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.

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

Add LCAO performance tests #1639

Merged
merged 7 commits into from
Jun 18, 2019
Merged

Add LCAO performance tests #1639

merged 7 commits into from
Jun 18, 2019

Conversation

Hyeondeok-Shin
Copy link
Contributor

These benchmarks are VMC and DMC runs for carbon molecules in order to measure LCAO performance for various system sizes. The simulated systems are various sizes of a carbon ring (C12, C18 ,C24, and C30) and C60 buckyball. All-electron tests (ae) with the cusp correction and tests with a pseudopotential (pp) are included in carbon ring tests.

Name Atoms Electrons Electrons per spin
C12 12 48(pp), 72(ae) 24(pp),36(ae)
C18 18 72(pp),108(ae) 36(pp),54(ae)
C24 24 96(pp),144(ae) 48(pp),72(ae)
C30 30 120(pp),180(ae) 60(pp),90(ae)
C60 60 240(pp) 120(pp)

More details (h5 files location, etc.) can be found in README.

Hyeondeok Shin

@qmc-robot
Copy link

Can one of the maintainers verify this patch?

@prckent prckent changed the title add LCAO performance tests Add LCAO performance tests Jun 16, 2019
@prckent
Copy link
Contributor

prckent commented Jun 16, 2019

OK to test

Copy link
Contributor

@prckent prckent left a comment

Choose a reason for hiding this comment

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

Thanks.

  1. Lets put all the .h5 files in the same directory/URL on Box, including the legacy CORAL graphite and our existing NiO spline and AFQMC files. I suggest a new name such as QMCPACK_Performance_Inputs or similar. (We need a host with a better user experience.) @ye-luo is the current files owner.
  2. How long are these expected to take to run?

Download the necessary C-molecule h5 orbital files of different sizes from
the following link

https://anl.box.com/s/pveyyzrc2wuvg5tmxjzzwxeo561vh3r0
Copy link
Contributor

Choose a reason for hiding this comment

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

Copy link
Contributor

Choose a reason for hiding this comment

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

Change of strategy. Single URL for all files from now on.

for details of the current tests and input files and to download them.

Currently the files must be downloaded via\\
\url{https://anl.box.com/s/pveyyzrc2wuvg5tmxjzzwxeo561vh3r0} and \\
\url{https://anl.box.com/s/j5d6jeazpgx5441s04ajtv77iogoyjsh}.
\url{https://anl.box.com/s/j5d6jeazpgx5441s04ajtv77iogoyjsh} (C-graphite file).

Copy link
Contributor

Choose a reason for hiding this comment

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

Please add a link to C h5 files.

Copy link
Contributor

Choose a reason for hiding this comment

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

See comment below. After updating we should only have a single unique box url.

@ye-luo
Copy link
Contributor

ye-luo commented Jun 17, 2019

Please remove C.BFD.xml and use softlink instead. There is a copy at tests/pseudopotentials_for_tests

@prckent
Copy link
Contributor

prckent commented Jun 17, 2019

@ye-luo We should use a single URL for all our files from now on. This is simpler and scales better. Please work with @Hyeondeok-Shin to update appropriately. (Will reconsider if there are issues with Box.com).

@prckent
Copy link
Contributor

prckent commented Jun 17, 2019

Confirming that these tests run OK (oxygen, PGI2019 compiler with MKL). We might consider more steps per block in future since the smaller systems are perhaps too fast. There are some interesting timing differences between ae and pp runs, but we should check more runs, compilers & platforms before drawing conclusions.

