v3.0.x: misc fixes for heterogeneous clusters

config/opal_configure_options.m4

@@ -286,7 +286,22 @@ fi
 AC_DEFINE_UNQUOTED(OPAL_ENABLE_DLOPEN_SUPPORT, $OPAL_ENABLE_DLOPEN_SUPPORT,
     [Whether we want to enable dlopen support])
 
-opal_want_heterogeneous=0
+#
+# Heterogeneous support
+#
+
+AC_MSG_CHECKING([if want heterogeneous support])
+AC_ARG_ENABLE([heterogeneous],
+    [AC_HELP_STRING([--enable-heterogeneous],
+                    [Enable features required for heterogeneous
+                     platform support (default: disabled)])])
+if test "$enable_heterogeneous" = "yes" ; then
+     AC_MSG_RESULT([yes])
+     opal_want_heterogeneous=1
+else
+     AC_MSG_RESULT([no])
+     opal_want_heterogeneous=0
+fi
 AC_DEFINE_UNQUOTED([OPAL_ENABLE_HETEROGENEOUS_SUPPORT],
                    [$opal_want_heterogeneous],
                    [Enable features required for heterogeneous support])

configure.ac

@@ -599,7 +599,7 @@ AC_CHECK_HEADERS([alloca.h aio.h arpa/inet.h dirent.h \
     sys/types.h sys/uio.h sys/un.h net/uio.h sys/utsname.h sys/vfs.h sys/wait.h syslog.h \
     termios.h ulimit.h unistd.h util.h utmp.h malloc.h \
     ifaddrs.h crt_externs.h regex.h mntent.h paths.h \
-    ioLib.h sockLib.h hostLib.h shlwapi.h sys/synch.h db.h ndbm.h zlib.h])
+    ioLib.h sockLib.h hostLib.h shlwapi.h sys/synch.h db.h ndbm.h zlib.h ieee754.h])
 
 AC_CHECK_HEADERS([sys/mount.h], [], [],
 [AC_INCLUDES_DEFAULT

opal/datatype/opal_convertor.c

@@ -481,7 +481,9 @@ size_t opal_convertor_compute_remote_size( opal_convertor_t* pConvertor )
     pConvertor->remote_size = pConvertor->local_size;
     if( OPAL_UNLIKELY(datatype->bdt_used & pConvertor->master->hetero_mask) ) {
         pConvertor->flags &= (~CONVERTOR_HOMOGENEOUS);
-        pConvertor->use_desc = &(datatype->desc);
+        if (!(pConvertor->flags & CONVERTOR_SEND && pConvertor->flags & OPAL_DATATYPE_FLAG_CONTIGUOUS)) {
+            pConvertor->use_desc = &(datatype->desc);
+        }
         if( 0 == (pConvertor->flags & CONVERTOR_HAS_REMOTE_SIZE) ) {
             /* This is for a single datatype, we must update it with the count */
             pConvertor->remote_size = opal_datatype_compute_remote_size(datatype,
@@ -570,6 +572,7 @@ int32_t opal_convertor_prepare_for_recv( opal_convertor_t* convertor,
     mca_cuda_convertor_init(convertor, pUserBuf);
 #endif
 
+    assert(! (convertor->flags & CONVERTOR_SEND));
     OPAL_CONVERTOR_PREPARE( convertor, datatype, count, pUserBuf );
 
     if( convertor->flags & CONVERTOR_WITH_CHECKSUM ) {

opal/datatype/opal_convertor.h

@@ -218,6 +218,7 @@ static inline void opal_convertor_get_unpacked_size( const opal_convertor_t* pCo
         return;
     }
     if( 0 == (CONVERTOR_HAS_REMOTE_SIZE & pConv->flags) ) {
+        assert(! (pConv->flags & CONVERTOR_SEND));
         opal_convertor_compute_remote_size( (opal_convertor_t*)pConv);
     }
     *pSize = pConv->remote_size;

opal/datatype/opal_copy_functions_heterogeneous.c

@@ -5,6 +5,7 @@
  *                         reserved.
  * Copyright (c) 2009      Oak Ridge National Labs.  All rights reserved.
  * Copyright (c) 2015-2017 Research Organization for Information Science
+ * Copyright (c) 2015-2017 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  * $COPYRIGHT$
  *
@@ -15,6 +16,10 @@
 
 #include "opal_config.h"
 
