[Bug inline-asm/70184] Explicit register variables holding function arguments overwritten by conversion libcall

[ramana at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70184

Ramana Radhakrishnan  changed:

   What|Removed |Added

   Target Milestone|--- |6.3

--- Comment #11 from Ramana Radhakrishnan  ---
Fixed in 6.3

[Bug inline-asm/70184] Explicit register variables holding function arguments overwritten by conversion libcall

[ktkachov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70184

ktkachov at gcc dot gnu.org changed:

   What|Removed |Added

 Status|NEW |RESOLVED
 Resolution|--- |FIXED

--- Comment #10 from ktkachov at gcc dot gnu.org ---
Fixed.

[Bug inline-asm/70184] Explicit register variables holding function arguments overwritten by conversion libcall

[ktkachov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70184

--- Comment #9 from ktkachov at gcc dot gnu.org ---
Author: ktkachov
Date: Thu Dec  1 13:47:13 2016
New Revision: 243109

URL: https://gcc.gnu.org/viewcvs?rev=243109=gcc=rev
Log:
[TER] PR target/48863: Don't replace expressions across local register variable
definitions

Backport from mainline
2016-11-24  Kyrylo Tkachov  

PR target/48863
PR inline-asm/70184
* tree-ssa-ter.c (temp_expr_table): Add reg_vars_cnt field.
(new_temp_expr_table): Initialise reg_vars_cnt.
(free_temp_expr_table): Release reg_vars_cnt.
(process_replaceable): Add reg_vars_cnt argument, set reg_vars_cnt
field of TAB.
(find_replaceable_in_bb): Use the above to record register variable
write occurrences and cancel replacement across them.

* gcc.target/arm/pr48863.c: New test.


Added:
branches/gcc-6-branch/gcc/testsuite/gcc.target/arm/pr48863.c
Modified:
branches/gcc-6-branch/gcc/ChangeLog
branches/gcc-6-branch/gcc/testsuite/ChangeLog
branches/gcc-6-branch/gcc/tree-ssa-ter.c

[Bug inline-asm/70184] Explicit register variables holding function arguments overwritten by conversion libcall

[ktkachov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70184

--- Comment #8 from ktkachov at gcc dot gnu.org ---
Author: ktkachov
Date: Thu Nov 24 15:22:34 2016
New Revision: 242840

URL: https://gcc.gnu.org/viewcvs?rev=242840=gcc=rev
Log:
[TER] PR target/48863 : Don't replace expressions across local register
variable definitions

PR target/48863
PR inline-asm/70184
* tree-ssa-ter.c (temp_expr_table): Add reg_vars_cnt field.
(new_temp_expr_table): Initialise reg_vars_cnt.
(free_temp_expr_table): Release reg_vars_cnt.
(process_replaceable): Add reg_vars_cnt argument, set reg_vars_cnt
field of TAB.
(find_replaceable_in_bb): Use the above to record register variable
write occurrences and cancel replacement across them.

* gcc.target/arm/pr48863.c: New test.


Added:
trunk/gcc/testsuite/gcc.target/arm/pr48863.c
Modified:
trunk/gcc/ChangeLog
trunk/gcc/testsuite/ChangeLog
trunk/gcc/tree-ssa-ter.c

[Bug inline-asm/70184] Explicit register variables holding function arguments overwritten by conversion libcall

[ktkachov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70184

--- Comment #7 from ktkachov at gcc dot gnu.org ---
(In reply to Richard Biener from comment #4)
> With TER we delay expanding of (unsigned b) until you require its expansion
> during asm op expansion (and thus may be interleaved with asm expansion
> code).
> 
> I believe we had this issue in the past for other archs and libcalls.  Don't
> remember what the solution there was though ;)  IIRC it was libcalls
> expanded during libcall expansion or so like for a + (b * c) and both
> + and * resulting in a libcall.

