Commit Graph

199 Commits

Author SHA1 Message Date
Jakob Stoklund Olesen
41e2073f62 Don't PRE compares.
CodeGenPrepare sinks compare instructions down to their uses to prevent
live flags and predicate registers across basic blocks.

PRE of a compare instruction prevents that, forcing the i1 compare
result into a general purpose register.  That is usually more expensive
than the redundant compare PRE was trying to eliminate in the first
place.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@153657 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-29 17:22:39 +00:00
Duncan Sands
190e5a3314 Nick pointed out on IRC that GVN's propagateEquality wasn't propagating
equalities into phi node operands for which the equality is known to
hold in the incoming basic block.  That's because replaceAllDominatedUsesWith
wasn't handling phi nodes correctly in general (that this didn't give wrong
results was just luck: the specific way GVN uses replaceAllDominatedUsesWith
precluded wrong changes to phi nodes).


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@152006 91177308-0d34-0410-b5e6-96231b3b80d8
2012-03-04 13:25:19 +00:00
Duncan Sands
2b4f491045 Have GVN also do condition propagation when the right-hand side is not
a constant.  This fixes PR1768.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@151713 91177308-0d34-0410-b5e6-96231b3b80d8
2012-02-29 11:12:03 +00:00
Duncan Sands
669011f50b When performing a conditional branch depending on the value of a comparison
%cmp (eg: A==B) we already replace %cmp with "true" under the true edge, and
with "false" under the false edge.  This change enhances this to replace the
negated compare (A!=B) with "false" under the true edge and "true" under the
false edge.  Reported to improve perlbench results by 1%.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@151517 91177308-0d34-0410-b5e6-96231b3b80d8
2012-02-27 08:14:30 +00:00
Duncan Sands
e170c76ccd Teach GVN that x+y is the same as y+x and that x<y is the same as y>x.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@151365 91177308-0d34-0410-b5e6-96231b3b80d8
2012-02-24 15:16:31 +00:00
Eli Bendersky
0f0c411079 Replace all instances of dg.exp file with lit.local.cfg, since all tests are run with LIT now and now Dejagnu. dg.exp is no longer needed.
Patch reviewed by Daniel Dunbar. It will be followed by additional cleanup patches.




git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@150664 91177308-0d34-0410-b5e6-96231b3b80d8
2012-02-16 06:28:33 +00:00
Duncan Sands
33756f96d7 Reduce the number of dom queries made by GVN's conditional propagation
logic by half: isOnlyReachableViaThisEdge was trying to be clever and
handle the case of a branch to a basic block which is contained in a
loop.  This costs a domtree lookup and is completely useless due to
GVN's position in the pass pipeline: all loops have preheaders at this
point, which means it is enough for isOnlyReachableViaThisEdge to check
that Dst has only one predecessor.  (I checked this theoretical argument
by running over the entire nightly testsuite, and indeed it is so!).


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@149838 91177308-0d34-0410-b5e6-96231b3b80d8
2012-02-05 18:25:50 +00:00
Nick Lewycky
9f47fb6637 Fix crasher in GVN due to my recent capture tracking changes.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145047 91177308-0d34-0410-b5e6-96231b3b80d8
2011-11-21 19:42:56 +00:00
Nick Lewycky
88990248d3 Refactor capture tracking (which already had a couple flags for whether returns
and stores capture) to permit the caller to see each capture point and decide
whether to continue looking.

Use this inside memdep to do an analysis that basicaa won't do. This lets us
solve another devirtualization case, fixing PR8908!


