Merge remote branch 'upstream/master'

ofiwg · Sep 10, 2015 · d2418b2 · d2418b2
2 parents 025c04d + 2b94096
commit d2418b2
Show file tree

Hide file tree

Showing 11 changed files with 266 additions and 48 deletions.
diff --git a/config/cron-run-all-md2nroff.pl b/config/cron-run-all-md2nroff.pl
@@ -34,8 +34,6 @@
     if (!defined($repo_arg));
 die "Must specify a git source branch"
     if (!defined($source_branch_arg));
-die "Must specify a git pages branch"
-    if (!defined($pages_branch_arg));
 
 #####################################################################
 
@@ -103,8 +101,10 @@ sub verbose {
 doit(0, "git clone --single-branch --branch $source_branch_arg $repo_arg source", "git-clone");
 
 # Next, git clone the pages branch of repo
-verbose("*** Cloning repo: $repo_arg / $pages_branch_arg...\n");
-doit(0, "git clone --single-branch --branch $pages_branch_arg $repo_arg pages", "git-clone2");
+if (defined($pages_branch_arg)) {
+    verbose("*** Cloning repo: $repo_arg / $pages_branch_arg...\n");
+    doit(0, "git clone --single-branch --branch $pages_branch_arg $repo_arg pages", "git-clone2");
+}
 
 #####################################################################
 
@@ -118,29 +118,31 @@ sub verbose {
 #####################################################################
 
 # Copy each of the markdown files to the pages branch checkout
-chdir("pages/master");
-foreach my $file (@markdown_files) {
-    doit(0, "cp ../../source/man/$file man/$file", "loop-cp");
+if (defined($pages_branch_arg)) {
+    chdir("pages/master");
+    foreach my $file (@markdown_files) {
+        doit(0, "cp ../../source/man/$file man/$file", "loop-cp");
+
+        # Is there a new man page?  If so, we need to "git add" it.
+        my $out = `git status --porcelain man/$file`;
+        doit(0, "git add man/$file", "loop-git-add")
+            if ($out =~ /^\?\?/);
+    }
 
-    # Is there a new man page?  If so, we need to "git add" it.
-    my $out = `git status --porcelain man/$file`;
-    doit(0, "git add man/$file", "loop-git-add")
-        if ($out =~ /^\?\?/);
+    # Git commit those files in the pages repo and push them to the
+    # upstream repo so that they go live.  If nothing changed, the commit
+    # and push will be no-ops.
+    chdir("..");
+    doit(1, "git commit --no-verify -a -m \"Updated Markdown man pages from $source_branch_arg\"",
+         "git-commit-first");
+    doit(1, "git push", "git-push-first");
 }
 
-# Git commit those files in the pages repo and push them to the
-# upstream repo so that they go live.  If nothing changed, the commit
-# and push will be no-ops.
-chdir("..");
-doit(1, "git commit --no-verify -a -m \"Updated Markdown man pages from $source_branch_arg\"",
-     "git-commit-first");
-doit(1, "git push", "git-push-first");
-
 #####################################################################
 
 # Now process each of the Markdown files in the source repo and
 # generate new nroff man pages.
-chdir("../source/man");
+chdir("$tmpdir/source/man");
 foreach my $file (@markdown_files) {
     doit(0, "../config/md2nroff.pl --source $file", "loop2-md2nroff");
 

