[BOLT] Drop converting return profile to call cont (#129477)

The workaround was not implemented for BAT case, and it is no longer
needed with pre-aggregated traces, alternatively, the effect can be
achieved with `infer-fall-throughs` with old pre-aggregated format
(branches + ranges).

Test Plan: updated callcont-fallthru.s

Author: aaupov

bolt/include/bolt/Profile/DataAggregator.h

@@ -197,10 +197,6 @@ class DataAggregator : public DataReader {
 
   BoltAddressTranslation *BAT{nullptr};
 
-  /// Whether pre-aggregated profile needs to convert branch profile into call
-  /// to continuation fallthrough profile.
-  bool NeedsConvertRetProfileToCallCont{false};
-
   /// Update function execution profile with a recorded trace.
   /// A trace is region of code executed between two LBR entries supplied in
   /// execution order.

bolt/lib/Profile/DataAggregator.cpp

@@ -720,23 +720,6 @@ bool DataAggregator::doBranch(uint64_t From, uint64_t To, uint64_t Count,
                : isReturn(Func.disassembleInstructionAtOffset(Offset));
   };
 
-  // Returns whether \p Offset in \p Func may be a call continuation excluding
-  // entry points and landing pads.
-  auto checkCallCont = [&](const BinaryFunction &Func, const uint64_t Offset) {
-    // No call continuation at a function start.
-    if (!Offset)
-      return false;
-
-    // FIXME: support BAT case where the function might be in empty state
-    // (split fragments declared non-simple).
-    if (!Func.hasCFG())
-      return false;
-
-    // The offset should not be an entry point or a landing pad.
-    const BinaryBasicBlock *ContBB = Func.getBasicBlockAtOffset(Offset);
-    return ContBB && !ContBB->isEntryPoint() && !ContBB->isLandingPad();
-  };
-
   // Mutates \p Addr to an offset into the containing function, performing BAT
   // offset translation and parent lookup.
   //
@@ -749,8 +732,7 @@ bool DataAggregator::doBranch(uint64_t From, uint64_t To, uint64_t Count,
 
     Addr -= Func->getAddress();
 
-    bool IsRetOrCallCont =
-        IsFrom ? checkReturn(*Func, Addr) : checkCallCont(*Func, Addr);
+    bool IsRet = IsFrom && checkReturn(*Func, Addr);
 
     if (BAT)
       Addr = BAT->translate(Func->getAddress(), Addr, IsFrom);
@@ -761,24 +743,16 @@ bool DataAggregator::doBranch(uint64_t From, uint64_t To, uint64_t Count,
       NumColdSamples += Count;
 
     if (!ParentFunc)
-      return std::pair{Func, IsRetOrCallCont};
+      return std::pair{Func, IsRet};
 
-    return std::pair{ParentFunc, IsRetOrCallCont};
+    return std::pair{ParentFunc, IsRet};
   };
 
-  uint64_t ToOrig = To;
   auto [FromFunc, IsReturn] = handleAddress(From, /*IsFrom*/ true);
-  auto [ToFunc, IsCallCont] = handleAddress(To, /*IsFrom*/ false);
+  auto [ToFunc, _] = handleAddress(To, /*IsFrom*/ false);
   if (!FromFunc && !ToFunc)
     return false;
 
-  // Record call to continuation trace.
-  if (NeedsConvertRetProfileToCallCont && FromFunc != ToFunc &&
-      (IsReturn || IsCallCont)) {
-    LBREntry First{ToOrig - 1, ToOrig - 1, false};
-    LBREntry Second{ToOrig, ToOrig, false};
-    return doTrace(First, Second, Count);
-  }
   // Ignore returns.
   if (IsReturn)
     return true;
@@ -1235,21 +1209,14 @@ std::error_code DataAggregator::parseAggregatedLBREntry() {
   ErrorOr<StringRef> TypeOrErr = parseString(FieldSeparator);
   if (std::error_code EC = TypeOrErr.getError())
     return EC;
-  // Pre-aggregated profile with branches and fallthroughs needs to convert
-  // return profile into call to continuation fall-through.
-  auto Type = AggregatedLBREntry::BRANCH;
-  if (TypeOrErr.get() == "B") {
-    NeedsConvertRetProfileToCallCont = true;
+  auto Type = AggregatedLBREntry::TRACE;
+  if (LLVM_LIKELY(TypeOrErr.get() == "T")) {
+  } else if (TypeOrErr.get() == "B") {
     Type = AggregatedLBREntry::BRANCH;
   } else if (TypeOrErr.get() == "F") {
-    NeedsConvertRetProfileToCallCont = true;
     Type = AggregatedLBREntry::FT;
   } else if (TypeOrErr.get() == "f") {
-    NeedsConvertRetProfileToCallCont = true;
     Type = AggregatedLBREntry::FT_EXTERNAL_ORIGIN;
-  } else if (TypeOrErr.get() == "T") {
-    // Trace is expanded into B and [Ff]
-    Type = AggregatedLBREntry::TRACE;
   } else {
     reportError("expected T, B, F or f");
     return make_error_code(llvm::errc::io_error);

