egcs-1.2 FCOMI stuff

[H.J. Lu]

> 
> 
>   In message < m10TGcb-000ErMC@ocean.lucon.org >you write:
>   > You have pointed out that GET_CODE (next) == JUMP_INSN is redundant.
>   > We have to check
>   > 
>   >    (GET_CODE (PATTERN (next)) == SET
>   >     && SET_DEST (PATTERN (next)) == pc_rtx
>   >     && GET_CODE (SET_SRC (PATTERN (next))) == IF_THEN_ELSE)
>   > 
>   > I believe they will also satisfy GET_CODE (next) == JUMP_INSN. There
>   > is no need to check it.
> This is odd.  In some places you added redundant checks with your original
> patch (the output_fp_conditional_move abort you added).  In this place you're
> removing a redudant check.

They are quite different. Here is

If B is ture, A must be true. So we don't need to check if A is true
when we have to check if B is true.

In output_fp_conditional_move, it is

A should be true. There is no B involved. It is not a bad idea to
check if A is really true.



-- 
H.J. Lu (hjl@gnu.org)