<sb0>
rjo, by the way the llvm execution engine is super slow here (used in the py2llvm unit tests) - is it the same for you?
<sb0>
it's actually slower than the python interpreter by a wide margin ...
<rjo>
sb0: Ran 4 tests in 0.127s here
<sb0>
5 tests in 1.239s here (with the new commits)
<GitHub113>
[ARTIQ] sbourdeauducq pushed 6 new commits to master: http://git.io/rBrLYw
<GitHub113>
ARTIQ/master dddc925 Sebastien Bourdeauducq: py2llvm: run passes before creating EE
<GitHub113>
ARTIQ/master bce687b Sebastien Bourdeauducq: py2llvm/base_types: implement 'not' on 'None'
<GitHub113>
ARTIQ/master 15dcf33 Sebastien Bourdeauducq: py2llvm: move GCD function into LLVM IR
<rjo>
Ran 5 tests in 0.281s
<rjo>
with numba i rarely see speed problems. only with some extremely large expressions.
<sb0>
ok... maybe JIT is broken somehow in my llvm setup
<rjo>
is that 1.2s on the netbook?
<sb0>
no, i7-3770 CPU @ 3.40GHz
<rjo>
ah.
<mithro>
in migen, how does one do something on the falling edge of a clock?
<sb0>
what do you need that for?
<GitHub117>
[migen] sbourdeauducq pushed 7 new commits to master: http://git.io/oIpzUA
<GitHub117>
migen/master 4def6ec Robert Jordens: flow/network: replace NetworkX MultiDiGraph with simple implementation
<GitHub117>
migen/master 7518a7b Robert Jordens: examples/dataflow: adapt to new simple MultiDiGraph implementation
<GitHub117>
migen/master 8489604 Robert Jordens: examples/dataflow/dma: fix simulation, run it for 100 cycles
<rjo>
sb0: cheers.
<rjo>
sb0: i tested that graph implementation only with the available examples, misoc and and a handful of my own use cases. it does not cover all the code, especially not the remove_{node,edge}.
<rjo>
but i guess since that was hardly used before any bugs can be fixed later ;). the code is also pretty trivial.
<GitHub78>
[ARTIQ] sbourdeauducq pushed 1 new commit to master: http://git.io/vhFyew