infinite-inlining across multiple runs of the inliner by keeping a tiny
history of internal-to-SCC inlining decisions.
This is still a bit gross, but I don't yet have any fundamentally better
ideas and numerous people are blocked on this to use new PM and ThinLTO
together.
The core of the idea is to detect when we are about to do an inline that
has a chance of re-splitting an SCC which we have split before with
a similar inlining step. That is a critical component in the inlining
forming a cycle and so far detects all of the various cyclic patterns
I can come up with as well as the original real-world test case (which
comes from a ThinLTO build of libunwind).
I've added some tests that I think really demonstrate what is going on
here. They are essentially state machines that march the inliner through
various steps of a cycle and check that we stop when the cycle is closed
and that we actually did do inlining to form that cycle.
A lot of thanks go to Eric Christopher and Sanjoy Das for the help
understanding this issue and improving the test cases.
The biggest "yuck" here is the layering issue -- the CGSCC pass manager
is providing somewhat magical state to the inliner for it to use to make
itself converge. This isn't great, but I don't honestly have a lot of
better ideas yet and at least seems nicely isolated.
I have tested this patch, and it doesn't block *any* inlining on the
entire LLVM test suite and SPEC, so it seems sufficiently narrowly
targeted to the issue at hand.
We have come up with hypothetical issues that this patch doesn't cover,
but so far none of them are practical and we don't have a viable
solution yet that covers the hypothetical stuff, so proceeding here in
the interim. Definitely an area that we will be back and revisiting in
the future.
Differential Revision: https://reviews.llvm.org/D36188
llvm-svn: 309784
functions.
In the prior commit, we provide ordering to the LCG between functions
and library function definitions that they might begin to call through
transformations. But we still would delete these library functions from
the call graph if they became dead during inlining.
While this immediately crashed, it also exposed a loss of information.
We shouldn't remove definitions of library functions that can still
usefully participate in the LCG-powered CGSCC optimization process. If
new call edges are formed, we want to have definitions to be called.
We can still remove these functions if truly dead using global-dce, etc,
but removing them during the CGSCC walk is premature.
This fixes a crash in the new PM when optimizing some unusual libraries
that end up with "internal" lib functions such as the code in the "R"
language's libraries.
llvm-svn: 308417
the invalidation propagation logic from an SCC to a Function.
I wrote the infrastructure to test this but didn't actually use it in
the unit test where it was designed to be used. =[ My bad. Once
I actually added it to the test case I discovered that it also hadn't
been properly implemented, so I've implemented it. The logic in the FAM
proxy for an SCC pass to propagate invalidation follows the same ideas
as the FAM proxy for a Module pass, but the implementation is a bit
different to reflect the fact that it is forwarding just for an SCC.
However, implementing this correctly uncovered a surprising "bug" (it
was conservatively correct but relatively very expensive) in how we
handle invalidation when splitting one SCC into multiple SCCs. We did an
eager invalidation when in reality we should be deferring invaliadtion
for the *current* SCC to the CGSCC pass manager and just invaliating the
newly constructed SCCs. Otherwise we end up invalidating too much too
soon. This was exposed by the inliner test case that I've updated. Now,
we invalidate *just* the split off '(test1_f)' SCC when doing the CG
update, and then the inliner finishes and invalidates the '(test1_g,
test1_h)' SCC's analyses. The first few attempts at fixing this hit
still more bugs, but all of those are covered by existing tests. For
example, the inliner should also preserve the FAM proxy to avoid
unnecesasry invalidation, and this is safe because the CG update
routines it uses handle any necessary adjustments to the FAM proxy.
Finally, the unittests for the CGSCC pass manager needed a bunch of
updates where we weren't correctly preserving the FAM proxy because it
hadn't been fully implemented and failing to preserve it didn't matter.
Note that this doesn't yet fix the current crasher due to MemSSA finding
a stale dominator tree, but without this the fix to that crasher doesn't
really make any sense when testing because it relies on the proxy
behavior.
llvm-svn: 307487
This restores the order of evaluation (& conditionalized evaluation) of
isTriviallyDeadInstruction, InlineHistoryIncludes, and shouldInline
(with the addition of a shouldInline call after
isTriviallyDeadInstruction) from before r305245.
llvm-svn: 305267
Other comments/implications are that this isn't intended behavior (nor
perserved/reimplemented in the new inliner) & complicates fixing the
'inlining' of trivially dead calls without consulting the cost function
first.
llvm-svn: 305052
This change is required because the notion of count is different for
sample profiling and getProfileCount will need to determine the
underlying profile type.
Differential revision: https://reviews.llvm.org/D33012
llvm-svn: 302597
- First time, during calculation of the cost in InlineCost.cpp
- Second time, during calculation of the cost in Inliner.cpp
This patches fixes this.
Differential Revision: https://reviews.llvm.org/D31137
llvm-svn: 298496
entire SCC before iterating on newly-introduced call edges resulting
from any inlined function bodies.
This more closely matches the behavior of the old PM's inliner. While it
wasn't really clear to me initially, this behavior is actually essential
to the inliner behaving reasonably in its current design.
Because the inliner is fundamentally a bottom-up inliner and all of its
cost modeling is designed around that it often runs into trouble within
an SCC where we don't have any meaningful bottom-up ordering to use. In
addition to potentially cyclic, infinite inlining that we block with the
inline history mechanism, it can also take seemingly simple call graph
patterns within an SCC and turn them into *insanely* large functions by
accidentally working top-down across the SCC without any of the
threshold limitations that traditional top-down inliners use.
Consider this diabolical monster.cpp file that Richard Smith came up
with to help demonstrate this issue:
```
template <int N> extern const char *str;
void g(const char *);
template <bool K, int N> void f(bool *B, bool *E) {
if (K)
g(str<N>);
if (B == E)
return;
if (*B)
f<true, N + 1>(B + 1, E);
else
f<false, N + 1>(B + 1, E);
}
template <> void f<false, MAX>(bool *B, bool *E) { return f<false, 0>(B, E); }
template <> void f<true, MAX>(bool *B, bool *E) { return f<true, 0>(B, E); }
extern bool *arr, *end;
void test() { f<false, 0>(arr, end); }
```
When compiled with '-DMAX=N' for various values of N, this will create an SCC
with a reasonably large number of functions. Previously, the inliner would try
to exhaust the inlining candidates in a single function before moving on. This,
unfortunately, turns it into a top-down inliner within the SCC. Because our
thresholds were never built for that, we will incrementally decide that it is
always worth inlining and proceed to flatten the entire SCC into that one
function.
What's worse, we'll then proceed to the next function, and do the exact same
thing except we'll skip the first function, and so on. And at each step, we'll
also make some of the constant factors larger, which is awesome.
The fix in this patch is the obvious one which makes the new PM's inliner use
the same technique used by the old PM: consider all the call edges across the
entire SCC before beginning to process call edges introduced by inlining. The
result of this is essentially to distribute the inlining across the SCC so that
every function incrementally grows toward the inline thresholds rather than
allowing the inliner to grow one of the functions vastly beyond the threshold.
The code for this is a bit awkward, but it works out OK.
We could consider in the future doing something more powerful here such as
prioritized order (via lowest cost and/or profile info) and/or a code-growth
budget per SCC. However, both of those would require really substantial work
both to design the system in a way that wouldn't break really useful
abstraction decomposition properties of the current inliner and to be tuned
across a reasonably diverse set of code and workloads. It also seems really
risky in many ways. I have only found a single real-world file that triggers
the bad behavior here and it is generated code that has a pretty pathological
pattern. I'm not worried about the inliner not doing an *awesome* job here as
long as it does *ok*. On the other hand, the cases that will be tricky to get
right in a prioritized scheme with a budget will be more common and idiomatic
for at least some frontends (C++ and Rust at least). So while these approaches
are still really interesting, I'm not in a huge rush to go after them. Staying
even closer to the existing PM's behavior, especially when this easy to do,
seems like the right short to medium term approach.
I don't really have a test case that makes sense yet... I'll try to find a
variant of the IR produced by the monster template metaprogram that is both
small enough to be sane and large enough to clearly show when we get this wrong
in the future. But I'm not confident this exists. And the behavior change here
*should* be unobservable without snooping on debug logging. So there isn't
really much to test.
