[llvm] r204795 - Follow-up to r204790: don't try to emit line tables if there are no functions with DI in the TU

Tue Apr 1 03:08:18 PDT 2014

Let's continue the discussion on the r204790 commit thread.

2014-04-01 13:55 GMT+04:00 Timur Iskhodzhanov <timurrrr at google.com>:
> 2014-04-01 4:07 GMT+04:00 Eric Christopher <echristo at gmail.com>:
>> The MI has a DebugLoc and none of the IR do?
>
> Frankly, I'm confused by all this stuff myself :)
>
> Looking at
>   test/DebugInfo/COFF/tail-call-without-lexical-scopes.ll
> I think the case is that IR *has* !dbg tags, yet
> LexicalScopes::empty() actually returned "true" for this MF.
>
>> Otherwise a function is considered a lexical scope for purposes of the
>> other check.
>>
>> -eric
>>
>> On Wed, Mar 26, 2014 at 1:47 PM, Timur Iskhodzhanov <timurrrr at google.com> wrote:
>>> the WinCodeViewLineTables::beginInstruction is called and the MI happens to
>>> have a DebugLoc.
>>>
>>> That's in fact exactly the scenario r204790 has fixed - processing of a
>>> function without a lexical scope but with at least one MI having DL.
>>>
>>> 2014-03-27 0:36 GMT+04:00 Eric Christopher <echristo at gmail.com>:
>>>
>>>> But how do you know where/etc they are?
>>>>
>>>> -eric
>>>>
>>>> On Wed, Mar 26, 2014 at 1:33 PM, Timur Iskhodzhanov <timurrrr at google.com>
>>>> wrote:
>>>> > 1 file/line number entry for a single jmp instruction.
>>>> >
>>>> >
>>>> > 2014-03-27 0:22 GMT+04:00 Eric Christopher <echristo at gmail.com>:
>>>> >
>>>> >> On Wed, Mar 26, 2014 at 1:20 PM, Timur Iskhodzhanov
>>>> >> <timurrrr at google.com>
>>>> >> wrote:
>>>> >> >
>>>> >> >
>>>> >> >
>>>> >> > 2014-03-27 0:14 GMT+04:00 Eric Christopher <echristo at gmail.com>:
>>>> >> >
>>>> >> >> On Wed, Mar 26, 2014 at 1:00 PM, Timur Iskhodzhanov
>>>> >> >> <timurrrr at google.com>
>>>> >> >> wrote:
>>>> >> >> >
>>>> >> >> >
>>>> >> >> >
>>>> >> >> > 2014-03-26 19:24 GMT+04:00 David Blaikie <dblaikie at gmail.com>:
>>>> >> >> >
>>>> >> >> >> On Wed, Mar 26, 2014 at 4:24 AM, Timur Iskhodzhanov
>>>> >> >> >> <timurrrr at google.com>
>>>> >> >> >> wrote:
>>>> >> >> >> > Author: timurrrr
>>>> >> >> >> > Date: Wed Mar 26 06:24:36 2014
>>>> >> >> >> > New Revision: 204795
>>>> >> >> >> >
>>>> >> >> >> > URL: http://llvm.org/viewvc/llvm-project?rev=204795&view=rev
>>>> >> >> >> > Log:
>>>> >> >> >> > Follow-up to r204790: don't try to emit line tables if there
>>>> >> >> >> > are
>>>> >> >> >> > no
>>>> >> >> >> > functions with DI in the TU
>>>> >> >> >>
>>>> >> >> >> Hmm - is there any better way to do this? Should we only be
>>>> >> >> >> building
>>>> >> >> >> line tables for those functions with DI? Then we'd just emit line
>>>> >> >> >> tables if we have any (non-empty) line tables to emit, right?
>>>> >> >> >
>>>> >> >> >
>>>> >> >> > Is there a handy function available for that? :)
>>>> >> >> >
>>>> >> >>
>>>> >> >> From DwarfDebug::beginFunction:
>>>> >> >>
>>>> >> >>   // If there's no debug info for the function we're not going to do
>>>> >> >> anything.
>>>> >> >>   if (!MMI->hasDebugInfo())
>>>> >> >>     return;
>>>> >> >>
>>>> >> >>   // Grab the lexical scopes for the function, if we don't have any
>>>> >> >> of
>>>> >> >> those
>>>> >> >>   // then we're not going to be able to do anything.
