diff --git a/VERSION b/VERSION
index daf0e897c0..87a9f53905 100644
--- a/VERSION
+++ b/VERSION
@@ -1,5 +1,6 @@
 # Copyright (c) 2008      Sun Microsystems, Inc.  All rights reserved.
 # Copyright (c) 2008-2011 Cisco Systems, Inc.  All rights reserved.
+# Copyright (c) 2011      NVIDIA Corporation.  All rights reserved.
 
 # This is the VERSION file for Open MPI, describing the precise
 # version of Open MPI in this distribution.  The various components of
@@ -102,4 +103,5 @@ libopen_pal_so_version=0:0:0
 libmca_common_sm_so_version=0:0:0
 libmca_common_mx_so_version=0:0:0
 libmca_common_portals_so_version=0:0:0
+libmca_common_cuda_so_version=0:0:0
 libmca_opal_common_hwloc_so_version=0:0:0
diff --git a/configure.ac b/configure.ac
index 78f33903d5..85c2eb087b 100644
--- a/configure.ac
+++ b/configure.ac
@@ -15,6 +15,7 @@
 # Copyright (c) 2006-2011 Los Alamos National Security, LLC.  All rights
 #                         reserved.
 # Copyright (c) 2009      Oak Ridge National Labs.  All rights reserved.
+# Copyright (c) 2011      NVIDIA Corporation.  All rights reserved.
 # $COPYRIGHT$
 # 
 # Additional copyrights may follow
@@ -129,6 +130,7 @@ m4_ifdef([project_ompi],
           # right now.
           AC_SUBST(libmca_common_sm_so_version)
           AC_SUBST(libmca_common_mx_so_version)
+          AC_SUBST(libmca_common_cuda_so_version)
           AC_SUBST(libmca_common_portals_so_version)])
 m4_ifdef([project_orte],
          [AC_SUBST(libopen_rte_so_version)])
diff --git a/ompi/class/ompi_free_list.c b/ompi/class/ompi_free_list.c
index 6793399f7d..d468a705be 100644
--- a/ompi/class/ompi_free_list.c
+++ b/ompi/class/ompi_free_list.c
@@ -12,6 +12,7 @@
  *                         All rights reserved.
  * Copyright (c) 2006-2007 Mellanox Technologies. All rights reserved.
  * Copyright (c) 2010      Cisco Systems, Inc. All rights reserved.
+ * Copyright (c) 2011      NVIDIA Corporation.  All rights reserved.
  * $COPYRIGHT$
  * 
  * Additional copyrights may follow
@@ -202,7 +203,7 @@ int ompi_free_list_grow(ompi_free_list_t* flist, size_t num_elements)
         if(elem_size != 0) {
             mpool_alloc_ptr = (unsigned char *) flist->fl_mpool->mpool_alloc(flist->fl_mpool,
                    num_elements * elem_size, flist->fl_payload_buffer_alignment,
-                   MCA_MPOOL_FLAGS_CACHE_BYPASS, &reg);
+                   MCA_MPOOL_FLAGS_CACHE_BYPASS | MCA_MPOOL_FLAGS_CUDA_REGISTER_MEM, &reg);
             if(NULL == mpool_alloc_ptr) {
                 free(alloc_ptr);
                 return OMPI_ERR_TEMP_OUT_OF_RESOURCE;
diff --git a/ompi/mca/btl/openib/Makefile.am b/ompi/mca/btl/openib/Makefile.am
index 3a6c3b7cf6..36ae1f8e19 100644
--- a/ompi/mca/btl/openib/Makefile.am
+++ b/ompi/mca/btl/openib/Makefile.am
@@ -11,6 +11,7 @@
 #                         All rights reserved.
 # Copyright (c) 2007-2010 Cisco Systems, Inc.  All rights reserved.
 # Copyright (c) 2010      Oracle and/or its affiliates.  All rights reserved.
+# Copyright (c) 2011      NVIDIA Corporation.  All rights reserved.
