bitwise rotation

[Shawn B.]

Greetings,

I am simulating an assembly language bit rotation in C# and it works
wonderfully

---------
....
public uint Value;
....
public uint RotateRight(byte count) {
Value = ((Value >> count) | (Value << (32-count)));
return Value;
}
---------

as long is the value is 32-bit or 64-bit (because of a limitation of the
shift operator). I need it to work on an 8-bit or 16-bit value, as well. I
presume I'd have to have knowledge of the intended size of the value and do
the shift and move the bit and the 0 or 7, or 0 or 15 bit position and clear
the extra bits if there are any. That seems like a lot of work.

Is there a way to simulate an 8 or 16-bit rotate with a 32-bit or 64-bit
value?


Thanks,
Shawn

 

[Robert Jeppesen]

How about:

public uint RotateRight16(uint Value, byte Count)
{
return ((Value >> Count) | (Value << (16-Count)))&0xffff;
}
public uint RotateRight8(uint Value, byte Count)
{
return ((Value >> Count) | (Value << (8-Count)))&0xff;
}

public uint RotateLeft16(uint Value, byte Count)
{
return ((Value << Count) | (Value >> (16-Count)))&0xffff;
}
public uint RotateLeft8(uint Value, byte Count)
{
return ((Value << Count) | (Value >> (8-Count)))&0xff;
}

 

[Robert Jeppesen]

Oops, forgot to make sure that value is within limits:

public static uint RotateRight16(uint Value, byte Count)
{
return (((Value&0xffff) >> Count) | (Value << (16-Count)))&0xffff;
}
public static uint RotateLeft16(uint Value, byte Count)
{
return ((Value << Count) | ((Value&0xffff) >> (16-Count)))&0xffff;
}

etc for 8-bit.

 

[Shawn B.]

Hard to believe it was so simple, I tried that... except where you're
16-count is, I left it at 64 bits with the 8-bit mask, that explains why it
was wrong.

Thanks very much,
Shawn

 

[Bob Powell [MVP]]

So, you're writing an emulator for which processor??

--
Bob Powell [MVP]
C#, System.Drawing

September's edition of Well Formed is now available.
http://www.bobpowell.net/currentissue.htm

Answer those GDI+ questions with the GDI+ FAQ
http://www.bobpowell.net/gdiplus_faq.htm

 

[Shawn B.]

Don't miss a thing, d'ya? =))

I'm writing a processor Simulator for the 8-bit 6502, 8-bit 65c02, and the
16-bit 65816. In particular, I'm doing this so I can debug (I have the
ability to modify the assembly language while the debugger is running
(similar to VB6 and VC++ Edit-and-Continue). I can only see raw
instructions. I'm not designing it to be a particular piece of hardware
(such as the Apple, Atari, Commadore, etc.). I can supply it plugins so my
plugins can treat an area of memory as a dedicated "something" such as text
display or graphics display, etc. For now, I'm concerned not even with the
text display. My end result, is actually more akin to being able to debug
my NES games I've been into writing lately. Also I want to be ready with my
own debugger for Andre LaMothe's upcoming XGameStation console (don't know
whether he'll have a debugger or not) but I want a debugger that allows me
to change the code while it's being debugged and my simulator allows for it
(as long as the source code file is available -- else if it was loading as a
..bin file then it won't be editable during a debug session).

So it's really more than a simulator. It's an integrated (non-seperable)
assembly, linker, dissassembler, debugger, simulator, etc. kind of a beast.
I chose C# because it's not the kind of thing I'd want to do in VB.NET (I'm
primarly a VB.NET programmer) but didn't feel like doing it in Managed C++
but also wanted it to target the .NET framework (as much as is reasonable
and I'll find out what that really means when its mature enough to start
using reliably).

I can do a decent job emulating 1-5MHz in C# on my PIII 1000MHz CPU with
GeForce II 64MB and 512MB Ram. I don't know how well it'll do emulating the
16-bit processor as I haven't deciphered the specs for the CPU yet (the
8-bit CPU is very easy compared to the 16-bit version of the processor). I
ultimately will make this Interfacable so I can write a Z80 core, as well.

These bit manipulations are the utmost inner-loop so I have to make them
tight. Looking at the JIT'd code for these methods seems to be decent but I
have to make too many function calls (read, jumps) for my tastes but doesn't
seem to be hurting me at the moment). Since I can't inline in .NET, I'll
have to live with it. Also, being an 8-bit and 16-bit CPU, it can't
generate the best code for the rotates and shifts because I have to emulate
8-bit and 16-bit shifts and rotates. I'm also trying to work out a better
way for simulating memory manipulation. Now I'm creating a string of (64kb
in length) but manipulating each byte by substrings has got to be expensive.
I'm looking into whether a byte array would be better.


