Re: [avr-gcc-list] compiler improvments

[HutchinsonAndy]

One big improvement in the handling of structures is to tell gcc how to move DI 
operands. ie movdi in avr.md file

Then the "silly example" ends up as something more palatable.

I have rewitten the various move routines to handle all modes but its a bit 
involved to post here.


    .file   "foo.c"
    .arch avr2
__SREG__ = 0x3f
__SP_H__ = 0x3e
__SP_L__ = 0x3d
__tmp_reg__ = 0
__zero_reg__ = 1
    .global __do_copy_data
    .global __do_clear_bss
    .text
.global foo
    .type   foo, @function
foo:
/* prologue: frame size=6 */
    push r28
    push r29
    in r28,__SP_L__
    in r29,__SP_H__
    sbiw r28,6
    in __tmp_reg__,__SREG__
    cli
    out __SP_H__,r29
    out __SREG__,__tmp_reg__
    out __SP_L__,r28
/* prologue end (size=10) */
    ldi r18,lo8(0)
    ldi r19,hi8(0)
    ldi r20,hlo8(0)
    ldi r21,hhi8(0)
    ldi r19,lo8(1) ;AWHr_mr
    std Y+1,r19 ;AWHs
    std Y+2,__zero_reg__ ;AWHs
    std Y+3,__zero_reg__ ;AWHs
    std Y+4,__zero_reg__ ;AWHs
    ldi r18,lo8(2) ;AWHr_mr
    std Y+5,r18 ;AWHs
    std Y+6,__zero_reg__ ;AWHs
    ldd r18,Y+1
    ldd r19,Y+2
    ldd r20,Y+3
    ldd r21,Y+4
    ldd r22,Y+5
    ldd r23,Y+6
/* epilogue: frame size=6 */
    adiw r28,6
    in __tmp_reg__,__SREG__
    cli
    out __SP_H__,r29
    out __SREG__,__tmp_reg__
    out __SP_L__,r28
    pop r29
    pop r28
    ret
/* epilogue end (size=9) */
/* function foo size 37 (18) */
    .size   foo, .-foo
/* File "foo.c": code   37 = 0x0025 (  18), prologues  10, epilogues   9 */


This involves expanding the existing move



HutchinsonAndy wrote:

>I am looking at the structure problems. Seems there are several issues.
>
>The compiler is trying to use a DI sized variable to hold structure - 8 bytes. 
>I am not sure why but perhaps it attempst to hold all structures in registers 
>and it has not been told that a DI register is not a good idea.
>
>Size DI moves are not defined for avr target, it is using QI moves (bytes 
>moves) - which is reasonable
>
>I beleive it wants to use a temporary DI register in order to return the value 
>in the right registers. This may be because of the overlap.
>
>The extra stack variable is simple padding out 6 bytes of structure to 8. 
>Compiler did not spot that subregister was unused.
>
>The unwanted push/pop occurs somewhere in reload. R8 get assigned by global 
>register allocation. However, it is latter replaced by R13 (on my build). The 
>GCC part that marks the register as "live" does not get updated. Hence the 
>problem.
>I am trying to find out what part of code does this.
>
>I am glad to see bug report is in on this one!
>
>I may well be able to do something about the other problems. It would seem 
>that curing the "DI" allocation and movement will cause all the other issues 
>to go away.
>
>
>
>
>
>
>__________________________________________________________________
>New! Unlimited Netscape Internet Service.
>Only $9.95 a month -- Sign up today at http://isp.netscape.com/register
>Act now to get a personalized email address!
>
>Netscape. Just the Net You Need.
>

__________________________________________________________________
New! Unlimited Netscape Internet Service.
Only $9.95 a month -- Sign up today at http://isp.netscape.com/register
Act now to get a personalized email address!

Netscape. Just the Net You Need.