diff --git a/config/ofiwg-bot@flatbed:crontab.txt b/config/ofiwg-bot@flatbed:crontab.txt
@@ -0,0 +1,38 @@
+#
+# Update the git clone of libfabric
+#
+# Split this into two lines so that we can have it run every 15 minutes
+# *except* 3:00am (when the nightly tarball is created, which does
+# its own git pull).
+#
+# Update master:
+0,15,30,45 0-2,4-23 * * * (cd $HOME/cron/git/libfabric; git pull --rebase > /dev/null 2>&1; cd $HOME/cron/git/fabtests; git pull --rebase > /dev/null 2>&1)
+15,30,45 3 * * * (cd $HOME/cron/git/libfabric; git pull --rebase > /dev/null 2>&1; cd $HOME/cron/git/fabtests; git pull --rebase > /dev/null 2>&1)
+#
+# Update v1.1.x:
+1,16,31,46 0-2,4-23 * * * (cd $HOME/cron/git/libfabric-1.1.x; git pull --rebase > /dev/null 2>&1)
+15,30,45 3 * * * (cd $HOME/cron/git/libfabric-1.1.x; git pull --rebase > /dev/null 2>&1)
+#
+# Using the script from newly-updated libfabric master clone,
+# generate new man pages
+#
+5,20,35,50 * * * * (PATH=$HOME/cron/local/bin:$PATH; /home/ofiwg-bot/cron/git/libfabric/config/cron-run-all-md2nroff.pl --repo git@github.com:ofiwg/libfabric.git --source-branch master --pages-branch gh-pages --logfile-dir ~/cron/build-logs)
+#
+# Also update man pages on v1.x branch
+# Do not auto-publish the new pages to gh-pages
+#
+6,21,36,51 * * * * (PATH=$HOME/cron/local/bin:$PATH; /home/ofiwg-bot/cron/git/libfabric/config/cron-run-all-md2nroff.pl --repo git@github.com:ofiwg/libfabric.git --source-branch v1.1.x --logfile-dir ~/cron/build-logs/v1.1.x)
+#
+# Make a nightly tarball snapshot from:
+# - libfabric master
+# - fabtests master
+#
+0 3 * * * (PATH=$HOME/cron/local/bin:$PATH; $HOME/cron/git/libfabric/config/cron-make-nightly-tarball.pl --libfabric-source-dir $HOME/cron/git/libfabric --fabtests-source-dir $HOME/cron/git/fabtests --download-dir=/var/www/html/downloads/ofiwg/nightly_tarballs/v1.2.x --logfile-dir ~/cron/build-logs/v1.2.x --libfabric-coverity-token `cat ~/coverity-libfabric-token.txt` --fabtests-coverity-token `cat ~/coverity-fabtests-token.txt`)
+#
+# Make a nightly tarball snapshot from:
+# - libfabric v1.1.x branch
+# - fabtests master
+#
+# No Coverity submissions for this build.
+#
+15 3 * * * (PATH=$HOME/cron/local/bin:$PATH; $HOME/cron/git/libfabric/config/cron-make-nightly-tarball.pl --libfabric-source-dir $HOME/cron/git/libfabric-1.1.x --fabtests-source-dir $HOME/cron/git/fabtests --download-dir=/var/www/html/downloads/ofiwg/nightly_tarballs/v1.1.x --logfile-dir ~/cron/build-logs/v1.1.x)
diff --git a/man/fi_rma.3.md b/man/fi_rma.3.md
@@ -160,7 +160,7 @@ struct fi_msg_rma {
 	const struct iovec *msg_iov;     /* local scatter-gather array */
 	void               **desc;       /* operation descriptor */
 	size_t             iov_count;    /* # elements in msg_iov */
-	const void         *addr;        /* optional endpoint address */
+	fi_addr_t          addr;        /* optional endpoint address */
 	const struct fi_rma_iov *rma_iov;/* remote SGL */
 	size_t             rma_iov_count;/* # elements in rma_iov */
 	void               *context;     /* user-defined context */

diff --git a/man/fi_sockets.7.md b/man/fi_sockets.7.md
@@ -51,6 +51,38 @@ performance numbers are lower compared to other providers implemented
 over high-speed fabric, and lower than what an application might see
 implementing to sockets directly.
 
+# RUNTIME PARAMETERS
+
+The sockets provider checks for the following environment variables -
+
+*FI_SOCKETS_PE_WAITTIME*
+: An integer value that specifies how many milliseconds to spin while waiting for progress in *FI_PROGRESS_AUTO* mode.
+
+*FI_SOCKETS_MAX_CONN_RETRY*
+: An integer value that specifies the number of socket connection retries before reporting as failure.
+
+*FI_SOCKETS_DEF_CONN_MAP_SZ*
+: An integer to specify the default connection map size. 
+
+*FI_SOCKETS_DEF_AV_SZ*
+: An integer to specify the default address vector size.
+
+*FI_SOCKETS_DEF_CQ_SZ*
+: An integer to specify the default completion queue size.
+
+*FI_SOCKETS_DEF_EQ_SZ*
+: An integer to specify the default event queue size.
+
+*FI_SOCKETS_DGRAM_DROP_RATE*
+: An integer value to specify the drop rate of dgram frame when endpoint is *FI_EP_DGRAM*. This is for debugging purpose only.
+
+*FI_SOCKETS_PE_AFFINITY*
+: If specified, progress thread is bound to the indicated range(s) of Linux virtual processor ID(s). This option is currently not supported on OS X. The usage is - id_start[-id_end[:stride]][,].
+
+# LARGE SCALE JOBS
+
+For large scale runs one can use these environment variables to set the default parameters e.g. size of the address vector(AV), completion queue (CQ), connection map etc. that satisfies the requriment of the particular benchmark. The recommended parameters for large scale runs are *FI_SOCKETS_MAX_CONN_RETRY*, *FI_SOCKETS_DEF_CONN_MAP_SZ*, *FI_SOCKETS_DEF_AV_SZ*, *FI_SOCKETS_DEF_CQ_SZ*, *FI_SOCKETS_DEF_EQ_SZ*.
+
 # SEE ALSO
 
