ports/openmpi
ports/openmpi
1
1
Форкнуть 0
Код Релизы Активность
74d3393a4f
openmpi/ompi/mca/bcol/iboffload/bcol_iboffload_mca.c
Nathan Hjelm 1a021b8f2d coll/ml: add support for blocking and non-blocking allreduce, reduce, and 
allgather.

The new collectives provide a signifigant performance increase over tuned for
small and medium messages. We are initially setting the priority lower than
tuned until this has had some time to soak in the trunk. Please set
coll_ml_priority to 90 for MTT runs.

Credit for this work goes to Manjunath Gorentla Venkata (ORNL), Pavel Shamis (ORNL),
and Nathan Hjelm (LANL).

Commit details (for reference):

Import ORNL's collectives for MPI_Allreduce, MPI_Reduce, and MPI_Allgather.

We need to take the basesmuma header into account when calculating the
ptpcoll small message thresholds. Add a define to bcol.h indicating the
maximum header size so we can take the header into account while not
making ptpcoll dependent on information from basesmuma.

This resolves an issue with allreduce where ptpcoll overwrites the
header of the next buffer in the basesmuma bank.

Fix reduce and make a sequential collective launcher in coll_ml_inlines.h

The root calculation for reduce was wrong for any root != 0. There are
four possibilities for the root:

 - The root is not the current process but is in the current hierarchy. In
   this case the root is the index of the global root as specified in the
   root vector.

 - The root is not the current process and is not in the next level of the
   hierarchy. In this case 0 must be the local root since this process will
   never communicate with the real root.

 - The root is not the current process but will be in next level of the
   hierarchy. In this case the current process must be the root.

 - I am the root. The root is my index.

Tested with IMB which rotates the root on every call to MPI_Reduce. Consider
IMB the reproducer for the issue this commit solves.

Make the bcast algorithm decision an enumerated variable

Resolve various asset failures when destructing coll ml requests.

Two issues:

 - Always reset the request to be invalid before returning it to the
   free list. This will avoid an asset in ompi_request_t's destructor.
   OMPI_REQUEST_FINI does this (and also releases the fortran handle
   index).

 - Never explicitly construct or destruct the superclass of an opal
   object. This screws up the class function tables and will cause
   either an assert failure or a segmentation fault when destructing
   coll ml requests.

Cleanup allgather.

I removed the duplicate non-blocking and blocking functions and modeled
the cleanup after what I found in allreduce. Also cleaned up the code
somewhat.

Don't bother copying from the send to the recieve buffer in
bcol_basesmuma_allreduce_intra_fanin_fanout if the pointers are the
same.

The eliminates a warning about memcpy and aliasing and avoids an
unnecessary call to memcpy.

Alwasy call CHECK_AND_RELEASE on memsync collectives.

There was a call to OBJ_RELEASE on the collective communicator but
because CHECK_AND_RECYLCE was never called there was not matching call
to OBJ_RELEASE. This caused coll ml to leak communicators.

Make allreduce use the sequential collective launcher in coll_ml_inlines.h

Just launch the next collective in the component progress.

I am a little unsure about this patch. There appears to be some sort
of race between collectives that causes buffer exhaustion in some cases
(IMB Allreduce is a reproducer). Changing progress to only launch the
next bcol seems to resolve the issue but might not be the best fix.

Note that I see little-no performance penalty for this change.

Fix allreduce when there are extra sources.

There was an issue with the buffer offset calculation when there are
extra sources. In the case of extra sources == 1 the offset was set
to buffer_size (just past the header of the next buffer). I adjusted
the buffer size to take into accoun the maximum header size (see the
earlier commit that added this) and simplified the offset calculation.

Make reduce/allreduce non-blocking. This is required for MPI_Comm_idup
to work correctly.

This has been tested with various layouts using the ibm testsuite and
imb and appears to have the same performance as the old blocking version.

Fix allgather for non-contiguous layouts and simplify parsing the
topology.

Some things in this patch:

 - There were several comments to the effect that level 0 of the
   hierarchy MUST contain all of the ranks. At least one function
   made this assumption but it was not true. I changed the sbgp
   components and the coll ml initization code to enforce this
   requirement.

 - Ensure that hierarchy level 0 has the ranks in the correct
   scatter gather order. This removes the need for a separate
   sort list and fixes the offset calculation for allgather.

