Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Gpu widget new branch #380

Merged
merged 114 commits into from Dec 19, 2021
Merged

Gpu widget new branch #380

merged 114 commits into from Dec 19, 2021

Conversation

dcmvdbekerom
Copy link
Member

@dcmvdbekerom dcmvdbekerom commented Oct 18, 2021 •
edited

Update of the old GPU widget over at #274.

It is back in a working state after some merging issues, so we can now pick it up where we left off.

@dcmvdbekerom
add sliders to plot
3e60262
@dcmvdbekerom
make space for multiple sliders
4f7266d
@dcmvdbekerom
remove unused inputs
919d964
@dcmvdbekerom
Update gpu.py
7ba3937
@dcmvdbekerom
fix error in determination of min/max Lorentzian widths
26e0ce2
@dcmvdbekerom
Update gpu.py
faf4c7d 
cupy already caches the compiled CUDA code so we don't have to take care of this ourselves
@dcmvdbekerom
add lbl/gpu.cu to manifest
f703928
@dcmvdbekerom
comment out WIP for patch
221d3dd
@dcmvdbekerom
calculate S0 for GPU
124a9d3
@dcmvdbekerom
add sliders back in
748b290
@dcmvdbekerom
gpu implementation of apply_slit()
8303ecc
@dcmvdbekerom
remove superfluous normalization
edc9ee1
@dcmvdbekerom
zero arrays by fill(0), not np.zeros
3dc8101
@dcmvdbekerom
Simplify block allocation and calc log_2vMm on device
2f800a8 
- Block allocation is now trivial, so prepare_blocks() is no longer needed.
- Since iso is copied to device, log_2vMm can be calculated on device.
- Reduced iso from int32 to int8.
@dcmvdbekerom
Update gpu.cu
c92d98a
@dcmvdbekerom
Update gpu.cu
5c06fa0
@dcmvdbekerom
Update gpu.cu
4372d9d
@dcmvdbekerom
Update gpu.cu
c05825c
@dcmvdbekerom
CPU functions are compiled from CUDA code
63063cb 
This means that there is only one place where the physics is described, which can be used both by GPU and CPU code.
@dcmvdbekerom
agnosticized arrays, kernels, and fft's
ce5920b
@dcmvdbekerom
agnosticized constant memory params
9bb1f95
@dcmvdbekerom
Update gpu.py
a66a802
@dcmvdbekerom
working setup and complation
9aecf47
@dcmvdbekerom
fixed setting init and iter params
280b35d 
cpu/gpu agnostic code now working!
@dcmvdbekerom
Update gpu.cpp
9af0273
@dcmvdbekerom
add switches for gpu emulation
c67ea93
@dcmvdbekerom
remove debugging prints
20d9fd4
@dcmvdbekerom
Update MANIFEST.in
fb41181
@dcmvdbekerom
Merge branch 'develop' into gpu_widget
a8f64cd
@dcmvdbekerom
revert to old definition of greater
ee29b32
@erwanp erwanp added this to the 0.11.1 milestone Nov 14, 2021
@erwanp
Copy link
Member

erwanp commented Nov 14, 2021 •
edited

@dcmvdbekerom I'm delaying this for the next minor versino, which will be within days / 15 days at most, given the expected minor Doc updates / compat with Radis-App and Radis-Lab

@erwanp erwanp modified the milestones: 0.11.1, 0.12 Nov 27, 2021
erwanp
erwanp reviewed Dec 6, 2021
View reviewed changes
examples/plot_gpu_widgets.py Outdated

sf.fetch_databank("hitemp")

s = sf.eq_spectrum_gpu_explore(
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@dcmvdbekerom test fails because this should now be :

Suggested change
s = sf.eq_spectrum_gpu_explore(
s = sf.eq_spectrum_gpu_interactive(

@erwanp
Copy link
Member

erwanp commented Dec 18, 2021 •
edited

@dcmvdbekerom good to merge ? We could release it in 0.12

@erwanp erwanp modified the milestones: 0.13, 0.12 Dec 18, 2021
@dcmvdbekerom
Copy link
Member Author

I'll work on it tomorrow, I think it should be ready by then!

@dcmvdbekerom
Copy link
Member Author

There's a problem with the pressure slider; the keyword is pressure, but internally the value pressure_mbar is used. Can't we just use pressure everywhere?

@erwanp
Copy link
Member

erwanp commented Dec 19, 2021

Pressure should be better (it's an artifact of the very first lines of code of Radis), but renaming it requires to be cautious. We'll do it separately with a dedicated PR. Can you have a workaround in the meantime?

@dcmvdbekerom
Copy link
Member Author

@erwanp I also changed all occurances of DLM to LDM; let's make sure this went well everywhere.
Other than that it's ready to merge once it passes the tests

@erwanp
Revert "Auxiliary commit to revert individual files from 9a1d62a"
eef59f1 
This reverts commit c18b063f6c82dbaecc7fb674e765fa9a9ac04dc2.

(files got corrupted)
@erwanp
make emulate=True in plot_gpu_widget example (Travis-CI and ReadTheDo…
ef92b58 
…cs do not have GPU available). Fixes test

update error message if wrong mode
@erwanp
emulate GPU; fixes tests
da022fc
@erwanp
Copy link
Member

erwanp commented Dec 19, 2021

Tested, works well locally. I cloned your branch on Radis-Lab and tested from there too, works well (with emulated-GPU).

I let you merge ! 🎉

@dcmvdbekerom dcmvdbekerom merged commit cbeb2cc into radis:develop Dec 19, 2021
@erwanp erwanp mentioned this pull request Dec 19, 2021
Merged
@erwanp erwanp mentioned this pull request Apr 13, 2022
Merged
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@erwanp erwanp erwanp approved these changes

@anandxkumar anandxkumar Awaiting requested review from anandxkumar

@CorentinGrimaldi CorentinGrimaldi Awaiting requested review from CorentinGrimaldi

Assignees
No one assigned
Labels
gpu interface not related to the physics of the code performance
Projects
None yet
Milestone
0.12
Development

Successfully merging this pull request may close these issues.

None yet
3 participants
@dcmvdbekerom @codecov-commenter @erwanp