thr3ads.net - llvm dev - [llvm-dev] Interleaved debug info on arm [Sep 2017]

If this information is useful, please help other people find it:
Share via:

Carlo Kok via llvm-dev

2017-Sep-18 17:45 UTC

[llvm-dev] Interleaved debug info on arm

When compiling code with lld -O0 --lto-O0 --eh-frame-hdr I get strange
interleaved line info all
over the place:
...
0x00000000000595ac 22 11 1 0 0 is_stmt 
0x00000000000595bc 25 7 1 0 0 is_stmt <<<
0x00000000000595c0 22 11 1 0 0 is_stmt 
0x00000000000595c4 25 7 1 0 0 is_stmt <<<
0x00000000000595c8 26 7 1 0 0 is_stmt

but the code only has 1 reference to line 25 and the references to line 22 all
precede that:

; Function Attrs: uwtable
define hidden void @ms_t1d_ConsoleApplication479.Program4_Testni32(i32) #0 !dbg
!34 {
  %2 = alloca i32
  store i32 %0, i32* %2
  call void @llvm.dbg.declare(metadata i32* %2, metadata !61, metadata !19),
!dbg !63
  %3 = alloca double, !dbg !63
  store double 0.000000e+00, double* %3, !dbg !63
  %4 = alloca %WriteLineObject*, !dbg !63
  store %WriteLineObject* null, %WriteLineObject** %4, !dbg !63
  br label %5, !dbg !63

; <label>:5:                                      ; preds = %1
  call void @llvm.dbg.declare(metadata double* %3, metadata !64, metadata !19),
!dbg !68
  call void @llvm.dbg.declare(metadata %WriteLineObject** %4, metadata !69,
metadata !19), !dbg !68
  store double 3.140000e+00, double* %3, !dbg !71
  %6 = call i8* @__newinst(i8* bitcast ({ { i64, [30 x i8]*, i8*, [141 x i8]*, {
i8*, void (%._ConsoleApplication479.Program*)*, void (i32)*, i8*,
%WriteLineString* (%._ConsoleApplication479.Program*)*, i32 ()*, i8** }* }*,
i8*, i8*, i8*, [16 x i8], i8*, i8*, i8*, i8* }*
@_rtti_t1d_ConsoleApplication479.Program to i8*), i32 16), !dbg !72
  %7 = bitcast i8* %6 to %._ConsoleApplication479.Program*, !dbg !72
  call void
@mi_t1d_ConsoleApplication479.Program5_.ctor(%._ConsoleApplication479.Program*
%7) #0, !dbg !72
  %8 = bitcast %._ConsoleApplication479.Program* %7 to %WriteLineObject*, !dbg
!72
  store %WriteLineObject* %8, %WriteLineObject** %4, !dbg !72
  call void @WriteLine(%WriteLineString* bitcast ({ i8*, i32, [14 x i8] }*
@.str2 to %WriteLineString*)) #1, !dbg !73
  %9 = load double, double* %3, !dbg !74
  %10 = sitofp i32 1 to double, !dbg !74
  %11 = fadd double %9, %10, !dbg !74
  store double %11, double* %3, !dbg !74
  %12 = load double, double* %3, !dbg !75
  %13 = sitofp i32 1 to double, !dbg !75
  %14 = fadd double %12, %13, !dbg !75
  store double %14, double* %3, !dbg !75
  ret void, !dbg !76
}



!72 = !DILocation(line: 22, column: 11, scope: !65)
!73 = !DILocation(line: 25, column: 7, scope: !65)

full bitcode: https://gist.github.com/carlokok/aa3fab175f8afe8e4776ad6c2f5cc80a

disasm: https://gist.github.com/carlokok/ab265461b37f87a10e0853ffc4d93c77

What could possibly cause this? It's wreaking havoc with the debugger
experience.

