Offset CallContLPs by MaxInputOffset + 1

Created using spr 1.3.4

Author: aaupov

bolt/lib/Profile/BoltAddressTranslation.cpp

@@ -87,6 +87,11 @@ void BoltAddressTranslation::write(const BinaryContext &BC, raw_ostream &OS) {
       continue;
 
     uint32_t NumSecondaryEntryPoints = 0;
+    // Offset call continuation landing pads by max input offset + 1 to prevent
+    // confusing them with real entry points. Note we can't use the input size
+    // as it's not available in BOLTed binary.
+    const BBHashMapTy &BBHashMap = getBBHashMap(InputAddress);
+    const uint32_t CallContLPOffset = std::prev(BBHashMap.end())->first + 1;
     for (const BinaryBasicBlock &BB : llvm::drop_begin(Function)) {
       if (BB.isEntryPoint()) {
         ++NumSecondaryEntryPoints;
@@ -104,8 +109,8 @@ void BoltAddressTranslation::write(const BinaryContext &BC, raw_ostream &OS) {
       if (!Instr || !BC.MIB->isCall(*Instr))
         continue;
       ++NumSecondaryEntryPoints;
-      SecondaryEntryPointsMap[OutputAddress].push_back(
-          Function.getOutputSize() + BB.getOffset());
+      SecondaryEntryPointsMap[OutputAddress].push_back(CallContLPOffset +
+                                                       BB.getOffset());
     }
     if (NumSecondaryEntryPoints)
       llvm::sort(SecondaryEntryPointsMap[OutputAddress]);
@@ -124,7 +129,6 @@ void BoltAddressTranslation::write(const BinaryContext &BC, raw_ostream &OS) {
     // Add entries for deleted blocks. They are still required for correct BB
     // mapping of branches modified by SCTC. By convention, they would have the
     // end of the function as output address.
-    const BBHashMapTy &BBHashMap = getBBHashMap(InputAddress);
     if (BBHashMap.size() != Function.size()) {
       const uint64_t EndOffset = Function.getOutputSize();
       std::unordered_set<uint32_t> MappedInputOffsets;

bolt/lib/Profile/DataAggregator.cpp

@@ -810,10 +810,13 @@ bool DataAggregator::doBranch(uint64_t From, uint64_t To, uint64_t Count,
 
     if (!Func.hasCFG()) {
       const uint64_t Address = Func.getAddress();
+      if (!BAT)
+        return false;
+      const uint32_t InputOffset = BAT->translate(Address, Offset, false);
       // Check if offset is a secondary entry point or a call continuation
       // landing pad (offset shifted by function size).
-      return BAT && !BAT->getSecondaryEntryPointId(Address, Offset) &&
-             !BAT->getSecondaryEntryPointId(Address, Func.getSize() + Offset);
+      return !BAT->getSecondaryEntryPointId(Address, InputOffset) &&
+             !BAT->getSecondaryEntryPointId(Address, Func.getSize() + InputOffset);
     }
 
     // The offset should not be an entry point or a landing pad.

bolt/test/X86/callcont-fallthru.s

@@ -21,34 +21,40 @@
 # RUN:   --print-cfg --print-only=main | FileCheck %s --check-prefix=CHECK3
 
 ## Check fallthrough to a landing pad case.
-# RUN: llvm-bolt %t.exe --pa -p %t.pa3 -o %t.out4 --enable-bat \
+# RUN: llvm-bolt %t.exe --pa -p %t.pa3 -o %t.out4 \
 # RUN:   --print-cfg --print-only=main | FileCheck %s --check-prefix=CHECK3
 
-## Check that a landing pad is emitted in BAT
-# RUN: llvm-bat-dump %t.out4 --dump-all | FileCheck %s --check-prefix=CHECK-BAT
+# RUN: llvm-mc -filetype=obj -triple x86_64-unknown-unknown --defsym GLOBL=1 \
+# RUN:   %s -o %t.o
+# RUN: %clangxx %cxxflags %t.o -o %t2 -Wl,-q -nostdlib 
+# RUN: llvm-bolt %t2 --pa -p %t.pa3 -o %t.bat --enable-bat \
+# RUN:   --print-cfg --print-only=main | FileCheck %s --check-prefix=CHECK3
 
-# CHECK-BAT:      1 secondary entry points:
+## Check that a landing pad is emitted in BAT
+# RUN: llvm-bat-dump %t.bat --dump-all | FileCheck %s --check-prefix=CHECK-BAT
 
-## Check BAT case of a fallthrough to a call continuation
-# link_fdata %s %t.out4 %t.pa.bat PREAGG
-# RUN: perf2bolt %t.out4 -p %t.pa.bat --pa -o %t.fdata
-# RUN: FileCheck %s --check-prefix=CHECK-BAT-CC --input-file=%t.fdata
-# CHECK-BAT-CC: main
+# CHECK-BAT:      secondary entry points:
 
 ## Check BAT case of a fallthrough to a secondary entry point or a landing pad
-# link_fdata %s %t.out4 %t.pa.bat2 PREAGG3
+# RUN: link_fdata %s %t.bat %t.pa.bat2 PREAGG3
 
 ## Secondary entry
-# RUN: perf2bolt %t.out4 -p %t.pa.bat2 --pa -o %t.fdata2
+# RUN: perf2bolt %t.bat -p %t.pa.bat2 --pa -o %t.fdata2
 # RUN: FileCheck %s --check-prefix=CHECK-BAT-ENTRY --input-file=%t.fdata2
 # CHECK-BAT-ENTRY: main
 
 ## Landing pad
-# RUN: llvm-strip --strip-unneeded %t.out4
-# RUN: perf2bolt %t.out4 -p %t.pa.bat2 --pa -o %t.fdata3
+# RUN: llvm-strip --strip-unneeded %t.bat
+# RUN: perf2bolt %t.bat -p %t.pa.bat2 --pa -o %t.fdata3
 # RUN: FileCheck %s --check-prefix=CHECK-BAT-LP --input-file=%t.fdata3
 # CHECK-BAT-LP: main
 
+## Check BAT case of a fallthrough to a call continuation
+# link_fdata %s %t.bat %t.pa.bat PREAGG
+# RUN: perf2bolt %t.bat -p %t.pa.bat --pa -o %t.fdata
+# RUN: FileCheck %s --check-prefix=CHECK-BAT-CC --input-file=%t.fdata
+# CHECK-BAT-CC: main
+
   .globl foo
   .type foo, %function
 foo:
@@ -107,6 +113,9 @@ Ltmp4:
 # CHECK3:      callq foo
 # CHECK3-NEXT: count: 0
 
+.ifdef GLOBL
+.globl Ltmp3
+.endif
 Ltmp3:
 	cmpl	$0x0, -0x18(%rbp)
 Ltmp3_br: