diff -r -u /tmp/perftest-1.3.0/Makefile perftest-1.3.0/Makefile

--- /tmp/perftest-1.3.0/Makefile	Thu Jan 20 01:37:35 2011

+++ perftest-1.3.0/Makefile	Fri Feb 11 04:12:45 2011

@@ -2,10 +2,11 @@

 MCAST_TESTS = send_bw send_lat

 TESTS = write_bw_postlist write_lat write_bw read_lat read_bw

 UTILS = clock_test

+BINS = ib_write_bw_postlist rdma_lat rdma_bw ib_send_lat ib_send_bw ib_write_lat ib_write_bw ib_read_lat ib_read_bw ib_clock_test

 all: ${RDMACM_TESTS} ${MCAST_TESTS} ${TESTS} ${UTILS}

-CFLAGS += -Wall -g -D_GNU_SOURCE -O2

+CFLAGS += -Wall -g -D_GNU_SOURCE -O3

 BASIC_FILES = get_clock.c

 EXTRA_FILES = perftest_resources.c

 MCAST_FILES = multicast_resources.c

@@ -12,13 +13,14 @@

 BASIC_HEADERS = get_clock.h

 EXTRA_HEADERS = perftest_resources.h

 MCAST_HEADERS = multicast_resources.h

+CC=gcc

 #The following seems to help GNU make on some platforms

 LOADLIBES += 

 LDFLAGS +=

-${RDMACM_TESTS}: LOADLIBES += -libverbs -lrdmacm

-${MCAST_TESTS}: LOADLIBES += -libverbs -libumad -lm

-${TESTS} ${UTILS}: LOADLIBES += -libverbs

+${RDMACM_TESTS}: LOADLIBES += -lsocket -libverbs -lrdmacm -lrt 

+${MCAST_TESTS}: LOADLIBES += -lsocket -libverbs -lrdmacm -lrt -libumad -lm

+${TESTS} ${UTILS}: LOADLIBES += -lsocket -lrt -libverbs

 ${RDMACM_TESTS}: %: %.c ${BASIC_FILES} ${BASIC_HEADERS}

 	$(CC) $(CPPFLAGS) $(CFLAGS) $(LDFLAGS) $< ${BASIC_FILES} $(LOADLIBES) $(LDLIBS) -o $@

@@ -27,6 +29,10 @@

 ${TESTS} ${UTILS}: %: %.c ${BASIC_FILES} ${EXTRA_FILES} ${BASIC_HEADERS} ${EXTRA_HEADERS}

 	$(CC) $(CPPFLAGS) $(CFLAGS) $(LDFLAGS) $< ${BASIC_FILES} ${EXTRA_FILES} $(LOADLIBES) $(LDLIBS) -o ib_$@

+install: all

+	$(INSTALL) -d $(DESTDIR)$(BINDIR)

+	$(INSTALL) -m 755 -s $(BINS) $(DESTDIR)$(BINDIR)

+

 clean:

 	$(foreach fname,${RDMACM_TESTS}, rm -f ${fname})

 	$(foreach fname,${MCAST_TESTS}, rm -f ib_${fname})

diff -r -u /tmp/perftest-1.3.0/clock_test.c perftest-1.3.0/clock_test.c

--- /tmp/perftest-1.3.0/clock_test.c	Sun Nov  1 03:09:16 2009

+++ perftest-1.3.0/clock_test.c	Fri Sep 30 08:08:29 2011

@@ -20,6 +20,10 @@

 		c1 = get_cycles();

 		sleep(1);

 		c2 = get_cycles();

+#if defined(__sparc)

+		printf("1 sec = %g usec\n", (double)((c2 - c1))/1000);

+#else

 		printf("1 sec = %g usec\n", (c2 - c1) / mhz);

+#endif

 	}

 }

diff -r -u /tmp/perftest-1.3.0/get_clock.c perftest-1.3.0/get_clock.c

--- /tmp/perftest-1.3.0/get_clock.c	Sun Dec 19 06:36:26 2010

+++ perftest-1.3.0/get_clock.c	Fri Feb 11 04:12:46 2011

@@ -45,6 +45,10 @@

 #include <unistd.h>

 #include <stdio.h>

 #include "get_clock.h"

+#if defined(__SVR4) && defined(__sun)

+#include <stdlib.h>

+#include <infiniband/ofa_solaris.h>

+#endif

 #ifndef DEBUG

 #define DEBUG 0

@@ -56,6 +60,8 @@

 #define MEASUREMENTS 200

 #define USECSTEP 10

 #define USECSTART 100

+#define NSECSTART 100000

+#define NSECSTEP 10000

 /*

  Use linear regression to calculate cycles per microsecond.

