X86: Add prefetch insertion based on Propeller profile

llvm/include/llvm/CodeGen/BasicBlockSectionsProfileReader.h

@@ -42,6 +42,22 @@ struct BBClusterInfo {
   unsigned PositionInCluster;
 };
 
+// The prefetch symbol is emitted immediately after the call of the given index,
+// in block `BBID` (First call has an index of 1). Zero callsite index means the
+// start of the block.
+struct CallsiteID {
+  UniqueBBID BBID;
+  unsigned CallsiteIndex;
+};
+
+// This represents a prefetch hint to be injected at site `SiteID`, targetting
+// `TargetID` in function `TargetFunction`.
+struct PrefetchHint {
+  CallsiteID SiteID;
+  StringRef TargetFunction;
+  CallsiteID TargetID;
+};
+
 // This represents the raw input profile for one function.
 struct FunctionPathAndClusterInfo {
   // BB Cluster information specified by `UniqueBBID`s.
@@ -50,9 +66,13 @@ struct FunctionPathAndClusterInfo {
   // the edge a -> b (a is not cloned). The index of the path in this vector
   // determines the `UniqueBBID::CloneID` of the cloned blocks in that path.
   SmallVector<SmallVector<unsigned>> ClonePaths;
+  // Code prefetch targets, specified by the callsite ID immediately after
+  // which beginning must be targetted for prefetching.
+  SmallVector<CallsiteID> PrefetchTargets;
+  SmallVector<PrefetchHint> PrefetchHints;
   // Node counts for each basic block.
   DenseMap<UniqueBBID, uint64_t> NodeCounts;
-  // Edge counts for each edge, stored as a nested map.
+  // Edge counts for each edge.
   DenseMap<UniqueBBID, DenseMap<UniqueBBID, uint64_t>> EdgeCounts;
   // Hash for each basic block. The Hashes are stored for every original block
   // (not cloned blocks), hence the map key being unsigned instead of
@@ -86,6 +106,15 @@ class BasicBlockSectionsProfileReader {
   uint64_t getEdgeCount(StringRef FuncName, const UniqueBBID &SrcBBID,
                         const UniqueBBID &SinkBBID) const;
 
+  // Returns the prefetch targets (identified by their containing callsite IDs)
+  // for function `FuncName`.
+  SmallVector<CallsiteID>
+  getPrefetchTargetsForFunction(StringRef FuncName) const;
+
+  // Returns the prefetch hints to be injected in function `FuncName`.
+  SmallVector<PrefetchHint>
+  getPrefetchHintsForFunction(StringRef FuncName) const;
+
 private:
   StringRef getAliasName(StringRef FuncName) const {
     auto R = FuncAliasMap.find(FuncName);
@@ -195,6 +224,12 @@ class BasicBlockSectionsProfileReaderWrapperPass : public ImmutablePass {
   uint64_t getEdgeCount(StringRef FuncName, const UniqueBBID &SrcBBID,
                         const UniqueBBID &DestBBID) const;
 
+  SmallVector<PrefetchHint>
+  getPrefetchHintsForFunction(StringRef FuncName) const;
+
+  SmallVector<CallsiteID>
+  getPrefetchTargetsForFunction(StringRef FuncName) const;
+
   // Initializes the FunctionNameToDIFilename map for the current module and
   // then reads the profile for the matching functions.
   bool doInitialization(Module &M) override;

llvm/include/llvm/CodeGen/InsertCodePrefetch.h

@@ -0,0 +1,25 @@
+//===- BasicBlockSectionUtils.h - Utilities for basic block sections     --===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef LLVM_CODEGEN_INSERTCODEPREFETCH_H
+#define LLVM_CODEGEN_INSERTCODEPREFETCH_H
+
+#include "llvm/ADT/STLExtras.h"
+#include "llvm/ADT/SmallString.h"
+#include "llvm/Support/CommandLine.h"
+#include "llvm/Support/UniqueBBID.h"
+
+namespace llvm {
+
+SmallString<128> getPrefetchTargetSymbolName(StringRef FunctionName,
+                                             const UniqueBBID &BBID,
+                                             unsigned SubblockIndex);
+
+} // end namespace llvm
+
+#endif // LLVM_CODEGEN_INSERTCODEPREFETCH_H

llvm/include/llvm/CodeGen/MachineBasicBlock.h

@@ -213,6 +213,8 @@ class MachineBasicBlock
   /// basic block sections and basic block labels.
   std::optional<UniqueBBID> BBID;
 
