Commit Graph

4566 Commits

Author SHA1 Message Date
Devang Patel
780c38d84f Add utility routines to remove dead debug info.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@59011 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-11 00:53:02 +00:00
Devang Patel
d4e7f44319 If the sign of exit condition and split condition does not match
then do not split loop index.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58995 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-10 19:48:34 +00:00
Bill Wendling
11bb6c6d2c Third time's a charm.
The previous patches didn't match correctly. Also, we need to make sure that
the conditional is the same before doing the transformation.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58978 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-10 06:59:06 +00:00
Mon P Wang
aeb06d2462 Added support for the following definition of shufflevector
<result> = shufflevector <n x <ty>> <v1>, <n x <ty>> <v2>, <m x i32> <mask> 



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58964 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-10 04:46:22 +00:00
Bill Wendling
8f24cc237f Correction for the last patch. Should match the conditional in the first part
of the select match, not the select instruction itself.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58947 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-09 23:37:53 +00:00
Bill Wendling
1fb43c32ed The method of doing the matching with a 'select' instruction was wrong. The
original code was matching like this:

	if (match(A, m_Not(m_Value(B))))

B was already matched as a 'select' instruction. However, this isn't matching
what we think it's matching. It would match B as a 'Value', so basically
anything would match to it. In this case, a Constant matched. B was replaced
with a constant representation. And then the wrong value would be used in the
SelectInst::Create statement, causing a crash.

After thinking on this for a moment, and after Nick L. told me how the pattern
matching stuff was supposed to work, the solution was to match NOT an m_Value,
but an m_Select.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58946 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-09 23:17:42 +00:00
Nuno Lopes
ea736cee2b fix leakage of ValueNumbering
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58933 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-09 12:45:23 +00:00
Bill Wendling
c143bcf9ac If the LHS of the FCMP is coming from a UIToFP instruction, then we don't want
to generate signed ICMP instructions to replace the FCMP. This would violate
the following:

define i1 @test1(i32 %val) {
  %1 = uitofp i32 %val to double
  %2 = fcmp ole double %1, 0.000000e+00
  ret i1 %2
}

would be transformed into:

define i1 @test1(i32 %val) {
  %1 = icmp slt i33 %val, 1
  ret i1 %1
}

which is obviously wrong. This patch modifes InstCombiner::FoldFCmp_IntToFP_Cst
to handle when the LHS comes from UIToFP.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58929 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-09 04:26:50 +00:00
Daniel Dunbar
746867c8b6 Rework r58829, allowing removal of dbg info intrinsics during alloca
promotion.
 - Eliminate uses after free and simplify tests.

Devang: Please check that this is still doing what you intended.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58887 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-08 04:12:17 +00:00
Bill Wendling
068a795b33 BCUI + 1 doesn't work. Use next instead.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58830 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-07 01:59:41 +00:00
Devang Patel
180ffaef18 Handle (delete) dbg intrinsics while promoting alloca.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58826 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-07 01:30:07 +00:00
Mon P Wang
e0b436ac41 Fixed scalarizing an extract subvector and prevent an infinite loop
when simplify a vector. 


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58820 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-06 22:52:21 +00:00
Devang Patel
33e868ff43 InstructionNamer preserves everything.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58787 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-06 01:00:16 +00:00
Devang Patel
b7c6bf1e07 Do now allow InlineAlways pass to remove dead functions.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58744 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-05 01:39:16 +00:00
Devang Patel
d91ac615e3 Check Attribute::NoInline.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58742 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-05 01:37:05 +00:00
Oscar Fuentes
436893d924 CMake: updated list of source files.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58736 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-05 00:11:22 +00:00
Dan Gohman
8f027c7fea Add a new pass to simplify specific half_powr function calls. This is
a specialized pass that it not likely to be generally useful.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58732 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-04 23:41:45 +00:00
Dale Johannesen
c0bc547c99 Allow SROA of vectors. Removing this caused a
huge performance regression in something we care
about.  This may not be final fix.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58718 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-04 20:54:03 +00:00
Devang Patel
8a88a14462 Fix unused variable warnings.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58651 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-03 23:14:09 +00:00
Devang Patel
265ca5dff5 Ignore conditions that are outside the loop.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58631 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-03 19:38:07 +00:00
Andrew Lenharth
06a1242207 add a period at the end of the comment, ignoring the fact that the comment would be hard pressed to be considered a sentence, but if it makes Bill happy...
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58630 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-03 19:29:29 +00:00
Devang Patel
58d43d4a41 Turn floating point IVs into integer IVs where possible.
This allows SCEV users to effectively calculate trip count.
LSR later on transforms back integer IVs to floating point IVs
later on to avoid int-to-float casts inside the loop.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58625 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-03 18:32:19 +00:00
Andrew Lenharth
97bd9a914b Ensure that we are checking only calls to the function we are interested in specializing
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58615 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-03 16:05:35 +00:00
Nick Lewycky
af38613e31 Replace explicit loop with utility function.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58593 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-03 03:49:14 +00:00
Nick Lewycky
6feb333695 Changes from Duncan's review:
* merge two weak functions by making them both alias a third non-weak fn
 * don't reimplement CallSite::hasArgument
 * whitelist the safe linkage types


