https://gcc.gnu.org/bugzilla/show_bug.cgi?id=112526
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Last reconfirmed| |2023-11-15
Ever confirmed|0 |1
Assignee|unassigned at gcc dot gnu.org |jakub at gcc dot gnu.org
Status|UNCONFIRMED |ASSIGNED
--- Comment #10 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
I think
--- gcc/config/i386/i386.md.jj 2023-11-14 21:38:38.667046713 +0100
+++ gcc/config/i386/i386.md 2023-11-15 16:06:30.638353592 +0100
@@ -9918,7 +9918,10 @@
&& REGNO (operands[0]) != REGNO (operands[3])
&& (REGNO (operands[0]) == REGNO (operands[4])
|| REGNO (operands[0]) == REGNO (operands[5])
- || peep2_reg_dead_p (3, operands[0]))"
+ || peep2_reg_dead_p (3, operands[0]))
+ && (REGNO (operands[2]) == REGNO (operands[4])
+ || REGNO (operands[2]) == REGNO (operands[5])
+ || peep2_reg_dead_p (3, operands[2]))"
[(set (match_dup 2) (match_dup 1))
(parallel [(set (match_dup 4)
(mult:DWIH (match_dup 2) (match_dup 3)))
ought to fix this. Unlike the normal imul case where the destination is always
%[re]dx:%[re]ax and operands[2] must be %[re]ax, so it is naturally dead at the
end and all we need is verify operands[0], in the mulx case the destination is
arbitrary pair of registers, so we need to check that both operands[0] and
operands[2] (the latter must be %[re]dx) are dead so that we can optimize,
because we stop initializing operands[0] altogether and set operands[2] to the
immediate rather than the previous value.
The question is if I manage to create a small reproducer or not for the
testsuite.
