[SOLVED] Fnv-1a hash algorithm crashes with range check error

[Hoch]

Hi,

I'm trying to execute a simple fnv-1a hash algorithm but it crashes with a range check error (Runerror(201)).

See attached screen capture.

This is the algorithm:

Code: Pascal  [Select][+][-]

1. function fnv1a32 (constref str_to_hash: string): longword;
2. var
3.   str_pos: longword;
4. begin
5.   result := 2166136261;
6.   for str_pos := 1 to length (str_to_hash) do
7.     result := (result xor byte(str_to_hash[str_pos])) * 16777619;
8. end;

The code crashes in this line:

Code: [Select]

result := (result xor byte(str_to_hash[str_pos])) * 16777619;

The problem is that the product of the multiplication is langer than a longword (32 bit).

I'm pretty sure that the translation from c to pascal is correct.

What I'm I doing wrong?...

Thanks in advance! 

 

[Red_prig]

It looks like the algorithm assumes arithmetic overflow, so turn off range check in the project settings

[Thaddy]

If the compiler detects the value can in theory be larger than 32 bit it is expanded to 64 bit.
You can try this:

Code: Pascal  [Select][+][-]

1. result := longword (result xor byte(str_to_hash[str_pos])) * 16777619);

IOW a longword cast before assigning the result.
You should have seen a hint or warning about the expansion to 64 bit.

[Hoch]

Thanks red_prig for the fast answer. That did the trick.

The runtime error also happens with other hash algorithms like murmurhash3 or xxhash.

Adding {$RANGECHECKS OFF} or {$OVERFLOWCHECKS OFF} solved the errors.

Thaddy solution also works fine and eliminates the runtime errors so you don't need to use the directives but I guess that it adds a little bit of logic to the loop...?

I also found that you can and the result bits with $FFFFFFFF to avoid the runtime errors. But you are adding an extra operation.

Thanks for the useful answers!