Author: Koundinya Veluri
Date: 03:26:10 01/15/02
Go up one level in this thread
On January 14, 2002 at 11:27:00, Severi Salminen wrote:
>
>>I'm using this for LastBit() for a8 = 0, b8 = 1, h1 = 63 orientation:
>>
>> bsr eax, [bb + 4]
>> jz L10
>> add eax, 32
>> jmp L11
>>L10:
>> bsr eax, [bb]
>>L11:
>>
>>This should work, but I'm relatively new to assembly and I wonder if this can
>>get any better...?
>
>Well, you could remove the other jmp and add one "sub eax,32". It would have
>less branches but not sure if it was faster. Try it:
>
>bsr eax,bb+4
>jnz exit
>bsr eax,bb
>sub eax,32
>exit:
>add eax,32
>
>...or something. Your orientation makes my head spin so double check the above.
Interesting idea, but the first one seems to be noticably (38%) faster on my
computer. I guess an unconditional jump is faster than an extra sub instruction.
There's something else strange that didn't work as I expected. I tried the
FirstBit() with the bsr/add/bsr method and another method with a conditional
jump (which does not make the assumption that zero means register is unchanged),
and the one with the conditional jump was much faster. Here's the code I used to
test it and the output I get after that. Is there something wrong with this?
#include <iostream>
using namespace std;
#include <windows.h>
#define ITERATIONS (1000000000)
inline const int FirstBit1(const unsigned __int64 bb);
inline const int FirstBit2(const unsigned __int64 bb);
inline const int LastBit1(const unsigned __int64 bb);
inline const int LastBit2(const unsigned __int64 bb);
int main()
{
unsigned int a, b, t1, t2, t3, t4;
cout << "Press enter to start...";
cin.ignore();
t1 = GetTickCount();
for(a = 0; a < ITERATIONS; ++a)
b = FirstBit1(1);
t1 = GetTickCount() - t1;
t2 = GetTickCount();
for(a = 0; a < ITERATIONS; ++a)
b = FirstBit2(1);
t2 = GetTickCount() - t2;
t3 = GetTickCount();
for(a = 0; a < ITERATIONS; ++a)
b = LastBit1(1);
t3 = GetTickCount() - t3;
t4 = GetTickCount();
for(a = 0; a < ITERATIONS; ++a)
b = LastBit2(1);
t4 = GetTickCount() - t4;
cout << endl << t1 << endl << t2 << "\n\n" << t3 << endl << t4 << "\n\n";
return 0;
}
const int FirstBit1(const unsigned __int64 bb)
{
__asm
{
bsf eax, [bb]
jnz L10
bsf eax, [bb + 4]
add eax, 32
L10:
}
}
const int FirstBit2(const unsigned __int64 bb)
{
__asm
{
bsf eax, [bb + 4]
add eax, 32
bsf eax, [bb]
}
}
const int LastBit1(const unsigned __int64 bb)
{
__asm
{
bsr eax, [bb + 4]
jz L10
add eax, 32
jmp L11
L10:
bsr eax, [bb]
L11:
}
}
const int LastBit2(const unsigned __int64 bb)
{
__asm
{
bsr eax, [bb + 4]
jnz L10
bsr eax, [bb]
sub eax, 32
L10:
add eax, 32
}
}
Output on my computer:
Press enter to start...
2383
5909
4747
6529
--------
Regards,
Koundinya
>
>Severi
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.