6502/llvm-6502
1
0
Fork 0
mirror of https://github.com/c64scene-ar/llvm-6502.git synced 2024-07-23 16:30:00 +00:00
Code Issues Projects Releases Wiki Activity
43cda021d9
llvm-6502/lib/Analysis
History
Jakob Stoklund Olesen 43cda021d9 Fix inline cost predictions with SCIENCE. 
After running a batch of measurements, it is clear that the inliner metrics
need some adjustments:

Own argument bonus:       20 -> 5
Outgoing argument penalty: 0 -> 5
Alloca bonus:             10 -> 5
Constant instr bonus:      7 -> 5
Dead successor bonus:     40 -> 5*(avg instrs/block)

The new cost metrics are generaly 25 points higher than before, so we may need
to move thresholds.

With this change, InlineConstants::CallPenalty becomes a political correction:

if (!isa<IntrinsicInst>(II) && !callIsSmall(CS.getCalledFunction()))
  NumInsts += InlineConstants::CallPenalty + CS.arg_size();

The code size is accurately modelled by CS.arg_size(). CallPenalty is added
because calls tend to take a long time, so it may not be worth it to inline a
function with lots of calls.

All of the political corrections are in the InlineConstants namespace:
IndirectCallBonus, CallPenalty, LastCallToStaticBonus, ColdccPenalty,
NoreturnPenalty.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@94615 91177308-0d34-0410-b5e6-96231b3b80d8
2010-01-26 23:21:56 +00:00
..
IPA make -fno-rtti the default unless a directory builds with REQUIRES_RTTI. 2010-01-24 20:43:08 +00:00
AliasAnalysis.cpp Partially address a README by having functionattrs consider calls to 2010-01-06 08:45:52 +00:00
AliasAnalysisCounter.cpp drop the pass name from the output. 2010-01-22 05:52:51 +00:00
AliasAnalysisEvaluator.cpp Change dbgs() back to errs() as Chris requested. 2009-12-23 22:49:57 +00:00
AliasDebugger.cpp adopt getAdjustedAnalysisPointer in a few more passes. 2010-01-20 20:09:02 +00:00
AliasSetTracker.cpp Change dbgs() back to errs() as Chris requested. 2009-12-23 22:49:57 +00:00
Analysis.cpp "In order to ease automatic bindings generation, it would be helpful if boolean values were distinguishable from integers. The attached patch introduces "typedef int LLVMBool;", and uses LLVMBool instead of int throughout the C API, wherever a boolean value is called for." 2010-01-09 22:27:07 +00:00
BasicAliasAnalysis.cpp add some new methods to adjust this pointers. Not used yet. 2010-01-20 19:26:14 +00:00
CaptureTracking.cpp Reuse the Threshold value to size these containers because it's 2009-12-09 18:48:53 +00:00
CFGPrinter.cpp
CMakeLists.txt add to cmake 2009-12-04 04:15:36 +00:00
ConstantFolding.cpp fix PR5978 by peeling the loop so that we avoid shifting the 2010-01-08 19:02:23 +00:00
DbgInfoPrinter.cpp Remove dead debug info intrinsics. 2010-01-05 01:10:40 +00:00
DebugInfo.cpp Add extra element to composite type. This new element will be used to record c++ class that holds current class's vtable. 2010-01-26 21:14:59 +00:00
DomPrinter.cpp Create Generic DOTGraphTraits Printer/Viewer 2010-01-16 10:56:41 +00:00
InlineCost.cpp Fix inline cost predictions with SCIENCE. 2010-01-26 23:21:56 +00:00
InstCount.cpp Change dbgs() back to errs() as Chris requested. 2009-12-23 23:29:28 +00:00
InstructionSimplify.cpp factor some logic out of instcombine into a new SimplifyAddInst method. 2009-11-27 17:42:22 +00:00
Interval.cpp
IntervalPartition.cpp
IVUsers.cpp Re-implement the main strength-reduction portion of LoopStrengthReduction. 2010-01-21 02:09:26 +00:00
LazyValueInfo.cpp Convert debug messages to use dbgs(). Generally this means 2009-12-23 20:43:58 +00:00
LibCallAliasAnalysis.cpp
LibCallSemantics.cpp
LiveValues.cpp
LoopDependenceAnalysis.cpp Convert debug messages to use dbgs(). Generally this means 2009-12-23 20:52:41 +00:00
LoopInfo.cpp Restore dump() methods to Loop and MachineLoop. 2010-01-05 21:08:02 +00:00
LoopPass.cpp eliminate a bunch more unneeded dynamic_cast's. 2010-01-22 05:37:10 +00:00
Makefile make -fno-rtti the default unless a directory builds with REQUIRES_RTTI. 2010-01-24 20:43:08 +00:00
MemoryBuiltins.cpp Simplify ComputeMultiple so that it doesn't depend on TargetData. 2009-11-18 00:58:27 +00:00
MemoryDependenceAnalysis.cpp The phi translated pointer can be computed when returning a partially cached result 2009-12-22 04:25:02 +00:00
PHITransAddr.cpp Change dbgs() back to errs() as Chris requested. 2009-12-23 23:27:15 +00:00
PointerTracking.cpp
PostDominators.cpp Convert debug messages to use dbgs(). Generally this means 2009-12-23 21:16:54 +00:00
ProfileEstimatorPass.cpp adopt getAdjustedAnalysisPointer in a few more passes. 2010-01-20 20:09:02 +00:00
ProfileInfo.cpp adopt getAdjustedAnalysisPointer in a few more passes. 2010-01-20 20:09:02 +00:00
ProfileInfoLoader.cpp
ProfileInfoLoaderPass.cpp adopt getAdjustedAnalysisPointer in a few more passes. 2010-01-20 20:09:02 +00:00
ProfileVerifierPass.cpp Convert debug messages to use dbgs(). Generally this means 2009-12-23 22:10:20 +00:00
README.txt
ScalarEvolution.cpp Make the unsigned-range code more consistent with the signed-range code, 2010-01-26 19:19:05 +00:00
ScalarEvolutionAliasAnalysis.cpp adopt getAdjustedAnalysisPointer in a few more passes. 2010-01-20 20:09:02 +00:00
ScalarEvolutionExpander.cpp When re-using an existing cast for a user, it's still necessary to call 2010-01-21 10:08:42 +00:00
SparsePropagation.cpp Convert debug messages to use dbgs(). Generally this means 2009-12-23 22:28:01 +00:00
Trace.cpp Convert debug messages to use dbgs(). Generally this means 2009-12-23 22:35:10 +00:00
ValueTracking.cpp teach ComputeNumSignBits to look through PHI nodes. 2010-01-07 23:44:37 +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.

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