[Bug target/54760] New: [SH] Add __builtin_thread_pointer, __builtin_set_thread_pointer

[Bug target/54760] New: [SH] Add __builtin_thread_pointer, __builtin_set_thread_pointer

[olegendo at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54760

             Bug #: 54760
           Summary: [SH] Add __builtin_thread_pointer,
                    __builtin_set_thread_pointer
    Classification: Unclassified
           Product: gcc
           Version: 4.8.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
        AssignedTo: olegendo@gcc.gnu.org
        ReportedBy: olegendo@gcc.gnu.org
            Target: sh*-*-*


SH normally uses GBR to store the address of the thread control block (TCB). 
Other targets have __builtin_thread_pointer and __builtin_set_thread_pointer to
modify such registers.  It has been proposed to make these machine independent
builtins:
http://gcc.gnu.org/ml/gcc-patches/2012-07/msg00428.html

but so far this has not made it into mainline.  In any case the builtin
functions should be added for SH to get/set the GBR.  Moreover, code that
accesses variables in the TCB like

struct tcb
{
  int a, b, c, d;
};

int test (void)
{
  return ((tcb*)__builtin_thread_pointer ())->c;
}

ideally would result in:

  rts
  mov.l  @(8,gbr),r0

[Bug target/54760] [SH] Add __builtin_thread_pointer, __builtin_set_thread_pointer

[olegendo at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54760

Oleg Endo <olegendo at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |NEW
   Last reconfirmed|                            |2012-09-30
     Ever Confirmed|0                           |1

[Bug target/54760] [SH] Add __builtin_thread_pointer, __builtin_set_thread_pointer

[olegendo at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54760

--- Comment #1 from Oleg Endo <olegendo at gcc dot gnu.org> 2012-10-06 11:20:18 UTC ---
Author: olegendo
Date: Sat Oct  6 11:20:11 2012
New Revision: 192155

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=192155
Log:
    PR target/54760
    * config/sh/sh.md (define_constants): Add UNSPECV_GBR.
    (get_thread_pointer, set_thread_pointer): New expanders.
    (load_gbr): Rename to store_gbr.  Remove GBR_REG use.
    (store_gbr): New insn.
    * config/sh/sh.c (prepare_move_operands): Use gen_store_gbr instead of
    gen_load_gbr in TLS_MODEL_LOCAL_EXEC case.
    (sh1_builtin_p): New function.
    (signature_args): Add SH_BLTIN_VP.
    (bdesc): Add __builtin_thread_pointer and __builtin_set_thread_pointer.

    PR target/54760
    * gcc.target/sh/pr54760-1.c: New.


Added:
    trunk/gcc/testsuite/gcc.target/sh/pr54760-1.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/sh/sh.c
    trunk/gcc/config/sh/sh.md
    trunk/gcc/testsuite/ChangeLog

[Bug target/54760] [SH] Add __builtin_thread_pointer, __builtin_set_thread_pointer

[olegendo at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54760

--- Comment #2 from Oleg Endo <olegendo at gcc dot gnu.org> 2012-10-08 02:00:57 UTC ---
Author: olegendo
Date: Mon Oct  8 02:00:46 2012
New Revision: 192193

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=192193
Log:
    PR target/54760
    * config/sh/sh.md (*mov<mode>_gbr_load, *mov<mode>_gbr_store): New
    insns and accompanying unnamed splits.
    * config/sh/predicates.md (general_movsrc_operand,
    general_movdst_operand): Reject GBR addresses.
    * config/sh/sh-protos.h (sh_find_equiv_gbr_addr): New declaration.
    * config/sh/sh.c (sh_address_cost, sh_legitimate_address_p,
    sh_secondary_reload): Handle GBR addresses.
    (base_reg_disp): New class.
    (sh_find_base_reg_disp, sh_find_equiv_gbr_addr): New functions.

    PR target/54760
    * gcc.target/sh/pr54760-2.c: New.
    * gcc.target/sh/pr54760-3.c: New.


