[MetaSchedule] Introduce `Union` and `OrderedUnion` in Database #12628

junrushao · 2022-08-28T00:44:56Z

Following up #12520 and #12626, this PR introduces two database classes:
UnionDatabase and OrderedUnionDatabase, both of which allow users to
organically compose multiple databases together, so that the high-level
IR (Relay, Relax) could select the best tuning records according to
running time or a preferred order given by users.

To each query, UnionDatabase returns the best record among all the
databases given; Instead, OrderedUnionDatabase returns he record from
the first database that responds to the query.

Used together, users may specify complicated dispatching patterns like
below:

Examples below demonstrate the usecases of and difference between
UnionDatabase and OrderDatabase.

Assumption:

db1, db2 do not have tuning records for the target workload.
Each of db3, db4, db5 has tuning records r3, r4, r5 for target
workload respectively.

#### Case 1. `UnionDatabase`:
merged_db = ms.database.UnionDatabase(
    db1, # no record
    db2, # no record
    db3, # has r3
    db4  # has r4
)
# returns the better one between r3 and r4
merged_db.query_tuning_record(..., target_workload)

### Case 2. `OrderedUnionDatabase`
merged_db = ms.database.OrderedUnionDatabase(
    db1, # no record
    db2, # no record
    db3, # has r3
    db4  # has r4
)
# returns r3
merged_db.query_tuning_record(..., target_workload)

### Case 3. Mix-use scenario
merged_db = ms.database.UnionDatabase(
    db1, # no record
    db2, # no record
    db3, # has r3
    ms.database.OrderedUnionDatabase( # returns r4
        db4,  # has r4
        db5,  # has r5
    )
)
# returns the better one between r3 and r4
merged_db.query_tuning_record(..., target_workload)

### Case 4. Another mix-use scenario
merged_db = ms.database.UnionDatabase(
    db1, # no record
    db2, # no record
    db3, # has r3
    ms.database.UnionDatabase( # returns the better one between r4 and r5
        db4,  # has r4
        db5,  # has r5
    )
)
# returns the best one among r3, r4 and r5
merged_db.query_tuning_record(..., target_workload)

### Case 5. Yet another mix-use scenario
merged_db = ms.database.OrderedUnionDatabase(
    db1, # no record
    db2, # no record
    ms.database.UnionDatabase( # returns the better one between r3 and r4
        db3, # has r3
        db4, # has r4
    )
    db5,  # has r5
)
# returns the better one between r3 and r4
merged_db.query_tuning_record(..., target_workload)

Co-authored-by: sunggg <49998730+sunggg@users.noreply.github.com>

cc @Hzfengsy @junrushao1994

junrushao · 2022-08-28T00:45:09Z

~~Pending #12626~~

junrushao · 2022-08-30T01:20:22Z

This is ready for review! CC: @sunggg @zxybazh @vinx13

zxybazh

LGTM. Nits on the comments.

include/tvm/meta_schedule/database.h

MasterJH5574

Thanks @junrushao for the continuous hard work! Looks very nice to me :-)

include/tvm/meta_schedule/database.h

src/meta_schedule/database/merged_database.cc

tests/python/unittest/test_link_params.py

sunggg

Thanks for the PR!
LGTM. One clarification question.

python/tvm/meta_schedule/database/merged_database.py

junrushao · 2022-08-30T16:08:01Z

Might need some deliberation on the API design

junrushao · 2022-08-31T06:56:51Z

Significantly amended the APIs and docs according to feedbacks from @sunggg @MasterJH5574 @spectrometerHBH @Kathryn-cat @tqchen. Please re-review :-)

sunggg

LGTM! Thanks for reflecting the feedback. :)
One nit.

