Feature: graph representation of the grid #776
Conversation
|
jenkins build this please |
michal-toth
changed the title
michal-toth
force-pushed
the
feature/graph-of-grid-second-try
branch
from
c8be590
to
dd7d098
Compare
|
jenkins build this please |
Contracting an edge joins two vertices into one. Their weight is added up, and the lists of edges is merged. For a common neighbor the edge weights are added up.
…tegration/implementation of wells. Note: Code debt: Zoltan-specific types are overwritten/hard-coded. Also wrappers have the type of grid hard-coded to CpGrid.
Each well is a set of IDs of its vertices. The graph contains only one ID (the smallest), but functions will now check wells if the sought ID is not in the graph itself. Wells are assumed disjoint, adding new well checks for intersections and possibly merges wells together. The check for intersections can be switched off to improve performance.
…ing remaining well cells
michal-toth
force-pushed
the
feature/graph-of-grid-second-try
branch
from
dd7d098
to
4cabe9a
Compare
Asking for ID not present in the graph or wells will now throw exception.
|
jenkins build this please |
|
jenkins build this serial please |
| std::sort(addToList.begin(),addToList.end(),[](const auto& a, const auto& b){return std::get<0>(a)<std::get<0>(b);}); | ||
| int origSize = cellList.size(); | ||
| int totsize = origSize+addToList.size(); | ||
| cellList.reserve(totsize); |
There was a problem hiding this comment.
I think we do know who many entries will be added in the function already at the beginning, right?
In that case we might be able to omit the extra container addToList and push_back to cellList directly.
I might be missing something here, though.
There was a problem hiding this comment.
The number of cells to be added to the cellList depends on which wells are in the list. Unless all wells are always in the list we do not know the final size beforehand.
|
There are some build failures (due to me, sorry): I think you need to explicitly include dune/istl/owneroverlapcopy.hh to fix this. |
I will add the include. However, I can not reproduce this error so I can not check if it helped. |
This is my second attempt to create a pull request after the first one got bugged (see #774 ). akva2 doubled it already (#775), so this one is to determine if I can create a working PR. The original description follows:
Graph representation of the grid.
Purpose: Loadbalancing of the grid with wells. Each well is represented with a single vertex in the graph repr. of the grid, therefore no well can be spread over multiple processes.
What is implemented so far:
GraphofGrid -class that stores the representation. Needs a grid (currently CpGrid) to build the graph. Has functions for searching for cells by their global IDs, adding wells, and vertex contraction.
Edge weights (transmissibilities) currently start equal to 1, and are possibly added up when contracted vertices have a common neighbor
wrappers for Zoltan callback functions (not tested with Zoltan yet though)
unit test for the above