The test case updates come from two incidental changes:
1) We now visit functions in an SCC in the opposite order. I don't think there
really is a "right" order here, so I just update the test cases.
2) We no longer compute some analyses when an SCC has no call instructions that
we consider for inlining.
llvm-svn: 297374
Summary:
As written in the comments above, LastCallToStaticBonus is already applied to
the cost if Caller has only one user, so it is redundant to reapply the bonus
here.
If the only user is not a caller, TotalSecondaryCost will not be adjusted
anyway because callerWillBeRemoved is false. If there's no caller at all, we
don't need to care about TotalSecondaryCost because
inliningPreventsSomeOuterInline is false.
Reviewers: chandlerc, eraman
Reviewed By: eraman
Subscribers: haicheng, davidxl, davide, llvm-commits, mehdi_amini
Differential Revision: https://reviews.llvm.org/D29169
llvm-svn: 295075
disturbing the graph or having to update edges.
This is motivated by porting argument promotion to the new pass manager.
Because of how LLVM IR Function objects work, in order to change their
signature a new object needs to be created. This is efficient and
straight forward in the IR but previously was very hard to implement in
LCG. We could easily replace the function a node in the graph
represents. The challenging part is how to handle updating the edges in
the graph.
LCG previously used an edge to a raw function to represent a node that
had not yet been scanned for calls and references. This was the core
of its laziness. However, that model causes this kind of update to be
very hard:
1) The keys to lookup an edge need to be `Function*`s that would all
need to be updated when we update the node.
2) There will be some unknown number of edges that haven't transitioned
from `Function*` edges to `Node*` edges.
All of this complexity isn't necessary. Instead, we can always build
a node around any function, always pointing edges at it and always using
it as the key to lookup an edge. To maintain the laziness, we need to
sink the *edges* of a node into a secondary object and explicitly model
transitioning a node from empty to populated by scanning the function.
This design seems much cleaner in a number of ways, but importantly
there is now exactly *one* place where the `Function*` has to be
updated!
Some other cleanups that fall out of this include having something to
model the *entry* edges more accurately. Rather than hand rolling parts
of the node in the graph itself, we have an explicit `EdgeSequence`
object that gives us exactly the functionality needed. We also have
a consistent place to define the edge iterators and can use them for
both the entry edges and the internal edges of the graph.
The API used to model the separation between a node and its edges is
intentionally very thin as most clients are expected to deal with nodes
that have populated edges. We model this exactly as an optional does
with an additional method to populate the edges when that is
a reasonable thing for a client to do. This is based on API design
suggestions from Richard Smith and David Blaikie, credit goes to them
for helping pick how to model this without it being either too explicit
or too implicit.
The patch is somewhat noisy due to shifting around iterator types and
new syntax for walking the edges of a node, but most of the
functionality change is in the `Edge`, `EdgeSequence`, and `Node` types.
Differential Revision: https://reviews.llvm.org/D29577
llvm-svn: 294653
a lazy-asserting PoisoningVH.
AssertVH is fundamentally incompatible with cache-invalidation of
analysis results. The invaliadtion happens after the AssertingVH has
already fired. Instead, use a PoisoningVH that will assert if the
dangling handle is ever used rather than merely be assigned or
destroyed.
This patch also removes all of the (numerous) doomed attempts to work
around this fundamental incompatibility. It is a pretty significant
simplification IMO.
The most interesting change is in the Inliner where we still do some
clearing because we don't want to rely on the coarse grained
invalidation strategy of the containing pass manager. However, I prefer
the approach that contains this logic to the cleanup phase of the
Inliner, and I think we could enhance the CGSCC analysis management
layer to make this even better in the future if desired.
The rest is straight cleanup.
I've also added a test for one of the harder cases to work around: when
a *module analysis* contains many AssertingVHes pointing at functions.
Differential Revision: https://reviews.llvm.org/D29006
llvm-svn: 292928
clearing its body. This is essential to avoid triggering asserting value
handles in analyses on the function's body.
I'm working on a test case for this behavior in LLVM, but Clang has
a great one that managed to trigger this on all of the bots already.
llvm-svn: 292770
new PM's inliner.
The bug happens when we refine an SCC after having computed a proxy for
the FunctionAnalysisManager, and then proceed to compute fresh analyses
for functions in the *new* SCC using the manager provided by the old
SCC's proxy. *And* when we manage to mutate a function in this new SCC
in a way that invalidates those analyses. This can be... challenging to
reproduce.
I've managed to contrive a set of functions that trigger this and added
a test case, but it is a bit brittle. I've directly checked that the
passes run in the expected ways to help avoid the test just becoming
silently irrelevant.
This gets the new PM back to passing the LLVM test suite after the PGO
improvements landed.
llvm-svn: 292757
This adds the following to the new PM based inliner in PGO mode:
* Use block frequency analysis to derive callsite's profile count and use
that to adjust thresholds of hot and cold callsites.
* Incrementally update the BFI of the caller after a callee gets inlined
into it. This incremental update is only within an invocation of the run
method - BFI is not preserved across calls to run.
Update the function entry count of the callee after inlining it into a
caller.
* I've tuned the thresholds for the hot and cold callsites using a hacked
up version of the old inliner that explicitly computes BFI on a set of
internal benchmarks and spec. Once the new PM based pipeline stabilizes
(IIRC Chandler mentioned there are known issues) I'll benchmark this
again and adjust the thresholds if required.
Inliner PGO support.
Differential revision: https://reviews.llvm.org/D28331
llvm-svn: 292666
when they are call edges at the leaf but may (transitively) be reached
via ref edges.
It turns out there is a simple rule: insert everything as a ref edge
which is a safe conservative default. Then we let the existing update
logic handle promoting some of those to call edges.
Note that it would be fairly cheap to make these call edges right away
if that is desirable by testing whether there is some existing call path
from the source to the target. It just seemed like slightly more
complexity in this code path that isn't strictly necessary. If anyone
feels strongly about handling this differently I'm happy to change it.
llvm-svn: 290649
skipping indirectly recursive inline chains.
To do this, we implicitly build an inline stack for each callsite and
check prior to inlining that doing so would not form a cycle. This uses
the exact same technique and even shares some code with the legacy PM
inliner.
This solution remains deeply unsatisfying to me because it means we
cannot actually iterate the inliner externally. Doing so would not be
able to easily detect and avoid such cycles. Some day I would very much
like to have a solution that works without this internal state to detect
cycles, but this is not that day.
llvm-svn: 290590
removing fully-dead comdats without removing dead entries in comdats
with live members.
This factors the core logic out of the current inliner's internals to
a reusable utility and leverages that in both places. The factored out
code should also be (minorly) more efficient in cases where we have very
few dead functions or dead comdats to consider.
I've added a test case to cover this behavior of the always inliner.
This is the last significant bug in the new PM's always inliner I've
found (so far).
llvm-svn: 290557
This doesn't implement *every* feature of the existing inliner, but
tries to implement the most important ones for building a functional
optimization pipeline and beginning to sort out bugs, regressions, and
other problems.
Notable, but intentional omissions:
- No alloca merging support. Why? Because it isn't clear we want to do
this at all. Active discussion and investigation is going on to remove
it, so for simplicity I omitted it.
- No support for trying to iterate on "internally" devirtualized calls.
Why? Because it adds what I suspect is inappropriate coupling for
little or no benefit. We will have an outer iteration system that
tracks devirtualization including that from function passes and
iterates already. We should improve that rather than approximate it
here.
- Optimization remarks. Why? Purely to make the patch smaller, no other
reason at all.
The last one I'll probably work on almost immediately. But I wanted to
skip it in the initial patch to try to focus the change as much as
possible as there is already a lot of code moving around and both of
these *could* be skipped without really disrupting the core logic.
A summary of the different things happening here:
1) Adding the usual new PM class and rigging.
2) Fixing minor underlying assumptions in the inline cost analysis or
inline logic that don't generally hold in the new PM world.
3) Adding the core pass logic which is in essence a loop over the calls
in the nodes in the call graph. This is a bit duplicated from the old
inliner, but only a handful of lines could realistically be shared.
