Feature/variable resolution damage #68
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Weibull flaw distribution. Needs more testing.
for constructing Fields of random generators.
but untested. This also includes a number of infrastructure extensions to suppport this model, like FileIO support for Fields of unsigned and such.
generalizing our random number generator a bit.
problems with unsigned Fields.
… ProbabilisticDamageModel.
default minimum flaws per node.
in the ProbabilisticDamagePolicy, rather than regenerating a new realization each time. Should be *much* faster.
MishaZakharchanka
approved these changes
Jun 2, 2021
mdavis36
requested changes
Jun 2, 2021
There was a problem hiding this comment.
Overall looks good, I only have a small change request for the nullptr otherwise I noticed some comment blocks that I'm not sure were purposely left behind along with a point of discussion on policy traits.
The random generator looks good to me, this will certainly be something that might be a tricky point for GPU implementation in the future but we can approach that when we get there, I know there are specific libraries dedicated to random number generation on the GPU we will be able to utilize.
Comment on lines
+632
to
+633
| // VERIFY2(DBReadVar(mFilePtr, varname.c_str(), &value) == 0, | ||
| // "SiloFileIO ERROR: unable to read variable " << pathName); |
Comment on lines
+293
to
+294
| #[x for x in self._fields if isinstance(x, eval("UnsignedField%s" % self.dimension))] + | ||
| #[x for x in self._fields if isinstance(x, eval("ULLField%s" % self.dimension))]) |
Comment on lines
+116
to
+117
| // mGen.discard(n * mGen.state_size); | ||
| // mNumCalls += n; |
mdavis36
approved these changes
Jun 3, 2021
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This change is focused on updates to our damage models to make them more efficient and able to properly handle damage in the presence of varying initial resolution. There are two major algorithmic changes:
GradyKippTensorDamageOwen was an existing algorithm to handle this situation, but was way to slow during problem generation. I've implemented support for per point random number generators, and used this to more effectively parallelize problem setup of this model. It is now orders of magnitude faster in typical problems (minutes to setup rather than hours). However, this model still requires explicitly holding all the activation flaw strains, which for hi-res impact problems can be many billions and overflow memory on achievable machine allocations.
We have a new model (ProbabilisticTensorDamage) which implements a similar algorithm to GradyKippTensorDamageOwen, but based on statistics is able to represent the span of flaws on each point mathematically. This reduces our storage per point to a few doubles and some relatively quick math to update the damage when the implicit flaw distribution is activated. The results are similar to the older algorithm, run just as quickly, but require very little memory.
I have added regression tests of the new model in TensileRod-1d, and tested it out for 1d and 2d tensile rods, as well as hypervelocity cratering impact models.
The bulk of the files touched in this merge are just to support Fields of std::uniform_distribution random number generators.