[llvm-profgen] Filter out invalid debug line

[wlei]

Summary

There appears some big inline numbers in the symbolizer's output which is invalid for compiler to consume. See the example below:

400618: movl  $1, %edx                          main:4294967285
40061d: subl  %eax, %edx                        main:4294967285
40061f: addl  %ecx, %edx                        bar:1 @ main:3.2
40062b: addl  %ecx, %eax                        bar:1 @ main:3.2
40062d: movl  %eax, %esi                        bar:4294967292 @ main:3.2
40062f: addl  $1, %ecx                          main:2.2
...

Here changed to filter them out in a centralized way and also adjusted some total count samples due to it.

[hoy]

llvm/test/tools/llvm-profgen/cs-preinline.test

13

Wondering what is causing the total sample to change but the body samples stay the same? The lines filtered out should not have mapped to the same context previously?

[wenlei]

I think I've seen negative line offsets due to use of macro. If macro is used the line info may point to location where the macro is defined, then it can lead to negative line offset if we use function start line as the base.

In such cases, negative offset should still work for correlating profile back to IR.. What is the corresponding line offset you saw on relevant IR? Do they match the offset in the profile?

[wlei]

In D110081#3010969, @wenlei wrote:

I think I've seen negative line offsets due to use of macro. If macro is used the line info may point to location where the macro is defined, then it can lead to negative line offset if we use function start line as the base.

In such cases, negative offset should still work for correlating profile back to IR.. What is the corresponding line offset you saw on relevant IR? Do they match the offset in the profile?

As we discussed offline, we might have the invalid number in non-macro case and our internal tools also haven't covered this which can be a separate improvements later.

llvm/test/tools/llvm-profgen/cs-preinline.test

13

Sorry for the late reply.

This is because previously we add total sample even it's an invalid line number. With current change, the invalid line is filtered out at the early stage, so we then won't have them sum to the total number. This is causing the difference.

I think it should be minor impact by the difference since the invalid line is rare and total number itself is kind of a estimation(like it counts against all instructions but not the source code)

[hoy]

llvm/tools/llvm-profgen/ProfiledBinary.cpp

543

Can we potentially lose contexts when an invalid line offset is one of the frames? Like A:-1 @ B:2 @ C:3, without this change, we could still have samples for B:2 @ C:3. But I think that's rare.

LGTM.

[wlei]

llvm/tools/llvm-profgen/ProfiledBinary.cpp

543

Can we potentially lose contexts when an invalid line offset is one of the frames? Like A:-1 @ B:2 @ C:3, without this change, we could still have samples for B:2 @ C:3. But I think that's rare.

I haven't seen those cases, it seems it only happened for leaf frame. Even it can be leaf call, there is no samples hit the callee. I can add a warning on the non-leaf frame invalid line of stack address.

[wenlei]

In D110081#3028373, @wlei wrote:

In D110081#3010969, @wenlei wrote:

I think I've seen negative line offsets due to use of macro. If macro is used the line info may point to location where the macro is defined, then it can lead to negative line offset if we use function start line as the base.

In such cases, negative offset should still work for correlating profile back to IR.. What is the corresponding line offset you saw on relevant IR? Do they match the offset in the profile?

As we discussed offline, we might have the invalid number in non-macro case and our internal tools also haven't covered this which can be a separate improvements later.

Okay, as long as this matches the behavior of original tool, this looks good. I'm also curious how google's tool deal with such negative offset though..

[wlei]

In D110081#3035683, @wenlei wrote:

In D110081#3028373, @wlei wrote:

In D110081#3010969, @wenlei wrote:

I think I've seen negative line offsets due to use of macro. If macro is used the line info may point to location where the macro is defined, then it can lead to negative line offset if we use function start line as the base.

In such cases, negative offset should still work for correlating profile back to IR.. What is the corresponding line offset you saw on relevant IR? Do they match the offset in the profile?

As we discussed offline, we might have the invalid number in non-macro case and our internal tools also haven't covered this which can be a separate improvements later.

Okay, as long as this matches the behavior of original tool, this looks good. I'm also curious how google's tool deal with such negative offset though..