 [`fabric`(7)](fabric.7.html),

diff --git a/man/man3/fi_rma.3 b/man/man3/fi_rma.3
@@ -1,4 +1,4 @@
-.TH fi_rma 3 "2015\-08\-19" "Libfabric Programmer\[aq]s Manual" "\@VERSION\@"
+.TH fi_rma 3 "2015\-09\-10" "Libfabric Programmer\[aq]s Manual" "\@VERSION\@"
 .SH NAME
 .PP
 fi_rma - Remote memory access operations
@@ -148,7 +148,7 @@ struct\ fi_msg_rma\ {
 \ \ \ \ const\ struct\ iovec\ *msg_iov;\ \ \ \ \ /*\ local\ scatter-gather\ array\ */
 \ \ \ \ void\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ **desc;\ \ \ \ \ \ \ /*\ operation\ descriptor\ */
 \ \ \ \ size_t\ \ \ \ \ \ \ \ \ \ \ \ \ iov_count;\ \ \ \ /*\ #\ elements\ in\ msg_iov\ */
-\ \ \ \ const\ void\ \ \ \ \ \ \ \ \ *addr;\ \ \ \ \ \ \ \ /*\ optional\ endpoint\ address\ */
+\ \ \ \ fi_addr_t\ \ \ \ \ \ \ \ \ \ addr;\ \ \ \ \ \ \ \ /*\ optional\ endpoint\ address\ */
 \ \ \ \ const\ struct\ fi_rma_iov\ *rma_iov;/*\ remote\ SGL\ */
 \ \ \ \ size_t\ \ \ \ \ \ \ \ \ \ \ \ \ rma_iov_count;/*\ #\ elements\ in\ rma_iov\ */
 \ \ \ \ void\ \ \ \ \ \ \ \ \ \ \ \ \ \ \ *context;\ \ \ \ \ /*\ user-defined\ context\ */

diff --git a/man/man7/fi_sockets.7 b/man/man7/fi_sockets.7
@@ -1,4 +1,4 @@
-.TH fi_sockets 7 "2015\-05\-05" "Libfabric Programmer\[aq]s Manual" "\@VERSION\@"
+.TH fi_sockets 7 "2015\-09\-08" "Libfabric Programmer\[aq]s Manual" "\@VERSION\@"
 .SH NAME
 .PP
 The Sockets Fabric Provider
@@ -44,6 +44,48 @@ implemented over TCP sockets.
 As a result, the performance numbers are lower compared to other
 providers implemented over high-speed fabric, and lower than what an
 application might see implementing to sockets directly.
+.SH RUNTIME PARAMETERS
+.PP
+The sockets provider checks for the following environment variables -
+.PP
+\f[I]FI_SOCKETS_PE_WAITTIME\f[] : An integer value that specifies how
+many milliseconds to spin while waiting for progress in
+\f[I]FI_PROGRESS_AUTO\f[] mode.
+.PP
+\f[I]FI_SOCKETS_MAX_CONN_RETRY\f[] : An integer value that specifies the
+number of socket connection retries before reporting as failure.
+.PP
+\f[I]FI_SOCKETS_DEF_CONN_MAP_SZ\f[] : An integer to specify the default
+connection map size.
+.PP
+\f[I]FI_SOCKETS_DEF_AV_SZ\f[] : An integer to specify the default
+address vector size.
+.PP
+\f[I]FI_SOCKETS_DEF_CQ_SZ\f[] : An integer to specify the default
+completion queue size.
+.PP
+\f[I]FI_SOCKETS_DEF_EQ_SZ\f[] : An integer to specify the default event
+queue size.
+.PP
+\f[I]FI_SOCKETS_DGRAM_DROP_RATE\f[] : An integer value to specify the
+drop rate of dgram frame when endpoint is \f[I]FI_EP_DGRAM\f[].
+This is for debugging purpose only.
+.PP
+\f[I]FI_SOCKETS_PE_AFFINITY\f[] : If specified, progress thread is bound
+to the indicated range(s) of Linux virtual processor ID(s).
+This option is currently not supported on OS X.
+The usage is - id_start[-id_end[:stride]][,].
+.SH LARGE SCALE JOBS
+.PP
+For large scale runs one can use these environment variables to set the
+default parameters e.g.
+size of the address vector(AV), completion queue (CQ), connection map
+etc.
+that satisfies the requriment of the particular benchmark.
+The recommended parameters for large scale runs are
+\f[I]FI_SOCKETS_MAX_CONN_RETRY\f[], \f[I]FI_SOCKETS_DEF_CONN_MAP_SZ\f[],
+\f[I]FI_SOCKETS_DEF_AV_SZ\f[], \f[I]FI_SOCKETS_DEF_CQ_SZ\f[],
+\f[I]FI_SOCKETS_DEF_EQ_SZ\f[].
 .SH SEE ALSO
 .PP
 \f[C]fabric\f[](7), \f[C]fi_provider\f[](7), \f[C]fi_getinfo\f[](3)

