Open mottelet opened 1 month ago
otaolafranc
commented
1 month ago Yes, this is a correct ouput, you can visualize the output using openFlipper https://gitlab.vci.rwth-aachen.de:9000/OpenFlipper-Free/OpenFlipper-Free which supports the ovm output format. the mesh output (cylinder_hex.ovm) should look like this (if everything went smoothly, I have faced problems where algoHex was compiled, but the output was not correct):
mottelet
commented
1 month ago I looks like this in Scilab:
mheistermann
commented
1 month ago For reference, here's some timing output from my M1 MacBook Pro (built with default settings, i.e., clang in Release mode):
Total time 60485 ms, n = 1
├ Field generation 3201 ms, n = 3
│ ├ SpH Projection 539 ms, n = 81
│ ├ Linear system setup 1247 ms, n = 81
│ │ ├ Smoothness term 813 ms, n = 81
│ │ ├ Normal constraints 83 ms, n = 81
│ │ ├ Full constraints 0 ms, n = 81
│ │ ├ Local linearisation 340 ms, n = 80
│ │ ╰ (unaccounted) 10 ms
│ ├ Linear system solve 1371 ms, n = 81
│ ╰ (unaccounted) 42 ms
├ Singular graph extraction 1217 ms, n = 1
│ ├ Refine singular region 324 ms, n = 1
│ ├ Interpolate for cell quaternions 867 ms, n = 1
│ ├ Extract singular edges 25 ms, n = 1
│ ╰ (unaccounted) 0 ms
├ Locally hexmeshable field generation 10760 ms, n = 1
│ ├ Singular graph repair 320 ms, n = 32
│ ├ Singularity relocation 6874 ms, n = 32
│ │ ├ Split edges 516 ms, n = 32
│ │ ├ Collapse edges 1054 ms, n = 32
│ │ ├ Swap edges 2360 ms, n = 32
│ │ ├ Smooth singular graph 2620 ms, n = 32
│ │ ├ Split for field alignment DOF 158 ms, n = 32
│ │ ╰ (unaccounted) 165 ms
│ ├ Check for termination 1927 ms, n = 1
│ ├ Optimize field locally 1458 ms, n = 32
│ ├ Optimize field with alignment 186 ms, n = 2
│ ├ Save mesh for visualization 14 ms, n = 33
│ ╰ (unaccounted) -19 ms
├ Post remeshing 15377 ms, n = 1
├ FrameField Integrability Optimization 13912 ms, n = 6
├ FrameField Dual Integrability Optimization 0 ms, n = 0
├ Parameterization 11073 ms, n = 2
├ Hexmesh extraction 4525 ms, n = 1
╰ (unaccounted) 416 ms
Thanks. Does the code uses multithreading ?
S.
mheistermann
commented
1 month ago There is a bit of multithreading in some places using OpenMP, more may be in the ipopt solve depending on backends used. The build that produced the numbers above did not have OpenMP enabled.
What kind of hardware are you running on, and what compiler are you using? Note that a non-release build (e.g. CMAKE_BUILD_TYPE empty or Debug) is significantly slower. For single-configuration cmake generators (e.g. makefiles), Release is default, but if you're using something like MSVC, you may have to manually select a release build in the GUI.
If it helps I can run this on a x86_64 desktop system to get some more easily comparable numbers.
mottelet
commented
1 month ago In the logs I saw that IPopt used mumps, although I had to build it with the coin-hsl library (otherwise AlgoHex does not build). I have compiled AlgoHex on a Linux (CentOS) server with 2 AMD EPYC 7552 48-Core Processor with gcc.
S.
mheistermann
commented
1 month ago Hi! I ran it on the most comparable server we have, with 2 EPYC 7742. For reproducibility, I built it using the Dockerfile in the dev/dockerfile-without-gurobi branch.