>>>> >> >>   LScopes.initialize(*MF);
>>>> >> >>   if (LScopes.empty())
>>>> >> >>     return;
>>>> >> >>
>>>> >> >> -eric
>>>> >> >
>>>> >> >
>>>> >> > This is almost exactly the beginning of
>>>> >> > WinCodeViewLineTables::beginFunction() as of yesterday :)
>>>> >> > So yes, I'm aware of this and I do/did use it.
>>>> >> >
>>>> >> > However, IIUC it turns out that "if we don't have any [lexical
>>>> >> > scopes]
>>>> >> > then
>>>> >> > we're not going to be able to do anything" we still can emit minimal
>>>> >> > DI
>>>> >> > for
>>>> >> > a function. [see r204790 + r204791]
>>>> >>
>>>> >> Without looking (my backlog is insane) what's the info you're emitting
>>>> >> for
>>>> >> it?
>>>> >>
>>>> >> -eric
>>>> >>
>>>> >> >
>>>> >> >> >> I assume these COFF line tables don't actually get referenced
>>>> >> >> >> from
>>>> >> >> >> any
>>>> >> >> >> other debug info (since the rest of our info is DWARF), but I've
>>>> >> >> >> been
>>>> >> >> >> thinking of cleaning up our other line table handling code a bit
>>>> >> >> >> so
>>>> >> >> >> that it's a bit more structured - letting the line table object
>>>> >> >> >> itself
>>>> >> >> >> decide whether it should be emitted (by virtue of being
>>>> >> >> >> non-empty)
>>>> >> >> >> and
>>>> >> >> >> then having line table clients (CUs in DwarfDebug - no analogy
>>>> >> >> >> for
>>>> >> >> >> COFF which has no line table client) query the line table to see
>>>> >> >> >> if
>>>> >> >> >> it'll be there and attach the DW_AT_stmt_list if there will be a
>>>> >> >> >> line
>>>> >> >> >> table.
>>>> >> >> >>
>>>> >> >> >> - David
>>>> >> >> >>
>>>> >> >> >> >
>>>> >> >> >> > Added:
>>>> >> >> >> >     llvm/trunk/test/DebugInfo/COFF/asan2.ll
>>>> >> >> >> > Modified:
>>>> >> >> >> >     llvm/trunk/lib/CodeGen/AsmPrinter/WinCodeViewLineTables.cpp
>>>> >> >> >> >
>>>> >> >> >> > Modified:
>>>> >> >> >> > llvm/trunk/lib/CodeGen/AsmPrinter/WinCodeViewLineTables.cpp
>>>> >> >> >> > URL:
>>>> >> >> >> >
>>>> >> >> >> >
>>>> >> >> >> >
>>>> >> >> >> > http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/AsmPrinter/WinCodeViewLineTables.cpp?rev=204795&r1=204794&r2=204795&view=diff
>>>> >> >> >> >
>>>> >> >> >> >
>>>> >> >> >> >
>>>> >> >> >> >
>>>> >> >> >> > ==============================================================================
>>>> >> >> >> > --- llvm/trunk/lib/CodeGen/AsmPrinter/WinCodeViewLineTables.cpp
>>>> >> >> >> > (original)
>>>> >> >> >> > +++ llvm/trunk/lib/CodeGen/AsmPrinter/WinCodeViewLineTables.cpp
>>>> >> >> >> > Wed
>>>> >> >> >> > Mar
>>>> >> >> >> > 26 06:24:36 2014
>>>> >> >> >> > @@ -304,11 +304,18 @@ void WinCodeViewLineTables::beginFunctio
>>>> >> >> >> >    }
>>>> >> >> >> >  }
>>>> >> >> >> >
>>>> >> >> >> > -void WinCodeViewLineTables::endFunction(const MachineFunction
>>>> >> >> >> > *)
>>>> >> >> >> > {
>>>> >> >> >> > +void WinCodeViewLineTables::endFunction(const MachineFunction
>>>> >> >> >> > *MF) {
>>>> >> >> >> >    if (!Asm || !CurFn)  // We haven't created any debug info
>>>> >> >> >> > for
>>>> >> >> >> > this
>>>> >> >> >> > function.