+#ifdef HAVE_IEEE754_H
+#include <ieee754.h>
+#endif
+
 #include <stddef.h>
 #include <stdint.h>
 
@@ -62,13 +67,78 @@ opal_dt_swap_bytes(void *to_p, const void *from_p, const size_t size, size_t cou
     }
 }
 
+#ifdef HAVE_IEEE754_H
+struct bit128 {
+    unsigned int mantissa3:32;
+    unsigned int mantissa2:32;
+    unsigned int mantissa1:32;
+    unsigned int mantissa0:16;
+    unsigned int exponent:15;
+    unsigned int negative:1;
+};
+
+struct bit80 {
+    unsigned int pad:32;
+    unsigned int empty:16;
+    unsigned int negative:1;
+    unsigned int exponent:15;
+    unsigned int mantissa0:32;
+    unsigned int mantissa1:32;
+};
+
+static inline void
+opal_dt_swap_long_double(void *to_p, const void *from_p, const size_t size, size_t count, uint32_t remoteArch)
+{
+#ifdef HAVE_IEEE754_H
+    size_t i;
+    long double*to = (long double *) to_p;
+
+    if ((opal_local_arch&OPAL_ARCH_LDISINTEL) && !(remoteArch&OPAL_ARCH_LDISINTEL)) {
+#ifdef __x86_64
+        for (i=0; i<count; i++, to++) {
+            union ieee854_long_double ld;
+            struct bit128 * b = (struct bit128 *)to;
+            ld.ieee.empty = 0;
+            ld.ieee.mantissa0 = 0x80000000 | (((unsigned int)b->mantissa0 << 15) & 0x7FFF8000) | ((b->mantissa1 >> 17) & 0x00007FFF);
+            ld.ieee.mantissa1 = ((b->mantissa1 << 15) & 0xFFFF8000) | ((b->mantissa2 << 17) & 0x000007FFF);
+            ld.ieee.exponent = b->exponent;
+            ld.ieee.negative = b->negative;
+            MEMCPY( to, &ld, sizeof(long double));
+        }
+#endif
+    } else if (!(opal_local_arch&OPAL_ARCH_LDISINTEL) && (remoteArch&OPAL_ARCH_LDISINTEL)) {
+#ifdef __sparcv9
+        for (i=0; i<count; i++, to++) {
+            union ieee854_long_double ld;
+            struct bit80 * b = (struct bit80 *)to;
+            ld.ieee.mantissa3 = 0;
+            ld.ieee.mantissa2 = 0;
+            ld.ieee.mantissa0 = (b->mantissa0 << 1) | (b->mantissa1 & 0x80000000);
+            ld.ieee.mantissa1 = (b->mantissa1 << 1) & 0xFFFFFFFE;
+            ld.ieee.exponent = b->exponent;
+            ld.ieee.negative = b->negative;
+            MEMCPY( to, &ld, sizeof(long double));
+        }
+#endif
+    }
+#else
+    assert(0);
+#endif
+}
+#else
+#define opal_dt_swap_long_double(to_p, from_p, size, count, remoteArch)
+#endif
+
 /**
  * BEWARE: Do not use the following macro with composed types such as
  * complex. As the swap is done using the entire type sizeof, the
  * wrong endianess translation will be done.  Instead, use the
  * COPY_2SAMETYPE_HETEROGENEOUS.
  */
 #define COPY_TYPE_HETEROGENEOUS( TYPENAME, TYPE )                                         \
+            COPY_TYPE_HETEROGENEOUS_INTERNAL( TYPENAME, TYPE, 0 )
+
+#define COPY_TYPE_HETEROGENEOUS_INTERNAL( TYPENAME, TYPE, LONG_DOUBLE )                   \
 static int32_t                                                                            \
 copy_##TYPENAME##_heterogeneous(opal_convertor_t *pConvertor, uint32_t count,             \
                                 const char* from, size_t from_len, ptrdiff_t from_extent, \
@@ -85,9 +155,15 @@ copy_##TYPENAME##_heterogeneous(opal_convertor_t *pConvertor, uint32_t count,
         (opal_local_arch & OPAL_ARCH_ISBIGENDIAN)) {                    \
         if( (to_extent == from_extent) && (to_extent == sizeof(TYPE)) ) { \
             opal_dt_swap_bytes(to, from, sizeof(TYPE), count);          \
+            if (LONG_DOUBLE) {                                          \
+                opal_dt_swap_long_double(to, from, sizeof(TYPE), count, pConvertor->remoteArch);\
+            }                                                           \
         } else {                                                        \
             for( i = 0; i < count; i++ ) {                              \
                 opal_dt_swap_bytes(to, from, sizeof(TYPE), 1);          \
+                if (LONG_DOUBLE) {                                      \
+                    opal_dt_swap_long_double(to, from, sizeof(TYPE), 1, pConvertor->remoteArch);\
+                }                                                       \
                 to += to_extent;                                        \
                 from += from_extent;                                    \
             }                                                           \
@@ -108,6 +184,9 @@ copy_##TYPENAME##_heterogeneous(opal_convertor_t *pConvertor, uint32_t count,
 }
 