include/tvm/meta_schedule/database.h

Following up apache#12520 and apache#12626, this PR introduces two database classes: `UnionDatabase` and `OrderedUnionDatabase`, both of which allow users to organically compose multiple databases together, so that the high-level IR (Relay, Relax) could select the best tuning records according to running time or a preferred order given by users. To each query, `UnionDatabase` returns the best record among all the databases given; Instead, `OrderedUnionDatabase` returns he record from the first database that responds to the query. Used together, users may specify complicated dispatching patterns like below: Examples below demonstrate the usecases of and difference between UnionDatabase and OrderDatabase. Assumption: * db1, db2 do not have tuning records for the target workload. * Each of db3, db4, db5 has tuning records r3, r4, r5 for target workload respectively. ```python #### Case 1. `UnionDatabase`: merged_db = ms.database.UnionDatabase( db1, # no record db2, # no record db3, # has r3 db4 # has r4 ) # returns the better one between r3 and r4 merged_db.query_tuning_record(..., target_workload) ### Case 2. `OrderedUnionDatabase` merged_db = ms.database.OrderedUnionDatabase( db1, # no record db2, # no record db3, # has r3 db4 # has r4 ) # returns r3 merged_db.query_tuning_record(..., target_workload) ### Case 3. Mix-use scenario merged_db = ms.database.UnionDatabase( db1, # no record db2, # no record db3, # has r3 ms.database.OrderedUnionDatabase( # returns r4 db4, # has r4 db5, # has r5 ) ) # returns the better one between r3 and r4 merged_db.query_tuning_record(..., target_workload) ### Case 4. Another mix-use scenario merged_db = ms.database.UnionDatabase( db1, # no record db2, # no record db3, # has r3 ms.database.UnionDatabase( # returns the better one between r4 and r5 db4, # has r4 db5, # has r5 ) ) # returns the best one among r3, r4 and r5 merged_db.query_tuning_record(..., target_workload) ### Case 5. Yet another mix-use scenario merged_db = ms.database.OrderedUnionDatabase( db1, # no record db2, # no record ms.database.UnionDatabase( # returns the better one between r3 and r4 db3, # has r3 db4, # has r4 ) db5, # has r5 ) # returns the better one between r3 and r4 merged_db.query_tuning_record(..., target_workload) ``` Co-authored-by: sunggg <49998730+sunggg@users.noreply.github.com>

junrushao · 2022-09-01T23:44:21Z

Thank you all for super valuable suggestions!

…he#12628) Following up apache#12520 and apache#12626, this PR introduces two database classes: `UnionDatabase` and `OrderedUnionDatabase`, both of which allow users to organically compose multiple databases together, so that the high-level IR (Relay, Relax) could select the best tuning records according to running time or a preferred order given by users. To each query, `UnionDatabase` returns the best record among all the databases given; Instead, `OrderedUnionDatabase` returns he record from the first database that responds to the query. Used together, users may specify complicated dispatching patterns like below: Examples below demonstrate the usecases of and difference between UnionDatabase and OrderDatabase. Assumption: * db1, db2 do not have tuning records for the target workload. * Each of db3, db4, db5 has tuning records r3, r4, r5 for target workload respectively. ```python #### Case 1. `UnionDatabase`: merged_db = ms.database.UnionDatabase( db1, # no record db2, # no record db3, # has r3 db4 # has r4 ) # returns the better one between r3 and r4 merged_db.query_tuning_record(..., target_workload) ### Case 2. `OrderedUnionDatabase` merged_db = ms.database.OrderedUnionDatabase( db1, # no record db2, # no record db3, # has r3 db4 # has r4 ) # returns r3 merged_db.query_tuning_record(..., target_workload) ### Case 3. Mix-use scenario merged_db = ms.database.UnionDatabase( db1, # no record db2, # no record db3, # has r3 ms.database.OrderedUnionDatabase( # returns r4 db4, # has r4 db5, # has r5 ) ) # returns the better one between r3 and r4 merged_db.query_tuning_record(..., target_workload) ### Case 4. Another mix-use scenario merged_db = ms.database.UnionDatabase( db1, # no record db2, # no record db3, # has r3 ms.database.UnionDatabase( # returns the better one between r4 and r5 db4, # has r4 db5, # has r5 ) ) # returns the best one among r3, r4 and r5 merged_db.query_tuning_record(..., target_workload) ### Case 5. Yet another mix-use scenario merged_db = ms.database.OrderedUnionDatabase( db1, # no record db2, # no record ms.database.UnionDatabase( # returns the better one between r3 and r4 db3, # has r3 db4, # has r4 ) db5, # has r5 ) # returns the better one between r3 and r4 merged_db.query_tuning_record(..., target_workload) ``` Co-authored-by: sunggg <49998730+sunggg@users.noreply.github.com>

