search

gunrock / io

Input (scripts, etc.) and output (scripts, performance results, etc.) for Gunrock and other graph engines
10 stars 6 forks source link

v1+ (and if possible for older versions too, mainly 0.4+) avg-process-time edges-visited, nodes-visited vs. num-edges, num-nodes #47

Closed jowens closed 4 years ago

jowens commented 4 years ago

edges-visited, nodes-visited vs. num-edges, num-nodes For all datasets (so, the legend here is the dataset): How much "stuff" was inputted (x-axis?) vs. how much "stuff" was actually processed (y-axis?). Note, for PR you may need to use edges-/nodes-queued instead of visited.

jowens commented 4 years ago

@neoblizz Not sure these are super meaningful. Perhaps some good debug info though!

gunrock_primitives_all_vertices-visited-vs-num-vertices gunrock_primitives_all_edges-visited-vs-num-edges

jowens commented 4 years ago

This is what a PDF attachment looks like (not inline), btw.

gunrock_primitives_all_vertices-visited-vs-num-vertices.pdf gunrock_primitives_all_edges-visited-vs-num-edges.pdf

jowens commented 4 years ago

V100 only, probably preferable (cleaner).

gunrock_primitives_all-V100_vertices-visited-vs-num-vertices gunrock_primitives_all-V100_edges-visited-vs-num-edges

neoblizz commented 4 years ago

All of the last few graphs are useful for me, is there a reason why there is no bfs?

jowens commented 4 years ago

This was under your category of "PR, SSSP, BFS and BC".

neoblizz commented 4 years ago

I am confused.

jowens commented 4 years ago

OK, when you said title "Figures for PR, SSSP, BFS and BC.", category "v1+ (and if possible for older versions too, mainly 0.4+)" at the top of https://github.com/gunrock/gunrock/issues/725, you meant also include BFS.

jowens commented 4 years ago

And now I see "Figures for PR, SSSP, BFS and BC" actually already includes "BFS", sooooo I'm a numbskull.

neoblizz commented 4 years ago

OK, when you said title "Figures for PR, SSSP, BFS and BC.", category "v1+ (and if possible for older versions too, mainly 0.4+)" at the top of gunrock/gunrock#725, you meant also include BFS.

Yes, and then I made more comments that were algorithm specific plots.

neoblizz commented 4 years ago

After BFS, we can push these.

jowens commented 4 years ago

OK, I'll work on this in the morning and have a whole 'nother set for you to look at. (I have to do some data cleanup, there's some zeroes I have to track down.)

jowens commented 4 years ago

And now with BFS!

gunrock_primitives_all-V100_vertices-visited-vs-num-vertices gunrock_primitives_all-V100_edges-visited-vs-num-edges

neoblizz commented 4 years ago

Ok! Oh, and also can we separate bfs from dobfs? Just consider dobfs as a separate application.

jowens commented 4 years ago

Do we even have that distinction in 1.0+ (bfs vs dobfs)?

jowens commented 4 years ago

I think everything is named "bfs" at this point (the only bfs we have is dobfs).

jowens commented 4 years ago

gunrock_primitives_all-V100_vertices-visited-vs-num-vertices gunrock_primitives_all-V100_edges-visited-vs-num-edges

neoblizz commented 4 years ago

I think everything is named "bfs" at this point (the only bfs we have is dobfs).

dobfs is not a separate application but a flag, but what I was asking if if you could detect that flag, and name it a separate app called dobfs.

bfs (direction-optimized=false)
bfs (direction-optimized=true) -> rename to dobfs
bc
pr
sssp
jowens commented 4 years ago

gunrock_primitives_all-V100_vertices-visited-vs-num-vertices gunrock_primitives_all-V100_edges-visited-vs-num-edges

neoblizz commented 4 years ago

They look great! Push these please.

jowens commented 4 years ago

Are you cool with what we have now?