llvm-project/llvm/lib/Analysis
Cameron Esfahani 17177d1e84 Value soft float calls as more expensive in the inliner.
Summary: When evaluating floating point instructions in the inliner, ask the TTI whether it is an expensive operation.  By default, it's not an expensive operation.  This keeps the default behavior the same as before.  The ARM TTI has been updated to return back TCC_Expensive for targets which don't have hardware floating point.

Reviewers: chandlerc, echristo

Reviewed By: echristo

Subscribers: t.p.northover, aemerson, llvm-commits

Differential Revision: http://reviews.llvm.org/D6936

llvm-svn: 228263
2015-02-05 02:09:33 +00:00
..
IPA Value soft float calls as more expensive in the inliner. 2015-02-05 02:09:33 +00:00
AliasAnalysis.cpp [PM] Separate the TargetLibraryInfo object from the immutable pass. 2015-01-15 10:41:28 +00:00
AliasAnalysisCounter.cpp [C++11] More 'nullptr' conversion. In some cases just using a boolean check instead of comparing to nullptr. 2014-04-15 04:59:12 +00:00
AliasAnalysisEvaluator.cpp AA metadata refactoring (introduce AAMDNodes) 2014-07-24 12:16:19 +00:00
AliasDebugger.cpp [C++11] Add 'override' keyword to virtual methods that override their base class. 2014-03-05 07:30:04 +00:00
AliasSetTracker.cpp AliasSet: Simplify mergeSetIn 2014-11-19 19:36:18 +00:00
Analysis.cpp [PM] Change the core design of the TTI analysis to use a polymorphic 2015-01-31 03:43:40 +00:00
AssumptionCache.cpp [PM] Actually add the new pass manager support for the assumption cache. 2015-01-22 21:53:09 +00:00
BasicAliasAnalysis.cpp [PM] Split the LoopInfo object apart from the legacy pass, creating 2015-01-17 14:16:18 +00:00
BlockFrequencyInfo.cpp [PM] Split the LoopInfo object apart from the legacy pass, creating 2015-01-17 14:16:18 +00:00
BlockFrequencyInfoImpl.cpp BFI: Saturate when combining edges to a successor 2014-12-05 19:13:42 +00:00
BranchProbabilityInfo.cpp [PM] Split the LoopInfo object apart from the legacy pass, creating 2015-01-17 14:16:18 +00:00
CFG.cpp Standardize {pred,succ,use,user}_empty() 2015-01-13 03:46:47 +00:00
CFGPrinter.cpp Modernize raw_fd_ostream's constructor a bit. 2014-08-25 18:16:47 +00:00
CFLAliasAnalysis.cpp Fix incorrect partial aliasing 2015-01-26 17:31:17 +00:00
CGSCCPassManager.cpp [PM] Remove the defunt CGSCC-specific debug flag. 2015-01-13 22:45:13 +00:00
CMakeLists.txt [LoopVectorize] Move LoopAccessAnalysis to its own module 2015-02-01 16:56:15 +00:00
CaptureTracking.cpp [cleanup] Re-sort all the #include lines in LLVM using 2015-01-14 11:23:27 +00:00
CodeMetrics.cpp [PM] Split the AssumptionTracker immutable pass into two separate APIs: 2015-01-04 12:03:27 +00:00
ConstantFolding.cpp [PM] Move TargetLibraryInfo into the Analysis library. 2015-01-15 02:16:27 +00:00
CostModel.cpp [multiversion] Thread a function argument through all the callers of the 2015-02-01 12:01:35 +00:00
Delinearization.cpp [PM] Split the LoopInfo object apart from the legacy pass, creating 2015-01-17 14:16:18 +00:00
DependenceAnalysis.cpp [PM] Split the LoopInfo object apart from the legacy pass, creating 2015-01-17 14:16:18 +00:00
DomPrinter.cpp [PM] Split DominatorTree into a concrete analysis result object which 2014-01-13 13:07:17 +00:00
DominanceFrontier.cpp Templatify DominanceFrontier. 2014-07-12 21:59:52 +00:00
IVUsers.cpp [PM] Split the LoopInfo object apart from the legacy pass, creating 2015-01-17 14:16:18 +00:00
InstCount.cpp [Modules] Fix potential ODR violations by sinking the DEBUG_TYPE 2014-04-22 02:48:03 +00:00
InstructionSimplify.cpp Fold fcmp in cases where value is provably non-negative. By Arch Robison. 2015-01-28 08:03:58 +00:00
Interval.cpp Revert "[C++11] Add predecessors(BasicBlock *) / successors(BasicBlock *) iterator ranges." 2014-07-21 17:06:51 +00:00
IntervalPartition.cpp [C++11] More 'nullptr' conversion. In some cases just using a boolean check instead of comparing to nullptr. 2014-04-15 04:59:12 +00:00
