[PATCH] D35096: Increase the import-threshold for crtical functions.

[Dehao Chen via Phabricator via llvm-commits]

danielcdh created this revision.
Herald added subscribers: eraman, mehdi_amini, sanjoy.

For interative sample-pgo, if a hot call site is inlined in the profiling binary, we should inline it in before profile annotation in the backend. Before that, the compile phase first collects all GUIDs that needs to be imported and creates virtual "hot" call edge in the summary. However, "hot" is not good enough to guarantee the callsites get inlined. This patch introduces "critical" call edge, and assign much higher importing threshold for those edges.


https://reviews.llvm.org/D35096

Files:
  include/llvm/IR/ModuleSummaryIndex.h
  lib/Analysis/ModuleSummaryAnalysis.cpp
  lib/Transforms/IPO/FunctionImport.cpp
  test/Bitcode/thinlto-function-summary-callgraph-profile-summary.ll
  test/Bitcode/thinlto-function-summary-callgraph-sample-profile-summary.ll


Index: test/Bitcode/thinlto-function-summary-callgraph-sample-profile-summary.ll
===================================================================
--- test/Bitcode/thinlto-function-summary-callgraph-sample-profile-summary.ll
+++ test/Bitcode/thinlto-function-summary-callgraph-sample-profile-summary.ll
@@ -29,7 +29,7 @@
 ; CHECK-NEXT:    <VERSION
 ; CHECK-NEXT:    <VALUE_GUID op0=25 op1=123/>
 ; op4=hot1 op6=cold op8=hot2 op10=hot4 op12=none1 op14=hot3 op16=none2 op18=none3 op20=123
-; CHECK-NEXT:    <PERMODULE_PROFILE {{.*}} op4=1 op5=3 op6=5 op7=1 op8=2 op9=3 op10=4 op11=3 op12=6 op13=2 op14=3 op15=3 op16=7 op17=2 op18=8 op19=2 op20=25 op21=3/>
+; CHECK-NEXT:    <PERMODULE_PROFILE {{.*}} op4=1 op5=3 op6=5 op7=1 op8=2 op9=3 op10=4 op11=3 op12=6 op13=2 op14=3 op15=3 op16=7 op17=2 op18=8 op19=2 op20=25 op21=4/>
 ; CHECK-NEXT:  </GLOBALVAL_SUMMARY_BLOCK>
 
 ; CHECK: <STRTAB_BLOCK
Index: test/Bitcode/thinlto-function-summary-callgraph-profile-summary.ll
===================================================================
--- test/Bitcode/thinlto-function-summary-callgraph-profile-summary.ll
+++ test/Bitcode/thinlto-function-summary-callgraph-profile-summary.ll
@@ -29,7 +29,7 @@
 ; CHECK-NEXT:    <VERSION
 ; CHECK-NEXT:    <VALUE_GUID op0=25 op1=123/>
 ; op4=hot1 op6=cold op8=hot2 op10=hot4 op12=none1 op14=hot3 op16=none2 op18=none3 op20=123
-; CHECK-NEXT:    <PERMODULE_PROFILE {{.*}} op4=1 op5=3 op6=5 op7=1 op8=2 op9=3 op10=4 op11=1 op12=6 op13=2 op14=3 op15=3 op16=7 op17=2 op18=8 op19=2 op20=25 op21=3/>
+; CHECK-NEXT:    <PERMODULE_PROFILE {{.*}} op4=1 op5=3 op6=5 op7=1 op8=2 op9=3 op10=4 op11=1 op12=6 op13=2 op14=3 op15=3 op16=7 op17=2 op18=8 op19=2 op20=25 op21=4/>
 ; CHECK-NEXT:  </GLOBALVAL_SUMMARY_BLOCK>
 
 ; CHECK: <STRTAB_BLOCK
Index: lib/Transforms/IPO/FunctionImport.cpp
===================================================================
--- lib/Transforms/IPO/FunctionImport.cpp
+++ lib/Transforms/IPO/FunctionImport.cpp
@@ -64,6 +64,12 @@
     "import-hot-multiplier", cl::init(3.0), cl::Hidden, cl::value_desc("x"),
     cl::desc("Multiply the `import-instr-limit` threshold for hot callsites"));
 
+static cl::opt<float> ImportCriticalMultiplier(
+    "import-critical-multiplier", cl::init(100.0), cl::Hidden,
+    cl::value_desc("x"),
+    cl::desc(
+        "Multiply the `import-instr-limit` threshold for critical callsites"));
+
 // FIXME: This multiplier was not really tuned up.
 static cl::opt<float> ImportColdMultiplier(
     "import-cold-multiplier", cl::init(0), cl::Hidden, cl::value_desc("N"),
@@ -207,6 +213,8 @@
         return ImportHotMultiplier;
       if (Hotness == CalleeInfo::HotnessType::Cold)
         return ImportColdMultiplier;
+      if (Hotness == CalleeInfo::HotnessType::Critical)
+        return ImportCriticalMultiplier;
       return 1.0;
     };
 
Index: lib/Analysis/ModuleSummaryAnalysis.cpp
===================================================================
--- lib/Analysis/ModuleSummaryAnalysis.cpp
+++ lib/Analysis/ModuleSummaryAnalysis.cpp
@@ -266,7 +266,7 @@
   // sample PGO, to enable the same inlines as the profiled optimized binary.
   for (auto &I : F.getImportGUIDs())
     CallGraphEdges[Index.getOrInsertValueInfo(I)].updateHotness(
-        CalleeInfo::HotnessType::Hot);
+        CalleeInfo::HotnessType::Critical);
 
   bool NonRenamableLocal = isNonRenamableLocal(F);
   bool NotEligibleForImport =
Index: include/llvm/IR/ModuleSummaryIndex.h
===================================================================
--- include/llvm/IR/ModuleSummaryIndex.h
+++ include/llvm/IR/ModuleSummaryIndex.h
@@ -47,7 +47,13 @@
 
 /// \brief Class to accumulate and hold information about a callee.
 struct CalleeInfo {
-  enum class HotnessType : uint8_t { Unknown = 0, Cold = 1, None = 2, Hot = 3 };
+  enum class HotnessType : uint8_t {
+    Unknown = 0,
+    Cold = 1,
+    None = 2,
+    Hot = 3,
+    Critical = 4
+  };
   HotnessType Hotness = HotnessType::Unknown;
 
   CalleeInfo() = default;


-------------- next part --------------
A non-text attachment was scrubbed...
Name: D35096.105557.patch
Type: text/x-patch
Size: 3987 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20170706/5133d98d/attachment.bin>