 - There were several passes over the hierarchy to determine
   properties of the hierarchy. I eliminated these extra passes
   and the memory allocation associated with them and calculate the
   tree properties on the fly. The same DFS recursion also handles
   the re-order of level 0.

All these changes have been verified with MPI_Allreduce, MPI_Reduce, and
MPI_Allgather. All functions now pass all IBM/Open MPI, and IMB tests.

coll/ml: correct pointer usage for MPI_BOTTOM

Since contiguous datatypes are copied via memcpy (bypassing the convertor) we
need to adjust for the lb of the datatype. This corrects problems found testing
code that uses MPI_BOTTOM (NULL) as the send pointer.

Add fallback collectives for allreduce and reduce.

cmr=v1.7.5:reviewer=pasha

This commit was SVN r30363.
2014-01-22 15:39:19 +00:00

450 строки
20 KiB
C
Исходник Ответственный История

/*
* Copyright (c) 2009-2012 Oak Ridge National Laboratory. All rights reserved.
* Copyright (c) 2009-2012 Mellanox Technologies. All rights reserved.
* $COPYRIGHT$
*
* Additional copyrights may follow
*
* $HEADER$
*/
#include "ompi_config.h"
#include <fcntl.h>
#include <unistd.h>
#include <sys/mman.h>
#include <sys/types.h>
#include "bcol_iboffload.h"
#include "bcol_iboffload_mca.h"
#include "ompi/constants.h"
#include "ompi/mca/common/ofacm/base.h"
#include "ompi/communicator/communicator.h"
#include "opal/util/show_help.h"
/*
* Local flags
*/
enum {
REGINT_NEG_ONE_OK = 0x01,
REGINT_GE_ZERO = 0x02,
REGINT_GE_ONE = 0x04,
REGINT_NONZERO = 0x08,
REGINT_MAX = 0x88
};
enum {
REGSTR_EMPTY_OK = 0x01,
REGSTR_MAX = 0x88
};
mca_base_var_enum_value_t mtu_values[] = {
{IBV_MTU_256, "256B"},
{IBV_MTU_512, "512B"},
{IBV_MTU_1024, "1k"},
{IBV_MTU_4096, "4k"},
{0, NULL}
};
/*
* utility routine for string parameter registration
*/
static int reg_string(const char* param_name,
const char* deprecated_param_name,
const char* param_desc,
const char* default_value, char **storage,
int flags)
{
int index;
/* the MCA variable system will not attempt to modify this value */
*storage = (char *) default_value;
index = mca_base_component_var_register(&mca_bcol_iboffload_component.super.bcol_version,
param_name, param_desc, MCA_BASE_VAR_TYPE_STRING,
NULL, 0, 0, OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_READONLY, storage);
if (NULL != deprecated_param_name) {
(void) mca_base_var_register_synonym(index, "ompi", "bcol", "iboffload", deprecated_param_name,
MCA_BASE_VAR_SYN_FLAG_DEPRECATED);
}
if (0 != (flags & REGSTR_EMPTY_OK) && 0 == strlen(*storage)) {
opal_output(0, "Bad parameter value for parameter \"%s\"",
param_name);
return OMPI_ERR_BAD_PARAM;
}
return OMPI_SUCCESS;
}
/*
* utility routine for integer parameter registration
*/
static int reg_int(const char* param_name,
const char* deprecated_param_name,
const char* param_desc,
int default_value, int *storage, int flags)
{
int index;
*storage = default_value;
index = mca_base_component_var_register(&mca_bcol_iboffload_component.super.bcol_version,
param_name, param_desc, MCA_BASE_VAR_TYPE_INT,
NULL, 0, 0, OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_READONLY, storage);
if (NULL != deprecated_param_name) {
(void) mca_base_var_register_synonym(index, "ompi", "bcol", "iboffload", deprecated_param_name,
MCA_BASE_VAR_SYN_FLAG_DEPRECATED);
}
