Commit Graph

926 Commits

Author SHA1 Message Date
Matt Arsenault
4598bd53ab Use type helper functions.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@190113 91177308-0d34-0410-b5e6-96231b3b80d8
2013-09-06 00:37:24 +00:00
Matt Arsenault
596aa123f4 Consistently use dbgs() in debug printing
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@190093 91177308-0d34-0410-b5e6-96231b3b80d8
2013-09-05 19:48:28 +00:00
Tim Northover
0415b1810b InstCombine: allow unmasked icmps to be combined with logical ops
"(icmp op i8 A, B)" is equivalent to "(icmp op i8 (A & 0xff), B)" as a
degenerate case. Allowing this as a "masked" comparison when analysing "(icmp)
&/| (icmp)" allows us to combine them in more cases.

rdar://problem/7625728

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@189931 91177308-0d34-0410-b5e6-96231b3b80d8
2013-09-04 11:57:17 +00:00
Tim Northover
7bfabdac4e InstCombine: look for masked compares with subset relation
Even in cases which aren't universally optimisable like "(A & B) != 0 && (A &
C) != 0", the masks can make one of the comparisons completely redundant. In
this case, since we've gone to the effort of spotting masked comparisons we
should combine them.

rdar://problem/7625728

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@189930 91177308-0d34-0410-b5e6-96231b3b80d8
2013-09-04 11:57:13 +00:00
Matt Arsenault
330943afb7 Teach InstCombineLoadCast about address spaces.
This is another one that doesn't matter much,
but uses the right GEP index types in the first
place.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@189854 91177308-0d34-0410-b5e6-96231b3b80d8
2013-09-03 21:05:48 +00:00
Matt Arsenault
aa31d35dc7 Use type form of getIntPtrType in alloca visitor.
This doesn't actually matter, since alloca is always
0 address space, but this is more consistent.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@189853 91177308-0d34-0410-b5e6-96231b3b80d8
2013-09-03 21:05:15 +00:00
Benjamin Kramer
a8517ee732 InstCombine: Check for zero shift amounts before subtracting one causing integer overflow.
PR17026. Also avoid undefined shifts and shift amounts larger than 64 bits
(those are always undef because we can't represent integer types that large).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@189672 91177308-0d34-0410-b5e6-96231b3b80d8
2013-08-30 14:35:35 +00:00
Matt Arsenault
b70d79e7e9 Fix typo.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@189524 91177308-0d34-0410-b5e6-96231b3b80d8
2013-08-28 22:17:26 +00:00
Matt Arsenault
52c7d8e4eb Teach InstCombine about address spaces
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@188926 91177308-0d34-0410-b5e6-96231b3b80d8
2013-08-21 19:53:10 +00:00
Jakub Staszak
c2d722efbf Use pop_back_val() instead of both back() and pop_back().
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@188723 91177308-0d34-0410-b5e6-96231b3b80d8
2013-08-19 22:47:55 +00:00
Matt Arsenault
8e3367ea36 Teach InstCombine visitGetElementPtr about address spaces
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@188721 91177308-0d34-0410-b5e6-96231b3b80d8
2013-08-19 22:17:40 +00:00
Matt Arsenault
c4ad982f0b Cleanup visitGetElementPtr to make address space change easier
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@188720 91177308-0d34-0410-b5e6-96231b3b80d8
2013-08-19 22:17:34 +00:00
Matt Arsenault
5c40cc2e1e commonPointerCast cleanups to make address space change easier
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@188719 91177308-0d34-0410-b5e6-96231b3b80d8
2013-08-19 22:17:18 +00:00
Matt Arsenault
89062b8387 Revert non-test parts of r188507
Re-add the inboundsless tests I didn't add originally

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@188710 91177308-0d34-0410-b5e6-96231b3b80d8
2013-08-19 21:40:31 +00:00
Jim Grosbach
bff3c587f6 InstCombine: Use isAllOnesValue() instead of explicit -1.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@188563 91177308-0d34-0410-b5e6-96231b3b80d8
2013-08-16 17:03:36 +00:00
Jim Grosbach
4658ce9c1f InstCombine: Simplify if(x!=0 && x!=-1).
When both constants are positive or both constants are negative,
InstCombine already simplifies comparisons like this, but when
it's exactly zero and -1, the operand sorting ends up reversed
and the pattern fails to match. Handle that special case.

