I am very glad to see so many discussions. Thank you!
At present I am using the PowerPC, because its bram is 128K, but the bram of microblze(64K) is not enough for us to run our programme. This is another problem.
About GCC£¬ I don't quite understand about Jon's meaning. I am just using the integrated tool EDK under Windows XP to "build the project", "update the bitstream", then I can get the file "download.bit", so I am not familiar with the command interface GCC, then how to start the GCC under EDK, is it the ¡°tools->software debugger¡±? How to compile with optimisation switched on (-O2)?
Lina
----------the former discussions----
How to speed up float computing
lina - 12:54am Jun 1, 2005 PST Guest User
Hi all, I am doing some programmes with EDK 6.3i on the Virtex-II Pro50. There are many math computing in the program, and I do them using the software libs provided by the board, as a result, it is very very slow that we could not endure. So I would like to know the effective and simple method to speed up the float computing. Please help me. Thank you very much.
Jon Beniston - 01:55am Jun 1, 2005 PST (#1 of 5) Guest User
Get EDK 7.1, which has support for h/w floating point.
lina - 04:14am Jun 1, 2005 PST (#2 of 5)
Thank you very much for your answer! We will try it.
Ben Jones - 08:34am Jun 1, 2005 PST (#3 of 5) Guest User
Hi Lina,
Get EDK 7.1, which has support for h/w floating point.
Just to clarify - this is almost but not quite true. If your code is running on MicroBlaze, then the new EDK does indeed include support for a floating-point unit in the FPGA fabric. If you are using the PowerPC (and since you're using V2-Pro, I suspect you probably are), then it doesn't.
I can suggest a few options:
1) Use a Microblaze core with FPU, either for your whole application or to accelerate those algorithms which use floating point arithmetic; 2) Get a PowerPC floating-point unit, for example:ca_fpu_ppc.html
3) Consider moving to Virtex-4 FX, which will soon have a PowerPC floating-point unit available.Hope this helps,
-Ben-
Joey - 10:01am Jun 1, 2005 PST (#4 of 5) Guest User
Hi Lina
I was also experiencing the same problems. Infact the computation was taking such a long time. At the end what we did was rewrite the whole program with very less computations. For example use "register" variables/keywords , compute commonly used variables before hand and so on and at the end we got a satisfactory computational speed. One more hint: Put the instructions in the ISOCM memory and dont use any Instruction side cache memory !! The results which I got were more than 10 times better Joe
"lina" schrieb im Newsbeitrag news: snipped-for-privacy@webx.sUNCHnE...
Hi all, I am doing some programmes with EDK 6.3i on the Virtex-II Pro50. There are many math computing in the program, and I do them using the software libs provided by the board, as a result, it is very very slow that we could not endure. So I would like to know the effective and simple method to speed up the float computing. Please help me. Thank you very much.
Jon Beniston - 11:55am Jun 1, 2005 PST (#5 of 5) Guest User
For example use "register" variables/keywords , compute commonly used variables before hand and so on and at the end we got a satisfactory computational speed.
With GCC there should be no need for this. Was you compiling with optimisation switched on (-O2)?
Cheers, Jon