Author: Eugene Nalimov
Date: 14:11:07 08/18/00
Go up one level in this thread
On August 18, 2000 at 14:02:55, J. Wesley Cleveland wrote:
>On August 17, 2000 at 14:28:28, Eugene Nalimov wrote:
>
>>See "Itanium Processor Microarchitecture Reference", Order number 245473-001.
>>Also available from Intel's web site for free.
>>
>>Main problem with your solution is intermix of integer (mov, and, shift by
>>constant) and MMX (parallel add, shift by variable amount) operations.
>>Int-to-MMX latency is 3 clock cycles. MMX-to-int latency is 4 clock cycles.
>>
>Even though I don't do MMX and integer operations on the same registers
>(assuming that the result of czx is integer) ?
You use. "shr.u r1, r2" is MMX operation.
Eugene
>>Eugene
>>
>>On August 17, 2000 at 14:12:25, J. Wesley Cleveland wrote:
>>
>>>On August 16, 2000 at 13:11:04, Eugene Nalimov wrote:
>>>
>>>>On August 15, 2000 at 23:27:46, Larry Griffiths wrote:
>>>>
>>>>>I looked through some literature for the Intel Itanium processor and did not see
>>>>>any Bit Scan Forward or Bit Scan reverse instructions. Does anyone know for
>>>>>sure if the Itanium has these instructions or something like them.
>>>>>My bitboard move generation would really suck without BSF or BSR.
>>>>>
>>>>>Larry.
>>>>
>>>>No. There are no such instructions. You have to use ifs, shifts, and lookup
>>>>table. Here is the example -- FirstOne()/LastOne() that I wrote for IA-64
>>>>version of Crafty (it is not in the official build yet):
>>>>
>>>I think you can do a lot better using the multimedia instructions (this is just
>>>from reading the docs).
>>>
>>>int FirstOne(BITBOARD arg1) {
>>> __asm {
>>> ld r1, arg1
>>> ld r2, 0fefefefefefefefeh ; probably is address of constant
>>> ld r3, 0101010101010101h
>>> padd1.uuu r4=r1,r2 ; each byte in r4 has fe if byte in arg1 was 0, else ff
>>> padd r4=r4,r3 ; now each non-zero byte in arg1 is zero
>>> czx1.l r2=r4 ; r2 has byte with first bit set or 8 if all 0
>>> shl r2=r2,3
>>> shr.u r1=r1,r2 ;first bit is now in low byte (or 0 if none set)
>>> and r1, 0ffh
>>> ld r0, btab[r1] ;not sure how to do this
>>> add r0=r0,r2
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.