Author: Matt Taylor
Date: 12:44:53 12/05/02
Go up one level in this thread
On December 05, 2002 at 15:41:02, Gerd Isenberg wrote:
>On December 05, 2002 at 12:48:02, Dieter Buerssner wrote:
>
>>On December 05, 2002 at 11:25:27, Gerd Isenberg wrote:
>>
>>>Most (All?) C-compilers have no problem with unary minus and constants:
>>>
>>>unsigned int A = -CONST;
>>
>>Assume 32 bit integers, and const is -2^31.
>>I believe, the above is undefined behaviour. In this case a cast will make it
>>defined behaviour:
>>
>> unsigned int A = -(unsigned)CONST.
>>
>
>Hi Dieter,
>
>I guess it's null, because 2^31 is INT_MIN. I had small constants in mind.
>So -1 is very common to to set all bits in an unsigned variable, don't care
>about the word size.
>
>unsigned int A = -1;
>if ( popCount(A) == 64)
>{
> // ah, nice native 64 bit ints
>}
>
>Where i have some problems with sometimes, is the implicite type of a direct
>constant expressions, calculated by the compiler, specially with propritary
>64 bit integers types.
>
>Eg.: MSC6.0
>
>unsigned __int64 FFFF = -1; // that' fine, all 64 bits set
>unsigned __int64 H8BB = 1<<63; // oups zero
>
>produces an internal unsigned int overflow and assigns zero( without any
>warning, even with warning level 4). Why isn't the compiler able to cast the 1
>implicitly to the 64bit type?
>
>unsigned __int64 H8BB = (unsigned __int64)1<<63; // that's ok
>unsigned __int64 H8BB = (-1)<<63; // also wrong
>
>
>>Totally unrelated to chess. If you want to write a function to convert int to
>>ASCII (without using sprintf) the typical way may be:
>>
>> int toconvert;
>> unsigned uval;
>> if (toconvert < 0)
>> uval = -toconvert; /* Oops, may not work for toconvert = INT_MIN */
>> /* and go on to convert the unsigned value */
>>
>>>Because -CONST is a kind synonym for the compiler which means implicitly
>>> (2**wordLengthInBits) - CONST ==> 0 - CONST
>>
>>All unsigned arithmetics is guaranteed to yield results mod 2 ^ bits.
>
>except imul and idiv ;-)
>
>>
>>Cheers,
>>Dieter
>
>See you,
>Gerd
Actually, I noticed an interesting "feature" of the VC compiler when I converted
an atoi assembly routine into C (before I realized it was just atoi). I was
using unsigned numbers, but the VC compiler generated a signed multiply. I first
thought it was a bug, but later I worked it on paper and came to the conclusion
that a signed 32x32 -> 32-bits is identical to unsigned 32x32 -> 32-bits. I
concluded that this was not really a bug, then, but a feature since the 32x32 ->
32 is probably a bit faster than a full 32x32 -> 64 on most chips.
Athlon sustains a rate of 1 32x32 -> 32 multiplies every 2 cycles, and Opteron
is supposed to do a multiply per cycle. I haven't tested the 32x32 -> 64 on my
Athlon, but I suppose it's worth looking into.
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.