Added:
    trunk/gcc/testsuite/gcc.target/sh/pr54760-2.c
    trunk/gcc/testsuite/gcc.target/sh/pr54760-3.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/sh/predicates.md
    trunk/gcc/config/sh/sh-protos.h
    trunk/gcc/config/sh/sh.c
    trunk/gcc/config/sh/sh.md
    trunk/gcc/testsuite/ChangeLog

[Bug target/54760] [SH] Add __builtin_thread_pointer, __builtin_set_thread_pointer

[olegendo at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54760

--- Comment #3 from Oleg Endo <olegendo at gcc dot gnu.org> 2012-10-09 18:59:19 UTC ---
Author: olegendo
Date: Tue Oct  9 18:59:11 2012
New Revision: 192269

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=192269
Log:
    PR target/54760
    * doc/extend.texi (Target Builtins): Add SH built-in section.
    Document __builtin_thread_pointer and __builtin_set_thread_pointer.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/doc/extend.texi

[Bug target/54760] [SH] Add __builtin_thread_pointer, __builtin_set_thread_pointer

[olegendo at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54760

Oleg Endo <olegendo at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |RESOLVED
         Resolution|                            |FIXED

--- Comment #4 from Oleg Endo <olegendo at gcc dot gnu.org> 2012-10-09 19:42:33 UTC ---
Done in 4.8.

[Bug target/54760] [SH] Add __builtin_thread_pointer, __builtin_set_thread_pointer

[olegendo at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54760

Oleg Endo <olegendo at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|RESOLVED                    |REOPENED
                 CC|                            |kkojima at gcc dot gnu.org
         Resolution|FIXED                       |

--- Comment #5 from Oleg Endo <olegendo at gcc dot gnu.org> 2012-10-11 20:29:12 UTC ---
It seems that the proposed target independent thread pointer stuff made it into
mainline.  I guess the SH specific built-ins can be removed now.  I will check
it out.

Another thing that I'm not sure about is whether 'ldc  rn,gbr'  and 'stc 
gbr,rn' instructions can go in the delay slot, or whether they will cause a
slot illegal exception.  Currently this 

void* test (void)
{
  return __builtin_thread_pointer ();
}

will result in

  rts
  stc   gbr,r0

I've looked in the source of the GDB sim and it seems that the sim will not
detect this as a slot illegal instruction.  However, the sim will also not
detect 'stc  sgr,rn' as a slot illegal instruction, but in my older
interrupt/exception code I was doing:

  stc  sgr,r15
  rte
  nop

because otherwise it would cause trouble.
The HW manuals unfortunately don't mention any details.  Basically all of them
say: stc, stc.l, ldc, ldc.l instructions might cause a slot illegal exception. 
But it doesn't say which register combinations are slot illegal and which are
legal.

Kaz, do you happen to know something regarding this matter?

[Bug target/54760] [SH] Add __builtin_thread_pointer, __builtin_set_thread_pointer

[kkojima at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54760

--- Comment #6 from Kazumoto Kojima <kkojima at gcc dot gnu.org> 2012-10-12 00:13:51 UTC ---
(In reply to comment #5)
> Kaz, do you happen to know something regarding this matter?

My SH4 software manual says for STC that all stc/stc.l instructions
except stc gbr,rn/stc gbr,@-rn are privileged instructions which
might cause an illegal exception in user mode.  LDC has a similar
remark.

[Bug target/54760] [SH] Add __builtin_thread_pointer, __builtin_set_thread_pointer

[olegendo at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54760

--- Comment #7 from Oleg Endo <olegendo at gcc dot gnu.org> 2012-10-12 00:26:42 UTC ---
(In reply to comment #6)
> (In reply to comment #5)
> > Kaz, do you happen to know something regarding this matter?
> 
> My SH4 software manual says for STC that all stc/stc.l instructions
> except stc gbr,rn/stc gbr,@-rn are privileged instructions which
> might cause an illegal exception in user mode.  LDC has a similar
> remark.

Yes, but e.g. in the manual rej09b0003_sh4a.pdf, pdf page 346 (STC insn), in
addition to the gbr user/privileged mode note:

Possible Exceptions:
* Data TLB multiple-hit exception
* General illegal instruction exception
* Slot illegal instruction exception      <<< but which insns?!?!
* Data TLB miss exception
* Data TLB protection violation exception
* Initial page write exception
* Data address error

I'd like to leave this open until this has been clarified/confirmed.

[Bug target/54760] [SH] Add __builtin_thread_pointer, __builtin_set_thread_pointer

[kkojima at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54760

--- Comment #8 from Kazumoto Kojima <kkojima at gcc dot gnu.org> 2012-10-12 00:43:09 UTC ---
(In reply to comment #7)
> * Slot illegal instruction exception      <<< but which insns?!?!

Ah, you could see a list in that manual rej09b0003_sh4a.pdf,
pdf page 108, Slot Illegal Instruction Exception:

... Previleged instructions:LDC,STC,RTE,LDTLB,SLEEP,but
    excluding LDC/STC instructions that access GBR ...