+  SmallVector<unsigned> PrefetchTargets;
+
   /// With basic block sections, this stores the Section ID of the basic block.
   MBBSectionID SectionID{0};
 
@@ -229,6 +231,12 @@ class MachineBasicBlock
   /// is only computed once and is cached.
   mutable MCSymbol *CachedMCSymbol = nullptr;
 
+  /// Contains the callsite indices in this block that are targets of code
+  /// prefetching. The index `i` specifies the `i`th call, with zero
+  /// representing the beginning of the block and ` representing the first call.
+  /// Must be in ascending order and without duplicates.
+  SmallVector<unsigned> PrefetchTargetCallsiteIndexes;
+
   /// Cached MCSymbol for this block (used if IsEHContTarget).
   mutable MCSymbol *CachedEHContMCSymbol = nullptr;
 
@@ -710,6 +718,14 @@ class MachineBasicBlock
 
   std::optional<UniqueBBID> getBBID() const { return BBID; }
 
+  const SmallVector<unsigned> &getPrefetchTargetCallsiteIndexes() const {
+    return PrefetchTargetCallsiteIndexes;
+  }
+
+  void setPrefetchTargetCallsiteIndexes(const SmallVector<unsigned> &V) {
+    PrefetchTargetCallsiteIndexes = V;
+  }
+
   /// Returns the section ID of this basic block.
   MBBSectionID getSectionID() const { return SectionID; }
 

llvm/include/llvm/CodeGen/Passes.h

@@ -69,6 +69,8 @@ LLVM_ABI MachineFunctionPass *createBasicBlockSectionsPass();
 
 LLVM_ABI MachineFunctionPass *createBasicBlockPathCloningPass();
 
+LLVM_ABI MachineFunctionPass *createInsertCodePrefetchPass();
+
 /// createMachineBlockHashInfoPass - This pass computes basic block hashes.
 LLVM_ABI MachineFunctionPass *createMachineBlockHashInfoPass();
 

llvm/include/llvm/CodeGen/TargetInstrInfo.h

@@ -2381,6 +2381,14 @@ class LLVM_ABI TargetInstrInfo : public MCInstrInfo {
     llvm_unreachable("unknown number of operands necessary");
   }
 
+  /// Inserts a code prefetch instruction before `InsertBefore` in block `MBB`
+  /// targetting `GV`.
+  virtual bool insertCodePrefetchInstr(MachineBasicBlock &MBB,
+                                       MachineBasicBlock::iterator InsertBefore,
+                                       const GlobalValue *GV) const {
+    return false;
+  }
+
 private:
   mutable std::unique_ptr<MIRFormatter> Formatter;
   unsigned CallFrameSetupOpcode, CallFrameDestroyOpcode;

llvm/include/llvm/InitializePasses.h

@@ -56,6 +56,7 @@ LLVM_ABI void initializeAssignmentTrackingAnalysisPass(PassRegistry &);
 LLVM_ABI void initializeAssumptionCacheTrackerPass(PassRegistry &);
 LLVM_ABI void initializeAtomicExpandLegacyPass(PassRegistry &);
 LLVM_ABI void initializeBasicBlockPathCloningPass(PassRegistry &);
+LLVM_ABI void initializeInsertCodePrefetchPass(PassRegistry &);
 LLVM_ABI void
 initializeBasicBlockSectionsProfileReaderWrapperPassPass(PassRegistry &);
 LLVM_ABI void initializeBasicBlockSectionsPass(PassRegistry &);

