Yiannis Tsiouris

We should really keep the best out of N times and not some other statistics e.g. the average or the median...

We should update the benchmarking infrastructure to be able to compare ErLLVM runtime performance to Erjang and create the corresponding diagrams. XXX: This might not be so straightforward as Erjang...

Extend suite with more interesting benchmarks

1

We should select some of the existing benchmarks, the most interesting ones and search for other benchmarks that fit! Maybe we should consider (some of) [these](http://shootout.alioth.debian.org/u64q/compare.php?lang=erlang). Furthermore, a mandelbrot benchmark...

Some benchmarks run for a couple of seconds but others for a couple of minutes. I think we should consider inputs that trigger similar execution times; in that way, the...

Hail fails with FailN1 for an EXT instance allocation in 2.16

15

Hey, A strange issue caught our attention when trying to update snf-ganeti to 2.16. Hail fails to allocate an instance with EXT disk template. This used to work in 2.15....

Feature kvm ifdown script

7

This patch-set adds support for an ifdown script that is responsible for deconfiguring network devices and cleanup changes made by the ifup script as described in the corresponding [design-doc](http://docs.ganeti.org/ganeti/2.15/html/design-ifdown.html). While...

Talking with [David Terei](http://www.scs.stanford.edu/~davidt/) convinced me that using Type-Based Alias Analysis (TBAA) in LLVM could enable various optimization transformations; the LLVM CodeGen does not apply them because it doesn't know...

Implement a liveness analysis on RTL variables (only tagged) in order to reduce LLVM stack usage. Each RTL tagged var is mapped explicitly to a stack slot (with 'alloca') because...