(I tried at first, and it really didn't help anything.) All told,
this is only about 100 lines of code, and most of that is the
mechanics of wiring up analyses from the new PM world.
4) Updating the LazyCallGraph (in the new PM) based on the *newly
inlined* calls and references. This is very minimal because we cannot
form cycles.
5) When inlining removes the last use of a function, eagerly nuking the
body of the function so that any "one use remaining" inline cost
heuristics are immediately refined, and queuing these functions to be
completely deleted once inlining is complete and the call graph
updated to reflect that they have become dead.
6) After all the inlining for a particular function, updating the
LazyCallGraph and the CGSCC pass manager to reflect the
function-local simplifications that are done immediately and
internally by the inline utilties. These are the exact same
fundamental set of CG updates done by arbitrary function passes.
7) Adding a bunch of test cases to specifically target CGSCC and other
subtle aspects in the new PM world.
Many thanks to the careful review from Easwaran and Sanjoy and others!
Differential Revision: https://reviews.llvm.org/D24226
llvm-svn: 290161
After r289755, the AssumptionCache is no longer needed. Variables affected by
assumptions are now found by using the new operand-bundle-based scheme. This
new scheme is more computationally efficient, and also we need much less
code...
llvm-svn: 289756
The core of the change is supposed to be NFC, however it also fixes
what I believe was an undefined behavior when calling:
va_start(ValueArgs, Desc);
with Desc being a StringRef.
Differential Revision: https://reviews.llvm.org/D25342
llvm-svn: 283671
There is really no reason for these to be separate.
The vectorizer started this pretty bad tradition that the text of the
missed remarks is pretty meaningless, i.e. vectorization failed. There,
you have to query analysis to get the full picture.
I think we should just explain the reason for missing the optimization
in the missed remark when possible. Analysis remarks should provide
information that the pass gathers regardless whether the optimization is
passing or not.
llvm-svn: 282542
(Re-committed after moving the template specialization under the yaml
namespace. GCC was complaining about this.)
This allows various presentation of this data using an external tool.
This was first recommended here[1].
As an example, consider this module:
1 int foo();
2 int bar();
3
4 int baz() {
5 return foo() + bar();
6 }
The inliner generates these missed-optimization remarks today (the
hotness information is pulled from PGO):
remark: /tmp/s.c:5:10: foo will not be inlined into baz (hotness: 30)
remark: /tmp/s.c:5:18: bar will not be inlined into baz (hotness: 30)
Now with -pass-remarks-output=<yaml-file>, we generate this YAML file:
--- !Missed
Pass: inline
Name: NotInlined
DebugLoc: { File: /tmp/s.c, Line: 5, Column: 10 }
Function: baz
Hotness: 30
Args:
- Callee: foo
- String: will not be inlined into
- Caller: baz
...
--- !Missed
Pass: inline
Name: NotInlined
DebugLoc: { File: /tmp/s.c, Line: 5, Column: 18 }
Function: baz
Hotness: 30
Args:
- Callee: bar
- String: will not be inlined into
- Caller: baz
...
This is a summary of the high-level decisions:
* There is a new streaming interface to emit optimization remarks.
E.g. for the inliner remark above:
ORE.emit(DiagnosticInfoOptimizationRemarkMissed(
DEBUG_TYPE, "NotInlined", &I)
<< NV("Callee", Callee) << " will not be inlined into "
<< NV("Caller", CS.getCaller()) << setIsVerbose());
NV stands for named value and allows the YAML client to process a remark
using its name (NotInlined) and the named arguments (Callee and Caller)
without parsing the text of the message.
Subsequent patches will update ORE users to use the new streaming API.
* I am using YAML I/O for writing the YAML file. YAML I/O requires you
to specify reading and writing at once but reading is highly non-trivial
for some of the more complex LLVM types. Since it's not clear that we
(ever) want to use LLVM to parse this YAML file, the code supports and
asserts that we're writing only.
On the other hand, I did experiment that the class hierarchy starting at
DiagnosticInfoOptimizationBase can be mapped back from YAML generated
here (see D24479).
* The YAML stream is stored in the LLVM context.
* In the example, we can probably further specify the IR value used,
i.e. print "Function" rather than "Value".
* As before hotness is computed in the analysis pass instead of
DiganosticInfo. This avoids the layering problem since BFI is in
Analysis while DiagnosticInfo is in IR.
[1] https://reviews.llvm.org/D19678#419445
Differential Revision: https://reviews.llvm.org/D24587
llvm-svn: 282539
This allows various presentation of this data using an external tool.
This was first recommended here[1].
As an example, consider this module:
1 int foo();
2 int bar();
3
4 int baz() {
5 return foo() + bar();
6 }
The inliner generates these missed-optimization remarks today (the
hotness information is pulled from PGO):
remark: /tmp/s.c:5:10: foo will not be inlined into baz (hotness: 30)
remark: /tmp/s.c:5:18: bar will not be inlined into baz (hotness: 30)
Now with -pass-remarks-output=<yaml-file>, we generate this YAML file:
--- !Missed
Pass: inline
Name: NotInlined
DebugLoc: { File: /tmp/s.c, Line: 5, Column: 10 }
Function: baz
Hotness: 30
Args:
- Callee: foo
- String: will not be inlined into
- Caller: baz
...
--- !Missed
Pass: inline
Name: NotInlined
DebugLoc: { File: /tmp/s.c, Line: 5, Column: 18 }
Function: baz
Hotness: 30
Args:
- Callee: bar
- String: will not be inlined into
- Caller: baz
...
This is a summary of the high-level decisions:
* There is a new streaming interface to emit optimization remarks.
E.g. for the inliner remark above:
ORE.emit(DiagnosticInfoOptimizationRemarkMissed(
DEBUG_TYPE, "NotInlined", &I)
<< NV("Callee", Callee) << " will not be inlined into "
<< NV("Caller", CS.getCaller()) << setIsVerbose());
NV stands for named value and allows the YAML client to process a remark
using its name (NotInlined) and the named arguments (Callee and Caller)
without parsing the text of the message.
Subsequent patches will update ORE users to use the new streaming API.
* I am using YAML I/O for writing the YAML file. YAML I/O requires you
to specify reading and writing at once but reading is highly non-trivial
for some of the more complex LLVM types. Since it's not clear that we
(ever) want to use LLVM to parse this YAML file, the code supports and
asserts that we're writing only.
On the other hand, I did experiment that the class hierarchy starting at
DiagnosticInfoOptimizationBase can be mapped back from YAML generated
here (see D24479).
* The YAML stream is stored in the LLVM context.
* In the example, we can probably further specify the IR value used,
i.e. print "Function" rather than "Value".
* As before hotness is computed in the analysis pass instead of
DiganosticInfo. This avoids the layering problem since BFI is in
Analysis while DiagnosticInfo is in IR.
[1] https://reviews.llvm.org/D19678#419445
Differential Revision: https://reviews.llvm.org/D24587
llvm-svn: 282499
Summary:
This is obviously an interesting case because it may motivate code
restructuring or LTO.
Reporting this requires instantiation of ORE in the loop where the call
sites are first gathered. I've checked compile-time
overhead *with* -Rpass-with-hotness and the worst slow-down was 6% in
mcf and quickly tailing off. As before without -Rpass-with-hotness
there is no overhead.
Because this could be a pretty noisy diagnostics, it is currently
qualified as 'verbose'. As of this patch, 'verbose' diagnostics are
only emitted with -Rpass-with-hotness, i.e. when the output is expected
to be filtered.
Reviewers: eraman, chandlerc, davidxl, hfinkel
Subscribers: tejohnson, Prazek, davide, llvm-commits
Differential Revision: https://reviews.llvm.org/D23415
llvm-svn: 279860
This is off for now while testing can take place to make sure that in
fact we do sufficient stack coloring to fully obviate the manual alloca
array merging.
Some context on why we should be using stack coloring rather than
merging allocas in this way:
LLVM relies very heavily on analyzing pointers as coming from different
allocas in order to make aliasing decisions. These are some of the most
powerful aliasing signals available in LLVM. So merging allocas is an
extremely destructive operation on the LLVM IR -- it takes away highly
valuable and hard to reconstruct information.
