Commit Graph

451 Commits

Author SHA1 Message Date
Stuart Hastings
67f071e697 Avoid combining GEPs that might overflow at runtime.
rdar://problem/9267970

Patch by Julien Lerouge!


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131339 91177308-0d34-0410-b5e6-96231b3b80d8
2011-05-14 05:55:10 +00:00
Eli Friedman
b6e7cd655c PR9838: Fix transform introduced in r127064 to not trigger when only one side of the icmp is an exact shift.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130954 91177308-0d34-0410-b5e6-96231b3b80d8
2011-05-05 21:59:18 +00:00
Duncan Sands
00676a6df1 Remove unused variable.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130705 91177308-0d34-0410-b5e6-96231b3b80d8
2011-05-02 18:41:29 +00:00
Duncan Sands
f24ed77d24 Move some rem transforms out of instcombine and into instsimplify.
This automagically provides a transform noticed by my super-optimizer
as occurring quite often: "rem x, (select cond, x, 1)" -> 0.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130694 91177308-0d34-0410-b5e6-96231b3b80d8
2011-05-02 16:27:02 +00:00
Benjamin Kramer
7d6eb5a018 InstCombine: Turn (zext A) udiv (zext B) into (zext (A udiv B)). Same for urem or constant B.
This obviously helps a lot if the division would be turned into a libcall
(think i64 udiv on i386), but div is also one of the few remaining instructions
on modern CPUs that become more expensive when the bitwidth gets bigger.

This also helps register pressure on i386 when dividing chars, divb needs
two 8-bit parts of a 16 bit register as input where divl uses two registers.

int foo(unsigned char a) { return a/10; }
int bar(unsigned char a, unsigned char b) { return a/b; }

compiles into (x86_64)
_foo:
  imull $205, %edi, %eax
  shrl  $11, %eax
  ret
_bar:
  movzbl        %dil, %eax
  divb  %sil, %al
  movzbl        %al, %eax
  ret

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130615 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-30 18:16:07 +00:00
Benjamin Kramer
23b02cd031 Use SimplifyDemandedBits on div instructions.
This folds away silly stuff like (a&255)/1000 -> 0.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130614 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-30 18:16:00 +00:00
Benjamin Kramer
3224806109 Balance parentheses.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130489 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-29 08:41:23 +00:00
Benjamin Kramer
c2e31c1461 InstCombine: turn (C1 << A) << C2) into (C1 << C2) << A)
Fixes PR9809.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130485 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-29 08:15:41 +00:00
Benjamin Kramer
b0884f3f13 We require threse bits to be zero, too.
This shouldn't happen in practice because the icmp would be a constant.
Add a check so we don't miscompile code if something goes wrong.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130446 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-28 21:38:51 +00:00
Benjamin Kramer
e12d583621 Fix a comment.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130428 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-28 20:09:57 +00:00
Benjamin Kramer
264ac878b2 InstCombine: Merge "(trunc x) == C1 & (and x, CA) == C2" into a single and+icmp.
This happens when GVN widens loads. Part of PR6627.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130405 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-28 16:58:40 +00:00
Duncan Sands
089a246333 Stop trying to have instcombine preserve LCSSA form: this was not
effective in avoiding recomputation of LCSSA form; the widespread
use of instsimplify (which looks through phi nodes) means it was
not preserving LCSSA form anyway; and instcombine is no longer
scheduled in the middle of the loop passes so this doesn't matter
anymore.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130301 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-27 10:55:12 +00:00
Chris Lattner
325eeb1cd7 Transform: "icmp eq (trunc (lshr(X, cst1)), cst" to "icmp (and X, mask), cst"
when X has multiple uses.  This is useful for exposing secondary optimizations,
but the X86 backend isn't ready for this when X has a single use.  For example,
this can disable load folding.

This is inching towards resolving PR6627.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130238 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-26 20:18:20 +00:00
Chris Lattner
5036ce4a64 some random cleanups, no functionality change.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130237 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-26 20:02:45 +00:00
Frits van Bommel
86f72a8002 Rename a misleadingly-named variable.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@129644 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-16 14:32:34 +00:00
Jay Foad
4130278fbe Fix bug when checking phi operands in InstCombiner::visitPHINode(),
found by code inspection.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@129641 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-16 14:17:37 +00:00
Chris Lattner
7a2bdde0a0 Fix a ton of comment typos found by codespell. Patch by
Luis Felipe Strano Moraes!



