Convert string number to negative number

[Steven Pringels 3]

Hi everybody,

This may sound like a strange request but I need to know how I convert
a number that is represented as string into a NEGATIVE value with the least
amount of CPU cycles. The result should be a numeric variable.

Dim extA as Ext
Dim A as String
A = "5"
A = "-" & A
extA = Val(A)

Eventually there is an operation on the result and that result is stored as
a string.

Or are there other tricks.

Steven

[Marco Pontello]

Just starting with what you wrote, this will save a string concatenation:

Dim extA as Ext
Dim A as String
A = "5"
extA = -Val(A)

But I'm not sure if I understood correctly...

Bye!

[Kev Peel]

There is no shortcut (that I know of) to do run-time calculations from a string. Search the forum for "math parser" or "expression evaluator"

[Steven Pringels 3]

Thanks Marco, Kev,

I was actually looking for something more low level (even ASM) e.g. by manipulating the bit order itself but I guess that the process involved
will be quite complex especially when dealing with floating point numbers
that are first represented as a string and then should be made negative.

[Michael Mattias]

Do something like this, except instead of hex digits, use decimal digits and manage your own carries.
Byte to HEX character pair and vice-versa conversions

Frankly, to suggest that -1& * VAL(string) is not fast enough is an insult to the compiler developers. They have been doing this for 20+ years and maybe, just maybe have learned a few things about doing it well and doing it quickly.

Of course, VAL() does support various forms of numbers (explicit decimal points, binary, octal and hex representations) as well as scientific notation (E or D in string), so if you can restrict the input to known decimal digits plus zero or one explicit decimal points, the substitution table thing ought to be worth a shot.

[Michael Mattias]

However, once you have an EXT (or single or double for that matter) you can negate it by simply setting the sign bit (use OR or BIT SET ) .

That is, once you have an EXT.

[Michael Mattias]

OR (guess because code not shown, application not described)....

If you want to do this fast, it suggests you have "many".

Because it's a number, it suggests you want to do some arithmetic on it.

Maybe you can do the negation when you do the arithmetic - once per number, instead of negating each potential operand individually.

[Kev Peel]

VAL doesn't support this (which I believe is what Steven meant):

 
calc = VAL("(2 + 2) * 5")

You need a math parser or roll-your-own.

[Mel Bishop]

Divide by minus one.

temp = val(string) / -1

[Mike Doty]

QuickPak Pro has/had QPValI and QPValL for siimilar things,but never for EXT.
There might be a math library available somewhere.
I requested a stripped down VAL, but the request was rejected.

This is just playing around to see the bytes.

#COMPILE EXE
#DIM ALL
UNION EXT_UNION
  s AS STRING * 10
  b(1 TO 10) AS BYTE
  e AS EXT
END UNION
FUNCTION PBMAIN () AS LONG
  DIM e AS EXT
  DIM s AS STRING
  DIM u AS EXT_UNION
  DIM x AS LONG
  u.e = -5
  FOR x = 1 TO 10: PRINT u.b(x);: NEXT:PRINT
  WAITKEY$
END FUNCTION

Hi Steven

[Steven Pringels 3]

Hi Michael, Mel,

I like both your solutions. The obvious is to /-1 but I'll check out the OR and BIT SET alternatives. Thanks !

[Mike Doty]

Can your strings be limited to a range smaller than EXT?
Are they always floating point?

String to long.

http://www.powerbasic.com/support/pbforums/showthread.php?t=25163

[Guy Dombrowski]

*-1 is faster than /-1

[Paul Dixon]

Not the fastest and has no error checking, assumes the string is +ve and has only characters 0 to 9 and decimal point. It may get you started

PBCC5 program
FUNCTION PBMAIN () AS LONG
#REGISTER NONE

LOCAL MyNumber AS ASCIIZ *100

MyNumber="12345.012345645675"
ten!=10
tenth##=1/10##
temp&=0
answer##=0


DIM extA AS EXT
DIM A AS STRING
A = MyNumber

TIX t&&
'A = "-" & A
extA = -VAL(A)

TIX END t&&
PRINT "time="t&&
PRINT extA



TIX t&&

!lea eax,MyNumber       'get address of string, it's zero terminated
!fld ten!               'some useful values for the FPU, 10
!fldz                   ' zero (the initial value for the answer)

'do the integer part
lp:
!movzx ecx,byte ptr [eax]   'get current byte
!inc eax                    'point to next byte
!cmp cl,0                   'is current byte zero? (end of string)
!je skip                    'yes, exit
!cmp cl,"."                 'is it a decimal point?
!je fraction                'yes, do the fraction part


!sub ecx,"0"                'no, convert ascii character to binary
!mov temp&,ecx              'transfer the byte to the FPU
!fmul st(0),st(1)           'multiply answer so far by 10
!fild temp&                 'get digit
!faddp st(1),st(0)          'add it in to the answer

!jmp short lp               'go back for next character

fraction:
!fldz                       'start the fractional part off at zero
!fld tenth##                'some useful values for the FPU, 1/10 (the scale for the fractional multiplier)
!fld1                       ' 1  (the fractional mutiplier)


lp2:
!movzx ecx,byte ptr [eax]   'get current byte
!inc eax                    'point to next byte
!cmp cl,0                   'is current byte zero? (end of string)
!je skip2                    'yes, exit

