Commit Graph

4249 Commits

Author SHA1 Message Date
Nick Lewycky
f201a06662 Factor out the multiply analysis code in ComputeMaskedBits and apply it to the
overflow checking multiply intrinsic as well.

Add a test for this, updating the test from grep to FileCheck.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@153028 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-18 23:28:48 +00:00
Chandler Carruth
f91f5af802 Start removing the use of an ad-hoc 'never inline' set and instead
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.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152903 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-16 06:10:13 +00:00
Chandler Carruth
9b081d9691 Pull the implementation of the code metrics out of the inline cost
analysis implementation. The header was already separated. Also cleanup
all the comments in the header to follow a nice modern doxygen form.

There is still plenty of cruft here, but some of that will fall out in
subsequent refactorings and this was an easy step in the right
direction. No functionality changed here.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152898 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-16 05:51:52 +00:00
Andrew Trick
75ae20366f LSR fix: Add isSimplifiedLoopNest to IVUsers analysis.
Only record IVUsers that are dominated by simplified loop
headers. Otherwise SCEVExpander will crash while looking for a
preheader.

I previously tried to work around this in LSR itself, but that was
insufficient. This way, LSR can continue to run if some uses are not
in simple loops, as long as we don't attempt to analyze those users.

Fixes <rdar://problem/11049788> Segmentation fault: 11 in LoopStrengthReduce

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152892 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-16 03:16:56 +00:00
Eric Christopher
75df9f23fa Do the right thing on NULL uint64 fields.
Patch by Clemens Hammacher!

Fixes PR12243

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152880 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-16 00:21:54 +00:00
Duncan Sands
f72e0ca715 Type sizes and fields offsets inside structs are unsigned. This is a highly
theoretical fix since it only matters for types with >= 2^63 bits (!) and also
only matters if pointers have more than 64 bits, which is not supported anyway.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152831 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-15 20:14:42 +00:00
Chandler Carruth
377c7f049d Make the swap code here a bit more obvious what its doing... We're
essentially sorting the pair's arguments. I'd love to actually call sort
here, but I'm just not that crazy. ;]

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152764 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-15 00:55:51 +00:00
Chandler Carruth
0e33d9fea2 Don't assume that the arguments are processed in some particular order.
This appears to not be the case with dragonegg at least in some
contexts. Hopefully will fix the bootstrap assert failure there.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152763 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-15 00:50:21 +00:00
Chandler Carruth
220d2d7b50 Remove all remnants of partial specialization in the cost computation
side of things. This is all dead code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152759 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-15 00:29:08 +00:00
Chandler Carruth
274d377ea6 Extend the inline cost calculation to account for bonuses due to
correlated pairs of pointer arguments at the callsite. This is designed
to recognize the common C++ idiom of begin/end pointer pairs when the
end pointer is a constant offset from the begin pointer. With the
C-based idiom of a pointer and size, the inline cost saw the constant
size calculation, and this provides the same level of information for
begin/end pairs.

In order to propagate this information we have to search for candidate
operations on a pair of pointer function arguments (or derived from
them) which would be simplified if the pointers had a known constant
offset. Then the callsite analysis looks for such pointer pairs in the
argument list, and applies the appropriate bonus.

This helps LLVM detect that half of bounds-checked STL algorithms
(such as hash_combine_range, and some hybrid sort implementations)
disappear when inlined with a constant size input. However, it's not
a complete fix due the inaccuracy of our cost metric for constants in
general. I'm looking into that next.

Benchmarks showed no significant code size change, and very minor
performance changes. However, specific code such as hashing is showing
significantly cleaner inlining decisions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152752 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-14 23:19:53 +00:00
Chandler Carruth
3d1d895c86 Refactor the inline cost bonus calculation for constants to use
a worklist rather than a recursive call.