@@ -63,14 +69,14 @@

 */

 static double sample_get_cpu_mhz(void)

 {

-	struct timeval tv1, tv2;

+        struct timeval tv1, tv2;

 	cycles_t start;

 	double sx = 0, sy = 0, sxx = 0, syy = 0, sxy = 0;

-	double tx, ty;

+	cycles_t tx, ty;

 	int i;

 	/* Regression: y = a + b x */

-	long x[MEASUREMENTS];

+	cycles_t x[MEASUREMENTS];

 	cycles_t y[MEASUREMENTS];

 	double a; /* system call overhead in cycles */

 	double b; /* cycles per microsecond */

@@ -78,7 +84,6 @@

 	for (i = 0; i < MEASUREMENTS; ++i) {

 		start = get_cycles();

-

 		if (gettimeofday(&tv1, NULL)) {

 			fprintf(stderr, "gettimeofday failed.\n");

 			return 0;

@@ -86,7 +91,7 @@

 		do {

 			if (gettimeofday(&tv2, NULL)) {

-				fprintf(stderr, "gettimeofday failed.\n");

+			        fprintf(stderr, "gettimeofday failed.\n");

 				return 0;

 			}

 		} while ((tv2.tv_sec - tv1.tv_sec) * 1000000 +

@@ -94,9 +99,10 @@

 		x[i] = (tv2.tv_sec - tv1.tv_sec) * 1000000 +

 			tv2.tv_usec - tv1.tv_usec;

+

 		y[i] = get_cycles() - start;

 		if (DEBUG_DATA)

-			fprintf(stderr, "x=%ld y=%Ld\n", x[i], (long long)y[i]);

+			fprintf(stderr, "x=%lld y=%lld\n", x[i], y[i]);

 	}

 	for (i = 0; i < MEASUREMENTS; ++i) {

@@ -134,10 +140,15 @@

 static double proc_get_cpu_mhz(int no_cpu_freq_fail)

 {

+#if !(defined(__SVR4) && defined(__sun))

 	FILE* f;

 	char buf[256];

+#else

+	sol_cpu_info_t	*info;

+#endif

 	double mhz = 0.0;

+#if !(defined(__SVR4) && defined(__sun))

 	f = fopen("/proc/cpuinfo","r");

 	if (!f)

 		return 0.0;

@@ -174,6 +185,13 @@

 		}

 	}

 	fclose(f);

+#else

+	if (sol_get_cpu_info(&info) > 0)

+		mhz = info[0].cpu_mhz;

+	else

+		return (0.0);

+	free(info);

+#endif

 	return mhz;

 }

@@ -184,8 +202,9 @@

 	sample = sample_get_cpu_mhz();

 	proc = proc_get_cpu_mhz(no_cpu_freq_fail);

-	if (!proc || !sample)

-		return 0;

+#if defined(__sparc)

+	return proc;

+#endif

 	delta = proc > sample ? proc - sample : sample - proc;

 	if (delta / proc > 0.01) {

@@ -194,5 +213,6 @@

 					sample, proc);

 			return sample;

 	}

+

 	return proc;

 }

diff -r -u /tmp/perftest-1.3.0/get_clock.h perftest-1.3.0/get_clock.h

--- /tmp/perftest-1.3.0/get_clock.h	Sun Nov  1 03:09:16 2009

+++ perftest-1.3.0/get_clock.h	Fri Feb 11 04:12:46 2011

@@ -36,8 +36,18 @@

 #ifndef GET_CLOCK_H

 #define GET_CLOCK_H

+#if defined(__sparc)

+#include <sys/times.h>

+#include <limits.h>

+#include <sys/time.h>

+typedef unsigned long long cycles_t;

-#if defined (__x86_64__) || defined(__i386__)

+static inline cycles_t get_cycles()

+{

+        return (gethrtime());

+}

+

+#elif defined (__x86_64__) || defined(__i386__)

 /* Note: only x86 CPUs which have rdtsc instruction are supported. */

 typedef unsigned long long cycles_t;

 static inline cycles_t get_cycles()

diff -r -u /tmp/perftest-1.3.0/rdma_bw.c perftest-1.3.0/rdma_bw.c

--- /tmp/perftest-1.3.0/rdma_bw.c	Wed Apr  7 09:44:56 2010

+++ perftest-1.3.0/rdma_bw.c	Fri Feb 11 04:12:46 2011

@@ -50,8 +50,11 @@

 #include <malloc.h>

 #include <getopt.h>

 #include <arpa/inet.h>

+#if !(defined(__SVR4) && defined(__sun))

