[REVIEW] Add indirection and replace algorithms with new renumbering#1484
Conversation
rlratzel
added
improvement
Codecov Report
@@ Coverage Diff @@
## branch-0.19 #1484 +/- ##
===============================================
- Coverage 60.72% 60.35% -0.38%
===============================================
Files 70 70
Lines 3132 3143 +11
===============================================
- Hits 1902 1897 -5
- Misses 1230 1246 +16
Continue to review full report at Codecov.
|
Iroy30
changed the title
| personalization, | ||
| nstart): | ||
| wid = Comms.get_worker_id(sID) | ||
| print("RANK: ", wid) |
There was a problem hiding this comment.
Do we want these print statements left here?
There was a problem hiding this comment.
No. Will remove them , thanks for catching
seunghwak
left a comment
seunghwak
left a comment
There was a problem hiding this comment.
What change should I make to call the C++ renumber code in both Pascal & non-Pascal architectures? (I am planning to update the C++ renumber code to use a binary search based approach for Pascal).
And once renumber_map is returned, currently, unrenumbering step is purely python and differences in partitioning mechanisms shouldn't affect the unrenumbering process, correct?
| def call_renumber(sID, | ||
| data, | ||
| num_edges, | ||
| is_mnmg, |
There was a problem hiding this comment.
In case of single node multi-gpu, shouldn't this be false? If this should be true in single-node multi-gpu, isn't this a misnomer?
Please disregard my question if we are not supporting Pascal only in the single-GPU mode (and I am assuming we're not using the C++ renumbering code on Pascal). |
|
@gpucibot merge |
7 participants
No description provided.