Commit Graph

492 Commits

Author SHA1 Message Date
Cameron Zwarich 58c8670ab2 Turn this pass ordering:
Natural Loop Information
 Loop Pass Manager
   Canonicalize natural loops
 Scalar Evolution Analysis
 Loop Pass Manager
   Induction Variable Users
   Canonicalize natural loops
   Induction Variable Users
   Loop Strength Reduction

into this:

Scalar Evolution Analysis
Loop Pass Manager
  Canonicalize natural loops
  Induction Variable Users
  Loop Strength Reduction

This fixes <rdar://problem/8869639>. I also filed PR9184 on doing this sort of
thing automatically, but it seems easier to just change the ordering of the
passes if this is the only case.

llvm-svn: 125254
2011-02-10 01:07:54 +00:00
Dan Gohman de7f699754 Don't split any loop backedges, including backedges of loops other than
the active loop. This is generally desirable, and it avoids trouble
in situations such as the testcase in PR9123, though the failure
mode depends on use-list order, so it is infeasible to test.

llvm-svn: 125065
2011-02-08 00:55:13 +00:00
Cameron Zwarich b703654edc Remove code for updating dominance frontiers and some outdated references to
dominance and post-dominance frontiers.

llvm-svn: 123725
2011-01-18 04:11:31 +00:00
Chris Lattner bf0aa927cc split dom frontier handling stuff out to its own DominanceFrontier header,
so that Dominators.h is *just* domtree.  Also prune #includes a bit.

llvm-svn: 122714
2011-01-02 22:09:33 +00:00
Dan Gohman 20d9ce21ef Move SCEV::dominates and properlyDominates to ScalarEvolution.
llvm-svn: 119570
2010-11-17 21:41:58 +00:00
Dan Gohman afd6db9932 Move SCEV::isLoopInvariant and hasComputableLoopEvolution to be member
functions of ScalarEvolution, in preparation for memoization and
other optimizations.

llvm-svn: 119562
2010-11-17 21:23:15 +00:00
Owen Anderson a4fefc1949 Passes do not need to recursively initialize passes that they preserve, if
they do not also require them.  This allows us to reduce inter-pass linkage
dependencies.

llvm-svn: 116854
2010-10-19 20:08:44 +00:00
Owen Anderson 6c18d1aac0 Get rid of static constructors for pass registration. Instead, every pass exposes an initializeMyPassFunction(), which
must be called in the pass's constructor.  This function uses static dependency declarations to recursively initialize
the pass's dependencies.

Clients that only create passes through the createFooPass() APIs will require no changes.  Clients that want to use the
CommandLine options for passes will need to manually call the appropriate initialization functions in PassInitialization.h
before parsing commandline arguments.

I have tested this with all standard configurations of clang and llvm-gcc on Darwin.  It is possible that there are problems
with the static dependencies that will only be visible with non-standard options.  If you encounter any crash in pass
registration/creation, please send the testcase to me directly.

llvm-svn: 116820
2010-10-19 17:21:58 +00:00
Owen Anderson 8ac477ffb5 Begin adding static dependence information to passes, which will allow us to
perform initialization without static constructors AND without explicit initialization
by the client.  For the moment, passes are required to initialize both their
(potential) dependencies and any passes they preserve.  I hope to be able to relax
the latter requirement in the future.

llvm-svn: 116334
2010-10-12 19:48:12 +00:00
Dan Gohman 2fd85d7cd2 Filter out illegal formulae after updating offsets, not before, so that
formulae which become illegal as a result of the offset updating don't
escape.

