RE: [avr-gcc-list] ATmega2560 and r30/31 save in function prologue

[Weddington, Eric]

Anatoly, Andy, can you look at this?

Stu, at minimum you can always submit a bug report at the WinAVR project
on SourceForge. I'll take care of getting it to the right place.

Eric Weddington

> -----Original Message-----
> From: 
> address@hidden 
> [mailto:address@hidden
> org] On Behalf Of Stu Bell
> Sent: Friday, May 02, 2008 4:23 PM
> To: avr-GCC
> Subject: RE: [avr-gcc-list] ATmega2560 and r30/31 save in 
> function prologue
> 
> I have chased this down.  The problem occurs on any processor (okay, I
> checked the m128 and m2560 and it occurred on both - YMMV).
> 
> If you have an ISR that calls a function, but that function is defined
> in a different compile module (file.c), the compiler will not save the
> "call-used" registers.  In my case, this means r30 and r31.
> 
> Interestingly enough, if the called function is in the same compile
> module (file.c), then the call-used registers *are* saved in the ISR!
> 
> This appears to have changed as of WinAVR 2008*.
> 
> I have attached (I hope) a zip file of my test.  The file test5.c
> contains the all-in-one compile that works.  The file test5.bad.c
> contains just the references to serial.h, which refers to 
> queue.h.  You
> need to tweak the Makefile to make the associated files.
> 
> I'm sending it here because I can't figure out to how to submit an
> AVR-GCC bug.  I can do the AVR-LIBC stuff no problem.  I'm 
> sending this
> to myself at home so I can log it this evening or tomorrow.
> 
> Caveat emptor!
> 
> Best regards, 
> 
> Stu Bell 
> DataPlay (DPHI, Inc.) 
> 
> 
> -----Original Message-----
> From: address@hidden
> [mailto:address@hidden 
> On Behalf Of
> Stu Bell
> Sent: Friday, May 02, 2008 1:24 PM
> To: avr-GCC
> Subject: [avr-gcc-list] ATmega2560 and r30/31 save in 
> function prologue
> 
> Hi all,
> 
> I'm chasing down a problem with WinAVR 20080430 and my 
> ATmega2560 code.
> In generated code comparisons of an ISR between 20071221 and 
> 20080430, I
> noticed that r30 and r31 are not saved on the stack.  Normally, this
> wouldn't be a problem as the ISR itself does not use r30/r31.
> 
> However, this ISR calls a function that does use r30/r31.  Again, it
> *should* be no problem, since the function has the responsibility to
> stor r30/r31 if it uses it....
>