Commit Graph

121 Commits

Author SHA1 Message Date
Daniel Berlin 9d0042b47c NewGVN: Fix memory congruence verification. The return true should be a return false. Merge the appropriate if statements so it doesn't happen again.
llvm-svn: 300584
2017-04-18 20:15:47 +00:00
Daniel Berlin ec9deb7f54 NewGVN: Don't waste time value numbering unreachable blocks
llvm-svn: 300565
2017-04-18 17:06:11 +00:00
Daniel Berlin 2f72b19b05 NewGVN: Don't propagate over phi backedges where undef causes us to
have >1 value, unless we can prove the phi node is cycle free.

Fixes PR 32607.

llvm-svn: 300299
2017-04-14 02:53:37 +00:00
Chandler Carruth 927d8e610a [IR] Redesign the case iterator in SwitchInst to actually be an iterator
and to expose a handle to represent the actual case rather than having
the iterator return a reference to itself.

All of this allows the iterator to be used with common STL facilities,
standard algorithms, etc.

Doing this exposed some missing facilities in the iterator facade that
I've fixed and required some work to the actual iterator to fully
support the necessary API.

Differential Revision: https://reviews.llvm.org/D31548

llvm-svn: 300032
2017-04-12 07:27:28 +00:00
Daniel Berlin 554dcd8c89 MemorySSA: Move to Analysis, from Transforms/Utils. It's used as
Analysis, it has Analysis passes, and once NewGVN is made an Analysis,
this removes the cross dependency from Analysis to Transform/Utils.
NFC.

llvm-svn: 299980
2017-04-11 20:06:36 +00:00
Daniel Berlin bf80cfe6b6 Revert "NewGVN: Don't propagate over phi backedges where undef causes us to have >1 value."
It's not ready yet this was an accidental commit :(

This reverts r299903

llvm-svn: 299904
2017-04-11 00:07:26 +00:00
Daniel Berlin 3938111fe7 NewGVN: Don't propagate over phi backedges where undef causes us to have >1 value.
Fixes PR 32607.

llvm-svn: 299903
2017-04-11 00:02:38 +00:00
Davide Italiano f58a30236b [NewGVN] Surround with parens to clarify allegedly ambiguous precedence.
This Placates GCC7 with -Werror. Also, clang-format the assertions
while I'm here.

llvm-svn: 299895
2017-04-10 23:08:35 +00:00
Daniel Berlin a823656ce7 NewGVN: Make CongruenceClass a real class in preparation for splitting
NewGVN into analysis and eliminator.

llvm-svn: 299792
2017-04-07 18:38:09 +00:00
Daniel Berlin 21279bd37a NewGVN: Rename some functions for consistency
llvm-svn: 299685
2017-04-06 18:52:58 +00:00
Daniel Berlin 08fe6e0f74 NewGVN: Fixup some small issues
llvm-svn: 299684
2017-04-06 18:52:55 +00:00
Daniel Berlin 5845e0549e NewGVN: Fix a small formatting issue in performSymbolicLoadEvaluation.
llvm-svn: 299683
2017-04-06 18:52:53 +00:00
Daniel Berlin 1316a94ebc NewGVN: This patch makes memory congruence work for all types of
memorydefs, not just stores.  Along the way, we audit and fixup issues
about how we were tracking memory leaders, and improve the verifier
to notice more memory congruency issues.

llvm-svn: 299682
2017-04-06 18:52:50 +00:00
Daniel Berlin 3082b8e062 MemorySSA: Fix and use optimized_def_chain
llvm-svn: 299566
2017-04-05 17:26:25 +00:00
Daniel Berlin 07daac8a36 NewGVN: Handle coercion of constant stores, loads, memory insts.
Summary:
Depends on D30928.

This adds support for coercion of stores and memory instructions that do not require insertion to process.
Another few tests down.
I added the relevant tests from rle.ll

Reviewers: davide

Subscribers: llvm-commits, Prazek

Differential Revision: https://reviews.llvm.org/D30929

llvm-svn: 299330
2017-04-02 13:23:44 +00:00
Daniel Berlin 9a9c9ff260 NewGVN: Don't try to kill off the stored value of stores when
processing the congruence class of the store.
Because we use the stored value of a store as the def, it isn't dead
just because it appears as a def when it comes from a store.