 # $COPYRIGHT$
 #
 # Additional copyrights may follow
@@ -118,6 +119,10 @@ mcacomponent_LTLIBRARIES = $(component)
 mca_btl_openib_la_SOURCES = $(component_sources)
 mca_btl_openib_la_LDFLAGS = -module -avoid-version $(btl_openib_LDFLAGS)
 mca_btl_openib_la_LIBADD = $(btl_openib_LIBS)
+if MCA_ompi_cuda_support
+mca_btl_openib_la_LIBADD += \
+    $(top_ompi_builddir)/ompi/mca/common/cuda/libmca_common_cuda.la
+endif
 
 noinst_LTLIBRARIES = $(lib)
 libmca_btl_openib_la_SOURCES = $(lib_sources)
diff --git a/ompi/mca/btl/openib/btl_openib_component.c b/ompi/mca/btl/openib/btl_openib_component.c
index 84c8914ca5..385ab42e71 100644
--- a/ompi/mca/btl/openib/btl_openib_component.c
+++ b/ompi/mca/btl/openib/btl_openib_component.c
@@ -16,6 +16,7 @@
  *                         reserved.
  * Copyright (c) 2006-2007 Voltaire All rights reserved.
  * Copyright (c) 2009-2011 Oracle and/or its affiliates.  All rights reserved.
+ * Copyright (c) 2011      NVIDIA Corporation.  All rights reserved.
  * $COPYRIGHT$
  *
  * Additional copyrights may follow
@@ -70,6 +71,7 @@ const char *ibv_get_sysfs_path(void);
 #include "ompi/constants.h"
 #include "ompi/proc/proc.h"
 #include "ompi/mca/btl/btl.h"
+#include "ompi/mca/common/cuda/common_cuda.h"
 #include "ompi/mca/mpool/base/base.h"
 #include "ompi/mca/mpool/rdma/mpool_rdma.h"
 #include "ompi/mca/btl/base/base.h"
@@ -542,6 +544,13 @@ static int openib_reg_mr(void *reg_data, void *base, size_t size,
         return OMPI_ERR_OUT_OF_RESOURCE;
     }
 
+#if OMPI_CUDA_SUPPORT
+    if (reg->flags & MCA_MPOOL_FLAGS_CUDA_REGISTER_MEM) {
+        mca_common_cuda_register(base, size,
+            openib_reg->base.mpool->mpool_component->mpool_version.mca_component_name);
+    }
+#endif
+
     return OMPI_SUCCESS;
 }
 
@@ -555,6 +564,14 @@ static int openib_dereg_mr(void *reg_data, mca_mpool_base_registration_t *reg)
                        __func__, strerror(errno)));
             return OMPI_ERROR;
         }
+
+#if OMPI_CUDA_SUPPORT
+        if (reg->flags & MCA_MPOOL_FLAGS_CUDA_REGISTER_MEM) {
+            mca_common_cuda_unregister(openib_reg->base.base,
+                openib_reg->base.mpool->mpool_component->mpool_version.mca_component_name);
+        }
+#endif
+
     }
     openib_reg->mr = NULL;
     return OMPI_SUCCESS;
diff --git a/ompi/mca/common/cuda/Makefile.am b/ompi/mca/common/cuda/Makefile.am
new file mode 100644
index 0000000000..d0deab6197
--- /dev/null
+++ b/ompi/mca/common/cuda/Makefile.am
@@ -0,0 +1,110 @@
+#
+# Copyright (c) 2004-2005 The Trustees of Indiana University and Indiana
+#                         University Research and Technology
+#                         Corporation.  All rights reserved.
+# Copyright (c) 2004-2005 The University of Tennessee and The University
+#                         of Tennessee Research Foundation.  All rights
+#                         reserved.
+# Copyright (c) 2004-2009 High Performance Computing Center Stuttgart, 
+#                         University of Stuttgart.  All rights reserved.
+# Copyright (c) 2004-2005 The Regents of the University of California.