if (0 != (flags & REGINT_NEG_ONE_OK) && -1 == *storage) {
return OMPI_SUCCESS;
}
if ((0 != (flags & REGINT_GE_ZERO) && *storage < 0) ||
(0 != (flags & REGINT_GE_ONE) && *storage < 1) ||
(0 != (flags & REGINT_NONZERO) && 0 == *storage)) {
opal_output(0, "Bad parameter value for parameter \"%s\"",
param_name);
return OMPI_ERR_BAD_PARAM;
}
return OMPI_SUCCESS;
}
/*
* utility routine for integer parameter registration
*/
static int reg_bool(const char* param_name,
const char* deprecated_param_name,
const char* param_desc,
bool default_value, bool *storage)
{
int index;
*storage = default_value;
index = mca_base_component_var_register(&mca_bcol_iboffload_component.super.bcol_version,
param_name, param_desc, MCA_BASE_VAR_TYPE_BOOL,
NULL, 0, 0, OPAL_INFO_LVL_9,
MCA_BASE_VAR_SCOPE_READONLY, storage);
if (NULL != deprecated_param_name) {
(void) mca_base_var_register_synonym(index, "ompi", "bcol", "iboffload", deprecated_param_name,
MCA_BASE_VAR_SYN_FLAG_DEPRECATED);
}
return OMPI_SUCCESS;
}
int mca_bcol_iboffload_verify_params(void)
{
if (mca_bcol_iboffload_component.min_rnr_timer > 31) {
opal_show_help("help-mpi-bcol-iboffload.txt", "invalid mca param value",
true, "bcol_iboffload_ib_min_rnr_timer > 31",
"bcol_iboffload_ib_min_rnr_timer reset to 31");
mca_bcol_iboffload_component.min_rnr_timer = 31;
} else if (mca_bcol_iboffload_component.min_rnr_timer < 0){
opal_show_help("help-mpi-bcol-iboffload.txt", "invalid mca param value",
true, "bcol_iboffload_ib_min_rnr_timer < 0",
"bcol_iboffload_ib_min_rnr_timer reset to 0");
mca_bcol_iboffload_component.min_rnr_timer = 0;
}
if (mca_bcol_iboffload_component.timeout > 31) {
opal_show_help("help-mpi-bcol-iboffload.txt", "invalid mca param value",
true, "bcol_iboffload_ib_timeout > 31",
"bcol_iboffload_ib_timeout reset to 31");
mca_bcol_iboffload_component.timeout = 31;
} else if (mca_bcol_iboffload_component.timeout < 0) {
opal_show_help("help-mpi-bcol-iboffload.txt", "invalid mca param value",
true, "bcol_iboffload_ib_timeout < 0",
"bcol_iboffload_ib_timeout reset to 0");
mca_bcol_iboffload_component.timeout = 0;
}
if (mca_bcol_iboffload_component.retry_count > 7) {
opal_show_help("help-mpi-bcol-iboffload.txt", "invalid mca param value",
true, "bcol_iboffload_ib_retry_count > 7",
"bcol_iboffload_ib_retry_count reset to 7");
mca_bcol_iboffload_component.retry_count = 7;
} else if (mca_bcol_iboffload_component.retry_count < 0) {
opal_show_help("help-mpi-bcol-iboffload.txt", "invalid mca param value",
true, "bcol_iboffload_ib_retry_count < 0",
"bcol_iboffload_ib_retry_count reset to 0");
mca_bcol_iboffload_component.retry_count = 0;
}
if (mca_bcol_iboffload_component.max_rdma_dst_ops > 7) {
opal_show_help("help-mpi-bcol-iboffload.txt", "invalid mca param value",
true, "bcol_iboffload_ib_rnr_retry > 7",
"bcol_iboffload_ib_rnr_retry reset to 7");
mca_bcol_iboffload_component.max_rdma_dst_ops = 7;
} else if (mca_bcol_iboffload_component.max_rdma_dst_ops < 0) {
opal_show_help("help-mpi-bcol-iboffload.txt", "invalid mca param value",
true, "bcol_iboffload_ib_rnr_retry < 0",
"bcol_iboffload_ib_rnr_retry reset to 0");
mca_bcol_iboffload_component.max_rdma_dst_ops = 0;
}
if (mca_bcol_iboffload_component.service_level > 15) {
opal_show_help("help-mpi-bcol-iboffload.txt", "invalid mca param value",
true, "bcol_iboffload_ib_service_level > 15",
"bcol_iboffload_ib_service_level reset to 15");
