invalid cost.
This then causes BasicTTImpl::getNumberOfParts to dereference an invalid
cost, which triggers an assert. This patch changes getNumberOfParts to
return 0 for such cases, since the definition of getNumberOfParts in
TargetTransformInfo.h states that we can use a return value of 0 to represent
an unknown answer.
Currently, LoopVectorize.cpp is the only place where we need to check for
0 as a return value, because all other instances will not currently
ask for the number of parts for <vscale x 1 x iX> types.
In addition, I have changed the target-independent interface for
getNumberOfParts to return 1 and assume there is a single register
that can fit the type. The loop vectoriser has lots of tests that are
target-independent and they relied upon the 0 value to mean the
answer is known and that we are not scalarising the vector.
I have added tests here that show we correctly return an invalid cost
for VF=vscale x 1 when the loop contains unusual types such as i7:
Transforms/LoopVectorize/AArch64/sve-inductions-unusual-types.ll
Differential Revision: https://reviews.llvm.org/D113772
std::hash returns a 64bit hash code while previously we were using only lower 32 bits which caused hash collision for large workloads.
Reviewed By: wenlei, wlei
Differential Revision: https://reviews.llvm.org/D113688
No need to count the final shuffle cost for the constants, gathering of
the constants is just a constant vector + extra inserts, if required.
Differential Revision: https://reviews.llvm.org/D113770
Need to adjust the types of GEPs indices when building the tree
entries/operands. Otherwise some of the nodes might differ and
vectorizer is unable to correctly find them and count their cost.
Differential Revision: https://reviews.llvm.org/D113792
rGf39978b84f1d3a1da6c32db48f64c8daae64b3ad led to and/or exposed
an issue with IndVarSimplification for a loop where a i32 phi node is
no longer replaced by a widened (i64) phi node, because the SCEVs of a
sign-extend no longer folded the same way. I'm unsure how to properly
explain this because it's all rather complicated, but in short: SCEVs
don't fold as nicely as they used to and this caused a difference.
While investigating this, I found that IndVarSimplify can actually
optimise the case in the way we want to if it chooses the widened IV to
be 'signed' (the i32 IV is both sign and zero-extended). Oddly enough,
there is some level of indeterminism in the way the algorithm works,
it just picks the sign of the 'first' zext/sext user, where the order of
the users-iterator is not guaranteed to be the same on each invocation
of the pass (e.g. shown by first running loop-rotate, which puts the
users in a different order).
While I think the fix is valid in the sense that consistently picking
_any_ order is better than having an nondeterministic order, I can
use a bit of advice from people more familiar in this area of the
code-base.
For example, I'm not sure if this fix is hiding another issue where the
IndVarSimplify pass could actually draw the same conclusions (i.e. that
it only needs an i64 phi node) if it does a bit more work, regardless
of whether it chooses the induction variable to be signed or unsigned.
I'm also not sure if choosing signed is better than unsigned, or whether
that just happens to be beneficial only in this individual case.
Any feedback would be much appreciated!
Reviewed By: reames
Differential Revision: https://reviews.llvm.org/D112573
Previously, any change in any function in an SCC would cause all
analyses for all functions in the SCC to be invalidated. With this
change, we now manually invalidate analyses for functions we modify,
then let the pass manager know that all function analyses should be
preserved since we've already handled function analysis invalidation.
So far this only touches the inliner, argpromotion, function-attrs, and
updateCGAndAnalysisManager(), since they are the most used.
This is part of an effort to investigate running the function
simplification pipeline less on functions we visit multiple times in the
inliner pipeline.
However, this causes major memory regressions especially on larger IR.
To counteract this, turn on the option to eagerly invalidate function
analyses. This invalidates analyses on functions immediately after
they're processed in a module or scc to function adaptor for specific
parts of the pipeline.
Within an SCC, if a pass only modifies one function, other functions in
the SCC do not have their analyses invalidated, so in later function
passes in the SCC pass manager the analyses may still be cached. It is
only after the function passes that the eager invalidation takes effect.
For the default pipelines this makes sense because the inliner pipeline
runs the function simplification pipeline after all other SCC passes
(except CoroSplit which doesn't request any analyses).
Overall this has mostly positive effects on compile time and positive effects on memory usage.
https://llvm-compile-time-tracker.com/compare.php?from=7f627596977624730f9298a1b69883af1555765e&to=39e824e0d3ca8a517502f13032dfa67304841c90&stat=instructionshttps://llvm-compile-time-tracker.com/compare.php?from=7f627596977624730f9298a1b69883af1555765e&to=39e824e0d3ca8a517502f13032dfa67304841c90&stat=max-rss
D113196 shows that we slightly regressed compile times in exchange for
some memory improvements when turning on eager invalidation. D100917
shows that we slightly improved compile times in exchange for major
memory regressions in some cases when invalidating less in SCC passes.
Turning these on at the same time keeps the memory improvements while
keeping compile times neutral/slightly positive.
Reviewed By: asbirlea, nikic
Differential Revision: https://reviews.llvm.org/D113304
This change is mostly about getting rid of some "uninteresting" cases in a follow on deeper heuristic change. If anyone sees actually interesting code differences out of this, please let me know. I'm not expecting this to have much impact at all.
Case 1 - With the single deoptimize non-latch exit, we can't have two exiting blocks sharing an exit block. We can only hit this with a poorly documented debug flag.
Case 2 - Why should we treat epilog cases differently from prolog cases? Or to say it differently, why should starting with a constant control whether a multiple exit loop gets unrolled?
Sorry for the lack of tests here. These are both *exceedingly* narrow cases in practice, and after a while trying, I couldn't come up with a test which did anything "useful" as opposed to simply exercise a random combination of force flags. Note that the legality cases for each are already exercised with force flags.