Carlo Kok
RemObjects Software

Adrian Prantl via llvm-dev

2017-Sep-18 18:17 UTC

head link

[llvm-dev] Interleaved debug info on arm

Have you tried looking at the --dump-after-all output to see which pass (if any)
is adding the extra line table entries?
-- adrian> On Sep 18, 2017, at 10:45 AM, Carlo Kok via llvm-dev <llvm-dev at
lists.llvm.org> wrote:
> 
> When compiling code with lld -O0 --lto-O0 --eh-frame-hdr I get strange
interleaved line info all
> over the place:
> ...
> 0x00000000000595ac 22 11 1 0 0 is_stmt 
> 0x00000000000595bc 25 7 1 0 0 is_stmt <<<
> 0x00000000000595c0 22 11 1 0 0 is_stmt 
> 0x00000000000595c4 25 7 1 0 0 is_stmt <<<
> 0x00000000000595c8 26 7 1 0 0 is_stmt
> 
> but the code only has 1 reference to line 25 and the references to line 22
all precede that:
> 
> ; Function Attrs: uwtable
> define hidden void @ms_t1d_ConsoleApplication479.Program4_Testni32(i32) #0
!dbg !34 {
>  %2 = alloca i32
>  store i32 %0, i32* %2
>  call void @llvm.dbg.declare(metadata i32* %2, metadata !61, metadata !19),
!dbg !63
>  %3 = alloca double, !dbg !63
>  store double 0.000000e+00, double* %3, !dbg !63
>  %4 = alloca %WriteLineObject*, !dbg !63
>  store %WriteLineObject* null, %WriteLineObject** %4, !dbg !63
>  br label %5, !dbg !63
> 
> ; <label>:5:                                      ; preds = %1
>  call void @llvm.dbg.declare(metadata double* %3, metadata !64, metadata
!19), !dbg !68
>  call void @llvm.dbg.declare(metadata %WriteLineObject** %4, metadata !69,
metadata !19), !dbg !68
>  store double 3.140000e+00, double* %3, !dbg !71
>  %6 = call i8* @__newinst(i8* bitcast ({ { i64, [30 x i8]*, i8*, [141 x
i8]*, { i8*, void (%._ConsoleApplication479.Program*)*, void (i32)*, i8*,
%WriteLineString* (%._ConsoleApplication479.Program*)*, i32 ()*, i8** }* }*,
i8*, i8*, i8*, [16 x i8], i8*, i8*, i8*, i8* }*
@_rtti_t1d_ConsoleApplication479.Program to i8*), i32 16), !dbg !72
>  %7 = bitcast i8* %6 to %._ConsoleApplication479.Program*, !dbg !72
>  call void
@mi_t1d_ConsoleApplication479.Program5_.ctor(%._ConsoleApplication479.Program*
%7) #0, !dbg !72
>  %8 = bitcast %._ConsoleApplication479.Program* %7 to %WriteLineObject*,
!dbg !72
>  store %WriteLineObject* %8, %WriteLineObject** %4, !dbg !72
>  call void @WriteLine(%WriteLineString* bitcast ({ i8*, i32, [14 x i8] }*
@.str2 to %WriteLineString*)) #1, !dbg !73
>  %9 = load double, double* %3, !dbg !74
>  %10 = sitofp i32 1 to double, !dbg !74
>  %11 = fadd double %9, %10, !dbg !74
>  store double %11, double* %3, !dbg !74
>  %12 = load double, double* %3, !dbg !75
>  %13 = sitofp i32 1 to double, !dbg !75
>  %14 = fadd double %12, %13, !dbg !75
>  store double %14, double* %3, !dbg !75
>  ret void, !dbg !76
> }
> 
> 
> 
> !72 = !DILocation(line: 22, column: 11, scope: !65)
> !73 = !DILocation(line: 25, column: 7, scope: !65)
> 
> full bitcode:
https://gist.github.com/carlokok/aa3fab175f8afe8e4776ad6c2f5cc80a
> 
> disasm: https://gist.github.com/carlokok/ab265461b37f87a10e0853ffc4d93c77
> 
> What could possibly cause this? It's wreaking havoc with the debugger
experience.
> 
> Carlo Kok
> RemObjects Software
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev

Rui Ueyama via llvm-dev

2017-Sep-19 01:01 UTC

head link

[llvm-dev] Interleaved debug info on arm

Does this happen only when you are using lld? Please try gold to see if it
works as you expected.

