A release fence acts as a publication barrier for stores within the current thread to become visible to other threads which might observe the release fence. It does not require the current thread to observe stores performed on other threads. As a result, we can allow store-load and load-load forwarding across a release fence.
We choose to be much more conservative about stores. In theory, nothing prevents us from shifting a store from after a release fence to before it, and then eliminating the preceeding (previously fenced) store. Doing this without actually moving the second store is likely also legal, but we chose to be conservative at this time.
The LangRef indicates only atomic loads and stores are effected by fences. This patch chooses to be far more conservative then that.
This is the GVN companion to http://reviews.llvm.org/D11434 which applied the same logic in EarlyCSE and has been baking in tree for a while now.
Differential Revision: http://reviews.llvm.org/D11436
llvm-svn: 264472
This reserves an MDKind for !llvm.loop, which allows callers to avoid a
string-based lookup. I'm not sure why it was missing.
There should be no functionality change here, just a small compile-time
speedup.
llvm-svn: 264371
We used to only allow SCEVAddRecExpr for pointer expressions in order to
be able to compute the bounds. However this is also trivially possible
for loop-invariant addresses (scUnknown) since then the bounds are the
address itself.
Interestingly, we used allow this for the special case when the
loop-invariant address happens to also be an SCEVAddRecExpr (in an outer
loop).
There are a couple more loops that are vectorized in SPEC after this.
My guess is that the main reason we don't see more because for example a
loop-invariant load is vectorized into a splat vector with several
vector-inserts. This is likely to make the vectorization unprofitable.
I.e. we don't notice that a later LICM will move all of this out of the
loop so the cost estimate should really be 0.
llvm-svn: 264243
Summary:
This changes the conversion functions from SCEV * to SCEVAddRecExpr from
ScalarEvolution and PredicatedScalarEvolution to return a SCEVAddRecExpr*
instead of a SCEV* (which removes the need of most clients to do a
dyn_cast right after calling these functions).
We also don't add new predicates if the transformation was not successful.
This is not entirely a NFC (as it can theoretically remove some predicates
from LAA when we have an unknown dependece), but I couldn't find an obvious
regression test for it.
Reviewers: sanjoy
Subscribers: sanjoy, mzolotukhin, llvm-commits
Differential Revision: http://reviews.llvm.org/D18368
llvm-svn: 264161
Summary:
replaceCongruentIVs can break LCSSA when trying to replace IV increments
since it tries to replace all uses of a phi node with another phi node
while both of the phi nodes are not necessarily in the processed loop.
This will cause an assert in IndVars.
To fix this, we add a check to make sure that the replacement maintains
LCSSA.
Reviewers: sanjoy
Subscribers: mzolotukhin, llvm-commits
Differential Revision: http://reviews.llvm.org/D18266
llvm-svn: 263941
Summary:
It can hurt performance to prefetch ahead too much. Be conservative for
now and don't prefetch ahead more than 3 iterations on Cyclone.
Reviewers: hfinkel
Subscribers: llvm-commits, mzolotukhin
Differential Revision: http://reviews.llvm.org/D17949
llvm-svn: 263772
Summary:
And use this TTI for Cyclone. As it was explained in the original RFC
(http://thread.gmane.org/gmane.comp.compilers.llvm.devel/92758), the HW
prefetcher work up to 2KB strides.
I am also adding tests for this and the previous change (D17943):
* Cyclone prefetching accesses with a large stride
* Cyclone not prefetching accesses with a small stride
* Generic Aarch64 subtarget not prefetching either
Reviewers: hfinkel
Subscribers: aemerson, rengolin, llvm-commits, mzolotukhin
Differential Revision: http://reviews.llvm.org/D17945
llvm-svn: 263771
Summary:
This form was replaced by a form taking an instruction instead of opcode and
return type in r258391. After committing this change (and some depending,
follow-up changes) it turned out in the review thread to be controversial. The
discussion didn't come to a conclusion yet. I'm re-adding the old form to fix
the API regression and to provide a better base for discussion, possibly on
llvm-dev.