As a consequence, inlined functions which happen to have array allocas
that this pattern matches will fail to be properly interleaved unless
SROA manages to hoist everything to an SSA register. Instead, the
inliner will have added an unnecessary dependence that one inlined
function execute after the other because they will have been rewritten
to refer to the same memory.
All that said, folks will reasonably want some time to experiment here
and make sure there are no significant regressions. A flag should give
us an easy knob to test.
For more context, see the thread here:
http://lists.llvm.org/pipermail/llvm-dev/2016-July/103277.htmlhttp://lists.llvm.org/pipermail/llvm-dev/2016-August/103285.html
Differential Revision: https://reviews.llvm.org/D23052
llvm-svn: 278892
Summary:
I think it is much better this way.
When I firstly saw line:
Cost += InlineConstants::LastCallToStaticBonus;
I though that this is a bug, because everywhere where the cost is being reduced
it is usuing -=.
Reviewers: eraman, tejohnson, mehdi_amini
Subscribers: llvm-commits, mehdi_amini
Differential Revision: https://reviews.llvm.org/D23222
llvm-svn: 278290
Summary:
The inliner not being a function pass requires the work-around of
generating the OptimizationRemarkEmitter and in turn BFI on demand.
This will go away after the new PM is ready.
BFI is only computed inside ORE if the user has requested hotness
information for optimization diagnostitics (-pass-remark-with-hotness at
the 'opt' level). Thus there is no additional overhead without the
flag.
Reviewers: hfinkel, davidxl, eraman
Subscribers: llvm-commits
Differential Revision: https://reviews.llvm.org/D22694
llvm-svn: 278185
Summary:
copypasta doc of ImportedFunctionsInliningStatistics class
\brief Calculate and dump ThinLTO specific inliner stats.
The main statistics are:
(1) Number of inlined imported functions,
(2) Number of imported functions inlined into importing module (indirect),
(3) Number of non imported functions inlined into importing module
(indirect).
The difference between first and the second is that first stat counts
all performed inlines on imported functions, but the second one only the
functions that have been eventually inlined to a function in the importing
module (by a chain of inlines). Because llvm uses bottom-up inliner, it is
possible to e.g. import function `A`, `B` and then inline `B` to `A`,
and after this `A` might be too big to be inlined into some other function
that calls it. It calculates this statistic by building graph, where
the nodes are functions, and edges are performed inlines and then by marking
the edges starting from not imported function.
If `Verbose` is set to true, then it also dumps statistics
per each inlined function, sorted by the greatest inlines count like
- number of performed inlines
- number of performed inlines to importing module
Reviewers: eraman, tejohnson, mehdi_amini
Subscribers: mehdi_amini, llvm-commits
Differential Revision: https://reviews.llvm.org/D22491
llvm-svn: 277089
This unblocks the new PM part of River's patch in
https://reviews.llvm.org/D22706
Conveniently, this same change was needed for D21921 and so these
changes are just spun out from there.
llvm-svn: 276515
Instead of directly using MaxFunctionCount and function entry count to determine callee hotness, use the isHotFunction/isColdFunction methods provided by ProfileSummaryInfo.
Differential revision: http://reviews.llvm.org/D21045
llvm-svn: 272321
The implemented heuristic has a large body of code which better sits
in its own function for better readability. It also allows adding more
heuristics easier in the future.
llvm-svn: 268107
The original commit was reverted because of a buildbot problem with LazyCallGraph::SCC handling (not related to the OptBisect handling).
Differential Revision: http://reviews.llvm.org/D19172
llvm-svn: 267231
This patch implements a optimization bisect feature, which will allow optimizations to be selectively disabled at compile time in order to track down test failures that are caused by incorrect optimizations.
The bisection is enabled using a new command line option (-opt-bisect-limit). Individual passes that may be skipped call the OptBisect object (via an LLVMContext) to see if they should be skipped based on the bisect limit. A finer level of control (disabling individual transformations) can be managed through an addition OptBisect method, but this is not yet used.
The skip checking in this implementation is based on (and replaces) the skipOptnoneFunction check. Where that check was being called, a new call has been inserted in its place which checks the bisect limit and the optnone attribute. A new function call has been added for module and SCC passes that behaves in a similar way.
Differential Revision: http://reviews.llvm.org/D19172
llvm-svn: 267022
Removed some unused headers, replaced some headers with forward class declarations.
Found using simple scripts like this one:
clear && ack --cpp -l '#include "llvm/ADT/IndexedMap.h"' | xargs grep -L 'IndexedMap[<]' | xargs grep -n --color=auto 'IndexedMap'
Patch by Eugene Kosov <claprix@yandex.ru>
Differential Revision: http://reviews.llvm.org/D19219
From: Mehdi Amini <mehdi.amini@apple.com>
llvm-svn: 266595
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
Summary:
Passes that call `getAnalysisIfAvailable<T>` also need to call
`addUsedIfAvailable<T>` in `getAnalysisUsage` to indicate to the
legacy pass manager that it uses `T`. This contract was being
violated by passes that used `createLegacyPMAAResults`. This change
fixes this by exposing a helper in AliasAnalysis.h,
`addUsedAAAnalyses`, that is complementary to createLegacyPMAAResults
and does the right thing when called from `getAnalysisUsage`.
Reviewers: chandlerc
Subscribers: mcrosier, llvm-commits
Differential Revision: http://reviews.llvm.org/D17010
llvm-svn: 260183
InlineCostAnalysis is an analysis pass without any need for it to be one.
Once it stops being an analysis pass, it doesn't maintain any useful state
and the member functions inside can be made free functions. NFC.
Differential Revision: http://reviews.llvm.org/D15701
llvm-svn: 256521
This reapplies r256277 with two changes:
- In emitFnAttrCompatCheck, change FuncName's type to std::string to fix
a use-after-free bug.
- Remove an unnecessary install-local target in lib/IR/Makefile.
Original commit message for r252949:
Provide a way to specify inliner's attribute compatibility and merging
rules using table-gen. NFC.
This commit adds new classes CompatRule and MergeRule to Attributes.td,
which are used to generate code to check attribute compatibility and
merge attributes of the caller and callee.
rdar://problem/19836465
llvm-svn: 256304
This reapplies r252990 and r252949. I've added member function getKind
to the Attr classes which returns the enum or string of the attribute.
Original commit message for r252949:
Provide a way to specify inliner's attribute compatibility and merging
rules using table-gen. NFC.
This commit adds new classes CompatRule and MergeRule to Attributes.td,
which are used to generate code to check attribute compatibility and
merge attributes of the caller and callee.
rdar://problem/19836465
llvm-svn: 256277
This uses the same criteria used in CFE's CodeGenPGO to identify hot and cold
callees and uses values of inlinehint-threshold and inlinecold-threshold
respectively as the thresholds for such callees.
Differential Revision: http://reviews.llvm.org/D15245
llvm-svn: 256222
This reapplies r252949. I've changed the type of FuncName to be
std::string instead of StringRef in emitFnAttrCompatCheck.
Original commit message for r252949:
Provide a way to specify inliner's attribute compatibility and merging
rules using table-gen. NFC.
This commit adds new classes CompatRule and MergeRule to Attributes.td,
which are used to generate code to check attribute compatibility and
merge attributes of the caller and callee.
rdar://problem/19836465
llvm-svn: 252990
rules using table-gen. NFC.
This commit adds new classes CompatRule and MergeRule to Attributes.td,
which are used to generate code to check attribute compatibility and
merge attributes of the caller and callee.
rdar://problem/19836465
llvm-svn: 252949
Place new and update dbg.declare calls immediately after the
corresponding alloca.
Current code in replaceDbgDeclareForAlloca puts the new dbg.declare
at the end of the basic block. LLVM codegen has problems emitting
debug info in a situation when dbg.declare appears after all uses of
the variable. This usually kinda works for inlining and ASan (two
users of this function) but not for SafeStack (see the pending change
in http://reviews.llvm.org/D13178).
llvm-svn: 248769
with the new pass manager, and no longer relying on analysis groups.
This builds essentially a ground-up new AA infrastructure stack for
LLVM. The core ideas are the same that are used throughout the new pass
manager: type erased polymorphism and direct composition. The design is
as follows:
- FunctionAAResults is a type-erasing alias analysis results aggregation
interface to walk a single query across a range of results from
different alias analyses. Currently this is function-specific as we
always assume that aliasing queries are *within* a function.