Note: I have not hit any cases with the memory code as it is where
this breaks anything, just because of what memory congruences we
actually allow.  In a followup that improves memory congruence,
this bug actually breaks real stuff (but the verifier catches it).

llvm-svn: 299300
2017-04-01 09:44:33 +00:00
Daniel Berlin 9b4984926c NewGVN: Clean up GVNExpression memory hierarchy, restructure hash computation a bit so we don't have to redefine it for loads, stores, and calls
llvm-svn: 299299
2017-04-01 09:44:29 +00:00
Daniel Berlin 871ecd90ca NewGVN: Use def_chain iterator in singleReachablePhiPath instead of recursion
llvm-svn: 299298
2017-04-01 09:44:24 +00:00
Davide Italiano e9781e7b2f [NewGVN] Adjust NDEBUG markers.
This avoids 'used but not defined' warnings in Release builds
with GCC.

llvm-svn: 298760
2017-03-25 02:40:02 +00:00
Daniel Berlin ffc30781f4 NewGVN: Small cleanup of two dominance related functions to make
them easier to understand.

llvm-svn: 298692
2017-03-24 06:33:51 +00:00
Daniel Berlin 0e9001131d NewGVN: Small cleanup of useless expression deletion, and don't uselessly create two expressions in symbolic store evaluation.
llvm-svn: 298691
2017-03-24 06:33:48 +00:00
Daniel Berlin 9d0796e5d0 NewGVN: Fix PR32403 - Handling of undef in phis was not quite correct
due to LLVM's view of phi nodes.  It would cause NewGVN not to fixpoint
in some interesting edge cases.

llvm-svn: 298687
2017-03-24 05:30:34 +00:00
Daniel Berlin 46b72e6de6 NewGVN: Now that we have a better verifier, we can prove that we can erase the predicateuser set each time we mark it touched
llvm-svn: 298199
2017-03-19 00:07:32 +00:00
Daniel Berlin d43f0ee7e1 NewGVN: Remove dead code (for now)
llvm-svn: 298198
2017-03-19 00:07:27 +00:00
Daniel Berlin 06329a98e3 NewGVN: Greatly enhance the ability of the NewGVN verifier to detect
issues, subsuming previous verifier.

llvm-svn: 298188
2017-03-18 15:41:40 +00:00
Daniel Berlin 41b39169e2 NewGVN: Fix PHI evaluation bug exposed by new verifier. We were checking whether the incoming block was reachable instead of whether the specific edge was reachable
llvm-svn: 298187
2017-03-18 15:41:36 +00:00
Daniel Berlin 64e689938d Split NewGVN class into a legacy pass and an impl, instead of a merged class.
llvm-svn: 297576
2017-03-12 04:46:45 +00:00
Daniel Berlin 5c338ff7a3 NewGVN: Rename InitialClass to TOP, which is what most people would expect it to be called
llvm-svn: 297494
2017-03-10 19:05:04 +00:00
Daniel Berlin e3e69e1680 NewGVN: Rewrite DCE during elimination so we do it as well as old GVN did.
llvm-svn: 297428
2017-03-10 00:32:33 +00:00
Daniel Berlin c0e008d807 NewGVN: Rename a few things for clarity
llvm-svn: 297427
2017-03-10 00:32:26 +00:00
Daniel Berlin 343576a6f4 NewGVN: Remove DebugUnknownExprs, just mark the instructions as unused
llvm-svn: 297047
2017-03-06 18:42:39 +00:00
Daniel Berlin 856fa14e06 NewGVN: Only call isInstructionTrivially dead once per instruction.
llvm-svn: 297046
2017-03-06 18:42:27 +00:00
Daniel Berlin 0350a87ce3 NewGVN: Be consistent in what order we compare operands for swapping.
NFC.

llvm-svn: 296935
2017-03-04 00:44:43 +00:00
Daniel Berlin 283a60875e NewGVN: Add debug counter for value numbering
llvm-svn: 296665
2017-03-01 19:59:26 +00:00
Daniel Berlin fccbda967a PredicateInfo: Support switch statements
Summary:
Depends on D29606 and D29682

Makes us pass GVN's edge.ll (we also will pass a few other testcases
they just need cleaning up).

