Add a GPU implementation of lax.linalg.eig.
#24663
Open
Conversation
This file contains 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
![copybara-service[bot]](https://avatars.githubusercontent.com/in/44061?s=60&v=4){kind=small}
|
Oh.... This is... Amazing! Thanks enormously for this, really, it's been on my secretive wishlist for so long... |
copybara-service
bot
force-pushed
the
test_691072237
branch
3 times, most recently
from
b653418
to
d24721b
Compare
This feature has been in the queue for a long time (see #1259), and some folks have found that they can use `pure_callback` to call the CPU version as a workaround. It has recently come up that there can be issues when using `pure_callback` with JAX calls in the body (#24255; this should be investigated separately). This change adds a native solution for computing `lax.linalg.eig` on GPU. By default, this is implemented by calling LAPACK on host directly because this has good performance for small to moderately sized problems (less than about 2048^2). For larger matrices, a GPU-backed implementation based on [MAGMA](https://icl.utk.edu/magma/) can have significantly better performance. (I should note that I haven't done a huge amount of benchmarking yet, but this was the breakeven point used by PyTorch, and I find roughly similar behavior so far.) We don't want to add MAGMA as a required dependency, but if a user has installed it, JAX can use it when the `jax_gpu_eig_magma` configuration variable is set to `"on"`. By default, we try to dlopen `libmagma.so`, but the path to a non-standard installation location can be specified using the `JAX_GPU_MAGMA_PATH` environment variable. For reasons that I don't yet totally understand, the MAGMA implementation segfaults deep in the MAGMA internals for complex128 inputs, so I've disabled that configuration for now. PiperOrigin-RevId: 691072237
copybara-service
bot
force-pushed
the
test_691072237
branch
from
d24721b
to
35199e1
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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.
Add a GPU implementation of
lax.linalg.eig.This feature has been in the queue for a long time (see #1259), and some folks have found that they can use
pure_callbackto call the CPU version as a workaround. It has recently come up that there can be issues when usingpure_callbackwith JAX calls in the body (#24255; this should be investigated separately).This change adds a native solution for computing
lax.linalg.eigon GPU. By default, this is implemented by calling LAPACK on host directly because this has good performance for small to moderately sized problems (less than about 2048^2). For larger matrices, a GPU-backed implementation based on MAGMA can have significantly better performance. (I should note that I haven't done a huge amount of benchmarking yet, but this was the breakeven point used by PyTorch, and I find roughly similar behavior so far.)We don't want to add MAGMA as a required dependency, but if a user has installed it, JAX can use it when the
jax_gpu_eig_magmaconfiguration variable is set to"on". By default, we try to dlopenlibmagma.so, but the path to a non-standard installation location can be specified using theJAX_GPU_MAGMA_PATHenvironment variable.For reasons that I don't yet totally understand, the MAGMA implementation segfaults deep in the MAGMA internals for complex128 inputs, so I've disabled that configuration for now.