Commit Graph

523 Commits

Author SHA1 Message Date
Nick Lewycky
b48a18903a Change CaptureTracking to pass a Use* instead of a Value* when a value is
captured. This allows the tracker to look at the specific use, which may be
especially interesting for function calls.

Use this to fix 'nocapture' deduction in FunctionAttrs. The existing one does
not iterate until a fixpoint and does not guarantee that it produces the same
result regardless of iteration order. The new implementation builds up a graph
of how arguments are passed from function to function, and uses a bottom-up walk
on the argument-SCCs to assign nocapture. This gets us nocapture more often, and
does so rather efficiently and independent of iteration order.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@147327 91177308-0d34-0410-b5e6-96231b3b80d8
2011-12-28 23:24:21 +00:00
Chandler Carruth
51f40a725b Make the unreachable probability much much heavier. The previous
probability wouldn't be considered "hot" in some weird loop structures
or other compounding probability patterns. This makes it much harder to
confuse, but isn't really a principled fix. I'd actually like it if we
could model a zero probability, as it would make this much easier to
reason about. Suggestions for how to do this better are welcome.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@147142 91177308-0d34-0410-b5e6-96231b3b80d8
2011-12-22 09:26:37 +00:00
Chandler Carruth
ddbc274169 Manually upgrade the test suite to specify the flag to cttz and ctlz.
I followed three heuristics for deciding whether to set 'true' or
'false':

- Everything target independent got 'true' as that is the expected
  common output of the GCC builtins.
- If the target arch only has one way of implementing this operation,
  set the flag in the way that exercises the most of codegen. For most
  architectures this is also the likely path from a GCC builtin, with
  'true' being set. It will (eventually) require lowering away that
  difference, and then lowering to the architecture's operation.
- Otherwise, set the flag differently dependending on which target
  operation should be tested.

Let me know if anyone has any issue with this pattern or would like
specific tests of another form. This should allow the x86 codegen to
just iteratively improve as I teach the backend how to differentiate
between the two forms, and everything else should remain exactly the
same.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@146370 91177308-0d34-0410-b5e6-96231b3b80d8
2011-12-12 11:59:10 +00:00
Andrew Trick
ecb35ece5c SCEV fix. In general, Add/Mul expressions should not inherit NSW/NUW.
This reverts r139450, fixes r139453, and adds much needed comments and a
unit test.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145367 91177308-0d34-0410-b5e6-96231b3b80d8
2011-11-29 02:16:38 +00:00
Andrew Trick
d2b5e2dd44 Filecheckize.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145363 91177308-0d34-0410-b5e6-96231b3b80d8
2011-11-29 02:05:23 +00:00
Chris Lattner
d2bf432b2b Upgrade syntax of tests using volatile instructions to use 'load volatile' instead of 'volatile load', which is archaic.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@145171 91177308-0d34-0410-b5e6-96231b3b80d8
2011-11-27 06:54:59 +00:00
Nick Lewycky
89d093d5b6 Don't forget to check FlagNW when determining whether an AddRecExpr will wrap
or not. Patch by Brendon Cahoon!


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@144173 91177308-0d34-0410-b5e6-96231b3b80d8
2011-11-09 07:11:37 +00:00
Benjamin Kramer
588d84c4df 2>&1 doesn't work here, it just creates an empty file called "&1"
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@143117 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-27 18:27:45 +00:00
Duncan Sands
f8a9eb1fa6 Restore commits 142790 and 142843 - they weren't breaking the build
bots.  Original commit messages:
- Reapply r142781 with fix. Original message:

  Enhance SCEV's brute force loop analysis to handle multiple PHI nodes in the
  loop header when computing the trip count.

  With this, we now constant evaluate:
    struct ListNode { const struct ListNode *next; int i; };
    static const struct ListNode node1 = {0, 1};
    static const struct ListNode node2 = {&node1, 2};
    static const struct ListNode node3 = {&node2, 3};
    int test() {
      int sum = 0;
      for (const struct ListNode *n = &node3; n != 0; n = n->next)
        sum += n->i;
      return sum;
    }

- Now that we look at all the header PHIs, we need to consider all the header PHIs
  when deciding that the loop has stopped evolving. Fixes miscompile in the gcc
  torture testsuite!



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142919 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-25 12:28:52 +00:00
Chandler Carruth
45baf6bb85 Fix the API usage in loop probability heuristics. It was incorrectly
classifying many edges as exiting which were in fact not. These mainly
formed edges into sub-loops. It was also not correctly classifying all
returning edges out of loops as leaving the loop. With this match most
of the loop heuristics are more rational.

