a) Have PTL_Elan running for short messages <= 2048.
b) Have RMS_RANK setting the ompi:vpid for simple matching with RMS.
c) Fix add_Procs
d) Fix bugs related to elan4 style of addressing elan memory
e) Pending -malign-double option to be added into configuration
f) Temporarily added data integrity check and debugging MACRO
TODO:
g) To add chained DMA support (2 day)
h) test support for arbitrary message sizes in polling mode (1 day)
i) test combined functionality with TCP and Elan (1 day)
f) Get threaded progress engine working and ready for release
(a couple of days)
This commit was SVN r1760.
Stuff accomplished:
1. configure.stub and other Makefile stuff.
2. Layout of initial ptl_ib datastructures.
3. module_open and module_init are working.
4. Set up a UD interface on module init.
5. OOB is used to exchange the UD queue pairs.
HACK ALERT!
-----------
* Forcibly opening only one ptl inside module init. Our machines are
equipped with multiple HCAs ...
* Using #defines for some constants. These should be runtime parameters
but ignoring these for now ...
still fiddling around with ompi_proc and other ib_peer datastructures.
-Sayantan.
This commit was SVN r1759.
- changed the msg structure so it has a pointer to the process name
and not the peer structure. This was done because on the receive side
once a message has been read in it is no longer associated with a peer
but rather just a process name.
- changed the lookup function to include a bool to determine whether to
get the main tcp oob lock or not. This is useful if the caller already
holds the lock.
- first cut of recv functions
- documentation updates
This commit was SVN r1747.
- moved pml_base_send_request_t allocation to pml - provide capability to cache
requests on per ptl basis - and provide hooks for ptl to init/cleanup ptl
specific data cached on the request
- renamed request/fragment super fields
- added ptl_send i/f which is called by pml for the first fragment, the
ptl_put i/f is called for subsequent fragments of the same request
This commit was SVN r1745.
- almost finished the send side of the oob
- added the lists that will hold the posted recieves and the
actual recieves
- made it so we send an addition iovec first that contains the
size of the total message so we can recieve into a temporary buffer
before the user has posted a recieve
This commit was SVN r1736.
- Various fixes / cleanups as a result of the code review
George makes a good point that we could consolidate a lot of the
malloc's / strdup's in this code for platforms where we do not have
a lot of memory (i.e., add up all the memory required and make one
big malloc vs. lots of little mallocs). This is something for
post-first-release.
- Convert another data structure to the OBJ_* interface.
- Split the internal prototypes and struct definitions into their own
header file that won't be included anywhere else in Open MPI (except
ompi_info).
This commit was SVN r1692.
