Issues with devel/boost-* on Sparc64

[Alexander Churanov]

HI folks!

As I know, currently devel/boost-libs port fails to build on sparc64.
I had a discussion of this in September. The root cause is unknown for
me. To investigate into this further I need either access to a sparc64
box or a person who has access and whom I may instruct with the
actions to perform.

Are there any plans to fix ports for sparc64 before the release?

Sincerely,
Alexander Churanov,
maintainer of devel/boost-*
_______________________________________________
freebsd-ports@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to "freebsd-ports-unsubscribe@..."

[Boris Kochergin]

Alexander Churanov wrote:

> HI folks!
>
> As I know, currently devel/boost-libs port fails to build on sparc64.
> I had a discussion of this in September. The root cause is unknown for
> me. To investigate into this further I need either access to a sparc64
> box or a person who has access and whom I may instruct with the
> actions to perform.
>
> Are there any plans to fix ports for sparc64 before the release?
>
> Sincerely,
> Alexander Churanov,
> maintainer of devel/boost-*
> _______________________________________________
> freebsd-ports@... mailing list
> http://lists.freebsd.org/mailman/listinfo/freebsd-ports
> To unsubscribe, send any mail to "freebsd-ports-unsubscribe@..."
>  

Ahoy. I can give you shell access to a reasonably-speedy sparc64 machine
(4 x 300 MHz) running 8.0-BETA.

-Boris
_______________________________________________
freebsd-ports@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to "freebsd-ports-unsubscribe@..."

[Alexander Churanov]

Boris,

This would be very handy. Just tell me at what time I can use it. I
mean at anytime during a week or just several hours when the box is
free? If there is a specific time, I should prepare for it.

Please, e-mail me the details privately.
Thank you for co-operation.

Sincerely,
Alexander Churanov,
maintainer of devel/boost-*
_______________________________________________
freebsd-ports@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to "freebsd-ports-unsubscribe@..."

[Mark Linimon-2]

On Sat, Nov 07, 2009 at 09:11:56PM -0500, Boris Kochergin wrote:
> Are there any plans to fix ports for sparc64 before the release?

I had been planning to do this but other breakage, and paying work, has
kept me from getting to it.  I'd be happy to see someone else take a look
at it, and I'll also be happy to re-run the sparc64 8.0 package set when
the fix is ready.

mcl
_______________________________________________
freebsd-ports@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to "freebsd-ports-unsubscribe@..."

[Alexander Churanov]

OK, Mark, no problems.

Alexander Churanov
_______________________________________________
freebsd-ports@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to "freebsd-ports-unsubscribe@..."

[Eygene Ryabinkin-3]

Boris, good day.

Sat, Nov 07, 2009 at 09:11:56PM -0500, Boris Kochergin wrote:

> Alexander Churanov wrote:
> > HI folks!
> >
> > As I know, currently devel/boost-libs port fails to build on sparc64.
> > I had a discussion of this in September. The root cause is unknown for
> > me. To investigate into this further I need either access to a sparc64
> > box or a person who has access and whom I may instruct with the
> > actions to perform.
> >
> > Are there any plans to fix ports for sparc64 before the release?
> >
> > Sincerely,
> > Alexander Churanov,
> > maintainer of devel/boost-*
> > _______________________________________________
> > freebsd-ports@... mailing list
> > http://lists.freebsd.org/mailman/listinfo/freebsd-ports
> > To unsubscribe, send any mail to "freebsd-ports-unsubscribe@..."
> >  
> Ahoy. I can give you shell access to a reasonably-speedy sparc64 machine
> (4 x 300 MHz) running 8.0-BETA.

I am willing to try.  How to proceed?  If you'll need my SSH key, it is
in the attachment.
--
Eygene
 _                ___       _.--.   #
 \`.|\..----...-'`   `-._.-'_.-'`   #  Remember that it is hard
 /  ' `         ,       __.--'      #  to read the on-line manual
 )/' _/     \   `-_,   /            #  while single-stepping the kernel.
 `-'" `"\_  ,_.-;_.-\_ ',  fsc/as   #
     _.-'_./   {_.'   ; /           #    -- FreeBSD Developers handbook
    {_.-``-'         {_/            #

ssh-rsa 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 rea@...

