EmbDev.net

Forum: FPGA, VHDL & Verilog floating point / real number


von soumava r. (Company: student) (soumava)


Rate this post
useful
not useful
i m working on a vhdl project which requires the use of floating point 
numbers ? please let me knew if there is any available source code for 
the same?

von Lothar M. (Company: Titel) (lkmiller) (Moderator)


Rate this post
useful
not useful
Up to now real numbers aren't synthesizable "automatically". You 
cannot write:
1
signal r, a, b, c, x, y, z : real;
2
3
  c <= a-b+3.3;
4
  z <= x+y;
5
  r <= c*z;
Instead you must implement the datapath by using IP-Cores or coding by 
yourself.

But:
WHAT is your actual problem?
WHY do you need floats?

von Oliver H. (ollih)


Rate this post
useful
not useful
Do you realy nead "floating point" or does fixpoint fits to your 
project?

So you could consider your integer variables as fixpoint values e.g. a 
8bit integer as a fixpoint with 4 integer and 4 fractional digits:
  deciaml    binary
     0.06125 0000_0001
     0.125   0000_0010
     0.5     0000_1000
     1.0     0001_0000
     3.75    0011_1100

von soumava r. (Company: student) (soumava)


Rate this post
useful
not useful
i am working on implementation of neural networks on FPGA. which 
requires addition and multiplication of real numbers.
i am and Electronics and communication student. i dont know what is IP 
core.
My professors have told me that i should define the package for using 
real numbers.

von Ale (Guest)


Rate this post
useful
not useful
... do you know what is a real number ? look for IEEE754 for example
... what range/precision do you need ? how big/small accurate should 
they be ?  (maybe fixed point is the way to go).

look at opencores.org, there are some fpu implementations there...

von pks (Guest)


Rate this post
useful
not useful
I think in case of an FPGA-implementation of a neural network (mlp?) you 
will have a big amount of parallel mult operations, which means a lot of 
ressource-requirement especially in case of float values. First you 
should check if you really need fp-precision.

Please log in before posting. Registration is free and takes only a minute.
Existing account
Do you have a Google/GoogleMail account? No registration required!
Log in with Google account
No account? Register here.