Several serious regressions on loop-intesive benchmarks like perlbench's
loop tests when built with -enable-block-placement are fixed by these
updated heuristics. Unfortunately they in turn uncover some other
regressions. There are still several improvemenst that should be made to
loop heuristics including trip-count, and early back-edge management.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142917 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-25 09:47:41 +00:00
Duncan Sands
526c80bae4 Speculatively revert commits 142790 and 142843 to see if it fixes
the dragonegg and llvm-gcc self-host buildbots.  Original commit
messages:
- Reapply r142781 with fix. Original message:

  Enhance SCEV's brute force loop analysis to handle multiple PHI nodes in the
  loop header when computing the trip count.

  With this, we now constant evaluate:
    struct ListNode { const struct ListNode *next; int i; };
    static const struct ListNode node1 = {0, 1};
    static const struct ListNode node2 = {&node1, 2};
    static const struct ListNode node3 = {&node2, 3};
    int test() {
      int sum = 0;
      for (const struct ListNode *n = &node3; n != 0; n = n->next)
        sum += n->i;
      return sum;
    }

- Now that we look at all the header PHIs, we need to consider all the header PHIs
when deciding that the loop has stopped evolving. Fixes miscompile in the gcc
torture testsuite!



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142916 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-25 09:26:43 +00:00
Nick Lewycky
3846163aee Now that we look at all the header PHIs, we need to consider all the header PHIs
when deciding that the loop has stopped evolving. Fixes miscompile in the gcc
torture testsuite!


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142843 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-24 21:02:38 +00:00
Chandler Carruth
de1c9bb450 Remove return heuristics from the static branch probabilities, and
introduce no-return or unreachable heuristics.

The return heuristics from the Ball and Larus paper don't work well in
practice as they pessimize early return paths. The only good hitrate
return heuristics are those for:
 - NULL return
 - Constant return
 - negative integer return

Only the last of these three can possibly require significant code for
the returning block, and even the last is fairly rare and usually also
a constant. As a consequence, even for the cold return paths, there is
little code on that return path, and so little code density to be gained
by sinking it. The places where sinking these blocks is valuable (inner
loops) will already be weighted appropriately as the edge is a loop-exit
branch.

All of this aside, early returns are nearly as common as all three of
these return categories, and should actually be predicted as taken!
Rather than muddy the waters of the static predictions, just remain
silent on returns and let the CFG itself dictate any layout or other
issues.

However, the return heuristic was flagging one very important case:
unreachable. Unfortunately it still gave a 1/4 chance of the
branch-to-unreachable occuring. It also didn't do a rigorous job of
finding those blocks which post-dominate an unreachable block.

This patch builds a more powerful analysis that should flag all branches
to blocks known to then reach unreachable. It also has better worst-case
runtime complexity by not looping through successors for each block. The
previous code would perform an N^2 walk in the event of a single entry
block branching to N successors with a switch where each successor falls
through to the next and they finally fall through to a return.

Test case added for noreturn heuristics. Also doxygen comments improved
along the way.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142793 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-24 12:01:08 +00:00
Nick Lewycky
7c3fc57472 Reapply r142781 with fix. Original message:
Enhance SCEV's brute force loop analysis to handle multiple PHI nodes in the
  loop header when computing the trip count.

  With this, we now constant evaluate:
    struct ListNode { const struct ListNode *next; int i; };
    static const struct ListNode node1 = {0, 1};
    static const struct ListNode node2 = {&node1, 2};
    static const struct ListNode node3 = {&node2, 3};
    int test() {
      int sum = 0;
      for (const struct ListNode *n = &node3; n != 0; n = n->next)
        sum += n->i;
      return sum;
    }


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142790 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-24 06:57:05 +00:00
Nick Lewycky
38af3d5a8d Speculatively revert r142781. Bots are showing
Assertion `i_nocapture < OperandTraits<PHINode>::operands(this) && "getOperand() out of range!"' failed.
coming out of indvars.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142786 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-24 04:00:25 +00:00
Nick Lewycky
795cb48f1a Enhance SCEV's brute force loop analysis to handle multiple PHI nodes in the
loop header when computing the trip count.

With this, we now constant evaluate:
  struct ListNode { const struct ListNode *next; int i; };
  static const struct ListNode node1 = {0, 1};
  static const struct ListNode node2 = {&node1, 2};
  static const struct ListNode node3 = {&node2, 3};
  int test() {
    int sum = 0;
    for (const struct ListNode *n = &node3; n != 0; n = n->next)
      sum += n->i;
    return sum;
  }


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142781 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-23 23:43:14 +00:00
Chandler Carruth
14edd314af Teach the BranchProbabilityInfo pass to print its results, and use that
to bring it under direct test instead of merely indirectly testing it in
the BlockFrequencyInfo pass.

The next step is to start adding tests for the various heuristics
employed, and to start fixing those heuristics once they're under test.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142778 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-23 21:21:50 +00:00
Nick Lewycky
614fef6d5a Make SCEV's brute force analysis stronger in two ways. Firstly, we should be
able to constant fold load instructions where the argument is a constant.
Second, we should be able to watch multiple PHI nodes through the loop; this
patch only supports PHIs in loop headers, more can be done here.

With this patch, we now constant evaluate:
  static const int arr[] = {1, 2, 3, 4, 5};
  int test() {
    int sum = 0;
    for (int i = 0; i < 5; ++i) sum += arr[i];
    return sum;
  }


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142731 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-22 19:58:20 +00:00
Chandler Carruth
941aa7b1e8 Generalize the reading of probability metadata to work for both branches
and switches, with arbitrary numbers of successors. Still optimized for
the common case of 2 successors for a conditional branch.

Add a test case for switch metadata showing up in the BlockFrequencyInfo pass.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142493 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-19 10:32:19 +00:00
Chandler Carruth
99d01c54a0 Teach the BranchProbabilityInfo analysis pass to read any metadata
encoding of probabilities. In the absense of metadata, it continues to
fall back on static heuristics.

This allows __builtin_expect, after lowering through llvm.expect
a branch instruction's metadata, to actually enter the branch
probability model. This is one component of resolving PR2577.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142492 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-19 10:30:30 +00:00
Chandler Carruth
6aa5c26c81 Add pass printing support to BlockFrequencyInfo pass. The implementation
layer already had support for printing the results of this analysis, but
the wiring was missing.

Now that printing the analysis works, actually bring some of this
analysis, and the BranchProbabilityInfo analysis that it wraps, under
test! I'm planning on fixing some bugs and doing other work here, so
having a nice place to add regression tests and a way to observe the
results is really useful.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@142491 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-19 10:12:41 +00:00
Andrew Trick
0c388583fb Missing test case for r141164.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@141166 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-05 06:23:32 +00:00
Nick Lewycky
1cbae18cf6 Reapply r140979 with fix! We never did get a testcase, but careful review of the
logic by David Meyer revealed this bug.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140992 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-03 07:10:45 +00:00
Nick Lewycky
4fcc80a486 Revert r140979 due to reports of bootstrap failure.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140980 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-03 05:14:59 +00:00
Nick Lewycky
8fde4f5842 Add one more case we compute a max trip count.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140979 91177308-0d34-0410-b5e6-96231b3b80d8
2011-10-03 01:03:57 +00:00
Eli Friedman
e6fadced87 PR10628: Fix getModRefInfo so it queries the underlying alias() implementation correctly while checking nocapture calls.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140666 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-28 00:34:27 +00:00
Eli Friedman
46cb5afdcd Enhance alias analysis for atomic instructions a bit. Upgrade a couple alias-analysis tests to the new atomic instructions.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@140557 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-26 20:15:28 +00:00
Andrew Trick
2e95d76f8c This test only makes sense with -enable-iv-rewrite.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@139576 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-13 02:45:26 +00:00
Eli Friedman
81ac8ddc67 Fix the logic in BasicAliasAnalysis::aliasGEP for comparing GEP's with variable differences so that it actually does something sane. Fixes PR10881.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@139276 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-08 02:23:31 +00:00
Owen Anderson
69acc93b3d Teach BasicAA about the aliasing properties of memset_pattern16.
Fixes PR10872 and <rdar://problem/10065079>.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@139204 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-06 23:33:25 +00:00
Nick Lewycky
c103a08a11 This transform only handles two-operand AddRec's. Prevent it from trying to
handle anything more complex. Fixes PR10383 again!


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@139186 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-06 21:42:18 +00:00
Nick Lewycky
fa151a79ed The logic inside getMulExpr to simplify {a,+,b}*{c,+,d} was wrong, which was
visible given a=b=c=d=1, on iteration #1 (the second iteration). Replace it with
correct math. Fixes PR10383!


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@139133 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-06 05:05:14 +00:00
Nick Lewycky
b2840fdcd8 Revert r139126 due to selfhost failures reported by buildbots.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@139130 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-06 02:43:13 +00:00
Nick Lewycky
77b7352db0 Teach SCEV to report a max backedge count in one interesting case in
HowFarToZero; the case for a canonical loop.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@139126 91177308-0d34-0410-b5e6-96231b3b80d8
2011-09-05 23:25:16 +00:00
Rafael Espindola
aecfdd3068 Move the loads after the calls so that the fix for
PR10292 doesn't show that the loads don't alias
the allocas.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@134852 91177308-0d34-0410-b5e6-96231b3b80d8
2011-07-09 23:53:58 +00:00
Rafael Espindola
4fe509431b Use CHECK-NEXT.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@134850 91177308-0d34-0410-b5e6-96231b3b80d8
2011-07-09 22:56:50 +00:00
Chris Lattner
a53616d08b Remove support for parsing the "type i32" syntax for defining a numbered
top level type without a specified number.  This syntax isn't documented
and blocks forward progress.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@133371 91177308-0d34-0410-b5e6-96231b3b80d8
2011-06-19 00:03:46 +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
Chris Lattner
d589099eec make the asmparser reject function and type redefinitions. 'Merging' hasn't been
needed since llvm-gcc 3.4 days.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@133248 91177308-0d34-0410-b5e6-96231b3b80d8
2011-06-17 07:06:44 +00:00
Chris Lattner
7a1b9bdd2b Remove support for using "foo" as symbols instead of %"foo". This is ancient
syntax and has been long obsolete.  As usual, updating the tests is the nasty
part of this.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@133242 91177308-0d34-0410-b5e6-96231b3b80d8
2011-06-17 06:36:20 +00:00
Chris Lattner
26b0000166 manually upgrade a bunch of tests to modern syntax, and remove some that
are either unreduced or only test old syntax.



git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@133228 91177308-0d34-0410-b5e6-96231b3b80d8
2011-06-17 03:14:27 +00:00
John McCall
410eac5536 Test case for r132797.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132962 91177308-0d34-0410-b5e6-96231b3b80d8
2011-06-14 03:02:05 +00:00
Dan Gohman
5f1312c36e Reapply r131781, now that the GVN bug with partially-aliasing loads
is disabled.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132632 91177308-0d34-0410-b5e6-96231b3b80d8
2011-06-04 06:50:18 +00:00
Dan Gohman
baddf6c8b0 Remove this test, which should have been reverted along with r131781.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132628 91177308-0d34-0410-b5e6-96231b3b80d8
2011-06-04 06:21:23 +00:00
Dan Gohman
9a9d218ed0 Revert r131781 again. Apparently there is more going on here.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132625 91177308-0d34-0410-b5e6-96231b3b80d8
2011-06-04 05:11:22 +00:00
Dan Gohman
c122c625ce Reapply r131781 (revert r131809), now that some BasicAA shortcomings
it exposed are fixed.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132611 91177308-0d34-0410-b5e6-96231b3b80d8
2011-06-04 00:46:31 +00:00
Dan Gohman
1fc18d71de Fix BasicAA's recursion detection so that it doesn't pessimize
queries in the case of a DAG, where a query reaches a node
visited earlier, but it's not on a cycle. This avoids
MayAlias results in cases where BasicAA is expected to
return MustAlias or PartialAlias in order to protect TBAA.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132609 91177308-0d34-0410-b5e6-96231b3b80d8
2011-06-04 00:31:50 +00:00
Dan Gohman
965fefa1ad When merging MustAlias and PartialAlias, chose PartialAlias instead
of conservatively choosing MayAlias.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@132579 91177308-0d34-0410-b5e6-96231b3b80d8
2011-06-03 20:17:36 +00:00
Dan Gohman
9adf151b3d Make DecomposeGEPExpression check SimplifyInstruction only
after checking for a GEP, so that it matches what GetUnderlyingObject
does. This fixes an obscure bug turned up by bugpoint in the testcase
for PR9931.


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131971 91177308-0d34-0410-b5e6-96231b3b80d8
2011-05-24 18:24:08 +00:00
Chris Lattner
4d9255057a I missed a checking with my GVN change.
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@131851 91177308-0d34-0410-b5e6-96231b3b80d8
2011-05-22 07:20:02 +00:00