llvm-6502/lib/Analysis
Adam Nemet 50b9e7f7d4 [getUnderlyingOjbects] Analyze loop PHIs further to remove false positives
Specifically, if a pointer accesses different underlying objects in each
iteration, don't look through the phi node defining the pointer.

The motivating case is the underlyling-objects-2.ll testcase.  Consider
the loop nest:

  int **A;
  for (i)
    for (j)
       A[i][j] = A[i-1][j] * B[j]

This loop is transformed by Load-PRE to stash away A[i] for the next
iteration of the outer loop:

  Curr = A[0];          // Prev_0
  for (i: 1..N) {
    Prev = Curr;        // Prev = PHI (Prev_0, Curr)
    Curr = A[i];
    for (j: 0..N)
       Curr[j] = Prev[j] * B[j]
  }

Since A[i] and A[i-1] are likely to be independent pointers,
getUnderlyingObjects should not assume that Curr and Prev share the same
underlying object in the inner loop.

If it did we would try to dependence-analyze Curr and Prev and the
analysis of the corresponding SCEVs would fail with non-constant
distance.

To fix this, the getUnderlyingObjects API is extended with an optional
LoopInfo parameter.  This is effectively what controls whether we want
the above behavior or the original.  Currently, I only changed to use
this approach for LoopAccessAnalysis.

The other testcase is to guard the opposite case where we do want to
look through the loop PHI.  If we step through an array by incrementing
a pointer, the underlying object is the incoming value of the phi as the
loop is entered.