 #define COPY_2SAMETYPE_HETEROGENEOUS( TYPENAME, TYPE )                                         \
+            COPY_2SAMETYPE_HETEROGENEOUS_INTERNAL( TYPENAME, TYPE, 0)
+
+#define COPY_2SAMETYPE_HETEROGENEOUS_INTERNAL( TYPENAME, TYPE, LONG_DOUBLE)                 \
 static int32_t                                                                            \
 copy_##TYPENAME##_heterogeneous(opal_convertor_t *pConvertor, uint32_t count,             \
                                 const char* from, size_t from_len, ptrdiff_t from_extent, \
@@ -122,11 +201,17 @@ copy_##TYPENAME##_heterogeneous(opal_convertor_t *pConvertor, uint32_t count,
                                                                         \
     if ((pConvertor->remoteArch & OPAL_ARCH_ISBIGENDIAN) !=             \
         (opal_local_arch & OPAL_ARCH_ISBIGENDIAN)) {                    \
-        if( (to_extent == from_extent) && (to_extent == sizeof(TYPE)) ) { \
+        if( (to_extent == from_extent) && (to_extent == (2 * sizeof(TYPE))) ) { \
             opal_dt_swap_bytes(to, from, sizeof(TYPE), 2 * count);      \
+            if (LONG_DOUBLE) {                                          \
+                opal_dt_swap_long_double(to, from, sizeof(TYPE), 2*count, pConvertor->remoteArch);\
+            }                                                           \
         } else {                                                        \
             for( i = 0; i < count; i++ ) {                              \
                 opal_dt_swap_bytes(to, from, sizeof(TYPE), 2);          \
+                if (LONG_DOUBLE) {                                      \
+                    opal_dt_swap_long_double(to, from, sizeof(TYPE), 2, pConvertor->remoteArch);\
+                }                                                       \
                 to += to_extent;                                        \
                 from += from_extent;                                    \
             }                                                           \
@@ -333,7 +418,7 @@ COPY_TYPE_HETEROGENEOUS( float16, float )
 #elif SIZEOF_DOUBLE == 16
 COPY_TYPE_HETEROGENEOUS( float16, double )
 #elif HAVE_LONG_DOUBLE && SIZEOF_LONG_DOUBLE == 16
-COPY_TYPE_HETEROGENEOUS( float16, long double )
+COPY_TYPE_HETEROGENEOUS_INTERNAL( float16, long double, 1)
 #else
 /* #error No basic type for copy function for opal_datatype_float16 found */
 #define copy_float16_heterogeneous NULL
@@ -354,7 +439,7 @@ COPY_2SAMETYPE_HETEROGENEOUS( double_complex, double )
 #endif
 
 #if HAVE_LONG_DOUBLE__COMPLEX
-COPY_2SAMETYPE_HETEROGENEOUS( long_double_complex, long double )
+COPY_2SAMETYPE_HETEROGENEOUS_INTERNAL( long_double_complex, long double, 1)
 #else
 /* #error No basic type for copy function for opal_datatype_long_double_complex found */
 #define copy_long_double_complex_heterogeneous NULL

opal/datatype/opal_datatype_fake_stack.c

@@ -91,6 +91,7 @@ int opal_convertor_create_stack_with_pos_general( opal_convertor_t* pConvertor,
     }
 