mca_bcol_iboffload_component.service_level = 15;
} else if (mca_bcol_iboffload_component.service_level < 0) {
opal_show_help("help-mpi-bcol-iboffload.txt", "invalid mca param value",
true, "bcol_iboffload_ib_service_level < 0",
"bcol_iboffload_ib_service_level reset to 0");
mca_bcol_iboffload_component.service_level = 0;
}
if(mca_bcol_iboffload_component.buffer_alignment <= 1 ||
(mca_bcol_iboffload_component.buffer_alignment & (mca_bcol_iboffload_component.buffer_alignment - 1))) {
opal_show_help("help-mpi-bcol-iboffload.txt", "wrong buffer alignment",
true, mca_bcol_iboffload_component.buffer_alignment, ompi_process_info.nodename, 64);
mca_bcol_iboffload_component.buffer_alignment = 64;
}
return OMPI_SUCCESS;
}
int mca_bcol_iboffload_register_params(void)
{
mca_base_var_enum_t *new_enum;
char *msg;
int ret = OMPI_SUCCESS, tmp;
#define CHECK(expr) do { \
tmp = (expr); \
if (OMPI_SUCCESS != tmp) ret = tmp; \
} while (0)
/* register openib component parameters */
CHECK(reg_int("k_nomial_radix", NULL,
"The radix of the K-nomial tree for scatther-gather type algorithms"
"(starts from 2)", 2, &mca_bcol_iboffload_component.k_nomial_radix,
REGINT_GE_ONE));
CHECK(reg_int("priority", NULL,
"IB offload component priority"
"(from 0(low) to 90 (high))", 90,
&mca_bcol_iboffload_component.super.priority, 0));
CHECK(reg_int("verbose", NULL,
"Output some verbose IB offload BTL information "
"(0 = no output, nonzero = output)", 0,
&mca_bcol_iboffload_component.verbose, 0));
CHECK(reg_bool("warn_default_gid_prefix", NULL,
"Warn when there is more than one active ports and at least one of them connected to the network with only default GID prefix configured (0 = do not warn; any other value = warn)",
true, &mca_bcol_iboffload_component.warn_default_gid_prefix));
CHECK(reg_bool("warn_nonexistent_if", NULL,
"Warn if non-existent devices and/or ports are specified in the bcol_iboffla_if_[in|ex]clude MCA parameters (0 = do not warn; any other value = warn)",
true, &mca_bcol_iboffload_component.warn_nonexistent_if));
CHECK(reg_int("max_pipeline_depth", NULL,
"The maximal number of fragments of the same collective request that can be transferred in parallel", 3,
(int *) &mca_bcol_iboffload_component.max_pipeline_depth, 0));
CHECK(reg_int("max_mqe_tasks", NULL,
"Maximum number of MQEs for each iboffload module",
1024, &mca_bcol_iboffload_component.max_mqe_tasks, 0));
CHECK(reg_int("max_mq_size", NULL,
"Maximum size of each MQ for each iboffload module",
1024, &mca_bcol_iboffload_component.max_mq_size, 0));
CHECK(reg_int("free_list_num", NULL,
"Intial size of free lists (must be >= 1)",
256, &mca_bcol_iboffload_component.free_list_num,
REGINT_GE_ONE));
CHECK(reg_int("free_list_max", NULL,
"Maximum size of free lists "
"(-1 = infinite, otherwise must be >= 0)",
-1, &mca_bcol_iboffload_component.free_list_max,
REGINT_NEG_ONE_OK | REGINT_GE_ONE));
CHECK(reg_int("free_list_inc", NULL,
"Increment size of free lists (must be >= 1)",
32, &mca_bcol_iboffload_component.free_list_inc,
REGINT_GE_ONE));
/* rdma mpool no longer exists - must use the grdma mpool component, should resolve errors in
* mtt testing
*/
/*
CHECK(reg_string("mpool", NULL,
"Name of the memory pool to be used (it is unlikely that you will ever want to change this",
"rdma", &mca_bcol_iboffload_component.mpool_name,
0));
*/
