[gccsdk] New VFP/NEON capable GCC release?

Lee Noar leenoar at sky.com
Thu Aug 13 10:10:30 PDT 2015


On 11/08/15 20:00, Chris Gransden wrote:
> In article <55C88DC7.6030105 at sky.com>,
>     Lee Noar <leenoar at sky.com> wrote:
>> On 08/08/15 20:03, Chris Gransden wrote:
>>> Also if gcc is run with an insufficient wimpslot it gives the error
>>> 'Application requires a minimum of 6000K to run.' Where as to compile
>>> anything it needs around 15000k.
>
>> I think the 6000k requirement came from the driver program. You can do
>> gcc -v in 6000k, it's when the compiler is called that 15000k is required.
>> I can change the test and error message from 6000k to 15000k, but it
>> will then complain about needing 15000k just to do gcc -v. OTOH, that's
>> not something you need to do very often if at all so it may be worth
>> changing to make the memory requirement clearer.
>
>> So do I make the change or leave it be?
>
> Setting the wimpslot to 6000k the next error message is,
>
> Wimpslot not big enough to run ELF program.
>
> Setting it to 14500k gives,
>
> Insufficient memory for application
>
> Is there any way for the ELF loader to automatically allocate the wimpslot
> in the same way that current versions of RISC OS do for AIF.

Well, maybe. Perhaps in the next version - I don't want to hold up this
release any longer. The problem is that library dependencies all require
their own slice of the wimpslot and until they're all loaded, you don't
know what it all adds up to. There's also some data stored at the top of
the wimpslot that would have to be shifted up as it was extended. I
can't remember if that data may be sensitive to being moved. I will look
into it though.

> I can't see it would do any harm setting it to 15000k. At least it will
> always run.

Well, it wont run unless you set the wimpslot manually to 15000k, but it
will report that it needs that much.

 > g++ needs 16000k to compile.

Okay, I'll set the test and error message to 16000k.

Lee.





More information about the gcc mailing list