Author: Andrew Dados
Date: 12:32:19 01/14/02
Go up one level in this thread
On January 14, 2002 at 14:57:29, Robert Hyatt wrote:
>On January 14, 2002 at 12:11:52, James Robertson wrote:
>
>>Here are several questions I've been wondering about:
>>
>>1) How does one insert inline assembler statements into your code such that it
>>can be compiled by gcc (obviously __asm doesn't work)?
>>
>>2) How is switch() code generated? For instance, if the compiler encounters:
>>switch (var) {
>>case 0: do_stuff0(); break;
>>case 1: do_stuff1(); break;
>>case 2: do_stuff2(); break;
>>}
>>will more efficient code be generated than if I write:
>>switch (var) {
>>case 297: do_stuff287; break;
>>case 0: do_stuff0; break;
>>case 9000: do_stuff9000; break;
>>}
>
>
>Note that in the latter case, you have oddball numbers. That makes the jump
>table idea not so good and it will generally try to turn that switch into a
>series of if-then-elses instead. Which will be slower.
>
>>If so, what does the compiler do to take advantage of the fact that 0, 1, and 2
>>are consecutive numbers?
>>
>
>It generates something like this (I will use a pseudo-asm for simplicity):
>
> load var,R1
> cmp R1,0
> blt skip ; if var < 0, exit
> cmp R1,2
> bgt skip ; ditto for var > 2
> load R2,#table ; address of branch table
> mult R1,4 ; scale to word index
> add R1,R2 ; R2 = table, table+4 or table+8
> branch R2
>table: branch case0
> branch case1
> branch case2
>case0: do stuff
> branch skip
>case1: do stuff
> branch skip
>case2: do stuff
>
>skip:
>
>The above is the best way to handle a switch (C) or case (pascal)
>type structure. The problem is that if the case variable values are
>scattered around, the branch table becomes huge. The first pascal compiler
>I used didn't understand this and a case 0: case 88000: would cause it to
>blow up as it produced an _enormous_ jump table that was too large to fit in
>the machine it ran on (a xerox sigma-9 with 128KW of memory total).
>
>The alternative is to use a series of if-then-else clauses but as you add
>cases, you add branches and slow things down. Most common case _must_ appear
>first for performance. While with the jump table, the order of the cases
>is totally irrelevant.
There is another technique requiring max k jumps, where 2^k>=number of switches
(minimum of k-1 jumps). It works with no jump table, but adding new cases is
tedious.
e.g for 4 n values (1,30,120,500):
if n>30 goto upper;
if n==1 {...; goto end;}
else {...; goto end;} //n==30 here;
label upper:
if n==120 {...; goto end;}
else {...; goto end;} //n==500 here;
So you have max 2 jumps here, min one.
I doubt anyone would want to hand-code huge switch in such a way, but trick is
worth mentioning, imo.
Sure you may want to validate all cases btw...
-Andrew-
>
>
>
>
>
>
>
>
>
>
>>3) How much overhead is added to a function call from a function pointer? Maybe
>>none? How about if you index an array of function pointers?
>
>A memory reference at most...
>
>>
>>4) What are the syntax differences between console code written for Linux versus
>>Unix (I don't have easy access to a Unix machine)?
>
>
>
>absolutely nothing. Linux _is_ UNIX....
>
>
>
>>
>>5) Just out of curiousity, when is VC7 being released?
>>
>>Ok, If anyone could help me with these I'd be grateful.
>>Thanks!
>>James
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.