Commit Graph

3537 Commits

Author SHA1 Message Date
Dan Gohman ec4557f324 Fix SplitCriticalEdge to properly update LCSSA form when splitting a
loop exit edge -- new PHIs may be needed not only for the additional
splits that are made to preserve LoopSimplify form, but also for the
original split. Factor out the code that inserts new PHIs so that it
can be used for both. Remove LoopRotation.cpp's code for manually
updating LCSSA form, as it is now redundant. This fixes PR4934.

llvm-svn: 81363
2009-09-09 18:18:18 +00:00
Mike Stump deaf572ca8 Reflow comment.
llvm-svn: 81361
2009-09-09 17:57:16 +00:00
Dan Gohman c56af25c01 Fix an 80-column violation.
llvm-svn: 81354
2009-09-09 17:17:19 +00:00
Chris Lattner 9ce1781ef4 remove an extremely dubious instcombine transformation of
extractelement(load).

llvm-svn: 81239
2009-09-08 18:48:01 +00:00
Dan Gohman 3ddbc242fb Re-apply r80926, with fixes: keep the domtree informed of new blocks
that get created during loop unswitching, and fix SplitBlockPredecessors'
LCSSA updating code to create new PHIs instead of trying to just move
existing ones.

Also, optimize Loop::verifyLoop, since it gets called a lot. Use
searches on a sorted list of blocks instead of calling the "contains"
function, as is done in other places in the Loop class, since "contains"
does a linear search. Also, don't call verifyLoop from LoopSimplify or
LCSSA, as the PassManager is already calling verifyLoop as part of
LoopInfo's verifyAnalysis.

llvm-svn: 81221
2009-09-08 15:45:00 +00:00
Chris Lattner d1b21c6092 remove a turd
llvm-svn: 81186
2009-09-08 03:47:41 +00:00
Chris Lattner d3210e1a20 instcombine transforms vector loads that are only used by
extractelement operations into a bitcast of the pointer,
then a gep, then a scalar load.  Disable this when the vector
only has one element, because it leads to infinite loops in
instcombine (PR4908).

This transformation seems like a really bad idea to me, as it
will likely disable CSE of vector load/stores etc and can be
better done in the code generator when profitable.  This
goes all the way back to the first days of packed types,
r25299 specifically.

I'll let those people who care about the performance of vector
code decide what to do with this.

llvm-svn: 81185
2009-09-08 03:44:51 +00:00
Chris Lattner f2ab40a46f Fix PR4882, by making MemCpyOpt not dereference removed stores to get the
context for the newly created operations.
 
Patch by Jakub Staszak!

llvm-svn: 81175
2009-09-08 00:27:14 +00:00
Dan Gohman 1b84908f92 Reappy r80998, now that the GlobalOpt bug that it exposed on MiniSAT is fixed.
llvm-svn: 81172
2009-09-07 23:54:19 +00:00
Duncan Sands 89720bbd11 Remove some not-really-used variables, as warned
about by icc (#593, partial).  Patch by Erick Tryzelaar.

llvm-svn: 81115
2009-09-06 12:41:19 +00:00
Daniel Dunbar 86c6a6ef0f Fix a possible crash call setIsInBounds.
- I think there are more instances of this, but I think they are fixed in Dan's
   incoming patch. This one was preventing me from doing a bugpoint reduction
   though.

llvm-svn: 81103
2009-09-06 02:31:36 +00:00
Evan Cheng 904199547b Revert r80926. It causes loop unswitch assertion and slow down some JIT tests significantly.
llvm-svn: 81101
2009-09-06 02:26:10 +00:00
Daniel Dunbar 10ea8bb8e0 Revert "Include optional subclass flags, such as inbounds, nsw, etc., ...", this
breaks MiniSAT on x86_64.

llvm-svn: 81098
2009-09-06 00:11:24 +00:00
Dan Gohman 0c2477c26b Include optional subclass flags, such as inbounds, nsw, etc., in the
Constant uniquing tables. This allows distinct ConstantExpr objects
with the same operation and different flags.

Even though a ConstantExpr "a + b" is either always overflowing or
never overflowing (due to being a ConstantExpr), it's still necessary
to be able to represent it both with and without overflow flags at
the same time within the IR, because the safety of the flag may
depend on the context of the use. If the constant really does overflow,
it wouldn't ever be safe to use with the flag set, however the use
may be in code that is never actually executed.

