[PATCH] Fix gcc.target/mips/extend-1.c

Richard Sandiford rdsandiford@googlemail.com
Tue Aug 18 22:24:00 GMT 2009


Adam Nemet <anemet@caviumnetworks.com> writes:
> Richard Sandiford writes:
>> But I don't think this
>> would work with things like "isa=!octeon -mbranch-likely", where we need
>> MIPS II.  (We don't need to worry about that for "isa(=|>=|<=)N
>> -mbranch-likely", because it's the test writer's responsibility
>> to make sure that the combinations are OK.)
>
> Wait I think I missing something logically here.  If it's the test writer's
> responsibility to write:
>
>   isa>=2 -mbranch-likely
>
> to ensure -mbranch-likely will actually work then if octeon should be omitted
> how is that he is not required to write:
>
>   isa>=2!octeon -mbranch-likely
>
> ?

Well, by that logic, anything with -mgp64 should have isa>=3!octeon
rather than plain "isa=!octeon".  But AIUI, the point of your patch
(which I agree with) is that the isa version is supposed to be optional
when specifying "!cpu".  And the current comments certainly imply that.
So you shouldn't be forced to respecify something that's obvious from
the other options simply because you're excluding a particular CPU.

OTOH, if you deliberately ask for "isa=1 -mbranch-likely", presumably
you're doing it for a reason.

Richard



More information about the Gcc-patches mailing list