1
1
Граф коммитов

96 Коммитов

Автор SHA1 Сообщение Дата
Tim Prins
889f6c79fe Properly initialize the freelist in the constructor.
This commit was SVN r17175.
2008-01-22 18:17:06 +00:00
George Bosilca
841ae01bc9 Add OPAL_LIKELY to the freelist.
This commit was SVN r17142.
2008-01-15 05:44:28 +00:00
George Bosilca
bd72057364 Cleanup the comments.
This commit was SVN r17094.
2008-01-09 20:34:37 +00:00
George Bosilca
906e8bf1d1 Replace the ompi_pointer_array with opal_pointer_array. The next step
(sometimes after the merge with the ORTE branch), the opal_pointer_array
will became the only pointer_array implementation (the orte_pointer_array
will be removed).

This commit was SVN r17007.
2007-12-21 06:02:00 +00:00
Rich Graham
e4646a4dd5 going through the ompi_free_list_init_ex, fl_payload_buffer_size and
fl_payload_buffer_alignment were not being set.

This commit was SVN r16641.
2007-11-02 17:51:32 +00:00
Rich Graham
27a748e7eb change all instances of ompi_free_list_init to ompi_free_list_init_new. Header
and payload data are specified separately at this stage.

This commit was SVN r16633.
2007-11-01 23:38:50 +00:00
Rich Graham
aa82acd34c continuing the incremental changes. fl_elem_class renamed fl_frag_class,
and ompi_free_list_init_new() and ompi_free_list_init_ex_new() were added.
Next step will be to start converting from ompi_free_list_init to()
ompi_free_list_init_new(), and then remove ompi_free_list_init(), and
rename ompi_free_list_init_new() back to ompi_free_list_init().  The merge
of the branch with the trunk was so substantial, it is far easeir to
re-implement the changes in the trunk, rather than trying to fix the bugs
the merge brought in ...

This commit was SVN r16630.
2007-11-01 17:25:12 +00:00
Rich Graham
52fb318950 starting to put in the changes for ompi_free_list_t. fl_elem_size is renamed
to fl_frag_size, fl_alignment is renamed to fl_frag_alignment, and
fl_payload_buffer_size and fl_payload_buffer_alignment are added.

This commit was SVN r16629.
2007-11-01 16:47:44 +00:00
George Bosilca
e724ca0a1f Remodel the ompi_free_list a little. The free_list_memory is in
fact a free_list_item so instead of having a struct, use typedef
to make them equivalent. Modify the parallel debuggers support
in order to allow them access to the internal types even when
we have an optimized build.

This commit was SVN r16567.
2007-10-25 16:47:54 +00:00
Shiqing Fan
a0660f4deb - Just some type casts.
This commit was SVN r16100.
2007-09-12 15:29:58 +00:00
Jeff Squyres
8ace07efed This commit brings in two major things:
1. Galen's fine-grain control of queue pair resources in the openib
   BTL.
1. Pasha's new implementation of asychronous HCA event handling.

Pasha's new implementation doesn't take much explanation, but the new
"multifrag" stuff does.  

Note that "svn merge" was not used to bring this new code from the
/tmp/ib_multifrag branch -- something Bad happened in the periodic
trunk pulls on that branch making an actual merge back to the trunk
effectively impossible (i.e., lots and lots of arbitrary conflicts and
artifical changes).  :-(

== Fine-grain control of queue pair resources ==

Galen's fine-grain control of queue pair resources to the OpenIB BTL
(thanks to Gleb for fixing broken code and providing additional
functionality, Pasha for finding broken code, and Jeff for doing all
the svn work and regression testing).

Prior to this commit, the OpenIB BTL created two queue pairs: one for
eager size fragments and one for max send size fragments.  When the
use of the shared receive queue (SRQ) was specified (via "-mca
btl_openib_use_srq 1"), these QPs would use a shared receive queue for
receive buffers instead of the default per-peer (PP) receive queues
and buffers.  One consequence of this design is that receive buffer
utilization (the size of the data received as a percentage of the
receive buffer used for the data) was quite poor for a number of
applications.