- AAResultBase is a CRTP utility providing stub implementations of
various parts of the alias analysis result concept, notably in several
cases in terms of other more general parts of the interface. This can
be used to implement only a narrow part of the interface rather than
the entire interface. This isn't really ideal, this logic should be
hoisted into FunctionAAResults as currently it will cause
a significant amount of redundant work, but it faithfully models the
behavior of the prior infrastructure.
- All the alias analysis passes are ported to be wrapper passes for the
legacy PM and new-style analysis passes for the new PM with a shared
result object. In some cases (most notably CFL), this is an extremely
naive approach that we should revisit when we can specialize for the
new pass manager.
- BasicAA has been restructured to reflect that it is much more
fundamentally a function analysis because it uses dominator trees and
loop info that need to be constructed for each function.
All of the references to getting alias analysis results have been
updated to use the new aggregation interface. All the preservation and
other pass management code has been updated accordingly.
The way the FunctionAAResultsWrapperPass works is to detect the
available alias analyses when run, and add them to the results object.
This means that we should be able to continue to respect when various
passes are added to the pipeline, for example adding CFL or adding TBAA
passes should just cause their results to be available and to get folded
into this. The exception to this rule is BasicAA which really needs to
be a function pass due to using dominator trees and loop info. As
a consequence, the FunctionAAResultsWrapperPass directly depends on
BasicAA and always includes it in the aggregation.
This has significant implications for preserving analyses. Generally,
most passes shouldn't bother preserving FunctionAAResultsWrapperPass
because rebuilding the results just updates the set of known AA passes.
The exception to this rule are LoopPass instances which need to preserve
all the function analyses that the loop pass manager will end up
needing. This means preserving both BasicAAWrapperPass and the
aggregating FunctionAAResultsWrapperPass.
Now, when preserving an alias analysis, you do so by directly preserving
that analysis. This is only necessary for non-immutable-pass-provided
alias analyses though, and there are only three of interest: BasicAA,
GlobalsAA (formerly GlobalsModRef), and SCEVAA. Usually BasicAA is
preserved when needed because it (like DominatorTree and LoopInfo) is
marked as a CFG-only pass. I've expanded GlobalsAA into the preserved
set everywhere we previously were preserving all of AliasAnalysis, and
I've added SCEVAA in the intersection of that with where we preserve
SCEV itself.
One significant challenge to all of this is that the CGSCC passes were
actually using the alias analysis implementations by taking advantage of
a pretty amazing set of loop holes in the old pass manager's analysis
management code which allowed analysis groups to slide through in many
cases. Moving away from analysis groups makes this problem much more
obvious. To fix it, I've leveraged the flexibility the design of the new
PM components provides to just directly construct the relevant alias
analyses for the relevant functions in the IPO passes that need them.
This is a bit hacky, but should go away with the new pass manager, and
is already in many ways cleaner than the prior state.
Another significant challenge is that various facilities of the old
alias analysis infrastructure just don't fit any more. The most
significant of these is the alias analysis 'counter' pass. That pass
relied on the ability to snoop on AA queries at different points in the
analysis group chain. Instead, I'm planning to build printing
functionality directly into the aggregation layer. I've not included
that in this patch merely to keep it smaller.
Note that all of this needs a nearly complete rewrite of the AA
documentation. I'm planning to do that, but I'd like to make sure the
new design settles, and to flesh out a bit more of what it looks like in
the new pass manager first.
Differential Revision: http://reviews.llvm.org/D12080
llvm-svn: 247167
The only place that tries to return a CallGraph by value
(CallGraphAnalysis::run) doesn't seem to be used right now, but it's a
reasonable bit of cleanup anyway.
llvm-svn: 244122
Create wrapper methods in the Function class for the OptimizeForSize and MinSize
attributes. We want to hide the logic of "or'ing" them together when optimizing
just for size (-Os).
Currently, we are not consistent about this and rely on a front-end to always set
OptimizeForSize (-Os) if MinSize (-Oz) is on. Thus, there are 18 FIXME changes here
that should be added as follow-on patches with regression tests.
This patch is NFC-intended: it just replaces existing direct accesses of the attributes
by the equivalent wrapper call.
Differential Revision: http://reviews.llvm.org/D11734
llvm-svn: 243994
Not sure if the optimizer will save the call as getCalledFunction()
is not a trivial access function but the code is clearer this way.
llvm-svn: 242641
This patch adds the safe stack instrumentation pass to LLVM, which separates
the program stack into a safe stack, which stores return addresses, register
spills, and local variables that are statically verified to be accessed
in a safe way, and the unsafe stack, which stores everything else. Such
separation makes it much harder for an attacker to corrupt objects on the
safe stack, including function pointers stored in spilled registers and
return addresses. You can find more information about the safe stack, as
well as other parts of or control-flow hijack protection technique in our
OSDI paper on code-pointer integrity (http://dslab.epfl.ch/pubs/cpi.pdf)
and our project website (http://levee.epfl.ch).
The overhead of our implementation of the safe stack is very close to zero
(0.01% on the Phoronix benchmarks). This is lower than the overhead of
stack cookies, which are supported by LLVM and are commonly used today,
yet the security guarantees of the safe stack are strictly stronger than
stack cookies. In some cases, the safe stack improves performance due to
better cache locality.
Our current implementation of the safe stack is stable and robust, we
used it to recompile multiple projects on Linux including Chromium, and
we also recompiled the entire FreeBSD user-space system and more than 100
packages. We ran unit tests on the FreeBSD system and many of the packages
and observed no errors caused by the safe stack. The safe stack is also fully
binary compatible with non-instrumented code and can be applied to parts of
a program selectively.
This patch is our implementation of the safe stack on top of LLVM. The
patches make the following changes:
- Add the safestack function attribute, similar to the ssp, sspstrong and
sspreq attributes.
- Add the SafeStack instrumentation pass that applies the safe stack to all
functions that have the safestack attribute. This pass moves all unsafe local
variables to the unsafe stack with a separate stack pointer, whereas all
safe variables remain on the regular stack that is managed by LLVM as usual.
- Invoke the pass as the last stage before code generation (at the same time
the existing cookie-based stack protector pass is invoked).
- Add unit tests for the safe stack.
Original patch by Volodymyr Kuznetsov and others at the Dependable Systems
Lab at EPFL; updates and upstreaming by myself.
Differential Revision: http://reviews.llvm.org/D6094
llvm-svn: 239761
COMDAT groups which have become rendered unused because of inline are
discardable if we can prove that we've made the group empty.
This fixes PR22285.
llvm-svn: 236539
Summary:
DataLayout keeps the string used for its creation.
As a side effect it is no longer needed in the Module.
This is "almost" NFC, the string is no longer
canonicalized, you can't rely on two "equals" DataLayout
having the same string returned by getStringRepresentation().
Get rid of DataLayoutPass: the DataLayout is in the Module
The DataLayout is "per-module", let's enforce this by not
duplicating it more than necessary.
One more step toward non-optionality of the DataLayout in the
module.
Make DataLayout Non-Optional in the Module
Module->getDataLayout() will never returns nullptr anymore.
Reviewers: echristo
Subscribers: resistor, llvm-commits, jholewinski
Differential Revision: http://reviews.llvm.org/D7992
From: Mehdi Amini <mehdi.amini@apple.com>
llvm-svn: 231270
Canonicalize access to function attributes to use the simpler API.
getAttributes().getAttribute(AttributeSet::FunctionIndex, Kind)
=> getFnAttribute(Kind)
getAttributes().hasAttribute(AttributeSet::FunctionIndex, Kind)
=> hasFnAttribute(Kind)
llvm-svn: 229202
The pass is really just a means of accessing a cached instance of the
TargetLibraryInfo object, and this way we can re-use that object for the
new pass manager as its result.
Lots of delta, but nothing interesting happening here. This is the
common pattern that is developing to allow analyses to live in both the
old and new pass manager -- a wrapper pass in the old pass manager
emulates the separation intrinsic to the new pass manager between the
result and pass for analyses.
llvm-svn: 226157
While the term "Target" is in the name, it doesn't really have to do
with the LLVM Target library -- this isn't an abstraction which LLVM
targets generally need to implement or extend. It has much more to do
with modeling the various runtime libraries on different OSes and with
different runtime environments. The "target" in this sense is the more
general sense of a target of cross compilation.
