-
Notifications
You must be signed in to change notification settings - Fork 6
Conversation
I think |
Codecov Report
@@ Coverage Diff @@
## master #19 +/- ##
==========================================
- Coverage 77.18% 77.16% -0.02%
==========================================
Files 8 8
Lines 434 438 +4
==========================================
+ Hits 335 338 +3
- Misses 99 100 +1
Continue to review full report at Codecov.
|
need to add a threshold to disable multi threading when the number of qubits is not large. Or should we just let user to decide this? since user can always close it by tweaking |
OK so it looks like the overhead of multi threads, should define a threshold to avoid them. https://gist.github.com/Roger-luo/26682142d70f2d24c8a8bea36518900f |
Cool, it is better to let user decide this. |
Please show the number of threads. It looks like always a lot of overhead. Lol |
It's in the benchmark report @GiggleLiu |
please check the GPU implementation for some inspiration of multi-threading. |
Codecov Report
@@ Coverage Diff @@
## master #19 +/- ##
==========================================
- Coverage 80.82% 79.47% -1.36%
==========================================
Files 8 8
Lines 558 570 +12
==========================================
+ Hits 451 453 +2
- Misses 107 117 +10
Continue to review full report at Codecov.
|
seems to have memory leak on Linux (works fine on Mac for me). Might related to JuliaLang/julia#31923 We need to wait for this issue to be fixed I think. |
I think it is fixed on master now: JuliaLang/julia#32217 |
we need more benchmark reports before we ship this feature. |
Any new benchmark reports? |
Not yet. I'll have time on this after JuliaCon |
I think if we merge this (after 1.3 is out), but we might need to ask our users to use julia 1.3 for best performance. (I'm not sure if thee new multi-threading support will be backported) |
Some updates on this. The I suspect we have some cache line problem when enabling multi-threading, the scaling is not good. |
We need to refactor this branch after #30 @GiggleLiu and I need to explore some more benchmark on cache line problem. Hopefully we can deliver this feature this year. |
@GiggleLiu This looks good now. I think we could merge this first. I'll make a few following PR to fix some performance issue later. |
Add multi threading support. Pending for benchmark results.