Author: Gerd Isenberg
Date: 11:06:15 09/02/02
Go up one level in this thread
On September 02, 2002 at 13:04:45, Robert Hyatt wrote:
>On September 02, 2002 at 08:51:21, Gerd Isenberg wrote:
>
>>On September 01, 2002 at 23:33:26, Robert Hyatt wrote:
>>
>>>On September 01, 2002 at 20:46:55, Pham Hong Nguyen wrote:
>>>
>>>>The following article and discussion may be useful for you!
>>>>
>>>>http://www.codeguru.com/cpp_mfc/switch.html
>>>
>>>
>>>That is basically wrong.
>>>
>>>Compile a simple piece of code with gcc, using -O and you will see
>>>why.
>>>
>>>gcc creates a jump table of addresses for each case. It computes the
>>>entry to load and jumps to that. No compares and branches. Someone
>>>doesn't know how compilers do things. If you have really oddball case
>>>value, it might have to resort to that. But for the kinds of values
>>>we use in chess, no way...
>>
>>In MSC there are at least one or two bound check compares before indexing the
>>table (even if there is no explicite default case). What about using a table of
>>function pointers, even with C++ member functions, specially if your case-code
>>already calls subroutines?
>>
>>class CNode {
>> ...
>> typedef BitBoard (CNode::*PTR_GETATTACK)(unsigned int sq) const;
>> static PTR_GETATTACK m_scPieceAtta[14]; // for each kind of piece
>>
>> __forceinline BitBoard GetAttack(unsigned int sq, unsigned int piece) const {
>> return (this->*m_scPieceAtta[piece])(sq); }
>> ...
>>};
>>
>>CNode::PTR_GETATTACK CNode::m_scPieceAtta[14] =
>>{
>> AssertAttack, // no valid piece code
>> AssertAttack, // no valid piece code
>> WPawnAttacks,
>> BPawnAttacks,
>> ....
>>};
>
>
>The drawback is that function calls are not free. To lose the upper/lower
>bound range checking, you take a hit on procedure call overhead. I have not
>looked, but many compilers have a way to turn off bounds checking in a switch/
>case/etc statement, if you don't mind taking the risk that the code will blow
>up if a out-of-bound case value is hit...
Agreed, but i use it to replace a function (none inlined, because quite huge
and/or referenced too often) containing only one switch/case statement as main
structure. So with the inlined dispatch function (only a few bytes) there is no
additional call/ret overhead.
Regards,
Gerd
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.