Deprecate PrefetchInsertion optimization

The optimization has O(n^2) complexity and is currently disabled for
forward array traversals, and recently we've had to disable it under
all array traversals for concurrent scavenge. As read barriers become
more prominent, the introduction of the dataAddr pointer in the header
will force us to further limit this optimization.

This optimization was originally introduced in 2010 for z196
z/Architecture processor to accelerate SPEC workloads. It showed
minimal improvement (1.5%) according to historical data dug up prior to
open sourcing the project. The hardware is hardly used today since it
is so old. In recent iterations of the hardware, software prefetching
has been discouraged due to advancements in hardware prefetching.
Similar observations were seen when we removed prefetching from our
zero memory routines.

Moreover, the phantom words past the end of the GC heap were removed
several years ago, forcing us to disable this optimization for forward
array traversals. As mentioned previously we have been encountering
bugs in this optimization recently due to changes in the JVM and the
fact that this optimization introduces loads on the heap which
previously did not exist.

In the PR we detail performance numbers for several workloads where
we believed this optimization may have provided a benefit. Across all
platforms measured we no longer see a benefit (and sometimes a
degradation) from running this optimization. Here we deprecate the
optimization as we still have access to the original source code in
the history if we ever need to resurrect this effort.

Signed-off-by: Filip Jeremic <fjeremic@ca.ibm.com>

compiler/control/OMROptions.cpp

@@ -1,5 +1,5 @@
 /*******************************************************************************
- * Copyright (c) 2000, 2020 IBM Corp. and others
+ * Copyright (c) 2000, 2021 IBM Corp. and others
  *
  * This program and the accompanying materials are made available under
  * the terms of the Eclipse Public License 2.0 which accompanies this
@@ -480,7 +480,6 @@ TR::OptionTable OMR::Options::_jitOptions[] = {
    {"disablePRBE",                        "O\tdisable partial redundancy branch elimination",  SET_OPTION_BIT(TR_DisablePRBE), "F"},
    {"disablePRE",                         "O\tdisable partial redundancy elimination",         TR::Options::disableOptimization, partialRedundancyElimination, 0, "P"},
    {"disablePreexistenceDuringGracePeriod","O\tdisable preexistence during CLP grace period",  SET_OPTION_BIT(TR_DisablePrexistenceDuringGracePeriod), "F"},
-   {"disablePrefetchInsertion",           "O\tdisable prefetch insertion",                     TR::Options::disableOptimization, prefetchInsertion, 0, "P"},
    {"disableProfiledInlining",            "O\tdisable inlining based on profiled this values", SET_OPTION_BIT(TR_DisableProfiledInlining), "F"},
    {"disableProfiledMethodInlining",      "O\tdisable inlining based on profiled methods", SET_OPTION_BIT(TR_DisableProfiledMethodInlining), "F"},
    {"disableProfiledNodeVersioning",      "O\tdisable profiled node versioning",               TR::Options::disableOptimization, profiledNodeVersioning, 0, "P"},
@@ -1183,7 +1182,6 @@ TR::OptionTable OMR::Options::_jitOptions[] = {
 #endif
    {"traceOSRLiveRangeAnalysis",        "L\ttrace OSR live range analysis",                TR::Options::traceOptimization, osrLiveRangeAnalysis, 0, "P"},
    {"tracePRE",                         "L\ttrace partial redundancy elimination",        TR::Options::traceOptimization, partialRedundancyElimination, 0, "P"},
-   {"tracePrefetchInsertion",           "L\ttrace prefetch insertion",                     TR::Options::traceOptimization, prefetchInsertion, 0, "P"},
    {"tracePREForSubNodeReplacement",    "L\ttrace partial redundancy elimination focussed on optimal subnode replacement", SET_OPTION_BIT(TR_TracePREForOptimalSubNodeReplacement), "P" },
    {"traceProfiledNodeVersioning",      "L\ttrace profiled node versioning",               TR::Options::traceOptimization, profiledNodeVersioning, 0, "P"},
 #ifdef J9_PROJECT_SPECIFIC
@@ -2626,17 +2624,6 @@ OMR::Options::jitPreProcess()
 
       self()->setOption(TR_EnableAnnotations);
 