!sub ecx,"0"                'no, convert ascii character to binary
!mov temp&,ecx              'transfer the byte to the FPU
!fmul st(0),st(1)           'update the fractional multiplier

!fild temp&                 'get digit
!fmul st(0),st(1)           'scale the digit
!faddp st(3),st(0)          'add the scaled digit to the current fraction and pop the digit

!jmp short lp2              'go back for next character

skip2:
!fstp st(0)             'clear the scale off the stack
!fstp st(0)             'clear the scale multiplier off the stack
!faddp st(1),st(0)      'add the fraction to the integer and pop the fraction

skip:
!fchs           'change the sign before storing answer
!fstp answer##  'save the answer

!fstp st(0)     'clear the "10" off the FPU stack


TIX END t&&
PRINT "time="t&&
PRINT answer##

WAITKEY$

END FUNCTION

Paul.

[Mike Doty]

Very nice, Paul.
I turned it into a function for him and ran a few tests.

  #REGISTER NONE
 
  FUNCTION PBMAIN () AS LONG
  DO
    INPUT "Number ";s$
    IF LEN(s$) = 0 THEN EXIT DO
    REPLACE "-" WITH "" IN s$
    e## = ExtNumber(s$)
    ? STR$(e##)
  LOOP
  END FUNCTION
 
FUNCTION ExtNumber(s AS STRING) AS EXT
 
 'Paul Dixon code  2/6/09 
 ' [URL]http://www.powerbasic.com/support/pbforums/showthread.php?t=39802[/URL]
 
 LOCAL MyNumber AS ASCIIZ *100
 LOCAL Ten    AS SINGLE
 LOCAL tenth  AS EXT
 LOCAL Temp   AS LONG
 LOCAL Answer AS EXT
 MyNumber = s  'added line, you can pass MyNumber AS ASCIIZ *100
 ten!=10
 tenth##=1/10##
 temp&=0
 answer##=0
 
!lea eax,MyNumber       'get address of string, it's zero terminated
!fld ten!               'some useful values for the FPU, 10
!fldz                   ' zero (the initial value for the answer)
'do the integer part
lp:
!movzx ecx,byte ptr [eax]   'get current byte
!inc eax                    'point to next byte
!cmp cl,0                   'is current byte zero? (end of string)
!je skip                    'yes, exit
!cmp cl,"."                 'is it a decimal point?
!je fraction                'yes, do the fraction part
 
!sub ecx,"0"                'no, convert ascii character to binary
!mov temp&,ecx              'transfer the byte to the FPU
!fmul st(0),st(1)           'multiply answer so far by 10
!fild temp&                 'get digit
!faddp st(1),st(0)          'add it in to the answer
!jmp short lp               'go back for next character
fraction:
!fldz                       'start the fractional part off at zero
!fld tenth##                'some useful values for the FPU, 1/10 (the scale for the fractional multiplier)
!fld1                       ' 1  (the fractional mutiplier)
 
lp2:
!movzx ecx,byte ptr [eax]   'get current byte
!inc eax                    'point to next byte
!cmp cl,0                   'is current byte zero? (end of string)
!je skip2                    'yes, exit
!sub ecx,"0"                'no, convert ascii character to binary
!mov temp&,ecx              'transfer the byte to the FPU
!fmul st(0),st(1)           'update the fractional multiplier
!fild temp&                 'get digit
!fmul st(0),st(1)           'scale the digit
!faddp st(3),st(0)          'add the scaled digit to the current fraction and pop the digit
!jmp short lp2              'go back for next character
skip2:
!fstp st(0)             'clear the scale off the stack
!fstp st(0)             'clear the scale multiplier off the stack
!faddp st(1),st(0)      'add the fraction to the integer and pop the fraction
skip:
!fchs           'change the sign before storing answer
!fstp answer##  'save the answer
!fstp st(0)     'clear the "10" off the FPU stack
FUNCTION = answer
END FUNCTION

[Michael Mattias]

>assumes the string is +ve and has only characters 0 to 9 and decimal point

That's one of the problems with these 'peformance enhancements'

The fact the data are stored as character strings without the correct sign suggests it is data not created by the application.....meaning it SHOULD be edited.. meaning perforce string operations are going to be required and negation will be but a tiny portion of the total processing time required anyway.

With faster negation code but no editing, you're just asking for faster "garbage out."

[Steven Pringels 3]

Waw !!! I'm impressed with the work you've all put in this. I got away for some 'hours' and a lot of stuff has been put into this.

Yes, there is more to this. It is part of a modified version of the Reverse Polish Notation algorithm that I'm using. The RPN parser doesn't know the difference between and operator (-) and a negative sign (-) hence I have to store it using a byte in a structure that indicates that the number is negative but then to solve the RPN I have to check for the negative-ness of a number. As everything is represented as a string (by design and because of the need for it). I requested a fast method to make a string number into negative Extended value.

Thanks to everyone who has put thought and work into this

[Mike Doty]

Michael, why do you preach? He asked for this and got it.

[Michael Mattias]

>why do you preach?

Because the only thing of real value any of us leave behind is the next generation.

[Mike Doty]

Why not create a thread of your own?