Andrew Whitworth wrote:
Problematic architectures (I'm looking at
your low-mem box for this, kid51!) and various HLLs.
See below.
This branch should not bring any appreciable changes in performance or
memory usage. If you see something that looks like a huge slowdown or
a significant jump in memory consumption, that's a bug.
I'm very glad that you mentioned this. It has often seemed in the past
that we ignored one or both of performance and memory usage with respect
to evaluating whether a branch should be merged into master.
So far, I have only had time to test how the branch handles 'make'. The
data below reports elasped time in seconds running several iterations of
'make' in master and in the branch. I used exactly the same
configuration options (or lack of options) that I have used every day
for years.
Linux/i386:
Branch: master
1 119
2 120
3 120
Mean: 119.7
Branch: whiteknight/imcc_compreg_pmc
1 127
2 120
3 118
Mean: 121.7
Branch: master
1 120
2 119
3 120
4 117
5 117
6 116
Mean: 118.2
Branch: whiteknight/imcc_compreg_pmc
1 119
2 121
3 120
4 119
5 118
6 122
Mean: 119.8
Darwin/PPC
Branch: master
1 844
2 865
3 884
Mean: 864.3
Branch: whiteknight/imcc_compreg_pmc
1 839
2 861
3 845
Mean: 848.3
This is encouraging. But I will have to run repetitive tests as well.
Given that we know my resource-constrained box is slow, I ask:
(a) Is there any subset of the tests run during 'make test' that you
think would be revelatory of time differences?
(b) How would I measure differences in memory usage between the two
branches (other than noting places where speed slows down simply because
memory is no longer available)?
Thank you very much.
kid51
_______________________________________________
http://lists.parrot.org/mailman/listinfo/parrot-dev