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@129558 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-15 05:18:47 +00:00
Eli Friedman
3ec01b7dac Add an instcombine for constructs like a | -(b != c); a select is more
canonical, and generally leads to better code.  Found while looking at
an article about saturating arithmetic.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@129545 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-14 22:41:27 +00:00
Bill Wendling
f93f7b2446 Reapply r129401 with patch for clang.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@129419 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-13 00:36:11 +00:00
Bill Wendling
f9b2dc66c8 Revert r129401 for now. Clang is using the old way of doing things.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@129403 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-12 22:59:27 +00:00
Bill Wendling
d5f323d70b Remove the unaligned load intrinsics in favor of using native unaligned loads.
Now that we have a first-class way to represent unaligned loads, the unaligned
load intrinsics are superfluous.

First part of <rdar://problem/8460511>.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@129401 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-12 22:46:31 +00:00
Jay Foad
562b84b3ae Don't include Operator.h from InstrTypes.h.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@129271 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-11 09:35:34 +00:00
Nadav Rotem
0286ca89f0 InstCombine optimizes gep(bitcast(x)) even when the bitcasts casts away address
space info. We crash with an assert in this case. This change checks that the
address space of the bitcasted pointer is the same as the gep ptr.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@128884 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-05 14:29:52 +00:00
Benjamin Kramer
f5b75932b8 While SimplifyDemandedBits constant folds this, we can't rely on it here.
It's possible to craft an input that hits the recursion limits in a way
that SimplifyDemandedBits doesn't simplify the icmp but ComputeMaskedBits
can infer which bits are zero.

No test case as it depends on too many other things. Fixes PR9609.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@128777 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-02 18:50:58 +00:00
Benjamin Kramer
406a650237 Fix comment.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@128745 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-01 22:29:18 +00:00
Benjamin Kramer
5337fabdc4 Tweaks to the icmp+sext-to-shifts optimization to address Frits' comments:
- Localize the check if an icmp has one use to a place where we know we're
  introducing something that's likely more expensive than a sext from i1.
- Add an assert to make sure a case that would lead to a miscompilation is
  folded away earlier.
- Fix a typo.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@128744 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-01 22:22:11 +00:00
Benjamin Kramer
ce1498bdbe Fix build.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@128733 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-01 20:15:16 +00:00
Benjamin Kramer
0baa94a13b InstCombine: Turn icmp + sext into bitwise/integer ops when the input has only one unknown bit.
int test1(unsigned x) { return (x&8) ? 0 : -1; }
int test3(unsigned x) { return (x&8) ? -1 : 0; }

before (x86_64):
_test1:
	andl	$8, %edi
	cmpl	$1, %edi
	sbbl	%eax, %eax
	ret
_test3:
	andl	$8, %edi
	cmpl	$1, %edi
	sbbl	%eax, %eax
	notl	%eax
	ret

after:
_test1:
	shrl	$3, %edi
	andl	$1, %edi
	leal	-1(%rdi), %eax
	ret
_test3:
	shll	$28, %edi
	movl	%edi, %eax
	sarl	$31, %eax
	ret

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@128732 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-01 20:09:10 +00:00
Benjamin Kramer
0a30c42008 InstCombine: Move (sext icmp) transforms into their own method. No intended functionality change.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@128731 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-01 20:09:03 +00:00
Nadav Rotem
0ff8a4fa35 Instcombile optimization: extractelement(cast) -> cast(extractelement)
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@128683 91177308-0d34-0410-b5e6-96231b3b80d8
2011-03-31 22:57:29 +00:00
Benjamin Kramer
7ebdc37d2b InstCombine: APFloat can't perform arithmetic on PPC double doubles, don't even try.
Thanks Eli!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@128676 91177308-0d34-0410-b5e6-96231b3b80d8
2011-03-31 21:35:49 +00:00
Benjamin Kramer
00e00d693c InstCombine: Fix transform to use the swapped predicate.
Thanks Frits!