llvm/lib/CodeGen/AsmPrinter/AsmPrinter.cpp

@@ -39,6 +39,7 @@
 #include "llvm/BinaryFormat/ELF.h"
 #include "llvm/CodeGen/GCMetadata.h"
 #include "llvm/CodeGen/GCMetadataPrinter.h"
+#include "llvm/CodeGen/InsertCodePrefetch.h"
 #include "llvm/CodeGen/LazyMachineBlockFrequencyInfo.h"
 #include "llvm/CodeGen/MachineBasicBlock.h"
 #include "llvm/CodeGen/MachineBlockHashInfo.h"
@@ -1985,7 +1986,33 @@ void AsmPrinter::emitFunctionBody() {
     // Print a label for the basic block.
     emitBasicBlockStart(MBB);
     DenseMap<StringRef, unsigned> MnemonicCounts;
+
+    SmallVector<unsigned> PrefetchTargets =
+        MBB.getPrefetchTargetCallsiteIndexes();
+    auto PrefetchTargetIt = PrefetchTargets.begin();
+    unsigned LastCallsiteIndex = 0;
+    // Helper to emit a symbol for the prefetch target and proceed to the next
+    // one.
+    auto EmitPrefetchTargetSymbolIfNeeded = [&]() {
+      if (PrefetchTargetIt != PrefetchTargets.end() &&
+          *PrefetchTargetIt == LastCallsiteIndex) {
+        MCSymbol *PrefetchTargetSymbol = OutContext.getOrCreateSymbol(
+            Twine("__llvm_prefetch_target_") + MF->getName() + Twine("_") +
+            utostr(MBB.getBBID()->BaseID) + Twine("_") +
+            utostr(static_cast<unsigned>(*PrefetchTargetIt)));
+        // If the function is weak-linkage it may be replaced by a strong
+        // version, in which case the prefetch targets should also be replaced.
+        OutStreamer->emitSymbolAttribute(
+            PrefetchTargetSymbol,
+            MF->getFunction().isWeakForLinker() ? MCSA_Weak : MCSA_Global);
+        OutStreamer->emitLabel(PrefetchTargetSymbol);
+        ++PrefetchTargetIt;
+      }
+    };
+
     for (auto &MI : MBB) {
+      EmitPrefetchTargetSymbolIfNeeded();
+
       // Print the assembly for the instruction.
       if (!MI.isPosition() && !MI.isImplicitDef() && !MI.isKill() &&
           !MI.isDebugInstr()) {
@@ -2123,8 +2150,11 @@ void AsmPrinter::emitFunctionBody() {
         break;
       }
 
-      if (MI.isCall() && MF->getTarget().Options.BBAddrMap)
-        OutStreamer->emitLabel(createCallsiteEndSymbol(MBB));
+      if (MI.isCall()) {
+        if (MF->getTarget().Options.BBAddrMap)
+          OutStreamer->emitLabel(createCallsiteEndSymbol(MBB));
+        LastCallsiteIndex++;
+      }
 
       if (TM.Options.EmitCallGraphSection && MI.isCall())
         handleCallsiteForCallgraph(FuncCGInfo, CallSitesInfoMap, MI);
@@ -2136,6 +2166,8 @@ void AsmPrinter::emitFunctionBody() {
       for (auto &Handler : Handlers)
         Handler->endInstruction();
     }
+    // Emit the last prefetch target in case the last instruction was a call.
+    EmitPrefetchTargetSymbolIfNeeded();
 
     // We must emit temporary symbol for the end of this basic block, if either
     // we have BBLabels enabled or if this basic blocks marks the end of a

llvm/lib/CodeGen/BasicBlockSections.cpp

@@ -106,7 +106,8 @@ class BasicBlockSections : public MachineFunctionPass {
 public:
   static char ID;
 
-  BasicBlockSectionsProfileReaderWrapperPass *BBSectionsProfileReader = nullptr;
+  // BasicBlockSectionsProfileReaderWrapperPass *BBSectionsProfileReader =
+  // nullptr;
 
   BasicBlockSections() : MachineFunctionPass(ID) {
     initializeBasicBlockSectionsPass(*PassRegistry::getPassRegistry());

llvm/lib/CodeGen/BasicBlockSectionsProfileReader.cpp

@@ -93,6 +93,19 @@ uint64_t BasicBlockSectionsProfileReader::getEdgeCount(
   return EdgeIt->second;
 }
 
