liblo-devel Mailing List for LibLO

Interesting, I'll try the amd64 approach when I get the chance.

Steve


On Sat, Nov 22, 2014 at 3:44 PM, Felipe Sateler <fsa...@gm...> wrote:
> On Sat, Nov 22, 2014 at 10:01 AM, Stephen Sinclair <rad...@gm...> wrote:
>> Hi Felipe,
>>
>> It does sound like this is worth addressing.  I would love to know if
>> there is a reliable way to test the condition, using an emulator or
>> such.  I had no success using one last time you brought this up, but I
>> could give it another shot.
>
> Maybe we could try getting you access on the sparc
> porterbox.Unfortunately that box has a 2.6.32 kernel so I think some
> SO_* features are not available, but you should be able to debug this
> issue as well.
>
> I can advocate you for that, but we need to follow the instructions in
> https://dsa.debian.org/doc/guest-account/
>
> I have no idea how feasible it is to launch a vm with sparc these days.
>
> Hmm, I just found this:
> http://stackoverflow.com/questions/5498102/how-to-debug-unaligned-accesses-on-amd64-using-visual-studio
>
> This may enable debugging in an amd64 machine.
> Unfortunately, one cannot just add the flag at startup because it
> seems like libc has some unaligned accesses of its own :(. So it would
> need to be added and removed to the relevant pieces of code.
> --
>
> Saludos,
> Felipe Sateler
>
> ------------------------------------------------------------------------------
> Download BIRT iHub F-Type - The Free Enterprise-Grade BIRT Server
> from Actuate! Instantly Supercharge Your Business Reports and Dashboards
> with Interactivity, Sharing, Native Excel Exports, App Integration & more
> Get technology previously reserved for billion-dollar corporations, FREE
> http://pubads.g.doubleclick.net/gampad/clk?id=157005751&iu=/4140/ostg.clktrk
> _______________________________________________
> liblo-devel mailing list
> lib...@li...
> https://lists.sourceforge.net/lists/listinfo/liblo-devel

On Sat, Nov 22, 2014 at 10:01 AM, Stephen Sinclair <rad...@gm...> wrote:
> Hi Felipe,
>
> It does sound like this is worth addressing.  I would love to know if
> there is a reliable way to test the condition, using an emulator or
> such.  I had no success using one last time you brought this up, but I
> could give it another shot.

Maybe we could try getting you access on the sparc
porterbox.Unfortunately that box has a 2.6.32 kernel so I think some
SO_* features are not available, but you should be able to debug this
issue as well.

I can advocate you for that, but we need to follow the instructions in
https://dsa.debian.org/doc/guest-account/

I have no idea how feasible it is to launch a vm with sparc these days.

Hmm, I just found this:
http://stackoverflow.com/questions/5498102/how-to-debug-unaligned-accesses-on-amd64-using-visual-studio

This may enable debugging in an amd64 machine.
Unfortunately, one cannot just add the flag at startup because it
seems like libc has some unaligned accesses of its own :(. So it would
need to be added and removed to the relevant pieces of code.
-- 

Saludos,
Felipe Sateler

Hi Felipe,

It does sound like this is worth addressing.  I would love to know if
there is a reliable way to test the condition, using an emulator or
such.  I had no success using one last time you brought this up, but I
could give it another shot.

Steve


On Fri, Nov 21, 2014 at 9:30 PM, Felipe Sateler <fsa...@gm...> wrote:
> Hi,
>
> On Fri, Nov 29, 2013 at 1:11 PM, Stephen Sinclair <rad...@gm...> wrote:
>> On Fri, Nov 29, 2013 at 3:44 PM, Felipe Sateler <fsa...@gm...> wrote:
>
>>> But we don't know which architectures have the alignment issues :(. It
>>> seems that sparc is the only one that actually dies on a 8-byte access
>>> at a 4-byte alignment. People keep telling me it is slower (even when
>>> allowed), but I haven't been able to find a reference for that.
>>
>> Detecting it at configure time using a crashing program is likely possible,
>> but for slowness it's obviously harder to detect.  I'd be fine with simply
>> providing it as a manual option in this case.
>
> During a recent discussion in debian-devel, it was asked on which
> architectures it is ok to use unaligned accesses[1].
> The short version from one ARM developer was:
>
> [2] Wookey:
>> High-level summary for software engineers: Don't do unaligned access
>> on armhf or armel. It's always inefficient, sometimes extremely
>> inefficent, and in various cases is not permitted at all.
>
> I thought I'd share that here to start the ball rolling on how to fix
> this for good.
>
> I remember discussing the following options:
>
> 1. The patch I sent using memcpy and a temporary variable to avoid
> nonaligned accesses.This approach should be correct, but I'm not sure
> if more places are necessary to fix.
> 2. Packing and unpacking into an 8-byte wide stream instead of a
> 4-byte wide stream. This should work as well but it seems wasteful in
> memory (not sure if that would ever be relevant).
>
> I think option 1 would be minimal effort to fix this.
>
> [1] https://lists.debian.org/debian-devel/2014/11/msg01011.html
> [2] https://lists.debian.org/debian-devel/2014/11/msg01028.html
>
>
> --
>
> Saludos,
> Felipe Sateler
>
> ------------------------------------------------------------------------------
> Download BIRT iHub F-Type - The Free Enterprise-Grade BIRT Server
> from Actuate! Instantly Supercharge Your Business Reports and Dashboards
> with Interactivity, Sharing, Native Excel Exports, App Integration & more
> Get technology previously reserved for billion-dollar corporations, FREE
> http://pubads.g.doubleclick.net/gampad/clk?id=157005751&iu=/4140/ostg.clktrk
> _______________________________________________
> liblo-devel mailing list
> lib...@li...
> https://lists.sourceforge.net/lists/listinfo/liblo-devel

Hi,

On Fri, Nov 29, 2013 at 1:11 PM, Stephen Sinclair <rad...@gm...> wrote:
> On Fri, Nov 29, 2013 at 3:44 PM, Felipe Sateler <fsa...@gm...> wrote:

>> But we don't know which architectures have the alignment issues :(. It
>> seems that sparc is the only one that actually dies on a 8-byte access
>> at a 4-byte alignment. People keep telling me it is slower (even when
>> allowed), but I haven't been able to find a reference for that.
>
> Detecting it at configure time using a crashing program is likely possible,
> but for slowness it's obviously harder to detect.  I'd be fine with simply
> providing it as a manual option in this case.

During a recent discussion in debian-devel, it was asked on which
architectures it is ok to use unaligned accesses[1].
The short version from one ARM developer was:

[2] Wookey:
> High-level summary for software engineers: Don't do unaligned access
> on armhf or armel. It's always inefficient, sometimes extremely
> inefficent, and in various cases is not permitted at all.

I thought I'd share that here to start the ball rolling on how to fix
this for good.

I remember discussing the following options:

1. The patch I sent using memcpy and a temporary variable to avoid
nonaligned accesses.This approach should be correct, but I'm not sure
if more places are necessary to fix.
2. Packing and unpacking into an 8-byte wide stream instead of a
4-byte wide stream. This should work as well but it seems wasteful in
memory (not sure if that would ever be relevant).

I think option 1 would be minimal effort to fix this.

[1] https://lists.debian.org/debian-devel/2014/11/msg01011.html
[2] https://lists.debian.org/debian-devel/2014/11/msg01028.html


-- 

Saludos,
Felipe Sateler