maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/lib/IR
History
Dehao Chen 9232f98279 Implement callsite-hotness based inline cost for Sample-based PGO 
Summary:
For sample-based PGO, using BFI to calculate callsite count is sometime not accurate. This is because with sampling based approach, if a callsite resides in a hot loop deeply nested in a bunch of cold branches, the callsite's BFI frequency would be inaccurately calculated due to lack of samples in the cold branch.

E.g.

if (A1 && A2 && A3 && ..... && A10) {
  for (i=0; i < 100000000; i++) {
    callsite();
  }
}

Assume that A1 to A100 are all 100% taken, and callsite has 1000 samples and thus is considerred hot. Because the loop's trip count is huge, it's normal that all branches outside the loop has no sample at all. As a result, we can only use static branch probability to derive the the frequency of the loop header. Assuming that static heuristic thinks each branch is 50% taken, then the count calculated from BFI will be 1/(2^10) of the actual value.

In order to get more accurate callsite count, we directly annotate the weight on the call instruction, and directly use it when checking callsite hotness.

Note that this mechanism can also be shared by instrumentation based callsite hotness analysis. The side benefit is that it breaks the dependency from Inliner to BFI as call count is embedded in the IR.

Reviewers: davidxl, eraman, dnovillo

Subscribers: llvm-commits

Differential Revision: http://reviews.llvm.org/D22118

llvm-svn: 275073
 		2016-07-11 16:48:54 +00:00
..
AsmWriter.cpp [codeview] Add DISubprogram::ThisAdjustment 2016-07-01 02:41:21 +00:00
AttributeImpl.h Apply clang-tidy's modernize-loop-convert to most of lib/IR. 2016-06-26 14:10:56 +00:00
Attributes.cpp Add getReturnedArgOperand to Call/InvokeInst, CallSite 2016-07-10 23:01:32 +00:00
AttributesCompatFunc.td Provide a way to specify inliner's attribute compatibility and merging. 2015-12-22 23:57:37 +00:00
AutoUpgrade.cpp [X86] Remove and autoupgrade 512-bit non-temporal store intrinsics. 2016-07-09 04:38:27 +00:00
BasicBlock.cpp Apply clang-tidy's modernize-loop-convert to most of lib/IR. 2016-06-26 14:10:56 +00:00
CMakeLists.txt Move ProfileSummary to IR. 2016-05-19 21:07:12 +00:00
Comdat.cpp [NFC] Remove some dead code: 2016-05-11 16:38:40 +00:00
ConstantFold.cpp [ConstantFold] NFC cleanup after previous change. 2016-05-26 07:08:09 +00:00
ConstantFold.h [NFC] Header cleanup 2016-04-18 09:17:29 +00:00
ConstantRange.cpp fix formatting, typo; NFC 2016-06-19 17:20:27 +00:00
Constants.cpp Apply clang-tidy's modernize-loop-convert to most of lib/IR. 2016-06-26 14:10:56 +00:00
ConstantsContext.h Increase SmallVector size for ConstantUniqueMap::getHashValue() (NFC) 2016-04-19 00:17:55 +00:00
Core.cpp Apply clang-tidy's modernize-loop-convert to most of lib/IR. 2016-06-26 14:10:56 +00:00
DIBuilder.cpp [DIBuilder] Remove dead code. NFC. 2016-07-02 13:18:38 +00:00
DataLayout.cpp Rename getLargestLegalIntTypeSize to getLargestLegalIntTypeSizeInBits(). NFC. 2016-05-13 18:38:35 +00:00
DebugInfo.cpp StripDebugInfo: uses isa<DbgInfoIntrinsic> instead of matching against llvm.dbg.* (NFC) 2016-05-14 04:58:35 +00:00
DebugInfoMetadata.cpp [codeview] Add DISubprogram::ThisAdjustment 2016-07-01 02:41:21 +00:00
DebugLoc.cpp [NFC] Header cleanup 2016-04-18 09:17:29 +00:00
DiagnosticInfo.cpp [Diag] Add getter shouldAlwaysPrint. NFC 2016-06-29 04:55:19 +00:00
DiagnosticPrinter.cpp MIR Serialization: Connect the machine function analysis pass to the MIR parser. 2015-06-15 20:30:22 +00:00
Dominators.cpp [PM] Remove support for omitting the AnalysisManager argument to new 2016-06-17 00:11:01 +00:00
Function.cpp Apply clang-tidy's modernize-loop-convert to most of lib/IR. 2016-06-26 14:10:56 +00:00
GCOV.cpp llvm-cov: Fix reading gcov data that does not have function names 2016-02-08 22:49:40 +00:00
GVMaterializer.cpp
…
Globals.cpp [IR] [DAE] Copy comdats during DAE, and don't copy comdats in GlobalObject::copyAttributesFrom. 2016-06-15 23:20:15 +00:00
IRBuilder.cpp Support arbitrary addrspace pointers in masked load/store intrinsics 2016-06-28 18:27:25 +00:00
IRPrintingPasses.cpp [PM] Remove support for omitting the AnalysisManager argument to new 2016-06-17 00:11:01 +00:00
InlineAsm.cpp [InlineAsm] Avoid creating extra string instances in ConstraintInfo::Parse() 2016-05-23 13:58:04 +00:00
Instruction.cpp The absence of noreturn doesn't ensure mayReturn 2016-06-25 00:55:12 +00:00
Instructions.cpp Add getReturnedArgOperand to Call/InvokeInst, CallSite 2016-07-10 23:01:32 +00:00
IntrinsicInst.cpp [NFC] Remove some dead code: 2016-05-11 16:38:40 +00:00
LLVMBuild.txt
…
LLVMContext.cpp IR: New representation for CFI and virtual call optimization pass metadata. 2016-06-24 21:21:32 +00:00
LLVMContextImpl.cpp Apply clang-tidy's modernize-loop-convert to most of lib/IR. 2016-06-26 14:10:56 +00:00
LLVMContextImpl.h [codeview] Add DISubprogram::ThisAdjustment 2016-07-01 02:41:21 +00:00
LegacyPassManager.cpp Remove doInitialization() and doFinalization() member declarations without definitions. 2016-04-28 19:21:30 +00:00
MDBuilder.cpp [MDBuilder] Simplify code using initializer lists. NFC. 2015-11-22 18:03:17 +00:00
Mangler.cpp Update to use new name alignTo(). 2016-01-14 21:06:47 +00:00
Metadata.cpp Implement callsite-hotness based inline cost for Sample-based PGO 2016-07-11 16:48:54 +00:00
MetadataImpl.h DI: Disallow uniquable DICompileUnits 2015-08-03 17:26:41 +00:00
Module.cpp Remove interface to get/set MaxFunctionCount 2016-06-20 21:36:38 +00:00
ModuleSummaryIndex.cpp [ThinLTO] Introduce typedef for commonly-used map type (NFC) 2016-04-25 21:09:51 +00:00
Operator.cpp [opaque pointer types] [NFC] GEP: replace get(Pointer)ElementType uses with get{Source,Result}ElementType. 2016-01-19 17:28:00 +00:00
OptBisect.cpp Removing unused function. 2016-04-22 22:24:20 +00:00
Pass.cpp Re-commit optimization bisect support (r267022) without new pass manager support. 2016-04-22 22:06:11 +00:00
PassManager.cpp [PM] Run clang-format over various parts of the new pass manager code 2016-06-17 07:15:29 +00:00
PassRegistry.cpp [NFC] Header cleanup 2016-04-18 09:17:29 +00:00
ProfileSummary.cpp Reapply r271728 after adding move cobstructor for ProfileSummaryInfo 2016-06-03 22:54:26 +00:00
Statepoint.cpp Introduce an GCRelocateInst class [NFC] 2016-04-12 18:05:10 +00:00
SymbolTableListTraitsImpl.h IR: Remove implicit iterator conversions from lib/IR, NFC 2015-10-08 23:49:46 +00:00
Type.cpp fix documentation comments; NFC 2016-06-02 20:37:52 +00:00
TypeFinder.cpp Apply clang-tidy's modernize-loop-convert to most of lib/IR. 2016-06-26 14:10:56 +00:00
Use.cpp Revert r240137 (Fixed/added namespace ending comments using clang-tidy. NFC) 2015-06-23 09:49:53 +00:00
User.cpp [IR] Teach `llvm::User` to co-allocate a descriptor. 2015-09-24 01:00:49 +00:00
Value.cpp BasicAA should look through functions with returned arguments 2016-07-11 01:32:20 +00:00
ValueSymbolTable.cpp Apply clang-tidy's modernize-loop-convert to most of lib/IR. 2016-06-26 14:10:56 +00:00
ValueTypes.cpp Add isScalarInteger helper to EVT/MVT 2016-03-01 18:01:28 +00:00
Verifier.cpp Add writeonly IR attribute 2016-07-04 08:01:29 +00:00