C / C++ / MFC

Re: Algorithm Request [modified]

Francesco Aruta28-Jun-06 4:46

Francesco Aruta

28-Jun-06 4:46

Cedric Moonen wrote:
For the number of digits after the floating point, I don't have any idea.

well you can do the same... just multiply the initial number by 10 ...

F.

TBEditor: a pandapowered app!

Viorel.28-Jun-06 4:27

Viorel.

28-Jun-06 4:27

I think you can investigate the following approach.

For the "before" part of positive values:

float f;
long m = (long)f;
int before;

if( m == 0) before = 0;
else if( m <= 9) before = 1;
else if( m <= 99) before = 2;
else if( m <= 999) before = 3;
else ...

and so on according to a maximal value.

Hope it is fast enough.

For the "after" part, you can try a similar idea, but I am afraid it will be some problems related to imprecise representation of decimal float numbers in binary format, but such issue you will see in case of string-convertion-based method too.

-- modified at 10:35 Wednesday 28th June, 2006

Sarath C28-Jun-06 4:42

Sarath C

28-Jun-06 4:42

can't we use mod (%) instead?

SaRath.

"It is your attitude, not your aptitude, that determines your altitude - Zig Ziglar."
My Blog | Understanding State Pattern in C++

Viorel.28-Jun-06 4:49

Viorel.

28-Jun-06 4:49

Sarath.<fon wrote:<="" div="">can't we use mod (%) instead?

I suppose it is not faster.

Re: Algorithm Request [modified]

kylur28-Jun-06 4:28

kylur

28-Jun-06 4:28

double x(123.3)
ceil(log10(x)) => 3 ie the # digits left of decimal.

I'm still working of # digits right of decimal...

Regards,
Kylur

Tnarol28-Jun-06 5:20

Tnarol

28-Jun-06 5:20

Be careful of the behaviour if x = 0 when using "log10".

For the number of digits after "." I think you're going to have some problems, simply because the computer representation of floats and doubles of numbers do not exactly correspond to the value you gave. For instance it could happen that 123.3 actually corresponds to 123.29999999999999 as soon as it is stored as a float. So trying to guess the number of digits after "." makes no sense. You could maybe get around this if you limit the possible amount of decimal numbers to a well chosen value.

Re: Algorithm Request [modified]

C^hris L^osinger28-Jun-06 5:41

C^hris L^osinger

28-Jun-06 5:41

you're going to run into floating-point precision errors, unless you round numbers to a maximum number of decimal places. ex. you're going to come across something like 1.9999999999...

Cleek | Image Toolkits | Thumbnail maker

Zac Howland28-Jun-06 6:18

Zac Howland

28-Jun-06 6:18

String operations are your best bet for the most accuracy and flexibility (easy to control the number of significant figures). Since you don't want to go that route, this might get you started:

To count left side of decimal

<br />
unsigned long countLeftOfDecimal(double x)<br />
{<br />
	if (1 > abs(x))<br />
  	{<br />
     		return 0; // x is between -1 and 1<br />
	}<br />
	else if (1 == abs(x))<br />
	{<br />
		return 1; // x is either -1 or 1<br />
	}<br />
	else<br />
	{<br />
		return ceil(log10(abs(x)));<br />
	}<br />
}<br />
<br />
unsigned long countRightOfDecimal(double x)<br />
{<br />
	long y = x;	// truncate the decimal<br />
	double x1 = x - (double)y;	// keep only the decimal<br />
	long count = 0;<br />
	while (0.0 != x1)<br />
	{<br />
		double x2 = x1 * 10.0;	// move the decimal 1 place<br />
		y = x2;			// truncate the decimal<br />
		x1 = x2 - y;		// keep only the decimal<br />
		++count;<br />
	}<br />
	return count;<br />
}<br />

Keep in mind that you will run into problems when you have numbers like 1.99999999999... with this algorithm -- which is why converting to a string for counting significant figures is a better solution, and may actually be faster than any non-string solution you come up with that doesn't have problems with infinite repeating decimals.

If you decide to become a software engineer, you are signing up to have a 1/2" piece of silicon tell you exactly how stupid you really are for 8 hours a day, 5 days a week

Zac

-- modified at 15:16 Wednesday 28th June, 2006

Rilhas9-Jul-06 6:25

Rilhas

9-Jul-06 6:25

I don't think there is an easy solution for this without passing the real values to a string. This is mainly due to the fact that the real numbers may have an exact representation as decimal and an inexact representation as native. And, of course, vice-versa.