Thanks for the help,
if interested, I'll post my binary class here for all to see.

I've completed the Rotates, Shift, converting numerics into binary (string
of 1's and 0's). I know now have to read the binary back to a numeric, and
allow a way to get/set/toggle individual bits.


Thanks,
Shawn

 

[Robert Jeppesen]

Do post it here!
I'd be interested to see, as I'm sure many others would.
You'll probably get a ton of optimization hints from it.

 

[Shawn B.]

Robert,

Youre suggestions have transformed my operations into 1-liners from 5
liners... you'll be the first to be notified (on these groups, of course).

I've opted to create a series of classes, Binary8, Binary16, Binary32,
Binary64 (all are unsigned).

I need to know, when you say "int x = 10;" is that a compiler trick or is
there a way to create my own type that can do that, for example, "Binary8 x
= 0xFF;"???


Thanks,
Shawn

 

[Jon Skeet [C# MVP]]

I need to know, when you say "int x = 10;" is that a compiler trick or is
there a way to create my own type that can do that, for example, "Binary8 x
= 0xFF;"???

It's not really a compiler trick, it's just that the type of 10 is int,
so the compiler can just assign that value to the variable.

I believe if you give the Binary8 class an implicit conversion from
int, you could get away with it.

I'm not sure whether or not you could get away with giving an implicit
conversion from byte, letting the compiler figure out that it can
convert a constant expression of type int to byte, and then from byte
to Binary8 - if that works though, it would remove the need for range
checking in your conversion.

 

[Robert Jeppesen]

In addition to Jon's response,
No, you can't overload the = operator.
Found a nice summary of what can be overloaded:
http://www.csharphelp.com/archives/archive135.html

 

[Jason Smith]

Instead of creating classes, use structs. Or, if you want the best
performance, just use the unsigned integers supported by .NET and C#.
Manipulate the integers using static methods.

Your BinaryX classes are really just unsigned integers with some added
functionality.

Robert,

Youre suggestions have transformed my operations into 1-liners from 5
liners... you'll be the first to be notified (on these groups, of course).

I've opted to create a series of classes, Binary8, Binary16, Binary32,
Binary64 (all are unsigned).

I need to know, when you say "int x = 10;" is that a compiler trick or is
there a way to create my own type that can do that, for example, "Binary8 x
= 0xFF;"???


Thanks,
Shawn

Do post it here!
I'd be interested to see, as I'm sure many others would.
You'll probably get a ton of optimization hints from it.

so

my

with

the

allows

me

for

it

loading
as a emulating

the

processor).

I

numeric,

and

of

the

value

and

 

[Bob Powell [MVP]]

Cool. I started on one for the Z80 a while back and I wanted to write a
Sinclair / Timex Spectrum emulator.

Some years ago I wrote a complete disassembler for the z80 for the Grundy
corporation (newbrain) before they went belly up so I got to know the
instruction set inside-out.

Great project. I'd like to keep informed of it if you don't mind.

Regards,

Bob.


--
Bob Powell [MVP]
C#, System.Drawing

September's edition of Well Formed is now available.
http://www.bobpowell.net/currentissue.htm

Answer those GDI+ questions with the GDI+ FAQ
http://www.bobpowell.net/gdiplus_faq.htm

 

[Shawn B.]

Yeah, I learned that (you can overload it in MC++)... instead I used the
"implicit" gig.

I am embarking now to overload all the operators for this struct and
implementing IConvertable as well as some other Interfaces and things. I
want to make it as complete as I possibly can since I intend to use it very
intensely.


Thanks,
Shawn

 

[Shawn B.]

Yes, I had already (in the 3rd major revision of this thing) changed it to
use structs. Each of the Binary8, Binary16, Binary32, and Binary64 are
Byte, ushort, uint, and ulong respectively. If I don't use unsigned then
the rotates crash if rotating a set high or low bit.

Yes, they are just instrinsic valuetypes with added functionality, for that
matter.


Thanks,
Shawn

 

[Shawn B.]