A difference to the original function is that it can't be called with GEPs
(similarly to how it was already the case for compares). In order to support
opaque pointers in the future, folding GEPs needs to be passed the source
element type, which is not possible with the current API.
Reviewers: dberlin, reames
Subscribers: dblaikie, eddyb
Differential Revision: http://reviews.llvm.org/D17901
llvm-svn: 263501
This doesn't change how many times we construct domtrees in the normal
pipeline, and it removes fragility and instability where basic-aa may
not be run in time to see domtrees because they happen to be constructed
afterward.
This isn't quite as clean as the change to memdep because there is
a mode where basic-aa specifically runs without domtrees -- in the
hacking version used by function-attrs with the legacy pass manager.
llvm-svn: 263234
This doesn't cause us to construct dominator trees any more often in the
normal pipeline, and removes an entire mode of memdep that needed to be
reasoned about and maintained. Perhaps more importantly, it removes the
ability for the results of memdep to be different because of accidental
pass scheduling goofs or the order of evaluation of 'getResult' calls.
Essentially, 'getCachedResult', unless across IR-unit boundaries, is
extremely dangerous. We need to work much harder to avoid it (or its
analog in the old pass manager).
llvm-svn: 263232
This was originally a pointer to support pass managers which didn't use
AnalysisManagers. However, that doesn't realistically come up much and
the complexity of supporting it doesn't really make sense.
In fact, *many* parts of the pass manager were just assuming the pointer
was never null already. This at least makes it much more explicit and
clear.
llvm-svn: 263219
work in the face of the limitations of DLLs and templated static
variables.
This requires passes that use the AnalysisBase mixin provide a static
variable themselves. So as to keep their APIs clean, I've made these
private and befriended the CRTP base class (which is the common
practice).
I've added documentation to AnalysisBase for why this is necessary and
at what point we can go back to the much simpler system.
This is clearly a better pattern than the extern template as it caught
*numerous* places where the template magic hadn't been applied and
things were "just working" but would eventually have broken
mysteriously.
llvm-svn: 263216
actually finish wiring up the old call graph.
There were bugs in the old call graph that hadn't been caught because it
wasn't being tested. It wasn't being tested because it wasn't in the
pipeline system and we didn't have a printing pass to run in tests. This
fixes all of that.
As for why I'm still keeping the old call graph alive its so that I can
port GlobalsAA to the new pass manager with out forking it to work with
the lazy call graph. That's clearly the right eventual design, but it
seems pragmatic to defer that until its necessary. The old call graph
works just fine for GlobalsAA.
llvm-svn: 263104
location in the opt tool to live along side the analysis in LLVM's
libraries.
No functionality changed here, but this will allow me to port the
printer to the new pass manager as well.
llvm-svn: 263101
There is another pass by the generic name 'CallGraphPrinter' which is
actually just a call graph printer tucked away inside the opt tool. I'd
like to bring it out and make it follow the same patterns as the rest of
the CallGraph code, but doing so would end up conflicting with the name
of the DOT printing pass. So this makes the DOT printing pass name be
more precise.
No functionality changed here.
llvm-svn: 263100
This is a fairly straightforward port to the new pass manager with one
exception. It removes a very questionable use of releaseMemory() in
the old pass to invalidate its caches between runs on a function.
I don't think this is really guaranteed to be safe. I've just used the
more direct port to the new PM to address this by nuking the results
object each time the pass runs. While this could cause some minor malloc
traffic increase, I don't expect the compile time performance hit to be
noticable, and it makes the correctness and other aspects of the pass
much easier to reason about. In some cases, it may make things faster by
making the sets and maps smaller with better locality. Indeed, the
measurements collected by Bruno (thanks!!!) show mostly compile time
improvements.
There is sadly very limited testing at this point as there are only two
tests of memdep, and both rely on GVN. I'll be porting GVN next and that
will exercise this heavily though.
Differential Revision: http://reviews.llvm.org/D17962
llvm-svn: 263082
MemoryDependenceAnalysis had a hard-coded exception to the general aliasing rules for malloc and calloc. The reasoning that applied there is equally valid in BasicAA and clarifies the remaining logic in MDA.