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@144580 91177308-0d34-0410-b5e6-96231b3b80d8
2011-11-14 22:49:42 +00:00
Duncan Sands
3f329cb781 Teach GVN to also propagate switch cases. For example, in this code
switch (n) {
    case 27:
      do_something(x);
    ...
  }
the call do_something(x) will be replaced with do_something(27).  In
gcc-as-one-big-file this results in the removal of about 500 lines of
bitcode (about 0.02%), so has about 1/10 of the effect of propagating
branch conditions.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141360 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-07 08:29:06 +00:00
Duncan Sands
02b5e72ac6 GVN does simple propagation of conditions: when it sees a conditional
branch "br i1 %x, label %if_true, label %if_false" then it replaces
"%x" with "true" in places only reachable via the %if_true arm, and
with "false" in places only reachable via the %if_false arm.  Except
that actually it doesn't: if value numbering shows that %y is equal
to %x then, yes, %y will be turned into true/false in this way, but
any occurrences of %x itself are not transformed.  Fix this.  What's
more, it's often the case that %x is an equality comparison such as
"%x = icmp eq %A, 0", in which case every occurrence of %A that is
only reachable via the %if_true arm can be replaced with 0.  Implement
this and a few other variations on this theme.  This reduces the number
of lines of LLVM IR in "GCC as one big file" by 0.2%.  It has a bigger
impact on Ada code, typically reducing the number of lines of bitcode
by around 0.4% by removing repeated compiler generated checks.  Passes
the LLVM nightly testsuite and the Ada ACATS testsuite.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141177 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-05 14:28:49 +00:00
Duncan Sands
452c58f4c4 Generalize GVN's conditional propagation logic slightly:
it's OK for the false/true destination to have multiple
predecessors as long as the extra ones are dominated by
the branch destination.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141176 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-05 14:17:01 +00:00
Duncan Sands
73c8415d90 Mark the eh.typeid.for intrinsic as being 'const', which it is inside
any given function.  As pointed out by John McCall, this is needed to
have redundant eh.typeid.for tests be eliminated in the presence of
cleanups.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@139360 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-09 07:50:37 +00:00
Duncan Sands
758ba1f4ed When inlining exception handling code into another function, ensure that
duplicate tests are eliminated (for example if the two functions both have
a catch clause catching the same type, ensure the redundant one is removed).
Note that it would probably be safe to say that eh.typeid.for is 'const',
but since two calls to it with the same argument can give different results
(but only if the calls are in different functions), it seems more correct to
mark it only 'pure'; this doesn't get in the way of the optimization.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@139236 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-07 16:44:14 +00:00
Jakub Staszak
9469d5cf48 Extra CHECK-NOT to make sure that GVN transform works properly.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@139012 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-02 17:40:39 +00:00
Jakub Staszak
8cec759339 Compare type size instead of type _store_ size to make sure that BitCastInst
will be valid. This fixes PR10820.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@139005 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-02 14:57:37 +00:00
Bill Wendling
aeaa9a8b25 Update some tests to the new EH scheme.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@138925 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-01 00:58:03 +00:00
Eli Friedman
56efe24431 Atomic load/store handling for the passes using memdep (GVN, DSE, memcpyopt).
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@137888 91177308-0d34-0410-b5e6-96231b3b80d8
2011-08-17 22:22:24 +00:00
Lang Hames
f86a547419 Added test cases for GVN signed intrinsics recognition, r134777.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@134778 91177308-0d34-0410-b5e6-96231b3b80d8
2011-07-09 00:36:54 +00:00
Lang Hames
1fb0955cab Make GVN look through extractvalues for recognised intrinsics. GVN can then CSE ops that match values produced by the intrinsics.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@134677 91177308-0d34-0410-b5e6-96231b3b80d8
2011-07-08 01:50:54 +00:00
Jay Foad
5e27a1d468 This is an automatically reduced test case that crashed in GVN, at some
point during the development of the phi operand changes.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@133436 91177308-0d34-0410-b5e6-96231b3b80d8
2011-06-20 14:46:47 +00:00
Chris Lattner
b85e4eba85 rip out a ton of intrinsic modernization logic from AutoUpgrade.cpp, which is
for pre-2.9 bitcode files.  We keep x86 unaligned loads, movnt, crc32, and the
target indep prefetch change.

As usual, updating the testsuite is a PITA.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@133337 91177308-0d34-0410-b5e6-96231b3b80d8
2011-06-18 06:05:24 +00:00
Dan Gohman
f14d33d4cb Add a testcase to demonstrate the problem where phi translation is
ignored for clobbering partial-alias loads.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132633 91177308-0d34-0410-b5e6-96231b3b80d8
2011-06-04 07:05:05 +00:00
Dan Gohman
a3351a0e5d Disable the main feature of 130180, the elimination of loads that are
redundant with partially-aliasing loads.

When computing what portion of a clobbering load value is needed,
it doesn't consider phi-translation which may have occurred
between the clobbing load and the redundant load.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132631 91177308-0d34-0410-b5e6-96231b3b80d8
2011-06-04 06:48:50 +00:00
Eli Friedman
7c6a5a2c39 When marking a block as being unanalyzable, use "Clobber" on the terminator instead of the first instruction in the block. This is a bit of a hack; "Clobber" isn't really the right marking in the first place. memdep doesn't really have any way of properly expressing "unanalyzable" at the moment. Using it on the terminator is much less ambiguous than using it on an arbitrary instruction, though.
In the given testcase, the "Clobber" was pointing to a load, and GVN was incorrectly assuming that meant that the "Clobber" load overlapped the load being analyzed (when they are actually unrelated).

The included testcase tests both this commit and r132434.

Part two of rdar://9429882.  (r132434 was mislabeled.)



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132442 91177308-0d34-0410-b5e6-96231b3b80d8
2011-06-02 00:08:52 +00:00
Chris Lattner
9e7bc05ad7 fix PR9841 by having GVN not process dead loads. This was
causing it to get into infinite loops when it would widen a 
load (which can necessarily leave around dead loads).