+#                         All rights reserved.
+# Copyright (c) 2011      NVIDIA Corporation.  All rights reserved.
+# $COPYRIGHT$
+# 
+# Additional copyrights may follow
+# 
+# $HEADER$
+#
+
+# A word of explanation...
+#
+# This library is linked against various MCA components because the
+# support for CUDA device pointers is needed in various places.
+# There's two cases:
+#
+# 1. libmca_common_cuda.la is a shared library.  By linking that shared
+# library to all components that need it, the OS linker will
+# automatically load it into the process as necessary, and there will
+# only be one copy (i.e., all the components will share *one* copy of
+# the code and data).
+#
+# 2. libmca_common_cuda.la is a static library.  In this case, it will
+# be rolled up into the top-level libmpi.la.  It will also be rolled
+# into each component, but then the component will also be rolled up
+# into the upper-level libmpi.la.  Linkers universally know how to
+# "figure this out" so that we end up with only one copy of the code
+# and data.
+#
+# Note that building this common component statically and linking
+# against other dynamic components is *not* supported!
+
+AM_CPPFLAGS = $(common_cuda_CPPFLAGS)
+
+# Header files
+
+headers = \
+        common_cuda.h
+
+# Source files
+
+sources = \
+        common_cuda.c
+
+dist_pkgdata_DATA = help-mpi-common-cuda.txt
+
+# As per above, we'll either have an installable or noinst result.
+# The installable one should follow the same MCA prefix naming rules
+# (i.e., libmca_<type>_<name>.la).  The noinst one can be named
+# whatever it wants, although libmca_<type>_<name>_noinst.la is
+# recommended.
+
+# To simplify components that link to this library, we will *always*
+# have an output libtool library named libmca_<type>_<name>.la -- even
+# for case 2) described above (i.e., so there's no conditional logic
+# necessary in component Makefile.am's that link to this library).
+# Hence, if we're creating a noinst version of this library (i.e.,
+# case 2), we sym link it to the libmca_<type>_<name>.la name
+# (libtool will do the Right Things under the covers).  See the
+# all-local and clean-local rules, below, for how this is effected.
+
+lib_LTLIBRARIES =
+noinst_LTLIBRARIES =
+comp_inst = libmca_common_cuda.la
+comp_noinst = libmca_common_cuda_noinst.la
+
+if MCA_BUILD_ompi_common_cuda_DSO
+lib_LTLIBRARIES += $(comp_inst)
+else
+noinst_LTLIBRARIES += $(comp_noinst)
+endif
+
+libmca_common_cuda_la_SOURCES = $(headers) $(sources)
+libmca_common_cuda_la_LDFLAGS = -version-info $(libmca_common_cuda_so_version)
+libmca_common_cuda_la_LIBADD = $(common_cuda_LIBS)
+
+libmca_common_cuda_noinst_la_SOURCES = $(libmca_common_cuda_la_SOURCES)
+libmca_common_cuda_noinst_la_LDFLAGS = $(common_cuda_LDFLAGS)
+libmca_common_cuda_noinst_la_LIBADD = $(common_cuda_LIBS)
+
+# Conditionally install the header files
+
+if WANT_INSTALL_HEADERS
+ompidir = $(includedir)/openmpi/$(subdir)
+ompi_HEADERS = $(headers)
+endif
+
+# These two rules will sym link the "noinst" libtool library filename
+# to the installable libtool library filename in the case where we are
+# compiling this component statically (case 2), described above).
+
+all-local:
+	if test -z "$(lib_LTLIBRARIES)"; then \
+	  rm -f "$(comp_inst)"; \
+	  $(LN_S) "$(comp_noinst)" "$(comp_inst)"; \
+	fi
+
+clean-local:
+	if test -z "$(lib_LTLIBRARIES)"; then \
+	  rm -f "$(comp_inst)"; \
+	fi
diff --git a/ompi/mca/common/cuda/common_cuda.c b/ompi/mca/common/cuda/common_cuda.c
new file mode 100644
index 0000000000..6e8dfdefb1
--- /dev/null
+++ b/ompi/mca/common/cuda/common_cuda.c
@@ -0,0 +1,149 @@
+/*
+ * Copyright (c) 2004-2006 The Trustees of Indiana University and Indiana
+ *                         University Research and Technology
+ *                         Corporation.  All rights reserved.
