[Bug rtl-optimization/80818] LRA clobbers live hard reg clobbered during rematerialization

[marxin at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80818

Martin Liška  changed:

   What|Removed |Added

 CC||marxin at gcc dot gnu.org

--- Comment #15 from Martin Liška  ---
Vladimir: Can you please update Known to work?

[Bug rtl-optimization/80818] LRA clobbers live hard reg clobbered during rematerialization

[krebbel at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80818

Andreas Krebbel  changed:

   What|Removed |Added

 Status|RESOLVED|REOPENED
   Last reconfirmed||2018-07-25
Version|8.0 |7.3.1
 Resolution|FIXED   |---
 Ever confirmed|0   |1

--- Comment #14 from Andreas Krebbel  ---
I ran into the same problem with current GCC 7 branch. Could you apply your fix
also to GCC 7 branch?

[Bug rtl-optimization/80818] LRA clobbers live hard reg clobbered during rematerialization

[law at redhat dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80818

Jeffrey A. Law  changed:

   What|Removed |Added

 Status|UNCONFIRMED |RESOLVED
 CC||law at redhat dot com
 Resolution|--- |FIXED

--- Comment #13 from Jeffrey A. Law  ---
Fixed by Vlad commit on the trunk.

[Bug rtl-optimization/80818] LRA clobbers live hard reg clobbered during rematerialization

[vmakarov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80818

--- Comment #12 from Vladimir Makarov  ---
Author: vmakarov
Date: Thu Dec  7 17:50:54 2017
New Revision: 255471

URL: https://gcc.gnu.org/viewcvs?rev=255471=gcc=rev
Log:
2017-12-07  Vladimir Makarov  

PR target/83252
PR rtl-optimization/80818
* lra.c (add_regs_to_insn_regno_info): Make a hard reg in CLOBBER
always early clobbered.
* lra-lives.c (process_bb_lives): Check input hard regs for early
clobbered non-operand hard reg.


Modified:
trunk/gcc/ChangeLog
trunk/gcc/lra-lives.c
trunk/gcc/lra.c

[Bug rtl-optimization/80818] LRA clobbers live hard reg clobbered during rematerialization

[vmakarov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80818

--- Comment #11 from Vladimir Makarov  ---
I am still working on this PR.  I hope to fix it on this week or on the next
one (the patch will need a lot of testing).

[Bug rtl-optimization/80818] LRA clobbers live hard reg clobbered during rematerialization

[vmakarov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80818

--- Comment #10 from Vladimir Makarov  ---
Author: vmakarov
Date: Wed Nov 29 22:19:25 2017
New Revision: 255258

URL: https://gcc.gnu.org/viewcvs?rev=255258=gcc=rev
Log:
2017-11-29  Vladimir Makarov  

PR rtl-optimization/80818
* lra.c (collect_non_operand_hard_regs): New arg insn.  Pass it
recursively.  Use insn code for clobber.
(lra_set_insn_recog_data): Pass the new arg to
collect_non_operand_hard_regs.
(add_regs_to_insn_regno_info): Pass insn instead of uid.  Use insn
code for clobber.
(lra_update_insn_regno_info): Pass insn to
add_regs_to_insn_regno_info.


Modified:
trunk/gcc/ChangeLog
trunk/gcc/lra.c

[Bug rtl-optimization/80818] LRA clobbers live hard reg clobbered during rematerialization

[vmakarov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80818

--- Comment #9 from Vladimir Makarov  ---
(In reply to Andreas Krebbel from comment #8)
> Hi Vladimir. What do you think about the additional patch?

Andreas, sorry for the delay with the answer.  The patch looks reasonable for
me.  If your additional patch works for you we could try to incorporate it into
my patch.  I'll do it and test the new patch and, if everything is ok, I'll
commit it tomorrow or on Wednesday.

[Bug rtl-optimization/80818] LRA clobbers live hard reg clobbered during rematerialization

[krebbel at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80818

Andreas Krebbel  changed:

   What|Removed |Added

   Priority|P3  |P2

--- Comment #8 from Andreas Krebbel  ---
Hi Vladimir. What do you think about the additional patch?

[Bug rtl-optimization/80818] LRA clobbers live hard reg clobbered during rematerialization

[krebbel at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80818

--- Comment #7 from Andreas Krebbel  ---
Created attachment 41699
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=41699=edit
Additional changes required ontop of the proposed patch

Changes which might be required on-top of your patch.

Your patch alone does not appear to fix the testcase for me. Neither the
reduced nor the full version. The hard reg clobber is added in
collect_non_operand_hard_regs not in add_regs_to_insn_regno_info.

Copying what your patch does in add_regs_to_insn_regno_info to
collect_non_operand_hard_regs does fix the problem for me.

[Bug rtl-optimization/80818] LRA clobbers live hard reg clobbered during rematerialization

[vmakarov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80818

--- Comment #6 from Vladimir Makarov  ---
Hi, Andreas.

Could you check the patch I attached on your bigger test case.  If it works, I
will commit it.

[Bug rtl-optimization/80818] LRA clobbers live hard reg clobbered during rematerialization

[vmakarov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80818

--- Comment #5 from Vladimir Makarov  ---
Created attachment 41686
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=41686=edit
A proposed patch

[Bug rtl-optimization/80818] LRA clobbers live hard reg clobbered during rematerialization

[krebbel at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80818

--- Comment #4 from Andreas Krebbel  ---
Created attachment 41609
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=41609=edit
Reduced testcase

gcc -O3 -march=z900 t.c

r33 is marked as unused in process_bb_lives for the compare (insn 10), although
the value is used by the add with carry (insn 18).