Yeah, it seems like google also ignore those negative offset : https://github.com/google/autofdo/blob/963a8c1f55ed86db6b909ee603a46742b3980139/source_info.h

// Return the offset using the line number and the discriminator.
static constexpr uint64_t GenerateOffset(uint32_t Line,
                                         uint32_t Discriminator) {
  return (static_cast<uint64_t>(Line) << 32) |
         static_cast<uint64_t>(Discriminator);
}

// Return the line number from the offset.
static constexpr uint32_t GetLineNumberFromOffset(uint64_t Offset) {
  return (Offset >> 32) & 0xffff;
}

llvm/test/tools/llvm-profgen/cs-preinline.test

 ; Test default llvm-profgen with preinline off
 ; RUN: llvm-profgen --format=text --perfscript=%S/Inputs/inline-cs-noprobe.perfscript --binary=%S/Inputs/inline-cs-noprobe.perfbin --output=%t
 ; RUN: FileCheck %s --input-file %t --check-prefix=CHECK-DEFAULT
 
 ; Test llvm-profgen with preinliner on will merge not inlinable profile into base profile.
 ; RUN: llvm-profgen --format=text --perfscript=%S/Inputs/inline-cs-noprobe.perfscript --binary=%S/Inputs/inline-cs-noprobe.perfbin --output=%t --csspgo-preinliner=1
 ; RUN: FileCheck %s --input-file %t --check-prefix=CHECK-PREINL
 
 ; Test preinliner threshold that prevents all possible inlining and merges everything into base profile.
 ; RUN: llvm-profgen --format=text --perfscript=%S/Inputs/inline-cs-noprobe.perfscript --binary=%S/Inputs/inline-cs-noprobe.perfbin --output=%t --csspgo-preinliner=1  -sample-profile-cold-inline-threshold=0
 ; RUN: FileCheck %s --input-file %t --check-prefix=CHECK-NO-PREINL
 
-; CHECK-DEFAULT:     [main:1 @ foo]:309:0
+; CHECK-DEFAULT:     [main:1 @ foo]:225:0
 ; CHECK-DEFAULT-NEXT: 2.1: 14
 ; CHECK-DEFAULT-NEXT: 3: 15
 ; CHECK-DEFAULT-NEXT: 3.1: 14 bar:14
 ; CHECK-DEFAULT-NEXT: 3.2: 1
 ; CHECK-DEFAULT-NEXT: !Attributes: 1
 ; CHECK-DEFAULT-NEXT:[main:1 @ foo:3.1 @ bar]:84:0
 ; CHECK-DEFAULT-NEXT: 1: 14
 ; CHECK-DEFAULT-NEXT: !Attributes: 1
 
-; CHECK-PREINL:     [foo]:309:0
+; CHECK-PREINL:     [foo]:225:0
 ; CHECK-PREINL-NEXT: 2.1: 14
 ; CHECK-PREINL-NEXT: 3: 15
 ; CHECK-PREINL-NEXT: 3.1: 14 bar:14
 ; CHECK-PREINL-NEXT: 3.2: 1
 ; CHECK-PREINL-NEXT: !Attributes: 1
 ; CHECK-PREINL-NEXT:[foo:3.1 @ bar]:84:0
 ; CHECK-PREINL-NEXT: 1: 14
 ; CHECK-PREINL-NEXT: !Attributes: 3
 
-; CHECK-NO-PREINL:     [foo]:309:0
+; CHECK-NO-PREINL:     [foo]:225:0
 ; CHECK-NO-PREINL-NEXT: 2.1: 14
 ; CHECK-NO-PREINL-NEXT: 3: 15
 ; CHECK-NO-PREINL-NEXT: 3.1: 14 bar:14
 ; CHECK-NO-PREINL-NEXT: 3.2: 1
 ; CHECK-NO-PREINL-NEXT: !Attributes: 1
 ; CHECK-NO-PREINL-NEXT:[bar]:84:0
 ; CHECK-NO-PREINL-NEXT: 1: 14
 ; CHECK-NO-PREINL-NEXT: !Attributes: 1

llvm/test/tools/llvm-profgen/duplication-factor.test

 ; RUN: llvm-profgen --format=text --perfscript=%S/Inputs/duplication-factor.perfscript --binary=%S/Inputs/duplication-factor.perfbin --output=%t --profile-summary-cold-count=0
 ; RUN: FileCheck %s --input-file %t
 ; RUN: llvm-profgen --format=text --perfscript=%S/Inputs/duplication-factor.perfscript --binary=%S/Inputs/duplication-factor.perfbin --output=%t --strip-context
 ; RUN: FileCheck %s --input-file %t --check-prefix=CHECK-STRIP-CTX
 
