From dd018df8a00e841fe38fabe38520b7d297a885c1 Mon Sep 17 00:00:00 2001 From: Igor Murashkin Date: Wed, 9 Aug 2017 10:38:31 -0700 Subject: optimizing: add block-scoped constructor fence merging pass Introduce a new "Constructor Fence Redundancy Elimination" pass. The pass currently performs local optimization only, i.e. within instructions in the same basic block. All constructor fences preceding a publish (e.g. store, invoke) get merged into one instruction. ============== OptStat#ConstructorFenceGeneratedNew: 43825 OptStat#ConstructorFenceGeneratedFinal: 17631 <+++ OptStat#ConstructorFenceRemovedLSE: 164 OptStat#ConstructorFenceRemovedPFRA: 9391 OptStat#ConstructorFenceRemovedCFRE: 16133 <--- Removes ~91.5% of the 'final' constructor fences in RitzBenchmark: (We do not distinguish the exact reason that a fence was created, so it's possible some "new" fences were also removed.) ============== Test: art/test/run-test --host --optimizing 476-checker-ctor-fence-redun-elim Bug: 36656456 Change-Id: I8020217b448ad96ce9b7640aa312ae784690ad99 --- compiler/optimizing/optimizing_compiler.cc | 7 +++++++ 1 file changed, 7 insertions(+) (limited to 'compiler/optimizing/optimizing_compiler.cc') diff --git a/compiler/optimizing/optimizing_compiler.cc b/compiler/optimizing/optimizing_compiler.cc index 435ca1cad4..b45f3c6b33 100644 --- a/compiler/optimizing/optimizing_compiler.cc +++ b/compiler/optimizing/optimizing_compiler.cc @@ -55,6 +55,7 @@ #include "compiled_method.h" #include "compiler.h" #include "constant_folding.h" +#include "constructor_fence_redundancy_elimination.h" #include "dead_code_elimination.h" #include "debug/elf_debug_writer.h" #include "debug/method_debug_info.h" @@ -516,6 +517,8 @@ static HOptimization* BuildOptimization( return new (arena) CHAGuardOptimization(graph); } else if (opt_name == CodeSinking::kCodeSinkingPassName) { return new (arena) CodeSinking(graph, stats); + } else if (opt_name == ConstructorFenceRedundancyElimination::kPassName) { + return new (arena) ConstructorFenceRedundancyElimination(graph, stats); #ifdef ART_ENABLE_CODEGEN_arm } else if (opt_name == arm::InstructionSimplifierArm::kInstructionSimplifierArmPassName) { return new (arena) arm::InstructionSimplifierArm(graph, stats); @@ -786,6 +789,8 @@ void OptimizingCompiler::RunOptimizations(HGraph* graph, IntrinsicsRecognizer* intrinsics = new (arena) IntrinsicsRecognizer(graph, stats); CHAGuardOptimization* cha_guard = new (arena) CHAGuardOptimization(graph); CodeSinking* code_sinking = new (arena) CodeSinking(graph, stats); + ConstructorFenceRedundancyElimination* cfre = + new (arena) ConstructorFenceRedundancyElimination(graph, stats); HOptimization* optimizations1[] = { intrinsics, @@ -823,6 +828,8 @@ void OptimizingCompiler::RunOptimizations(HGraph* graph, // can satisfy. For example, the code generator does not expect to see a // HTypeConversion from a type to the same type. simplify4, + cfre, // Eliminate constructor fences after code sinking to avoid + // complicated sinking logic to split a fence with many inputs. }; RunOptimizations(optimizations2, arraysize(optimizations2), pass_observer); -- cgit v1.2.3-59-g8ed1b