> 1) What guest OS / OSes are running on that virtual machine?
> Are they simulated linux, windows, or other ?
Yes, I have ubuntu linux on the virtual machine.
> 2) Would Maarten try running the simulator directly
> (without the overhead of make) on the virtualbox machine:
> time ../../sim/ucsim/z80.src/sz80.src/sz80.exe -tr2k
> gen/ucr2k/gcc-torture-execute-961017-2/gcc-torture-execute-961017-2.ihx <
Well, it is a little shorter, but there is also less to
do as both make and python are not involved now.
> On cygwin, "make" has an order of magnitude more overhead vs. building
> on linux
> as result of differences in how memory is managed in linux vs windows.
Cygwin is not involved here. But I do have an idea why
the simulator is slow though not necessarily slower on a
VM. It uses a lot of malloc's because every target
memory cell is encapsulated in a C++ class with a
constructor. And malloc is notoriously slow.
> On 4/26/2012 12:32 PM, Philipp Klaus Krause wrote:
> > gcc-torture-execute-961017-2 (f: 0, t: 0, c: 1, b: 3278, t:
> > 8656305)
> > Summary for 'ucgbz80': 0 failures, 6849 tests, 1539 test cases, 4559615
> > bytes, 28665374 ticks
> > real 0m47.556s
> > user 0m37.402s
> > sys 0m8.937s
> > This is on a virtual machine in virtualbox on a Pentium 4 @ 3.0GHz.
> > And for the ucr2k:
> > Running ucr2k regression tests
> > gcc-torture-execute-961017-2 (f: 0, t: 0, c: 1, b: 2574, t:
> > 3936475)
> > Summary for 'ucr2k': 0 failures, 6853 tests, 1539 test cases, 3765745
> > bytes, 15590124 ticks
> > real 0m23.536s
> > user 0m17.881s
> > sys 0m5.124s
> > ...
> > Maarten
> > Weirdly I see a similar problem when running Debian GNU/Hurd inside a
> > qemu-kvm vm: Running the simulator takes a really long time, but neither
> > the host nor the guest show particularly high cpu usage. Maybe there's
> > something in our simulator that just happen to be very inefficient when
> > running on an OS inside a VM?
> > Philipp