DO NOT MERGE: Experiment for TaskletFusion in Intra-Map Optimization #2454
Conversation
This file contains hidden or 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
NOT WORKING: 5.7612245082855225
philip-paul-mueller
force-pushed
the
better_in_map_dataflow_optimization_order
branch
from
831fb2d to
9cade71
Compare
DOES NOT WORK: 5.77982020
philip-paul-mueller
force-pushed
the
better_in_map_dataflow_optimization_order
branch
from
bcb3fd5 to
41385a5
Compare
NOT WORKING: 5.89192s
philip-paul-mueller
force-pushed
the
better_in_map_dataflow_optimization_order
branch
from
41385a5 to
8a389ee
Compare
SEEMS WORKING: 4.57165s
philip-paul-mueller
force-pushed
the
better_in_map_dataflow_optimization_order
branch
from
9e499a4 to
244dc10
Compare
…tion. But it also has an additional simplify that was present when TF was run in stage 1, but not in the other version. PERFORMANCE: 4.5106589s
philip-paul-mueller
force-pushed
the
better_in_map_dataflow_optimization_order
branch
from
eb6ebf3 to
ac2c5ce
Compare
DOES NOT WORK: 5.90685105s
philip-paul-mueller
force-pushed
the
better_in_map_dataflow_optimization_order
branch
from
9837be5 to
b68bae5
Compare
This reverts commit b68bae5.
philip-paul-mueller
force-pushed
the
better_in_map_dataflow_optimization_order
branch
from
3347571 to
e05939e
Compare
The inliner ignores Tasklets with empty Memlets and the PR also moved the `TaskletIntoMap` before the TF. This is like the version that did not worked, i.e. was super fast, but with the new TF we should recover it. WORKING: 4.604942321777344s
philip-paul-mueller
force-pushed
the
better_in_map_dataflow_optimization_order
branch
from
e05939e to
aa6afa2
Compare
This reverts commit aa6afa2.
philip-paul-mueller
force-pushed
the
better_in_map_dataflow_optimization_order
branch
from
2faa11d to
c404174
Compare
… Tasklet fusion again. I have noticed that in the normal/fast version the Tasklets with the literals are not inlined into other Tasklets. Now let's do that. DOES NOT WORK: 6.070111036300659s
philip-paul-mueller
force-pushed
the
better_in_map_dataflow_optimization_order
branch
from
b417158 to
cef7ebb
Compare
…sion run Tasklet fusion again." This reverts commit c404174. WORKS: 4.4253344535s
…fa3b1b754e`. But this time it is run right after the inlining. I think there will be a next iteration. DOWS NOT WORK: 6.1715042
philip-paul-mueller
force-pushed
the
better_in_map_dataflow_optimization_order
branch
from
a7435da to
eb8c62a
Compare
…he currently disacbled code and see if it is slow. As expected it is kind of fast. WORKS: 4.620257
philip-paul-mueller
force-pushed
the
better_in_map_dataflow_optimization_order
branch
from
eb8c62a to
2cf8002
Compare
I would have expected that it is slow now but it is not. WORKS: 4.66531s
philip-paul-mueller
force-pushed
the
better_in_map_dataflow_optimization_order
branch
from
2cf8002 to
1047b06
Compare
This seems to do the trick. If between the `MoveDataflowIntoIf` and the second `TaskletFusion` is a call to simplify, it is slow. DOES NOT WORK: 6.2248911s
philip-paul-mueller
force-pushed
the
better_in_map_dataflow_optimization_order
branch
2 times, most recently
from
1c769ea to
ebf9ce7
Compare
…somehow executed in a wrong way. In that experiment I wanted to run TF a second time after everything is done. It was fast, which puzzeled me, because it contradicted by previous findings. However, I looked at the archived SDFG and at least that thing showed that TF was not run a second time. Thus I have decided to redo that part, so now let's run TF for a second time and see what happens. SLOW: 6.137449s
philip-paul-mueller
force-pushed
the
better_in_map_dataflow_optimization_order
branch
from
5812c74 to
4ada9a6
Compare
…sklets with zero input degree and the SDFG must not be the top level one, i.e. real constant tasklet, that are not at the top level. WORKS: 4.68273s
philip-paul-mueller
force-pushed
the
better_in_map_dataflow_optimization_order
branch
from
4ada9a6 to
967c2a0
Compare
…ly ignoreing empty memlet (which the previous experiment should have been done), we only consider them. SLOW: 6.658746s
philip-paul-mueller
force-pushed
the
better_in_map_dataflow_optimization_order
branch
2 times, most recently
from
0fa4c80 to
5dd1a69
Compare
…ll that have empty memlets. FAST: 4.5654s
philip-paul-mueller
force-pushed
the
better_in_map_dataflow_optimization_order
branch
from
19e69b9 to
cef303d
Compare
… that are nested, but in a more liberal way. FAST: 4.63308429
philip-paul-mueller
changed the title
philip-paul-mueller
changed the title
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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 is for experimental purposes.
Can be force pushed any time.
Instead you should use PR#2457 that is made to be public.