_______________________________________________
freebsd-ports@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to "freebsd-ports-unsubscribe@..."

[Eygene Ryabinkin-3]

Sat, Nov 07, 2009 at 09:11:56PM -0500, Boris Kochergin wrote:
> > As I know, currently devel/boost-libs port fails to build on sparc64.
> > I had a discussion of this in September. The root cause is unknown for
> > me. To investigate into this further I need either access to a sparc64
> > box or a person who has access and whom I may instruct with the
> > actions to perform.

From the preliminary investigations I done on the Boris's machine,
it turns out that as(1) chokes on the "cas" instruction that looks
like "cas [%l0+12], %g2, %g1".  The instruction itself looks sane,
so I'll try to understand if as(1) has some bugs inside it or something
isn't good with the instruction itself.
--
Eygene
 _                ___       _.--.   #
 \`.|\..----...-'`   `-._.-'_.-'`   #  Remember that it is hard
 /  ' `         ,       __.--'      #  to read the on-line manual
 )/' _/     \   `-_,   /            #  while single-stepping the kernel.
 `-'" `"\_  ,_.-;_.-\_ ',  fsc/as   #
     _.-'_./   {_.'   ; /           #    -- FreeBSD Developers handbook
    {_.-``-'         {_/            #
_______________________________________________
freebsd-ports@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to "freebsd-ports-unsubscribe@..."

[Eygene Ryabinkin-3]

Mon, Nov 09, 2009 at 04:12:01PM +0300, Eygene Ryabinkin wrote:

> Sat, Nov 07, 2009 at 09:11:56PM -0500, Boris Kochergin wrote:
> > > As I know, currently devel/boost-libs port fails to build on sparc64.
> > > I had a discussion of this in September. The root cause is unknown for
> > > me. To investigate into this further I need either access to a sparc64
> > > box or a person who has access and whom I may instruct with the
> > > actions to perform.
>
> From the preliminary investigations I done on the Boris's machine,
> it turns out that as(1) chokes on the "cas" instruction that looks
> like "cas [%l0+12], %g2, %g1".  The instruction itself looks sane,
> so I'll try to understand if as(1) has some bugs inside it or something
> isn't good with the instruction itself.

Hmm, looks like only memory address taken straight from the register is
allowed for the Sparc.  So, [%l0] will be good, but [%l0+12] -- won't.
So it seems to me that g++ is generating improper assembly code.  Will
look into it a bit further.
--
Eygene
 _                ___       _.--.   #
 \`.|\..----...-'`   `-._.-'_.-'`   #  Remember that it is hard
 /  ' `         ,       __.--'      #  to read the on-line manual
 )/' _/     \   `-_,   /            #  while single-stepping the kernel.
 `-'" `"\_  ,_.-;_.-\_ ',  fsc/as   #
     _.-'_./   {_.'   ; /           #    -- FreeBSD Developers handbook
    {_.-``-'         {_/            #
_______________________________________________
freebsd-ports@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to "freebsd-ports-unsubscribe@..."

[Alexander Churanov]

Folks,

I've identified the root cause of the issue.
Boost folks are using custom routine written in assembly language for SPARC.

If the line 57 of boost/smart_ptr/detail/sp_counted_base.hpp is
commented, the implementation switches to spin-locked and code
compiles successfully.

I will contact Boost developers in order to fix this in upstream.
I'll also create a patch for devel/boost-libs port for using in the meanwhile.

Alexander Churanov
_______________________________________________
freebsd-ports@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to "freebsd-ports-unsubscribe@..."

[Alexander Churanov]

Guys,

Boost mailing list contains a record about using CAS instruction for
shared_ptr. It is of 2004. I'll investigate into it further. It is
necessary to understand why that library was working before 1.39.

Alexander Churanov
_______________________________________________
freebsd-ports@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to "freebsd-ports-unsubscribe@..."