+ * Copyright (c) 2004-2005 The University of Tennessee and The University
+ *                         of Tennessee Research Foundation.  All rights
+ *                         reserved.
+ * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
+ *                         University of Stuttgart.  All rights reserved.
+ * Copyright (c) 2004-2006 The Regents of the University of California.
+ *                         All rights reserved.
+ * Copyright (c) 2011      NVIDIA Corporation.  All rights reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+
+#include "ompi_config.h"
+
+#include <errno.h>
+#include <unistd.h>
+#include <cuda.h>
+
+#include "opal/align.h"
+#include "opal/mca/base/mca_base_param.h"
+#include "opal/util/output.h"
+#include "orte/util/show_help.h"
+#include "common_cuda.h"
+
+static bool initialized = false;
+static int mca_common_cuda_verbose;
+static int mca_common_cuda_output = 0;
+static bool mca_common_cuda_enabled = false;
+static bool mca_common_cuda_register_memory = true;
+static bool mca_common_cuda_warning = true;
+
+void mca_common_cuda_init(void)
+{
+    int id, value;
+    CUresult res;
+    CUcontext cuContext;
+
+    if (initialized) {
+        return;
+    }
+
+    /* Set different levels of verbosity in the cuda related code. */
+    id = mca_base_param_reg_int_name("mpi", "common_cuda_verbose", 
+                                     "Set level of common cuda verbosity",
+                                     false, false, 0, &mca_common_cuda_verbose);
+    mca_common_cuda_output = opal_output_open(NULL);
+    opal_output_set_verbosity(mca_common_cuda_output, mca_common_cuda_verbose);
+
+    /* Control whether system buffers get CUDA pinned or not.  Allows for 
+     * performance analysis. */
+    id = mca_base_param_reg_int_name("mpi", "common_cuda_register_memory",
+                                     "Whether to cuMemHostRegister preallocated BTL buffers",
+                                     false, false, 
+                                     (int) mca_common_cuda_register_memory, &value);
+    mca_common_cuda_register_memory = OPAL_INT_TO_BOOL(value);
+
+    /* Control whether we see warnings when CUDA memory registration fails.  This is
+     * useful when CUDA support is configured in, but we are running a regular MPI
+     * application without CUDA. */
+    id = mca_base_param_reg_int_name("mpi", "common_cuda_warning",
+                                     "Whether to print warnings when CUDA registration fails",
+                                     false, false, 
+                                     (int) mca_common_cuda_warning, &value);
+    mca_common_cuda_warning = OPAL_INT_TO_BOOL(value);
+
+    /* Check to see if this process is running in a CUDA context.  If
+     * so, all is good.  If not, then disable CUDA support. */
+    res = cuCtxGetCurrent(&cuContext);
+    if (CUDA_SUCCESS != res) {
+        if (mca_common_cuda_warning) {
+            orte_show_help("help-mpi-common-cuda.txt", "cuCtxGetCurrent failed",
+                           true, res);
+        }
+        mca_common_cuda_enabled = false;
+        mca_common_cuda_register_memory = false;
+        initialized = true;
+        return;
+    } else {
+        mca_common_cuda_enabled = true;
+        opal_output_verbose(20, mca_common_cuda_output,
+                            "CUDA: cuCtxGetCurrent succeeded");
+    }
+
+    opal_output_verbose(30, mca_common_cuda_output,
+                        "CUDA: initialized");
+    initialized = true;
+}
+
+
+/**
+ * Call the CUDA register function so we pin the memory in the CUDA
+ * space.
