1
1

36 Коммитов

Автор SHA1 Сообщение Дата
Jon Mason
a3bf503e01 Remove error on rdma cm
If there are multiple QP's, RDMACM will not send a message if the
qpnum != 0.  In doing so, it will log an error unecessarily.  This
removes that.

This commit was SVN r18363.
2008-05-02 20:12:01 +00:00
Jon Mason
3989981578 Enable support of num_proc > num_nodes
Add the logic to support using port numbers, instead of simply using
the IP address of the sending node to determine which endpoint to
connect.  Since each process calls the cpc query function, it will
generate its own port to listen on thus enablign this to work.

This commit was SVN r18362.
2008-05-02 16:20:28 +00:00
Jeff Squyres
ba5615a18f Merge in /tmp-public/cpc3 branch to trunk. oob/xoob still remains the
default CPC.

This commit was SVN r18356.
2008-05-02 11:52:33 +00:00
Ralph Castain
d70e2e8c2b Merge the ORTE devel branch into the main trunk. Details of what this means will be circulated separately.
Remains to be tested to ensure everything came over cleanly, so please continue to withhold commits a little longer

This commit was SVN r17632.
2008-02-28 01:57:57 +00:00
Pavel Shamis
a0d12a9c92 Adding support for APM over different ports
This commit was SVN r17521.
2008-02-20 13:44:05 +00:00
Pavel Shamis
3ba3f70624 Adding apm support for xrc.
This commit was SVN r17382.
2008-02-06 10:19:51 +00:00
Pavel Shamis
f0c478e7e0 XRC - replacing the new old API with new one.
This commit was SVN r17369.
2008-02-04 14:03:38 +00:00
Jeff Squyres
2227d5ec4a Add configure check for struct ibv_device.transport type, which was added in OFED v1.2. Still need to fix up oob and rdma_cm cpc's to do something better with this information...
This commit was SVN r17198.
2008-01-24 12:14:21 +00:00
Gleb Natapov
52c94fa7ea Fix compilation warnings.
This commit was SVN r17169.
2008-01-21 15:07:39 +00:00
Jon Mason
a0d4122606 The new cpc selection framework is now in place. The patch below allows
for dynamic selection of cpc methods based on what is available.  It
also allows for inclusion/exclusions of methods.  It even futher allows
for modifying the priorities of certain cpc methods to better determine
the optimal cpc method.

This patch also contains XRC compile time disablement (per Jeff's
patch).

At a high level, the cpc selections works by walking through each cpc
and allowing it to test to see if it is permissable to run on this
mpirun.  It returns a priority if it is permissable or a -1 if not.  All
of the cpc names and priorities are rolled into a string.  This string
is then encapsulated in a message and passed around all the ompi
processes.  Once received and unpacked, the list received is compared
to a local copy of the list.  The connection method is chosen by
comparing the lists passed around to all nodes via modex with the list
generated locally.  Any non-negative number is a potentially valid
connection method.  The method below of determining the optimal
connection method is to take the cross-section of the two lists.  The
highest single value (and the other side being non-negative) is selected
as the cpc method.

svn merge -r 16948:17128 https://svn.open-mpi.org/svn/ompi/tmp-public/openib-cpc/ .

This commit was SVN r17138.
2008-01-14 23:22:03 +00:00
Jon Mason
597c7e68f1 Minor cleanups
This commit was SVN r17100.
2008-01-09 21:54:11 +00:00
Pavel Shamis
fbf7bcd9a9 We need to prepost on srq/xrc before reply with ENDPOINT_XOOB_CONNECT_XRC_RESPONSE.
This commit was SVN r17066.
2008-01-08 10:30:16 +00:00
Pavel Shamis
fcbca510d8 The ib_inline_max should be updated only when SEND qp is created.
This commit was SVN r16973.
2007-12-17 10:30:30 +00:00
Jon Mason
e05cd7b0e4 To modify the default connection method, a "btl_openib_connect <arg>"
should be passed via commandline.  However, there is a slight coding
bug in the openib connect code.  When registering the name of the
option, mca_base_param_reg_string will prepend the relevant info
("btl_openib_" in this case).  The existing code will require
"btl_openib_btl_openib_connect" instead of "btl_openib_connect".
This patch corrects this.

This commit was SVN r16937.
2007-12-11 20:36:36 +00:00
Pavel Shamis
57728986f8 Fixing XRC multiport/multisubnet support.
This commit was SVN r16819.
2007-12-03 09:49:53 +00:00
Gleb Natapov
a774cd98f8 Put send completions to low prio CQ. Receive is more important.
This commit was SVN r16817.
2007-12-02 14:46:37 +00:00
Jeff Squyres
8c0060701c Stub out the ibcm CPC.
This commit was SVN r16800.
2007-11-29 13:23:17 +00:00
Pavel Shamis
1c314ef4c3 If XRC qp was specified in btl_openib_receive_queues we automatically should
choose xoob connection module.