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@128628 91177308-0d34-0410-b5e6-96231b3b80d8
2011-03-31 10:46:03 +00:00
Benjamin Kramer
68b4bd0a63 InstCombine: fold fcmp (fneg x), (fneg y) -> fcmp x, y
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@128627 91177308-0d34-0410-b5e6-96231b3b80d8
2011-03-31 10:12:22 +00:00
Benjamin Kramer
0db50189dc InstCombine: fold fcmp pred (fneg x), C -> fcmp swap(pred) x, -C
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@128626 91177308-0d34-0410-b5e6-96231b3b80d8
2011-03-31 10:12:15 +00:00
Benjamin Kramer
b194bdc03b InstCombine: Shrink "fcmp (fpext x), C" to "fcmp x, C" if C can be losslessly converted to the type of x.
Fixes PR9592.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@128625 91177308-0d34-0410-b5e6-96231b3b80d8
2011-03-31 10:12:07 +00:00
Benjamin Kramer
cd0274ca18 InstCombine: fold fcmp (fpext x), (fpext y) -> fcmp x, y.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@128624 91177308-0d34-0410-b5e6-96231b3b80d8
2011-03-31 10:11:58 +00:00
Benjamin Kramer
546739656e InstCombine: If the divisor of an fdiv has an exact inverse, turn it into an fmul.
Fixes PR9587.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@128546 91177308-0d34-0410-b5e6-96231b3b80d8
2011-03-30 15:42:35 +00:00
Jay Foad
3ecfc861b4 Remove PHINode::reserveOperandSpace(). Instead, add a parameter to
PHINode::Create() giving the (known or expected) number of operands.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@128537 91177308-0d34-0410-b5e6-96231b3b80d8
2011-03-30 11:28:46 +00:00
Jay Foad
d8b4fb4aab (Almost) always call reserveOperandSpace() on newly created PHINodes.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@128535 91177308-0d34-0410-b5e6-96231b3b80d8
2011-03-30 11:19:20 +00:00
Benjamin Kramer
9822b869fc InstCombine: Add a few missing combines for ANDs and ORs of sign bit tests.
On x86 we now compile "if (a < 0 && b < 0)" into
	testl	%edi, %esi
	js	IF.THEN

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@128496 91177308-0d34-0410-b5e6-96231b3b80d8
2011-03-29 22:06:41 +00:00
Nick Lewycky
2bf026e375 Remove tabs I accidentally added.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@128413 91177308-0d34-0410-b5e6-96231b3b80d8
2011-03-28 17:48:26 +00:00
Jay Foad
fd4a5497b6 Make more use of PHINode::getNumIncomingValues().
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@128406 91177308-0d34-0410-b5e6-96231b3b80d8
2011-03-28 13:03:10 +00:00
Frits van Bommel
f56762a96d Add some debug output when -instcombine uses RAUW. This can make debug output for those cases much clearer since without this it only showed that the original instruction was removed, not what it was replaced with.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@128399 91177308-0d34-0410-b5e6-96231b3b80d8
2011-03-27 23:32:31 +00:00
Nick Lewycky
675619ca38 Teach the transformation that moves binary operators around selects to preserve
the subclass optional data.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@128388 91177308-0d34-0410-b5e6-96231b3b80d8
2011-03-27 19:51:23 +00:00
Benjamin Kramer
d655e6e9dc Use APInt's umul_ov instead of rolling our own overflow detection.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@128380 91177308-0d34-0410-b5e6-96231b3b80d8
2011-03-27 15:04:38 +00:00
Nick Lewycky
98cd75027c Add a small missed optimization: turn X == C ? X : Y into X == C ? C : Y. This
removes one use of X which helps it pass the many hasOneUse() checks.

In my analysis, this turns up very often where X = A >>exact B and that can't be
simplified unless X has one use (except by increasing the lifetime of A which is
generally a performance loss).


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@128373 91177308-0d34-0410-b5e6-96231b3b80d8
2011-03-27 07:30:57 +00:00
Devang Patel
813c9a0f19 Try to not lose variable's debug info during instcombine.
This is done by lowering dbg.declare intrinsic into dbg.value intrinsic.
Radar 9143931.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@127834 91177308-0d34-0410-b5e6-96231b3b80d8
2011-03-17 22:18:16 +00:00
Eric Christopher
b9b80c3268 If we don't know how long a string is we can't fold an _chk version to the
normal version.

Fixes rdar://9123638


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@127636 91177308-0d34-0410-b5e6-96231b3b80d8
2011-03-15 00:25:41 +00:00
Jin-Gu Kang
948d9e7ec7 This case is solved by Scalar Replacement of Aggregates (DT) and
Early CSE pass so this patch reverts it to original source code.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@127574 91177308-0d34-0410-b5e6-96231b3b80d8
2011-03-14 01:21:00 +00:00
Jin-Gu Kang
198e97c423 Add comment as following:
load and store reference same memory location, the memory location
is represented by getelementptr with two uses (load and store) and
the getelementptr's base is alloca with single use. At this point,
instructions from alloca to store can be removed.
(this pattern is generated when bitfield is accessed.)
For example,
%u = alloca %struct.test, align 4               ; [#uses=1]
%0 = getelementptr inbounds %struct.test* %u, i32 0, i32 0;[#uses=2]
%1 = load i8* %0, align 4                       ; [#uses=1]
%2 = and i8 %1, -16                             ; [#uses=1]
%3 = or i8 %2, 5                                ; [#uses=1]
store i8 %3, i8* %0, align 4


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@127565 91177308-0d34-0410-b5e6-96231b3b80d8
2011-03-13 14:05:51 +00:00