 #include <byteswap.h>

+#endif

 #include <time.h>

+#include <inttypes.h>

 #include <infiniband/verbs.h>

 #include <rdma/rdma_cma.h>

 		fprintf(stderr, "%d:%s: %s for %s:%d\n", 

-				pid, __func__, gai_strerror(n),

+				(int)pid, __func__, gai_strerror(n),

 				data->servername, data->port);

 		goto err4;

 	}

@@ -157,7 +160,7 @@

 		if (rdma_resolve_addr(data->cm_id, NULL,

 					 (struct sockaddr *)&sin, 2000)) {

 			fprintf(stderr, "%d:%s: rdma_resolve_addr failed\n",

-					 pid, __func__ );

+					 (int)pid, __func__ );

 			goto err2;

 		}

@@ -172,7 +175,7 @@

 		if (event->event != RDMA_CM_EVENT_ADDR_RESOLVED) {

 			fprintf(stderr, "%d:%s: unexpected CM event %d\n", 

-				pid, __func__, event->event);

+				(int)pid, __func__, event->event);

 			goto err1;

 		}

 		rdma_ack_cm_event(event);

@@ -180,7 +183,7 @@

 retry_route:

 		if (rdma_resolve_route(data->cm_id, 2000)) {

 			fprintf(stderr, "%d:%s: rdma_resolve_route failed\n", 

-						pid, __func__);

+						(int)pid, __func__);

 			goto err2;

 		}

@@ -195,7 +198,7 @@

 		if (event->event != RDMA_CM_EVENT_ROUTE_RESOLVED) {

 			fprintf(stderr, "%d:%s: unexpected CM event %d\n", 

-					pid, __func__, event->event);

+					(int)pid, __func__, event->event);

 			rdma_ack_cm_event(event);

 			goto err1;

 		}

@@ -202,7 +205,7 @@

 		rdma_ack_cm_event(event);

 		ctx = pp_init_ctx(data->cm_id, data);

 		if (!ctx) {

-			fprintf(stderr, "%d:%s: pp_init_ctx failed\n", pid, __func__);

+			fprintf(stderr, "%d:%s: pp_init_ctx failed\n", (int)pid, __func__);

 			goto err2;

 		}

 		data->my_dest.psn = lrand48() & 0xffffff;

@@ -218,7 +221,7 @@

 		conn_param.private_data_len = sizeof(data->my_dest);

 		if (rdma_connect(data->cm_id, &conn_param)) {

-			fprintf(stderr, "%d:%s: rdma_connect failure\n", pid, __func__);

+			fprintf(stderr, "%d:%s: rdma_connect failure\n", (int)pid, __func__);

 			goto err2;

 		}

@@ -227,13 +230,13 @@

 		if (event->event != RDMA_CM_EVENT_ESTABLISHED) {

 			fprintf(stderr, "%d:%s: unexpected CM event %d\n", 

- 					pid, __func__, event->event);

+ 					(int)pid, __func__, event->event);

 			goto err1;

 		}

 		if (!event->param.conn.private_data || 

 		    (event->param.conn.private_data_len < sizeof(*data->rem_dest))) {

 			fprintf(stderr, "%d:%s: bad private data ptr %p len %d\n",  

-				pid, __func__, event->param.conn.private_data, 

+				(int)pid, __func__, event->param.conn.private_data, 

 				event->param.conn.private_data_len);

 			goto err1;

 		}

@@ -257,7 +260,7 @@

 		}

 		if (sockfd < 0) {

 			fprintf(stderr, "%d:%s: Couldn't connect to %s:%d\n", 

-				 pid, __func__, data->servername, data->port);

+				 (int)pid, __func__, data->servername, data->port);

 			goto err3;

 		}

 		ctx = pp_init_ctx(data->ib_dev, data);

@@ -287,13 +290,13 @@

 	int parsed;

 	if (!data->use_cma) {

-		sprintf(msg, "%04x:%06x:%06x:%08x:%016Lx", data->my_dest.lid, 

+		sprintf(msg, "%04x:%06x:%06x:%08x:%016llx", data->my_dest.lid, 

 				data->my_dest.qpn, data->my_dest.psn,

 				data->my_dest.rkey, data->my_dest.vaddr);

 		if (write(data->sockfd, msg, sizeof msg) != sizeof msg) {

 			perror("client write");

 			fprintf(stderr, "%d:%s: Couldn't send local address\n", 

-					pid, __func__);

