[llvm][ctx_profile] Add instrumentation

This adds instrumenting callsites to PGOInstrumentation, *if* contextual
profiling is requested. The latter also enables inserting counters in
the entry basic block and disables value profiling (the latter is a
point in time change)

This change adds the skeleton of the contextual profiling lowering pass,
just so we can introduce the flag controlling that and the API to check
that. The actual lowering pass will be introduced in a subsequent patch.

(Tracking Issue: llvm#89287, RFC referenced there)

llvm/include/llvm/Transforms/Instrumentation/PGOCtxProfLowering.h

@@ -0,0 +1,24 @@
+//===-- PGOCtxProfLowering.h - Contextual PGO Instr. Lowering ---*- C++ -*-===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+//
+// This file declares the PGOCtxProfLoweringPass class.
+//
+//===----------------------------------------------------------------------===//
+#ifndef LLVM_TRANSFORMS_INSTRUMENTATION_PGOCTXPROFLOWERING_H
+#define LLVM_TRANSFORMS_INSTRUMENTATION_PGOCTXPROFLOWERING_H
+
+namespace llvm {
+class Type;
+
+class PGOCtxProfLoweringPass {
+public:
+  explicit PGOCtxProfLoweringPass() = default;
+  static bool isContextualIRPGOEnabled();
+};
+} // namespace llvm
+#endif

llvm/lib/Transforms/Instrumentation/CMakeLists.txt

@@ -14,6 +14,7 @@ add_llvm_component_library(LLVMInstrumentation
   InstrProfiling.cpp
   KCFI.cpp
   LowerAllowCheckPass.cpp
+  PGOCtxProfLowering.cpp
   PGOForceFunctionAttrs.cpp
   PGOInstrumentation.cpp
   PGOMemOPSizeOpt.cpp

llvm/lib/Transforms/Instrumentation/PGOCtxProfLowering.cpp

@@ -0,0 +1,19 @@
+//===- PGOCtxProfLowering.cpp - Contextual PGO Instr. Lowering ------------===//
+//
+// 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
+//
+//===----------------------------------------------------------------------===//
+//
+
+#include "llvm/Transforms/Instrumentation/PGOCtxProfLowering.h"
+#include "llvm/Support/CommandLine.h"
+
+using namespace llvm;
+
+static cl::list<std::string> ContextRoots("profile-context-root");
+
+bool PGOCtxProfLoweringPass::isContextualIRPGOEnabled() {
+  return !ContextRoots.empty();
+}

llvm/lib/Transforms/Instrumentation/PGOInstrumentation.cpp

@@ -110,6 +110,7 @@
 #include "llvm/Transforms/Instrumentation.h"
 #include "llvm/Transforms/Instrumentation/BlockCoverageInference.h"
 #include "llvm/Transforms/Instrumentation/CFGMST.h"
+#include "llvm/Transforms/Instrumentation/PGOCtxProfLowering.h"
 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
 #include "llvm/Transforms/Utils/MisExpect.h"
 #include "llvm/Transforms/Utils/ModuleUtils.h"
@@ -333,6 +334,16 @@ extern cl::opt<bool> EnableVTableValueProfiling;
 extern cl::opt<InstrProfCorrelator::ProfCorrelatorKind> ProfileCorrelate;
 } // namespace llvm
 
+bool shouldInstrumentEntryBB() {
+  return PGOInstrumentEntry ||
+         PGOCtxProfLoweringPass::isContextualIRPGOEnabled();
+}
+
+bool isValueProfilingDisabled() {
+  return DisableValueProfiling ||
+         PGOCtxProfLoweringPass::isContextualIRPGOEnabled();
+}
+
 // Return a string describing the branch condition that can be
 // used in static branch probability heuristics:
 static std::string getBranchCondString(Instruction *TI) {
@@ -379,7 +390,7 @@ static GlobalVariable *createIRLevelProfileFlagVar(Module &M, bool IsCS) {
   uint64_t ProfileVersion = (INSTR_PROF_RAW_VERSION | VARIANT_MASK_IR_PROF);
   if (IsCS)
     ProfileVersion |= VARIANT_MASK_CSIR_PROF;
-  if (PGOInstrumentEntry)
+  if (shouldInstrumentEntryBB())
     ProfileVersion |= VARIANT_MASK_INSTR_ENTRY;
   if (DebugInfoCorrelate || ProfileCorrelate == InstrProfCorrelator::DEBUG_INFO)
     ProfileVersion |= VARIANT_MASK_DBG_CORRELATE;
@@ -861,7 +872,7 @@ static void instrumentOneFunc(
   }
 
   FuncPGOInstrumentation<PGOEdge, PGOBBInfo> FuncInfo(
-      F, TLI, ComdatMembers, true, BPI, BFI, IsCS, PGOInstrumentEntry,
+      F, TLI, ComdatMembers, true, BPI, BFI, IsCS, shouldInstrumentEntryBB(),
       PGOBlockCoverage);
 
   auto Name = FuncInfo.FuncNameVar;
@@ -883,6 +894,33 @@ static void instrumentOneFunc(
   unsigned NumCounters =
       InstrumentBBs.size() + FuncInfo.SIVisitor.getNumOfSelectInsts();
 
