[llvm] r200422 - Turn on CU ranges if we've got multiple compile units in the same

[Eric Christopher]

On Wed, Jan 29, 2014 at 2:53 PM, David Blaikie <dblaikie at gmail.com> wrote:
>
>
>
> On Wed, Jan 29, 2014 at 2:41 PM, Eric Christopher <echristo at gmail.com>
> wrote:
>>
>> On Wed, Jan 29, 2014 at 2:37 PM, Eric Christopher <echristo at gmail.com>
>> wrote:
>> > On Wed, Jan 29, 2014 at 2:32 PM, David Blaikie <dblaikie at gmail.com>
>> > wrote:
>> >>
>> >>
>> >>
>> >> On Wed, Jan 29, 2014 at 2:26 PM, Eric Christopher <echristo at gmail.com>
>> >> wrote:
>> >>>
>> >>> On Wed, Jan 29, 2014 at 2:23 PM, David Blaikie <dblaikie at gmail.com>
>> >>> wrote:
>> >>> >
>> >>> >
>> >>> >
>> >>> > On Wed, Jan 29, 2014 at 2:06 PM, Eric Christopher
>> >>> > <echristo at gmail.com>
>> >>> > wrote:
>> >>> >>
>> >>> >> Author: echristo
>> >>> >> Date: Wed Jan 29 16:06:27 2014
>> >>> >> New Revision: 200422
>> >>> >>
>> >>> >> URL: http://llvm.org/viewvc/llvm-project?rev=200422&view=rev
>> >>> >> Log:
>> >>> >> Turn on CU ranges if we've got multiple compile units in the same
>> >>> >> module since there's no range guarantee that we could make given
>> >>> >> output order. This also fixes up the testcases that have multiple
>> >>> >> CUs to have the correct range offset.
>> >>> >
>> >>> >
>> >>> > What if LTO together a debug and non-debug module together? We'll
>> >>> > still
>> >>> > have
>> >>> > one CU, but we'll have functions that aren't in that CU - and
>> >>> > incorrectly
>> >>> > describe that using an "all text" range?
>> >>> >
>> >>>
>> >>> ... Short of saying "always use DW_AT_ranges" I don't have a good
>> >>> solution for this. I'm somewhat down with that, but...
>> >>
>> >>
>> >> Perhaps you were addressing this with the "there's no range guarantee
>> >> that
>> >> we could make given output order" (I'm not /quite/ sure how to parse
>> >> that
>> >> sentence) - but I thought we'd settled on the idea that we could rely
>> >> on
>> >> output order and just coalesce functions/ranges that came one after the
>> >> other. Or did you come up with some cases that breaks down for/that we
>> >> care
>> >> about?
>> >>
>> >
>> > If you have functions that aren't associated with a particular CU
>> > there's no way of knowing which functions those are. We can only rely
>> > on the output order to determine whether or not a function is not
>> > going to appear in debug info output. At least not at the moment. I
>> > imagine if we were to do so we'd run into the assert at
>> > DwarfDebug.cpp:1583 right now if we even tried linking the two modules
>> > together. :\
>>
>> FWIW we don't seem to. Wacky.
>
>
> Guess we must be exiting around DwarfDebug.cpp:15671, but not sure.

Ah yes. Likely.

>
>>
>> Anyhow, continuing down the unified range path.
>
>
> OK - sorry, wasn't clear whether you were pursuing that or whether this
> patch was an alternative. (given that I don't think this patch should be
> needed if the unified range stuff works, I assumed this was an alternative)
>

Though I think the above case means that if I do run into a function
without debug info I'll have to end the coalesced range there as well.
Another corner case.

Anyhow, this patch is an alternative for now in the "obviously correct
and the whole thing can be removed when we get the better
implementation".