No functionality changed.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152706 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-14 07:32:53 +00:00
Chris Lattner
5161de6ebb enhance jump threading to preserve TBAA information when PRE'ing loads,
fixing rdar://11039258, an issue that came up when inspecting clang's 
bootstrapped codegen.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152635 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-13 18:07:41 +00:00
Duncan Sands
bd0fe56425 Generalize the "trunc(ptrtoint(x)) - trunc(ptrtoint(y)) ->
trunc(ptrtoint(x-y))" optimization introduced by Chandler.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152626 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-13 14:07:05 +00:00
Duncan Sands
0aa85eb231 Uniformize the InstructionSimplify interface by ensuring that all routines
take a TargetLibraryInfo parameter.  Internally, rather than passing TD, TLI
and DT parameters around all over the place, introduce a struct for holding
them.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152623 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-13 11:42:19 +00:00
Eli Friedman
5b8f0ddc7e Fix regression from r151466: an we can't replace uses of an instruction reachable from the entry block with uses of an instruction not reachable from the entry block. PR12231.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152595 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-13 01:06:07 +00:00
Chandler Carruth
90c14fcb7e Address some review comments from Duncan. This moves the iterative
offset accumulation to use a boring APInt instead of ConstantExprs.
I didn't go all the way to an 'int64_t' because I wanted APInt to handle
any magic required to properly wrap the arithmetic when the pointer
width is <64 bits. If there is a significant penalty from using APInt
here, first off WTF, and secondly let me know and I'll do the math by
hand.

I've left one layer still operating w/ ConstantExpr because it makes the
interface quite a bit simpler, and that one isn't iterative so has much
lower cost.

I suppose this may potentially speed up some strang compilation
situations, but I don't really expect much. It should have no functional
impact either way.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152590 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-13 00:06:15 +00:00
Chandler Carruth
fc72ae613a Teach instsimplify how to constant fold pointer differences.
Typically instcombine has handled this, but pointer differences show up
in several contexts where we would like to get constant folding, and
cannot afford to run instcombine. Specifically, I'm working on improving
the constant folding of arguments used in inline cost analysis with
instsimplify.

Doing this in instsimplify implies some algorithm changes. We have to
handle multiple layers of all-constant GEPs because instsimplify cannot
fold them into a single GEP the way instcombine can. Also, we're only
interested in all-constant GEPs. The result is that this doesn't really
replace the instcombine logic, it's just complimentary and focused on
constant folding.

Reviewed on IRC by Benjamin Kramer.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152555 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-12 11:19:31 +00:00
Stepan Dyatkovskiy
3d3abe0852 llvm::SwitchInst
Renamed methods caseBegin, caseEnd and caseDefault with case_begin, case_end, and case_default.
Added some notes relative to case iterators.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152532 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-11 06:09:17 +00:00
Benjamin Kramer
4210b34e59 Make helper static, so it can be inlined into its sole caller.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152515 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-10 22:41:06 +00:00
Bill Wendling
798d013bcb As Duncan pointed out, pointers tend not to be in floating point format...for now.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152499 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-10 18:20:55 +00:00
Bill Wendling
c17731d65d Make this transformation slightly less agressive and more correct.
The 'CmpInst::isFalseWhenEqual' function returns 'false' for values other than
simply equality. For instance, it returns 'false' for <= or >=. This isn't the
correct behavior for this transformation, which is checking for strict equality
and non-equality. It was causing the gcc.c-torture/execute/frame-address.c test
to fail because it would completely (and incorrectly) optimize a whole function
into a 'ret i32 0'.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152497 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-10 17:56:03 +00:00
Chandler Carruth
84dfc32ff9 Refactor some methods to look through bitcasts and GEPs on pointers into
a common collection of methods on Value, and share their implementation.
We had two variations in two different places already, and I need the
third variation for inline cost estimation.

Reviewed by Duncan Sands on IRC, but further comments here welcome.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152490 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-10 08:39:09 +00:00
Nick Lewycky
00cbccceb3 Factor out the analysis of addition and subtraction in ComputeMaskedBits. Reuse
it to analyze extractvalue(llvm.[us](add|sub).with.overflow.*) intrinsics!


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152398 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-09 09:23:50 +00:00
Chandler Carruth
e8187e0294 Undo a previous restriction on the inline cost calculation which Nick
introduced. Specifically, there are cost reductions for all
constant-operand icmp instructions against an alloca, regardless of
whether the alloca will in fact be elligible for SROA. That means we
don't want to abort the icmp reduction computation when we abort the
SROA reduction computation. That in turn frees us from the need to keep
a separate worklist and defer the ICmp calculations.

