LNG Arithmetic Bug?

[XRaySpeX]

Can anyone tell me why the enclosed simple, but large, DEFLNG arithemetic does not give the correct answer? Is it a bug or can I just not calculate?

It happens with other large values but I cannot sus out a pattern except that Scale is a power of 2 & from my experience Floating Point errors usually occur at powers of 2.

I have replaced IF (OCCUR >= YPOINTS * Scale) by IF (OCCUR / Scale >= YPOINTS) & it works correctly.

This was happening with QB64 v0.980, so I came back here today to get the latest version, but it's still happening.

[bplus]

Looks like the multiplication exceeds limits of LONG type

see:

Code: QB64: [Select]

1. DEFLNG A-Z
2.  
3. OCCUR = 1649329259: Scale = 16777216: YPOINTS = 350
4. Compare1 = OCCUR - YPOINTS * Scale
5. Compare2 = OCCUR >= YPOINTS * Scale
6.  
7. PRINT "OCCUR="; OCCUR; "Scale="; Scale; "YPOINTS="; YPOINTS
8. PRINT
9. PRINT "Why is OCCUR - YPOINTS * Scale ="; Compare1; " & not "; -4222696341; "?"
10. PRINT "Why is OCCUR >= YPOINTS * Scale = "; Compare2; "(TRUE) & not"; FALSE; "(FALSE)"; "?"
11. PRINT
12. PRINT "In Numbers: "; 1649329259 - 16777216 * 350
13. PRINT
14. PRINT "By replacing IF (OCCUR>=YPOINTS*Scale) by IF (OCCUR/Scale>=YPOINTS), it works OK"
15.  
16. FOR i = 1 TO YPOINTS STEP 10
17.     PRINT i * Scale
18. NEXT
19. PRINT "press any to continue..."
20. SLEEP
21. CLS
22.  
23. _DEFINE A-Z AS _INTEGER64
24. OCCUR = 1649329259: Scale = 16777216: YPOINTS = 350
25. FOR i = 1 TO YPOINTS STEP 10
26.     PRINT i * Scale
27. NEXT
28.  

[XRaySpeX]

Duh! I should have seen that. It's so obvious.

Thanks, @bplus.

[XRaySpeX]

Thanks, @bplus, you've given me an idea.

By switching from LONG to INTEGER64 I can extend the scope of my primes program from 2.1 E9 to 9.2 E18 :).

[bplus]

Primes, yum!