BTW:

I am at the point now where I can say:

Binary8 bin = "10101110";
Binary8 bin2 = bin - 0x03;
Binary16 bin3 = bin << 4;
....
Binary32 bin4 = bin3.RotateLeft(7);
Bin4 >>= 1;

bin4.Bit(7) = False; // I'm Thinking about being able to say Bit(x) = ON;
Bit(x) = OFF; Not sure yet, perhaps even
// Bit(x) = 1; Bit(x) = 0;

MessageBox.Show(bin4.ToString());

The ToString() method returns a string of 1's and 0's representing the
binary form of the value.
Binary8.ToString() returns similar "00011111";
Binary8.ToString(False) returns similar "11111"; // no trailing padding of
0's.

I still have to work out the "&" and "|" operators to the the same effect as
"bin.And(0x8A);" or something like that.

Still interested?


Thanks,
Shawn

 

[Shawn B.]

Well, I've posted it twice and no one is jumpting to optimize it... perhaps
it's just so well working that there's no need? he he




Thanks,
Shawn

 

[Robert Jeppesen]

Hi Shawn,

I've looked at it, and no optimization suggestions other than
some obvious simple ones:

public int GetBit(byte index) {
if (index > (byte)SIZE-1) {
throw new IndexOutOfRangeException("Bit " + index.ToString() + " does not exist.");
}
uint val = (uint)Value;
byte shift = (byte)(index - 1);
if (index == 0)
return (int)(val & 0x01);
else
return (int)(val >> shift) & 0x01;
}

This is very verbose, and you're creating lots of temp variables
I'd rather write something like (untested):

public int GetBit(byte index) {
if (index > (byte)SIZE-1) {
throw new IndexOutOfRangeException("Bit " + index.ToString() + " does not exist.");
}
return (int)((Value >> index-1) & 0x01;
}

The check to avoid a shift probably just hurts performance anyway.
Also for SetBit, the two if else must result in unwanted branches:

public void SetBit(byte index, int value) {
if (index > (byte)SIZE-1) {
throw new IndexOutOfRangeException("Bit " + index.ToString() + " does not exist.");
}
byte shift = index;
if (index == 0) {
if (value == 0)
Value &= unchecked((ushort)~0x01);
else
Value |= 0x01;
}
else {
if (value == 0)
Value &= ((ushort)~(0x01 << shift));
else
Value |= (ushort)(0x01 << shift);
}
}

How about this instead (also untested):

public void SetBit(byte index, int value) {
if (index > (byte)SIZE-1) {
throw new IndexOutOfRangeException("Bit " + index.ToString() + " does not exist.");
}
if (value == 0)
Value &= ((ushort)~(0x01 << index));
else
Value |= (ushort)(0x01 << index);
}

This is the only type of optimizations I came up with, so I thought it wasn't worth posting.

 

[Robert Jeppesen]

Oh, nice classes by the way. They seem pretty fast.

 

[Shawn B.]

Robert,

You're first suggestion crossed my mind, I have already started eliminating
the temp variables thus far (thus far only in the 8-bit operator overloads
region area). Just be happy you didn't see the earlier version I first
posted, it didn't cast because I kept getting compile errors, it ran
everything through a Convert.ToUIntX() call but I've since removed those...

I also plan to move the Shift function logic into the overload to avoid the
extra function call.

The GetBit optimization hint you provided is what I've been wanting to do
but didn't quite put it they way you did... I'll try that and see how it
works. I don't know what to do about that check in it, though. I feel like
it's necessary. Perhaps I could just make the if based on some calculation
and then just throw an exception at the end of the function and avoid that
early check, perhaps that would be a better approach. Such as (if (Value ==
0 && Value < SIZE-1) {)

There is one thing I need help on.


I don't like the GetBit, SetBit approach. I would rather be able to say x =
BinaryX.Bit[0]; BinaryX.Bit[0] = 1; I created a Bit class that had an
indexer but it had problems. Not the least of which was a circular
reference that ended up being a memory leak, not what I wanted in the
innermost loop of an emulator. The second being that I could easily say x =
x.Bit[0] but not easily say x.Bit[0] = 1.

Do you have any ideas how I might approach this? I was using a property
get/set at first but that didn't work, instead I just created a public
module variable "public Bit Bit;" and then it had a constructor to pass a
reference of the Binary class into it so it would be able to change the bit
if a bit was set. Not a good approach. Do you have any ideas?


Thanks for your help,
Shawn

 

[Shawn B.]

Also,

I'm using byte in some of the parameters such as GetBit(byte index), I'm
wondering, if there incurrs a performance penalty because the processor
thinks in 32-bit and not 8-bit, perhaps I should just make that int? Do you
think it matters? I know in assembly language it takes longer to "think"
8-bit than 32-bit (I program windows in asm as a hobby).


Thanks,
Shawn