 ; Test when the duplication factor is 10
 
-; CHECK:[main:1 @ foo]:1716:0
+; CHECK:[main:1 @ foo]:1576:0
 ; CHECK: 2.1: 150
 ; CHECK: 2.2: 160
 ; CHECK: 2.3: 176 bar:16
 
-; CHECK:[main:1 @ foo:2.3 @ bar]:192:0
+; CHECK:[main:1 @ foo:2.3 @ bar]:176:0
 ; CHECK: 1: 16
 
-; CHECK-STRIP-CTX: main:1892:0
+; CHECK-STRIP-CTX: main:1736:0
 ; CHECK-STRIP-CTX:  0: 0
 ; CHECK-STRIP-CTX:  2: 0
-; CHECK-STRIP-CTX:  1: foo:1892
+; CHECK-STRIP-CTX:  1: foo:1736
 ; CHECK-STRIP-CTX:   2.1: 150
 ; CHECK-STRIP-CTX:   2.2: 160
 ; CHECK-STRIP-CTX:   2.3: 160
 ; CHECK-STRIP-CTX:   2.4: 0
 ; CHECK-STRIP-CTX:   3: 0
-; CHECK-STRIP-CTX:   2.3: bar:192
+; CHECK-STRIP-CTX:   2.3: bar:176
 ; CHECK-STRIP-CTX:    1: 16
 
 ; Here we manually generate the duplication-factor in discriminator for the test.
 ; clang -O3 -g -fdebug-info-for-profiling test.c -emit-llvm -S -o test.ll
 ; # Manually update the discriminator for a duplication-factor whose value is 10 in test.ll. For example: a discriminator value 2(0010) will be changed to 2562(101000000010)
 ; clang test.ll
 
 

llvm/test/tools/llvm-profgen/fname-canonicalization.test

 ; Test unique linkage name decoding and canonicalization from dwarf debug info
 ; RUN: llvm-profgen --format=text --perfscript=%S/Inputs/unique-linkage-name-dwarf.perfscript --binary=%S/Inputs/unique-linkage-name-dwarf.perfbin --output=%t --profile-summary-cold-count=0
 ; RUN: FileCheck %s --input-file %t --check-prefix=CHECK-DWARF-FNAME
 
-; CHECK-DWARF-FNAME:[main:1 @ foo]:309:0
+; CHECK-DWARF-FNAME:[main:1 @ foo]:225:0
 ; CHECK-DWARF-FNAME: 2: 14
 ; CHECK-DWARF-FNAME: 3: 29 bar.__uniq.26267048767521081047744692097241227776:14
 ; CHECK-DWARF-FNAME:[main:1 @ foo:3 @ bar.__uniq.26267048767521081047744692097241227776]:84:0
 ; CHECK-DWARF-FNAME: 1: 14
 
 
 ; Test unique linkage name decoding and canonicalization from pseudo probe info
 ; RUN: llvm-profgen --format=text --perfscript=%S/Inputs/unique-linkage-name-probe.perfscript --binary=%S/Inputs/unique-linkage-name-probe.perfbin --output=%t --profile-summary-cold-count=0

llvm/test/tools/llvm-profgen/inline-cs-noprobe.test

 ; RUN: llvm-profgen --format=text --perfscript=%S/Inputs/inline-cs-noprobe.perfscript --binary=%S/Inputs/inline-cs-noprobe.perfbin --output=%t --skip-symbolization --profile-summary-cold-count=0
 | FileCheck %s --input-file %t --check-prefix=CHECK-UNWINDER
 ; RUN: llvm-profgen --format=text --perfscript=%S/Inputs/inline-cs-noprobe.perfscript --binary=%S/Inputs/inline-cs-noprobe.perfbin --output=%t --profile-summary-cold-count=0
 ; RUN: FileCheck %s --input-file %t
 ; RUN: llvm-profgen --format=text --perfscript=%S/Inputs/inline-cs-noprobe.perfscript --binary=%S/Inputs/inline-cs-noprobe.perfbin --output=%t --profile-summary-cold-count=0 --strip-context
 ; RUN: FileCheck %s --input-file %t --check-prefix=CHECK-STRIP-CTX
 