+					(int)pid, __func__);

 			goto err;

 		}

@@ -300,7 +303,7 @@

 		if (read(data->sockfd, msg, sizeof msg) != sizeof msg) {

 			perror("client read");

 			fprintf(stderr, "%d:%s: Couldn't read remote address\n", 

-					pid, __func__);

+					(int)pid, __func__);

 			goto err;

 		}

@@ -310,13 +313,13 @@

 		if (!data->rem_dest)

 			goto err;

-		parsed = sscanf(msg, "%x:%x:%x:%x:%Lx", &data->rem_dest->lid,

+		parsed = sscanf(msg, "%x:%x:%x:%x:%llx", &data->rem_dest->lid,

 				&data->rem_dest->qpn, &data->rem_dest->psn,

 				&data->rem_dest->rkey, &data->rem_dest->vaddr);

 		if (parsed != 5) {

 			fprintf(stderr, "%d:%s: Couldn't parse line <%.*s>\n",

-					pid, __func__, (int)sizeof msg, msg);

+					(int)pid, __func__, (int)sizeof msg, msg);

 			free(data->rem_dest);

 			goto err;

 		}

 		goto err5;

-		fprintf(stderr, "%d:%s: %s for port %d\n", pid, __func__, 

+		fprintf(stderr, "%d:%s: %s for port %d\n", (int)pid, __func__, 

 					gai_strerror(n), data->port);

 		goto err5;

 	}

@@ -357,12 +360,12 @@

 		sin.sin_family = AF_INET;

 		sin.sin_port = htons(data->port);

 		if (rdma_bind_addr(data->cm_id, (struct sockaddr *)&sin)) {

-			fprintf(stderr, "%d:%s: rdma_bind_addr failed\n", pid, __func__);

+			fprintf(stderr, "%d:%s: rdma_bind_addr failed\n", (int)pid, __func__);

 			goto err3;

 		}

 		if (rdma_listen(data->cm_id, 0)) {

-			fprintf(stderr, "%d:%s: rdma_listen failed\n", pid, __func__);

+			fprintf(stderr, "%d:%s: rdma_listen failed\n", (int)pid, __func__);

 			goto err3;

 		}

@@ -371,13 +374,13 @@

 		if (event->event != RDMA_CM_EVENT_CONNECT_REQUEST) {

 			fprintf(stderr, "%d:%s: bad event waiting for connect request %d\n", 

-				pid, __func__, event->event);

+				(int)pid, __func__, event->event);

 			goto err2;

 		}

 		if (!event->param.conn.private_data ||

 		    (event->param.conn.private_data_len < sizeof(*data->rem_dest))) {

-			fprintf(stderr, "%d:%s: bad private data len %d\n", pid,

+			fprintf(stderr, "%d:%s: bad private data len %d\n", (int)pid,

 				__func__, event->param.conn.private_data_len);

 			goto err2;

 		}

@@ -405,18 +408,18 @@

 		conn_param.private_data = &data->my_dest;

 		conn_param.private_data_len = sizeof(data->my_dest);

 		if (rdma_accept(child_cm_id, &conn_param)) {

-			fprintf(stderr, "%d:%s: rdma_accept failed\n", pid, __func__);

+			fprintf(stderr, "%d:%s: rdma_accept failed\n", (int)pid, __func__);

 			goto err1;

 		}	

 		rdma_ack_cm_event(event);

 		if (rdma_get_cm_event(data->cm_channel, &event)) {

-			fprintf(stderr, "%d:%s: rdma_get_cm_event error\n", pid, __func__);

+			fprintf(stderr, "%d:%s: rdma_get_cm_event error\n", (int)pid, __func__);

 			rdma_destroy_id(child_cm_id);

 			goto err3;

 		}

 		if (event->event != RDMA_CM_EVENT_ESTABLISHED) {

 			fprintf(stderr, "%d:%s: bad event waiting for established %d\n", 

-				pid, __func__, event->event);

+				(int)pid, __func__, event->event);

 			goto err1;

 		}

 		rdma_ack_cm_event(event);	