On Mon, Sep 18, 2017 at 10:45 AM, Carlo Kok via llvm-dev <
llvm-dev at lists.llvm.org> wrote:
> When compiling code with lld -O0 --lto-O0 --eh-frame-hdr I get strange
> interleaved line info all
> over the place:
> ...
> 0x00000000000595ac 22 11 1 0 0 is_stmt
> 0x00000000000595bc 25 7 1 0 0 is_stmt <<<
> 0x00000000000595c0 22 11 1 0 0 is_stmt
> 0x00000000000595c4 25 7 1 0 0 is_stmt <<<
> 0x00000000000595c8 26 7 1 0 0 is_stmt
>
> but the code only has 1 reference to line 25 and the references to line 22
> all precede that:
>
> ; Function Attrs: uwtable
> define hidden void @ms_t1d_ConsoleApplication479.Program4_Testni32(i32)
> #0 !dbg !34 {
>   %2 = alloca i32
>   store i32 %0, i32* %2
>   call void @llvm.dbg.declare(metadata i32* %2, metadata !61, metadata
> !19), !dbg !63
>   %3 = alloca double, !dbg !63
>   store double 0.000000e+00, double* %3, !dbg !63
>   %4 = alloca %WriteLineObject*, !dbg !63
>   store %WriteLineObject* null, %WriteLineObject** %4, !dbg !63
>   br label %5, !dbg !63
>
> ; <label>:5:                                      ; preds = %1
>   call void @llvm.dbg.declare(metadata double* %3, metadata !64, metadata
> !19), !dbg !68
>   call void @llvm.dbg.declare(metadata %WriteLineObject** %4, metadata
> !69, metadata !19), !dbg !68
>   store double 3.140000e+00, double* %3, !dbg !71
>   %6 = call i8* @__newinst(i8* bitcast ({ { i64, [30 x i8]*, i8*, [141 x
> i8]*, { i8*, void (%._ConsoleApplication479.Program*)*, void (i32)*, i8*,
> %WriteLineString* (%._ConsoleApplication479.Program*)*, i32 ()*, i8** }*
> }*, i8*, i8*, i8*, [16 x i8], i8*, i8*, i8*, i8* }*
@_rtti_t1d_ConsoleApplication479.Program
> to i8*), i32 16), !dbg !72
>   %7 = bitcast i8* %6 to %._ConsoleApplication479.Program*, !dbg !72
>   call void
@mi_t1d_ConsoleApplication479.Program5_.ctor(%._ConsoleApplication479.Program*
> %7) #0, !dbg !72
>   %8 = bitcast %._ConsoleApplication479.Program* %7 to %WriteLineObject*,
> !dbg !72
>   store %WriteLineObject* %8, %WriteLineObject** %4, !dbg !72
>   call void @WriteLine(%WriteLineString* bitcast ({ i8*, i32, [14 x i8] }*
> @.str2 to %WriteLineString*)) #1, !dbg !73
>   %9 = load double, double* %3, !dbg !74
>   %10 = sitofp i32 1 to double, !dbg !74
>   %11 = fadd double %9, %10, !dbg !74
>   store double %11, double* %3, !dbg !74
>   %12 = load double, double* %3, !dbg !75
>   %13 = sitofp i32 1 to double, !dbg !75
>   %14 = fadd double %12, %13, !dbg !75
>   store double %14, double* %3, !dbg !75
>   ret void, !dbg !76
> }
>
>
>
> !72 = !DILocation(line: 22, column: 11, scope: !65)
> !73 = !DILocation(line: 25, column: 7, scope: !65)
>
> full bitcode: https://gist.github.com/carlokok/
> aa3fab175f8afe8e4776ad6c2f5cc80a
>
> disasm: https://gist.github.com/carlokok/ab265461b37f87a10e0853ffc4d93c77
>
> What could possibly cause this? It's wreaking havoc with the debugger
> experience.
>
> Carlo Kok
> RemObjects Software
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20170918/a0fcd85b/attachment.html>

