Closed 1uc closed 2 weeks ago
✔️ 2d75d7e1027e47ca736471761bfc05f1c0965fa8 -> Azure artifacts URL
All modified and coverable lines are covered by tests :white_check_mark:
Project coverage is 67.22%. Comparing base (
8cb9ec8
) to head (a350cee
). Report is 4 commits behind head on master.:exclamation: Current head a350cee differs from pull request most recent head 0b45c4d
Please upload reports for the commit 0b45c4d to get more accurate results.
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
Logfiles from GitLab pipeline #216817 (:white_check_mark:) have been uploaded here!
Status and direct links:
@nrnhines @pramodk could you please check if it at least partly solves the performance issue?
Here are the numbers:
I am using https://github.com/nrnhines/266806/ and with single thread . I am using Morphology_1/mod_files/cdp5StCmod.mod
without https://github.com/nrnhines/266806/commit/b3815a0b3df60bf6d9a5aadb843a4e900c35748f i.e. use diam
directly :
## Master Branch
NEURON RUN took 30.94808864593506
Path Min time/rank Max time/rank Avg time/rank Time %
state-cdp5StCmod 18.409768 18.409768 18.409768 59.448159
## This PR
NEURON RUN took 21.428975582122803
Path Min time/rank Max time/rank Avg time/rank Time %
state-cdp5StCmod 9.139341 9.139341 9.139341 42.610451
And this time is similar to the one if we cache diam
as a RANGE variable (as done in https://github.com/nrnhines/266806/commit/b3815a0b3df60bf6d9a5aadb843a4e900c35748f).
While implementing the same solution in NMODL, I noticed, that calling function from Python leads to a SEGFAULT. I'm debugging, but more changes are needed for this PR.
This has now been fixed, essentially the m_offset
for MechanismInstance
was incorrect for dptr_field
.
✔️ 7cb0affc9b9ae614e89069feb79be070298c25e4 -> Azure artifacts URL
Logfiles from GitLab pipeline #216999 (:white_check_mark:) have been uploaded here!
Status and direct links:
@pramodk It now contains a fix for the SEGFAULT, I think it's principled, but it could use re-review. Sorry, CI seems to have gotten stuck, I'll rerun.
Logfiles from GitLab pipeline #217187 (:white_check_mark:) have been uploaded here!
Status and direct links:
Logfiles from GitLab pipeline #217276 (:white_check_mark:) have been uploaded here!
Status and direct links:
✔️ a350ceec411f51b28c489e753ea232b44f2d4277 -> Azure artifacts URL
@nrnhines : I have tested this and confirmed diam
related perf regression is addressed. Do you to take a quick look at the change?
Issues
0 New issues
0 Accepted issues
Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code
✔️ 0b45c4dcf3cf695145f6970a3bf2b43399241d84 -> Azure artifacts URL
Logfiles from GitLab pipeline #218223 (:no_entry:) have been uploaded here!
Status and direct links:
I have neurodamus people about Gitlab failures. As this CI was green and failures are completely unrelated, I think we should skip/ignore gitlab CI here.
See #2787.
Optimizes
nocmodl
generated code to use the mechanism cache when accessingdiam
andarea
as described in #2913.For performance testing I use:
The name
two_radii
is to be able to grep fordiam
, we added the ionca
to compare with how it's solved for ions. The logic is to useinv
to create a very expensive multiplication with1.0
.The python file to measure the performance is:
Edit: I realized I was measuring with
Debug
. After switching toRelease
I needed to refine the measurement slightly (it now scales linearly withtstop
). The time after fix is5.6s
and before25s
.