+ */
+void mca_common_cuda_register(void *ptr, size_t amount, char *msg) {
+    int res;
+
+    if (!initialized) {
+        mca_common_cuda_init();
+    }
+
+    if (mca_common_cuda_enabled && mca_common_cuda_register_memory) {
+        res = cuMemHostRegister(ptr, amount, 0);
+        if (res != CUDA_SUCCESS) {
+            /* If registering the memory fails, print a message and continue.
+             * This is not a fatal error. */
+            orte_show_help("help-mpi-common-cuda.txt", "cuMemHostRegister failed",
+                           true, ptr, amount, res, msg);
+        } else {
+            opal_output_verbose(20, mca_common_cuda_output,
+                                "CUDA: cuMemHostRegister OK on mpool %s: "
+                                "address=%p, bufsize=%d",
+                                msg, ptr, (int)amount);
+        }
+    }
+}
+
+/**
+ * Call the CUDA unregister function so we unpin the memory in the CUDA
+ * space.
+ */
+void mca_common_cuda_unregister(void *ptr, char *msg) {
+    int res;
+
+    if (!initialized) {
+        mca_common_cuda_init();
+    }
+
+    if (mca_common_cuda_enabled && mca_common_cuda_register_memory) {
+        res = cuMemHostUnregister(ptr);
+        if (res != CUDA_SUCCESS) {
+            /* If unregistering the memory fails, print a message and continue.
+             * This is not a fatal error. */
+            orte_show_help("help-mpi-common-cuda.txt", "cuMemHostUnregister failed",
+                           true, ptr, res, msg);
+        } else {
+            opal_output_verbose(20, mca_common_cuda_output,
+                                "CUDA: cuMemHostUnregister OK on mpool %s: "
+                                "address=%p",
+                                msg, ptr);
+        }
+    }
+}
diff --git a/ompi/mca/common/cuda/common_cuda.h b/ompi/mca/common/cuda/common_cuda.h
new file mode 100644
index 0000000000..b23efbe934
--- /dev/null
+++ b/ompi/mca/common/cuda/common_cuda.h
@@ -0,0 +1,29 @@
+/*
+ * Copyright (c) 2004-2006 The Trustees of Indiana University and Indiana
+ *                         University Research and Technology
+ *                         Corporation.  All rights reserved.
+ * Copyright (c) 2004-2005 The University of Tennessee and The University
+ *                         of Tennessee Research Foundation.  All rights
+ *                         reserved.
+ * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
+ *                         University of Stuttgart.  All rights reserved.
+ * Copyright (c) 2004-2006 The Regents of the University of California.
+ *                         All rights reserved.
+ * Copyright (c) 2011      NVIDIA Corporation.  All rights reserved.
+ * $COPYRIGHT$
+ *
+ * Additional copyrights may follow
+ *
+ * $HEADER$
+ */
+
+#ifndef OMPI_MCA_COMMON_CUDA_H
+#define OMPI_MCA_COMMON_CUDA_H
+
+OMPI_DECLSPEC void mca_common_cuda_init(void);
+
+OMPI_DECLSPEC void mca_common_cuda_register(void *ptr, size_t amount, char *msg);
+
+OMPI_DECLSPEC void mca_common_cuda_unregister(void *ptr, char *msg);
+
+#endif /* OMPI_MCA_COMMON_CUDA_H */
diff --git a/ompi/mca/common/cuda/configure.m4 b/ompi/mca/common/cuda/configure.m4
new file mode 100644
index 0000000000..60acfe4cf2
--- /dev/null
+++ b/ompi/mca/common/cuda/configure.m4
@@ -0,0 +1,34 @@
+# -*- shell-script -*-
+#
+# Copyright (c) 2011      NVIDIA Corporation.  All rights reserved.
+# $COPYRIGHT$
+#
+# Additional copyrights may follow
+#
+# $HEADER$
+#
+
+#
+# If CUDA support was requested, then build the CUDA support library.
+# This code checks the variable CUDA_SUPPORT which was set earlier in
+# the configure sequence by the opal_configure_options.m4 code.