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58568 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-02 16:46:26 +00:00
Duncan Sands
5baf8ece83 Get this building on 64 bit machines (error:
cast from ‘const llvm::PointerType*’ to ‘unsigned int’
loses precision).


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58561 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-02 09:00:33 +00:00
Oscar Fuentes
2b82b7e77e CMake: added a source file.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58559 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-02 06:01:39 +00:00
Nick Lewycky
579a024661 Add a new MergeFunctions pass. It finds identical functions and merges them.
This triggers only 60 times in llvm-test (look at .llvm.bc, not .linked.rbc)
and so it probably wont be turned on by default. Also, may of those are likely
to go away when PR2973 is fixed.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58557 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-02 05:52:50 +00:00
Nick Lewycky
8e39432ea9 Fix demanded bits analysis with srem by negative number. Based on a patch
by Richard Osborne.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58555 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-02 02:41:50 +00:00
Dan Gohman
21440acc06 Fix this recently moved code to use the correct type. CI is now a
ConstantInt, and SI is the original cast instruction. This fixes
PR2996.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58549 91177308-0d34-0410-b5e6-96231b3b80d8
2008-11-02 00:17:33 +00:00
Daniel Dunbar
0656466734 Fix warning.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58486 91177308-0d34-0410-b5e6-96231b3b80d8
2008-10-31 01:50:01 +00:00
Dan Gohman
1975d03183 Canonicalize sext(i1) to i1?-1:0, and update various instcombine
optimizations accordingly.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58457 91177308-0d34-0410-b5e6-96231b3b80d8
2008-10-30 20:40:10 +00:00
Daniel Dunbar
c5e1ec47c7 Add InlineCost class for represent the estimated cost of inlining a
function.
 - This explicitly models the costs for functions which should
   "always" or "never" be inlined. This fixes bugs where such costs
   were not previously respected.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58450 91177308-0d34-0410-b5e6-96231b3b80d8
2008-10-30 19:26:59 +00:00
Chris Lattner
f8131c99de Fix PR2967 by not deleting volatile load/stores that occur before unreachable.
I don't really see this as being needed, but there is little harm from doing
it.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58385 91177308-0d34-0410-b5e6-96231b3b80d8
2008-10-29 17:46:26 +00:00
Daniel Dunbar
1a99dbfe3b Factor shouldInline method out of Inliner.
- No functionality change.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58355 91177308-0d34-0410-b5e6-96231b3b80d8
2008-10-29 01:02:02 +00:00
Daniel Dunbar
f184c66e71 Assorted comment/naming fixes, 80-col violations, and reindentation.
- No functionality change.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58352 91177308-0d34-0410-b5e6-96231b3b80d8
2008-10-28 23:24:26 +00:00
Dan Gohman
b493b2774c (A & sext(C)) | (B & ~sext(C) -> C ? A : B
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58351 91177308-0d34-0410-b5e6-96231b3b80d8
2008-10-28 22:38:57 +00:00
Torok Edwin
3d54502304 export an ID for the instructionNamer, allowing analysis/transformation passes
that need it to require it by ID.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58238 91177308-0d34-0410-b5e6-96231b3b80d8
2008-10-27 10:16:27 +00:00
Chris Lattner
0fd77a579b Rewrite all the 'PromoteLocallyUsedAlloca[s]' logic. With the power of
LargeBlockInfo, we can now dramatically simplify their implementation
and speed them up at the same time.  Now the code has time proportional
to the number of uses of the alloca, not the size of the block.