Use this new-found freedom and some judicious function boundaries to
factor the innards of computing the cost factor of any given instruction
out of the loop over the instructions and into static helper functions.
This greatly simplifies the code, and hopefully makes it more clear what
is happening here.

Reviewed by Eric Christopher. There is some concern that we'd like to
ensure this doesn't get out of hand, and I plan to benchmark the effects
of this change over the next few days along with some further fixes to
the inline cost.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152368 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-09 02:49:36 +00:00
Stepan Dyatkovskiy
c10fa6c801 Taken into account Duncan's comments for r149481 dated by 2nd Feb 2012:
http://lists.cs.uiuc.edu/pipermail/llvm-commits/Week-of-Mon-20120130/136146.html

Implemented CaseIterator and it solves almost all described issues: we don't need to mix operand/case/successor indexing anymore. Base iterator class is implemented as a template since it may be initialized either from "const SwitchInst*" or from "SwitchInst*".

ConstCaseIt is just a read-only iterator.
CaseIt is read-write iterator; it allows to change case successor and case value.

Usage of iterator allows totally remove resolveXXXX methods. All indexing convertions done automatically inside the iterator's getters.

Main way of iterator usage looks like this:
SwitchInst *SI = ... // intialize it somehow

for (SwitchInst::CaseIt i = SI->caseBegin(), e = SI->caseEnd(); i != e; ++i) {
  BasicBlock *BB = i.getCaseSuccessor();
  ConstantInt *V = i.getCaseValue();
  // Do something.
}

If you want to convert case number to TerminatorInst successor index, just use getSuccessorIndex iterator's method.
If you want initialize iterator from TerminatorInst successor index, use CaseIt::fromSuccessorIndex(...) method.

There are also related changes in llvm-clients: klee and clang.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152297 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-08 07:06:20 +00:00
Chandler Carruth
6f130bf368 Rotate two of the functions used to count bonuses for the inline cost
analysis to be methods on the cost analysis's function info object
instead of the code metrics object. These really are just users of the
code metrics, they're building the information for the function's
analysis.

This is the first step of growing the amount of information we collect
about a function in order to cope with pair-wise simplifications due to
allocas.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152283 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-08 02:04:19 +00:00
Nick Lewycky
891495e676 No functionality change. Type::isSized() can be expensive, so avoid calling it
until after other inexpensive tests.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152195 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-07 02:27:53 +00:00
Eli Friedman
923bb4117a A few more cases of missing masking in ComputeMaskedBits; found by inspection.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152070 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-05 23:22:40 +00:00
Eli Friedman
049d08f5c9 Make sure we don't return bits outside the mask in ComputeMaskedBits. PR12189.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152066 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-05 23:09:40 +00:00
Benjamin Kramer
8979e5f372 LVI: Recognize the form instcombine canonicalizes range checks into when forming constant ranges.
This could probably be made a lot smarter, but this is a common case and doesn't require LVI to scan a lot
of code. With this change CVP can optimize away the "shift == 0" case in Hashing.h that only gets hit when
"shift" is in a range not containing 0.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@151919 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-02 15:34:43 +00:00
Eli Friedman
cd38485b8a Duncan pointed out that if the alignment isn't explicitly specified, it defaults to the ABI alignment. Given that, make this code a bit more aggressive in such cases.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@151584 91177308-0d34-0410-b5e6-96231b3b80d8
2012-02-27 23:16:46 +00:00
Eli Friedman
1680a24e53 Teach BasicAA about the LLVM IR rules that allow reading past the end of an object given sufficient alignment. Fixes PR12098.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@151553 91177308-0d34-0410-b5e6-96231b3b80d8
2012-02-27 20:46:07 +00:00
Rafael Espindola
23b6ec906a Fix this assert. IP can point to an instruction with strange dominance
properties (invoke). Just assert that the instruction we return dominates
the insertion point.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@151511 91177308-0d34-0410-b5e6-96231b3b80d8
2012-02-27 02:13:03 +00:00
Rafael Espindola
5465c9422f Don't call dominates on unreachable instructions. Should fix the dragonegg
build. Testcase is still reducing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@151474 91177308-0d34-0410-b5e6-96231b3b80d8
2012-02-26 05:30:08 +00:00
Rafael Espindola
32c185eb27 And update the comment...
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@151472 91177308-0d34-0410-b5e6-96231b3b80d8
2012-02-26 02:36:56 +00:00
Rafael Espindola
5b04cfb785 Enable the assert that got all this dominator work started.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@151471 91177308-0d34-0410-b5e6-96231b3b80d8
2012-02-26 02:29:18 +00:00
Rafael Espindola
c9ae8cc24c Change the implementation of dominates(inst, inst) to one based on what the
verifier does. This correctly handles invoke.
Thanks to Duncan, Andrew and Chris for the comments.
Thanks to Joerg for the early testing.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@151469 91177308-0d34-0410-b5e6-96231b3b80d8
2012-02-26 02:19:19 +00:00
Nick Lewycky
f7087ea508 Reinstate the optimization from r151449 with a fix to not turn 'gep %x' into
'gep null' when the icmp predicate is unsigned (or is signed without inbounds).