@@ -436,7 +439,7 @@

 		}

 		if (sockfd < 0) {

-			fprintf(stderr, "%d:%s: Couldn't listen to port %d\n", pid,

+			fprintf(stderr, "%d:%s: Couldn't listen to port %d\n", (int)pid,

 						__func__, data->port);

 			goto err4;

 		}

@@ -445,7 +448,7 @@

 		connfd = accept(sockfd, NULL, 0);

 		if (connfd < 0) {

 			perror("server accept");

-			fprintf(stderr, "%d:%s: accept() failed\n", pid, __func__);

+			fprintf(stderr, "%d:%s: accept() failed\n", (int)pid, __func__);

 			close(sockfd);

 			goto err4;

 		}

@@ -485,7 +488,7 @@

 		if (n != sizeof msg) {

 			perror("server read");

 			fprintf(stderr, "%d:%s: %d/%d Couldn't read remote address\n", 

-						pid, __func__, n, (int) sizeof msg);

+						(int)pid, __func__, n, (int) sizeof msg);

 			goto err;

 		}

@@ -495,23 +498,23 @@

 		if (!data->rem_dest)

 			goto err;

-		parsed = sscanf(msg, "%x:%x:%x:%x:%Lx", &data->rem_dest->lid,

+		parsed = sscanf(msg, "%x:%x:%x:%x:%llx", &data->rem_dest->lid,

 			      &data->rem_dest->qpn, &data->rem_dest->psn,

 			      &data->rem_dest->rkey, &data->rem_dest->vaddr);

 		if (parsed != 5) {

-			fprintf(stderr, "%d:%s: Couldn't parse line <%.*s>\n", pid,

+			fprintf(stderr, "%d:%s: Couldn't parse line <%.*s>\n", (int)pid,

 						 __func__, (int)sizeof msg, msg);

 			free(data->rem_dest);

 			goto err;

 		}

-		sprintf(msg, "%04x:%06x:%06x:%08x:%016Lx", data->my_dest.lid,

+		sprintf(msg, "%04x:%06x:%06x:%08x:%016llx", data->my_dest.lid,

 					 data->my_dest.qpn, data->my_dest.psn,

 					 data->my_dest.rkey, data->my_dest.vaddr);

 		if (write(data->sockfd, msg, sizeof msg) != sizeof msg) {

 			perror("server write");

 			fprintf(stderr, "%d:%s: Couldn't send local address\n", 

-					pid, __func__);

+					(int)pid, __func__);

 			free(data->rem_dest);

 			goto err;

 		}

@@ -538,7 +541,7 @@

 	ctx->buf = memalign(page_size, ctx->size * 2);

 	if (!ctx->buf) {

 		fprintf(stderr, "%d:%s: Couldn't allocate work buf.\n",

-					 pid, __func__);

+					 (int)pid, __func__);

 		return NULL;

 	}

@@ -548,7 +551,7 @@

 		cm_id = (struct rdma_cm_id *)ptr;

 		ctx->context = cm_id->verbs;

 		if (!ctx->context) {

-			fprintf(stderr, "%d:%s: Unbound cm_id!!\n", pid, 

+			fprintf(stderr, "%d:%s: Unbound cm_id!!\n", (int)pid, 

 							__func__);

 			return NULL;

 		}

@@ -558,7 +561,7 @@

 		ctx->context = ibv_open_device(ib_dev);

 		if (!ctx->context) {

 			fprintf(stderr, "%d:%s: Couldn't get context for %s\n", 

-				pid, __func__, ibv_get_device_name(ib_dev));

+				(int)pid, __func__, ibv_get_device_name(ib_dev));

 			return NULL;

 		}

 	}

@@ -565,7 +568,7 @@

 	ctx->pd = ibv_alloc_pd(ctx->context);

 	if (!ctx->pd) {

-		fprintf(stderr, "%d:%s: Couldn't allocate PD\n", pid, __func__);

+		fprintf(stderr, "%d:%s: Couldn't allocate PD\n", (int)pid, __func__);

 		return NULL;

 	}

@@ -575,7 +578,7 @@

 	ctx->mr = ibv_reg_mr(ctx->pd, ctx->buf, ctx->size * 2,

 			     IBV_ACCESS_REMOTE_WRITE | IBV_ACCESS_LOCAL_WRITE);

 	if (!ctx->mr) {

-		fprintf(stderr, "%d:%s: Couldn't allocate MR\n", pid, __func__);

+		fprintf(stderr, "%d:%s: Couldn't allocate MR\n", (int)pid, __func__);

 		return NULL;

 	}