Spanning tree instrumentation

src/coreclr/inc/corjit.h

@@ -305,6 +305,7 @@ class ICorJitInfo : public ICorDynamicInfo
         TypeHandleHistogramTypeHandle = (DescriptorMin * 3) | TypeHandle, // TypeHandle that is part of a type histogram
         Version = (DescriptorMin * 4) | None, // Version is encoded in the Other field of the schema
         NumRuns = (DescriptorMin * 5) | None, // Number of runs is encoded in the Other field of the schema
+        EdgeIntCount = (DescriptorMin * 6) | FourByte, // 4 byte edge counter, using unsigned 4 byte int
     };
 
     struct PgoInstrumentationSchema

src/coreclr/jit/block.h

@@ -882,9 +882,14 @@ struct BasicBlock : private LIR::Range
     void ensurePredListOrder(Compiler* compiler);
     void reorderPredList(Compiler* compiler);
 
-    BlockSet    bbReach; // Set of all blocks that can reach this one
-    BasicBlock* bbIDom;  // Represent the closest dominator to this block (called the Immediate
-                         // Dominator) used to compute the dominance tree.
+    BlockSet bbReach; // Set of all blocks that can reach this one
+
+    union {
+        BasicBlock* bbIDom; // Represent the closest dominator to this block (called the Immediate
+                            // Dominator) used to compute the dominance tree.
+        void* bbProbeList;  // Used early on by fgInstrument
+        void* bbInfo;       // Used early on by fgIncorporate
+    };
 
     unsigned bbPostOrderNum; // the block's post order number in the graph.
 

src/coreclr/jit/compiler.cpp

@@ -4403,6 +4403,14 @@ void Compiler::compCompile(void** methodCodePtr, ULONG* methodCodeSize, JitFlags
     //
     DoPhase(this, PHASE_INCPROFILE, &Compiler::fgIncorporateProfileData);
 
+    // If we're going to instrument code, we may need to prepare before
+    // we import.
+    //
+    if (compileFlags->IsSet(JitFlags::JIT_FLAG_BBINSTR))
+    {
+        DoPhase(this, PHASE_IBCPREP, &Compiler::fgPrepareToInstrumentMethod);
+    }
+
     // Import: convert the instrs in each basic block to a tree based intermediate representation
     //
     DoPhase(this, PHASE_IMPORTATION, &Compiler::fgImport);

src/coreclr/jit/compiler.h

@@ -71,11 +71,13 @@ XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
  */
 
 struct InfoHdr;            // defined in GCInfo.h
-struct escapeMapping_t;    // defined in flowgraph.cpp
+struct escapeMapping_t;    // defined in fgdiagnostic.cpp
 class emitter;             // defined in emit.h
 struct ShadowParamVarInfo; // defined in GSChecks.cpp
 struct InitVarDscInfo;     // defined in register_arg_convention.h
-class FgStack;             // defined in flowgraph.cpp
+class FgStack;             // defined in fgbasic.cpp
+class Instrumentor;        // defined in fgprofile.cpp
+class SpanningTreeVisitor; // defined in fgprofile.cpp
 #if FEATURE_ANYCSE
 class CSE_DataFlow; // defined in OptCSE.cpp
 #endif
@@ -5535,15 +5537,6 @@ class Compiler
 
     void fgAdjustForAddressExposedOrWrittenThis();
 
-    bool                                   fgProfileData_ILSizeMismatch;
-    ICorJitInfo::PgoInstrumentationSchema* fgPgoSchema;
-    BYTE*                                  fgPgoData;
-    UINT32                                 fgPgoSchemaCount;
-    HRESULT                                fgPgoQueryResult;
-    UINT32                                 fgNumProfileRuns;
-    UINT32                                 fgPgoBlockCounts;
-    UINT32                                 fgPgoClassProfiles;
-
     unsigned fgStressBBProf()
     {
 #ifdef DEBUG
@@ -5562,13 +5555,32 @@ class Compiler
     }
 
