Patch: IA64 inline integer division

Steve Ellcey sje@cup.hp.com
Wed Mar 10 20:53:00 GMT 2004


> Something I thought might work well (in that it doesn't completely
> obfuscate the rtl, such that optimizers are happy), is to create an
> HFmode floating point format which consists of just exponent and sign.
> 
> You then need two more things, to make compress_float_constant dtrt:
> 
>   * Constants of this mode should be LEGITIMATE_CONSTANT_P.
>   * Create extendhf[sdx]f2 expanders that accept these constants in
>     operand 1.  Internally, the expander can force the constant into
>     a register, but can't require that the caller do so.
> 
> 
> r~

I believe I understand this idea in theory but I am having some problems
figuring out the specifics of how I would implement this.  Could you
expand on your idea a little?  For a start I don't understand your
comment about only needing an exponent and a sign.  I am also trying to
figure out what the extendhf[sdx]f2 expanders would look like.

For extendsfdf2 we have:

(define_insn "extendsfdf2"
  [(set (match_operand:DF 0 "fr_register_operand" "=f")
        (float_extend:DF (match_operand:SF 1 "fr_register_operand" "f")))]
  ""
  "fnorm.d %0 = %1"
  [(set_attr "itanium_class" "fmac")])


but for an extendhfdf2, if I want to go from a constant into a
fr_register_operand, I need two instructions.  One to put the constant
into a gr_register_operand and one to move the gr_register_operand into
an fr_register_operand.  So do I have to make extendhfdf2 a
define_expand instead of a define_insn?

Steve Ellcey
sje@cup.hp.com



More information about the Gcc-patches mailing list