[Bug 1001607] Cortex-M4F architectural Floating Point Support

[bugzilla-daemon@bugs.ecos.sourceware.org]

Please do not reply to this email, use the link below.

http://bugs.ecos.sourceware.org/show_bug.cgi?id=1001607

--- Comment #62 from Ilija Kocho <ilijak@siva.com.mk> ---
Hi Jifl

(In reply to comment #57)

[snip]

I applied CYGTST_KERNEL_SKIP_MULTI_THREAD_FP_TEST and performed fixes to the
tests. In addition I renamed thread_switch_fpu.cxx to fpint_thread_switch.cxx -
sounds better to me and now all tests have same prefix.

> > > (In reply to comment #50)
[snip]

> > Provided that that LAZY uses FPU enabled/disbled state in order to
> > distinguish between _FP_ and _INT_ threads, suppose that _INT_ thread is
> > interrupted by _FP_ ISR. Then Usage Fault VSR will enable FPU and FPU will
> > remain enabled after ISR returns [in thread context], effectively converting
> > the tread to _FP_. Gradually, this ISR "_FP_ missioner" ISR may convert all
> > threads to _FP_ so we're not lazy any more.
> 
> Yes you're right of course. But I can't help feel it wouldn't be difficult
> to fix this in the exception and interrupt vsrs - e.g. set
> HAL_SAVEDREGISTERS_WITH_FPU as the saved register type if the FPU enabled
> bit is set in FPU_CPACR on entry to the exception or interrupt VSR, and then
> ensure the FPU is enabled/disabled accordingly before exit. I just get the
> feeling this should be able to be solved with little overhead.

I think that for interrupts it would be a waste.
However this led me to idea to add check ASPEN bit of FPCCR to the
GDB_STUB_SAVEDREG_FPU_EXCEPTION_SET() so now RedBoot can determine if autosave
is enabled or disabled in runtime. Single RedBoot image to debug LAZY, ALL and
NONE. Implemented.

[snip]

> 
> Yes! In fact I have no problems with the FP patch 130210 and as far as I'm
> concerned that patch can be committed. However I do have an issue with your
> changes for the code build flag which you added in comment #56, which is
> that this will also prevent use of the DSP instructions by GCC I believe?

There's no problem with using DSP instructions. CDL will allow you to set
-mcpu=cortex-m4 either manually or provide as default setting by platform (we
can do it when we adopt new compiler). As a side effect you can set cortex-m4
alone by sequence of enable/disable FPU.

> 
> Also of course as mentioned above, we still need to work out the last few
> niggles with those kernel tests.

I hope the tests are clean now. Some of the "features" were inherited from
original files. My next window for check-in is the upcoming weekend.

Ilija

-- 
You are receiving this mail because:
You are on the CC list for the bug.