Re: binutils-2.15, gas and mips

[Jie Zhang]

Khaled Daham wrote:
> When trying to use the 0f format specifier for immediates an assert is
> triggered ( tested with binutils-2.15 and 2.14 )
>
> ../configure  --prefix=/tmp/mips --target=mips3 --enable-languages=c
> ( this was just a test build, the target I am using is not in binutils
> without a patch, and no it does not touch that part of load_register, same
> assert )
>
> only .s file given as argument to as
>
> Simple test case.
>
> .ent test
> test:
>  li $2, 0f1.0
> .end test
>
>
> this results in
>
> /tmp/test.s: Assembler messages:
> /tmp/test.s:3: Internal error!
> Assertion failure in load_register at ../../gas/config/tc-mips.c line
> 3626.
> Please report this bug.
>
> generally there doesnt seem to be a case for handling O_big and
> X_add_number <= 0 in tc-mips.c wich should indicate that
> generic_floating_point_number is holding the data we want.

It cannot be said as a bug for current gas, since there is no load float 
instruction in mips ISAs, the assertion is always true.

However, you can solve this problem by two methods. You can make 
compiler transform the float number into its internal representation, 
which is indeed an integer. Or you can add a new load float pseudo 
instruction into gas and treat it specifically as you like.


Jie