     bool fgHaveProfileData();
-    void fgComputeProfileScale();
     bool fgGetProfileWeightForBasicBlock(IL_OFFSET offset, BasicBlock::weight_t* weight);
+
+    Instrumentor* fgCountInstrumentor;
+    Instrumentor* fgClassInstrumentor;
+
+    PhaseStatus fgPrepareToInstrumentMethod();
     PhaseStatus fgInstrumentMethod();
     PhaseStatus fgIncorporateProfileData();
     void        fgIncorporateBlockCounts();
+    void        fgIncorporateEdgeCounts();
 
 public:
+    bool                                   fgProfileData_ILSizeMismatch;
+    ICorJitInfo::PgoInstrumentationSchema* fgPgoSchema;
+    BYTE*                                  fgPgoData;
+    UINT32                                 fgPgoSchemaCount;
+    HRESULT                                fgPgoQueryResult;
+    UINT32                                 fgNumProfileRuns;
+    UINT32                                 fgPgoBlockCounts;
+    UINT32                                 fgPgoEdgeCounts;
+    UINT32                                 fgPgoClassProfiles;
+
+    void WalkSpanningTree(SpanningTreeVisitor* visitor);
+    void fgSetProfileWeight(BasicBlock* block, BasicBlock::weight_t weight);
+    void fgComputeProfileScale();
+
     // fgIsUsingProfileWeights - returns true if we have real profile data for this method
     //                           or if we have some fake profile data for the stress mode
     bool fgIsUsingProfileWeights()

src/coreclr/jit/compphases.h

@@ -28,6 +28,7 @@ CompPhaseNameMacro(PHASE_IMPORTATION,            "Importation",
 CompPhaseNameMacro(PHASE_INDXCALL,               "Indirect call transform",        "INDXCALL", false, -1, true)
 CompPhaseNameMacro(PHASE_PATCHPOINTS,            "Expand patchpoints",             "PPOINT",   false, -1, true)
 CompPhaseNameMacro(PHASE_POST_IMPORT,            "Post-import",                    "POST-IMP", false, -1, false)
+CompPhaseNameMacro(PHASE_IBCPREP,                "Profile instrumentation prep",   "IBCPREP",  false, -1, false)
 CompPhaseNameMacro(PHASE_IBCINSTR,               "Profile instrumentation",        "IBCINSTR", false, -1, false)
 CompPhaseNameMacro(PHASE_INCPROFILE,             "Profile incorporation",          "INCPROF",  false, -1, false)
 CompPhaseNameMacro(PHASE_MORPH_INIT,             "Morph - Init",                   "MOR-INIT" ,false, -1, false)

src/coreclr/jit/fgbasic.cpp

@@ -170,7 +170,10 @@ void Compiler::fgInit()
     fgPgoSchemaCount     = 0;
     fgNumProfileRuns     = 0;
     fgPgoBlockCounts     = 0;
+    fgPgoEdgeCounts      = 0;
     fgPgoClassProfiles   = 0;
+    fgCountInstrumentor  = nullptr;
+    fgClassInstrumentor  = nullptr;
     fgPredListSortVector = nullptr;
 }
 
@@ -450,7 +453,10 @@ void Compiler::fgReplaceSwitchJumpTarget(BasicBlock* blockSwitch, BasicBlock* ne
         {
             // Remove the old edge [oldTarget from blockSwitch]
             //
-            fgRemoveAllRefPreds(oldTarget, blockSwitch);
+            if (fgComputePredsDone)
+            {
+                fgRemoveAllRefPreds(oldTarget, blockSwitch);
+            }
 
             //
             // Change the jumpTab entry to branch to the new location
@@ -460,7 +466,12 @@ void Compiler::fgReplaceSwitchJumpTarget(BasicBlock* blockSwitch, BasicBlock* ne
             //
             // Create the new edge [newTarget from blockSwitch]
             //
-            flowList* newEdge = fgAddRefPred(newTarget, blockSwitch);
+            flowList* newEdge = nullptr;
+
+            if (fgComputePredsDone)
+            {
+                newEdge = fgAddRefPred(newTarget, blockSwitch);
+            }
 