This is in preparation for porting this analysis to the new pass
manager.
No functionality changed, and updates inbound for Clang and Polly.
llvm-svn: 226078
a cache of assumptions for a single function, and an immutable pass that
manages those caches.
The motivation for this change is two fold. Immutable analyses are
really hacks around the current pass manager design and don't exist in
the new design. This is usually OK, but it requires that the core logic
of an immutable pass be reasonably partitioned off from the pass logic.
This change does precisely that. As a consequence it also paves the way
for the *many* utility functions that deal in the assumptions to live in
both pass manager worlds by creating an separate non-pass object with
its own independent API that they all rely on. Now, the only bits of the
system that deal with the actual pass mechanics are those that actually
need to deal with the pass mechanics.
Once this separation is made, several simplifications become pretty
obvious in the assumption cache itself. Rather than using a set and
callback value handles, it can just be a vector of weak value handles.
The callers can easily skip the handles that are null, and eventually we
can wrap all of this up behind a filter iterator.
For now, this adds boiler plate to the various passes, but this kind of
boiler plate will end up making it possible to port these passes to the
new pass manager, and so it will end up factored away pretty reasonably.
llvm-svn: 225131
This is to be consistent with StringSet and ultimately with the standard
library's associative container insert function.
This lead to updating SmallSet::insert to return pair<iterator, bool>,
and then to update SmallPtrSet::insert to return pair<iterator, bool>,
and then to update all the existing users of those functions...
llvm-svn: 222334
A function with discardable linkage cannot be discarded if its a member
of a COMDAT group without considering all the other COMDAT members as
well. This sort of thing is already handled by GlobalOpt/GlobalDCE.
This fixes PR21206.
llvm-svn: 219335
This adds an immutable pass, AssumptionTracker, which keeps a cache of
@llvm.assume call instructions within a module. It uses callback value handles
to keep stale functions and intrinsics out of the map, and it relies on any
code that creates new @llvm.assume calls to notify it of the new instructions.
The benefit is that code needing to find @llvm.assume intrinsics can do so
directly, without scanning the function, thus allowing the cost of @llvm.assume
handling to be negligible when none are present.
The current design is intended to be lightweight. We don't keep track of
anything until we need a list of assumptions in some function. The first time
this happens, we scan the function. After that, we add/remove @llvm.assume
calls from the cache in response to registration calls and ValueHandle
callbacks.
There are no new direct test cases for this pass, but because it calls it
validation function upon module finalization, we'll pick up detectable
inconsistencies from the other tests that touch @llvm.assume calls.
This pass will be used by follow-up commits that make use of @llvm.assume.
llvm-svn: 217334
This feeds AA through the IFI structure into the inliner so that
AddAliasScopeMetadata can use AA->getModRefBehavior to figure out which
functions only access their arguments (instead of just hard-coding some
knowledge of memory intrinsics). Most of the information is only available from
BasicAA; this is important for preserving alias scoping information for
target-specific intrinsics when doing the noalias parameter attribute to
metadata conversion.
llvm-svn: 216866
Summary:
This adds two new diagnostics: -pass-remarks-missed and
-pass-remarks-analysis. They take the same values as -pass-remarks but
are intended to be triggered in different contexts.
-pass-remarks-missed is used by LLVMContext::emitOptimizationRemarkMissed,
which passes call when they tried to apply a transformation but
couldn't.
-pass-remarks-analysis is used by LLVMContext::emitOptimizationRemarkAnalysis,
which passes call when they want to inform the user about analysis
results.
The patch also:
1- Adds support in the inliner for the two new remarks and a
test case.
2- Moves emitOptimizationRemark* functions to the llvm namespace.
3- Adds an LLVMContext argument instead of making them member functions
of LLVMContext.
Reviewers: qcolombet
Subscribers: llvm-commits
Differential Revision: http://reviews.llvm.org/D3682
llvm-svn: 209442
override the default cold threshold.
When we use command line argument to set the inline threshold, the default
cold threshold will not be used. This is in line with how we use
OptSizeThreshold. When we want a higher threshold for all functions, we
do not have to set both inline threshold and cold threshold.
llvm-svn: 207245
definition below all of the header #include lines, lib/Transforms/...
edition.
This one is tricky for two reasons. We again have a couple of passes
that define something else before the includes as well. I've sunk their
name macros with the DEBUG_TYPE.
Also, InstCombine contains headers that need DEBUG_TYPE, so now those
headers #define and #undef DEBUG_TYPE around their code, leaving them
well formed modular headers. Fixing these headers was a large motivation
for all of these changes, as "leaky" macros of this form are hard on the
modules implementation.
llvm-svn: 206844
Summary:
This patch adds backend support for -Rpass=, which indicates the name
of the optimization pass that should emit remarks stating when it
made a transformation to the code.
Pass names are taken from their DEBUG_NAME definitions.
When emitting an optimization report diagnostic, the lack of debug
information causes the diagnostic to use "<unknown>:0:0" as the
location string.
This is the back end counterpart for
http://llvm-reviews.chandlerc.com/D3226
Reviewers: qcolombet
CC: llvm-commits
Differential Revision: http://llvm-reviews.chandlerc.com/D3227
llvm-svn: 205774
This requires a number of steps.
1) Move value_use_iterator into the Value class as an implementation
detail
2) Change it to actually be a *Use* iterator rather than a *User*
iterator.
3) Add an adaptor which is a User iterator that always looks through the
Use to the User.
4) Wrap these in Value::use_iterator and Value::user_iterator typedefs.
5) Add the range adaptors as Value::uses() and Value::users().
6) Update *all* of the callers to correctly distinguish between whether
they wanted a use_iterator (and to explicitly dig out the User when
needed), or a user_iterator which makes the Use itself totally
opaque.
Because #6 requires churning essentially everything that walked the
Use-Def chains, I went ahead and added all of the range adaptors and
switched them to range-based loops where appropriate. Also because the
renaming requires at least churning every line of code, it didn't make
any sense to split these up into multiple commits -- all of which would
touch all of the same lies of code.
The result is still not quite optimal. The Value::use_iterator is a nice
regular iterator, but Value::user_iterator is an iterator over User*s
rather than over the User objects themselves. As a consequence, it fits
a bit awkwardly into the range-based world and it has the weird
extra-dereferencing 'operator->' that so many of our iterators have.
I think this could be fixed by providing something which transforms
a range of T&s into a range of T*s, but that *can* be separated into
another patch, and it isn't yet 100% clear whether this is the right
move.
However, this change gets us most of the benefit and cleans up
a substantial amount of code around Use and User. =]
llvm-svn: 203364
225 is the default value of inline-threshold. This change will make sure
we have the same inlining behavior as prior to r200886.
As Chandler points out, even though we don't have code in our testing
suite that uses cold attribute, there are larger applications that do
use cold attribute.
r200886 + this commit intend to keep the same behavior as prior to r200886.
We can later on tune the inlinecold-threshold.
The main purpose of r200886 is to help performance of instrumentation based
PGO before we actually hook up inliner with analysis passes such as BPI and BFI.
For instrumentation based PGO, we try to increase inlining of hot functions and
reduce inlining of cold functions by setting inlinecold-threshold.
Another option suggested by Chandler is to use a boolean flag that controls
if we should use OptSizeThreshold for cold functions. The default value
of the boolean flag should not change the current behavior. But it gives us
less freedom in controlling inlining of cold functions.
llvm-svn: 200898
Added command line option inlinecold-threshold to set threshold for inlining
functions with cold attribute. Listen to the cold attribute when it would
decrease the inline threshold.
llvm-svn: 200886
CallGraph.
This makes the CallGraph a totally generic analysis object that is the
container for the graph data structure and the primary interface for
querying and manipulating it. The pass logic is separated into its own
class. For compatibility reasons, the pass provides wrapper methods for
most of the methods on CallGraph -- they all just forward.