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131847 91177308-0d34-0410-b5e6-96231b3b80d8
2011-05-22 07:03:34 +00:00
Chris Lattner
0e959394ad remove a trivial test, make some other tests less trivial.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131846 91177308-0d34-0410-b5e6-96231b3b80d8
2011-05-22 07:02:43 +00:00
Chris Lattner
208d92c143 make this test less trivial.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131845 91177308-0d34-0410-b5e6-96231b3b80d8
2011-05-22 06:59:33 +00:00
Chris Lattner
0a9e3d613b final step needed to resolve PR6627, which allows us to flatten the code down to
a nice and tidy:
  %x1 = load i32* %0, align 4
  %1 = icmp eq i32 %x1, 1179403647
  br i1 %1, label %if.then, label %if.end

instead of doing lots of loads and branches.  May the FreeBSD bootloader
long fit in its allocated space.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130416 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-28 18:15:47 +00:00
Chris Lattner
4034e14985 teach GVN to widen integer loads when they are overaligned, when doing an
wider load would allow elimination of subsequent loads, and when the wider
load is still a native integer type.  This eliminates a ton of loads on 
various benchmarks involving struct fields, though it is somewhat hobbled
by clang not being very aggressive about field alignment.

This is yet another step along the way towards resolving PR6627.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130390 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-28 07:29:08 +00:00
Chris Lattner
130131ea78 remove support for llvm.invariant.end from memdep. It is a
work-in-progress that is not progressing, and it has issues.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130247 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-26 21:50:51 +00:00
Chris Lattner
9750acc079 Improve the bail-out predicate to really only kick in when phi
translation fails.  We were bailing out in some cases that would
cause us to miss GVN'ing some non-local cases away.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130206 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-26 17:41:02 +00:00
Chris Lattner
1f821512fc Enhance MemDep: When alias analysis returns a partial alias result,
return it as a clobber.  This allows GVN to do smart things.

Enhance GVN to be smart about the case when a small load is clobbered
by a larger overlapping load.  In this case, forward the value.  This
allows us to compile stuff like this:

int test(void *P) {
  int tmp = *(unsigned int*)P;
  return tmp+*((unsigned char*)P+1);
}

into:

_test:                                  ## @test
	movl	(%rdi), %ecx
	movzbl	%ch, %eax
	addl	%ecx, %eax
	ret

which has one load.  We already handled the case where the smaller
load was from a must-aliased base pointer.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@130180 91177308-0d34-0410-b5e6-96231b3b80d8
2011-04-26 01:21:15 +00:00
Owen Anderson
f056838956 Give GVN back the ability to perform simple conditional propagation on conditional branch values.
I still think that LVI should be handling this, but that capability is some ways off in the future,
and this matters for some significant benchmarks.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@122378 91177308-0d34-0410-b5e6-96231b3b80d8
2010-12-21 23:54:34 +00:00
Dan Gohman
f4177aa019 Preserve TBAA tags when doing load PRE.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@121921 91177308-0d34-0410-b5e6-96231b3b80d8
2010-12-15 23:53:55 +00:00
Owen Anderson
a04a0649e1 Completely rework the datastructure GVN uses to represent the value number to leader mapping. Previously,
this was a tree of hashtables, and a query recursed into the table for the immediate dominator ad infinitum
if the initial lookup failed.  This led to really bad performance on tall, narrow CFGs.

We can instead replace it with what is conceptually a multimap of value numbers to leaders (actually
represented by a hashtable with a list of Value*'s as the value type), and then
determine which leader from that set to use very cheaply thanks to the DFS numberings maintained by
DominatorTree.  Because there are typically few duplicates of a given value, this scan tends to be
quite fast.  Additionally, we use a custom linked list and BumpPtr allocation to avoid any unnecessary
allocation in representing the value-side of the multimap.

This change brings with it a 15% (!) improvement in the total running time of GVN on 403.gcc, which I
think is pretty good considering that includes all the "real work" being done by MemDep as well.

The one downside to this approach is that we can no longer use GVN to perform simple conditional progation,
but that seems like an acceptable loss since we now have LVI and CorrelatedValuePropagation to pick up
the slack.  If you see conditional propagation that's not happening, please file bugs against LVI or CVP.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@119714 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-18 18:32:40 +00:00
Dan Gohman
ce56262211 Add support for PHI-translating sext, zext, and trunc instructions,
enabling more PRE. PR8586.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@119704 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-18 17:05:13 +00:00
Duncan Sands
cd6636c737 Teach InstructionSimplify about phi nodes. I chose to have it simply
offload the work to hasConstantValue rather than do something more
complicated (such handling mutually recursive phis) because (1) it is
not clear it is worth it; and (2) if it is worth it, maybe such logic
would be better placed in hasConstantValue.  Adjust some GVN tests
which are now cleaned up much further (eg: all phi nodes are removed).