-; CHECK:[main:1 @ foo]:309:0
+; CHECK:[main:1 @ foo]:225:0
 ; CHECK: 2.1: 14
 ; CHECK: 3: 15
 ; CHECK: 3.1: 14 bar:14
 ; CHECK: 3.2: 1
 ; CHECK:[main:1 @ foo:3.1 @ bar]:84:0
 ; CHECK: 1: 14
 
-; CHECK-STRIP-CTX: main:379:0
+; CHECK-STRIP-CTX: main:295:0
 ; CHECK-STRIP-CTX:  0: 0
 ; CHECK-STRIP-CTX:  2: 0
-; CHECK-STRIP-CTX:  1: foo:379
+; CHECK-STRIP-CTX:  1: foo:295
 ; CHECK-STRIP-CTX:   2.1: 14
 ; CHECK-STRIP-CTX:   3: 15
 ; CHECK-STRIP-CTX:   3.2: 1
 ; CHECK-STRIP-CTX:   4: 0
 ; CHECK-STRIP-CTX:   7: 0
 ; CHECK-STRIP-CTX:   3.1: bar:84
 ; CHECK-STRIP-CTX:    1: 14
 

llvm/test/tools/llvm-profgen/inline-noprobe.test

 ; RUN: llvm-profgen --format=text --perfscript=%S/Inputs/inline-noprobe.perfscript --binary=%S/Inputs/inline-noprobe.perfbin --output=%t --skip-symbolization
 ; RUN: FileCheck %s --input-file %t --check-prefix=CHECK-RAW-PROFILE
 ; RUN: llvm-profgen --format=text --perfscript=%S/Inputs/inline-noprobe.perfscript --binary=%S/Inputs/inline-noprobe.perfbin --output=%t
 ; RUN: FileCheck %s --input-file %t --check-prefix=CHECK
 
-CHECK: main:836:0
+CHECK: main:669:0
 CHECK:  0: 0
 CHECK:  2: 0
-CHECK:  1: foo:836
+CHECK:  1: foo:669
 CHECK:   2.1: 42
 CHECK:   3: 62
 CHECK:   3.2: 21
 CHECK:   4: 0
-CHECK:   3.1: bar:252
+CHECK:   3.1: bar:210
 CHECK:    1: 42
 CHECK:   3.2: bar:63
 CHECK:    1: 21
 
 CHECK-RAW-PROFILE:      3
 CHECK-RAW-PROFILE-NEXT: 650-691:21
 CHECK-RAW-PROFILE-NEXT: 669-677:20
 CHECK-RAW-PROFILE-NEXT: 669-691:21

llvm/test/tools/llvm-profgen/multi-load-segs.test

 ; RUN: llvm-profgen --perfscript=%S/Inputs/multi-load-segs.perfscript --binary=%S/Inputs/multi-load-segs.perfbin --output=%t --format=text
 ; RUN: FileCheck %s --input-file %t
 
-;;  %S/Inputs/multi-load-segs.perfbin is an ELF image with two executable load segments. 
+;;  %S/Inputs/multi-load-segs.perfbin is an ELF image with two executable load segments.
 ; running llvm-readelf -l %S/Inputs/multi-load-segs.perfbin gives:
 ;;  LOAD           0x000000 0x0000000000200000 0x0000000000200000 0x00075c 0x00075c R   0x1000
 ;;  LOAD           0x000760 0x0000000000201760 0x0000000000201760 0x0004c0 0x0004c0 R E 0x1000
 ;;  LOAD           0x000c20 0x0000000000202c20 0x0000000000202c20 0x0001f0 0x0001f0 RW  0x1000
 ;;  LOAD           0x000e10 0x0000000000203e10 0x0000000000203e10 0x000040 0x000058 RW  0x1000
 ;;  LOAD           0x200000 0x0000000000400000 0x0000000000400000 0x0005e8 0x0005e8 R E 0x200000
 