This also eliminates code that tried to batch up different allocas which
are used in the same blocks, and eliminates the 'retry list' logic which
was baroque and no unneccesary.  In addition to being a speedup for crazy
cases, this is also a nice cleanup:

PromoteMemoryToRegister.cpp |  270 +++++++++++++++-----------------------------
 1 file changed, 96 insertions(+), 174 deletions(-)



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58229 91177308-0d34-0410-b5e6-96231b3b80d8
2008-10-27 07:05:53 +00:00
Chris Lattner
33210608be Add a new LargeBlockInfo helper, which is just a wrapper around
a trivial dense map.  Use this in RewriteSingleStoreAlloca to
avoid aggressively rescanning blocks over and over again.  This
fixes PR2925, speeding up mem2reg on the testcase in that bug
from 4.56s to 0.02s in a debug build on my machine.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58227 91177308-0d34-0410-b5e6-96231b3b80d8
2008-10-27 06:05:26 +00:00
Nick Lewycky
79cce5c51b Add value range analyzing of Add and Sub.
Understand that mul %x, 1 = %x.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58069 91177308-0d34-0410-b5e6-96231b3b80d8
2008-10-24 04:00:26 +00:00
Daniel Dunbar
394f0441e0 Change create*Pass factory functions to return Pass* instead of
LoopPass*.
 - Although less precise, this means they can be used in clients
   without RTTI (who would otherwise need to include LoopPass.h, which
   eventually includes things using dynamic_cast). This was the
   simplest solution that presented itself, but I am happy to use a
   better one if available.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@58010 91177308-0d34-0410-b5e6-96231b3b80d8
2008-10-22 23:32:42 +00:00
Dan Gohman
0db69dcc87 Use Function::getEntryBlock() instead of Function::front(), for clarity.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57870 91177308-0d34-0410-b5e6-96231b3b80d8
2008-10-21 03:10:28 +00:00
Dan Gohman
b9a31a15f9 Fix a bug that prevented llvm-extract -delete from working.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57864 91177308-0d34-0410-b5e6-96231b3b80d8
2008-10-21 01:08:07 +00:00
Dan Gohman
a1fcd77ccf Use 0 instead of false to return a null pointer.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57660 91177308-0d34-0410-b5e6-96231b3b80d8
2008-10-17 00:56:52 +00:00
Dan Gohman
2276a7bc8e Teach instcombine's visitLoad to scan back several instructions
to find opportunities for store-to-load forwarding or load CSE,
in the same way that visitStore scans back to do DSE. Also, define
a new helper function for testing whether the addresses of two
memory accesses are known to have the same value, and use it in
both visitStore and visitLoad.

These two changes allow instcombine to eliminate loads in code
produced by front-ends that frequently emit obviously redundant
addressing for memory references.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57608 91177308-0d34-0410-b5e6-96231b3b80d8
2008-10-15 23:19:35 +00:00
Evan Cheng
403006205f Combine (fcmp cc0 x, y) | (fcmp cc1 x, y) into a single fcmp when possible.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57515 91177308-0d34-0410-b5e6-96231b3b80d8
2008-10-14 18:44:08 +00:00
Evan Cheng
4990b257dd - Somehow I forgot about one / une.
- Renumber fcmp predicates to match their icmp counterparts.
- Try swapping operands to expose more optimization opportunities.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57513 91177308-0d34-0410-b5e6-96231b3b80d8
2008-10-14 18:13:38 +00:00
Evan Cheng
8db90725cb Optimize anding of two fcmp into a single fcmp if the operands are the same. e.g. uno && ueq -> ueq
ord && olt -> olt
     ord && ueq -> oeq


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57507 91177308-0d34-0410-b5e6-96231b3b80d8
2008-10-14 17:15:11 +00:00
Matthijs Kooijman
7e6d9b96ad Make InstructionCombining::getBitCastOperand() recognize GEP instructions and
constant expression with all zero indices as being the same as a bitcast.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@57442 91177308-0d34-0410-b5e6-96231b3b80d8
2008-10-13 15:17:01 +00:00