I see that tree-ssa-ter.c:find_replaceable_in_bb already tries to avoid
replacing across calls but it doesn't take into account expressions that expand
into libcalls and doesn't consider register variables.
I don't think there's an easy way to query from gimple if an arbitrary
expression will result in a libcall, maybe the solution here is to not allow
TER across definitions of register variables? Or is that too big a hammer?

[Bug inline-asm/70184] Explicit register variables holding function arguments overwritten by conversion libcall

[ramana at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70184

Ramana Radhakrishnan  changed:

   What|Removed |Added

 Status|UNCONFIRMED |NEW
   Last reconfirmed||2016-03-31
 CC||ramana at gcc dot gnu.org
 Ever confirmed|0   |1

--- Comment #6 from Ramana Radhakrishnan  ---
Confirmed - but not sure if this is a dup of PR48863

[Bug inline-asm/70184] Explicit register variables holding function arguments overwritten by conversion libcall

[mikpelinux at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70184

Mikael Pettersson  changed:

   What|Removed |Added

 CC||mikpelinux at gmail dot com

--- Comment #5 from Mikael Pettersson  ---
Looks similar to PR48863.

[Bug inline-asm/70184] Explicit register variables holding function arguments overwritten by conversion libcall

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70184

--- Comment #4 from Richard Biener  ---
With TER we delay expanding of (unsigned b) until you require its expansion
during asm op expansion (and thus may be interleaved with asm expansion code).

I believe we had this issue in the past for other archs and libcalls.  Don't
remember what the solution there was though ;)  IIRC it was libcalls
expanded during libcall expansion or so like for a + (b * c) and both
+ and * resulting in a libcall.

[Bug inline-asm/70184] Explicit register variables holding function arguments overwritten by conversion libcall

[ktkachov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70184

--- Comment #3 from ktkachov at gcc dot gnu.org ---
(In reply to Andrew Pinski from comment #2)
> Does -fno-ter fix the issue?
> How does the tree level look?

You mean -fno-tree-ter?
That gives:
foo:
@ args = 0, pretend = 0, frame = 0
@ frame_needed = 0, uses_anonymous_args = 0
push{r4, lr}
mov r4, r0
mov r0, r1
bl  __aeabi_f2uiz
mov r1, r0
mov r0, r4
.syntax divided
@ 7 "foo.c" 1
str r1, [r0]

@ 0 "" 2
.syntax unified
pop {r4, pc}

Which looks correct, thanks!

The optimized tree dump looks right to me with or without -fno-tree-ter

;; Function foo (foo, funcdef_no=0, decl_uid=4090, cgraph_uid=0,
symbol_order=0)

foo (unsigned int a, float b)
{
  register unsigned int r1 __asm__ (*r1);
  register unsigned int r0 __asm__ (*r0);
  unsigned int c;

  :
  c_2 = (unsigned int) b_1(D);
  r0 = a_4(D);
  r1 = c_2;
  __asm__ __volatile__("str %1, [%0]
" :  : "r0" "r" r0, "r1" "r" r1);
  return;

}

[Bug inline-asm/70184] Explicit register variables holding function arguments overwritten by conversion libcall

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70184

--- Comment #2 from Andrew Pinski  ---
Does -fno-ter fix the issue?
How does the tree level look?

[Bug inline-asm/70184] Explicit register variables holding function arguments overwritten by conversion libcall

[ktkachov at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=70184

--- Comment #1 from ktkachov at gcc dot gnu.org ---
Things are bad already at expand time:
2: r111:SI=r0:SI
3: r112:SF=r1:SF
4: NOTE_INSN_FUNCTION_BEG
7: r0:SI=r111:SI
8: r0:SF=r112:SF
9: r0:SI=call [`__aeabi_f2uiz'] argc:0
  REG_CALL_DECL `__aeabi_f2uiz'
  REG_EH_REGION 0x8000
   10: r113:SI=r0:SI
  REG_EQUAL uns_fix(r112:SF)
   11: r1:SI=r113:SI
   12: asm_operands