Re: [PATCH] [lra] force reg update after spilling to memory [PR120424]

[Georg-Johann Lay]

This patch introduces an ICE in lra-eliminations.cc:1200 for
an existing test case.
In $builddir/gcc:

$ make -k check-gcc RUNTESTFLAGS="--target_board=atmega128-sim 
avr-torture.exp=pr118591-1.c"

FAIL: gcc.target/avr/torture/pr118591-1.c   -O1  (internal compiler 
error: in update_reg_eliminate, at lra-eliminations.cc:1200)
FAIL: gcc.target/avr/torture/pr118591-1.c   -O2  (internal compiler 
error: in update_reg_eliminate, at lra-eliminations.cc:1200)
...

Configured with: --target=avr --disable-nls --with-dwarf2 --with-gnu-as 
--with-gnu-ld --enable-languages=c,c++

Please don't hesitate to ask me when you have problems reproducing the 
ICE.  Notes on how to reproduce can also be found at

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=118591#c3

Configured with: --target=avr --disable-nls --with-dwarf2 --with-gnu-as 
--with-gnu-ld --enable-languages=c,c++

Johann

--

Am 07.06.25 um 09:03 schrieb Denis Chertykov:
Alexandre Oliva <ol...@adacore.com> writes:

On Jun  6, 2025, Alexandre Oliva <ol...@adacore.com> wrote:

Now, since lra_update_fp2sp_elimination checks that
!elimination_fp2sp_occured_p, we *could* disable the fp2sp elimination,
if it's selected, right away, so that it is not applied after we've
disabled it, and then we don't have to worry about disabling it half-way
or reversing it later.

[lra] inactivate disabled fp2sp elimination

Even after we disable the fp2sp elimination when it is the active
elimination for the fp, spilling might use it before
update_reg_eliminate runs and inactivates it for good.  If it is used,
update_reg_eliminate will fail the check that fp2sp was not used.

Since we keep track of uses of this specific elimination, and
lra_update_fp2sp_elimination checks it before disabling it, we know it
hasn't been used, so we can inactivate it without any ill effects.

This fixes the pr118591-1.c avr-none regression exposed by the
PR120424 fix.

Regstrapped on x86_64-linux-gnu.  Also testing with gcc-14 on
arm-vx7r2.  Ok to install?


for  gcc/ChangeLog

        * lra-eliminations.cc (lra_update_fp2sp_elimination):
        Inactivate the unused fp2sp elimination right away.

Thank you.
The patch fixes PR118591.


Denis


---
  gcc/lra-eliminations.cc |   15 ++++++++++++---
  1 file changed, 12 insertions(+), 3 deletions(-)

diff --git a/gcc/lra-eliminations.cc b/gcc/lra-eliminations.cc
index bb708b007a4ee..6c8c91086f323 100644
--- a/gcc/lra-eliminations.cc
+++ b/gcc/lra-eliminations.cc
@@ -1415,6 +1415,14 @@ lra_update_fp2sp_elimination (int *spilled_pseudos)
    if (frame_pointer_needed || !targetm.frame_pointer_required ())
      return 0;
    gcc_assert (!elimination_fp2sp_occured_p);
+  ep = elimination_map[FRAME_POINTER_REGNUM];
+  if (ep->to == STACK_POINTER_REGNUM)
+    {
+      elimination_map[FRAME_POINTER_REGNUM] = NULL;
+      setup_can_eliminate (ep, false);
+    }
+  else
+    ep = NULL;
    if (lra_dump_file != NULL)
      fprintf (lra_dump_file,
             "        Frame pointer can not be eliminated anymore\n");
@@ -1422,9 +1430,10 @@ lra_update_fp2sp_elimination (int *spilled_pseudos)
    CLEAR_HARD_REG_SET (set);
    add_to_hard_reg_set (&set, Pmode, HARD_FRAME_POINTER_REGNUM);
    n = spill_pseudos (set, spilled_pseudos);
-  for (ep = reg_eliminate; ep < &reg_eliminate[NUM_ELIMINABLE_REGS]; ep++)
-    if (ep->from == FRAME_POINTER_REGNUM && ep->to == STACK_POINTER_REGNUM)
-      setup_can_eliminate (ep, false);
+  if (!ep)
+    for (ep = reg_eliminate; ep < &reg_eliminate[NUM_ELIMINABLE_REGS]; ep++)
+      if (ep->from == FRAME_POINTER_REGNUM && ep->to == STACK_POINTER_REGNUM)
+       setup_can_eliminate (ep, false);
    return n;
  }