https://gcc.gnu.org/bugzilla/show_bug.cgi?id=105209
Richard Biener <rguenth at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |NEW
Known to fail| |10.3.1, 12.0
Last reconfirmed| |2022-04-11
Ever confirmed|0 |1
--- Comment #1 from Richard Biener <rguenth at gcc dot gnu.org> ---
Confirmed.
(gdb) p debug_rtx (out_insn)
(insn 37 36 38 5 (parallel [
(set (reg:DI 93)
(plus:DI (reg:DI 71 [ _3 ])
(reg:DI 82 [ _25 ])))
(trap_if (ne (plus:TI (sign_extend:TI (reg:DI 71 [ _3 ]))
(sign_extend:TI (reg:DI 82 [ _25 ])))
(sign_extend:TI (plus:DI (reg:DI 71 [ _3 ])
(reg:DI 82 [ _25 ]))))
(const_int 0 [0]))
]) "t.c":21:24 13 {addvdi3}
(expr_list:REG_DEAD (reg:DI 82 [ _25 ])
(expr_list:REG_DEAD (reg:DI 71 [ _3 ])
(nil))))
looks like store_data_bypass_p_1 does not handle trap_if in PATTERN.
#1 0x000000000124c228 in store_data_bypass_p_1 (out_insn=0x7ffff6522980,
in_set=0x7ffff663d948) at /home/rguenther/src/trunk/gcc/recog.cc:4233
#2 0x000000000124c29e in store_data_bypass_p (out_insn=0x7ffff6522980,
in_insn=0x7ffff65229c0) at /home/rguenther/src/trunk/gcc/recog.cc:4251
#3 0x0000000001ddc520 in internal_insn_latency (insn_code=9, insn2_code=2,
insn=0x7ffff6522980, insn2=0x7ffff65229c0) at insn-automata.cc:3171
#4 0x0000000001ddc9a4 in insn_latency (insn=0x7ffff6522980,
insn2=0x7ffff65229c0) at insn-automata.cc:3382
#5 0x0000000001f5aec9 in dep_cost_1 (link=0x2d8af28, dw=0)
at /home/rguenther/src/trunk/gcc/haifa-sched.cc:1488
#6 0x0000000001f5af4e in dep_cost (link=0x2d8af28)
at /home/rguenther/src/trunk/gcc/haifa-sched.cc:1510
#7 0x0000000001f5b46a in priority (insn=0x7ffff6522980, force_recompute=false)
at /home/rguenther/src/trunk/gcc/haifa-sched.cc:1661
#8 0x0000000001f6871d in set_priorities (head=0x7ffff6522900,
tail=0x7ffff65223c0) at /home/rguenther/src/trunk/gcc/haifa-sched.cc:7166
#9 0x00000000012b9868 in compute_priorities ()
at /home/rguenther/src/trunk/gcc/sched-rgn.cc:3025
note that this is likely an error in the alpha machine description which
uses store_data_bypass_p in the ev4 pipeline description and wrongly(?)
tags addvsi3 with such part of the pipeline.