diff --git a/prov/sockets/src/sock_atomic.c b/prov/sockets/src/sock_atomic.c
@@ -569,11 +569,17 @@ static int sock_ep_atomic_valid(struct fid_ep *ep, enum fi_datatype datatype,
 		    op == FI_BXOR || op == FI_MSWAP)
 			return -FI_ENOENT;
 		break;
-		
+
 	case FI_FLOAT_COMPLEX:
 	case FI_DOUBLE_COMPLEX:
 	case FI_LONG_DOUBLE_COMPLEX:
-		return -FI_ENOENT;
+		if (op == FI_BOR      || op == FI_BAND     ||
+		    op == FI_BXOR     || op == FI_MSWAP    ||
+		    op == FI_MIN      || op == FI_MAX      ||
+		    op == FI_CSWAP_LE || op == FI_CSWAP_LT ||
+		    op == FI_CSWAP_GE || op == FI_CSWAP_GT)
+			return -FI_ENOENT;
+        break;
 	default:
 		break;
 	}

diff --git a/prov/sockets/src/sock_av.c b/prov/sockets/src/sock_av.c
@@ -366,14 +366,14 @@ static int sock_av_insertsym(struct fid_av *av, const char *node, size_t nodecnt
 		      const char *service, size_t svccnt, fi_addr_t *fi_addr,
 		      uint64_t flags, void *context)
 {
-	int ret = 0, success = 0, err_code = 0;
+	int ret = 0, success = 0, err_code = 0, len1, len2;
 	int var_port, var_host;
 	char base_host[FI_NAME_MAX] = {0};
 	char tmp_host[FI_NAME_MAX] = {0};
 	char tmp_port[FI_NAME_MAX] = {0};
 	int hostlen, offset = 0, fmt, i, j;
 
-	if (!node || !service) {
+	if (!node || !service || node[0] == '\0') {
 		SOCK_LOG_ERROR("Node/service not provided\n");
 		return -FI_EINVAL;
 	}
@@ -386,20 +386,26 @@ static int sock_av_insertsym(struct fid_av *av, const char *node, size_t nodecnt
 		fmt = 0;
 	else 
 		fmt = offset;
-
+
+	assert((hostlen-offset) < FI_NAME_MAX);
 	strncpy(base_host, node, hostlen - (offset));
 	var_port = atoi(service);
 	var_host = atoi(node + hostlen - offset);
 
 	for (i = 0; i < nodecnt; i++) {
 		for (j = 0; j < svccnt; j++) {
-			sprintf(tmp_host, "%s%0*d", base_host, fmt, var_host + i);
-			sprintf(tmp_port, "%d", var_port + j);
-			if ((ret = _sock_av_insertsvc(av, tmp_host, tmp_port, fi_addr, flags, 
+			len1 = snprintf(tmp_host, FI_NAME_MAX, "%s%0*d", base_host, fmt, var_host + i);
+			len2 = snprintf(tmp_port, FI_NAME_MAX,  "%d", var_port + j);
+			if (len1 > 0 && len1 < FI_NAME_MAX && len2 > 0 && len2 < FI_NAME_MAX) {
+				if ((ret = _sock_av_insertsvc(av, tmp_host, tmp_port, fi_addr, flags, 
 						      context, i * nodecnt + j)) == 1) {
-				success++;
+					success++;
+				} else {
+					err_code = ret;
+				}	
 			} else {
-				err_code = ret;
+				SOCK_LOG_ERROR("Node/service value is not valid\n");
+				err_code = FI_ETOOSMALL;
 			}
 		}
 	}