Thoughts on the Predicate* hiearchy of classes especially welcome :)
(it's not clear to me how best to organize it, and currently, the getBlock* seems ... uglier than maybe wasting a field somewhere or something).

Reviewers: davide

Subscribers: llvm-commits

Differential Revision: https://reviews.llvm.org/D29747

llvm-svn: 295889
2017-02-22 22:20:58 +00:00
Daniel Berlin f7d9580a08 NewGVN: Start making use of predicateinfo pass.
Summary: This begins using the predicateinfo pass in NewGVN.

Reviewers: davide

Subscribers: llvm-commits, Prazek

Differential Revision: https://reviews.llvm.org/D29682

llvm-svn: 295583
2017-02-18 23:06:50 +00:00
Daniel Berlin b355c4ff5f NewGVN: Make ranking prefer undef to constants. Fix direction of
shouldSwapOperands to be correct.

llvm-svn: 295582
2017-02-18 23:06:47 +00:00
Davide Italiano bca05df38b [NewGVN] isOnlyReachableViaThisEdge() is dead now. NFCI.
llvm-svn: 295503
2017-02-17 22:12:30 +00:00
Davide Italiano 6db0ecafd7 [NewGVN] createVariableOrConstant is not required anymore. NFCI.
llvm-svn: 295500
2017-02-17 21:55:47 +00:00
Daniel Berlin 4d54796f87 NewGVN: Reverse order of congruence class elimination to maximize trivial deadness
llvm-svn: 294926
2017-02-12 23:24:45 +00:00
Daniel Berlin 508a1dec94 NewGVN: Use shouldSwapOperands in one more place
llvm-svn: 294925
2017-02-12 23:24:42 +00:00
Daniel Berlin 31e1b8fe48 Revert accidental commit titled "testing"
This reverts commit r294919

llvm-svn: 294923
2017-02-12 22:40:10 +00:00
Daniel Berlin 86eab15f2b NewGVN: Apply the fast math flags fix in r267113 to NewGVN as well.
llvm-svn: 294922
2017-02-12 22:25:20 +00:00
Daniel Berlin eccb8740d1 NewGVN: Fix missed call that should be to shouldSwapOperands
llvm-svn: 294920
2017-02-12 22:02:47 +00:00
Daniel Berlin 3fecad0d3e testing
llvm-svn: 294919
2017-02-12 22:02:20 +00:00
Daniel Berlin 22a4a01ffa NewGVN: Reverse sense of this test to make it clearer
llvm-svn: 294851
2017-02-11 15:20:15 +00:00
Daniel Berlin 1529bb93c9 NewGVN: Add missing initialization of NumFuncArgs lost due to bad merge.
llvm-svn: 294850
2017-02-11 15:13:49 +00:00
Daniel Berlin 1c08767f88 NewGVN: Rank and order commutative operands consistently.
llvm-svn: 294849
2017-02-11 15:07:01 +00:00
Daniel Berlin b79f53669a NewGVN: Clean up how we handle the INITIAL class so that everything in
it is dead or unreachable, as it should be.
This also makes the leader of INITIAL undef, enabling us to handle
irreducibility properly.

Summary:
This lets us verify, more than we do now, that we didn't screw up
value numbering.

Reviewers: davide

Subscribers: Prazek, llvm-commits

Differential Revision: https://reviews.llvm.org/D29842

llvm-svn: 294844
2017-02-11 12:48:50 +00:00
Benjamin Kramer efcf06f5f2 Move symbols from the global namespace into (anonymous) namespaces. NFC.
llvm-svn: 294837
2017-02-11 11:06:55 +00:00