+#
+
+AC_DEFUN([MCA_ompi_common_cuda_CONFIG],[
+    AC_CONFIG_FILES([ompi/mca/common/cuda/Makefile])
+
+    # Use CUDA_SUPPORT which was filled in by the opal configure code.
+    AM_CONDITIONAL([MCA_ompi_cuda_support], [test "x$CUDA_SUPPORT" = "x1"])
+    AC_DEFINE_UNQUOTED([OMPI_CUDA_SUPPORT],$CUDA_SUPPORT,
+                       [Whether we want cuda memory registration support in OMPI code])
+    AS_IF([test "x$CUDA_SUPPORT" = "x1"],
+          [$1],
+          [$2])
+
+    # Copy over the includes and libs needed to build CUDA
+    common_cuda_CPPFLAGS=$opal_datatype_CPPFLAGS
+    common_cuda_LIBS=$opal_datatype_LIBS
+    AC_SUBST([common_cuda_CPPFLAGS])
+    AC_SUBST([common_cuda_LIBS])
+
+])dnl
diff --git a/ompi/mca/common/cuda/help-mpi-common-cuda.txt b/ompi/mca/common/cuda/help-mpi-common-cuda.txt
new file mode 100644
index 0000000000..3fcde6a581
--- /dev/null
+++ b/ompi/mca/common/cuda/help-mpi-common-cuda.txt
@@ -0,0 +1,27 @@
+# -*- text -*-
+#
+# Copyright (c) 2011      NVIDIA.  All rights reserved.
+# $COPYRIGHT$
+# 
+# Additional copyrights may follow
+# 
+# $HEADER$
+#
+[cuCtxGetCurrent failed]
+WARNING: The call to cuCtxGetCurrent() failed while initializing the
+CUDA support.  Support for CUDA registered memory is disabled.
+  cuCtxGetCurrent return value:   %d
+
+NOTE: You can turn off this warning by setting the MCA parameter
+      mpi_common_cuda_warning to 0.
+#
+[cuMemHostRegister failed]
+The call to cuMemHostRegister(%p, %d, 0) failed.
+  cuMemHostRegister return value:   %d
+  Memory Pool: %s
+#
+[cuMemHostUnregister failed]
+The call to cuMemHostUnregister(%p) failed.
+  cuMemHostUnregister return value:   %d
+  Memory Pool: %s
+#
diff --git a/ompi/mca/mpool/mpool.h b/ompi/mca/mpool/mpool.h
index 1542389c33..f5f4ea4ba4 100644
--- a/ompi/mca/mpool/mpool.h
+++ b/ompi/mca/mpool/mpool.h
@@ -11,6 +11,7 @@
   * Copyright (c) 2004-2005 The Regents of the University of California.
   *                         All rights reserved.
   * Copyright (c) 2009      Sun Microsystems, Inc.  All rights reserved.
+  * Copyright (c) 2011      NVIDIA Corporation.  All rights reserved.
   * $COPYRIGHT$
   * 
   * Additional copyrights may follow
@@ -36,6 +37,7 @@
 #define MCA_MPOOL_FLAGS_MPI_ALLOC_MEM 0x4
 #define MCA_MPOOL_FLAGS_INVALID 0x8
 #define MCA_MPOOL_FLAGS_SO_MEM 0x10
+#define MCA_MPOOL_FLAGS_CUDA_REGISTER_MEM 0x20
 
 struct mca_mpool_base_resources_t;
 
diff --git a/ompi/mca/mpool/sm/Makefile.am b/ompi/mca/mpool/sm/Makefile.am
index 6badd207b8..5e8383e688 100644
--- a/ompi/mca/mpool/sm/Makefile.am
+++ b/ompi/mca/mpool/sm/Makefile.am
@@ -10,6 +10,7 @@
 # Copyright (c) 2004-2005 The Regents of the University of California.
 #                         All rights reserved.
 # Copyright (c) 2010      Cisco Systems, Inc.  All rights reserved.
+# Copyright (c) 2011      NVIDIA Corporation.  All rights reserved.