Follow up for rdar://14689217

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@188512 91177308-0d34-0410-b5e6-96231b3b80d8
2013-08-16 00:15:20 +00:00
Matt Arsenault
a630cb032c Don't do FoldCmpLoadFromIndexedGlobal for non inbounds GEPs
This path wasn't tested before without a datalayout,
so add some more tests and re-run with and without one.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@188507 91177308-0d34-0410-b5e6-96231b3b80d8
2013-08-15 23:11:07 +00:00
Matt Arsenault
fdc2660214 Fix always creating GEP with i32 indices
Use the pointer size if datalayout is available.
Use i64 if it's not, which is consistent with what other
places do when the pointer size is unknown.

The test doesn't really test this in a useful way
since it will be transformed to that later anyway,
but this now tests it for non-zero arrays and when
datalayout isn't available. The cases in
visitGetElementPtrInst should save an extra re-visit to
the newly created GEP since it won't need to cleanup after
itself.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@188339 91177308-0d34-0410-b5e6-96231b3b80d8
2013-08-14 00:24:38 +00:00
Matt Arsenault
3ea117e1bc Use type helper functions instead of cast
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@188338 91177308-0d34-0410-b5e6-96231b3b80d8
2013-08-14 00:24:34 +00:00
Matt Arsenault
c2a484164d Use array initializer, space around operator
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@188337 91177308-0d34-0410-b5e6-96231b3b80d8
2013-08-14 00:24:05 +00:00
Richard Sandiford
23331c30ae Fix big-endian handling of integer-to-vector bitcasts in InstCombine
These functions used to assume that the lsb of an integer corresponds
to vector element 0, whereas for big-endian it's the other way around:
the msb is in the first element and the lsb is in the last element.

Fixes MultiSource/Benchmarks/mediabench/gsm/toast for z.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@188155 91177308-0d34-0410-b5e6-96231b3b80d8
2013-08-12 07:26:09 +00:00
Matt Arsenault
00d7baad90 Fix missing -*- C++ -*-s
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187758 91177308-0d34-0410-b5e6-96231b3b80d8
2013-08-06 00:16:21 +00:00
Owen Anderson
605b3427a9 Preserve fast-math flags when folding (fsub x, (fneg y)) to (fadd x, y).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187462 91177308-0d34-0410-b5e6-96231b3b80d8
2013-07-30 23:53:17 +00:00
Matt Arsenault
f34dc428fa Change behavior of calling bitcasted alias functions.
It will now only convert the arguments / return value and call
the underlying function if the types are able to be bitcasted.
This avoids using fp<->int conversions that would occur before.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187444 91177308-0d34-0410-b5e6-96231b3b80d8
2013-07-30 20:45:05 +00:00
Owen Anderson
107c578126 Fix variable name.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187253 91177308-0d34-0410-b5e6-96231b3b80d8
2013-07-26 22:06:21 +00:00
Owen Anderson
0c326f07ca When InstCombine tries to fold away (fsub x, (fneg y)) into (fadd x, y), it is
also worthwhile for it to look through FP extensions and truncations, whose
application commutes with fneg.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187249 91177308-0d34-0410-b5e6-96231b3b80d8
2013-07-26 21:40:29 +00:00
Stephen Lin
3b6bb79578 Correct case of m_UIToFp to m_UIToFP to match instruction name, add m_SIToFP for consistency.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@187225 91177308-0d34-0410-b5e6-96231b3b80d8
2013-07-26 17:55:00 +00:00
Stephen Lin
a98ce503b9 InstCombine: call FoldOpIntoSelect for all floating binops, not just fmul
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@186759 91177308-0d34-0410-b5e6-96231b3b80d8
2013-07-20 07:13:13 +00:00
Stephen Lin
54bf58a032 Restore r181216, which was partially reverted in r182499.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@186533 91177308-0d34-0410-b5e6-96231b3b80d8
2013-07-17 20:06:03 +00:00
Craig Topper
b9df53a40b Use llvm::array_lengthof to replace sizeof(array)/sizeof(array[0]).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@186301 91177308-0d34-0410-b5e6-96231b3b80d8
2013-07-15 04:27:47 +00:00
Craig Topper
a0ec3f9b7b Use SmallVectorImpl& instead of SmallVector to avoid repeating small vector size.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@186274 91177308-0d34-0410-b5e6-96231b3b80d8
2013-07-14 04:42:23 +00:00
Nick Lewycky
75681bb302 Add a microoptimization for urem.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@186235 91177308-0d34-0410-b5e6-96231b3b80d8
2013-07-13 01:16:47 +00:00
Joey Gouly
ebe1147722 Fix a crash in EvaluateInDifferentElementOrder where it would generate an
undef vector of the wrong type.