Carlo Kok via llvm-dev

2017-Sep-19 05:02 UTC

head link

[llvm-dev] Interleaved debug info on arm

September 19, 2017 3:02 AM, "Rui Ueyama"  wrote:
 Does this happen only when you are using lld? Please try gold to see if it
works as you expected.
Yes this only happens when using lld (I'm on aug1's head)

If I compile with lld with bitcode as input:
-O0 --lto-O0
--eh-frame-hdr --dynamic-linker "/lib/ld-linux-armhf.so.3"
0x0000000000059578 15 0 1 0 0 is_stmt
0x0000000000059588 22 11 1 0 0 is_stmt prologue_end
0x000000000005958c 21 11 1 0 0 is_stmt
0x00000000000595a0 15 5 1 0 0 is_stmt
0x00000000000595a4 22 11 1 0 0 is_stmt
0x00000000000595b4 25 7 1 0 0 is_stmt
0x00000000000595b8 22 11 1 0 0 is_stmt
0x00000000000595bc 25 7 1 0 0 is_stmt
0x00000000000595c4 26 7 1 0 0 is_stmt
0x00000000000595dc 27 7 1 0 0 is_stmt
0x00000000000595f0 30 5 1 0 0 is_stmt
If I precompile the bitcode with llc -O0 -filetype=obj
-O0 --lto-O0
--eh-frame-hdr --dynamic-linker "/lib/ld-linux-armhf.so.3"
0x00000000000401ac 15 0 1 0 0 is_stmt
0x00000000000401c4 15 5 1 0 0 is_stmt prologue_end
0x00000000000401d0 21 11 1 0 0 is_stmt
0x00000000000401e0 22 11 1 0 0 is_stmt
0x00000000000401fc 25 7 1 0 0 is_stmt
0x0000000000040204 26 7 1 0 0 is_stmt
0x0000000000040234 27 7 1 0 0 is_stmt
0x0000000000040250 30 5 1 0 0 is_stmt
Carlo Kok
RemObjects Software
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20170919/6a91125d/attachment.html>

Carlo Kok via llvm-dev

2017-Sep-19 05:13 UTC

head link

[llvm-dev] Interleaved debug info on arm

September 18, 2017 8:17 PM, "Adrian Prantl" <aprantl at
apple.com> wrote:
> Have you tried looking at the --dump-after-all output to see which pass (if
any) is adding the
> extra line table entries?
> -- adrian

You mean  --print-after-all ? I don't see any way to pass that to lld; The
debug info is fine when using llc.

Carlo Kok
RemObjects Software

Adrian Prantl via llvm-dev

2017-Sep-19 15:35 UTC

head link

[llvm-dev] Interleaved debug info on arm

> On Sep 18, 2017, at 10:13 PM, Carlo Kok <ck at remobjects.com> wrote:
> 
> September 18, 2017 8:17 PM, "Adrian Prantl" <aprantl at
apple.com> wrote:
> 
>> Have you tried looking at the --dump-after-all output to see which pass
(if any) is adding the
>> extra line table entries?
>> -- adrian
> 
> 
> You mean  --print-after-all ? I don't see any way to pass that to lld;
The debug info is fine when using llc.
I have no experience with lld, but worst-case you should be able to hard-code
the default to true and recompile liblto/lld.

-- adrian> 
> Carlo Kok
> RemObjects Software

Maybe Matching Threads

Search for more maybe matching threads

llvm dev - Sep 2017 - Interleaved debug info on arm

[llvm-dev] Interleaved debug info on arm

[llvm-dev] Interleaved debug info on arm

[llvm-dev] Interleaved debug info on arm

[llvm-dev] Interleaved debug info on arm

[llvm-dev] Interleaved debug info on arm

[llvm-dev] Interleaved debug info on arm

Maybe Matching Threads