Re: genflags and genemit disagree

[hjl at lucon dot org (H.J. Lu)]

> 
> iI
> Hi,
> while looking for ways to add prototypes to i386.h (which I'll write about
> in another message) I stumbled over an odity regarding gen_call_pop() and
> gen_call_value_pop(). The bug is that either genflags produces a wrong
> prototype or genemit the wrong function.
> 
> As I don't understand rtl yet, I thought I'ld ask here for somebody to
> explain.
> 
> Here is the relevant part form i386.md:
> 
> (define_expand "call_pop"
>  [(parallel [(call (match_operand:QI 0 "indirect_operand" "")
>                     (match_operand:SI 1 "general_operand" ""))
>               (set (reg:SI 7)
>                    (plus:SI (reg:SI 7)
>                             (match_operand:SI 3 "immediate_operand"
> "")))])]
> How many parameters do result from this description ? I don't have the
> generated code here, but if I remember correctly, genflags produces a
> prototype with three rtx as parameter, while genemit produces a function
> with four rtx.
> 
> As I suspect genflags of being the culprit, could somebody please have
> a look at num_operands() in genemit.c and tell me where it could fail with
> a rtl as above ?
> 

Just do a

# cd gcc
# grep MD_CALL_PROTOTYPES *

and check out all references to MD_CALL_PROTOTYPES.



H.J.