This commit was SVN r16796.
2007-11-28 10:33:32 +00:00
Pavel Shamis
2ffbe8776a Fixing compilation problems in openib
This commit was SVN r16792.
2007-11-28 09:38:49 +00:00
Gleb Natapov
218adb2a96 Account for eager rdma credit fragments when creating send queue. Create XRC
receive QP with zero receive and send queue length. We don't going to use this
QP for send and receives a posted to SRQs.

This commit was SVN r16791.
2007-11-28 07:22:01 +00:00
Gleb Natapov
601952a952 Don't shared endpoint->qps array, only pointer to actual QP. Calculate send
queue size for shared QP based on all endpoints that want to use it.

This commit was SVN r16790.
2007-11-28 07:21:07 +00:00
Gleb Natapov
b46c9cc7bc Make xrc use srq_qp unions instead of the xrc_qp which is exactly like srq_qp.
This commit was SVN r16789.
2007-11-28 07:20:26 +00:00
Gleb Natapov
be0981fc07 Change a type of xrc_recv_qp to "struct ibv_qp".
This commit was SVN r16788.
2007-11-28 07:19:36 +00:00
Gleb Natapov
bd47da4699 Initial XRC support by Mellanox.
This commit was SVN r16787.
2007-11-28 07:18:59 +00:00
Gleb Natapov
b49788c499 Receive queue is not used in case of SRQ QP, so don't create one.
This commit was SVN r16786.
2007-11-28 07:17:22 +00:00
Gleb Natapov
923666b75c Process pending put/get frags on endpoint connection establishment.
This commit was SVN r16785.
2007-11-28 07:16:52 +00:00
Gleb Natapov
5a4e953aaa Allow share the same qp for different buffer sizes. Needed for XRC support.
This commit was SVN r16783.
2007-11-28 07:15:20 +00:00
Gleb Natapov
5463eb892c Send all explicit credits for PP QPs of all orders over smallest PP qp.
This commit was SVN r16781.
2007-11-28 07:13:34 +00:00
Gleb Natapov
a9f864d15c If there is an eager rdma credit, but there is no WQE to send a packet we add it
to a pending queue of eager rdma QP instead of correct pending list. This patch
fixes this by getting reed of "eager rdma qp" notion. Packet is always send
over its order QP. The patch also adds two pending queues for high and low prio
packets. Only high prio packets are sent over eager RDMA channel.

This commit was SVN r16780.
2007-11-28 07:12:44 +00:00
Jeff Squyres
94b1e9cff9 Update to use BTL_VERBOSE and BTL_ERROR instead of opal_output'ing to
the mca_btl_base_output stream directly (and relying on it to be -1 if
we didn't want any output).

This commit was SVN r16449.
2007-10-15 17:53:02 +00:00
Gleb Natapov
9c04b127f5 Forget to put this fix in previous commit.
This commit was SVN r16271.
2007-09-30 15:33:20 +00:00
Gleb Natapov
3a15d645be Remove lcl_qp_attr from endpoint qp description. It is used during init only.
This commit was SVN r16270.
2007-09-30 15:29:35 +00:00
Gleb Natapov
d8f3063895 Create only one CQ for all BTLs on the same HCA. Many BTLs can be created for
one HCA. Multiple ports, LMC, multiple BTLs per one LID. Having only one CQ for
all of them substantially reduce polling time.

This commit was SVN r15933.
2007-08-20 12:28:25 +00:00
Brad Benton
1ddba9ec65 Lock the endpoint before doing endpoint_state processing. This ensures
that the subsequent unlock is valid.

This commit was SVN r15890.
2007-08-16 18:11:29 +00:00
Tim Prins
5a795128af Change it so that different components in orte use unique rml tags
This commit was SVN r15881.
2007-08-16 14:02:35 +00:00
Jeff Squyres
50bae9c603 Bring in the modular-wireup stuff for the openib BTL (from
/tmp/jms-modular-wireup branch):

 * This commit moves all the openib BTL connection code out of
   btl_openib_endpoint.c and into a connect "pseudo-component" area,
   meaning that different schemes for doing OFA connection schemes can
   be chosen via function pointer (i.e., MCA parameter) at run-time.
 * The connect/connect.h file includes comments describing the
   specific interface for the connect pseudo-component.
 * Two pseudo-components are in this commit (more can certainly be
   added).
   * oob: use the same old oob/rml scheme for creating OFA connections
     that we've had forever; this now just puts the logic into this
     self-contained pseudo-component.
   * rdma_cm: a currently-empty set of functions (that currently
     return NOT_IMPLEMENTED) that will someday use the RDMA connection
     manager to make OFA connections.

This commit was SVN r15786.
2007-08-06 23:40:35 +00:00