This is for rdar://8529692. No testcase yet, because the given cases
hit use-list ordering differences.

llvm-svn: 116093
2010-10-08 19:33:26 +00:00
Dan Gohman 5947e1626a Delete the FormulaSorter class and inline its one method into its
one user. This code will be restructured soon and FormulaSorter
is getting in the way.

llvm-svn: 116012
2010-10-07 23:52:18 +00:00
Dan Gohman 1b61fd9bff Fix a spello.
llvm-svn: 116011
2010-10-07 23:43:09 +00:00
Dan Gohman 34f37e0d04 Charge a formula for explicit multiplies on scaled registers too,
not just base registers.

llvm-svn: 116010
2010-10-07 23:41:58 +00:00
Dan Gohman 49d638b45a Use size_t for consistency.
llvm-svn: 116009
2010-10-07 23:37:58 +00:00
Dan Gohman 8e72611058 When merging one use into another, transfer the offsets from
the old use to the new one.

llvm-svn: 116008
2010-10-07 23:36:45 +00:00
Dan Gohman a7b68d6d95 Fix LSR to keep the RegUseTracker up to date when combining users.
This doesn't usually matter, because the other heuristics usually
succeed regardless, but it's good to keep the register use
bookkeeping consistent.

llvm-svn: 116005
2010-10-07 23:33:43 +00:00
Owen Anderson df7a4f2515 Now with fewer extraneous semicolons!
llvm-svn: 115996
2010-10-07 22:25:06 +00:00
Dan Gohman 16ef49686c Delete an unused function.
llvm-svn: 114841
2010-09-27 16:58:21 +00:00
Gabor Greif a06741b356 do not rely on the implicit-dereference semantics of dyn_cast_or_null
llvm-svn: 114278
2010-09-18 11:55:34 +00:00
Dan Gohman 110ed64fbb Revert 112442 and 112440 until the compile time problems introduced
by 112440 are resolved.

llvm-svn: 112692
2010-09-01 01:45:53 +00:00
Dan Gohman 002ff89cbd Optionally rerun dedicated-register filtering after applying
other filtering techniques, as those may allow it to filter
out more obviously unprofitable candidates.

llvm-svn: 112441
2010-08-29 16:39:22 +00:00
Dan Gohman f031792cc6 Fix several areas in LSR to do a better job keeping the main
LSRInstance data structures up to date. This fixes some
pessimizations caused by stale data which will be exposed
in an upcoming change.

llvm-svn: 112440
2010-08-29 16:32:54 +00:00
Dan Gohman e9e0873b08 Refactor the three main groups of code out of
NarrowSearchSpaceUsingHeuristics into separate functions.

llvm-svn: 112439
2010-08-29 16:09:42 +00:00
Dan Gohman 37a0f68036 Delete a bogus check.
llvm-svn: 112438
2010-08-29 15:30:29 +00:00
Dan Gohman b6a520d63c Add some comments.
llvm-svn: 112437
2010-08-29 15:27:08 +00:00
Dan Gohman bf673e0652 Move this debug output into GenerateAllReuseFormula, to declutter
the high-level logic.

llvm-svn: 112436
2010-08-29 15:21:38 +00:00
Dan Gohman d366b6d5c8 Delete an unused declaration.
llvm-svn: 112435
2010-08-29 15:19:11 +00:00
Dan Gohman 4f13bbfefc Do one lookup instead of two.
llvm-svn: 112434
2010-08-29 15:18:49 +00:00
Dan Gohman 129a816ee6 Process the step before the start, because it's usually the simpler
of the two.

llvm-svn: 111495
2010-08-19 01:02:31 +00:00
Dan Gohman 89fdbaf99a Instead of having CollectSubexpr's categorize operands as interesting or
uninteresting, just put all the operands on one list and make
GenerateReassociations make the decision about what's interesting.
This is simpler, and it avoids an extra ScalarEvolution::getAddExpr call.

llvm-svn: 111133
2010-08-16 15:50:00 +00:00
Dan Gohman 9b7632df26 Put add operands in ScalarEvolution-canonical order, when convenient.
This isn't necessary, because ScalarEvolution sorts them anyway,
but it's tidier this way.

llvm-svn: 111132
2010-08-16 15:39:27 +00:00
Dan Gohman 081ffcd00b Fix LSR's ExtractImmediate and ExtractSymbol to avoid calling
ScalarEvolution::getAddExpr, which can be pretty expensive, when nothing
has changed, which is pretty common.