JumpInstrTableInfo.cpp Add Forward Control-Flow Integrity. 2014-11-11 21:08:02 +00:00
LLVMBuild.txt Update libdeps since TLI was moved from Target to Analysis in r226078. 2015-01-15 05:21:00 +00:00
LazyCallGraph.cpp Revert r225854: [PM] Move the LazyCallGraph printing functionality to 2015-01-14 00:27:45 +00:00
LazyValueInfo.cpp [PM] Separate the TargetLibraryInfo object from the immutable pass. 2015-01-15 10:41:28 +00:00
LibCallAliasAnalysis.cpp [C++11] More 'nullptr' conversion. In some cases just using a boolean check instead of comparing to nullptr. 2014-04-15 04:59:12 +00:00
LibCallSemantics.cpp Move EH personality type classification to Analysis/LibCallSemantics.h 2015-01-28 01:17:38 +00:00
Lint.cpp [PM] Separate the TargetLibraryInfo object from the immutable pass. 2015-01-15 10:41:28 +00:00
Loads.cpp Revert r220349 to re-instate r220277 with a fix for PR21330 -- quite 2014-11-25 08:20:27 +00:00
LoopAccessAnalysis.cpp [LoopVectorize] Move LoopAccessAnalysis to its own module 2015-02-01 16:56:15 +00:00
LoopInfo.cpp [PM] Port LoopInfo to the new pass manager, adding both a LoopAnalysis 2015-01-20 10:58:50 +00:00
LoopPass.cpp [PM] Split the LoopInfo object apart from the legacy pass, creating 2015-01-17 14:16:18 +00:00
Makefile
MemDepPrinter.cpp [REFACTOR] Push logic from MemDepPrinter into getNonLocalPointerDependency 2015-01-09 00:26:45 +00:00
MemoryBuiltins.cpp [PM] Move TargetLibraryInfo into the Analysis library. 2015-01-15 02:16:27 +00:00
MemoryDependenceAnalysis.cpp Refine memory dependence's notion of volatile semantics 2015-01-26 18:54:27 +00:00
ModuleDebugInfoPrinter.cpp [C++11] Change DebugInfoFinder to use range-based loops 2014-03-18 09:41:07 +00:00
NoAliasAnalysis.cpp Simplify code. No functionality change. 2014-10-05 12:21:57 +00:00
PHITransAddr.cpp [PM] Split the AssumptionTracker immutable pass into two separate APIs: 2015-01-04 12:03:27 +00:00
PostDominators.cpp [Modules] Fix potential ODR violations by sinking the DEBUG_TYPE 2014-04-22 02:48:03 +00:00
PtrUseVisitor.cpp Update SetVector to rely on the underlying set's insert to return a pair<iterator, bool> 2014-11-19 07:49:26 +00:00
README.txt
RegionInfo.cpp [cleanup] Re-sort all the #include lines in LLVM using 2015-01-14 11:23:27 +00:00
RegionPass.cpp [cleanup] Re-sort all the #include lines in LLVM using 2015-01-14 11:23:27 +00:00
RegionPrinter.cpp Templatify RegionInfo so it works on MachineBasicBlocks 2014-07-19 18:29:29 +00:00
ScalarEvolution.cpp Make ScalarEvolution less aggressive with respect to no-wrap flags. 2015-01-22 00:48:47 +00:00
ScalarEvolutionAliasAnalysis.cpp AA metadata refactoring (introduce AAMDNodes) 2014-07-24 12:16:19 +00:00
ScalarEvolutionExpander.cpp [PM] Split the AssumptionTracker immutable pass into two separate APIs: 2015-01-04 12:03:27 +00:00
ScalarEvolutionNormalization.cpp Fix typos in comments, NFC 2014-08-29 21:53:01 +00:00
ScopedNoAliasAA.cpp [cleanup] Re-sort all the #include lines in LLVM using 2015-01-14 11:23:27 +00:00
SparsePropagation.cpp [Modules] Fix potential ODR violations by sinking the DEBUG_TYPE 2014-04-22 02:48:03 +00:00
StratifiedSets.h Update SetVector to rely on the underlying set's insert to return a pair<iterator, bool> 2014-11-19 07:49:26 +00:00
TargetLibraryInfo.cpp [PM] Rework how the TargetLibraryInfo pass integrates with the new pass 2015-01-24 02:06:09 +00:00
TargetTransformInfo.cpp Value soft float calls as more expensive in the inliner. 2015-02-05 02:09:33 +00:00
Trace.cpp Put the functionality for printing a value to a raw_ostream as an 2014-01-09 02:29:41 +00:00
TypeBasedAliasAnalysis.cpp IR: Split Metadata from Value 2014-12-09 18:38:53 +00:00
ValueTracking.cpp ValueTracking: Make isSafeToSpeculativelyExecute a little cleaner 2015-02-01 19:10:19 +00:00

README.txt

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))

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