[PILC][profcheck] Bias branch weights when optimizing sqrt (#167742)

Biasing towards the native `sqrt`​ not returning NaN.  


Issue #147390

Author: mtrofin

llvm/lib/Transforms/Scalar/PartiallyInlineLibCalls.cpp

@@ -19,6 +19,7 @@
 #include "llvm/Analysis/TargetTransformInfo.h"
 #include "llvm/IR/Dominators.h"
 #include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/MDBuilder.h"
 #include "llvm/InitializePasses.h"
 #include "llvm/Support/DebugCounter.h"
 #include "llvm/Transforms/Scalar.h"
@@ -27,6 +28,10 @@
 
 using namespace llvm;
 
+namespace llvm {
+extern cl::opt<bool> ProfcheckDisableMetadataFixes;
+} // namespace llvm
+
 #define DEBUG_TYPE "partially-inline-libcalls"
 
 DEBUG_COUNTER(PILCounter, "partially-inline-libcalls-transform",
@@ -94,7 +99,14 @@ static bool optimizeSQRT(CallInst *Call, Function *CalledFunc,
                     : Builder.CreateFCmpOGE(Call->getOperand(0),
                                             ConstantFP::get(Ty, 0.0));
   CurrBBTerm->setCondition(FCmp);
-
+  if (!ProfcheckDisableMetadataFixes &&
+      CurrBBTerm->getFunction()->getEntryCount()) {
+    // Presume the quick path - where we don't call the library call - is the
+    // frequent one
+    MDBuilder MDB(CurrBBTerm->getContext());
+    CurrBBTerm->setMetadata(LLVMContext::MD_prof,
+                            MDB.createLikelyBranchWeights());
+  }
   // Add phi operands.
   Phi->addIncoming(Call, &CurrBB);
   Phi->addIncoming(LibCall, LibCallBB);

llvm/test/Transforms/PartiallyInlineLibCalls/X86/good-prototype.ll

@@ -1,12 +1,12 @@
 ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --check-globals
 ; RUN: opt -S -passes=partially-inline-libcalls -mtriple=x86_64-unknown-linux-gnu < %s | FileCheck %s
 
-define float @f(float %val) {
+define float @f(float %val) !prof !0 {
 ; CHECK-LABEL: @f(
 ; CHECK-NEXT:  entry:
 ; CHECK-NEXT:    [[RES:%.*]] = tail call float @sqrtf(float [[VAL:%.*]]) #[[ATTR0:[0-9]+]]
 ; CHECK-NEXT:    [[TMP0:%.*]] = fcmp oge float [[VAL]], 0.000000e+00
-; CHECK-NEXT:    br i1 [[TMP0]], label [[ENTRY_SPLIT:%.*]], label [[CALL_SQRT:%.*]]
+; CHECK-NEXT:    br i1 [[TMP0]], label [[ENTRY_SPLIT:%.*]], label [[CALL_SQRT:%.*]], !prof [[PROF1:![0-9]+]]
 ; CHECK:       call.sqrt:
 ; CHECK-NEXT:    [[TMP1:%.*]] = tail call float @sqrtf(float [[VAL]])
 ; CHECK-NEXT:    br label [[ENTRY_SPLIT]]
@@ -19,11 +19,11 @@ entry:
   ret float %res
 }
 
-define float @f_writeonly(float %val) {
+define float @f_writeonly(float %val) !prof !0 {
 ; CHECK-LABEL: @f_writeonly(
 ; CHECK-NEXT:    [[RES:%.*]] = tail call float @sqrtf(float [[VAL:%.*]]) #[[ATTR0]]
 ; CHECK-NEXT:    [[TMP1:%.*]] = fcmp oge float [[VAL]], 0.000000e+00
-; CHECK-NEXT:    br i1 [[TMP1]], label [[DOTSPLIT:%.*]], label [[CALL_SQRT:%.*]]
+; CHECK-NEXT:    br i1 [[TMP1]], label [[DOTSPLIT:%.*]], label [[CALL_SQRT:%.*]], !prof [[PROF1]]
 ; CHECK:       call.sqrt:
 ; CHECK-NEXT:    [[TMP2:%.*]] = tail call float @sqrtf(float [[VAL]]) #[[ATTR1:[0-9]+]]
 ; CHECK-NEXT:    br label [[DOTSPLIT]]
@@ -45,8 +45,13 @@ define float @f_readonly(float %val) {
 }
 
 declare float @sqrtf(float)
+
+!0 = !{!"function_entry_count", i32 10}
 ;.
 ; CHECK: attributes #[[ATTR0]] = { memory(none) }
 ; CHECK: attributes #[[ATTR1]] = { memory(write) }
 ; CHECK: attributes #[[ATTR2]] = { memory(read) }
 ;.
+; CHECK: [[META0:![0-9]+]] = !{!"function_entry_count", i32 10}
+; CHECK: [[PROF1]] = !{!"branch_weights", i32 1048575, i32 1}
+;.