RE: [avr-gcc-list] something I don't understand with stack manipulation

[Larry Barello]

The interrupts are enabled *after* the instruction following the sei(), reti
or SREG update.  The order, below, was developed to minimize interrupt
latency.

Setting up stack frames and accessing the variables is expensive on the AVR.
I guess it can't be good at everything.  Fortunately one can pass a lot of
parameters before hitting the stack.  Things that force use of stack:
automatics that need to be referenced by address (&, buffers), variable call
parameters (printf), large structures.

Other C compilers (IAR, ICC, CodeVision) use a separate stack for
parameters, thus avoiding the SREG update issue.  However, with the
exception of IAR, the other compilers generate significantly more code than
GCC - GCC is wicked good at overall optimization.

Cheers!

-----Original Message-----
From: Bernard Fouché

Hi there.

I'm (still!) looking at the asm code produced by avr-gcc 3.4.2 with -Os.
When the compiler needs to give stack space to a local variable (for
instance a 10 bytes space), it produces:

    1d6e:       cd b7           in      r28, 0x3d       ; 61
    1d70:       de b7           in      r29, 0x3e       ; 62
    1d72:       2a 97           sbiw    r28, 0x0a       ; 10
    1d74:       0f b6           in      r0, 0x3f        ; 63
    1d76:       f8 94           cli
    1d78:       de bf           out     0x3e, r29       ; 62
    1d7a:       0f be           out     0x3f, r0        ; 63