-      TR::Compilation* comp = TR::comp();
-      if (comp && TR::Compiler->om.canGenerateArraylets())
-         {
-         _disabledOptimizations[prefetchInsertion] = true;
-         }
-
-#if defined(TR_HOST_ARM64)
-      // Prefetch is not supported on ARM64 yet
-      _disabledOptimizations[prefetchInsertion] = true;
-#endif
-
       self()->setOption(TR_DisableThunkTupleJ2I); // JSR292:TODO: Figure out how to do this without confusing startPCIfAlreadyCompiled
 
       self()->setOption(TR_DisableSeparateInitFromAlloc);

compiler/optimizer/CMakeLists.txt

@@ -1,5 +1,5 @@
 ###############################################################################
-# Copyright (c) 2017, 2019 IBM Corp. and others
+# Copyright (c) 2017, 2021 IBM Corp. and others
 #
 # This program and the accompanying materials are made available under
 # the terms of the Eclipse Public License 2.0 which accompanies this
@@ -74,7 +74,6 @@ compiler_library(optimizer
 	${CMAKE_CURRENT_LIST_DIR}/OSRDefAnalysis.cpp
 	${CMAKE_CURRENT_LIST_DIR}/PartialRedundancy.cpp
 	${CMAKE_CURRENT_LIST_DIR}/PreExistence.cpp
-	${CMAKE_CURRENT_LIST_DIR}/PrefetchInsertion.cpp
 	${CMAKE_CURRENT_LIST_DIR}/Reachability.cpp
 	${CMAKE_CURRENT_LIST_DIR}/ReachingDefinitions.cpp
 	${CMAKE_CURRENT_LIST_DIR}/OMRRecognizedCallTransformer.cpp

compiler/optimizer/OMROptimizationGroups.enum

@@ -1,5 +1,5 @@
 /*******************************************************************************
- * Copyright (c) 2000, 2016 IBM Corp. and others
+ * Copyright (c) 2000, 2021 IBM Corp. and others
  *
  * This program and the accompanying materials are made available under
  * the terms of the Eclipse Public License 2.0 which accompanies this
@@ -57,7 +57,6 @@
    OPTIMIZATION(lateLocalGroup)
    OPTIMIZATION(eachLocalAnalysisPassGroup)
    OPTIMIZATION(stripMiningGroup)
-   OPTIMIZATION(prefetchInsertionGroup)
    OPTIMIZATION(sequentialLoadAndStoreColdGroup)
    OPTIMIZATION(sequentialLoadAndStoreWarmGroup)
    OPTIMIZATION(methodHandleInvokeInliningGroup)

compiler/optimizer/OMROptimizationManager.cpp

@@ -1,5 +1,5 @@
 /*******************************************************************************
- * Copyright (c) 2000, 2019 IBM Corp. and others
+ * Copyright (c) 2000, 2021 IBM Corp. and others
  *
  * This program and the accompanying materials are made available under
  * the terms of the Eclipse Public License 2.0 which accompanies this
@@ -189,9 +189,6 @@ OMR::OptimizationManager::OptimizationManager(TR::Optimizer *o, OptimizationFact
       case OMR::stripMining:
          _flags.set(requiresStructure | checkStructure | dumpStructure);
          break;
-      case OMR::prefetchInsertion:
-         _flags.set(requiresStructure | checkStructure | dumpStructure);
-         break;
       case OMR::osrDefAnalysis:
          if (self()->comp()->getOption(TR_DisableOSRSharedSlots))
             _flags.set(doesNotRequireAliasSets | doesNotRequireTreeDumps | supportsIlGenOptLevel);

compiler/optimizer/OMROptimizations.enum

@@ -1,5 +1,5 @@
 /*******************************************************************************
- * Copyright (c) 2000, 2019 IBM Corp. and others
+ * Copyright (c) 2000, 2021 IBM Corp. and others
  *
  * This program and the accompanying materials are made available under
  * the terms of the Eclipse Public License 2.0 which accompanies this
@@ -99,7 +99,6 @@
    OPTIMIZATION(profiledNodeVersioning)
    OPTIMIZATION(allocationSinking)
    OPTIMIZATION(stripMining)
-   OPTIMIZATION(prefetchInsertion)
    OPTIMIZATION(samplingJProfiling)
    OPTIMIZATION(trivialDeadTreeRemoval)
    OPTIMIZATION(osrDefAnalysis)

compiler/optimizer/OMROptimizer.cpp

@@ -1,5 +1,5 @@
 /*******************************************************************************
- * Copyright (c) 2000, 2019 IBM Corp. and others
+ * Copyright (c) 2000, 2021 IBM Corp. and others
  *
  * This program and the accompanying materials are made available under
  * the terms of the Eclipse Public License 2.0 which accompanies this
@@ -104,7 +104,6 @@
 #include "optimizer/SinkStores.hpp"
 #include "optimizer/PartialRedundancy.hpp"
 #include "optimizer/OSRDefAnalysis.hpp"
-#include "optimizer/PrefetchInsertion.hpp"
 #include "optimizer/StripMiner.hpp"
 #include "optimizer/FieldPrivatizer.hpp"
 #include "optimizer/ReorderIndexExpr.hpp"
@@ -444,13 +443,6 @@ const OptimizationStrategy stripMiningOpts[] =
    { endGroup                             }
    };
 
-const OptimizationStrategy prefetchInsertionOpts[] =
-   {
-   { inductionVariableAnalysis            },
-   { prefetchInsertion                    },
-   { endGroup                             }
-   };
-
 const OptimizationStrategy blockManipulationOpts[] =
    {
 //   { generalLoopUnroller,       IfLoops   }, //Unroll Loops
@@ -605,7 +597,6 @@ static const OptimizationStrategy omrHotStrategyOpts[] =
    { OMR::globalCopyPropagation,                             },
    { OMR::loopCanonicalizationGroup,                         }, // canonicalize loops (improve fall throughs)
    { OMR::expressionsSimplification,                         },
-   { OMR::prefetchInsertionGroup,                            }, // created IL should not be moved
    { OMR::partialRedundancyEliminationGroup                  },
    { OMR::globalDeadStoreElimination,                        },
    { OMR::inductionVariableAnalysis,                         },
@@ -849,8 +840,6 @@ OMR::Optimizer::Optimizer(TR::Compilation *comp, TR::ResolvedMethodSymbol *metho
       new (comp->allocator()) TR::OptimizationManager(self(), TR_OSRExceptionEdgeRemoval::create, OMR::osrExceptionEdgeRemoval);
    _opts[OMR::regDepCopyRemoval] =
       new (comp->allocator()) TR::OptimizationManager(self(), TR::RegDepCopyRemoval::create, OMR::regDepCopyRemoval);
-   _opts[OMR::prefetchInsertion] =
-      new (comp->allocator()) TR::OptimizationManager(self(), TR_PrefetchInsertion::create, OMR::prefetchInsertion);
    _opts[OMR::stripMining] =
       new (comp->allocator()) TR::OptimizationManager(self(), TR_StripMiner::create, OMR::stripMining);
    _opts[OMR::fieldPrivatization] =
@@ -904,8 +893,6 @@ OMR::Optimizer::Optimizer(TR::Compilation *comp, TR::ResolvedMethodSymbol *metho
       new (comp->allocator()) TR::OptimizationManager(self(), NULL, OMR::veryExpensiveGlobalValuePropagationGroup, veryExpensiveGlobalValuePropagationOpts);
    _opts[OMR::loopSpecializerGroup] =
       new (comp->allocator()) TR::OptimizationManager(self(), NULL, OMR::loopSpecializerGroup, loopSpecializerOpts);
-   _opts[OMR::prefetchInsertionGroup] =
-      new (comp->allocator()) TR::OptimizationManager(self(), NULL, OMR::prefetchInsertionGroup, prefetchInsertionOpts);
    _opts[OMR::lateLocalGroup] =
       new (comp->allocator()) TR::OptimizationManager(self(), NULL, OMR::lateLocalGroup, lateLocalOpts);
    _opts[OMR::eachLocalAnalysisPassGroup] =

compiler/optimizer/OptimizationStrategies.hpp

@@ -1,5 +1,5 @@
 /*******************************************************************************
- * Copyright (c) 2000, 2019 IBM Corp. and others
+ * Copyright (c) 2000, 2021 IBM Corp. and others
  *
  * This program and the accompanying materials are made available under
  * the terms of the Eclipse Public License 2.0 which accompanies this
@@ -55,7 +55,6 @@ extern const OptimizationStrategy loopCanonicalizationOpts[];
 extern const OptimizationStrategy blockManipulationOpts[];
 extern const OptimizationStrategy eachLocalAnalysisPassOpts[];
 extern const OptimizationStrategy stripMiningOpts[];
-extern const OptimizationStrategy prefetchInsertionOpts[];
 extern const OptimizationStrategy methodHandleInvokeInliningOpts[];
 
 //arrays of optimizations