JIT: add ability to pad profile counters and adjust scalable count th…

src/coreclr/inc/clrconfigvalues.h

@@ -616,6 +616,10 @@ RETAIL_CONFIG_DWORD_INFO(UNSUPPORTED_TieredPGO_InstrumentOnlyHotCode, W("TieredP
 
 // By default, we only use optimizations in instrumented tiers for hot R2R code only.
 RETAIL_CONFIG_DWORD_INFO(UNSUPPORTED_TieredPGO_InstrumentedTierAlwaysOptimized, W("TieredPGO_InstrumentedTierAlwaysOptimized"), 0, "Always use optimizations inside instrumented tiers")
+
+// If scalable counters are used, set the threshold for approximate counting.
+RETAIL_CONFIG_DWORD_INFO(UNSUPPORTED_TieredPGO_ScalableCountThreshold, W("TieredPGO_ScalableCountThreshold"), 13, "Log2 threshold where counting becomes approximate")
+
 #endif
 
 ///

src/coreclr/jit/fgprofile.cpp

@@ -562,6 +562,10 @@ void BlockCountInstrumentor::BuildSchemaElements(BasicBlock* block, Schema& sche
     {
         numCountersPerProbe = 2;
     }
+    else if (JitConfig.JitCounterPadding() > 0)
+    {
+        numCountersPerProbe = (unsigned)JitConfig.JitCounterPadding();
+    }
 
     // Remember the schema index for this block.
     //
@@ -1743,6 +1747,10 @@ void EfficientEdgeCountInstrumentor::BuildSchemaElements(BasicBlock* block, Sche
     {
         numCountersPerProbe = 2;
     }
+    else if (JitConfig.JitCounterPadding() > 0)
+    {
+        numCountersPerProbe = (unsigned)JitConfig.JitCounterPadding();
+    }
 
     // Walk the bbSparseProbeList, emitting one schema element per...
     //

src/coreclr/jit/jitconfigvalues.h

@@ -577,6 +577,7 @@ CONFIG_STRING(JitEnablePatchpointRange, W("JitEnablePatchpointRange"))
 // Profile instrumentation options
 CONFIG_INTEGER(JitInterlockedProfiling, W("JitInterlockedProfiling"), 0)
 CONFIG_INTEGER(JitScalableProfiling, W("JitScalableProfiling"), 1)
+CONFIG_INTEGER(JitCounterPadding, W("JitCounterPadding"), 0) // number of unused extra slots per counter
 CONFIG_INTEGER(JitMinimalJitProfiling, W("JitMinimalJitProfiling"), 1)
 CONFIG_INTEGER(JitMinimalPrejitProfiling, W("JitMinimalPrejitProfiling"), 0)
 

src/coreclr/vm/eeconfig.cpp

@@ -240,6 +240,7 @@ HRESULT EEConfig::Init()
 #if defined(FEATURE_PGO)
     fTieredPGO = false;
     tieredPGO_InstrumentOnlyHotCode = false;
+    tieredPGO_ScalableCountThreshold = 13;
 #endif
 
 #if defined(FEATURE_READYTORUN)
@@ -782,6 +783,13 @@ HRESULT EEConfig::sync()
         fTieredPGO |= CLRConfig::GetConfigValue(CLRConfig::INTERNAL_WritePGOData) != 0;
         tieredPGO_InstrumentOnlyHotCode = CLRConfig::GetConfigValue(CLRConfig::UNSUPPORTED_TieredPGO_InstrumentOnlyHotCode) == 1;
 
+        DWORD scalableCountThreshold = CLRConfig::GetConfigValue(CLRConfig::UNSUPPORTED_TieredPGO_ScalableCountThreshold);
+
+        if ((scalableCountThreshold > 0) && (scalableCountThreshold < 20))
+        {
+            tieredPGO_ScalableCountThreshold = scalableCountThreshold;
+        }
+
         // We need quick jit for TieredPGO
         if (!fTieredCompilation_QuickJit)
         {

src/coreclr/vm/eeconfig.h

@@ -94,6 +94,7 @@ class EEConfig
 #if defined(FEATURE_PGO)
     bool          TieredPGO(void) const { LIMITED_METHOD_CONTRACT;  return fTieredPGO; }
     bool          TieredPGO_InstrumentOnlyHotCode(void) const { LIMITED_METHOD_CONTRACT;  return tieredPGO_InstrumentOnlyHotCode; }
+    DWORD         TieredPGO_ScalableCountThreshold() const { LIMITED_METHOD_CONTRACT;  return tieredPGO_ScalableCountThreshold; }
 #endif
 
 #if defined(FEATURE_READYTORUN)
@@ -658,6 +659,7 @@ class EEConfig
 #if defined(FEATURE_PGO)
     bool fTieredPGO;
     bool tieredPGO_InstrumentOnlyHotCode;
+    DWORD tieredPGO_ScalableCountThreshold;
 #endif
 
 #if defined(FEATURE_READYTORUN)

src/coreclr/vm/jithelpers.cpp

@@ -6004,8 +6004,8 @@ HCIMPLEND
 
 // Helpers for scalable approximate counters
 //
-// Here 13 means we count accurately up to 2^13 = 8192 and
-// then start counting probabialistically.
+// Here threshold = 13 means we count accurately up to 2^13 = 8192 and
+// then start counting probabilistically.
 //
 // See docs/design/features/ScalableApproximateCounting.md
 //
@@ -6016,22 +6016,22 @@ HCIMPL1(void, JIT_CountProfile32, volatile LONG* pCounter)
 
     LONG count = *pCounter;
     LONG delta = 1;
+    DWORD threshold = g_pConfig->TieredPGO_ScalableCountThreshold();
 
     if (count > 0)
     {
         DWORD logCount = 0;
         BitScanReverse(&logCount, count);
 
-        if (logCount >= 13)
+        if (logCount >= threshold)
         {
-            delta = 1 << (logCount - 12);
+            delta = 1 << (logCount - (threshold - 1));
             const unsigned rand = HandleHistogramProfileRand();
             const bool update = (rand & (delta - 1)) == 0;
             if (!update)
             {
                 return;
             }
-
         }
     }
 
@@ -6046,15 +6046,16 @@ HCIMPL1(void, JIT_CountProfile64, volatile LONG64* pCounter)
 
     LONG64 count = *pCounter;
     LONG64 delta = 1;
+    DWORD threshold = g_pConfig->TieredPGO_ScalableCountThreshold();
 
     if (count > 0)
     {
         DWORD logCount = 0;
         BitScanReverse64(&logCount, count);
 
-        if (logCount >= 13)
+        if (logCount >= threshold)
         {
-            delta = 1LL << (logCount - 12);
+            delta = 1LL << (logCount - (threshold - 1));
             const unsigned rand = HandleHistogramProfileRand();
             const bool update = (rand & (delta - 1)) == 0;
             if (!update)