The new design allows multiple QPs to be specified at runtime.  Each
QP can be setup to use PP or SRQ receive buffers as well as giving
fine-grained control over receive buffer size, number of receive
buffers to post, when to replenish the receive queue (low water mark)
and for SRQ QPs, the number of outstanding sends can also be
specified.  The following is an example of the syntax to describe QPs
to the OpenIB BTL using the new MCA parameter btl_openib_receive_queues:

{{{
-mca btl_openib_receive_queues \
     "P,128,16,4;S,1024,256,128,32;S,4096,256,128,32;S,65536,256,128,32"
}}}

Each QP description is delimited by ";" (semicolon) with individual
fields of the QP description delimited by "," (comma).  The above
example therefore describes 4 QPs.

The first QP is:

    P,128,16,4

Meaning: per-peer receive buffer QPs are indicated by a starting field
of "P"; the first QP (shown above) is therefore a per-peer based QP.
The second field indicates the size of the receive buffer in bytes
(128 bytes).  The third field indicates the number of receive buffers
to allocate to the QP (16).  The fourth field indicates the low
watermark for receive buffers at which time the BTL will repost
receive buffers to the QP (4).

The second QP is:

    S,1024,256,128,32

Shared receive queue based QPs are indicated by a starting field of
"S"; the second QP (shown above) is therefore a shared receive queue
based QP.  The second, third and fourth fields are the same as in the
per-peer based QP.  The fifth field is the number of outstanding sends
that are allowed at a given time on the QP (32).  This provides a
"good enough" mechanism of flow control for some regular communication
patterns.

QPs MUST be specified in ascending receive buffer size order.  This
requirement may be removed prior to 1.3 release.

This commit was SVN r15474.
2007-07-18 01:15:59 +00:00
George Bosilca
6e8d25fdaf rearrange the code a bit.
This commit was SVN r15297.
2007-07-05 22:47:31 +00:00
Rainer Keller
6f9251ed39 - Small fixes by PGI -Minform=inform
This commit was SVN r14524.
2007-04-26 08:16:07 +00:00
Gleb Natapov
e5450613b5 Add new SM BTL parameter btl_sm_cb_max_num. If set to value greater then zero
it limits the number of circular buffers allocated between each pair of peers.
This allows for more tight memory usage control.

This commit was SVN r14120.
2007-03-22 12:21:42 +00:00
Brian Barrett
f6be04ff37 be a bit more careful with parens than the r13992 fix
This commit was SVN r13996.

The following SVN revision numbers were found above:
  r13992 --> open-mpi/ompi@3cbac958eb
2007-03-09 16:39:23 +00:00
Brian Barrett
3cbac958eb fix warning about types
This commit was SVN r13992.
2007-03-09 02:32:22 +00:00
Jeff Squyres
b94a39236b Submitted by Gleb, reviewed by Rich:
Queue_empty is determined by the reader, and is it's local view.
However, the writer may continue writing to this queue.  The decision
to go on to the next cb_fifo is done in an atomic region, checking the
writer's view.  The writer also "changes it's view" in an atomic
region protected by the same lock.

This commit was SVN r13968.
2007-03-08 16:51:59 +00:00
Gleb Natapov
be018944d2 Clean up circular buffer implementation. Get rid of _same_base_address()
functions by pre-calculating everything in advance.

This commit was SVN r13923.
2007-03-05 14:27:26 +00:00
Gleb Natapov
90fb58de4f When frags are allocated from mpool by free_list the frag structure is also
allocated from mpool memory (which is registered memory for RDMA transports)
This is not a problem for a small jobs, but for a big number of ranks an
amount of waisted memory is big.

This commit was SVN r13921.
2007-03-05 14:17:50 +00:00
Rainer Keller
0889ebd59f - Eliminate warnings, that PGI-6.2.5 issues with -Minform=inform
This commit was SVN r13840.
2007-02-28 08:36:34 +00:00
Pavel Shamis
edeab0e912 Adding Mellanox Technologies copyright to files touched by Mellanox.
This commit was SVN r13669.
2007-02-15 18:03:20 +00:00
Gleb Natapov
4d4b0a022a Add error callback to sm BTL. Call it when allocation of the initial circular
buffer fails. If cb is already allocated, but it is full and allocation of
additional cb fails, we spin waiting for receiver to free space in existing
cb.

