-
Notifications
You must be signed in to change notification settings - Fork 7
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
0.3.0: Performance Improvement #2
Commits on Jun 12, 2023
-
refactor(jit): make all jitted functions "inline=True"
This should be more beneficial, as followed by the discussions in Jax repository, see jax-ml/jax#6584 jax-ml/jax#6681 jax-ml/jax#9298 jax-ml/jax#9342
Configuration menu - View commit details
-
Copy full SHA for d4b9201 - Browse repository at this point
Copy the full SHA d4b9201View commit details -
perf(pipeline): try to render 4/2/1 rows per batch using vmap to redu…
…ce fori_loop iterations
Configuration menu - View commit details
-
Copy full SHA for 2af8030 - Browse repository at this point
Copy the full SHA 2af8030View commit details -
feat(_meta_utils): simple way to add multiple trace annotations toget…
…her for functions add `@ad_tracing_name` to most functions to assist profiling also bump to Python 3.10 BREAKING CHANGE: Now requires Python 3.10
Configuration menu - View commit details
-
Copy full SHA for 696bb1a - Browse repository at this point
Copy the full SHA 696bb1aView commit details -
perf(pipeline): big refactor to not updating per rows, but renders al…
…l rows then concat and merge
Configuration menu - View commit details
-
Copy full SHA for 1ace375 - Browse repository at this point
Copy the full SHA 1ace375View commit details -
perf(pipeline): using scan + unroll (equiv map + unroll)
This is very similar to map + vmap (minibatch processing) as the inner loop is too complex
Configuration menu - View commit details
-
Copy full SHA for 03b4da9 - Browse repository at this point
Copy the full SHA 03b4da9View commit details -
Configuration menu - View commit details
-
Copy full SHA for 69a2735 - Browse repository at this point
Copy the full SHA 69a2735View commit details -
Configuration menu - View commit details
-
Copy full SHA for 688cafd - Browse repository at this point
Copy the full SHA 688cafdView commit details -
perf: try to eliminate all
lax.cond
under `vmap`, `lax.cond` are lowered to `select_n` in HLO which leads to execution in both branches, thus fails to 1) save computation when possible; 2) prevent unexpected values to be produced/unexpected branches to be executed (defensive), thus let the non-dummy branch to be executed anyway and only rule-out garbage value at the final stage all together to try to improve performance. See google/brax#8409 for more details about unconditional executation of cond under vmap
Configuration menu - View commit details
-
Copy full SHA for e868968 - Browse repository at this point
Copy the full SHA e868968View commit details -
fix(pipeline): gl_FrontFacing: fix its determination in pipeline
`True` if NOT back-facing
Configuration menu - View commit details
-
Copy full SHA for e51b618 - Browse repository at this point
Copy the full SHA e51b618View commit details -
perf: added extra stage in pipeline, aiming to interpolate and shade …
…only one fragment per pixel
Configuration menu - View commit details
-
Copy full SHA for 7ec527b - Browse repository at this point
Copy the full SHA 7ec527bView commit details -
docs(changelog): expose option
loop_unroll
; dependency version changeBump minimum Python version from 3.9 to 3.10; lower minimum jax & jaxlib to 0.3.25.
Configuration menu - View commit details
-
Copy full SHA for c04fdde - Browse repository at this point
Copy the full SHA c04fddeView commit details -
Configuration menu - View commit details
-
Copy full SHA for 3f15f30 - Browse repository at this point
Copy the full SHA 3f15f30View commit details