[HECnet] Trying to build SIMH from git (on OpenIndiana)

[Dave McGuire]

On 03/13/2013 05:32 PM, Mark Pizzolato - Info Comm wrote:
  First of all, Cory...have you tried using Sun's compiler?   It kicks the snot out
of GCC for SPARC code generation.   I did give you a copy.

I would suggest that this would be a good idea.   

I have an old SparcStation which I fire up once every few years to test simh on a Big Endian platform.   It has a Linux disk and a Solaris disk (I don't recall the version).   I do not have a sun C compiler.   The last time I fired it up (more than a year ago) the basic testing worked out fine but I noticed that some things were so slow it was practically unusable.     I isolated some slow behavior down to an operation which walked across a 64MB chunk of VAX memory one byte at a time.   The 64MB array is stored in a single malloc'ed area.   This walk across the 64MB took some 20 seconds, while the same operation on my Windows desktop takes far less than any time I could measure on my watch.   My theory was that it was spending all of its time in the alignment trap handler, but I didn't have enough knowledge to dig into the details.

The test case is demonstrated by the following commands:

                    $ BIN/vax
                    MicroVAX 3900 simulator V4.0-0 Beta
                    sim> set cpu 64m
                    sim> save somefile         

The save operation writes the current simulator state to a file.

  It sounds to me like you need a fast SPARC, or access to one.

  I can give you an account on dev.neurotica.com, which is an
UltraSPARC-T1-based machine running Solaris 10 with Sun's C compiler
installed.   Several people are using it for free software projects;
that's its purpose in life.

  Let me know if you're interested.

                            -Dave

-- 
Dave McGuire, AK4HZ
New Kensington, PA