-; CHECK:      [main:2 @ _Z10sort_arrayv:6 @ _Z11bubble_sortPii]:465:0
+; CHECK:      [main:2 @ _Z10sort_arrayv:6 @ _Z11bubble_sortPii]:434:0
 ; CHECK-NEXT:  4: 31
 ; CHECK-NEXT:  5: 31
 ; CHECK-NEXT:  7: 31
 ; CHECK-NEXT:  8: 31
 ; CHECK-NEXT:    !Attributes: 1

llvm/test/tools/llvm-profgen/recursion-compression-noprobe.test

 ; Firstly test uncompression(--compress-recursion=0)
 ; RUN: llvm-profgen --format=text --perfscript=%S/Inputs/recursion-compression-noprobe.perfscript --binary=%S/Inputs/recursion-compression-noprobe.perfbin --output=%t --compress-recursion=0 --profile-summary-cold-count=0
 ; RUN: FileCheck %s --input-file %t -check-prefix=CHECK-UNCOMPRESS
 ; RUN: llvm-profgen --format=text --perfscript=%S/Inputs/recursion-compression-noprobe.perfscript --binary=%S/Inputs/recursion-compression-noprobe.perfbin --output=%t --profile-summary-cold-count=0
 ; RUN: FileCheck %s --input-file %t
 ; RUN: llvm-profgen --format=text --perfscript=%S/Inputs/recursion-compression-noprobe.perfscript --binary=%S/Inputs/recursion-compression-noprobe.perfbin --output=%t --compress-recursion=0 --profile-summary-cold-count=0 --csprof-max-context-depth=2
 ; RUN: FileCheck %s --input-file %t -check-prefix=CHECK-MAX-CTX-DEPTH
 
-; CHECK-UNCOMPRESS:[main:1 @ foo:3 @ fa:2 @ fb]:48:0
+; CHECK-UNCOMPRESS:[main:1 @ foo:3 @ fa:2 @ fb]:37:0
 ; CHECK-UNCOMPRESS: 1: 11
 ; CHECK-UNCOMPRESS: 2: 1 fa:1
-; CHECK-UNCOMPRESS:[main:1 @ foo:3 @ fa]:24:0
+; CHECK-UNCOMPRESS:[main:1 @ foo:3 @ fa]:21:0
 ; CHECK-UNCOMPRESS: 1: 1
 ; CHECK-UNCOMPRESS: 2: 13 fb:11
 ; CHECK-UNCOMPRESS:[main:1 @ foo]:7:0
 ; CHECK-UNCOMPRESS: 2: 1
 ; CHECK-UNCOMPRESS: 3: 2 fa:1
-; CHECK-UNCOMPRESS:[main:1 @ foo:3 @ fa:2 @ fb:2 @ fa]:7:0
+; CHECK-UNCOMPRESS:[main:1 @ foo:3 @ fa:2 @ fb:2 @ fa]:6:0
 ; CHECK-UNCOMPRESS: 1: 1
-; CHECK-UNCOMPRESS: 2: 2 fb:1
-; CHECK-UNCOMPRESS:[main:1 @ foo:3 @ fa:2 @ fb:2 @ fa:2 @ fb]:2:0
-; CHECK-UNCOMPRESS: 2: 1 fa:1
+; CHECK-UNCOMPRESS: 2: 1
 ; CHECK-UNCOMPRESS:[main:1 @ foo:3 @ fa:2 @ fb:2 @ fa:2 @ fb:2 @ fa]:2:0
 ; CHECK-UNCOMPRESS: 4: 1
 
