EnzymeAD
TA pass julia type info via param attr
Codecov Report
Attention: 2 lines in your changes are missing coverage. Please review.
Comparison is base (
1829706) 75.93% compared to head (d2df802) 75.68%.
| Files | Patch % | Lines |
|---|---|---|
| src/compiler.jl | 92.85% | 2 Missing :warning: |
:exclamation: Your organization needs to install the Codecov GitHub app to enable full functionality.
Additional details and impacted files
@@ Coverage Diff @@
## main #1300 +/- ##
==========================================
- Coverage 75.93% 75.68% -0.26%
==========================================
Files 35 35
Lines 10543 10507 -36
==========================================
- Hits 8006 7952 -54
- Misses 2537 2555 +18
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
On @swilliamson7 this seems to yield another 50s improvement from 187->132
We still spent most of our time in type-tree.
Across those three towers we spend: 83s out of 132s in constructing the type-tree. The closest other thing is ~10s in julia_activity_rule
Do you know what Julia code is calling all the constructions (specifically the hot regions)
On Tue, Feb 20, 2024 at 11:28 PM Valentin Churavy @.***> wrote:
On @swilliamson7 https://github.com/swilliamson7 this seems to yield another 50s improvement from 187->132
We still spent most of our time in type-tree.
image.png (view on web) https://github.com/EnzymeAD/Enzyme.jl/assets/145258/5effb262-a407-4e6f-a691-b21bd6db80f7
image.png (view on web) https://github.com/EnzymeAD/Enzyme.jl/assets/145258/fe376526-eab9-4de9-a414-f3776fb31f38
Across those three towers we spend: 83s out of 132s in constructing the type-tree. The closest other thing is ~10s in julia_activity_rule
— Reply to this email directly, view it on GitHub https://github.com/EnzymeAD/Enzyme.jl/pull/1300#issuecomment-1955869501, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAJTUXDYZQJGFZLGQJHXH5TYUVZWHAVCNFSM6AAAAABDR4LYTCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSNJVHA3DSNJQGE . You are receiving this because you authored the thread.Message ID: @.***>
Specifically I want to understand whether these functions are called too many times or if they are too slow
On Wed, Feb 21, 2024 at 7:45 AM Billy Moses @.***> wrote:
Do you know what Julia code is calling all the constructions (specifically the hot regions)
On Tue, Feb 20, 2024 at 11:28 PM Valentin Churavy < @.***> wrote:
On @swilliamson7 https://github.com/swilliamson7 this seems to yield another 50s improvement from 187->132
We still spent most of our time in type-tree.
image.png (view on web) https://github.com/EnzymeAD/Enzyme.jl/assets/145258/5effb262-a407-4e6f-a691-b21bd6db80f7
image.png (view on web) https://github.com/EnzymeAD/Enzyme.jl/assets/145258/fe376526-eab9-4de9-a414-f3776fb31f38
Across those three towers we spend: 83s out of 132s in constructing the type-tree. The closest other thing is ~10s in julia_activity_rule
— Reply to this email directly, view it on GitHub https://github.com/EnzymeAD/Enzyme.jl/pull/1300#issuecomment-1955869501, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAJTUXDYZQJGFZLGQJHXH5TYUVZWHAVCNFSM6AAAAABDR4LYTCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSNJVHA3DSNJQGE . You are receiving this because you authored the thread.Message ID: @.***>