$ grep molec ctest.out
      Start  88: performance-C-molecule-C12-e72-ae-1-16
39/74 Test  #88: performance-C-molecule-C12-e72-ae-1-16 ....................   Passed    1.38 sec
      Start  89: performance-C-molecule-C12-e72-ae-1-16-time
40/74 Test  #89: performance-C-molecule-C12-e72-ae-1-16-time ...............   Passed    0.02 sec
      Start  90: performance-C-molecule-C12-e72-ae-J3-1-16
41/74 Test  #90: performance-C-molecule-C12-e72-ae-J3-1-16 .................   Passed    1.54 sec
      Start  91: performance-C-molecule-C12-e72-ae-J3-1-16-time
42/74 Test  #91: performance-C-molecule-C12-e72-ae-J3-1-16-time ............   Passed    0.02 sec
      Start  92: performance-C-molecule-C12-e48-pp-1-16
43/74 Test  #92: performance-C-molecule-C12-e48-pp-1-16 ....................   Passed    0.37 sec
      Start  93: performance-C-molecule-C12-e48-pp-1-16-time
44/74 Test  #93: performance-C-molecule-C12-e48-pp-1-16-time ...............   Passed    0.02 sec
      Start  94: performance-C-molecule-C12-e48-pp-J3-1-16
45/74 Test  #94: performance-C-molecule-C12-e48-pp-J3-1-16 .................   Passed    0.45 sec
      Start  95: performance-C-molecule-C12-e48-pp-J3-1-16-time
46/74 Test  #95: performance-C-molecule-C12-e48-pp-J3-1-16-time ............   Passed    0.03 sec
      Start  96: performance-C-molecule-C18-e108-ae-1-16
47/74 Test  #96: performance-C-molecule-C18-e108-ae-1-16 ...................   Passed    4.33 sec
      Start  97: performance-C-molecule-C18-e108-ae-1-16-time
48/74 Test  #97: performance-C-molecule-C18-e108-ae-1-16-time ..............   Passed    0.02 sec
      Start  98: performance-C-molecule-C18-e108-ae-J3-1-16
49/74 Test  #98: performance-C-molecule-C18-e108-ae-J3-1-16 ................   Passed    4.58 sec
      Start  99: performance-C-molecule-C18-e108-ae-J3-1-16-time
50/74 Test  #99: performance-C-molecule-C18-e108-ae-J3-1-16-time ...........   Passed    0.02 sec
      Start 100: performance-C-molecule-C18-e72-pp-1-16
51/74 Test #100: performance-C-molecule-C18-e72-pp-1-16 ....................   Passed    0.64 sec
      Start 101: performance-C-molecule-C18-e72-pp-1-16-time
52/74 Test #101: performance-C-molecule-C18-e72-pp-1-16-time ...............   Passed    0.02 sec
      Start 102: performance-C-molecule-C18-e72-pp-J3-1-16
53/74 Test #102: performance-C-molecule-C18-e72-pp-J3-1-16 .................   Passed    0.78 sec
      Start 103: performance-C-molecule-C18-e72-pp-J3-1-16-time
54/74 Test #103: performance-C-molecule-C18-e72-pp-J3-1-16-time ............   Passed    0.02 sec
      Start 104: performance-C-molecule-C24-e144-ae-1-16
55/74 Test #104: performance-C-molecule-C24-e144-ae-1-16 ...................   Passed   12.22 sec
      Start 105: performance-C-molecule-C24-e144-ae-1-16-time
56/74 Test #105: performance-C-molecule-C24-e144-ae-1-16-time ..............   Passed    0.02 sec
      Start 106: performance-C-molecule-C24-e144-ae-J3-1-16
57/74 Test #106: performance-C-molecule-C24-e144-ae-J3-1-16 ................   Passed   11.97 sec
      Start 107: performance-C-molecule-C24-e144-ae-J3-1-16-time
58/74 Test #107: performance-C-molecule-C24-e144-ae-J3-1-16-time ...........   Passed    0.02 sec
      Start 108: performance-C-molecule-C24-e96-pp-1-16
59/74 Test #108: performance-C-molecule-C24-e96-pp-1-16 ....................   Passed    1.08 sec
      Start 109: performance-C-molecule-C24-e96-pp-1-16-time
60/74 Test #109: performance-C-molecule-C24-e96-pp-1-16-time ...............   Passed    0.02 sec
      Start 110: performance-C-molecule-C24-e96-pp-J3-1-16
61/74 Test #110: performance-C-molecule-C24-e96-pp-J3-1-16 .................   Passed    1.30 sec
      Start 111: performance-C-molecule-C24-e96-pp-J3-1-16-time
62/74 Test #111: performance-C-molecule-C24-e96-pp-J3-1-16-time ............   Passed    0.02 sec
      Start 112: performance-C-molecule-C30-e180-ae-1-16
63/74 Test #112: performance-C-molecule-C30-e180-ae-1-16 ...................   Passed   28.05 sec
      Start 113: performance-C-molecule-C30-e180-ae-1-16-time
64/74 Test #113: performance-C-molecule-C30-e180-ae-1-16-time ..............   Passed    0.03 sec
      Start 114: performance-C-molecule-C30-e180-ae-J3-1-16
65/74 Test #114: performance-C-molecule-C30-e180-ae-J3-1-16 ................   Passed   28.94 sec
      Start 115: performance-C-molecule-C30-e180-ae-J3-1-16-time
66/74 Test #115: performance-C-molecule-C30-e180-ae-J3-1-16-time ...........   Passed    0.03 sec
      Start 116: performance-C-molecule-C30-e120-pp-1-16
67/74 Test #116: performance-C-molecule-C30-e120-pp-1-16 ...................   Passed    1.71 sec
      Start 117: performance-C-molecule-C30-e120-pp-1-16-time
68/74 Test #117: performance-C-molecule-C30-e120-pp-1-16-time ..............   Passed    0.02 sec
      Start 118: performance-C-molecule-C30-e120-pp-J3-1-16
69/74 Test #118: performance-C-molecule-C30-e120-pp-J3-1-16 ................   Passed    1.95 sec
      Start 119: performance-C-molecule-C30-e120-pp-J3-1-16-time
70/74 Test #119: performance-C-molecule-C30-e120-pp-J3-1-16-time ...........   Passed    0.03 sec
      Start 120: performance-C-molecule-C60-e240-pp-1-16
71/74 Test #120: performance-C-molecule-C60-e240-pp-1-16 ...................   Passed   11.00 sec
      Start 121: performance-C-molecule-C60-e240-pp-1-16-time
72/74 Test #121: performance-C-molecule-C60-e240-pp-1-16-time ..............   Passed    0.03 sec
      Start 122: performance-C-molecule-C60-e240-pp-J3-1-16
73/74 Test #122: performance-C-molecule-C60-e240-pp-J3-1-16 ................   Passed   11.80 sec
      Start 123: performance-C-molecule-C60-e240-pp-J3-1-16-time
74/74 Test #123: performance-C-molecule-C60-e240-pp-J3-1-16-time ...........   Passed    0.03 sec

@Hyeondeok-Shin
Copy link
Contributor Author

@prckent @ye-luo Now box URLs for h5 are replaced to the same upper folder for all performance tests. H5 file located sub-folders are using the name of a system in the tests, so users can easily find h5 files.
As @ye-luo pointed out, now a carbon PP is using the softlink.
Current C-molecule is using exactly the same # of steps and blocks with the NiO performance tests. I'll keep watching the performance tests and figure out how much these steps & blocks should increase if it needs.

@prckent prckent merged commit 79d2cdf into QMCPACK:develop Jun 18, 2019
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

4 participants