This will allow the new pass manager infrastructure to provide its own
analysis pass that constructs the same CallGraph object and makes it
available. The idea is that in the new pass manager, the analysis pass's
'run' method returns a concrete analysis 'result'. Here, that result is
a 'CallGraph'. The 'run' method will typically do only minimal work,
deferring much of the work into the implementation of the result object
in order to be lazy about computing things, but when (like DomTree)
there is *some* up-front computation, the analysis does it prior to
handing the result back to the querying pass.
I know some of this is fairly ugly. I'm happy to change it around if
folks can suggest a cleaner interim state, but there is going to be some
amount of unavoidable ugliness during the transition period. The good
thing is that this is very limited and will naturally go away when the
old pass infrastructure goes away. It won't hang around to bother us
later.
Next up is the initial new-PM-style call graph analysis. =]
llvm-svn: 195722
Duncan pointed out a mistake in my fix in r186425 when only one of the allocas
being compared had the target-default alignment. This is essentially his
suggested solution. Thanks!
llvm-svn: 186510
For safety, the inliner cannot decrease the allignment on an alloca when
merging it with another.
I've included two variants of the test case for this: one with DataLayout
available, and one without. When DataLayout is not available, if only one of
the allocas uses the default alignment (getAlignment() == 0), then they cannot
be safely merged.
llvm-svn: 186425
SSPStrong applies a heuristic to insert stack protectors in these situations:
* A Protector is required for functions which contain an array, regardless of
type or length.
* A Protector is required for functions which contain a structure/union which
contains an array, regardless of type or length. Note, there is no limit to
the depth of nesting.
* A protector is required when the address of a local variable (i.e., stack
based variable) is exposed. (E.g., such as through a local whose address is
taken as part of the RHS of an assignment or a local whose address is taken as
part of a function argument.)
This patch implements the SSPString attribute to be equivalent to
SSPRequired. This will change in a subsequent patch.
llvm-svn: 173230
into their new header subdirectory: include/llvm/IR. This matches the
directory structure of lib, and begins to correct a long standing point
of file layout clutter in LLVM.
There are still more header files to move here, but I wanted to handle
them in separate commits to make tracking what files make sense at each
layer easier.
The only really questionable files here are the target intrinsic
tablegen files. But that's a battle I'd rather not fight today.
I've updated both CMake and Makefile build systems (I think, and my
tests think, but I may have missed something).
I've also re-sorted the includes throughout the project. I'll be
committing updates to Clang, DragonEgg, and Polly momentarily.
llvm-svn: 171366
directly.
This is in preparation for removing the use of the 'Attribute' class as a
collection of attributes. That will shift to the AttributeSet class instead.
llvm-svn: 171253
Better controls the inlining of functions when the caller function has MinSize attribute.
Basically, when the caller function has this attribute, we do not "force" the inlining
of callee functions carrying the InlineHint attribute (i.e., functions defined with
inline keyword)
llvm-svn: 170065
Sooooo many of these had incorrect or strange main module includes.
I have manually inspected all of these, and fixed the main module
include to be the nearest plausible thing I could find. If you own or
care about any of these source files, I encourage you to take some time
and check that these edits were sensible. I can't have broken anything
(I strictly added headers, and reordered them, never removed), but they
may not be the headers you'd really like to identify as containing the
API being implemented.
Many forward declarations and missing includes were added to a header
files to allow them to parse cleanly when included first. The main
module rule does in fact have its merits. =]
llvm-svn: 169131
We use the enums to query whether an Attributes object has that attribute. The
opaque layer is responsible for knowing where that specific attribute is stored.
llvm-svn: 165488
This disables malloc-specific optimization when -fno-builtin (or -ffreestanding)
is specified. This has been a problem for a long time but became more severe
with the recent memory builtin improvements.
Since the memory builtin functions are used everywhere, this required passing
TLI in many places. This means that functions that now have an optional TLI
argument, like RecursivelyDeleteTriviallyDeadFunctions, won't remove dead
mallocs anymore if the TLI argument is missing. I've updated most passes to do
the right thing.
Fixes PR13694 and probably others.
llvm-svn: 162841
As a side note, I really dislike array_pod_sort... Do we really still
care about any STL implementations that get this so wrong? Does libc++?
llvm-svn: 153834
interfaces. These methods were used in the old inline cost system where
there was a persistent cache that had to be updated, invalidated, and
cleared. We're now doing more direct computations that don't require
this intricate dance. Even if we resume some level of caching, it would
almost certainly have a simpler and more narrow interface than this.
llvm-svn: 153813
on a per-callsite walk of the called function's instructions, in
breadth-first order over the potentially reachable set of basic blocks.
This is a major shift in how inline cost analysis works to improve the
accuracy and rationality of inlining decisions. A brief outline of the
algorithm this moves to:
- Build a simplification mapping based on the callsite arguments to the
function arguments.
- Push the entry block onto a worklist of potentially-live basic blocks.
- Pop the first block off of the *front* of the worklist (for
breadth-first ordering) and walk its instructions using a custom
InstVisitor.
- For each instruction's operands, re-map them based on the
simplification mappings available for the given callsite.
- Compute any simplification possible of the instruction after
re-mapping, and store that back int othe simplification mapping.
- Compute any bonuses, costs, or other impacts of the instruction on the
cost metric.
- When the terminator is reached, replace any conditional value in the
terminator with any simplifications from the mapping we have, and add
any successors which are not proven to be dead from these
simplifications to the worklist.
- Pop the next block off of the front of the worklist, and repeat.
- As soon as the cost of inlining exceeds the threshold for the
callsite, stop analyzing the function in order to bound cost.
The primary goal of this algorithm is to perfectly handle dead code
paths. We do not want any code in trivially dead code paths to impact
inlining decisions. The previous metric was *extremely* flawed here, and
would always subtract the average cost of two successors of
a conditional branch when it was proven to become an unconditional
branch at the callsite. There was no handling of wildly different costs
between the two successors, which would cause inlining when the path
actually taken was too large, and no inlining when the path actually
taken was trivially simple. There was also no handling of the code
*path*, only the immediate successors. These problems vanish completely
now. See the added regression tests for the shiny new features -- we
skip recursive function calls, SROA-killing instructions, and high cost
complex CFG structures when dead at the callsite being analyzed.
Switching to this algorithm required refactoring the inline cost
interface to accept the actual threshold rather than simply returning
a single cost. The resulting interface is pretty bad, and I'm planning
to do lots of interface cleanup after this patch.
Several other refactorings fell out of this, but I've tried to minimize
them for this patch. =/ There is still more cleanup that can be done
here. Please point out anything that you see in review.
I've worked really hard to try to mirror at least the spirit of all of
the previous heuristics in the new model. It's not clear that they are
all correct any more, but I wanted to minimize the change in this single
patch, it's already a bit ridiculous. One heuristic that is *not* yet
mirrored is to allow inlining of functions with a dynamic alloca *if*
the caller has a dynamic alloca. I will add this back, but I think the
most reasonable way requires changes to the inliner itself rather than
just the cost metric, and so I've deferred this for a subsequent patch.
The test case is XFAIL-ed until then.
As mentioned in the review mail, this seems to make Clang run about 1%
to 2% faster in -O0, but makes its binary size grow by just under 4%.
I've looked into the 4% growth, and it can be fixed, but requires
changes to other parts of the inliner.
llvm-svn: 153812
size bloat. Unfortunately, I expect this to disable the majority of the
benefit from r152737. I'm hopeful at least that it will fix PR12345. To
explain this requires... quite a bit of backstory I'm afraid.
TL;DR: The change in r152737 actually did The Wrong Thing for
linkonce-odr functions. This change makes it do the right thing. The
benefits we saw were simple luck, not any actual strategy. Benchmark
numbers after a mini-blog-post so that I've written down my thoughts on
why all of this works and doesn't work...
To understand what's going on here, you have to understand how the
"bottom-up" inliner actually works. There are two fundamental modes to
the inliner:
1) Standard fixed-cost bottom-up inlining. This is the mode we usually
think about. It walks from the bottom of the CFG up to the top,
looking at callsites, taking information about the callsite and the
called function and computing th expected cost of inlining into that
callsite. If the cost is under a fixed threshold, it inlines. It's
a touch more complicated than that due to all the bonuses, weights,
etc. Inlining the last callsite to an internal function gets higher
weighth, etc. But essentially, this is the mode of operation.