All of the interesting logic from this routine has been removed, inline the single check into the sole non-assert caller. The assert use has little value with the restructured code and is simply dropped.
A bunch of scalars can be treated as a splat not only if all elements
are the same but also if some of them are undefvalues.
Differential Revision: https://reviews.llvm.org/D113774
This reverts commit c35e8185d8.
mstorsjo reported in the revision thread that one VNCoercion assertion
is violated and seemly in relate to this commit. As per "If a test case
that demonstrates a problem is reported in the commit thread, please
revert and investigate offline", this commit is reverted.
If the vector intrinsic has scalar argument, we currently still create
a tree entry for this argument. This entry is not used, just consumes
resources and increases the cost of the tree.
Differential Revision: https://reviews.llvm.org/D113806
The interface is a convenience function to ask if a block requires
predication when widening, but it's important that there are two
separate concepts to consider:
(A) The block was predicated in the original loop.
(B) The block was unpredicated in the original loop, but requires
predication because of tail folding.
In the case of (B) we know that at least one lane of the vector will
be executed, which means we can implementing a load from a uniform address
with a scalar load + splat (D112552). In the case of predication because
of (A), we cannot do this, because the scalar load itself requires
predication.
The name 'blockNeedsPredication' does not make the distinction between
(A) and (B), hence the reason to rename it.
Reviewed By: david-arm
Differential Revision: https://reviews.llvm.org/D113392
ProfileCount could model invalid values, but a user had no indication
that the getCount method could return bogus data. Optional<ProfileCount>
addresses that, because the user must dereference the optional. In
addition, the patch removes concept duplication.
Differential Revision: https://reviews.llvm.org/D113839
The if-check above deleted part guarantees that StoreOffset <= LoadOffset
and that StoreOffset + StoreSize >= LoadOffset + LoadSize, and given that
LoadOffset + LoadSize > LoadOffset when LoadSize > 0. Thus, this shows
StoreOffset + StoreSize > LoadOffset is guaranteed given LoadSize > 0,
while it could be meaningless to have a type with nonpositive size, so that
the check could be removed.
Part of revision D100179
Reviewed By: nikic
This is one of those wonderful "in theory X doesn't matter, but in practice is does" changes. In this particular case, we shift the IVs inserted by the runtime unroller to clamp iteration count of the loops* from decrementing to incrementing.
Why does this matter? A couple of reasons:
* SCEV doesn't have a native subtract node. Instead, all subtracts (A - B) are represented as A + -1 * B and drops any flags invalidated by such. As a result, SCEV is slightly less good at reasoning about edge cases involving decrementing addrecs than incrementing ones. (You can see this in the inferred flags in some of the test cases.)
* Other parts of the optimizer produce incrementing IVs, and they're common in idiomatic source language. We do have support for reversing IVs, but in general if we produce one of each, the pair will persist surprisingly far through the optimizer before being coalesced. (You can see this looking at nearby phis in the test cases.)
Note that if the hardware prefers decrementing (i.e. zero tested) loops, LSR should convert back immediately before codegen.
* Mostly irrelevant detail: The main loop of the prolog case is handled independently and will simple use the original IV with a changed start value. We could in theory use this scheme for all iteration clamping, but that's a larger and more invasive change.
The unrolling code was previously inserting new cloned blocks at the end of the function. The result of this with typical loop structures is that the new iterations are placed far from the initial iteration.
With unrolling, the general assumption is that the a) the loop is reasonable hot, and b) the first Count-1 copies of the loop are rarely (if ever) loop exiting. As such, placing Count-1 copies out of line is a fairly poor code placement choice. We'd much rather fall through into the hot (non-exiting) path. For code with branch profiles, later layout would fix this, but this may have a positive impact on non-PGO compiled code.
However, the real motivation for this change isn't performance. Its readability and human understanding. Having to jump around long distances in an IR file to trace an unrolled loop structure is error prone and tedious.
Need to fix ther cost estimation for split loads, since we look at the
subregs already, no need to permute them, need just to estimate
subregister insert, if it is smaller than the real register. Also, using
split loads, it might be profitable already to vectorize smaller trees
with gathering of the loads.
Differential Revision: https://reviews.llvm.org/D107188
For the scalar/splat case, this fold is subsumed by
foldLogOpOfMaskedICmps(). However, the conjugated fold for "or"
also supports splats with undef. Make both code paths consistent
by using m_ZeroInt() for the "and" implementation as well.
https://alive2.llvm.org/ce/z/tN63cuhttps://alive2.llvm.org/ce/z/ufB_Ue
When folding and/or of icmps, look through add of a constant and
adjust the icmp range instead. Effectively, this decomposes
X + C1 < C2 style range checks back into a normal range. This allows
us to fold comparisons involving two range checks or one range check
and some other condition. We had a fold for a really specific case
of this (or of range check and eq, and only one one side!) while
this handles it in fully generality.
Differential Revision: https://reviews.llvm.org/D113510
Since there is just a single check for LHS in ~(A | B) & C | ...
transforms and multiple RHS checks inside with more coming I am
removing m_OneUse checks for LHS and adding new checks for RHS.
This is non essential as long as there is total benefit.
In addition (~(A | B) & C) | (~(A | C) & B) --> (B ^ C) & ~A
checks were overly restrictive, it should be good without any
additional checks.
Differential Revision: https://reviews.llvm.org/D113141
A problem was noted in the post-commit review for
c36b7e21bd / D113035 :
If the source type is not integer or integer vector,
then we could crash when trying to ComputeNumSignBits().
Philip Reames