Re: [PATCH 1/2] lzma support: decompression lib, initrd support

[Paulo Marques <pmarques@grupopie.com>]

Andrew Morton wrote:
> Christian Leber <christian@leber.de> wrote:
> [...]
>>+	for (pb = 0; prop0 >= (9 * 5); pb++, prop0 -= (9 * 5));
>>+	for (lp = 0; prop0 >= 9; lp++, prop0 -= 9);
> 
> Put the ";" on a line of its own.
> 
> I'd have thought the above could be done arithmetically?

I just tried a small test program to see the speed/code size difference 
to this code, which is the arithmetic equivalent:

   pb = prop0 / (9 * 5);
   prop0 %= (9 * 5);
   lp = prop0 / 9;
   prop0 %= 9;

This code runs a lot faster than the original. This is not very 
important since it runs only once AFAICT.

As for the code size, it is smaller if compiled with -Os, but larger 
when compiled with -O2 or -O3.

When compiled with -Os, gcc uses the idiv instruction and it even uses 
its reminder so that it only does 2 idiv instructions to do the 4 
operations above.

With -O2 or -O3, it does a hard to follow division "by hand" using 
several instructions, rendering the code about 2.5x larger (but 
amazingly a lot faster).

The tests were done with gcc 3.3.2.

-- 
Paulo Marques - www.grupopie.com

An expert is a person who has made all the mistakes that can be
made in a very narrow field.
Niels Bohr (1885 - 1962)