CHECK(reg_string("mpool", NULL,
"Name of the memory pool to be used (it is unlikely that you will ever want to change this",
"grdma", &mca_bcol_iboffload_component.mpool_name,
0));
CHECK(reg_int("cq_size", "cq_size",
"Size of the OpenFabrics completion "
"queue (will automatically be set to a minimum of "
"(2 * number_of_peers * bcol_iboffload_rd_num))",
1024, &mca_bcol_iboffload_component.cq_size, REGINT_GE_ONE));
CHECK(reg_int("exchange_tree_order", NULL,
"The order of the exchange tree. "
"Must be power of two.",
2, &mca_bcol_iboffload_component.exchange_tree_order, REGINT_GE_ONE));
CHECK(reg_int("knomial_tree_order", NULL,
"The order of the knomial exchange tree. ",
3, &mca_bcol_iboffload_component.knomial_tree_order, REGINT_GE_ONE));
CHECK(reg_int("max_inline_data", "max_inline_data",
"Maximum size of inline data segment "
"(-1 = run-time probe to discover max value, "
"otherwise must be >= 0). "
"If not explicitly set, use max_inline_data from "
"the INI file containing device-specific parameters",
128, (int *) &mca_bcol_iboffload_component.max_inline_data,
REGINT_NEG_ONE_OK | REGINT_GE_ZERO));
#if 0
CHECK(reg_string("pkey", "ib_pkey_val",
"OpenFabrics partition key (pkey) value. "
"Unsigned integer decimal or hex values are allowed (e.g., \"3\" or \"0x3f\") and will be masked against the maximum allowable IB paritition key value (0x7fff)",
"0", &pkey, 0));
/* Pasha
mca_bcol_iboffload_component.pkey_val =
ompi_btl_openib_ini_intify(pkey) & MCA_BTL_IB_PKEY_MASK;
free(pkey);
*/
#endif
CHECK(reg_string("receive_queues", NULL,
"Colon-delimited, comma delimited list of receive queues: P,4096,8,6,4:P,32768,8,6,4",
"P,512,256,192,128", &mca_bcol_iboffload_component.receive_queues,
0));
CHECK(reg_int("qp_ous_rd_atom", NULL,
"InfiniBand outstanding atomic reads (must be >= 0)", 4,
(int *) &mca_bcol_iboffload_component.qp_ous_rd_atom, REGINT_GE_ZERO));
asprintf(&msg, "OpenFabrics MTU, in bytes (if not specified in INI files). Valid values are: %d=256 bytes, %d=512 bytes, %d=1024 bytes, %d=2048 bytes, %d=4096 bytes",
IBV_MTU_256,
IBV_MTU_512,
IBV_MTU_1024,
IBV_MTU_2048,
IBV_MTU_4096);
if (NULL == msg) {
/* Don't try to recover from this */
return OMPI_ERR_OUT_OF_RESOURCE;
}
CHECK(mca_base_var_enum_create("infiniband mtu", mtu_values, &new_enum));
mca_bcol_iboffload_component.mtu = IBV_MTU_1024;
tmp = mca_base_component_var_register(&mca_bcol_iboffload_component.super.bcol_version,
"mtu", msg, MCA_BASE_VAR_TYPE_INT, new_enum, 0, 0,
OPAL_INFO_LVL_9, MCA_BASE_VAR_SCOPE_READONLY,
&mca_bcol_iboffload_component.mtu);
OBJ_RELEASE(new_enum);
free(msg);
if (0 > tmp) ret = tmp;
tmp = mca_base_var_register_synonym(tmp, "ompi", "bcol", "iboffload", "ib_mtu",
MCA_BASE_VAR_SYN_FLAG_DEPRECATED);
if (0 > tmp) ret = tmp;
CHECK(reg_int("ib_min_rnr_timer", NULL, "InfiniBand minimum "
"\"receiver not ready\" timer, in seconds "
"(must be >= 0 and <= 31)",
1 , &mca_bcol_iboffload_component.min_rnr_timer, 0));
CHECK(reg_int("ib_timeout", NULL, "InfiniBand transmit timeout, plugged into formula: 4.096 microseconds * "
"(2^bcol_iboffload_ib_timeout) (must be >= 0 and <= 31)",
20, &mca_bcol_iboffload_component.timeout, 0));
CHECK(reg_int("ib_retry_count", NULL, "InfiniBand transmit retry count "
"(must be >= 0 and <= 7)",
7, &mca_bcol_iboffload_component.retry_count, 0));
CHECK(reg_int("ib_rnr_retry", NULL, "InfiniBand \"receiver not ready\" "