[Alexander Churanov]

Folks,

It seems that I have found a workaround. It is to store the address of
the variable in a register and then issue the "CAS" instruction, which
only accepts register for an address. I am currently testing it.

Alexander Churanov
_______________________________________________
freebsd-ports@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to "freebsd-ports-unsubscribe@..."

[Eygene Ryabinkin-3]

Alexander, good day.

Wed, Nov 11, 2009 at 11:04:59PM +0300, Alexander Churanov wrote:
> It seems that I have found a workaround. It is to store the address of
> the variable in a register and then issue the "CAS" instruction, which
> only accepts register for an address. I am currently testing it.

I hadn't seen the code you're testing, but it seems to me that the
attached patch should automatically take care of making the address to
be in some register (if it isn't already in the register).  It compiles
well, and even works as a stand-alone function, but I hadn't yet tested
if boost built with such a atomic primitive really works fine.  Have
some $JOB to do prior to the testing.

If you're storing the address in the register manually -- there is no
need for it: you can use assembly constraints; this also makes less
headache, because compiler decides what register should be allocated by
its own and typically it is doing this job well.  Anyway, I hadn't seen
your patch, so may be you already doing this.
--
Eygene
 _                ___       _.--.   #
 \`.|\..----...-'`   `-._.-'_.-'`   #  Remember that it is hard
 /  ' `         ,       __.--'      #  to read the on-line manual
 )/' _/     \   `-_,   /            #  while single-stepping the kernel.
 `-'" `"\_  ,_.-;_.-\_ ',  fsc/as   #
     _.-'_./   {_.'   ; /           #    -- FreeBSD Developers handbook
    {_.-``-'         {_/            #

---

--- boost/smart_ptr/detail/sp_counted_base_gcc_sparc.hpp.orig 2009-03-02 11:22:16.000000000 -0500
+++ boost/smart_ptr/detail/sp_counted_base_gcc_sparc.hpp 2009-11-12 09:51:25.000000000 -0500
@@ -30,10 +30,10 @@
 
 inline int32_t compare_and_swap( int32_t * dest_, int32_t compare_, int32_t swap_ )
 {
-    __asm__ __volatile__( "cas %0, %2, %1"
-                        : "+m" (*dest_), "+r" (swap_)
-                        : "r" (compare_)
-                        : "memory" );
+    __asm__ __volatile__( "casa [%3] 0x80, %2, %1"
+ : "=m" (*dest_), "=r" (swap_)
+ : "r" (compare_), "r" (dest_), "1" (swap_), "m" (*dest_)
+ : "memory" );
 
     return swap_;
 }

---

_______________________________________________
freebsd-ports@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to "freebsd-ports-unsubscribe@..."

[Alexander Churanov]

Eygene,

Your solution is clearly more appropriate. I'll test it.
Mine is attached.

Alexander Churanov


_______________________________________________
freebsd-ports@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to "freebsd-ports-unsubscribe@..."

[Alexander Churanov]

Foks,

The current status of the issue is as follows:

*) I've contacted a boost team, they suggested a smaller patch

*) I've applied a patch, this allows code to build, but one the tests for
the patched code (in "libs/smart_ptr/test") fail. This is not one of the
tests that were expected to fail when there are issues with concurrency.

*) I've filed a bug in the boost issue tracker.

patch:
http://alexanderchuranov.com/boost-port/boost-1.41-sparc64/patch-boost_smart_ptr_detail_sp_counted_base_gcc_sparc.hpp
test log:
http://alexanderchuranov.com/boost-port/boost-1.41-sparc64/bjam-smart-ptr-test.log
issue: https://svn.boost.org/trac/boost/ticket/3678
boost mail thread:
http://old.nabble.com/-1.41.0--Beta-1-available-td26196304i20.html

Currently I'm waiting for the boost team to provide us with further
directions.

Alexander Churanov
_______________________________________________
freebsd-ports@... mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-ports
To unsubscribe, send any mail to "freebsd-ports-unsubscribe@..."