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@151467 91177308-0d34-0410-b5e6-96231b3b80d8
2012-02-26 02:09:49 +00:00
Rafael Espindola
8c727f9200 Don't call dominates on unreachable instructions.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@151466 91177308-0d34-0410-b5e6-96231b3b80d8
2012-02-26 01:50:14 +00:00
Nick Lewycky
6fd3428afa Roll these back to r151448 until I figure out how they're breaking
MultiSource/Applications/lua.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@151463 91177308-0d34-0410-b5e6-96231b3b80d8
2012-02-25 23:01:19 +00:00
Nick Lewycky
28e215ba63 An argument and a local identified object (eg. a noalias call) could turn out
equal if both are null. In the test, scope type %t and global @y by adding a
'gep' prefix to them.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@151452 91177308-0d34-0410-b5e6-96231b3b80d8
2012-02-25 20:19:07 +00:00
Nick Lewycky
6bfb9d6a13 Fix five-letter typo in comment.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@151450 91177308-0d34-0410-b5e6-96231b3b80d8
2012-02-25 19:12:58 +00:00
Nick Lewycky
1e4e1c768b Teach instsimplify to be more aggressive when analyzing comparisons of pointers
by using llvm::isIdentifiedObject. Also teach it to handle GEPs that have
the same base pointer and constant operands. Fixes PR11238!


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@151449 91177308-0d34-0410-b5e6-96231b3b80d8
2012-02-25 19:07:42 +00:00
Nick Lewycky
55f4ab84e5 Move isKnownNonNull from private implementation detail of BasicAA to a public
function that others can use, next to llvm::isIdentifiedObject.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@151446 91177308-0d34-0410-b5e6-96231b3b80d8
2012-02-25 10:56:28 +00:00
Chris Lattner
009e2650d6 fix PR12075, a regression in a recent transform I added. In unreachable code, gep chains can be infinite. Just like "stripPointerCasts", use a set to keep track of visited instructions so we don't recurse infinitely.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@151383 91177308-0d34-0410-b5e6-96231b3b80d8
2012-02-24 19:01:58 +00:00
Rafael Espindola
7481d07a07 Fix typo.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@151238 91177308-0d34-0410-b5e6-96231b3b80d8
2012-02-23 05:38:51 +00:00
Chad Rosier
90f20044ad Remove extra semi-colons.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@151169 91177308-0d34-0410-b5e6-96231b3b80d8
2012-02-22 17:25:00 +00:00
Rafael Espindola
b84d540bc9 Improve comment. Thanks for Andrew for the suggestion.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@151127 91177308-0d34-0410-b5e6-96231b3b80d8
2012-02-22 03:44:46 +00:00
Rafael Espindola
919a503458 Semantically revert 151015. Add a comment on why we should be able to assert
the dominance once the dominates method is fixed and why we can use the builder's
insertion point.
Fixes pr12048.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@151125 91177308-0d34-0410-b5e6-96231b3b80d8
2012-02-22 03:21:39 +00:00
Rafael Espindola
e16da6c020 s/the the/the/
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@151079 91177308-0d34-0410-b5e6-96231b3b80d8
2012-02-21 19:27:16 +00:00