-eric

> - Dave
>
>>
>>
>> -eric
>>
>> >
>> >> That way we'd still get high/low whenever it's contiguous, and
>> >> otherwise
>> >> we'd use ranges.
>> >>
>> >
>> > Still don't have this way of doing it in yet, but that's an idea yes.
>> > I'll see how it works in practice.
>> >
>> > -eric
>> >
>> >>>
>> >>>
>> >>> -eric
>> >>>
>> >>> >>
>> >>> >>
>> >>> >> Modified:
>> >>> >>     llvm/trunk/lib/CodeGen/AsmPrinter/DwarfDebug.cpp
>> >>> >>
>> >>> >> llvm/trunk/test/DebugInfo/X86/stmt-list-multiple-compile-units.ll
>> >>> >>
>> >>> >> Modified: llvm/trunk/lib/CodeGen/AsmPrinter/DwarfDebug.cpp
>> >>> >> URL:
>> >>> >>
>> >>> >>
>> >>> >> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/AsmPrinter/DwarfDebug.cpp?rev=200422&r1=200421&r2=200422&view=diff
>> >>> >>
>> >>> >>
>> >>> >>
>> >>> >> ==============================================================================
>> >>> >> --- llvm/trunk/lib/CodeGen/AsmPrinter/DwarfDebug.cpp (original)
>> >>> >> +++ llvm/trunk/lib/CodeGen/AsmPrinter/DwarfDebug.cpp Wed Jan 29
>> >>> >> 16:06:27
>> >>> >> 2014
>> >>> >> @@ -1119,10 +1119,12 @@ void DwarfDebug::endSections() {
>> >>> >>    }
>> >>> >>
>> >>> >>    // For now only turn on CU ranges if we've explicitly asked for
>> >>> >> it,
>> >>> >> -  // we have -ffunction-sections enabled, or we've emitted a
>> >>> >> function
>> >>> >> -  // into a unique section. At this point all sections should be
>> >>> >> finalized
>> >>> >> -  // except for dwarf sections.
>> >>> >> -  HasCURanges = DwarfCURanges || UsedNonDefaultText ||
>> >>> >> +  // we have -ffunction-sections enabled, we've emitted a function
>> >>> >> +  // into a unique section, or we're using LTO. If we're using LTO
>> >>> >> then
>> >>> >> +  // we can't know that any particular function in the module is
>> >>> >> correlated
>> >>> >> +  // to a particular CU and so we need to be conservative. At this
>> >>> >> point
>> >>> >> all
>> >>> >> +  // sections should be finalized except for dwarf sections.
>> >>> >> +  HasCURanges = DwarfCURanges || UsedNonDefaultText ||
>> >>> >> (CUMap.size() >
>> >>> >> 1)
>> >>> >> ||
>> >>> >>                  TargetMachine::getFunctionSections();
>> >>> >>  }
>> >>> >>
>> >>> >>
>> >>> >> Modified:
>> >>> >> llvm/trunk/test/DebugInfo/X86/stmt-list-multiple-compile-units.ll
>> >>> >> URL:
>> >>> >>
>> >>> >>
>> >>> >> http://llvm.org/viewvc/llvm-project/llvm/trunk/test/DebugInfo/X86/stmt-list-multiple-compile-units.ll?rev=200422&r1=200421&r2=200422&view=diff
>> >>> >>
>> >>> >>
>> >>> >>
>> >>> >> ==============================================================================
>> >>> >> ---
>> >>> >> llvm/trunk/test/DebugInfo/X86/stmt-list-multiple-compile-units.ll
>> >>> >> (original)
>> >>> >> +++
>> >>> >> llvm/trunk/test/DebugInfo/X86/stmt-list-multiple-compile-units.ll
>> >>> >> Wed
>> >>> >> Jan 29 16:06:27 2014
>> >>> >> @@ -8,11 +8,11 @@
>> >>> >>  ; CHECK: .debug_info contents:
>> >>> >>  ; CHECK: DW_TAG_compile_unit
>> >>> >>  ; CHECK: DW_AT_stmt_list [DW_FORM_sec_offset]   (0x00000000)
>> >>> >> -; CHECK: DW_AT_low_pc [DW_FORM_addr]       (0x0000000000000000)
>> >>> >> +; CHECK: DW_AT_ranges [DW_FORM_sec_offset]      (0x00000000)
>> >>> >>
>> >>> >>  ; CHECK: DW_TAG_compile_unit
>> >>> >>  ; CHECK: DW_AT_stmt_list [DW_FORM_sec_offset]   (0x0000003c)
>> >>> >> -; CHECK: DW_AT_low_pc [DW_FORM_addr]       (0x0000000000000000)
>> >>> >> +; CHECK: DW_AT_ranges [DW_FORM_sec_offset]      (0x00000020)
>> >>> >>
>> >>> >>  ; CHECK: .debug_line contents:
>> >>> >>  ; CHECK-NEXT: Line table prologue:
>> >>> >> @@ -25,12 +25,12 @@
>> >>> >>
>> >>> >>  ; DWARF3: .debug_info contents:
>> >>> >>  ; DWARF3: DW_TAG_compile_unit
>> >>> >> -; DWARF3: DW_AT_stmt_list [DW_FORM_data4]   (0x00000000)
>> >>> >> -; DWARF3: DW_AT_low_pc [DW_FORM_addr]       (0x0000000000000000)
>> >>> >> +; DWARF3: DW_AT_stmt_list [DW_FORM_data4]    (0x00000000)
>> >>> >> +; DWARF3: DW_AT_ranges [DW_FORM_data4]       (0x00000000)
>> >>> >>
>> >>> >>  ; DWARF3: DW_TAG_compile_unit
>> >>> >>  ; DWARF3: DW_AT_stmt_list [DW_FORM_data4]   (0x0000003c)
>> >>> >> -; DWARF3: DW_AT_low_pc [DW_FORM_addr]       (0x0000000000000000)
>> >>> >> +; DWARF3: DW_AT_ranges [DW_FORM_data4]      (0x00000020)
>> >>> >>
>> >>> >>  ; DWARF3: .debug_line contents:
>> >>> >>  ; DWARF3-NEXT: Line table prologue:
>> >>> >>
>> >>> >>
>> >>> >> _______________________________________________
>> >>> >> llvm-commits mailing list
>> >>> >> llvm-commits at cs.uiuc.edu
>> >>> >> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>> >>> >
>> >>> >
>> >>
>> >>
>
>