junrushao force-pushed the feature/2022-08-27/merged-database branch 4 times, most recently from d1d3f4b to 4544cd0 Compare August 30, 2022 01:14

junrushao marked this pull request as ready for review August 30, 2022 01:15

zxybazh approved these changes Aug 30, 2022

View reviewed changes

include/tvm/meta_schedule/database.h Outdated Show resolved Hide resolved

include/tvm/meta_schedule/database.h Outdated Show resolved Hide resolved

MasterJH5574 approved these changes Aug 30, 2022

View reviewed changes

junrushao force-pushed the feature/2022-08-27/merged-database branch from 4544cd0 to aec10ac Compare August 30, 2022 07:11

github-actions bot requested a review from Hzfengsy August 30, 2022 07:15

sunggg approved these changes Aug 30, 2022

View reviewed changes

python/tvm/meta_schedule/database/merged_database.py Outdated Show resolved Hide resolved

junrushao changed the title ~~[MetaSchedule] Introduce MergedDatabase~~ [MetaSchedule] Introduce Union and OrderedUnion in Database Aug 31, 2022

junrushao force-pushed the feature/2022-08-27/merged-database branch 3 times, most recently from 3ed6cea to 370d247 Compare August 31, 2022 06:55

junrushao requested review from masahi, zxybazh, sunggg and MasterJH5574 and removed request for sunggg August 31, 2022 06:57

junrushao force-pushed the feature/2022-08-27/merged-database branch from 370d247 to 2170231 Compare August 31, 2022 08:42

github-actions bot removed the request for review from sunggg August 31, 2022 08:43

sunggg approved these changes Aug 31, 2022

View reviewed changes

include/tvm/meta_schedule/database.h Show resolved Hide resolved

junrushao force-pushed the feature/2022-08-27/merged-database branch 3 times, most recently from 2388106 to 3e83d11 Compare September 1, 2022 19:49

junrushao force-pushed the feature/2022-08-27/merged-database branch from 3e83d11 to ccbd64e Compare September 1, 2022 19:54

junrushao force-pushed the feature/2022-08-27/merged-database branch from ccbd64e to 10d0192 Compare September 1, 2022 19:55

junrushao merged commit eecb7fd into apache:main Sep 1, 2022

AndrewZhaoLuo mentioned this pull request Oct 4, 2022

TVM v0.10.0.rc0 Release Candidate Notes #12979

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[MetaSchedule] Introduce `Union` and `OrderedUnion` in Database #12628

[MetaSchedule] Introduce `Union` and `OrderedUnion` in Database #12628

junrushao commented Aug 28, 2022 •

edited by github-actions bot

Loading

junrushao commented Aug 28, 2022 •

edited

Loading

junrushao commented Aug 30, 2022

zxybazh left a comment

MasterJH5574 left a comment

sunggg left a comment

junrushao commented Aug 30, 2022

junrushao commented Aug 31, 2022

sunggg left a comment

junrushao commented Sep 1, 2022

[MetaSchedule] Introduce Union and OrderedUnion in Database #12628

[MetaSchedule] Introduce Union and OrderedUnion in Database #12628

Conversation

junrushao commented Aug 28, 2022 • edited by github-actions bot Loading

junrushao commented Aug 28, 2022 • edited Loading

junrushao commented Aug 30, 2022

zxybazh left a comment

Choose a reason for hiding this comment

MasterJH5574 left a comment

Choose a reason for hiding this comment

sunggg left a comment

Choose a reason for hiding this comment

junrushao commented Aug 30, 2022

junrushao commented Aug 31, 2022

sunggg left a comment

Choose a reason for hiding this comment

junrushao commented Sep 1, 2022

[MetaSchedule] Introduce `Union` and `OrderedUnion` in Database #12628

[MetaSchedule] Introduce `Union` and `OrderedUnion` in Database #12628

junrushao commented Aug 28, 2022 •

edited by github-actions bot

Loading

junrushao commented Aug 28, 2022 •

edited

Loading