+  if (PGOCtxProfLoweringPass::isContextualIRPGOEnabled()) {
+    auto *CSIntrinsic =
+        Intrinsic::getDeclaration(M, Intrinsic::instrprof_callsite);
+    auto Visit = [&](llvm::function_ref<void(CallBase * CB)> Visitor) {
+      for (auto &BB : F)
+        for (auto &Instr : BB)
+          if (auto *CS = dyn_cast<CallBase>(&Instr)) {
+            if ((CS->getCalledFunction() &&
+                 CS->getCalledFunction()->isIntrinsic()) ||
+                dyn_cast<InlineAsm>(CS->getCalledOperand()))
+              continue;
+            Visitor(CS);
+          }
+    };
+    uint32_t TotalNrCallsites = 0;
+    Visit([&TotalNrCallsites](auto *) { ++TotalNrCallsites; });
+    uint32_t CallsiteIndex = 0;
+
+    Visit([&](auto *CB) {
+      IRBuilder<> Builder(CB);
+      Builder.CreateCall(CSIntrinsic,
+                         {Name, CFGHash, Builder.getInt32(TotalNrCallsites),
+                          Builder.getInt32(CallsiteIndex++),
+                          CB->getCalledOperand()});
+    });
+  }
+
   uint32_t I = 0;
   if (PGOTemporalInstrumentation) {
     NumCounters += PGOBlockCoverage ? 8 : 1;
@@ -914,7 +952,7 @@ static void instrumentOneFunc(
                                        FuncInfo.FunctionHash);
   assert(I == NumCounters);
 
-  if (DisableValueProfiling)
+  if (isValueProfilingDisabled())
     return;
 
   NumOfPGOICall += FuncInfo.ValueSites[IPVK_IndirectCallTarget].size();
@@ -1676,7 +1714,7 @@ void SelectInstVisitor::visitSelectInst(SelectInst &SI) {
 
 // Traverse all valuesites and annotate the instructions for all value kind.
 void PGOUseFunc::annotateValueSites() {
-  if (DisableValueProfiling)
+  if (isValueProfilingDisabled())
     return;
 
   // Create the PGOFuncName meta data.
@@ -2015,9 +2053,8 @@ static bool annotateAllFunctions(
 
   // If the profile marked as always instrument the entry BB, do the
   // same. Note this can be overwritten by the internal option in CFGMST.h
-  bool InstrumentFuncEntry = PGOReader->instrEntryBBEnabled();
-  if (PGOInstrumentEntry.getNumOccurrences() > 0)
-    InstrumentFuncEntry = PGOInstrumentEntry;
+  bool InstrumentFuncEntry =
+      PGOReader->instrEntryBBEnabled() || shouldInstrumentEntryBB();
   bool HasSingleByteCoverage = PGOReader->hasSingleByteCoverage();
   for (auto &F : M) {
     if (skipPGOUse(F))

llvm/test/Transforms/PGOProfile/ctx-instrumentation.ll

@@ -0,0 +1,61 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals all --version 4
+; RUN: opt -passes=pgo-instr-gen -profile-context-root=an_entrypoint \
+; RUN:   -S < %s | FileCheck --check-prefix=INSTRUMENT %s
+
+declare void @bar()
+
+;.
+; INSTRUMENT: @__llvm_profile_raw_version = hidden constant i64 360287970189639690, comdat
+; INSTRUMENT: @__profn_foo = private constant [3 x i8] c"foo"
+; INSTRUMENT: @__profn_an_entrypoint = private constant [13 x i8] c"an_entrypoint"
+;.
+define void @foo(i32 %a) {
+; INSTRUMENT-LABEL: define void @foo(
+; INSTRUMENT-SAME: i32 [[A:%.*]]) {
+; INSTRUMENT-NEXT:    call void @llvm.instrprof.increment(ptr @__profn_foo, i64 146835647075900052, i32 2, i32 0)
+; INSTRUMENT-NEXT:    [[T:%.*]] = icmp eq i32 [[A]], 0
+; INSTRUMENT-NEXT:    br i1 [[T]], label [[YES:%.*]], label [[NO:%.*]]
+; INSTRUMENT:       yes:
+; INSTRUMENT-NEXT:    call void @llvm.instrprof.increment(ptr @__profn_foo, i64 146835647075900052, i32 2, i32 1)
+; INSTRUMENT-NEXT:    br label [[EXIT:%.*]]
+; INSTRUMENT:       no:
+; INSTRUMENT-NEXT:    call void @llvm.instrprof.callsite(ptr @__profn_foo, i64 146835647075900052, i32 1, i32 0, ptr @bar)
+; INSTRUMENT-NEXT:    call void @bar()
+; INSTRUMENT-NEXT:    br label [[EXIT]]
+; INSTRUMENT:       exit:
+; INSTRUMENT-NEXT:    ret void
+;
+  %t = icmp eq i32 %a, 0
+  br i1 %t, label %yes, label %no
+yes:
+  br label %exit
+no:
+  call void @bar()
+  br label %exit
+exit:
+  ret void
+}
+
+define void @an_entrypoint(i32 %a) {
+; INSTRUMENT-LABEL: define void @an_entrypoint(
+; INSTRUMENT-SAME: i32 [[A:%.*]]) {
+; INSTRUMENT-NEXT:    call void @llvm.instrprof.increment(ptr @__profn_an_entrypoint, i64 784007058953177093, i32 2, i32 0)
+; INSTRUMENT-NEXT:    [[T:%.*]] = icmp eq i32 [[A]], 0
+; INSTRUMENT-NEXT:    br i1 [[T]], label [[YES:%.*]], label [[NO:%.*]]
+; INSTRUMENT:       yes:
+; INSTRUMENT-NEXT:    call void @llvm.instrprof.increment(ptr @__profn_an_entrypoint, i64 784007058953177093, i32 2, i32 1)
+; INSTRUMENT-NEXT:    ret void
+; INSTRUMENT:       no:
+; INSTRUMENT-NEXT:    ret void
+;
+  %t = icmp eq i32 %a, 0
+  br i1 %t, label %yes, label %no
+
+yes:
+  ret void
+no:
+  ret void
+}
+;.
+; INSTRUMENT: attributes #[[ATTR0:[0-9]+]] = { nounwind }
+;.