LGTM'd by Nick Lewycky on IRC.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@186224 91177308-0d34-0410-b5e6-96231b3b80d8
2013-07-12 23:08:06 +00:00
Benjamin Kramer
0a230e0d98 Don't use a potentially expensive shift if all we want is one set bit.
No functionality change.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@186095 91177308-0d34-0410-b5e6-96231b3b80d8
2013-07-11 16:05:50 +00:00
David Majnemer
fcb7b97892 InstCombine: Fix typo in comment for visitICmpInstWithInstAndIntCst
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185916 91177308-0d34-0410-b5e6-96231b3b80d8
2013-07-09 09:24:35 +00:00
David Majnemer
fecf0d7a01 InstCombine: variations on 0xffffffff - x >= 4
The following transforms are valid if -C is a power of 2:
(icmp ugt (xor X, C), ~C) -> (icmp ult X, C)
(icmp ult (xor X, C), -C) -> (icmp uge X, C)

These are nice, they get rid of the xor.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185915 91177308-0d34-0410-b5e6-96231b3b80d8
2013-07-09 09:20:58 +00:00
David Majnemer
36b6f7409d InstCombine: X & -C != -C -> X <= u ~C
Tests were added in r185910 somehow.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185912 91177308-0d34-0410-b5e6-96231b3b80d8
2013-07-09 08:09:32 +00:00
David Majnemer
11c29bafd5 Commit r185909 was a misapplied patch, fix it
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185910 91177308-0d34-0410-b5e6-96231b3b80d8
2013-07-09 07:58:32 +00:00
David Majnemer
377a5c1a87 InstCombine: add more transforms
C1-X <u C2 -> (X|(C2-1)) == C1
C1-X >u C2 -> (X|C2) == C1
X-C1 <u C2 -> (X & -C2) == C1
X-C1 >u C2 -> (X & ~C2) == C1


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185909 91177308-0d34-0410-b5e6-96231b3b80d8
2013-07-09 07:50:59 +00:00
David Majnemer
53fc39992d InstCombine: Fold X-C1 <u 2 -> (X & -2) == C1
Back in r179493 we determined that two transforms collided with each
other.  The fix back then was to reorder the transforms so that the
preferred transform would give it a try and then we would try the
secondary transform.  However, it was noted that the best approach would
canonicalize one transform into the other, removing the collision and
allowing us to optimize IR given to us in that form.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185808 91177308-0d34-0410-b5e6-96231b3b80d8
2013-07-08 11:53:08 +00:00
David Majnemer
b1b738e036 InstCombine: (icmp eq B, 0) | (icmp ult A, B) -> (icmp ule A, B-1)
This transform allows us to turn IR that looks like:
  %1 = icmp eq i64 %b, 0
  %2 = icmp ult i64 %a, %b
  %3 = or i1 %1, %2
  ret i1 %3

into:
  %0 = add i64 %b, -1
  %1 = icmp uge i64 %0, %a
  ret i1 %1

which means we go from lowering:
        cmpq    %rsi, %rdi
        setb    %cl
        testq   %rsi, %rsi
        sete    %al
        orb     %cl, %al
        ret

to lowering:
        decq    %rsi
        cmpq    %rdi, %rsi
        setae   %al
        ret


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185677 91177308-0d34-0410-b5e6-96231b3b80d8
2013-07-05 00:31:17 +00:00
David Majnemer
e7006bb04b InstCombine: Reimplementation of visitUDivOperand
This transform was originally added in r185257 but later removed in
r185415.  The original transform would create instructions speculatively
and then discard them if the speculation was proved incorrect.  This has
been replaced with a scheme that splits the transform into two parts:
preflight and fold.  While we preflight, we build up fold actions that
inform the folding stage on how to act.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185667 91177308-0d34-0410-b5e6-96231b3b80d8
2013-07-04 21:17:49 +00:00
Craig Topper
6227d5c690 Use SmallVectorImpl::iterator/const_iterator instead of SmallVector to avoid specifying the vector size.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185606 91177308-0d34-0410-b5e6-96231b3b80d8
2013-07-04 01:31:24 +00:00
Hal Finkel
b19dd2bcaf Revert r185257 (InstCombine: Be more agressive optimizing 'udiv' instrs with 'select' denoms)
I'm reverting this commit because:

 1. As discussed during review, it needs to be rewritten (to avoid creating and
then deleting instructions).

 2. This is causing optimizer crashes. Specifically, I'm seeing things like
this:

    While deleting: i1 %
    Use still stuck around after Def is destroyed:  <badref> = select i1 <badref>, i32 0, i32 1
    opt: /src/llvm-trunk/lib/IR/Value.cpp:79: virtual llvm::Value::~Value(): Assertion `use_empty() && "Uses remain when a value is destroyed!"' failed.