This commit was SVN r13635.
2007-02-13 12:01:36 +00:00
Jeff Squyres
33619d6b43 Minor fixes for the ompi_bitmap class that were found while
investivating #817:

 * Remove use of legal_numbits member and always just use the full
   size of the array.  There was a corner case where legal_numbits was
   not an even multiple of the number of bits in the array where bits
   would not get freed properly, ususally causing wasted fortran
   MPI handles, or, as in the case of #817, wasted attribute keyvals
   (i.e., the user freed them, but the bitmap didn't reflect the
   free).
 * Re-order some error checks to ensure that we don't segv (we don't
   currently trigger this problem anywhere; I just noticed it while
   doing the other attribute keyval and legal_numbits work).

Since this change affects all Fortran MPI handles, I ran all the intel
and ibm tests and all still pass with this change.

This commit was SVN r13561.
2007-02-08 18:20:36 +00:00
Gleb Natapov
4e5deec496 Fix previous patch. In case of different sm base use pointer to tail after
recalculating it.

This commit was SVN r13557.
2007-02-08 14:59:18 +00:00
Gleb Natapov
c56497cf46 Fix race condition, that happens in circular buffer if consumer drained full
cb before producer set cb_overflow flag.

This commit was SVN r13552.
2007-02-08 07:16:14 +00:00
George Bosilca
575075ea77 typo.
This commit was SVN r13510.
2007-02-06 16:53:44 +00:00
Jeff Squyres
c91fcd7fbd Fix a bunch of minor typos submitted by Bernhard Fischer.
This commit was SVN r13505.
2007-02-06 12:00:30 +00:00
Rainer Keller
235f87fd14 - Small cleanups, getting rid of variables, using
index early, etc.

This commit was SVN r13222.
2007-01-19 23:28:04 +00:00
Rainer Keller
125ba1acfa - Reduce the amount of warnings with -Wshadow -- mainly due to
usage of index and abs in inline-fcts in header files.

This commit was SVN r13217.
2007-01-19 19:48:06 +00:00
Gleb Natapov
190e7a27cd Merge with gleb-mpool branch. All RDMA components use same mpool now (rdma).
udapl/openib/vapi/gm mpools a deprecated. rdma mpool has parameter that allows
to limit its size mpool_rdma_rcache_size_limit (default is 0 - unlimited).

This commit was SVN r12878.
2006-12-17 12:26:41 +00:00
George Bosilca
b51b87a4aa The correct way to compute the difference between the actual size and the
expected size, based on the comment few lines before.

This commit was SVN r12235.
2006-10-20 19:33:55 +00:00
George Bosilca
d7268557a8 Complete the SM BTL changes. Now all displacements are ptrdiff_t and there is
no warnings about any issue with signed/unsigned.

This commit was SVN r12234.
2006-10-20 19:28:12 +00:00
George Bosilca
06563b5dec Last set of explicit conversions. We are now close to the zero warnings on
all platforms. The only exceptions (and I will not deal with them
anytime soon) are on Windows:
- the write functions which require the length to be an int when it's
  a size_t on all UNIX variants.
- all iovec manipulation functions where the iov_len is again an int
  when it's a size_t on most of the UNIXes.
As these only happens on Windows, so I think we're set for now :)

This commit was SVN r12215.
2006-10-20 03:57:44 +00:00
George Bosilca
e81d38f322 Remove a function that was just a proof of concept. The same approach is
not used by the TotalView support.

This commit was SVN r12214.
2006-10-20 03:34:16 +00:00
George Bosilca
33f300f636 I don't know what it was supposed to do but I'm quite sure it didn't do it correctly.
Just follow inc_num and you will understand. Now _resize will grow the list to match
the required number of elements as described in the comment in the .h file.

