Allow control of patching via environment variables #20
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
Under some circumstances (usually involving InfiniBand network stacks), fork() has bad consequences. Introduce environment variables to control the behaviour of patch_numba_codegen_if_needed (which previously unconditionally forked a subprocess), to both skip the check and (optionally) unconditionally patch Numba codegen.
gmarkall
approved these changes
Jun 17, 2022
There was a problem hiding this comment.
Many thanks for the PR. This looks good to me and behaves as expected - some example invocations:
$ PTXCOMPILER_APPLY_NUMBA_CODEGEN_PATCH=1 NUMBA_CUDA_LOG_LEVEL=DEBUG python -c "from ptxcompiler.patch import patch_numba_codegen_if_needed; patch_numba_codegen_if_needed()"
== CUDA (ptxcompiler) [1701] DEBUG -- PTXCOMPILER_APPLY_NUMBA_CODEGEN_PATCH=1
== CUDA (ptxcompiler) [1701] DEBUG -- Patching Numba codegen for forward compatibility
$ PTXCOMPILER_APPLY_NUMBA_CODEGEN_PATCH=0 NUMBA_CUDA_LOG_LEVEL=DEBUG python -c "from ptxcompiler.patch import patch_numba_codegen_if_needed; patch_numba_codegen_if_needed()"
== CUDA (ptxcompiler) [282] DEBUG -- PTXCOMPILER_APPLY_NUMBA_CODEGEN_PATCH=0
== CUDA (ptxcompiler) [671] DEBUG -- CUDA Driver version 11.7
== CUDA (ptxcompiler) [671] DEBUG -- CUDA Runtime version 11.7
== CUDA (ptxcompiler) [671] DEBUG -- Driver version sufficient: not patching Numba codegen
$ PTXCOMPILER_CHECK_NUMBA_CODEGEN_PATCH_NEEDED=1 PTXCOMPILER_APPLY_NUMBA_CODEGEN_PATCH=0 NUMBA_CUDA_LOG_LEVEL=DEBUG python -c "from ptxcompiler.patch import patch_numba_codegen_if_needed; patch_numba_codegen_if_needed()"
== CUDA (ptxcompiler) [291] DEBUG -- PTXCOMPILER_CHECK_NUMBA_CODEGEN_PATCH_NEEDED=1
== CUDA (ptxcompiler) [291] DEBUG -- PTXCOMPILER_APPLY_NUMBA_CODEGEN_PATCH=0
== CUDA (ptxcompiler) [652] DEBUG -- CUDA Driver version 11.7
== CUDA (ptxcompiler) [652] DEBUG -- CUDA Runtime version 11.7
== CUDA (ptxcompiler) [652] DEBUG -- Driver version sufficient: not patching Numba codegen
$ PTXCOMPILER_CHECK_NUMBA_CODEGEN_PATCH_NEEDED=0 PTXCOMPILER_APPLY_NUMBA_CODEGEN_PATCH=0 NUMBA_CUDA_LOG_LEVEL=DEBUG python -c "from ptxcompiler.patch import patch_numba_codegen_if_needed; patch_numba_codegen_if_needed()"
== CUDA (ptxcompiler) [284] DEBUG -- PTXCOMPILER_CHECK_NUMBA_CODEGEN_PATCH_NEEDED=0
== CUDA (ptxcompiler) [284] DEBUG -- PTXCOMPILER_APPLY_NUMBA_CODEGEN_PATCH=0
== CUDA (ptxcompiler) [284] DEBUG -- Driver version sufficient: not patching Numba codegen
$ PTXCOMPILER_CHECK_NUMBA_CODEGEN_PATCH_NEEDED=0 PTXCOMPILER_APPLY_NUMBA_CODEGEN_PATCH=1 NUMBA_CUDA_LOG_LEVEL=DEBUG python -c "from ptxcompiler.patch import patch_numba_codegen_if_needed; patch_numba_codegen_if_needed()"
== CUDA (ptxcompiler) [285] DEBUG -- PTXCOMPILER_CHECK_NUMBA_CODEGEN_PATCH_NEEDED=0
== CUDA (ptxcompiler) [285] DEBUG -- PTXCOMPILER_APPLY_NUMBA_CODEGEN_PATCH=1
== CUDA (ptxcompiler) [285] DEBUG -- Patching Numba codegen for forward compatibility
$ PTXCOMPILER_CHECK_NUMBA_CODEGEN_PATCH_NEEDED=0 NUMBA_CUDA_LOG_LEVEL=DEBUG python -c "from ptxcompiler.patch import patch_numba_codegen_if_needed; patch_numba_codegen_if_needed()"
== CUDA (ptxcompiler) [281] DEBUG -- PTXCOMPILER_CHECK_NUMBA_CODEGEN_PATCH_NEEDED=0
== CUDA (ptxcompiler) [281] DEBUG -- Driver version sufficient: not patching Numba codegen
$ PTXCOMPILER_CHECK_NUMBA_CODEGEN_PATCH_NEEDED=1 NUMBA_CUDA_LOG_LEVEL=DEBUG python -c "from ptxcompiler.patch import patch_numba_codegen_if_needed; patch_numba_codegen_if_needed()"
== CUDA (ptxcompiler) [282] DEBUG -- PTXCOMPILER_CHECK_NUMBA_CODEGEN_PATCH_NEEDED=1
== CUDA (ptxcompiler) [668] DEBUG -- CUDA Driver version 11.7
== CUDA (ptxcompiler) [668] DEBUG -- CUDA Runtime version 11.7
== CUDA (ptxcompiler) [668] DEBUG -- Driver version sufficient: not patching Numba codegen
$ NUMBA_CUDA_LOG_LEVEL=DEBUG python -c "from ptxcompiler.patch import patch_numba_codegen_if_needed; patch_numba_codegen_if_needed()"
== CUDA (ptxcompiler) [669] DEBUG -- CUDA Driver version 11.7
== CUDA (ptxcompiler) [669] DEBUG -- CUDA Runtime version 11.7
== CUDA (ptxcompiler) [669] DEBUG -- Driver version sufficient: not patching Numba codegen
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.
Under some circumstances (usually involving InfiniBand network
stacks), fork() has bad consequences. Introduce environment variables
to control the behaviour of patch_numba_codegen_if_needed (which
previously unconditionally forked a subprocess), to both skip the
check and (optionally) unconditionally patch Numba codegen.