64 bit version of xilinx ISE

The Linux/i386 version of ISE 6.2 works on 64bit Linux. I believe that even 32bit programs get another Gig of address space on 64bit Linux. (32bit Linux typically gives a process 2Gig of user mode address space, 64bit Linux gives it 3gig., but my memory may be fuzzy here.)

Followup to: By author: Stephen Williams In newsgroup: comp.arch.fpga

Make that 3 GB and 4 GB, respectively.

-hpa

I'm surprised that no one from Xilinx has answered this question. Early last year, when 6.1 was in the works, I was told by someone at Xilinx that they were holding off the Linux native version until they had a 64 version. Obviously that didn't happen, the current version is Linux native but not 64 bit (unless the Solaris version is 64 bit, does anyone know?). I would expect that the 7.1 release will probably have 64 bit support now that AMD64s are common. The memory requirements for routing the largest V4s must be bumping up against the 32 bit limit unless they've gotten much smarter with their routing algorithms (which they may have). Only map and par need to be ported to 64 bits so the job shouldn't be that hard.

Here is a response to this inquiry on Xilinx 64-bit support so that there is no misunderstanding about it however I must first give the disclaimer that plans can and do change so this information may not follow exactly how things will get rolled out. I am also not going to claim to be an expert in operating systems so some of the information here may be incorrect so take what you hear here with a grain of salt as they say. One last thing, the main motivation to going to 64-bit is not really speed but more memory space (not sure if everyone understands this). 32-bit limits you to 4 Gigs of memory space which for the larger and more complex FPGAs, is starting to "bump its head" on this memory limit for all phases of design. In the past, significant speed gains has not been seen by moving from 32-bit to 64-bit and many times the opposite has been seen as well as memory requirements can be raised. This is not to say that things can not or will not be faster when working on 64-bit systems/OS/software but to set a realistic expectation, I would not count on it. At this point and time, I would suggest to look for a 64-bit OS and applications mainly if you expect to need the more memory space, rather than trying to buy additional speed of processing. The memory space is a for sure thing, the speed is not so sure. Place & route and other algorithms have got smarter when it comes to memory usage and Linux is better (requires less) memory than Windows Operating system to run our software and we have been able to increase the memory ceiling from 2 GB to 3 GB to 4 GB however at the same time FPGAs continue to grow significantly from family to family and soon if not already will outgrow the limits of a 32-bit system.

The current released version of the software, ISE 6.3 is all 32-bit however there is an internal 64-bit version for Solaris that can be released upon request. The 64-bit version can be slower and consume more memory than the 32-bit version and since the 32-bit version of Solaris can allocate 4 Gigs to applications, we have found that it is not necessary for all but the rarest of occasions an thus have held on to it. As for Linux, as mentioned, the current version is 32-bit. How much memory that can be allocated to it depends on the Linux kernel. Some 32-bit Linux distribution kernels by default allocate 2 GB to applications, other 3 GB (RHEL allocates 3 GB). 32-bit Linux kernels can also be reconfigured to allocate over 3 Gigs but generally do not come "out of the box" that way. If you have a 64-bit version of Linux and run the current 32-bit ISE application, then you generally get a full 4 GB of memory space as you do for Solaris. Similarly, this is generally more than enough for most current FPGA targets in most situations. A native 64-bit version of ISE for Linux Red Hat Enterprise Edition is intended to be released in the 7.1i software. It will support the Opteron processor with support for the Intel Xeon-64 pending. Speed and memory characteristics are yet to be determined so I can not comment on that yet however there should be a seemingly unlimited memory space available to that version. How much memory you can have will more be determined on what the hardware and your wallet can support than the OS and ISE software. That software is due out the first quarter of next year. In following releases, Windows 64-bit support can be expected likely on the same processors as the Linux support.

-- Brian

General Schvantzkoph wrote:

Do you know if they are using GCC or are they using the Pathscale compilers? The Pathscale compilers should give much better results for the AMD64 architecture.

...

...

For floating point code, yes. Not so much integer code. Mapping, P&R, etc are all pure integer workloads.

Tommy