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

Fast generate-implant-diffusion.py #5

Open
jamesavery opened this issue May 4, 2022 · 0 comments
Open

Fast generate-implant-diffusion.py #5

jamesavery opened this issue May 4, 2022 · 0 comments

Comments

@jamesavery
Copy link
Owner

Currently, the diffusion field generation (for use in the field histogramming) is written in Python using scipy.ndimage, and is very slow. But really, shouldn't it be fast? I believe that it should.
carljohnsen added a commit that referenced this issue May 4, 2022
@carljohnsen
Preliminary work on faster gaussian filter #5
e600be3
carljohnsen added a commit that referenced this issue May 5, 2022
@carljohnsen
Seems correct, but differs from ndimage #5
e2bd556
carljohnsen added a commit that referenced this issue May 5, 2022
@carljohnsen
It works! #5
ac75982
carljohnsen added a commit that referenced this issue May 5, 2022
@carljohnsen
Works, and is faster! #5
1e8a6cd
carljohnsen added a commit that referenced this issue May 5, 2022
@carljohnsen
Fast and small, and still correct! #5
cd66f4d
carljohnsen added a commit that referenced this issue May 5, 2022
@carljohnsen
#5 works without any odd errors
5c3b831
carljohnsen added a commit that referenced this issue May 5, 2022
@carljohnsen
It works and is fast! #5
2416c66
carljohnsen added a commit that referenced this issue May 8, 2022
@carljohnsen
Seems close enough #5
4f1755c
carljohnsen added a commit that referenced this issue May 9, 2022
@carljohnsen
Forgot to include the "glowing" implant #5
b8f8656
carljohnsen added a commit that referenced this issue May 9, 2022
@carljohnsen
Cleanup on aisle gauss diffusion #5
71bad2a
carljohnsen referenced this issue Mar 7, 2024
@carljohnsen
#9 Implemented first steps for diffusion based of hippo paper work
45228c4
carljohnsen added a commit that referenced this issue Mar 7, 2024
@carljohnsen
#5 CPU version is faster if guarded by loop ranges
b41402e
carljohnsen added a commit that referenced this issue Mar 7, 2024
@carljohnsen
#5 The timing of ndi should include conversions
d324edc
carljohnsen added a commit that referenced this issue Mar 7, 2024
@carljohnsen
#5 Added preliminary GPU version, which has branch-free critical loop
87b9da0
carljohnsen added a commit that referenced this issue Mar 7, 2024
@carljohnsen
#5 Added preliminary on-disk diffusion implementation
212a70a
carljohnsen added a commit that referenced this issue Mar 7, 2024
@carljohnsen
#5 the on-disk version seems to be working
3a7f106
carljohnsen added a commit that referenced this issue Mar 8, 2024
@carljohnsen
#5 gpu in memory seems to be working.
bf4d2d8
carljohnsen added a commit that referenced this issue Mar 8, 2024
@carljohnsen
#5 GPU on-disk version seems to be working
3274992
carljohnsen added a commit that referenced this issue Mar 13, 2024
@carljohnsen
#5 2x diffusion was too much to do in-memory for a 3080 GPU. (19 GB)
07f25cc
carljohnsen added a commit that referenced this issue Apr 11, 2024
@carljohnsen
#5 Removed redundant padding argument from diffusion_core
72d13f7
carljohnsen added a commit that referenced this issue Apr 11, 2024
@carljohnsen
#5 Added functions for uint16 as a result type
0bb6cd3
carljohnsen added a commit that referenced this issue Apr 11, 2024
@carljohnsen
#5 Changed diffusion to work with uint16 instead of uint8
d38fa70
carljohnsen added a commit that referenced this issue Apr 11, 2024
@carljohnsen
#5 Added progress printing
230359a
carljohnsen added a commit that referenced this issue Apr 11, 2024
@carljohnsen
#5 Doing pragma acc parallel loop on large loops creates memory viola…
101f049 
…tions.