2) Deferred bottom-up inlining (a term I just made up). This is the
interesting mode for this patch an r152737. Initially, this works
just like mode #1, but once we have the cost of inlining into the
callsite, we don't just compare it with a fixed threshold. First, we
check something else. Let's give some names to the entities at this
point, or we'll end up hopelessly confused. We're considering
inlining a function 'A' into its callsite within a function 'B'. We
want to check whether 'B' has any callers, and whether it might be
inlined into those callers. If so, we also check whether inlining 'A'
into 'B' would block any of the opportunities for inlining 'B' into
its callers. We take the sum of the costs of inlining 'B' into its
callers where that inlining would be blocked by inlining 'A' into
'B', and if that cost is less than the cost of inlining 'A' into 'B',
then we skip inlining 'A' into 'B'.
Now, in order for #2 to make sense, we have to have some confidence that
we will actually have the opportunity to inline 'B' into its callers
when cheaper, *and* that we'll be able to revisit the decision and
inline 'A' into 'B' if that ever becomes the correct tradeoff. This
often isn't true for external functions -- we can see very few of their
callers, and we won't be able to re-consider inlining 'A' into 'B' if
'B' is external when we finally see more callers of 'B'. There are two
cases where we believe this to be true for C/C++ code: functions local
to a translation unit, and functions with an inline definition in every
translation unit which uses them. These are represented as internal
linkage and linkonce-odr (resp.) in LLVM. I enabled this logic for
linkonce-odr in r152737.
Unfortunately, when I did that, I also introduced a subtle bug. There
was an implicit assumption that the last caller of the function within
the TU was the last caller of the function in the program. We want to
bonus the last caller of the function in the program by a huge amount
for inlining because inlining that callsite has very little cost.
Unfortunately, the last caller in the TU of a linkonce-odr function is
*not* the last caller in the program, and so we don't want to apply this
bonus. If we do, we can apply it to one callsite *per-TU*. Because of
the way deferred inlining works, when it sees this bonus applied to one
callsite in the TU for 'B', it decides that inlining 'B' is of the
*utmost* importance just so we can get that final bonus. It then
proceeds to essentially force deferred inlining regardless of the actual
cost tradeoff.
The result? PR12345: code bloat, code bloat, code bloat. Another result
is getting *damn* lucky on a few benchmarks, and the over-inlining
exposing critically important optimizations. I would very much like
a list of benchmarks that regress after this change goes in, with
bitcode before and after. This will help me greatly understand what
opportunities the current cost analysis is missing.
Initial benchmark numbers look very good. WebKit files that exhibited
the worst of PR12345 went from growing to shrinking compared to Clang
with r152737 reverted.
- Bootstrapped Clang is 3% smaller with this change.
- Bootstrapped Clang -O0 over a single-source-file of lib/Lex is 4%
faster with this change.
Please let me know about any other performance impact you see. Thanks to
Nico for reporting and urging me to actually fix, Richard Smith, Duncan
Sands, Manuel Klimek, and Benjamin Kramer for talking through the issues
today.
llvm-svn: 153506
to instead rely on much more generic and powerful instruction
simplification in the function cloner (and thus inliner).
This teaches the pruning function cloner to use instsimplify rather than
just the constant folder to fold values during cloning. This can
simplify a large number of things that constant folding alone cannot
begin to touch. For example, it will realize that 'or' and 'and'
instructions with certain constant operands actually become constants
regardless of what their other operand is. It also can thread back
through the caller to perform simplifications that are only possible by
looking up a few levels. In particular, GEPs and pointer testing tend to
fold much more heavily with this change.
This should (in some cases) have a positive impact on compile times with
optimizations on because the inliner itself will simply avoid cloning
a great deal of code. It already attempted to prune proven-dead code,
but now it will be use the stronger simplifications to prove more code
dead.
llvm-svn: 153403
directly query the function information which this set was representing.
This simplifies the interface of the inline cost analysis, and makes the
always-inline pass significantly more efficient.
Previously, always-inline would first make a single set of every
function in the module *except* those marked with the always-inline
attribute. It would then query this set at every call site to see if the
function was a member of the set, and if so, refuse to inline it. This
is quite wasteful. Instead, simply check the function attribute directly
when looking at the callsite.
The normal inliner also had similar redundancy. It added every function
in the module with the noinline attribute to its set to ignore, even
though inside the cost analysis function we *already tested* the
noinline attribute and produced the same result.
The only tricky part of removing this is that we have to be able to
correctly remove only the functions inlined by the always-inline pass
when finalizing, which requires a bit of a hack. Still, much less of
a hack than the set of all non-always-inline functions was. While I was
touching this function, I switched a heavy-weight set to a vector with
sort+unique. The algorithm already had a two-phase insert and removal
pattern, we were just needlessly paying the uniquing cost on every
insert.
This probably speeds up some compiles by a small amount (-O0 compiles
with lots of always-inline, so potentially heavy libc++ users), but I've
not tried to measure it.
I believe there is no functional change here, but yell if you spot one.
None are intended.
Finally, the direction this is going in is to greatly simplify the
inline cost query interface so that we can replace its implementation
with a much more clever one. Along the way, all the APIs get simplified,
so it seems incrementally good.
llvm-svn: 152903
which are small enough to themselves be inlined. Delaying in this manner
can be harmful if the function is inelligible for inlining in some (or
many) contexts as it pessimizes the code of the function itself in the
event that inlining does not eventually happen.
Previously the check was written to only do this delaying of inlining
for static functions in the hope that they could be entirely deleted and
in the knowledge that all callers of static functions will have the
opportunity to inline if it is in fact profitable. However, with C++ we
get two other important sources of functions where the definition is
always available for inlining: inline functions and templated functions.
This patch generalizes the inliner to allow linkonce-ODR (the linkage
such C++ routines receive) to also qualify for this delay-based
inlining.
Benchmarking across a range of large real-world applications shows
roughly 2% size increase across the board, but an average speedup of
about 0.5%. Some benhcmarks improved over 2%, and the 'clang' binary
itself (when bootstrapped with this feature) shows a 1% -O0 performance
improvement when run over all Sema, Lex, and Parse source code smashed
into a single file. A clean re-build of Clang+LLVM with a bootstrapped
Clang shows approximately 2% improvement, but that measurement is often
noisy.
llvm-svn: 152737
candidate set for subsequent inlining, try to simplify the arguments to
the inner call site now that inlining has been performed.
The goal here is to propagate and fold constants through deeply nested
call chains. Without doing this, we loose the inliner bonus that should
be applied because the arguments don't match the exact pattern the cost
estimator uses.
Reviewed on IRC by Benjamin Kramer.
llvm-svn: 152556
are optimization hints, but at -O0 we're not optimizing. This becomes a problem
when the alwaysinline attribute is abused.
rdar://10921594
llvm-svn: 151429
case where a static caller is itself inlined everywhere else, and
thus may go away if it doesn't get too big due to inlining other
things into it. If there are references to the caller other than
calls, it will not be removed; account for this.
This results in same-day completion of the case in PR8853.
llvm-svn: 122821
optimization.
Consider:
static void foo() {
A = alloca
...
}
static void bar() {
B = alloca
...
call foo();
}
void main() {
bar()
}
The inliner proceeds bottom up, but lets pretend it decides not to inline foo
into bar. When it gets to main, it inlines bar into main(), and says "hey, I
just inlined an alloca "B" into main, lets remember that. Then it keeps going
and finds that it now contains a call to foo. It decides to inline foo into
main, and says "hey, foo has an alloca A, and I have an alloca B from another
inlined call site, lets reuse it". The problem with this of course, is that
the lifetime of A and B are nested, not disjoint.
Unfortunately I can't create a reasonable testcase for this: the one in the
PR is both huge and extremely sensitive, because you minor tweaks end up
causing foo to get inlined into bar too early. We already have tests for the
basic alloca merging optimization and this does not break them.
llvm-svn: 120995
halting analysis, it is illegal to delete a call to a read-only function.
The correct solution is almost certainly to add a "must halt" attribute and
only allow deletions in its presence.
XFAIL the relevant testcase for now.
llvm-svn: 102831
Chandler Carruth