This also makes it possible to merge all the flags tests into a single test.

llvm-svn: 80998
2009-09-04 12:08:11 +00:00
Dan Gohman 4c1bdcf5d7 Add a verifyAnalysis to LoopInfo, LoopSimplify, and LCSSA form that verify
that these passes are properly preserved.

Fix several transformation passes that claimed to preserve LoopSimplify
form but weren't.

llvm-svn: 80926
2009-09-03 16:31:42 +00:00
Dan Gohman 22571485b3 Change PHINode::hasConstantValue to have a DominatorTree argument
instead of a bool argument, and to do the dominator check itself.
This makes it eaiser to use when DominatorTree information is
available.

llvm-svn: 80920
2009-09-03 15:34:35 +00:00
Duncan Sands 0edc7100ba Keep track of how many memmove calls were turned into
memcpy calls.

llvm-svn: 80915
2009-09-03 13:37:16 +00:00
Chris Lattner 27266f164f In C++, code is not allowed to call main. In C it is, this
simplifylibcalls optimization is thus valid for C++ but not C.
It's not important enough to worry about for C++ apps, so just
remove it.

rdar://7191924

llvm-svn: 80887
2009-09-03 05:19:59 +00:00
Gabor Greif 2d60e1ec0c back out my recent commit (r80858), it seems to break self-hosting buildbot's stage 2 configure
llvm-svn: 80871
2009-09-03 02:02:59 +00:00
Gabor Greif 14dfba6d66 re-commit r66920 (which has been backed out in r66953) I may have more luck this time. I'll back out if needed...
llvm-svn: 80858
2009-09-03 00:18:58 +00:00
Chris Lattner 4916267c97 fix PR4815: some cases where DeleteDeadInstruction can delete
the instruction BBI points to.

llvm-svn: 80768
2009-09-02 06:31:02 +00:00
Chris Lattner 09a79dcfdf clean up this code a bit.
llvm-svn: 80767
2009-09-02 06:15:37 +00:00
Chris Lattner 2dd09dbdf7 eliminate VISIBILITY_HIDDEN from Transforms/Scalar. PR4861
llvm-svn: 80766
2009-09-02 06:11:42 +00:00
Chris Lattner 64b5842986 fix PR4837, some bugs folding vector compares. These
return a vector of i1, not i1 itself.

llvm-svn: 80761
2009-09-02 05:12:37 +00:00
Chris Lattner 1145e33bc6 enhance memcpy opt to turn memmoves into memcpy when the src/dest
don't alias.  Remove an old and poorly reduced testcase that fails
with this transform for reasons unrelated to the original test.

llvm-svn: 80693
2009-09-01 17:56:32 +00:00
Chris Lattner b5557a7b42 random code cleanups, no functionality change.
llvm-svn: 80682
2009-09-01 17:09:55 +00:00
Chris Lattner ff5f1e4d70 fix some cases where instcombine would change hte IR but not return true
from runOnFunction

llvm-svn: 80562
2009-08-31 06:57:37 +00:00
Chris Lattner 19dd315e67 improve -debug output, so that -debug is more likely to print when
instcombine is changing stuff.

llvm-svn: 80538
2009-08-31 05:17:58 +00:00
Chris Lattner 4e3e930743 fix a bug I introduced with my 'instcombine builder' refactoring
changes: SimplifyDemandedBits can't use the builder yet because it
has the wrong insertion point.  This fixes a crash building
MultiSource/Benchmarks/PAQ8p

llvm-svn: 80537
2009-08-31 04:36:22 +00:00
Chris Lattner 73913f4cd3 Fix PR4748: don't fold gep(bitcast(x)) into bitcast(gep) when x
is itself a bitcast.  Since we have gep(bitcast(bitcast(y))) in this
case, just wait for the two bitcasts to get zapped.  This prevents
instcombine from confusing some aliasing stuff, and allows it to
directly eliminate the load in the testcase.