 # $COPYRIGHT$
 # 
 # Additional copyrights may follow
@@ -45,6 +46,10 @@ mca_mpool_sm_la_SOURCES = $(sources)
 mca_mpool_sm_la_LDFLAGS = -module -avoid-version
 mca_mpool_sm_la_LIBADD = \
     $(top_ompi_builddir)/ompi/mca/common/sm/libmca_common_sm.la
+if MCA_ompi_cuda_support
+mca_mpool_sm_la_LIBADD += \
+    $(top_ompi_builddir)/ompi/mca/common/cuda/libmca_common_cuda.la
+endif
 
 noinst_LTLIBRARIES = $(component_noinst)
 libmca_mpool_sm_la_SOURCES = $(sources)
diff --git a/ompi/mca/mpool/sm/mpool_sm_module.c b/ompi/mca/mpool/sm/mpool_sm_module.c
index f573f799c3..3319d2621b 100644
--- a/ompi/mca/mpool/sm/mpool_sm_module.c
+++ b/ompi/mca/mpool/sm/mpool_sm_module.c
@@ -12,6 +12,7 @@
  * Copyright (c) 2009      Cisco Systems, Inc.  All rights reserved.
  * Copyright (c) 2011      Los Alamos National Security, LLC.  
  *                         All rights reserved. 
+ * Copyright (c) 2011      NVIDIA Corporation.  All rights reserved.
  * $COPYRIGHT$
  * 
  * Additional copyrights may follow
@@ -23,6 +24,7 @@
 #include <string.h>
 #include "ompi/mca/mpool/sm/mpool_sm.h"
 #include "ompi/mca/common/sm/common_sm.h"
+#include "ompi/mca/common/cuda/common_cuda.h"
 #ifdef HAVE_UNISTD_H
 #include <unistd.h>
 #endif
@@ -95,6 +97,13 @@ void* mca_mpool_sm_alloc(
         opal_maffinity_base_bind(&mseg, 1, mpool_sm->mem_node);
     }
 
+#if OPAL_CUDA_SUPPORT
+    if (flags & MCA_MPOOL_FLAGS_CUDA_REGISTER_MEM) {
+        mca_common_cuda_register(mseg.mbs_start_addr, size,
+                                 mpool->mpool_component->mpool_version.mca_component_name);
+    }
+#endif
+
     return mseg.mbs_start_addr;
 }
 
diff --git a/ompi/mca/pml/ob1/pml_ob1_sendreq.h b/ompi/mca/pml/ob1/pml_ob1_sendreq.h
index 6b00c96ee1..fd510e4b0a 100644
--- a/ompi/mca/pml/ob1/pml_ob1_sendreq.h
+++ b/ompi/mca/pml/ob1/pml_ob1_sendreq.h
@@ -10,6 +10,7 @@
  * Copyright (c) 2004-2005 The Regents of the University of California.
  *                         All rights reserved.
  * Copyright (c) 2009      Sun Microsystems, Inc.  All rights reserved.
+ * Copyright (c) 2011      NVIDIA Corporation.  All rights reserved.
  * $COPYRIGHT$
  * 
  * Additional copyrights may follow
@@ -394,6 +395,13 @@ mca_pml_ob1_send_request_start_btl( mca_pml_ob1_send_request_t* sendreq,
                                                          MCA_PML_OB1_HDR_FLAGS_CONTIG);
             }
         } else {
+#if OPAL_CUDA_SUPPORT
+            /* Do not send anything with first rendezvous message as copying GPU
+             * memory into RNDV message is expensive. */
+            if (sendreq->req_send.req_base.req_convertor.flags & CONVERTOR_CUDA) {
+                size = 0;
+            }
+#endif
             rc = mca_pml_ob1_send_request_start_rndv(sendreq, bml_btl, size, 0);
         }
     }