This commit was SVN r12074.
2006-10-10 14:47:51 +00:00
Pavel Shamis
e400da01bc Fix of typo error introduced in revision 12053.
Reviewed by: Jeff

This commit was SVN r12071.
2006-10-10 11:58:29 +00:00
Brian Barrett
51b2a0fd3f A couple of changes to improve shared memory behavior when resources get
constrained:

  * Make sure we always have a number of eager fragments available
    that scales with the number of processes communicating with
    a given proc over shared memory
  * Use FREE_LIST_GET instead of FREE_LIST_WAIT to return an
    error to the PML when resource exhaustion occurs
  * Don't dereference the frag during alloc unless we're sure
    it's not NULL

Reviewed by: Galen

Refs trac:413

This commit was SVN r12053.

The following Trac tickets were found above:
  Ticket 413 --> https://svn.open-mpi.org/trac/ompi/ticket/413
2006-10-06 21:13:49 +00:00
George Bosilca
2411ad74e4 Fix the bug #315. If there are multiple threads waiting for a free_list item
then use broadcast in order to wake them up. If there is only one then use signal
(which is supposed to be faster) and of course if there are no threads
waiting then just continue.

This commit was SVN r12049.
2006-10-06 16:17:50 +00:00
George Bosilca
2029284820 Typo.
This commit was SVN r11695.
2006-09-18 17:57:55 +00:00
George Bosilca
3f0a7cad9e The last patch for Windows support. Mostly casting and conversion to C++ friendly headers.
This commit was SVN r11400.
2006-08-24 16:38:08 +00:00
David Daniel
59f2d86c36 * Move Gleb's rcache work from the gleb-rcache branch to the trunk
This commit was SVN r11198.
2006-08-15 18:40:08 +00:00
Gleb Natapov
383694c68d Add support to get alignemnt buffers from free_list_t. Convert openib BTL to new interface.
This commit was SVN r10899.
2006-07-20 14:39:05 +00:00
George Bosilca
b2a9d15db6 Broadcast the condition (not signal it) as we add multiple elements to the
free list.

This commit was SVN r10850.
2006-07-17 17:07:20 +00:00
George Bosilca
6b7467ea4d NULL is not an option ...
This commit was SVN r10779.
2006-07-13 07:38:35 +00:00
George Bosilca
7602066c4d The next and prev items cannot be NULL. The limit is the sentinel item.
This commit was SVN r10778.
2006-07-13 07:32:13 +00:00
George Bosilca
a43eb4b43e It's not about how much memory we use, but about how we use it.
Keeping the cache misses as low as possible is always a good approach.
The opal_list_t is widely used, it should be a highly optimized class.
The same functionality can be reached with one one sentinel instead
of 2 currently used.
I don't have anything against the STL version, but so far nothing can
compare with the Knuth algorithm. I replace the current implementation
with a modified version of the Knuth algorithm (the one described in
The Art of Computer Programming). As expected, the latency went down.

This commit was SVN r10776.
2006-07-13 04:56:15 +00:00
Gleb Natapov
012d95d195 If ompi_free_list_grow fails wait until resources are available instead of
spinning without progress.

This commit was SVN r10520.
2006-06-27 09:23:51 +00:00
Galen Shipman
8855e5b73a Fixes for DR as well as better diagnostic..
Successfully passing the intel test suite with/without induced errors/drops. 

This commit was SVN r10518.
2006-06-26 22:29:29 +00:00
George Bosilca
9eb023a5c2 OK my last commit was ... kind of wrong. It only worked if the element_size
was smaller than the CACHE_LINE_SIZE. Here is the version that works.

In fact this works on 2 steps. First we set the element size to something
multiple of the desired alignment. Then when we allocate memory, we compute
the total size, and we will align each of the elements (we allocate
multiple of them every time) to the CACHE_LINE_SIZE.

This commit was SVN r10479.
2006-06-22 14:47:07 +00:00
George Bosilca
c71f6c9765 All elements will be aligned to the CACHE_LINE_SIZE define (currently 128
bytes). The simplest way to make sure they are aligned is to update
the size of the basic element to a multiple of the desired alignment.
It will use a little bit more memory, but the improvements on the SM BTL
seems quite interesting.

This commit was SVN r10478.
2006-06-22 14:07:14 +00:00