llvm-svn: 80508
2009-08-30 20:38:21 +00:00
Chris Lattner c2f2cf896e misc cleanup
llvm-svn: 80507
2009-08-30 20:36:46 +00:00
Chris Lattner a3e620caba add getPointerAddressSpace() to GEP instruction, use the method
in a few scalar xforms to simplify things.

llvm-svn: 80506
2009-08-30 20:06:40 +00:00
Chris Lattner c856539edf eliminate InsertCastBefore, use the builder instead.
llvm-svn: 80505
2009-08-30 20:01:10 +00:00
Chris Lattner 606da5fed8 eliminate InsertBitCastBefore, just use the builder instead.
llvm-svn: 80504
2009-08-30 19:47:22 +00:00
Chris Lattner 5966341a2e convert a bunch more calls to InsertNewInstBefore to use
the new Instcombine builder.

llvm-svn: 80501
2009-08-30 18:50:58 +00:00
Chris Lattner 8326d529da fix typo
llvm-svn: 80500
2009-08-30 17:53:59 +00:00
Chris Lattner 022a582de2 give instcombine a custom IRBuilder that adds new instructions to the
workslist and is set to insert new instructions before the current one.
Convert a bunch of stuff that used to call InsertNewInstBefore over to
use it, greatly simplifying code and making it more natural.

There is still a lot more to go, but this is a good start.

llvm-svn: 80492
2009-08-30 07:44:24 +00:00
Chris Lattner a0c89ee1da add a new InstCombineWorklist::AddValue method that works even
if the operand is not an instruction.

Simplify most uses of AddOperandsToWorkList to use AddValue and
inline it into the one remaining callsite.

llvm-svn: 80488
2009-08-30 06:27:41 +00:00
Chris Lattner bacd05c2eb move AddUsersToWorkList to the worklist processing class, make the
argument stronger typed.

llvm-svn: 80487
2009-08-30 06:22:51 +00:00
Chris Lattner 795bfdbb55 rename AddUsesToWorkList -> AddOperandsToWorkList. The
former looks too much like AddUsersToWorkList and keeps
confusing me.

Remove AddSoonDeadInstToWorklist and change its two callers
to do the same thing in a simpler way.

llvm-svn: 80486
2009-08-30 06:20:05 +00:00
Chris Lattner 905976b1db inline the trivial AddToWorkList/RemoveFromWorkList methods
into their callers.  simplify ReplaceInstUsesWith.  Make
EraseInstFromFunction only add operands to the worklist if
there aren't too many of them (this was a scalability win
for crazy programs that was only infrequently enforced).
Switch more code to using EraseInstFromFunction instead of
duplicating it inline.  Change some fcmp/icmp optimizations
to modify fcmp/icmp in place instead of creating a new one
and deleting the old one just to change the predicate.

llvm-svn: 80483
2009-08-30 06:13:40 +00:00
Chris Lattner 93ad6170fd fix a bug I introduced in r80478 found by the build bot.
llvm-svn: 80482
2009-08-30 05:56:44 +00:00
Chris Lattner 97fd3599e1 refactor instcombine's worklist processing stuff out to its own class.
llvm-svn: 80481
2009-08-30 05:55:36 +00:00
Chris Lattner b2995e1eb1 more cleanups: remove some redundant code, and simplify some
other places.

llvm-svn: 80478
2009-08-30 05:30:55 +00:00
Chris Lattner 06c687b59e eliminate the temporary SrcGEPOperands smallvector.
llvm-svn: 80477
2009-08-30 05:08:50 +00:00
Chris Lattner e26bf17423 simplify/detangle some control flow.
llvm-svn: 80476
2009-08-30 05:00:50 +00:00
Chris Lattner d7b6e913fe simplify and cleanup some code, remove some code that just
does constant folding of gep's: this is already handled in
a more general way.

No functionality change.

llvm-svn: 80475
2009-08-30 04:49:01 +00:00
Dan Gohman 0dfe73ac9e Remove an unnecessary Context argument.
llvm-svn: 80454
2009-08-29 23:39:38 +00:00
Chris Lattner bda82c20f3 Fix PR3913, patch by Jakub Staszak!
llvm-svn: 80327
2009-08-28 00:43:14 +00:00
Owen Anderson 109ca5a14a Make this into a static method.
llvm-svn: 80170
2009-08-26 22:55:11 +00:00