>>>> >> >> >> >      return;
>>>> >> >> >> >
>>>> >> >> >> > -  if (!CurFn->Instrs.empty()) {
>>>> >> >> >> > +  const Function *GV = MF->getFunction();
>>>> >> >> >> > +  assert(FnDebugInfo.count(GV) == true);
>>>> >> >> >>
>>>> >> >> >> "count == true" seems a bit confusing for a few reasons. Should
>>>> >> >> >> that
>>>> >> >> >> be either just "assert(x.count(y))" or "assert(x.count(y) != 0)"
>>>> >> >> >> (or
>>>> >> >> >> greater than zero or 1 or whatever). "== true" always reads as
>>>> >> >> >> weird/redundant to me.
>>>> >> >> >>
>>>> >> >> >> > +  assert(CurFn == &FnDebugInfo[GV]);
>>>> >> >> >> > +
>>>> >> >> >> > +  if (CurFn->Instrs.empty()) {
>>>> >> >> >> > +    FnDebugInfo.erase(GV);
>>>> >> >> >> > +    VisitedFunctions.pop_back();
>>>> >> >> >> > +  } else {
>>>> >> >> >> >      // Define end label for subprogram.
>>>> >> >> >> >      MCSymbol *FunctionEndSym =
>>>> >> >> >> > Asm->OutStreamer.getContext().CreateTempSymbol();
>>>> >> >> >> >      Asm->OutStreamer.EmitLabel(FunctionEndSym);
>>>> >> >> >> >
>>>> >> >> >> > Added: llvm/trunk/test/DebugInfo/COFF/asan2.ll
>>>> >> >> >> > URL:
>>>> >> >> >> >
>>>> >> >> >> >
>>>> >> >> >> >
>>>> >> >> >> > http://llvm.org/viewvc/llvm-project/llvm/trunk/test/DebugInfo/COFF/asan2.ll?rev=204795&view=auto
>>>> >> >> >> >
>>>> >> >> >> >
>>>> >> >> >> >
>>>> >> >> >> >
>>>> >> >> >> > ==============================================================================
>>>> >> >> >> > --- llvm/trunk/test/DebugInfo/COFF/asan2.ll (added)
>>>> >> >> >> > +++ llvm/trunk/test/DebugInfo/COFF/asan2.ll Wed Mar 26 06:24:36
>>>> >> >> >> > 2014
>>>> >> >> >> > @@ -0,0 +1,52 @@
>>>> >> >> >> > +; RUN: llc -mcpu=core2 -mtriple=i686-pc-win32 -O0 < %s |
>>>> >> >> >> > FileCheck
>>>> >> >> >> > --check-prefix=X86 %s
>>>> >> >> >> > +
>>>> >> >> >> > +; This LL file was generated by running clang on the following
>>>> >> >> >> > code
>>>> >> >> >> > with
>>>> >> >> >> > +; -fsanitize=address
>>>> >> >> >> > +; D:\asan.c:
>>>> >> >> >> > +; 1 unsigned char c = 42;
>>>> >> >> >> > +
>>>> >> >> >> > +; Just make sure we don't try to emit the line table.
>>>> >> >> >> > +; X86-NOT: .section        .debug$S,"rn"
>>>> >> >> >> > +
>>>> >> >> >> > +; ModuleID = 'asan.c'
>>>> >> >> >> > +target datalayout = "e-m:w-p:32:32-i64:64-f80:32-n8:16:32-S32"
>>>> >> >> >> > +target triple = "i686-pc-win32"
>>>> >> >> >> > +
>>>> >> >> >> > + at c = global { i8, [63 x i8] } { i8 42, [63 x i8]
>>>> >> >> >> > zeroinitializer
>>>> >> >> >> > },
>>>> >> >> >> > align 32
>>>> >> >> >> > + at llvm.global_ctors = appending global [1 x { i32, void ()* }]
>>>> >> >> >> > [{
>>>> >> >> >> > i32,
>>>> >> >> >> > void ()* } { i32 1, void ()* @asan.module_ctor }]
>>>> >> >> >> > + at __asan_gen_ = private constant [7 x i8] c"asan.c\00", align 1
>>>> >> >> >> > + at __asan_gen_1 = private unnamed_addr constant [2 x i8]
>>>> >> >> >> > c"c\00",
>>>> >> >> >> > align 1
>>>> >> >> >> > + at 0 = internal global [1 x { i32, i32, i32, i32, i32, i32 }] [{
