[Bug target/89954] missed optimization for signed extension for x86-64

[ubizjak at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89954

--- Comment #10 from Uroš Bizjak <ubizjak at gmail dot com> ---
(In reply to Hongtao.liu from comment #9)
> (In reply to Hongtao.liu from comment #8)
> > (In reply to Uroš Bizjak from comment #7)
> > > Created attachment 51496 [details]
> > > Prototype patch
> > 
> > +;; convert (sign_extend:WIDE (any_logic:NARROW (memory, immediate)))
> > +;; to (any_logic:WIDE (sign_extend (memory)), (sign_extend (immediate))).
> > +;; This eliminates sign extension after logic operation.
> > +
> > +(define_split
> > +  [(set (match_operand:SWI248 0 "register_operand")
> > +	(sign_extend:SWI248
> > +	  (any_logic:QI (match_operand:QI 1 "memory_operand")
> > +			(match_operand:QI 2 "const_int_operand"))))]
> > +  ""
> > +  [(set (match_dup 0) (any_logic:SWI248 (match_dup 0) (match_dup 2)))]
> > +  "convert_move (operands[0], operands[1], false);")
> > 
> > Shouldn't we make sure (any_logic op1 const_int) would change sign bit of
> > op1?
> Typo, would not change sign bit of op1.

It doesn't matter, since all constants are implicitly sign-extended by the
insn.

For example, convert sext:HI (xor:QI (op1, -1)) to xor:HI (sext:HI (op1), -1):

(a = op1, b = -1)

sext:HI (xor:QI (op1, -1)):

a          0xxxxxxx
b          1xxxxxxx
r 11111111 1xxxxxxx

a          1xxxxxxx
b          1xxxxxxx
r 00000000 0xxxxxxx

xor:HI (sext:HI (op1), -1):

a 00000000 0xxxxxxx
b 11111111 1xxxxxxx
r 11111111 1xxxxxxx

a 11111111 1xxxxxxx
b 11111111 1xxxxxxx
r 00000000 0xxxxxxx