Loops in floats

[AlfaRomeo]

Hi, I want to convert a routine from QB64 to PureBasic but it includes a loop with a STEP in floating point..
How I could convert this:

Code: Select all

FOR a = 0 TO 50 STEP .04
...
code
...
NEXT

Thanks in advance

[RASHAD]

Hi
Try

Code: Select all

Define a.f
Repeat
  Debug a
  a.f+0.04 
Until a => 50

Code: Select all

Define a.f
While a < 50.04
  Debug a
  a.f+0.04 
Wend

[STARGÅTE]

Please note that loops with floats are not numerically stable.
Depending on the increment value, you sometimes reach the final value and sometimes not, even if you use doubles!

Code: Select all

Define a.d, da.d

a = 0.0  :  da = 0.1
While a <= 2
	Debug StrD(a, 3)
	a + da
Wend
Debug "---"

a = 0.0  :  da = 0.2
While a <= 2
	Debug StrD(a, 3)
	a + da
Wend
Debug "---"

It is more stable to use:

Code: Select all

Define ia.i, a.d
For ia = 0 To 5000 Step 4  :  a = ia / 100.0
	; code with "a"
Next

[AlfaRomeo]

With the code:

Define ia.i, a.d
For ia = 100 To 5000 Step 4 : a = ia / 100.0
; code with "a"
Next

it worked like a charm

Thank you all

[jacdelad]

I would prefer the RASHAD version since it avoids the division. But beside speed, both work.

[wilbert]

I would prefer the RASHAD version since it avoids the division. But beside speed, both work.

If the goal is to avoid division you could also do

Code: Select all

Define ia.i, a.d
For ia = 100 To 5000 Step 4 : a = ia * 0.01
; code with "a"
Next

[jacdelad]

Yes, that's good too. But RASHADs version uses an addition, this one a multiplication. I guess the addition is still a bit faster.
So many solutions for one problem, that's one of the points why I like programming.

[STARGÅTE]

I guess the addition is still a bit faster.

No, not on modern CPUs/FPUs. Nowadays, also MUL or FMUL need only a single clock cycle, similar to ADD or FADD.
But of course, adding 0.04 directly to the iteration variable is faster than iterating a dummy integer variable and then scale it.