Re: [sdcc-devel] Commit: better static array intialization for DS390

[Michael H. <mic...@ju...>]

I turn it on for the z80 port and added a RLE based init scheme - works
pretty well.

Two bugs:
1. Negative values break the run and cause this to fall back to the old
method.
2. The old method adds an offset of one to all the addresses, such that:

static char fish[] = {
  1, 2, 3
}

generates
   ld _fish+1,1
   ld _fish+2,2
   ld _fish+3,3

-- Michael


On Fri, 24 Aug 2001, Kevin Vigor wrote:

> I just comitted a change to improve static array initialization
> (currently only for the DS390 port).
>
> If a port has the arrayInitializerSuppported bool in the port structure
> set, instead of breaking the array initialization down into a series of
> pointer sets, it instead creates a new iCode, ARRAYINIT, which takes one
> operand (the array to initialize) and a list of values to initialize the
> array with. This allows the port specific code generator to handle this
> case, and hopefully generate better code.
>
> See ds390/gen.c:genArrayInit for the '390 implementation. The 8051
> version should be similar, but can't make the assumption that the array
> in in XDATA space.
>
> The generated code looks much better (and is generated much faster for
> large static arrays), but I haven't actually tested it on real hardware
> yet. If problems occur, please change the arrayInitializerSuppported
> value in the DS390 port structure (ds390/main.c) to FALSE, which should
> fix the problem, and let me know.
>
>         Peace,
>                 Kevin
>
> _______________________________________________
> sdcc-devel mailing list
> sdc...@li...
> http://lists.sourceforge.net/lists/listinfo/sdcc-devel
>