https://gcc.gnu.org/bugzilla/show_bug.cgi?id=126299
Mikael Pettersson <mikpelinux at gmail dot com> changed:
What |Removed |Added
----------------------------------------------------------------------------
CC| |mikpelinux at gmail dot com
--- Comment #2 from Mikael Pettersson <mikpelinux at gmail dot com> ---
Created attachment 65100
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=65100&action=edit
selfcontained executable reproducer for f1
This reproduces the wrong-code with 2.11BSD libc and the apout emulator.
As an experiment I excluded the multi-word modes from the mode iterators in
"cbranch<mode>4", so only QI and HI remained for the backend, and that fixed
this test case. The generated code for f1 went from the bogus
_f1:
tst 02(sp)
bne L_3
tst 04(sp)
L_3:
bge L_1
mov $01,_x
L_1:
rts pc
to the better-looking
_f1:
tst 02(sp)
bge L_1
mov $01,_x
L_1:
rts pc
--- ./gcc/config/pdp11/pdp11.md.~1~ 2026-07-19 00:32:07.000000000 +0200
+++ ./gcc/config/pdp11/pdp11.md 2026-07-20 14:37:09.743045489 +0200
@@ -437,8 +437,8 @@ (define_insn_and_split "cbranchdf4"
(define_insn_and_split "cbranch<mode>4"
[(set (pc)
(if_then_else (match_operator 0 "ordered_comparison_operator"
- [(match_operand:QHSDint 1 "general_operand" "g")
- (match_operand:QHSDint 2 "general_operand" "g")])
+ [(match_operand:PDPint 1 "general_operand" "g")
+ (match_operand:PDPint 2 "general_operand" "g")])
(label_ref (match_operand 3 "" ""))
(pc)))]
""
Not tested beyond this test case.