For example, float 0.2 has an inexact native representation. The exact representation is 0.200000002980232. But the resolution present in a float is not enough for all the decimals.

This just means that you probably won't have much luch analysing the bits in the float (or double). Their memory layout can be found in http://www.psc.edu/general/software/packages/ieee/ieee.html[^].

In any case, finding the values you want by using strings or not, this can always get complicated:

float f=0.00031233478376329384729344178300483284723740923762730534976862e7f;
double d=f;
printf("f=%f, d=%.15g", f, d);

This code compiles, but due to the lack of precision in the floating point representations the results are not what one would expect.

So, the only fast solution I can imagine involves the assumption that you have not that many number variations, they are always around zero (controlled exponent), and that you do not have rounding errors. With these assumptions you could build a fast look up table.

For example, assume that you will have numbers between -999.99 and +999.99. This results in 199998 diferent numbers plus the zero, i.e., 199999. So you could program the following algorithm:

// build table
float f;
int i;
i=0;
for(f=-999.99; f<=-999.99; f+=0.01f) {
table[i].left=ComputeLeftByUsingStrings(f);
table[i].right=ComputeRightByUsingStrings(f);
i++;
}

// get the left/right values for float f fast!
f*=100.0f; // shift so that the smallest decimal becomes an integer
int i=(int)f; // convert to int to use as offset to the table
i+=99999; // correct the start offset to start at 0 for the most negative number
int left=table[i].left; // left will contain the number of characters to the left
int right=table[i].right; // right will contain the number of characters to the right

I hope this helps!

Rilhas

FileMapping problem

Kharfax28-Jun-06 3:51

28-Jun-06 3:51

Hi people, I need your help again Big Grin | :-D

Im working with Memory mapped files for the first time.
I need to open a file map, and then go trought each byte of the mapped file and add one to it...

so...

<br />
<br />
LPVOID buffer = NULL;<br />
<br />
buffer = MapViewOfFile(hFileMappingObject, dwDesiredAccess | FILE_MAP_WRITE, dwFileOffsetHigh, dwFileOffsetLow, dwNumberOfBytesToMap);<br />
<br />
LPBYTE buf2 = (LPBYTE)buffer;<br />
<br />
for(SIZE_T i = 0; i < dwNumberOfBytesToMap; i++ ) {<br />
      (buf2[i])++;<br />
}<br />
<br />

well, that is exploding my test application in uggly ways.

Any suggestion is welcomed

Thanks in advance Big Grin | :-D

Matt Godbolt28-Jun-06 4:07

Matt Godbolt

28-Jun-06 4:07

When you say 'exploding' can you be a bit more specific? Smile | :)

Are you getting a valid 'buffer' back or is it NULL? Is the error itself a read access violation, or a write violation? Is it the first value of 'i' that errors, or one of the later ones? What's 'dwDesiredAccess' - does it contain 'FILE_MAP_READ'? How are you getting hFileMappingObject? Is the file you're modifying read only?

The code looks sensible enough to me there, so I suspect it'll be something outside of the snippet you've posted there Smile | :)

Matt Godbolt
Engineer, ProFactor Software
StyleManager project

Kharfax28-Jun-06 4:38

28-Jun-06 4:38

It hangs up in the first i value, but I tried reading and it works fine.
So the problem is when I try to write the filemap...

Is strange, because I'm in a windows hook, and the filemap is beign oppened by the notepad, who provides read/write access to the file.

I'll check filemap access rights

Thanks for the feedback Matt

Jun Du28-Jun-06 7:12

Jun Du

28-Jun-06 7:12

(buf2[i])++;

looks suspecious. Check two things:

1) What MapViewOfFile() returned is the starting address of a memory chunk in your address space. You may not be able to apply [i] to buf2.

2) If the memory is not initialized, you cannot apply ++ to its content.

Best,
Jun

Kharfax28-Jun-06 7:37

28-Jun-06 7:37

If I read buf2 contents, I can see the contents of the file.
When I try to write in buf2, the program hangs up.

What is the method to read some specific data of the file mapping?
Let me explain me better, if I have a CFile object I can use Seek method to move around.
In a Memory mapped file, how can I do that?

Kharfax28-Jun-06 8:41