Author: Eugene Nalimov
Date: 21:47:21 02/12/02
Go up one level in this thread
For VC it doesn't matter what you will write --
b = ((i & 0xFFFF0000) != 0);
or
if ((i & 0xFFFF0000) != 0)
b = 1;
else
b = 0;
Also it would generate not 'setne', but instruction sequence that is more
efficient on some x86 CPUs due to lack of partial registers stalls:
; Listing generated by Microsoft (R) Optimizing Compiler Version 13.00.9351
TITLE t.c
.386P
include listing.inc
if @Version gt 510
.model FLAT
else
_TEXT SEGMENT PARA USE32 PUBLIC 'CODE'
_TEXT ENDS
_DATA SEGMENT DWORD USE32 PUBLIC 'DATA'
_DATA ENDS
CONST SEGMENT DWORD USE32 PUBLIC 'CONST'
CONST ENDS
_BSS SEGMENT DWORD USE32 PUBLIC 'BSS'
_BSS ENDS
$$SYMBOLS SEGMENT BYTE USE32 'DEBSYM'
$$SYMBOLS ENDS
_TLS SEGMENT DWORD USE32 PUBLIC 'TLS'
_TLS ENDS
FLAT GROUP _DATA, CONST, _BSS
ASSUME CS: FLAT, DS: FLAT, SS: FLAT
endif
INCLUDELIB LIBC
INCLUDELIB OLDNAMES
PUBLIC _Lookup8
EXTRN _tab:BYTE
; Function compile flags: /Ogty
_TEXT SEGMENT
_i$ = 8
_Lookup8 PROC NEAR
; File c:\temp\t.c
; Line 7
mov edx, DWORD PTR _i$[esp-4]
mov eax, edx
and eax, -65536 ; ffff0000H
neg eax
sbb eax, eax
neg eax
; Line 8
shl eax, 4
push esi
mov esi, eax
; Line 9
mov ecx, eax
shr edx, cl
; Line 10
mov eax, edx
and eax, 65280 ; 0000ff00H
neg eax
sbb eax, eax
neg eax
; Line 12
lea ecx, DWORD PTR [eax*8]
shr edx, cl
; Line 13
movzx eax, BYTE PTR _tab[edx]
add eax, ecx
add eax, esi
pop esi
; Line 14
ret 0
_Lookup8 ENDP
_TEXT ENDS
Eugene
On February 11, 2002 at 18:49:49, Angrim wrote:
>On February 11, 2002 at 13:03:57, Eugene Nalimov wrote:
><snip>
>>(3) They did not try to remove poorly predicted branch from their implementation
>>of the LOOKUP algorithm. For example, on x86 there is no branches in the
>>following variant:
>>
>> extern unsigned char tab[256];
>>
>> unsigned Lookup8 (unsigned i)
>> {
>> unsigned b, n;
>>
>> b = ((i & 0xFFFF0000) != 0);
>> n = b * 16;
>> i >>= b * 16;
>> b = ((i & 0x0000FF00) != 0);
>> n += b * 8;
>> i >>= b * 8;
>> return n + tab[i];
>> }
>>
>>Eugene
>
>Cool, your posts are a nice source to learn more about program
>optimization. I had thought that code like
> b= (i!=0);
>would be the same as
> if(i!=0) b=1; else b=0;
>but now I know that the compiler(even gcc) is smart enough to use
>"test" and "setne" instructions to avoid branches here.
>
>Now to glance through my engine a bit and see if I can find some
>places that I can use this information.
>
>Angrim
This page took 0.01 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.