Fixes rdar://problem/19566729

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@235634 91177308-0d34-0410-b5e6-96231b3b80d8
2015-04-23 20:09:20 +00:00
..
IPA [Inliner] Don't inline functions with frameescape calls 2015-04-14 20:38:14 +00:00
AliasAnalysis.cpp Add new getModRefInfo API to determine whether an Instruction and a call modify the same memory 2015-04-13 23:25:41 +00:00
AliasAnalysisCounter.cpp Use 'override/final' instead of 'virtual' for overridden methods 2015-04-11 02:11:45 +00:00
AliasAnalysisEvaluator.cpp [CallSite] Make construction from Value* (or Instruction*) explicit. 2015-04-10 14:50:08 +00:00
AliasDebugger.cpp Make DataLayout Non-Optional in the Module 2015-03-04 18:43:29 +00:00
AliasSetTracker.cpp [CallSite] Make construction from Value* (or Instruction*) explicit. 2015-04-10 14:50:08 +00:00
Analysis.cpp Divergence analysis for GPU programs 2015-04-10 05:03:50 +00:00
AssumptionCache.cpp
BasicAliasAnalysis.cpp [opaque pointer type] Change GetElementPtrInst::getIndexedType to take the pointee type 2015-03-30 21:41:43 +00:00
BlockFrequencyInfo.cpp Remove superfluous .str() and replace std::string concatenation with Twine. 2015-03-27 17:51:30 +00:00
BlockFrequencyInfoImpl.cpp Remove 4,096 loop scale limitation. 2015-04-01 17:42:27 +00:00
BranchProbabilityInfo.cpp Add range iterators for post order and inverse post order. Use them 2015-04-15 17:41:42 +00:00
CaptureTracking.cpp
CFG.cpp
CFGPrinter.cpp Remove superfluous .str() and replace std::string concatenation with Twine. 2015-03-27 17:51:30 +00:00
CFLAliasAnalysis.cpp Use 'override/final' instead of 'virtual' for overridden methods 2015-04-11 02:11:45 +00:00
CGSCCPassManager.cpp
CMakeLists.txt Move IDF Calculation to a separate file, expose an interface to it. 2015-04-21 19:13:02 +00:00
CodeMetrics.cpp Re-sort includes with sort-includes.py and insert raw_ostream.h where it's used. 2015-03-23 19:32:43 +00:00
ConstantFolding.cpp [opaque pointer type] API migration for GEP constant factories 2015-04-02 18:55:32 +00:00
CostModel.cpp
Delinearization.cpp
DependenceAnalysis.cpp Fix a memory corruption in Dependency Analysis. 2015-03-10 14:32:02 +00:00
DivergenceAnalysis.cpp Divergence analysis for GPU programs 2015-04-10 05:03:50 +00:00
DominanceFrontier.cpp
DomPrinter.cpp
InstCount.cpp
InstructionSimplify.cpp [opaque pointer type] API migration for GEP constant factories 2015-04-02 18:55:32 +00:00
Interval.cpp
IntervalPartition.cpp
IteratedDominanceFrontier.cpp Move IDF Calculation to a separate file, expose an interface to it. 2015-04-21 19:13:02 +00:00
IVUsers.cpp DataLayout is mandatory, update the API to reflect it with references. 2015-03-10 02:37:25 +00:00
LazyCallGraph.cpp
LazyValueInfo.cpp [ConstantRange] Split makeICmpRegion in two. 2015-03-18 00:41:24 +00:00
LibCallAliasAnalysis.cpp Make DataLayout Non-Optional in the Module 2015-03-04 18:43:29 +00:00
LibCallSemantics.cpp Stop calling DwarfEHPrepare from WinEHPrepare 2015-03-12 00:36:20 +00:00
Lint.cpp Fix doxygen comments from r232268 2015-03-16 17:49:03 +00:00
LLVMBuild.txt
Loads.cpp DataLayout is mandatory, update the API to reflect it with references. 2015-03-10 02:37:25 +00:00
LoopAccessAnalysis.cpp [getUnderlyingOjbects] Analyze loop PHIs further to remove false positives 2015-04-23 20:09:20 +00:00
LoopInfo.cpp Re-sort includes with sort-includes.py and insert raw_ostream.h where it's used. 2015-03-23 19:32:43 +00:00
LoopPass.cpp Purge unused includes throughout libSupport. 2015-03-23 18:07:13 +00:00
Makefile
MemDepPrinter.cpp [CallSite] Make construction from Value* (or Instruction*) explicit. 2015-04-10 14:50:08 +00:00
MemDerefPrinter.cpp Move Value.isDereferenceablePointer to ValueTracking [NFC] 2015-04-23 17:36:48 +00:00
MemoryBuiltins.cpp DataLayout is mandatory, update the API to reflect it with references. 2015-03-10 02:37:25 +00:00
MemoryDependenceAnalysis.cpp Revamp PredIteratorCache interface to be cleaner. 2015-04-21 21:11:50 +00:00
ModuleDebugInfoPrinter.cpp DebugInfo: Drop rest of DIDescriptor subclasses 2015-04-21 18:44:06 +00:00
NoAliasAnalysis.cpp Make DataLayout Non-Optional in the Module 2015-03-04 18:43:29 +00:00
PHITransAddr.cpp [opaque pointer type] more gep API migration 2015-03-14 19:53:33 +00:00
PostDominators.cpp
PtrUseVisitor.cpp
README.txt
RegionInfo.cpp
RegionPass.cpp Change range-based for-loops to be -Wrange-loop-analysis clean. 2015-04-15 01:21:15 +00:00
RegionPrinter.cpp One more -Wrange-loop-analysis cleanup. 2015-04-15 21:40:50 +00:00
ScalarEvolution.cpp Fix a type mismatch assert in SCEV division 2015-04-22 15:06:40 +00:00
ScalarEvolutionAliasAnalysis.cpp Make DataLayout Non-Optional in the Module 2015-03-04 18:43:29 +00:00
ScalarEvolutionExpander.cpp [SCEV] Strengthen SCEVExpander::isHighCostExpansion. 2015-04-14 03:20:32 +00:00
ScalarEvolutionNormalization.cpp
ScopedNoAliasAA.cpp Make DataLayout Non-Optional in the Module 2015-03-04 18:43:29 +00:00
SparsePropagation.cpp
StratifiedSets.h
TargetLibraryInfo.cpp [WinEH] Run cleanup handlers when an exception is thrown 2015-03-30 22:58:10 +00:00
TargetTransformInfo.cpp Divergence analysis for GPU programs 2015-04-10 05:03:50 +00:00
Trace.cpp
TypeBasedAliasAnalysis.cpp Teach TBAA analysis to report errors on cyclic TBAA metadata rather than hanging. 2015-03-13 07:09:33 +00:00
ValueTracking.cpp [getUnderlyingOjbects] Analyze loop PHIs further to remove false positives 2015-04-23 20:09:20 +00:00

Analysis Opportunities:

//===---------------------------------------------------------------------===//

In test/Transforms/LoopStrengthReduce/quadradic-exit-value.ll, the
ScalarEvolution expression for %r is this:

  {1,+,3,+,2}<loop>

Outside the loop, this could be evaluated simply as (%n * %n), however
ScalarEvolution currently evaluates it as

  (-2 + (2 * (trunc i65 (((zext i64 (-2 + %n) to i65) * (zext i64 (-1 + %n) to i65)) /u 2) to i64)) + (3 * %n))

In addition to being much more complicated, it involves i65 arithmetic,
which is very inefficient when expanded into code.

//===---------------------------------------------------------------------===//

In formatValue in test/CodeGen/X86/lsr-delayed-fold.ll,

ScalarEvolution is forming this expression:

((trunc i64 (-1 * %arg5) to i32) + (trunc i64 %arg5 to i32) + (-1 * (trunc i64 undef to i32)))

This could be folded to

(-1 * (trunc i64 undef to i32))

//===---------------------------------------------------------------------===//