[llvm-dev] addition of vendor dwarf operator extension.

Wed Dec 4 05:29:53 PST 2019

I understand the specific savings, but do you have a sense of how much this
occurs overall for DWARF of some non-trivial program (like the clang
binary)?

I know it's verbose - but it's not especially expensive to maintain the
emission of all those bytes (they'd be in a constant array you'd copy into
the location expression, I assume - written once) so if there's not much
overall size savings and not much parsing savings I'd be inclined to wait
until such a thing were standardized rather than introducing an extension.

That's just my tendency on that - quite possible other folks (CC'd) feel
differently.

On Wed, Dec 4, 2019 at 12:32 AM Chirag Patel <Chirag at raincode.com> wrote:

> Below is a snippet of the dwarf info that we generate,
>
>
>
> 0x00000057:   DW_TAG_variable
>
>                 DW_AT_name      ("__gbloffset__")
>
>                 DW_AT_type      (0x000001e5 "int")
>
>                 DW_AT_external  (true)
>
>                 DW_AT_decl_file ("…")
>
>                 DW_AT_decl_line (8)
>
>                 DW_AT_location  (DW_OP_addr 0) // pre linkage
>
>                 DW_AT_linkage_name      ("_gblsection__")
>
>                 DW_AT_endianity (DW_END_big)
>
>
>
> 0x00000170:   DW_TAG_variable
>
>                 DW_AT_name      ("VAR1")
>
>                 DW_AT_type      (0x0000010b "fixed.dec.display.72")
>
>                 DW_AT_decl_file ("…")
>
>                 DW_AT_decl_line (10)
>
>                 DW_AT_location  (DW_OP_addr 0x0, DW_OP_call4 0x57,
> DW_OP_deref_size, 4,
>
> DW_OP_dup, DW_OP_constu 0xff, DW_OP_lit0, DW_OP_shl, DW_OP_and,
> DW_OP_lit24, DW_OP_shl, DW_OP_swap, DW_OP_dup, DW_OP_constu 0xff,
> DW_OP_lit8, DW_OP_shl, DW_OP_and, DW_OP_lit8, DW_OP_shl, DW_OP_swap,
> DW_OP_dup, DW_OP_constu 0xff, DW_OP_lit16, DW_OP_shl, DW_OP_and,
> DW_OP_lit8, DW_OP_shr, DW_OP_swap, DW_OP_constu 0xff, DW_OP_lit24,
> DW_OP_shl, DW_OP_and, DW_OP_lit24, DW_OP_shr, DW_OP_swap, DW_OP_or,
> DW_OP_or, DW_OP_or, DW_OP_plus)
>
>                 DW_AT_linkage_name      ("VAR1")
>
>
>
> In above code for location calculation of “VAR1” we change the endianity
> of “__gbloffset__” var, it is 32 bit value on 64-bit machine.
>
> In above case DW_OP_byte_swap will replace near about 38 operations(for 32 bit value), that’s 38-bytes per usage savings on mem footprint and at runtime as well. I do not have any profiling data.
>
>
>
> The use case for DW_OP_byte_swap is, for typed entry we know the size, for raw data, dwarf operator will consider the generic type and pop the top of the stack, byte swap it and push the data back on the stack.
>
>
>
> Regards,
>
>
>
> *Chirag Patel*
>
> Software Engineer | Raincode Labs India
>
> *Tel*: (+91) 080 41159811
>
> *Mob*: (+91) 9049336744
>
> *www.raincodelabs.com <http://www.raincodelabs.com/>*
>
> [image: linkedin-button] <https://in.linkedin.com/in/chirag-patel->
>
>
>
> *From:* David Blaikie <dblaikie at gmail.com>
> *Sent:* 03 December 2019 21:31
> *To:* Chirag Patel <Chirag at raincode.com>
> *Cc:* Adrian Prantl <aprantl at apple.com>; Paul Robinson <
> paul.robinson at sony.com>; Jonas Devlieghere <jdevlieghere at apple.com>;
> llvm-dev at lists.llvm.org
> *Subject:* Re: [llvm-dev] addition of vendor dwarf operator extension.
>
>
>
> Does this have a significant impact on total DWARF size or performance
> that you've observed/have data about?
>
>
>
> On Tue, Dec 3, 2019 at 12:06 AM Chirag Patel <Chirag at raincode.com> wrote:
>
> Hello,
>
>
>
> Thanks for prompt reply.
>
>
>
> Motivation,
>
> In case of debugging of bi-endian dwarf and value known at runtime some of
> the raw values needs to be byte swapped to change the endianity. We have a
> llvm based compiler which generates such dwarf, TAG_variable marked as big
> endian and debugging on little endian host using lldb. The listed operator
> on dwarf issues does that, one operator instead of list of operators.
>
>
>
> Proposal,
>
> Addition of listed dwarf operator as vender extension in llvm/lldb till it
> gets released.
>
>
>
> Regards,
>
>
>
> *Chirag Patel*
>
> Software Engineer | Raincode Labs India
>
> *Tel*: (+91) 080 41159811
>
> *Mob*: (+91) 9049336744
>
> *www.raincodelabs.com <http://www.raincodelabs.com/>*
>
> [image: linkedin-button] <https://in.linkedin.com/in/chirag-patel->
>
>
>
> *From:* David Blaikie <dblaikie at gmail.com>
> *Sent:* 02 December 2019 19:51
> *To:* Chirag Patel <Chirag at raincode.com>; Adrian Prantl <aprantl at apple.com>;
> Paul Robinson <paul.robinson at sony.com>; Jonas Devlieghere <
> jdevlieghere at apple.com>
> *Cc:* llvm-dev at lists.llvm.org
> *Subject:* Re: [llvm-dev] addition of vendor dwarf operator extension.
>
>
>
> There are some vendor extensions already in LLVM/LLDB, though I guess the
> main question would be: what's the motivation?
>
>
>
> On Mon, Dec 2, 2019 at 3:14 AM Chirag Patel via llvm-dev <
> llvm-dev at lists.llvm.org> wrote:
>
> Hello all,
>
>
>
> There is one enhancement request open for dwarfv5,
> http://dwarfstd.org/ShowIssue.php?issue=191107.1
>
> The request is for addition of dwarf expression operator to swap the top
> of the dwarf stack, the response seems positive but it may take some time
> till v6.
>
> I like to add that operator as vendor extension but I am not sure how to
> go about it for llvm/lldb. Currently I am using it as vendor specific
> extension on our local fork.
>
>
>
> Does adding it as vender specific extension seems like a good idea? If yes
> please provide me the details about adding specific vendor.
>
>
>
> Regards,
>
>
>
> *Chirag Patel*
>
> Software Engineer | Raincode Labs India
>
> *Tel*: (+91) 080 41159811
>
> *Mob*: (+91) 9049336744
>
> *www.raincodelabs.com <http://www.raincodelabs.com/>*
>
> [image: linkedin-button] <https://in.linkedin.com/in/chirag-patel->
>
>
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> https://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/20191204/c62fa9c4/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image001.png
Type: image/png
Size: 7399 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20191204/c62fa9c4/attachment.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image002.png
Type: image/png
Size: 7407 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20191204/c62fa9c4/attachment-0001.png>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: image003.png
Type: image/png
Size: 7436 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20191204/c62fa9c4/attachment-0002.png>