https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66814
Uroš Bizjak <ubizjak at gmail dot com> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |NEW
Last reconfirmed| |2015-07-08
CC| |kyukhin at gcc dot gnu.org
Target Milestone|--- |5.2
Ever confirmed|0 |1
--- Comment #1 from Uroš Bizjak <ubizjak at gmail dot com> ---
Following peephole2 is too broad:
(define_peephole2
[(set (match_operand:SWI124 0 "nonimmediate_operand")
(not:SWI124 (match_operand:SWI124 1 "nonimmediate_operand")))]
"optimize_insn_for_speed_p ()
&& ((TARGET_NOT_UNPAIRABLE
&& (!MEM_P (operands[0])
|| !memory_displacement_operand (operands[0], <MODE>mode)))
|| (TARGET_NOT_VECTORMODE
&& long_memory_operand (operands[0], <MODE>mode)))
&& peep2_regno_dead_p (0, FLAGS_REG)"
[(parallel [(set (match_dup 0)
(xor:SWI124 (match_dup 1) (const_int -1)))
(clobber (reg:CC FLAGS_REG))])])
Needs additional GENERAL_REG_P predicates.
Probably several other peephole2 patterns need similar treatment.
CC AVX512F maintainer.