             // Now set the correct value of newEdge->flDupCount
             // and replace any other jumps in jumpTab[] that go to oldTarget.
@@ -479,7 +490,10 @@ void Compiler::fgReplaceSwitchJumpTarget(BasicBlock* blockSwitch, BasicBlock* ne
                     //
                     // Increment the flDupCount
                     //
-                    newEdge->flDupCount++;
+                    if (fgComputePredsDone)
+                    {
+                        newEdge->flDupCount++;
+                    }
                 }
                 i++; // Check the next entry in jumpTab[]
             }
@@ -1780,11 +1794,6 @@ unsigned Compiler::fgMakeBasicBlocks(const BYTE* codeAddr, IL_OFFSET codeSize, F
         }
     }
 
-    if (compIsForInlining())
-    {
-        fgComputeProfileScale();
-    }
-
     do
     {
         unsigned   jmpAddr = DUMMY_INIT(BAD_IL_OFFSET);
@@ -3591,22 +3600,27 @@ BasicBlock* Compiler::fgSplitBlockAtBeginning(BasicBlock* curr)
 //              'succ' might be the fall-through path or the branch path from 'curr'.
 //
 // Arguments:
-//    curr - A block which branches conditionally to 'succ'
+//    curr - A block which branches to 'succ'
 //    succ - The target block
 //
 // Return Value:
 //    Returns a new block, that is a successor of 'curr' and which branches unconditionally to 'succ'
 //
 // Assumptions:
-//    'curr' must have a bbJumpKind of BBJ_COND or BBJ_SWITCH
+//    'curr' must have a bbJumpKind of BBJ_COND, BBJ_ALWAYS, or BBJ_SWITCH
 //
 // Notes:
 //    The returned block is empty.
+//    Can be invoked before pred lists are built.
 
 BasicBlock* Compiler::fgSplitEdge(BasicBlock* curr, BasicBlock* succ)
 {
-    assert(curr->bbJumpKind == BBJ_COND || curr->bbJumpKind == BBJ_SWITCH);
-    assert(fgGetPredForBlock(succ, curr) != nullptr);
+    assert(curr->bbJumpKind == BBJ_COND || curr->bbJumpKind == BBJ_SWITCH || curr->bbJumpKind == BBJ_ALWAYS);
+
+    if (fgComputePredsDone)
+    {
+        assert(fgGetPredForBlock(succ, curr) != nullptr);
+    }
 
     BasicBlock* newBlock;
     if (succ == curr->bbNext)
@@ -3639,20 +3653,30 @@ BasicBlock* Compiler::fgSplitEdge(BasicBlock* curr, BasicBlock* succ)
         }
         fgAddRefPred(newBlock, curr);
     }
-    else
+    else if (curr->bbJumpKind == BBJ_SWITCH)
     {
-        assert(curr->bbJumpKind == BBJ_SWITCH);
-
         // newBlock replaces 'succ' in the switch.
         fgReplaceSwitchJumpTarget(curr, newBlock, succ);
 
         // And 'succ' has 'newBlock' as a new predecessor.
         fgAddRefPred(succ, newBlock);
     }
+    else
+    {
+        assert(curr->bbJumpKind == BBJ_ALWAYS);
+        fgReplacePred(succ, curr, newBlock);
+        curr->bbJumpDest = newBlock;
+        newBlock->bbFlags |= BBF_JMP_TARGET;
+        fgAddRefPred(newBlock, curr);
+    }
 
     // This isn't accurate, but it is complex to compute a reasonable number so just assume that we take the
     // branch 50% of the time.
-    newBlock->inheritWeightPercentage(curr, 50);
+    //
+    if (curr->bbJumpKind != BBJ_ALWAYS)
+    {
+        newBlock->inheritWeightPercentage(curr, 50);
+    }
 
     // The bbLiveIn and bbLiveOut are both equal to the bbLiveIn of 'succ'
     if (fgLocalVarLivenessDone)