[Bug target/54760] [SH] Add __builtin_thread_pointer, __builtin_set_thread_pointer

[olegendo at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54760

--- Comment #9 from Oleg Endo <olegendo at gcc dot gnu.org> 2012-10-12 00:45:45 UTC ---
Author: olegendo
Date: Fri Oct 12 00:45:36 2012
New Revision: 192388

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=192388
Log:
    PR target/54760
    * config/sh/sh.c (bdesc): Remove thread pointer built-ins.
    * config/sh/sh.md (get_thread_pointer, set_thread_pointer): Append mode
    name 'si'.


Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/sh/sh.c
    trunk/gcc/config/sh/sh.md

[Bug target/54760] [SH] Add __builtin_thread_pointer, __builtin_set_thread_pointer

[olegendo at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54760

--- Comment #10 from Oleg Endo <olegendo at gcc dot gnu.org> 2012-10-12 00:50:44 UTC ---
(In reply to comment #8)
> (In reply to comment #7)
> > * Slot illegal instruction exception      <<< but which insns?!?!
> 
> Ah, you could see a list in that manual rej09b0003_sh4a.pdf,
> pdf page 108, Slot Illegal Instruction Exception:
> 
> ... Previleged instructions:LDC,STC,RTE,LDTLB,SLEEP,but
>     excluding LDC/STC instructions that access GBR ...

Ouch, right.  Thanks.  Should be good then as it is I guess (except for the new
target independent built-in expansion failure ...)

[Bug target/54760] [SH] Add __builtin_thread_pointer, __builtin_set_thread_pointer

[olegendo at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54760

--- Comment #11 from Oleg Endo <olegendo at gcc dot gnu.org> 2012-10-15 22:04:42 UTC ---
Author: olegendo
Date: Mon Oct 15 22:04:37 2012
New Revision: 192480

URL: http://gcc.gnu.org/viewcvs?root=gcc&view=rev&rev=192480
Log:
    PR target/54760
    * config/sh/sh.c (sh_find_base_reg_disp): Stop searching insns when
    hitting a call insn if GBR is marked as call used.
    * config/sh/iterators.md (QIHISIDI): New mode iterator.
    * config/sh/predicates.md (gbr_address_mem): New predicate.
    * config/sh/sh.md (*movdi_gbr_load, *movdi_gbr_store): New
    insn_and_split.
    Use QIHISIDI instead of QIHISI in unnamed GBR addressing splits.

    PR target/54760
    * gcc.target/sh/pr54760-2.c: Add long long and unsigned long long test
    functions.
    * gcc.target/sh/pr54760-4.c: New.


Added:
    trunk/gcc/testsuite/gcc.target/sh/pr54760-4.c
Modified:
    trunk/gcc/ChangeLog
    trunk/gcc/config/sh/iterators.md
    trunk/gcc/config/sh/predicates.md
    trunk/gcc/config/sh/sh.c
    trunk/gcc/config/sh/sh.md
    trunk/gcc/testsuite/ChangeLog
    trunk/gcc/testsuite/gcc.target/sh/pr54760-2.c

[Bug target/54760] [SH] Add __builtin_thread_pointer, __builtin_set_thread_pointer

[amylaar at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54760

Jorn Wolfgang Rennecke <amylaar at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |amylaar at gcc dot gnu.org

--- Comment #12 from Jorn Wolfgang Rennecke <amylaar at gcc dot gnu.org> 2012-10-23 19:25:02 UTC ---
(In reply to comment #9)

> Log:
>     PR target/54760
>     * config/sh/sh.c (bdesc): Remove thread pointer built-ins.

You left sh1_builtin_p behind.

This caused PR55042.

[Bug target/54760] [SH] Add __builtin_thread_pointer, __builtin_set_thread_pointer

[olegendo at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54760

--- Comment #13 from Oleg Endo <olegendo at gcc dot gnu.org> 2012-10-27 13:41:25 UTC ---
(In reply to comment #12)
> (In reply to comment #9)
> 
> > Log:
> >     PR target/54760
> >     * config/sh/sh.c (bdesc): Remove thread pointer built-ins.
> 
> You left sh1_builtin_p behind.
> 
> This caused PR55042.

Fixed in rev 192877.

[Bug target/54760] [SH] Add __builtin_thread_pointer, __builtin_set_thread_pointer

[olegendo at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=54760

Oleg Endo <olegendo at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|REOPENED                    |RESOLVED
         Resolution|                            |FIXED

--- Comment #14 from Oleg Endo <olegendo at gcc dot gnu.org> 2013-03-09 12:35:56 UTC ---
As far as I can tell, this is one is done.