-; CHECK-MAX-CTX-DEPTH:[foo:3 @ fa:2 @ fb]:47:0
+; CHECK-MAX-CTX-DEPTH:[foo:3 @ fa:2 @ fb]:36:0
 ; CHECK-MAX-CTX-DEPTH: 1: 11
-; CHECK-MAX-CTX-DEPTH:[main:1 @ foo:3 @ fa]:13:0
+; CHECK-MAX-CTX-DEPTH:[main:1 @ foo:3 @ fa]:10:0
 ; CHECK-MAX-CTX-DEPTH: 1: 1
 ; CHECK-MAX-CTX-DEPTH: 2: 2
 ; CHECK-MAX-CTX-DEPTH:[fa:2 @ fb:2 @ fa]:8:0
 ; CHECK-MAX-CTX-DEPTH: 1: 1
 ; CHECK-MAX-CTX-DEPTH: 2: 1
 ; CHECK-MAX-CTX-DEPTH: 4: 1
 ; CHECK-MAX-CTX-DEPTH:[main:1 @ foo]:7:0
 ; CHECK-MAX-CTX-DEPTH: 2: 1
 ; CHECK-MAX-CTX-DEPTH: 3: 2 fa:1
-; CHECK-MAX-CTX-DEPTH:[fb:2 @ fa:2 @ fb]:1:0
 
 
-; CHECK: [main:1 @ foo:3 @ fa:2 @ fb]:48:0
+; CHECK: [main:1 @ foo:3 @ fa:2 @ fb]:37:0
 ; CHECK:  1: 11
 ; CHECK:  2: 1 fa:1
-; CHECK: [main:1 @ foo:3 @ fa]:24:0
+; CHECK: [main:1 @ foo:3 @ fa]:21:0
 ; CHECK:  1: 1
 ; CHECK:  2: 13 fb:11
-; CHECK: [main:1 @ foo:3 @ fa:2 @ fb:2 @ fa]:9:0
+; CHECK: [main:1 @ foo:3 @ fa:2 @ fb:2 @ fa]:8:0
 ; CHECK:  1: 1
-; CHECK:  2: 2 fb:1
+; CHECK:  2: 1
 ; CHECK:  4: 1
 ; CHECK: [main:1 @ foo]:7:0
 ; CHECK:  2: 1
 ; CHECK:  3: 2 fa:1
-; CHECK: [main:1 @ foo:3 @ fa:2 @ fb:2 @ fa:2 @ fb]:1:0
-
 
 ; original code:
 ; clang -O3 -g test.c -o a.out
 #include <stdio.h>
 
 int fb(int n) {
   if(n > 10) return fb(n / 2);
   return fa(n - 1);

llvm/tools/llvm-profgen/ProfileGenerator.cpp

void CSProfileGenerator::generateProfile() {
   populateInferredFunctionSamples();
 
   postProcessProfiles();
 }
 
 void ProfileGenerator::updateBodySamplesforFunctionProfile(
     FunctionSamples &FunctionProfile, const SampleContextFrame &LeafLoc,
     uint64_t Count) {
-  // Filter out invalid negative(int type) lineOffset
-  if (LeafLoc.Callsite.LineOffset & 0x80000000)
-    return;
   // Use the maximum count of samples with same line location
   uint32_t Discriminator = getDiscriminator(LeafLoc.Callsite.Discriminator);
   ErrorOr<uint64_t> R =
       FunctionProfile.findSamplesAt(LeafLoc.Callsite.LineOffset, Discriminator);
 
   uint64_t PreviousCount = R ? R.get() : 0;
   if (PreviousCount <= Count) {
     FunctionProfile.addBodySamples(LeafLoc.Callsite.LineOffset, Discriminator,

llvm/tools/llvm-profgen/ProfiledBinary.cpp

if (UseCanonicalFnName)
       FunctionName = FunctionSamples::getCanonicalFnName(FunctionName);
 
     uint32_t Discriminator = CallerFrame.Discriminator;
     uint32_t LineOffset = CallerFrame.Line - CallerFrame.StartLine;
     if (UseProbeDiscriminator) {
       LineOffset =
           PseudoProbeDwarfDiscriminator::extractProbeIndex(Discriminator);
       Discriminator = 0;
+    } else {
+      // Filter out invalid negative(int type) lineOffset
+      if (LineOffset & 0x80000000)
+        return SampleContextFrameVector();
     }
 
     LineLocation Line(LineOffset, Discriminator);
     auto It = NameStrings.insert(FunctionName.str());
     CallStack.emplace_back(*It.first, Line);
   }
 
   return CallStack;