+SmallVector<CallsiteID>
+BasicBlockSectionsProfileReader::getPrefetchTargetsForFunction(
+    StringRef FuncName) const {
+  return ProgramPathAndClusterInfo.lookup(getAliasName(FuncName))
+      .PrefetchTargets;
+}
+
+SmallVector<PrefetchHint>
+BasicBlockSectionsProfileReader::getPrefetchHintsForFunction(
+    StringRef FuncName) const {
+  return ProgramPathAndClusterInfo.lookup(getAliasName(FuncName)).PrefetchHints;
+}
+
 // Reads the version 1 basic block sections profile. Profile for each function
 // is encoded as follows:
 //   m <module_name>
@@ -148,6 +161,36 @@ uint64_t BasicBlockSectionsProfileReader::getEdgeCount(
 //                            +-->: 5 :
 //                                ....
 // ****************************************************************************
+// This profile can also specify prefetch targets (starting with 't') which
+// instruct the compiler to emit a prefetch symbol for the given target.
+// A prefetch target is specified by a pair "<bbid>,<subblock_index>" where
+// bbid specifies the target basic block and subblock_index is a zero-based
+// index. Callsite 0 refers to the region at the beginning of the block up to
+// the first callsite. Callsite `i > 0` refers to the region immediately after
+// the `i`-th callsite up to the `i+1`-th callsite (or the end of the block).
+// The prefetch target is always emitted at the beginning of the subblock.
+// This is the beginning of the basic block for `i = 0` and immediately after
+// the `i`-th call for every `i > 0`.
+//
+// Example: A basic block in function "foo" with BBID 10 and two call
+// instructions (call_A, call_B). This block is conceptually split into
+// subblocks, with the prefetch target symbol emitted at the beginning of each
+// subblock.
+//
+// +----------------------------------+
+// | __llvm_prefetch_target_foo_10_0: | <- Callsite 0 (before call_A)
+// |  Instruction 1                   |
+// |  Instruction 2                   |
+// |  call_A (Callsite 0)             |
+// | __llvm_prefetch_target_foo_10_1: | <--- Callsite 1 (after call_A,
+// |                                  |                  before call_B)
+// |  Instruction 3                   |
+// |  call_B (Callsite 1)             |
+// | __llvm_prefetch_target_foo_10_2: | <--- Callsite 2 (after call_B,
+// |                                  |                  before call_C)
+// |  Instruction 4                   |
+// +----------------------------------+
+//
 Error BasicBlockSectionsProfileReader::ReadV1Profile() {
   auto FI = ProgramPathAndClusterInfo.end();
 
