Reducing Computation Time

Hey I have somewhat tricky task of reducing computation time for one line of code in my professor's project. Is there any way to reduce the following code so that it computes at a faster rate (maybe 0.01s), right now the time to beat is 0.08s. The code is as follows:

(32 bit IEEE demultiplexed)

x = hex2float( dec2hex( fread(fid, 1, 'ubit32', 0, 'ieee-be') ))

Basicslly looking for quicker way to convert the binary file to float. If you're not sure, I'm open to specifics suggestions. Also, would it be better if I wrote a code like this in C first and then let matlab read from there? Thank you.

Answer 1

[Jan]

d = fread(fid, 1, 'uint32', 'ieee-be');
x = hex2float(sprintf('%x', d));

I do not have hex2float. Could you provide some example data? I assume typecast(uint32(d), 'single') will help also, if the format of fread is set accordingly. But without hex2float I cannot test this.

See also: http://stackoverflow.com/questions/5232726/32-bit-hex-to-32-bit-floating-point-ieee-754-conversion-in-matlab

Anyhow, 0.08 sec sounds extremely slow. I assume this is cause by the rest of the code.

[Walter Roberson]

hex2float() is in the FEX.

[Dejan Cvijanovic]

Sorry, I was only given that piece of code to work with, cannot supply/comment on rest of code. All I know is what is listed, need to read in a binary file and produce

[Dejan Cvijanovic]

*produce float numbers eventually

[Jan]

Thanks, Walter. I cannot find it in the FEX: http://www.mathworks.com/matlabcentral/fileexchange/?sort=date_desc_updated&term=hex2float

Answer 2

[Walter Roberson]

Is there a reason you are not just using

x = fread(fid, 1, '*float32', 0, 'ieee-be')

Also if you are doing a lot of calculations, it is almost always faster to use fread() to read a number of items at once. (If you stick with going through hex, then dec2hex() can be used on arrays.)

[Dejan Cvijanovic]

Yes, the file Im reading is binary with multiple formats. Does that change anything or is this still applicable?

[Walter Roberson]

Multiple binary formats should not affect my suggestion for reading a single 32 bit float.

[Dejan Cvijanovic]

So just to confirm, this code reads in binary and converts to float and conserves the same answer if I had convert to hex first?

[Walter Roberson]

The only way to know for sure is to try it.

I do not have access to MATLAB from home to experiment with. And even if I said that it worked, you would need to try it to be *certain*.

[Dejan Cvijanovic]

What about using 'ubit32=>float32' instead of '*float32'.....because binary is not in float, right?

[Jan]

'ubit32' is very inefficient, when it really splits the reading to 32 bits. 'uint32' reads and processes them in a block.

Answer 3

[Dejan Cvijanovic]

Can anyone confirm correctness of last two posts?

[Jan]

I'm sure *you* can confirm them. Simply compare the results and to be sure try it with a bunch of random numbers or bit-patterns.