In principal, this can expose slightly more optimization opportunities, but since essentially all of our aliasing aware memory optimization passes go through MDA, this will likely be NFC in practice.
Differential Revision: http://reviews.llvm.org/D15912
llvm-svn: 263075
Building on the previous change, this generalizes
ScalarEvolution::getRangeViaFactoring to work with
{Ext(C?A:B)+k0,+,Ext(C?A:B)+k1} where Ext can be a zero extend, sign
extend or truncate operation, and k0 and k1 are constants.
llvm-svn: 262979
This change generalizes ScalarEvolution::getRangeViaFactoring to work
with {Ext(C?A:B),+,Ext(C?A:B)} where Ext can be a zero extend, sign
extend or truncate operation.
llvm-svn: 262978
This is much more clear and less surprising IMO. It also makes things
more consistent with the increasingly large chunk of LLVM code that
assumes true-on-success.
llvm-svn: 262826
duplicated comments.
In several cases these had diverged making them especially nice to
canonicalize. I checked to make sure we weren't losing important
information of course.
llvm-svn: 262825
the new pass manager.
The port will involve substantial edits here, and would likely introduce
bad formatting if formatted in isolation, so just get all the formatting
up to snuff. I'll also go through and try to freshen the doxygen here as
well as modernizing some of the code.
llvm-svn: 262821
The diff is relatively large since I took a chance to rearrange the code I had to touch in a more obvious way, but the key bit is merely using the !range metadata when we can't analyze the instruction further. The previous !range metadata code was essentially just dead since no binary operator or cast will have !range metadata (per Verifier) and it was otherwise dropped on the floor.
llvm-svn: 262751
This experiment was originally about trying to use facts implied dominating conditions to infer more precise known bits. While the compile time was found to be acceptable on several large code bases, we never found sufficiently profitable examples to justify turning on the code by default. Given this, it's time to abandon the experiment.
Several folks have commented that they've found this useful for experimentation, but nothing has come of those experiments. Given how easy the patch is to apply, there's no reason to leave the code in tree.
For anyone interested in further investigation in this area, I recommend finding the summary email I sent on one of the original review threads. In particular, I now believe the use-list based approach is strictly worse than the dom-tree-walking approach.
llvm-svn: 262646
After r262438 we can have provably positive NSW SCEV expressions whose
zero extensions cannot be simplified (since r262438 makes SCEV better at
computing constant ranges). This means demoting sexts of positive add
recurrences eagerly can result in an unsimplified zero extension where
we could have had a simplified sign extension. This change fixes the
issue by teaching SCEV to demote sext of a positive SCEV expression to a
zext only if the sext could not be simplified.
llvm-svn: 262638
This patch provides the following infrastructure for PGO enhancements in inliner:
Enable the use of block level profile information in inliner
Incremental update of block frequency information during inlining
Update the function entry counts of callees when they get inlined into callers.
Differential Revision: http://reviews.llvm.org/D16381
llvm-svn: 262636
parts of the AA interface out of the base class of every single AA
result object.
Because this logic reformulates the query in terms of some other aspect
of the API, it would easily cause O(n^2) query patterns in alias
analysis. These could in turn be magnified further based on the number
of call arguments, and then further based on the number of AA queries
made for a particular call. This ended up causing problems for Rust that
were actually noticable enough to get a bug (PR26564) and probably other
places as well.
When originally re-working the AA infrastructure, the desire was to
regularize the pattern of refinement without losing any generality.
While I think it was successful, that is clearly proving to be too
costly. And the cost is needless: we gain no actual improvement for this
generality of making a direct query to tbaa actually be able to
re-use some other alias analysis's refinement logic for one of the other
APIs, or some such. In short, this is entirely wasted work.
To the extent possible, delegation to other API surfaces should be done
at the aggregation layer so that we can avoid re-walking the
aggregation. In fact, this significantly simplifies the logic as we no
longer need to smuggle the aggregation layer into each alias analysis
(or the TargetLibraryInfo into each alias analysis just so we can form
argument memory locations!).
