Skip to content
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

romio: implement binding generation for ROMIO #7228

Merged
merged 12 commits into from
Jan 22, 2025

Conversation

hzhou
Copy link
Contributor

@hzhou hzhou commented Dec 1, 2024

Pull Request Description

[note: split preparation PR in #7232]

Integrate ROMIO into MPICH's binding generation framework. This will unify and simplify the IO binding tasks such as large count support, fortran support, man page support.

We'll retain the current stand-alone nature of ROMIO as much as we can in this PR. There shouldn't be any noticeable difference for users who build ROMIO separate from MPICH.

  • A refactor in ROMIO splits the top routines (e.g. MPI_File_open) into a binding layer and an implementation layer -- MPI_File_open in open.c and MPI_File_open_impl in io_impl.c
  • Building ROMIO without FROM_MPICH defined stays exactly the same
  • Building from MPICH will generate the binding layer in src/binding/c/io.c, which calls into the impl layer that is defined in ROMIO.
  • ROMIO is effectively treated the same way as an internal convenience library such as mpl.
  • We no longer need #include "mpio.h" in mpi.h
  • We no longer need worry about profiling and Fortran bindings in ROMIO
  • We no longer need maintain man pages and html docs in ROMIO

Author Checklist

  • Provide Description
    Particularly focus on why, not what. Reference background, issues, test failures, xfail entries, etc.
  • Commits Follow Good Practice
    Commits are self-contained and do not do two things at once.
    Commit message is of the form: module: short description
    Commit message explains what's in the commit.
  • Passes All Tests
    Whitespace checker. Warnings test. Additional tests via comments.
  • Contribution Agreement
    For non-Argonne authors, check contribution agreement.
    If necessary, request an explicit comment from your companies PR approval manager.

@hzhou hzhou force-pushed the 2308_romio_binding branch 14 times, most recently from 966f859 to d00e9f7 Compare December 5, 2024 05:56
@hzhou
Copy link
Contributor Author

hzhou commented Dec 5, 2024

test:mpich/ch3/tcp ✔️

test:mpich/custom
netmod: ch4:ofi
config: mpi-abi ❌

@hzhou hzhou force-pushed the 2308_romio_binding branch 2 times, most recently from 35655c4 to 3c43546 Compare December 6, 2024 04:17
@hzhou
Copy link
Contributor Author

hzhou commented Dec 6, 2024

test:mpich/custom
netmod: ch4:ofi
config: mpi-abi ✔️

@hzhou hzhou mentioned this pull request Dec 6, 2024
4 tasks
@hzhou hzhou force-pushed the 2308_romio_binding branch 4 times, most recently from 06fc378 to a4c17e9 Compare December 6, 2024 23:59
@hzhou hzhou marked this pull request as ready for review December 6, 2024 23:59
@hzhou hzhou requested review from raffenet and roblatham00 December 9, 2024 14:01
@hzhou hzhou force-pushed the 2308_romio_binding branch from a4c17e9 to eaebe60 Compare January 9, 2025 16:36
@hzhou
Copy link
Contributor Author

hzhou commented Jan 9, 2025

test:mpich/ch3/tcp

test:mpich/custom
netmod: ch4:ofi
config: mpi-abi

1 similar comment
@hzhou
Copy link
Contributor Author

hzhou commented Jan 9, 2025

test:mpich/ch3/tcp

test:mpich/custom
netmod: ch4:ofi
config: mpi-abi

Copy link
Contributor

@raffenet raffenet left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

There's a minor conflict that looks simple enough to resolve. @roblatham00 want to give a review? I didn't scrutinize everything since the generated code makes the actual change set pretty huge.

Copy link
Contributor

@roblatham00 roblatham00 left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I did not review every line but I resolved the conflict and ROMIO's I/O tests pass

hzhou added 12 commits January 22, 2025 16:55
Need double cast to suppress warnings when converting between pointers
and integers.

Also remove an unused macro, HAVE_INT_LT_POINTER.
Define the complete MPI-IO interface in mpi.h.

In all_romio_symbols.c, we can use MPI_Request instead of MPIO_Reuqest
since MPIO_Request is not longer defined in mpi.h (since it no longer
includes mpio.h).
Add io_api.txt and generate io bindings
This function is now defined in src/binding/c/io_api.txt.
We don't have error checking for async_thing, amode, whence.
Wrap the code that generates c_bindings.c in a function. This prepares
for later adding io binding generations. ROMIO bindings will be thinner
than non-IO functions. So we will a separate function to generate IO
bindings.
Refactor code to function dump_body_of_routine.

This is a preparation for generating io bindings. Since ROMIO is
designed on top of the rest of MPI, we will handle differently for
parameter filtering and error checking, but the part that handles large
count and calling the internal impl functions can be shared.
MPICH will generate the binding layer.

Now that libromio just supply internal code, we no longer need worry
about libromio vs libpromio. We don't separate man pages and include
mpio.h in mpi.h either.
We no longer have any convenience libraries that need go into libmpi.so
when libpmpi.so is also built. Thus, the legacy mpi_convenience_libs
variable is no longer in use. However, we need differentiate libromio
from the other convenience libraries for the ABI build. ROMIO depends on
MPI, thus will need separate version for ABI and non-ABI build. Thus, we
create two variables -- mpi_convenience_libs and abi_convenience_libs.
Note: the former reuses the name but now with different meanings.
Generate io bindings including prototypes in mpi_proto.h and
mpir_impl.h.

src/binding/c/io.c caintains all the IO binding functions that calls the
corresponding _impl functions. But it is not included in the Makefile
yet.

Dump the prototypes for io impl functions separately to allow inserting
necessary missing declarations such as MPIR_Ext_cs_enter/exit.

IO functions need use MPIR_Ext_cs_enter/exit for global cs. Now it gets
more complicated, refactor into dump_global_cs_enter/exit for better
readability.
Because ROMIO is built on top of the rest of MPI, it is difficult to use
the abi wrappers because we need either -

a. convert from ABI into MPICH abi and then convert back to ABI before
   calling into ROMIO. This seems silly.

Or -

b. build ROMIO on MPICH abi. But in libmpi_abi.so, MPICH abi is not
   available.

Thus we need -

c. Switch ROMIO to call MPICH impl functions instead of `MPI/PMPI`
   functions. We'll need to build a layer in ROMIO switch between using
   `MPI/PMPI` or using `MPIR_Xxx_impl` depending on build options.
   This will result in ROMIO having full access to MPICH internals!
   But let's delay work until we have a convincing need.

Thus, this commit implements io abi bindings without the abi wrappers.
We implemented a separate path rather than making the existing code
more complex.
@hzhou hzhou force-pushed the 2308_romio_binding branch from 59b031e to 014d053 Compare January 22, 2025 22:57
@hzhou hzhou merged commit 837914f into pmodels:main Jan 22, 2025
4 checks passed
@hzhou hzhou deleted the 2308_romio_binding branch January 22, 2025 23:11
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants