The two faces of SHR

Like Marek Mauder in his blog post from 2009, I have run into a problem because of the different implementations of the right shift operator (shr in Pascal) in different languages.

There I was happily translating some code from Java to Object Pascal when I encountered Java's >>> operator. A quick look at the Java docs tells me this is a logical right shift. No problem, just use shr. Next I found Java's >> operator. This time its an arithmetic (sign preserving) right shift. Problem - no Pascal equivalent, and since operation was on signed integers I couldn't just use shr because there's a bug waiting to happen should any of the integers go negative.

Marek (above) said that replacing

a >> 1

with

a div 2;

worked for him.

This is fine for a single bit shift, so I thought I would see if it scaled for shifts of more than one bit. I tried dividing by 2 to the power of the number of bits to shift:

function SAR(Value: LongInt; Shift: Byte): LongInt;
begin
  Shift := Shift and 31;
  if Shift = 0 then
    Exit(Value);
  Result := LongInt(LongWord(Value) shr Shift);
  if Value < 0 then
    Result := LongInt(LongWord(Result) or ($FFFFFFFF shl (32 - Shift)));
end;

I can see several ways to optimise the function but all of them make the code less readable. So I'm not going to do the optimisations unless I have to.

I should be quite trivial to create 8, 16 and 64 bit integer overloads of the routine.

Comments

Popular posts from this blog

Initialising dynamic arrays

Deleting elements from a dynamic array

New String Property Editor Planned For RAD Studio 12 Yukon 🤞