diff --git a/prov/sockets/src/sock_fabric.c b/prov/sockets/src/sock_fabric.c
@@ -59,6 +59,7 @@ int sock_cm_def_map_sz = SOCK_CMAP_DEF_SZ;
 int sock_av_def_sz = SOCK_AV_DEF_SZ;
 int sock_cq_def_sz = SOCK_CQ_DEF_SZ;
 int sock_eq_def_sz = SOCK_EQ_DEF_SZ;
+char *sock_pe_affinity_str = NULL;
 #if ENABLE_DEBUG
 int sock_dgram_drop_rate = 0;
 #endif
@@ -73,6 +74,7 @@ const struct fi_fabric_attr sock_fabric_attr = {
 static struct dlist_entry sock_fab_list;
 static struct dlist_entry sock_dom_list;
 static fastlock_t sock_list_lock;
+static int read_default_params = 0;
 
 void sock_dom_add_to_list(struct sock_domain *domain)
 {
@@ -307,6 +309,23 @@ static struct fi_ops sock_fab_fi_ops = {
 	.ops_open = fi_no_ops_open,
 };
 
+static void sock_read_default_params() {
+	if (!read_default_params) {
+		fi_param_get_int(&sock_prov, "pe_waittime", &sock_pe_waittime);
+		fi_param_get_int(&sock_prov, "max_conn_retry", &sock_conn_retry);
+		fi_param_get_int(&sock_prov, "def_conn_map_sz", &sock_cm_def_map_sz);
+		fi_param_get_int(&sock_prov, "def_av_sz", &sock_av_def_sz);
+		fi_param_get_int(&sock_prov, "def_cq_sz", &sock_cq_def_sz);
+		fi_param_get_int(&sock_prov, "def_eq_sz", &sock_eq_def_sz);
+		if(fi_param_get_str(&sock_prov, "pe_affinity", &sock_pe_affinity_str) != FI_SUCCESS)
+			sock_pe_affinity_str = NULL;
+#if ENABLE_DEBUG
+		fi_param_get_int(&sock_prov, "dgram_drop_rate", &sock_dgram_drop_rate);
+#endif
+		read_default_params = 1;
+	}
+}
+
 static int sock_fabric(struct fi_fabric_attr *attr,
 		       struct fid_fabric **fabric, void *context)
 {
@@ -319,6 +338,8 @@ static int sock_fabric(struct fi_fabric_attr *attr,
 	if (!fab)
 		return -FI_ENOMEM;
 
+	sock_read_default_params();
+
 	fastlock_init(&fab->lock);
 	dlist_init(&fab->service_list);
 
@@ -583,35 +604,32 @@ SOCKETS_INI
 {
 	fi_param_define(&sock_prov, "pe_waittime", FI_PARAM_INT,
                         "How many milliseconds to spin while waiting for progress");
-	fi_param_get_int(&sock_prov, "pe_waittime", &sock_pe_waittime);
 
 	fi_param_define(&sock_prov, "max_conn_retry", FI_PARAM_INT,
 			"Number of connection retries before reporting as failure");
-	fi_param_get_int(&sock_prov, "max_conn_retry", &sock_conn_retry);
 
 	fi_param_define(&sock_prov, "def_conn_map_sz", FI_PARAM_INT,
 			"Default connection map size");
-	fi_param_get_int(&sock_prov, "def_conn_map_sz", &sock_cm_def_map_sz);
 
 	fi_param_define(&sock_prov, "def_av_sz", FI_PARAM_INT,
 			"Default address vector size");
-	fi_param_get_int(&sock_prov, "def_av_sz", &sock_av_def_sz);
 
 	fi_param_define(&sock_prov, "def_cq_sz", FI_PARAM_INT,
 			"Default completion queue size");
-	fi_param_get_int(&sock_prov, "def_cq_sz", &sock_cq_def_sz);
 
 	fi_param_define(&sock_prov, "def_eq_sz", FI_PARAM_INT,
 			"Default event queue size");
-	fi_param_get_int(&sock_prov, "def_eq_sz", &sock_eq_def_sz);
+
+	fi_param_define(&sock_prov, "pe_affinity", FI_PARAM_STRING,
+                        "If specified, bind the progress thread to the indicated range(s) of Linux virtual processor ID(s). "
+                        "This option is currently not supported on OS X. Usage: id_start[-id_end[:stride]][,]");
 
 	fastlock_init(&sock_list_lock);
 	dlist_init(&sock_fab_list);
 	dlist_init(&sock_dom_list);
 #if ENABLE_DEBUG
 	fi_param_define(&sock_prov, "dgram_drop_rate", FI_PARAM_INT,
 			"Drop every Nth dgram frame (debug only)");
-	fi_param_get_int(&sock_prov, "dgram_drop_rate", &sock_dgram_drop_rate);
 #endif
 	return (&sock_prov);
 }