However, we also have some delegation logic inside of BasicAA and some
of it even makes sense. When the delegation logic is baking in specific
knowledge of aliasing properties of the LLVM IR, as opposed to simply
reformulating the query to utilize a different alias analysis interface
entry point, it makes a lot of sense to restrict that logic to
a different layer such as BasicAA. So one aspect of the delegation that
was in every AA base class is that when we don't have operand bundles,
we re-use function AA results as a fallback for callsite alias results.
This relies on the IR properties of calls and functions w.r.t. aliasing,
and so seems a better fit to BasicAA. I've lifted the logic up to that
point where it seems to be a natural fit. This still does a bit of
redundant work (we query function attributes twice, once via the
callsite and once via the function AA query) but it is *exactly* twice
here, no more.
The end result is that all of the delegation logic is hoisted out of the
base class and into either the aggregation layer when it is a pure
retargeting to a different API surface, or into BasicAA when it relies
on the IR's aliasing properties. This should fix the quadratic query
pattern reported in PR26564, although I don't have a stand-alone test
case to reproduce it.
It also seems general goodness. Now the numerous AAs that don't need
target library info don't carry it around and depend on it. I think
I can even rip out the general access to the aggregation layer and only
expose that in BasicAA as it is the only place where we re-query in that
manner.
However, this is a non-trivial change to the AA infrastructure so I want
to get some additional eyes on this before it lands. Sadly, it can't
wait long because we should really cherry pick this into 3.8 if we're
going to go this route.
Differential Revision: http://reviews.llvm.org/D17329
llvm-svn: 262490
For some reason MSVC seems to think I'm calling getConstant() from a
static context. Try to avoid this issue by explicitly specifying
'this->' (though I'm not confident that this will actually work).
llvm-svn: 262451
Have ScalarEvolution::getRange re-consider cases like "{C?A:B,+,C?P:Q}"
by factoring out "C" and computing RangeOf{A,+,P} union RangeOf({B,+,Q})
instead.
The latter can be easier to compute precisely in cases like
"{C?0:N,+,C?1:-1}" N is the backedge taken count of the loop; since in
such cases the latter form simplifies to [0,N+1) union [0,N+1).
llvm-svn: 262438
manager proxies and use those rather than repeating their definition
four times.
There are real differences between the two directions: outer AMs are
const and don't need to have invalidation tracked. But every proxy in
a particular direction is identical except for the analysis manager type
and the IR unit they proxy into. This makes them prime candidates for
nice templates.
I've started introducing explicit template instantiation declarations
and definitions as well because we really shouldn't be emitting all this
everywhere. I'm going to go back and add the same for the other
templates like this in a follow-up patch.
I've left the analysis manager as an opaque type rather than using two
IR units and requiring it to be an AnalysisManager template
specialization. I think its important that users retain the ability to
provide their own custom analysis management layer and provided it has
the appropriate API everything should Just Work.
llvm-svn: 262127
Most of this is fairly straight forward. Add handling for min/max via existing matcher utility and ConstantRange routines. Add handling for clamp by exploiting condition constraints on inputs.
Note that I'm only handling two constant ranges at this point. It would be reasonable to consider treating overdefined as a full range if the instruction is typed as an integer, but that should be a separate change.
Differential Revision: http://reviews.llvm.org/D17184
llvm-svn: 262085
analyses in the new pass manager.
These just handle really basic stuff: turning a type name into a string
statically that is nice to print in logs, and getting a static unique ID
for each analysis.
Sadly, the format of passes in anonymous namespaces makes using their
names in tests really annoying so I've customized the names of the no-op
passes to keep tests sane to read.
This is the first of a few simplifying refactorings for the new pass
manager that should reduce boilerplate and confusion.
llvm-svn: 262004
Summary: Check that we're using SCEV for the same loop we're simulating. Otherwise, we might try to use the iteration number of the current loop in SCEV expressions for inner/outer loops IVs, which is clearly incorrect.
