On Dec  9, 2021, Jeff Law <jeffreya...@gmail.com> wrote:

>> I found a similar pattern of issuing clobbers for multi-word moves, but
>> not when reload_in_progress, in expr.c:emit_move_complex_parts.  I don't
>> have a testcase, but I'm tempted to propose '!lra_in_progress &&' for it
>> as well.  Can you think of any reason not to?

> The only reason I can think of is we're in stage3 :-)  It'd be a lot
> easier to green light that if we could trigger an issue.

I have not found the cycles to try to construct a testcase to trigger
the issue, but before moving on, I have regstrapped this on
x86_64-linux-gnu, so, at least for now, I propose it for the next
release cycle.  Ok to install then?


[PR103302] skip multi-part clobber during lra for complex parts too

From: Alexandre Oliva <ol...@adacore.com>

As with the earlier patch, avoid emitting clobbers that we used to
avoid during reload also during LRA, now when moving complex
multi-part values.  We don't have a testcase for this one.


for  gcc/ChangeLog

        PR target/103302
        * expr.c (emit_move_complex_parts): Skip clobbers during lra.
---
 gcc/expr.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/gcc/expr.c b/gcc/expr.c
index 0365625e7b835..30d1735ec29ce 100644
--- a/gcc/expr.c
+++ b/gcc/expr.c
@@ -3736,7 +3736,7 @@ emit_move_complex_parts (rtx x, rtx y)
   /* Show the output dies here.  This is necessary for SUBREGs
      of pseudos since we cannot track their lifetimes correctly;
      hard regs shouldn't appear here except as return values.  */
-  if (!reload_completed && !reload_in_progress
+  if (!reload_completed && !reload_in_progress && !lra_in_progress
       && REG_P (x) && !reg_overlap_mentioned_p (x, y))
     emit_clobber (x);
 


-- 
Alexandre Oliva, happy hacker                https://FSFLA.org/blogs/lxo/
   Free Software Activist                       GNU Toolchain Engineer
Disinformation flourishes because many people care deeply about injustice
but very few check the facts.  Ask me about <https://stallmansupport.org>

Reply via email to