llvm dev - Nov 2016 - BoF: Debug info for optimized code.

On 11/1/2016 4:28 PM, Martin J. O'Riordan via llvm-dev
wrote:
> I do not even pretend to know much about Dwarf and the representation of
debug information, but it does appear that there is little or no support for the
idea that a single "instruction" can correspond to multiple diverse
lines in the source file.
There is.  There is even a patch for LLVM:
https://reviews.llvm.org/D16697

-Krzysztof

-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, 
hosted by The Linux Foundation

Thanks Krzysztof, I hadn't noticed this.

The patch refers to the target providing an 'op_index' register, but
this seems like something that can only be handled by an integrated assembler. 
We use an external assembler and I am curious if there are new directives that
we need to support for this?  At the moment our assembler is unable to accept
'.loc' directives between each operation in a VLIW instruction, is this
something that we need to implement to get this level of VLIW debug support?

Thanks,

	MartinO

-----Original Message-----
From: llvm-dev [mailto:llvm-dev-bounces at lists.llvm.org] On Behalf Of
Krzysztof Parzyszek via llvm-dev
Sent: 01 November 2016 21:35
To: llvm-dev at lists.llvm.org
Subject: Re: [llvm-dev] BoF: Debug info for optimized code.

On 11/1/2016 4:28 PM, Martin J. O'Riordan via llvm-dev
wrote:
> I do not even pretend to know much about Dwarf and the representation of
debug information, but it does appear that there is little or no support for the
idea that a single "instruction" can correspond to multiple diverse
lines in the source file.
There is.  There is even a patch for LLVM:
https://reviews.llvm.org/D16697

-Krzysztof

--
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted by The
Linux Foundation _______________________________________________
LLVM Developers mailing list
llvm-dev at lists.llvm.org
http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev

That would certainly seem to be the conceptually simplest way to go: treat
each operation in the VLIW instruction the same as a scalar instruction,
and have ".loc" be able to address them.

You'd then need a way to indicate to the debugger whether it is known
correct to execute the operations sequentially instead of in parallel (I
believe Itanium specifies that they is always true), or if they *must* be
done in parallel to be correct e.g. "r1<-r2; r2<-r1" to swap
values ("true"
VLIW).

Even when the architecture says the result is the same as if all operations
were done in parallel from the same set of input registers, in practice
many or most instructions will work correctly if the operations are
executed sequentially, and it would be nice for debuggers to be able to do
this for single-stepping -- and to be able to execute the operations in
source program order, in the event they have been mixed up within the
instruction (perhaps because each operation position is limited to certain
operation types).

So it would be useful to have some kind of "operation bundle"
information,
where each bundle is some arbitrary subset of the operations in the
instruction, which *must* be executed in parallel.

Note: this is somehow the opposite to another notion of "bundle| in VLIW,
where it means "everything in a bundle MAY be executed in parallel, but
different bundles MUST be executed sequentially".


On Wed, Nov 2, 2016 at 12:03 PM, Martin J. O'Riordan via llvm-dev <
llvm-dev at lists.llvm.org> wrote:
> Thanks Krzysztof, I hadn't noticed this.
>
> The patch refers to the target providing an 'op_index' register,
but this
> seems like something that can only be handled by an integrated assembler.
> We use an external assembler and I am curious if there are new directives
> that we need to support for this?  At the moment our assembler is unable to
> accept '.loc' directives between each operation in a VLIW
instruction, is
> this something that we need to implement to get this level of VLIW debug
> support?
>
> Thanks,
>
>         MartinO
>
> -----Original Message-----
> From: llvm-dev [mailto:llvm-dev-bounces at lists.llvm.org] On Behalf Of
> Krzysztof Parzyszek via llvm-dev
> Sent: 01 November 2016 21:35
> To: llvm-dev at lists.llvm.org
> Subject: Re: [llvm-dev] BoF: Debug info for optimized code.
>
> On 11/1/2016 4:28 PM, Martin J. O'Riordan via llvm-dev wrote:
> > I do not even pretend to know much about Dwarf and the representation
of
> debug information, but it does appear that there is little or no support
> for the idea that a single "instruction" can correspond to
multiple diverse
> lines in the source file.
>
> There is.  There is even a patch for LLVM:
> https://reviews.llvm.org/D16697
>
> -Krzysztof
>
> --
> Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted
> by The Linux Foundation _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
> _______________________________________________
> 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/20161102/cee0f98a/attachment.html>

Hi Martin,
Yes, the patch only changes the format of line information.  There will 
be more work needed for fully implementing it across all tools.
Here your concern still stands---more focus on debug information for 
VLIW architectures would be welcome.  I was only pointing out that the 
necessary capacity of the debug information to carry this data does in 
fact exist, and that at least one step for getting it into LLVM has been 
attempted (the patch was reverted shortly after commit).

-Krzysztof


On 11/2/2016 4:03 AM, Martin J. O'Riordan via llvm-dev
wrote:
> Thanks Krzysztof, I hadn't noticed this.
>
> The patch refers to the target providing an 'op_index' register,
but this seems like something that can only be handled by an integrated
assembler.  We use an external assembler and I am curious if there are new
directives that we need to support for this?  At the moment our assembler is
unable to accept '.loc' directives between each operation in a VLIW
instruction, is this something that we need to implement to get this level of
VLIW debug support?
>
> Thanks,
>
> 	MartinO
>
> -----Original Message-----
> From: llvm-dev [mailto:llvm-dev-bounces at lists.llvm.org] On Behalf Of
Krzysztof Parzyszek via llvm-dev
> Sent: 01 November 2016 21:35
> To: llvm-dev at lists.llvm.org
> Subject: Re: [llvm-dev] BoF: Debug info for optimized code.
>
> On 11/1/2016 4:28 PM, Martin J. O'Riordan via llvm-dev wrote:
>> I do not even pretend to know much about Dwarf and the representation
of debug information, but it does appear that there is little or no support for
the idea that a single "instruction" can correspond to multiple
diverse lines in the source file.
>
> There is.  There is even a patch for LLVM:
> https://reviews.llvm.org/D16697
>
> -Krzysztof
>
> --
> Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, hosted
by The Linux Foundation _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev
>
-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, 
hosted by The Linux Foundation