Reviewers: chandlerc, hfinkel
Subscribers: sanjoy, llvm-commits, mzolotukhin
Differential Revision: http://reviews.llvm.org/D17632
llvm-svn: 261958
This creates the new-style LoopPassManager and wires it up with dummy
and print passes.
This version doesn't support modifying the loop nest at all. It will
be far easier to discuss and evaluate the approaches to that with this
in place so that the boilerplate is out of the way.
llvm-svn: 261831
This is a part of the refactoring to unify isSafeToLoadUnconditionally and isDereferenceablePointer functions. In subsequent change I'm going to eliminate isDerferenceableAndAlignedPointer from Loads API, leaving isSafeToLoadSpecualtively the only function to check is load instruction can be speculated.
Reviewed By: hfinkel
Differential Revision: http://reviews.llvm.org/D16180
llvm-svn: 261736
pattern that triggers it. This essentially requires an immutable
function analysis, as that will survive anything we do to invalidate it.
When we have such patterns, the function analysis manager will not get
cleared between runs of the proxy.
If we actually need an assert about how things are queried, we can add
more elaborate machinery for computing it, but so far I'm not aware of
significant value provided.
Thanks to Justin Lebar for noticing this when he made a (seemingly
innocuous) change to FunctionAttrs that is enough to trigger it in one
test there. Now it is covered by a direct test of the pass manager code.
llvm-svn: 261627
These are really handles that ensure the analyses get cleared at
appropriate places, and as such copying doesn't really make sense.
Instead, they should look more like unique ownership objects. Make that
the case.
Relatedly, if you create a temporary of one and move out of it
its destructor shouldn't actually clear anything. I don't think there is
any code that can trigger this currently, but it seems like a more
robust implementation.
If folks want, I can add a unittest that forces this to be exercised,
but that seems somewhat pointless -- whether a temporary is ever created
in the innards of AnalysisManager is not really something we should be
adding a reliance on, but I didn't want to leave a timebomb in the code
here.
If anyone has a cleaner way to represent this, I'm all ears, but
I wanted to assure myself that this wasn't in fact responsible for
another bug I'm chasing down (it wasn't) and figured I'd commit that.
llvm-svn: 261594
Rename makeNoWrapRegion to a more obvious makeGuaranteedNoWrapRegion,
and add a comment about the counter-intuitive aspects of the function.
This is to help prevent cases like PR26628.
llvm-svn: 261532
I missed == and != when I removed implicit conversions between iterators
and pointers in r252380 since they were defined outside ilist_iterator.
Since they depend on getNodePtrUnchecked(), they indirectly rely on UB.
This commit removes all uses of these operators. (I'll delete the
operators themselves in a separate commit so that it can be easily
reverted if necessary.)
There should be NFC here.
llvm-svn: 261498
Before this patch simplified SCEV expressions for PHI nodes were only returned
the very first time getSCEV() was called, but later calls to getSCEV always
returned the non-simplified value, which had "temporarily" been stored in the
ValueExprMap, but was never removed and consequently blocked the caching of the
simplified PHI expression.
llvm-svn: 261485
the algorithm easily degrades into quadratic memory and time complexity.
The easiest example is a long chain of BBs that don't otherwise use a
location. The caching will add an entry for every intermediate block and
limiting the number of results doesn't help as no results are produced
until a definition is found.
Introduce a limit similar to the existing instructions-per-block limit.
This limit counts the total number of blocks checked. If the limit is
reached, entries are considered unknown. The initial value is 1000,
which avoids regressions for normal sized functions while still
limiting edge cases to reasnable memory consumption and execution time.
Differential Revision: http://reviews.llvm.org/D16123
llvm-svn: 261430
No functional change intended. Copying small (<= 64 bits) APInts isn't
expensive but bloats code by generating the slow path everywhere. Moving
doesn't care about the size of the value.
llvm-svn: 261426
it to actually test the new pass manager AA wiring.
This patch was extracted from the (somewhat too large) D12357 and
rebosed on top of the slightly different design of the new pass manager
AA wiring that I just landed. With this we can start testing the AA in
a thorough way with the new pass manager.
Some minor cleanups to the code in the pass was necessitated here, but
otherwise it is a very minimal change.