(insn 10 23 24 2 (parallel [
(set (reg:CCL3 33 %cc)
(compare:CCL3 (reg/v:DI 73 [orig:64 a ] [64])
(reg/v:DI 65 [ b ])))
(clobber (reg/v:DI 73 [orig:64 a ] [64]))
]) "t.c":6 1632 {*subdi3_cconly2}
 (expr_list:REG_UNUSED (reg:CC 33 %cc)
(expr_list:REG_UNUSED (reg/v:DI 73 [orig:64 a ] [64])
(expr_list:REG_DEAD (reg/v:DI 65 [ b ])
(nil)

[Bug rtl-optimization/80818] LRA clobbers live hard reg clobbered during rematerialization

[vmakarov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80818

--- Comment #3 from Vladimir Makarov  ---
(In reply to Andreas Krebbel from comment #0)
> Created attachment 41383 [details]
> Experimental patch

Andreas, thank you for working on the issue.  You are right.  The hard reg in
the clobber should be not be an earlyclobber one.  But, unfortunately the patch
is not fully correct.  Static data exists in one examplar for insn with the
same insn code and as consequence should not be dependent on other operands.

I'll think about the correct solution. Do you have a test case?  The patch
might be ready soon after I get the test.  Although it will take some time to
test the patch.

[Bug rtl-optimization/80818] LRA clobbers live hard reg clobbered during rematerialization

[krebbel at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80818

--- Comment #2 from Andreas Krebbel  ---
I think this is a different problem.  In my case the reason for adding an
invalid rematerialization appears to be that the live ranges of an hard
registers are not correctly calculated.

I've just verified with GCC head r249394:

For the two insns:

(insn 470 25077 476 10 (set (reg:CCU 33 %cc)
(compare:CCU (reg:DI 3 %r3 [orig:1277 _1997 ] [1277])
(mem/u/c:DI (reg:DI 5 %r5 [14737]) [5  S8 A64]))) "t.f90":48 1242
{*cmpdi_ccu}
 (nil))

(insn 472 27815 25079 10 (parallel [
(set (reg:SI 7 %r7 [orig:1282 _2003 ] [1282])
(plus:SI (plus:SI (gtu:SI (reg:CCU 33 %cc)
(const_int 0 [0]))
(reg:SI 7 %r7 [orig:1256 _1967 ] [1256]))
(reg:SI 2 %r2 [orig:1258 _1969 ] [1258])))
(clobber (reg:CC 33 %cc))
]) "t.f90":48 1661 {*addsi3_alc}
 (nil))

LRA records the following registers:

470:
{early_clobber_alts = 0, biggest_mode = DImode, type = OP_IN, subreg_p = 0, 
  early_clobber = 0, regno = 1277, next = 0x3cdb480}
{early_clobber_alts = 0, biggest_mode = DImode, type = OP_IN, subreg_p = 0, 
  early_clobber = 0, regno = 14737, next = 0x0}

hardregs:

{early_clobber_alts = 0, biggest_mode = CCUmode, type = OP_OUT, subreg_p = 0, 
  early_clobber = 0, regno = 33, next = 0x0}


472:

{early_clobber_alts = 0, biggest_mode = SImode, type = OP_OUT, subreg_p = 0, 
  early_clobber = 0, regno = 1282, next = 0x3cdb400}
{early_clobber_alts = 0, biggest_mode = SImode, type = OP_IN, subreg_p = 0, 
  early_clobber = 0, regno = 1256, next = 0x3cdb3e0}
{early_clobber_alts = 0, biggest_mode = SImode, type = OP_IN, subreg_p = 0, 
  early_clobber = 0, regno = 1258, next = 0x3cdb3c0}
{early_clobber_alts = 0, biggest_mode = CCUmode, type = OP_IN, subreg_p = 0, 
  early_clobber = 0, regno = 33, next = 0x0}

hardregs:
{early_clobber_alts = 18446744073709551615, biggest_mode = CCmode, type =  
OP_OUT, subreg_p = 0, early_clobber = 1, regno = 33, next = 0x0}

Marking r33 as early_clobber output makes it dead before insn 472 here:

lra-lives.c(915)
  /* Mark early clobber outputs dead.  */
  for (reg = curr_id->regs; reg != NULL; reg = reg->next)
if (reg->type == OP_OUT
&& reg_early_clobber_p (reg, n_alt) && ! reg->subreg_p)
  need_curr_point_incr
|= mark_regno_dead (reg->regno, reg->biggest_mode,
curr_point);

Reaching insn 470 with r33 being dead the reg is consequently marked as
"unused":

(insn 470 25077 476 10 (set (reg:CCU 33 %cc)
(compare:CCU (reg:DI 1277 [ _1997 ])
(mem/u/c:DI (reg:DI 14737) [5  S8 A64]))) "t.f90":48 1242
{*cmpdi_ccu}
 (expr_list:REG_UNUSED (reg:CC 33 %cc)
(expr_list:REG_DEAD (reg:DI 14737)
(nil

Since r33 is also an input operand to insn 472 it should not be marked as
early_clobber. The attached patch prevents that and fixes the problem for me.

[Bug rtl-optimization/80818] LRA clobbers live hard reg clobbered during rematerialization

[ktkachov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=80818

ktkachov at gcc dot gnu.org changed:

   What|Removed |Added

 CC||ktkachov at gcc dot gnu.org

--- Comment #1 from ktkachov at gcc dot gnu.org ---
Looks somewhat similar to PR 80754 ?