>>>> >> >> >> > i32,
>>>> >> >> >> > i32, i32, i32, i32, i32 } { i32 ptrtoint ({ i8, [63 x i8] }* @c
>>>> >> >> >> > to
>>>> >> >> >> > i32), i32
>>>> >> >> >> > 1, i32 64, i32 ptrtoint ([2 x i8]* @__asan_gen_1 to i32), i32
>>>> >> >> >> > ptrtoint ([7 x
>>>> >> >> >> > i8]* @__asan_gen_ to i32), i32 0 }]
>>>> >> >> >> > + at llvm.global_dtors = appending global [1 x { i32, void ()* }]
>>>> >> >> >> > [{
>>>> >> >> >> > i32,
>>>> >> >> >> > void ()* } { i32 1, void ()* @asan.module_dtor }]
>>>> >> >> >> > +
>>>> >> >> >> > +define internal void @asan.module_ctor() {
>>>> >> >> >> > +  call void @__asan_init_v3()
>>>> >> >> >> > +  call void @__asan_register_globals(i32 ptrtoint ([1 x { i32,
>>>> >> >> >> > i32,
>>>> >> >> >> > i32, i32, i32, i32 }]* @0 to i32), i32 1)
>>>> >> >> >> > +  ret void
>>>> >> >> >> > +}
>>>> >> >> >> > +
>>>> >> >> >> > +declare void @__asan_init_v3()
>>>> >> >> >> > +
>>>> >> >> >> > +declare void @__asan_before_dynamic_init(i32)
>>>> >> >> >> > +
>>>> >> >> >> > +declare void @__asan_after_dynamic_init()
>>>> >> >> >> > +
>>>> >> >> >> > +declare void @__asan_register_globals(i32, i32)
>>>> >> >> >> > +
>>>> >> >> >> > +declare void @__asan_unregister_globals(i32, i32)
>>>> >> >> >> > +
>>>> >> >> >> > +define internal void @asan.module_dtor() {
>>>> >> >> >> > +  call void @__asan_unregister_globals(i32 ptrtoint ([1 x {
>>>> >> >> >> > i32,
>>>> >> >> >> > i32,
>>>> >> >> >> > i32, i32, i32, i32 }]* @0 to i32), i32 1)
>>>> >> >> >> > +  ret void
>>>> >> >> >> > +}
>>>> >> >> >> > +
>>>> >> >> >> > +!llvm.dbg.cu = !{!0}
>>>> >> >> >> > +!llvm.module.flags = !{!3, !4}
>>>> >> >> >> > +!llvm.ident = !{!5}
>>>> >> >> >> > +
>>>> >> >> >> > +!0 = metadata !{i32 786449, metadata !1, i32 12, metadata
>>>> >> >> >> > !"clang
>>>> >> >> >> > version 3.5.0 ", i1 false, metadata !"", i32 0, metadata !2,
>>>> >> >> >> > metadata
>>>> >> >> >> > !2,
>>>> >> >> >> > metadata !2, metadata !2, metadata !2, metadata !"", i32 2} ; [
>>>> >> >> >> > DW_TAG_compile_unit ] [D:\/asan.c] [DW_LANG_C99]
>>>> >> >> >> > +!1 = metadata !{metadata !"asan.c", metadata !"D:\5C"}
>>>> >> >> >> > +!2 = metadata !{}
>>>> >> >> >> > +!3 = metadata !{i32 2, metadata !"Dwarf Version", i32 4}
>>>> >> >> >> > +!4 = metadata !{i32 1, metadata !"Debug Info Version", i32 1}
>>>> >> >> >> > +!5 = metadata !{metadata !"clang version 3.5.0 "}
>>>> >> >> >> >
>>>> >> >> >> >
>>>> >> >> >> > _______________________________________________
>>>> >> >> >> > llvm-commits mailing list
>>>> >> >> >> > llvm-commits at cs.uiuc.edu
>>>> >> >> >> > http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>>> >> >> >
>>>> >> >> >
>>>> >> >> >
>>>> >> >> > _______________________________________________
>>>> >> >> > llvm-commits mailing list
>>>> >> >> > llvm-commits at cs.uiuc.edu
>>>> >> >> > http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>>> >> >> >
>>>> >> >
>>>> >> >
>>>> >
>>>> >
>>>
>>>