   I'd guess that these will go away once we're no longer creating/deleting
instructions here, but just in case, I'm adding a regression test.

Because the code is bring rewritten, I've just XFAIL'd the original regression test. Original commit message:

	InstCombine: Be more agressive optimizing 'udiv' instrs with 'select' denoms

	Real world code sometimes has the denominator of a 'udiv' be a
	'select'.  LLVM can handle such cases but only when the 'select'
	operands are symmetric in structure (both select operands are a constant
	power of two or a left shift, etc.).  This falls apart if we are dealt a
	'udiv' where the code is not symetric or if the select operands lead us
	to more select instructions.

	Instead, we should treat the LHS and each select operand as a distinct
	divide operation and try to optimize them independently.  If we can
	to simplify each operation, then we can replace the 'udiv' with, say, a
	'lshr' that has a new select with a bunch of new operands for the
	select.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185415 91177308-0d34-0410-b5e6-96231b3b80d8
2013-07-02 05:21:11 +00:00
Benjamin Kramer
edac9151fd InstCombine: Also turn selects fed by an and into arithmetic when the types don't match.
Inserting a zext or trunc is sufficient. This pattern is somewhat common in
LLVM's pointer mangling code.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185270 91177308-0d34-0410-b5e6-96231b3b80d8
2013-06-29 21:17:04 +00:00
David Majnemer
c22a4eeec6 InstCombine: FoldGEPICmp shouldn't change sign of base pointer comparison
Changing the sign when comparing the base pointer would introduce all
sorts of unexpected things like:
  %gep.i = getelementptr inbounds [1 x i8]* %a, i32 0, i32 0
  %gep2.i = getelementptr inbounds [1 x i8]* %b, i32 0, i32 0
  %cmp.i = icmp ult i8* %gep.i, %gep2.i
  %cmp.i1 = icmp ult [1 x i8]* %a, %b
  %cmp = icmp ne i1 %cmp.i, %cmp.i1
  ret i1 %cmp

into:
  %cmp.i = icmp slt [1 x i8]* %a, %b
  %cmp.i1 = icmp ult [1 x i8]* %a, %b
  %cmp = xor i1 %cmp.i, %cmp.i1
  ret i1 %cmp

By preserving the original sign, we now get:
  ret i1 false

This fixes PR16483.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185259 91177308-0d34-0410-b5e6-96231b3b80d8
2013-06-29 10:28:04 +00:00
David Majnemer
df703257c1 InstCombine: Small whitespace cleanup in FoldGEPICmp
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185258 91177308-0d34-0410-b5e6-96231b3b80d8
2013-06-29 09:45:35 +00:00
David Majnemer
f723e5d1c2 InstCombine: Be more agressive optimizing 'udiv' instrs with 'select' denoms
Real world code sometimes has the denominator of a 'udiv' be a
'select'.  LLVM can handle such cases but only when the 'select'
operands are symmetric in structure (both select operands are a constant
power of two or a left shift, etc.).  This falls apart if we are dealt a
'udiv' where the code is not symetric or if the select operands lead us
to more select instructions.

Instead, we should treat the LHS and each select operand as a distinct
divide operation and try to optimize them independently.  If we can
to simplify each operation, then we can replace the 'udiv' with, say, a
'lshr' that has a new select with a bunch of new operands for the
select.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185257 91177308-0d34-0410-b5e6-96231b3b80d8
2013-06-29 08:40:07 +00:00
David Majnemer
b41f4bbfbd InstCombine: Optimize (1 << X) Pred CstP2 to X Pred Log2(CstP2)
We may, after other optimizations, find ourselves with IR that looks
like:

  %shl = shl i32 1, %y
  %cmp = icmp ult i32 %shl, 32

Instead, we should just compare the shift count:

  %cmp = icmp ult i32 %y, 5


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185242 91177308-0d34-0410-b5e6-96231b3b80d8
2013-06-28 23:42:03 +00:00
Matt Arsenault
1bde770ccd Fix using arg_end() - arg_begin() instead of arg_size()
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@185121 91177308-0d34-0410-b5e6-96231b3b80d8
2013-06-28 00:25:40 +00:00