Instead, parallelize over chunks of num_threads
carljohnsen added a commit that referenced this issue Apr 11, 2024
@carljohnsen
#5 More uint8 > uint16 related changes
c3cfb0c
carljohnsen added a commit that referenced this issue Apr 11, 2024
@carljohnsen
#5 out of memory for 1x when trying to use more than 1 gb of internal…
d5ab408 
… memory (due to extremely large gaussian kernel)
carljohnsen added a commit that referenced this issue Apr 11, 2024
@carljohnsen
#5 fixed bug in diffusion that would only show when running on 1x scale
b73c6f5
carljohnsen added a commit that referenced this issue Apr 11, 2024
@carljohnsen
#5 Found a bug when pre-creating the buffers in OpenACC
30026cf
carljohnsen added a commit that referenced this issue Apr 11, 2024
@carljohnsen
#5 Added down to scale 8 to use the on-disk diffusion for debugging p…
7cde694 
…urposes.
carljohnsen added a commit that referenced this issue Apr 11, 2024
@carljohnsen
#5 Moved plotting to before saving the diffusion field, to combat out…
dfd2b16 
… of memory when storing
carljohnsen added a commit that referenced this issue Jul 23, 2024
@carljohnsen
#5 Changed diffusion to use the new to_int
55cc66e
carljohnsen added a commit that referenced this issue Jul 23, 2024
@carljohnsen
#5 Added new default parameters to generate_gauss. 1x works now.
795c75e
carljohnsen added a commit that referenced this issue Jul 27, 2024
@carljohnsen
#5 Added x-axis caching edition of the 1d gauss kernel.
f5ecec4
carljohnsen added a commit that referenced this issue Jul 27, 2024
@carljohnsen
#5 Added original 1d gauss for comparison
877742a
carljohnsen added a commit that referenced this issue Jul 29, 2024
@carljohnsen
#5 Added y-axis specific gauss kernel
efccf9b
carljohnsen added a commit that referenced this issue Jul 29, 2024
@carljohnsen
#5 Optimized the y-specialized diffusion
fc90eaa
carljohnsen added a commit that referenced this issue Jul 29, 2024
@carljohnsen
#5 Introduced padding on the x-axis to be a multiple of veclen, to av…
adb78d5 
…oid having boundary checks in the critical loops.
carljohnsen added a commit that referenced this issue Jul 29, 2024
@carljohnsen
#5 Renamed loop variable i to tid in diffusion core y to better refle…
64816f2 
…ct that it's meant as thread id.
carljohnsen added a commit that referenced this issue Jul 29, 2024
@carljohnsen
#5 Renamed i,j,k to z,y,x in diffusion_core_x for consistency
b76c123
carljohnsen added a commit that referenced this issue Jul 29, 2024
@carljohnsen
#5 Made diffusion_core_y faster by increasing number of blocks
2a586b2
carljohnsen added a commit that referenced this issue Jul 29, 2024
@carljohnsen
#5 diffusion_core_x now exploits aligned x axis. Plus, increased the …
9707bbd 
…number of launched blocks.
carljohnsen added a commit that referenced this issue Jul 29, 2024
@carljohnsen
#5 Added diffusion_core_z
e144ec1
carljohnsen added a commit that referenced this issue Jul 29, 2024
@carljohnsen
#5 diffusion_core_x was incorrect, the indices were off.
c45f76e
carljohnsen added a commit that referenced this issue Jul 30, 2024
@carljohnsen
#5 Implemented the new diffusion core functions in the out-of-core im…
4e21798 
…plementation
carljohnsen added a commit that referenced this issue Aug 5, 2024
@carljohnsen
#5 Asynchronous conversion from uint8 to float
0158dfa
carljohnsen added a commit that referenced this issue Sep 9, 2024
@carljohnsen
#5 GPU diffusion was giving incorrect results.
63abddf
carljohnsen added a commit that referenced this issue Sep 9, 2024
@carljohnsen
#5 Handled correctly calling the optimized diffusion functions (radiu…
656865a 
…s < 16 (i.e. kernel <= 31))
carljohnsen added a commit that referenced this issue Sep 10, 2024
@carljohnsen
#5 Consistent veclen across functions.
8c90bdf 
Should be placed in a configuration somewhere, so that if it's updated, it's applied across.
carljohnsen added a commit that referenced this issue Sep 10, 2024
@carljohnsen
#5 Added padding for all dimensions, as this is a requirement for the…
ed3c47f 
… optimized 1d diffusions for each axis.
carljohnsen added a commit that referenced this issue Sep 10, 2024
@carljohnsen
#5 Async/multi-device WIP: translating the out of core implementation…
9d27dcf 
… into one that decomposes all axes, rather than just z.
carljohnsen added a commit that referenced this issue Sep 11, 2024
@carljohnsen
#5 Changed the OpenACC calls / directives for the out of core diffusion.
15aa220
carljohnsen added a commit that referenced this issue Sep 11, 2024
@carljohnsen
#5 Cleaned up old commented code.
809fe63
carljohnsen added a commit that referenced this issue Sep 11, 2024
@carljohnsen
#5 Changed from constexpr to const due ot acc_get_num_devices not bei…
d788b73 
…ng a constexpr
carljohnsen added a commit that referenced this issue Sep 11, 2024
@carljohnsen
#5 Added explicit shared/firstprivate clauses to the OpenMP section t…
f8f25b0 
…o ensure no false sharing.
carljohnsen added a commit that referenced this issue Sep 11, 2024
@carljohnsen
#5 Removed the async stream id argument, as the streams are now impli…
4299a97 
…cit inside the explicit threads.
carljohnsen added a commit that referenced this issue Sep 11, 2024
@carljohnsen
#5 Missing -fopenmp resulted in the OpenMP pragmas not being picked up.
f694357
carljohnsen added a commit that referenced this issue Sep 11, 2024
@carljohnsen
#5 Swap in OpenMP terms doesn't just swap the set of privately held p…
a7d8e21 
…ointers, they are swapped across threads.

This won't work when double buffering, since 1. they will be swapped num_threads times, and 2. is not thread safe.
carljohnsen added a commit that referenced this issue Sep 11, 2024
@carljohnsen
#5 Updated the automatic choosing of whether to use the in-memory or …
5d5741a 
…out-of-core diffusion implementation.
carljohnsen added a commit that referenced this issue Sep 11, 2024
@carljohnsen
#5 Added progress printing.
360321d
carljohnsen added a commit that referenced this issue Sep 11, 2024
@carljohnsen
#5 Fixed faulty copy back from staging area to main-memory copy in ou…
63aaeac 
…t of core diffusion.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
1 participant
@jamesavery