"retry count; applies *only* to SRQ/XRC queues. PP queues "
"use RNR retry values of 0 because Open MPI performs "
"software flow control to guarantee that RNRs never occur "
"(must be >= 0 and <= 7; 7 = \"infinite\")",
7, &mca_bcol_iboffload_component.rnr_retry, 0));
CHECK(reg_int("ib_max_rdma_dst_ops", NULL, "InfiniBand maximum pending RDMA "
"destination operations "
"(must be >= 0)",
4, &mca_bcol_iboffload_component.max_rdma_dst_ops, REGINT_GE_ZERO));
CHECK(reg_int("ib_service_level", NULL, "InfiniBand service level "
"(must be >= 0 and <= 15)",
0, &mca_bcol_iboffload_component.service_level, 0));
CHECK(reg_int("buffer_alignment", NULL,
"Prefered communication buffer alignment, in bytes "
"(must be > 0 and power of two)",
64, &mca_bcol_iboffload_component.buffer_alignment, REGINT_GE_ZERO));
/* register parmeters controlling message fragementation */
CHECK(reg_int("min_frag_size", NULL,
"Minimum fragment size",
getpagesize(), &mca_bcol_iboffload_component.super.min_frag_size,
REGINT_GE_ONE));
CHECK(reg_int("max_frag_size", NULL,
"Maximum fragment size",
FRAG_SIZE_NO_LIMIT, &mca_bcol_iboffload_component.super.max_frag_size,
REGINT_NONZERO));
CHECK(reg_bool("can_use_user_buffers", NULL,
"User memory can be used by the collective algorithms",
true, &mca_bcol_iboffload_component.super.can_use_user_buffers));
CHECK(reg_int("barrier_mode", NULL,
"Barrier mode: 0 - Recursive doubling; 1 - Recursive K-ing",
0, &mca_bcol_iboffload_component.barrier_mode, REGINT_GE_ZERO));
CHECK(reg_int("max_progress_pull", NULL,
"Max number of progress pull checks",
8, &mca_bcol_iboffload_component.max_progress_pull, REGINT_GE_ZERO));
CHECK(reg_int("use_brucks_smsg_alltoall_rdma", NULL,
"Use brucks algorithm for smsg alltoall and RDMA semantics 1 = No Temp buffer recycling"
"1 = Alg with no Temp Buffer Recycling (faster), 2 = Alg with temp Buffer Recycling (slower)",
0, &mca_bcol_iboffload_component.use_brucks_smsg_alltoall_rdma, 0));
CHECK(reg_int("use_brucks_smsg_alltoall_sr", NULL,
"Use brucks algorithm for smsg alltoall and Send/Recv semantics "
"1 = Alg with RTR (faster), 2 = Alg with RNR (slower)",
0, &mca_bcol_iboffload_component.use_brucks_smsg_alltoall_sr, 0));
CHECK(reg_int("alltoall_bruck_radix", NULL,
"Radix for Bruck algorithm for smsg alltoall",
3, &mca_bcol_iboffload_component.k_alltoall_bruck_radix, 0));
CHECK(reg_int("k_alltoall_bruck_radix", NULL,
"Temp Buffer alignment for Bruck algorithm for smsg alltoall",
64, &mca_bcol_iboffload_component.tmp_buf_alignment, 0));
/*
CHECK(reg_string("if_include", NULL,
"Comma-delimited list of devices/ports to be used (e.g. \"mthca0,mthca1:2\"; empty value means to use all ports found). Mutually exclusive with bcol_iboffload_if_exclude.",
NULL, &mca_bcol_iboffload_component.if_include,
0));
CHECK(reg_string("if_exclude", NULL,
"Comma-delimited list of device/ports to be excluded (empty value means to not exclude any ports). Mutually exclusive with bcol_iboffload_if_include.",
NULL, &mca_bcol_iboffload_component.if_exclude,
0));
*/
CHECK(mca_bcol_iboffload_verify_params());
/* Register any MCA params for the connect pseudo-components */
if (OMPI_SUCCESS == ret) {
ret = ompi_common_ofacm_base_register(&mca_bcol_iboffload_component.super.bcol_version);
}
return ret;
}
Показать git blame Копировать постоянную ссылку