On Mon, 1 Sep 2014, Andreas Hansson wrote:
Hi Nilay,
That is a very good point, and thanks for spending some cycles on this.
I¹m not pushing for a transition, I merely thought it made more sense, but
I forgot about the hello world tests.
Does the ³hello world² actually add any value to the regressions? Would it
not be better to: 1) run a more extensive regression using Ruby + an o3
CPU model (linux boot etc), or 2) use a more extensive synthetic tester
(e.g. memtester with actual sharing, which is something we¹re working
on...) for some of these protocols?
I am fine with adding more tests. I do sometimes test by booting Linux so
as to ensure things are in a working state. I am not sure if we would
like to see the time for regressions going up. I am unable to recall the
inner workings of the testers that we use for ruby, but I am sure they
test sharing.
As a side note, I¹ve managed to make the memory system (src/mem)
completely ISA independent, so we could compile the entire memory
directory once for all ISAs. Unfortunately we also need to compile it once
for every coherency protocol in Ruby. I¹m not sure there is any sensible
way around it, but it would be good to get your thoughts on this.
If I remember correctly, there is one particular file (MachineType.hh)
that is the stumbling block in compiling all protocols together. I might
look at this again once I am done with another ruby thing I am working on
currently.
Thanks
Nilay
_______________________________________________
gem5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/gem5-dev