llvm-svn: 111042
2010-08-13 21:17:19 +00:00
Owen Anderson a7aed18624 Reapply r110396, with fixes to appease the Linux buildbot gods.
llvm-svn: 110460
2010-08-06 18:33:48 +00:00
Owen Anderson bda59bd247 Revert r110396 to fix buildbots.
llvm-svn: 110410
2010-08-06 00:23:35 +00:00
Owen Anderson 755aceb5d0 Don't use PassInfo* as a type identifier for passes. Instead, use the address of the static
ID member as the sole unique type identifier.  Clean up APIs related to this change.

llvm-svn: 110396
2010-08-05 23:42:04 +00:00
Dan Gohman ba81fc16a5 Fix whitespace.
llvm-svn: 110223
2010-08-04 17:43:57 +00:00
Oscar Fuentes 40b31ad3ee Prefix `next' iterator operation with `llvm::'.
Fixes potential ambiguity problems on VS 2010.

Patch by nobled!

llvm-svn: 110029
2010-08-02 06:00:15 +00:00
Owen Anderson a57b97e7e7 Fix batch of converting RegisterPass<> to INTIALIZE_PASS().
llvm-svn: 109045
2010-07-21 22:09:45 +00:00
Dan Gohman 1415208292 Don't merge uses when they are targetting fixup sites with
different widths. In a use with a narrower fixup, formulae
may be wider than the fixup, in which case the high bits
aren't necessarily meaningful, so it isn't safe to reuse
them for uses with wider fixups.

This fixes PR7618, though the testcase is too large for a
reasonable regression test, since it heavily dependes on
hitting LSR's heuristics in a certain way.

llvm-svn: 108455
2010-07-15 20:24:58 +00:00
Dan Gohman a1501b9c50 Use dbgs() instead of errs() in a DEBUG.
llvm-svn: 108453
2010-07-15 20:12:42 +00:00
Dan Gohman 4afd412d6b Watch out for a constant offset cancelling out a base register, forming
a zero. This situation arrises in Fortran code with induction variables
that start at 1 instead of 0. This fixes PR7651.

llvm-svn: 108424
2010-07-15 15:14:45 +00:00
Gabor Greif 8ae3095286 use getArgOperand instead of getOperand
llvm-svn: 107270
2010-06-30 09:15:28 +00:00
Dan Gohman fb9712bdae In GenerateReassociations, don't bother thinking about individual
SCEVUnknown values which are loop-variant, as LSR can't do anything
interesting with these values in any case. This fixes very slow compile
times on loops which have large numbers of such values.

llvm-svn: 106897
2010-06-25 22:32:18 +00:00
Dan Gohman 963b1c142e A few minor micro-optimizations.
llvm-svn: 106764
2010-06-24 16:57:52 +00:00
Dan Gohman 47ddf76d89 Teach getExactSDiv to evaluate x/1 to x up front, as it's a common
enough special case, and it theoretically allows more folding because
it works even when x is unanalyzable.

llvm-svn: 106763
2010-06-24 16:51:25 +00:00
Dan Gohman ab5422200b Fix copy+pasto issues in isMulSExtable.
llvm-svn: 106759
2010-06-24 16:45:11 +00:00
Dan Gohman 1081f1a0f5 Fix OptimizeMax to handle an odd case where one of the max operands
is another max which folds. This fixes PR7454.

llvm-svn: 106594
2010-06-22 23:07:13 +00:00
Dan Gohman dd41bba517 Use A.append(...) instead of A.insert(A.end(), ...) when A is a
SmallVector, and other SmallVector simplifications.

llvm-svn: 106452
2010-06-21 19:47:52 +00:00
Dan Gohman 32655906e4 Add a TODO comment.
llvm-svn: 106397
2010-06-19 21:30:18 +00:00
Dan Gohman 51d00092b6 Include the use kind along with the expression in the key of the
use sharing map. The reconcileNewOffset logic already forces a
separate use if the kinds differ, so incorporating the kind in the
key means we can track more sharing opportunities.

More sharing means fewer total uses to track, which means smaller
problem sizes, which means the conservative throttles don't kick
in as often.

llvm-svn: 106396
2010-06-19 21:29:59 +00:00