On Wed, Dec 14, 2016 at 4:33 PM, Jakub Jelinek <ja...@redhat.com> wrote:
> On Wed, Dec 14, 2016 at 04:30:39PM +0100, Uros Bizjak wrote:
>> Apparently, Qt source uses these two builtins. As the removal brings
>> much pain, I'd like to commit the following patch to restore
>> __builtin_clzs and __builtin_ctzs on gcc-6 branch.
>>
>> 2016-12-14 Uros Bizjak <ubiz...@gmail.com>
>>
>> PR target/59874
>> * config/i386/i386.c (enum ix86_builtins): Add IX86_BUILTIN_CLZS
>> and IX86_BUILTIN_CTZS.
>> (bdesc_args): Add __builtin_clzs and __builtin_ctzs.
>> (ix86_fold_builtin): Handle IX86_BUILTIN_CTZS and IX86_BUILTIN_CLZS.
>>
>> Bootstrapped and regression tested on x86_64-linux-gnu.
>
> Ok for 6. For 7 we ask them to change it to the new builtins?
I'm investigating automatic conversion to HImode insn, something along
the lines:
+(define_insn_and_split "*ctzhi2"
+ [(set (match_operand:SI 0 "register_operand")
+ (ctz:SI
+ (zero_extend:SI (match_operand:HI 1 "nonimmediate_operand"))))
+ (clobber (reg:CC FLAGS_REG))]
+ "TARGET_BMI
+ && can_create_pseudo_p ()"
+ "#"
+ "&& 1"
+ [(const_int 0)]
+{
+ rtx tmp = gen_reg_rtx (HImode);
+
+ emit_insn (gen_bmi_tzcnt_hi (tmp, operands[1]));
+ emit_insn (gen_zero_extendhisi2 (operands[0], tmp));
+ DONE;
+})
But it looks that __builtin_clzs and __builtin_ctzs will also have to stay.
Uros.