Differential Revision: http://reviews.llvm.org/D17372
llvm-svn: 261403
These atomic operations are conceptually both a load and store from the same location. As such, we can treat them as the most conservative of those two components which in practice, means we can treat them like stores. An cmpxchg or atomicrmw captures the values, but not the locations accessed.
Note: We can probably be more aggressive about the comparison value in an cmpxhg since to have it be in memory, it must already be captured, but I figured it was better to avoid that for the moment.
Note 2: It turns out that since we don't actually support cmpxchg of pointer type, writing a negative test is impossible.
Differential Revision: http://reviews.llvm.org/D17400
llvm-svn: 261245
reference-edge SCCs.
This essentially builds a more normal call graph as a subgraph of the
"reference graph" that was the old model. This allows both to exist and
the different use cases to use the aspect which addresses their needs.
Specifically, the pass manager and other *ordering* constrained logic
can use the reference graph to achieve conservative order of visit,
while analyses reasoning about attributes and other properties derived
from reachability can reason about the direct call graph.
Note that this isn't necessarily complete: it doesn't model edges to
declarations or indirect calls. Those can be found by scanning the
instructions of the function if desirable, and in fact every user
currently does this in order to handle things like calls to instrinsics.
If useful, we could consider caching this information in the call graph
to save the instruction scans, but currently that doesn't seem to be
important.
An important realization for why the representation chosen here works is
that the call graph is a formal subset of the reference graph and thus
both can live within the same data structure. All SCCs of the call graph
are necessarily contained within an SCC of the reference graph, etc.
The design is to build 'RefSCC's to model SCCs of the reference graph,
and then within them more literal SCCs for the call graph.
The formation of actual call edge SCCs is not done lazily, unlike
reference edge 'RefSCC's. Instead, once a reference SCC is formed, it
directly builds the call SCCs within it and stores them in a post-order
sequence. This is used to provide a consistent platform for mutation and
update of the graph. The post-order also allows for very efficient
updates in common cases by bounding the number of nodes (and thus edges)
considered.
There is considerable common code that I'm still looking for the best
way to factor out between the various DFS implementations here. So far,
my attempts have made the code harder to read and understand despite
reducing the duplication, which seems a poor tradeoff. I've not given up
on figuring out the right way to do this, but I wanted to wait until
I at least had the system working and tested to continue attempting to
factor it differently.
This also requires introducing several new algorithms in order to handle
all of the incremental update scenarios for the more complex structure
involving two edge colorings. I've tried to comment the algorithms
sufficiently to make it clear how this is expected to work, but they may
still need more extensive documentation.
I know that there are some changes which are not strictly necessarily
coupled here. The process of developing this started out with a very
focused set of changes for the new structure of the graph and
algorithms, but subsequent changes to bring the APIs and code into
consistent and understandable patterns also ended up touching on other
aspects. There was no good way to separate these out without causing
*massive* merge conflicts. Ultimately, to a large degree this is
a rewrite of most of the core algorithms in the LCG class and so I don't
think it really matters much.
Many thanks to the careful review by Sanjoy Das!
Differential Revision: http://reviews.llvm.org/D16802
llvm-svn: 261040
The root issue appears to be a confusion around what makeNoWrapRegion actually does. It seems likely we need two versions of this function with slightly different semantics.
llvm-svn: 260981
Summary:
Extending findExistingExpansion can use existing value in ExprValueMap.
This patch gives 0.3~0.5% performance improvements on
benchmarks(test-suite, spec2000, spec2006, commercial benchmark)
Reviewers: mzolotukhin, sanjoy, zzheng
Differential Revision: http://reviews.llvm.org/D15559
llvm-svn: 260938
into the new pass manager and fix the latent bugs there.
This lets everything live together nicely, but it isn't really useful
yet. I never finished wiring the AA layer up for the new pass manager,
and so subsequent patches will change this to do that wiring and get AA
stuff more fully integrated into the new pass manager. Turns out this is
necessary even to get functionattrs ported over. =]
llvm-svn: 260836
Sanjoy Das