#####Hexmesh volume / tetmesh voluem: 9286.71/9314.11 0.997058
Total time 88865 ms, n = 1
├ Field generation 7214 ms, n = 3
│ ├ SpH Projection 49 ms, n = 81
│ ├ Linear system setup 3015 ms, n = 81
│ │ ├ Smoothness term 1904 ms, n = 81
│ │ ├ Normal constraints 210 ms, n = 81
│ │ ├ Full constraints 0 ms, n = 81
│ │ ├ Local linearisation 875 ms, n = 80
│ │ ╰ (unaccounted) 23 ms
│ ├ Linear system solve 3877 ms, n = 81
│ ╰ (unaccounted) 271 ms
├ Singular graph extraction 588 ms, n = 1
│ ├ Refine singular region 504 ms, n = 1
│ ├ Interpolate for cell quaternions 25 ms, n = 1
│ ├ Extract singular edges 58 ms, n = 1
│ ╰ (unaccounted) 0 ms
├ Locally hexmeshable field generation 14829 ms, n = 1
│ ├ Singular graph repair 462 ms, n = 32
│ ├ Singularity relocation 8865 ms, n = 32
│ │ ├ Split edges 754 ms, n = 32
│ │ ├ Collapse edges 1468 ms, n = 32
│ │ ├ Swap edges 2971 ms, n = 32
│ │ ├ Smooth singular graph 3289 ms, n = 32
│ │ ├ Split for field alignment DOF 211 ms, n = 32
│ │ ╰ (unaccounted) 170 ms
│ ├ Check for termination 2330 ms, n = 1
│ ├ Optimize field locally 2883 ms, n = 32
│ ├ Optimize field with alignment 382 ms, n = 2
│ ├ Save mesh for visualization 23 ms, n = 33
│ ╰ (unaccounted) -118 ms
├ Post remeshing 19540 ms, n = 1
├ FrameField Integrability Optimization 25712 ms, n = 6
├ FrameField Dual Integrability Optimization 0 ms, n = 0
├ Parameterization 17018 ms, n = 2
├ Hexmesh extraction 3262 ms, n = 1
╰ (unaccounted) 699 msOutput from /usr/bin/time -v (information on parallelism):
Command being timed: "/usr/local/bin/HexMeshing -i /app/demo/HexMeshing/cylinder.ovm"
User time (seconds): 1795.91
System time (seconds): 57.68
Percent of CPU this job got: 2071%
Elapsed (wall clock) time (h:mm:ss or m:ss): 1:29.48
Average shared text size (kbytes): 0
Average unshared data size (kbytes): 0
Average stack size (kbytes): 0
Average total size (kbytes): 0
Maximum resident set size (kbytes): 657108
Average resident set size (kbytes): 0
Major (requiring I/O) page faults: 0
Minor (reclaiming a frame) page faults: 464244
Voluntary context switches: 2522
Involuntary context switches: 8954505
Swaps: 0
File system inputs: 0
File system outputs: 3544
Socket messages sent: 0
Socket messages received: 0
Signals delivered: 0
Page size (bytes): 4096
Exit status: 0
Here's what I got just now:
#####Hexmesh volume / tetmesh voluem: 9288.21/9315.11 0.997112
Total time 311114 ms, n = 1
├ Field generation 70001 ms, n = 3
│ ├ SpH Projection 1967 ms, n = 81
│ ├ Linear system setup 7641 ms, n = 81
│ │ ├ Smoothness term 4876 ms, n = 81
│ │ ├ Normal constraints 582 ms, n = 81
│ │ ├ Full constraints 2 ms, n = 81
│ │ ├ Local linearisation 2106 ms, n = 80
│ │ ╰ (unaccounted) 74 ms
│ ├ Linear system solve 59991 ms, n = 81
│ ╰ (unaccounted) 401 ms
├ Singular graph extraction 1207 ms, n = 1
│ ├ Refine singular region 990 ms, n = 1
│ ├ Interpolate for cell quaternions 56 ms, n = 1
│ ├ Extract singular edges 160 ms, n = 1
│ ╰ (unaccounted) 0 ms
├ Locally hexmeshable field generation 58354 ms, n = 1
│ ├ Singular graph repair 2395 ms, n = 32
│ ├ Singularity relocation 31850 ms, n = 32
│ │ ├ Split edges 2668 ms, n = 32
│ │ ├ Collapse edges 4780 ms, n = 32
│ │ ├ Swap edges 11869 ms, n = 32
│ │ ├ Smooth singular graph 11049 ms, n = 32
│ │ ├ Split for field alignment DOF 756 ms, n = 32
│ │ ╰ (unaccounted) 725 ms
│ ├ Check for termination 6620 ms, n = 1
│ ├ Optimize field locally 16293 ms, n = 32
│ ├ Optimize field with alignment 1215 ms, n = 2
│ ├ Save mesh for visualization 112 ms, n = 33
│ ╰ (unaccounted) -133 ms
├ Post remeshing 63788 ms, n = 1
├ FrameField Integrability Optimization 62949 ms, n = 6
├ FrameField Dual Integrability Optimization 0 ms, n = 0
├ Parameterization 45502 ms, n = 2
├ Hexmesh extraction 7346 ms, n = 1
╰ (unaccounted) 1965 ms
save json data file into
Closed: Tue Jun 18 15:26:33 2024
Command being timed: "./Build/bin/HexMeshing -i ../demo/HexMeshing/cylinder.ovm -o ./cylinder-hex.ovm"
User time (seconds): 5889.53
System time (seconds): 17.64
Percent of CPU this job got: 1894%
Elapsed (wall clock) time (h:mm:ss or m:ss): 5:11.85
Average shared text size (kbytes): 0
Average unshared data size (kbytes): 0
Average stack size (kbytes): 0
Average total size (kbytes): 0
Maximum resident set size (kbytes): 775736
Average resident set size (kbytes): 0
Major (requiring I/O) page faults: 0
Minor (reclaiming a frame) page faults: 614154
Voluntary context switches: 473543
Involuntary context switches: 119049
Swaps: 0
File system inputs: 0
File system outputs: 3656
Socket messages sent: 0
Socket messages received: 0
Signals delivered: 0
Page size (bytes): 4096
Exit status: 0
Hello,
I have succeeded building algoHex and I have the following output/elapsed time for the proposed example in the build instructions (./Build/bin/HexMeshing -i ../demo/HexMeshing/cylinder.ovm -o /path/to/cylinder_hex.ovm). Soes it seem ok ? How can I visualize cylinder_hex.ovm once the computations are done ?