@@ -308,6 +351,67 @@ Error BasicBlockSectionsProfileReader::ReadV1Profile() {
       }
       continue;
     }
+    case 'i': { // Prefetch hint specifier.
+      // Skip the profile when we the profile iterator (FI) refers to the
+      // past-the-end element.
+      if (FI == ProgramPathAndClusterInfo.end())
+        continue;
+      if (Values.size() != 2)
+        return createProfileParseError(Twine("Prefetch hint expected: " + S));
+      SmallVector<StringRef, 2> PrefetchSiteStr;
+      Values[0].split(PrefetchSiteStr, ',');
+      if (PrefetchSiteStr.size() != 2)
+        return createProfileParseError(Twine("Prefetch site expected: ") +
+                                       Values[0]);
+      auto SiteBBID = parseUniqueBBID(PrefetchSiteStr[0]);
+      if (!SiteBBID)
+        return SiteBBID.takeError();
+      unsigned long long SiteCallsiteIndex;
+      if (getAsUnsignedInteger(PrefetchSiteStr[1], 10, SiteCallsiteIndex))
+        return createProfileParseError(Twine("unsigned integer expected: '") +
+                                       PrefetchSiteStr[1]);
+
+      SmallVector<StringRef, 3> PrefetchTargetStr;
+      Values[1].split(PrefetchTargetStr, ',');
+      if (PrefetchTargetStr.size() != 3)
+        return createProfileParseError(
+            Twine("Prefetch target target expected: ") + Values[1]);
+      auto TargetBBID = parseUniqueBBID(PrefetchTargetStr[1]);
+      if (!TargetBBID)
+        return TargetBBID.takeError();
+      unsigned long long TargetCallsiteIndex;
+      if (getAsUnsignedInteger(PrefetchTargetStr[2], 10, TargetCallsiteIndex))
+        return createProfileParseError(Twine("unsigned integer expected: '") +
+                                       PrefetchTargetStr[2]);
+      FI->second.PrefetchHints.push_back(PrefetchHint{
+          CallsiteID{*SiteBBID, static_cast<unsigned>(SiteCallsiteIndex)},
+          PrefetchTargetStr[0],
+          CallsiteID{*TargetBBID, static_cast<unsigned>(TargetCallsiteIndex)}});
+      continue;
+    }
+    case 't': { // Prefetch target specifier.
+      // Skip the profile when we the profile iterator (FI) refers to the
+      // past-the-end element.
+      if (FI == ProgramPathAndClusterInfo.end())
+        continue;
+      SmallVector<StringRef, 2> PrefetchTargetStr;
+      if (Values.size() != 1)
+        return createProfileParseError(Twine("Prefetch target expected: ") + S);
+      Values[0].split(PrefetchTargetStr, ',');
+      if (PrefetchTargetStr.size() != 2)
+        return createProfileParseError(Twine("Prefetch target expected: ") +
+                                       Values[0]);
+      auto TargetBBID = parseUniqueBBID(PrefetchTargetStr[0]);
+      if (!TargetBBID)
+        return TargetBBID.takeError();
+      unsigned long long CallsiteIndex;
+      if (getAsUnsignedInteger(PrefetchTargetStr[1], 10, CallsiteIndex))
+        return createProfileParseError(Twine("signed integer expected: '") +
+                                       PrefetchTargetStr[1]);
+      FI->second.PrefetchTargets.push_back(
+          CallsiteID{*TargetBBID, static_cast<unsigned>(CallsiteIndex)});
+      continue;
+    }
     default:
       return createProfileParseError(Twine("invalid specifier: '") +
                                      Twine(Specifier) + "'");
@@ -514,6 +618,18 @@ uint64_t BasicBlockSectionsProfileReaderWrapperPass::getEdgeCount(
   return BBSPR.getEdgeCount(FuncName, SrcBBID, SinkBBID);
 }
 
+SmallVector<CallsiteID>
+BasicBlockSectionsProfileReaderWrapperPass::getPrefetchTargetsForFunction(
+    StringRef FuncName) const {
+  return BBSPR.getPrefetchTargetsForFunction(FuncName);
+}
+
+SmallVector<PrefetchHint>
+BasicBlockSectionsProfileReaderWrapperPass::getPrefetchHintsForFunction(
+    StringRef FuncName) const {
+  return BBSPR.getPrefetchHintsForFunction(FuncName);
+}
+
 BasicBlockSectionsProfileReader &
 BasicBlockSectionsProfileReaderWrapperPass::getBBSPR() {
   return BBSPR;

llvm/lib/CodeGen/CMakeLists.txt

@@ -79,6 +79,7 @@ add_llvm_component_library(LLVMCodeGen
   IndirectBrExpandPass.cpp
   InitUndef.cpp
   InlineSpiller.cpp
+  InsertCodePrefetch.cpp
   InterferenceCache.cpp
   InterleavedAccessPass.cpp
   InterleavedLoadCombinePass.cpp

llvm/lib/CodeGen/CodeGenPrepare.cpp

@@ -22,6 +22,7 @@
 #include "llvm/ADT/SmallPtrSet.h"
 #include "llvm/ADT/SmallVector.h"
 #include "llvm/ADT/Statistic.h"
+#include "llvm/ADT/StringExtras.h"
 #include "llvm/Analysis/BlockFrequencyInfo.h"
 #include "llvm/Analysis/BranchProbabilityInfo.h"
 #include "llvm/Analysis/FloatingPointPredicateUtils.h"