bolt/test/X86/callcont-fallthru.s

@@ -4,31 +4,12 @@
 # RUN: %clang %cflags -fpic -shared -xc /dev/null -o %t.so
 ## Link against a DSO to ensure PLT entries.
 # RUN: %clangxx %cxxflags %s %t.so -o %t -Wl,-q -nostdlib
-# RUN: link_fdata %s %t %t.pa1 PREAGG1
-# RUN: link_fdata %s %t %t.pa2 PREAGG2
-# RUN: link_fdata %s %t %t.pa3 PREAGG3
 # RUN: link_fdata %s %t %t.pat PREAGGT1
 # RUN: link_fdata %s %t %t.pat2 PREAGGT2
 # RUN: link_fdata %s %t %t.patplt PREAGGPLT
 
-## Check normal case: fallthrough is not LP or secondary entry.
 # RUN: llvm-strip --strip-unneeded %t -o %t.strip
 # RUN: llvm-objcopy --remove-section=.eh_frame %t.strip %t.noeh
-# RUN: llvm-bolt %t.strip --pa -p %t.pa1 -o %t.out \
-# RUN:   --print-cfg --print-only=main | FileCheck %s
-
-## Check that getFallthroughsInTrace correctly handles a trace starting at plt
-## call continuation
-# RUN: llvm-bolt %t.strip --pa -p %t.pa2 -o %t.out2 \
-# RUN:   --print-cfg --print-only=main | FileCheck %s --check-prefix=CHECK2
-
-## Check that we don't treat secondary entry points as call continuation sites.
-# RUN: llvm-bolt %t --pa -p %t.pa3 -o %t.out \
-# RUN:   --print-cfg --print-only=main | FileCheck %s --check-prefix=CHECK3
-
-## Check fallthrough to a landing pad case.
-# RUN: llvm-bolt %t.strip --pa -p %t.pa3 -o %t.out \
-# RUN:   --print-cfg --print-only=main | FileCheck %s --check-prefix=CHECK3
 
 ## Check pre-aggregated traces attach call continuation fallthrough count
 # RUN: llvm-bolt %t.noeh --pa -p %t.pat -o %t.out \
@@ -77,7 +58,6 @@ Ltmp0_br:
 ## Check PLT traces are accepted
 # PREAGGPLT: T #Ltmp0_br# #puts@plt# #puts@plt# 3
 ## Target is an external-origin call continuation
-# PREAGG1: B X:0 #Ltmp1# 2 0
 # PREAGGT1: T X:0 #Ltmp1# #Ltmp4_br# 2
 # CHECK:      callq puts@PLT
 # CHECK-NEXT: count: 2
@@ -97,18 +77,15 @@ Ltmp4_br:
 	movl	$0xa, -0x18(%rbp)
 	callq	foo
 ## Target is a binary-local call continuation
-# PREAGG1: B #Lfoo_ret# #Ltmp3# 1 0
 # PREAGGT1: T #Lfoo_ret# #Ltmp3# #Ltmp3_br# 1
 # CHECK:      callq foo
 # CHECK-NEXT: count: 1
 
 ## PLT call continuation fallthrough spanning the call
-# PREAGG2: F #Ltmp1# #Ltmp3_br# 3
 # CHECK2:      callq foo
 # CHECK2-NEXT: count: 3
 
 ## Target is a secondary entry point (unstripped) or a landing pad (stripped)
-# PREAGG3: B X:0 #Ltmp3# 2 0
 # PREAGGT2: T X:0 #Ltmp3# #Ltmp3_br# 2
 # CHECK3:      callq foo
 # CHECK3-NEXT: count: 0