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@119043 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-14 13:30:18 +00:00
Duncan Sands
dd609e912b Testcase to go along with commit 118923 ("Have GVN simplify instructions
as it goes").  Before -std-compile-opts only got it down to
  %a = tail call i32 @foo(i32 0) readnone
  %x = tail call i32 @foo(i32 %a) readnone
  %y = tail call i32 @foo(i32 %a) readnone
  %z = icmp eq i32 %x, %y
  ret i1 %z
while now -basicaa -gvn alone reduce it to
  %a = call i32 @foo(i32 0) readnone
  %x = call i32 @foo(i32 %a) readnone
  ret i1 true



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@119009 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-13 21:33:19 +00:00
Dan Gohman
075fb5d68f Enhance GVN to do more precise alias queries for non-local memory
references. For example, this allows gvn to eliminate the load in
this example:

  void foo(int n, int* p, int *q) {
    p[0] = 0;
    p[1] = 1;
    if (n) {
      *q = p[0];
    }
  }


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@118714 91177308-0d34-0410-b5e6-96231b3b80d8
2010-11-10 20:37:15 +00:00
Dan Gohman
c1be92f3bb Make BasicAliasAnalysis a normal AliasAnalysis implementation which
does normal initialization and normal chaining. Change the default
AliasAnalysis implementation to NoAlias.

Update StandardCompileOpts.h and friends to explicitly request
BasicAliasAnalysis.

Update tests to explicitly request -basicaa.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@116720 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-18 18:04:47 +00:00
Owen Anderson
7267e14327 Now that the profitable bits of EnableFullLoadPRE have been enabled by default, rip out the remainder.
Anyone interested in more general PRE would be better served by implementing it separately, to get real
anticipation calculation, etc.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@115337 91177308-0d34-0410-b5e6-96231b3b80d8
2010-10-01 20:02:55 +00:00
Owen Anderson
722cc1f414 We do want to allow LoadPRE to perform LICM-like transformations: we already consider PHI nodes to be negligible for
code size (making this transform code size neutral), and it allows us to hoist values out of loops, which is always
a good thing.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@115205 91177308-0d34-0410-b5e6-96231b3b80d8
2010-09-30 20:53:04 +00:00
Owen Anderson
b0ba0f4170 LoadPRE was not properly checking that the load it was PRE'ing post-dominated the block it was being hoisted to.
Splitting critical edges at the merge point only addressed part of the issue; it is also possible for non-post-domination
to occur when the path from the load to the merge has branches in it.  Unfortunately, full anticipation analysis is
time-consuming, so for now approximate it.  This is strictly more conservative than real anticipation, so we will miss
some cases that real PRE would allow, but we also no longer insert loads into paths where they didn't exist before. :-)

This is a very slight net positive on SPEC for me (0.5% on average).  Most of the benchmarks are largely unaffected, but
when it pays off it pays off decently: 181.mcf improves by 4.5% on my machine.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@114785 91177308-0d34-0410-b5e6-96231b3b80d8
2010-09-25 05:26:18 +00:00
Duncan Sands
cdd4f8c7cb Correct bogus module triple specifications.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@112469 91177308-0d34-0410-b5e6-96231b3b80d8
2010-08-30 10:48:29 +00:00
Rafael Espindola
1e81966626 Remove arm_apcscc from the test files. It is the default and doing this
matches what llvm-gcc and clang now produce.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@106221 91177308-0d34-0410-b5e6-96231b3b80d8
2010-06-17 15:18:27 +00:00
Chris Lattner
7944c21cae Fix PR7052, patch by Jakub Staszak!
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@103347 91177308-0d34-0410-b5e6-96231b3b80d8
2010-05-08 20:01:44 +00:00
Nick Lewycky
8e13af309b Fix intrinsic signature in this test.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101674 91177308-0d34-0410-b5e6-96231b3b80d8
2010-04-17 21:12:55 +00:00
Bob Wilson
84bd6b0c31 Re-commit my previous SSAUpdater changes. The previous version naively tried
to determine where to place PHIs by iteratively comparing reaching definitions
at each block.  That was just plain wrong.  This version now computes the
dominator tree within the subset of the CFG where PHIs may need to be placed,
and then places the PHIs in the iterated dominance frontier of each definition.
The rest of the patch is mostly the same, with a few more performance
improvements added in.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@101612 91177308-0d34-0410-b5e6-96231b3b80d8
2010-04-17 03:08:24 +00:00
Chris Lattner
d2075586c8 add newlines at the end of files.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@100705 91177308-0d34-0410-b5e6-96231b3b80d8
2010-04-07 22:53:17 +00:00