     /* remove from the main loop all the complete datatypes */
+    assert (! (pConvertor->flags & CONVERTOR_SEND));
     remote_size    = opal_convertor_compute_remote_size( pConvertor );
     count          = (int32_t)(starting_point / remote_size);
     resting_place -= (remote_size * count);

opal/mca/btl/tcp/btl_tcp.c

@@ -12,7 +12,7 @@
  *                         All rights reserved.
  * Copyright (c) 2006-2015 Los Alamos National Security, LLC.  All rights
  *                         reserved.
- * Copyright (c) 2016      Research Organization for Information Science
+ * Copyright (c) 2016-2017 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  * Copyright (c) 2016      Intel, Inc. All rights reserved.
  *
@@ -381,6 +381,7 @@ int mca_btl_tcp_put (mca_btl_base_module_t *btl, struct mca_btl_base_endpoint_t
 
     frag->segments[1].seg_addr.lval = remote_address;
     frag->segments[1].seg_len = size;
+    if (endpoint->endpoint_nbo) MCA_BTL_BASE_SEGMENT_HTON(frag->segments[1]);
 
     frag->base.des_flags = MCA_BTL_DES_FLAGS_BTL_OWNERSHIP | MCA_BTL_DES_SEND_ALWAYS_CALLBACK;
     frag->base.des_cbfunc = fake_rdma_complete;

opal/mca/btl/tcp/btl_tcp_frag.c

@@ -12,7 +12,7 @@
  *                         All rights reserved.
  * Copyright (c) 2014      Los Alamos National Security, LLC. All rights
  *                         reserved.
- * Copyright (c) 2015-2016 Research Organization for Information Science
+ * Copyright (c) 2015-2017 Research Organization for Information Science
  *                         and Technology (RIST). All rights reserved.
  * Copyright (c) 2015-2016 Cisco Systems, Inc.  All rights reserved.
  * $COPYRIGHT$
@@ -291,6 +291,7 @@ bool mca_btl_tcp_frag_recv(mca_btl_tcp_frag_t* frag, int sd)
                 goto repeat;
             } else if (frag->iov_idx == 2) {
                 for( i = 0; i < frag->hdr.count; i++ ) {
+                    if (btl_endpoint->endpoint_nbo) MCA_BTL_BASE_SEGMENT_NTOH(frag->segments[i]);
                     frag->iov[i+2].iov_base = (IOVBASE_TYPE*)frag->segments[i].seg_addr.pval;
                     frag->iov[i+2].iov_len = frag->segments[i].seg_len;
                 }

orte/mca/oob/tcp/oob_tcp_hdr.h

@@ -13,6 +13,8 @@
  *                         All rights reserved.
  * Copyright (c) 2010-2011 Cisco Systems, Inc.  All rights reserved.
  * Copyright (c) 2014-2017 Intel, Inc.  All rights reserved.
+ * Copyright (c) 2017      Research Organization for Information Science
+ *                         and Technology (RIST). All rights reserved.
  *
  * $COPYRIGHT$
  *
@@ -32,12 +34,12 @@
  * the message came from an external (to
  * this component) source
  */
-typedef enum {
-    MCA_OOB_TCP_IDENT,
-    MCA_OOB_TCP_PROBE,
-    MCA_OOB_TCP_PING,
-    MCA_OOB_TCP_USER
-} mca_oob_tcp_msg_type_t;
+typedef uint8_t mca_oob_tcp_msg_type_t;
+
+#define MCA_OOB_TCP_IDENT 1
+#define MCA_OOB_TCP_PROBE 2
+#define MCA_OOB_TCP_PING  3
+#define MCA_OOB_TCP_USER  4
 
 #define ORTE_MAX_RTD_SIZE  31
 
@@ -54,14 +56,14 @@ typedef struct {
      * and let some other module try to send it
      */
     orte_process_name_t     dst;
-    /* type of message */
-    mca_oob_tcp_msg_type_t type;
     /* the rml tag where this message is headed */
     orte_rml_tag_t tag;
     /* the seq number of this message */
     uint32_t seq_num;
     /* number of bytes in message */
     uint32_t nbytes;
+    /* type of message */
+    mca_oob_tcp_msg_type_t type;
     /* routed module to be used */
     char routed[ORTE_MAX_RTD_SIZE+1];
 } mca_oob_tcp_hdr_t;
@@ -71,7 +73,6 @@ typedef struct {
 #define MCA_OOB_TCP_HDR_NTOH(h)                \
     ORTE_PROCESS_NAME_NTOH((h)->origin);        \
     ORTE_PROCESS_NAME_NTOH((h)->dst);           \
-    (h)->type = ntohl((h)->type);               \
     (h)->tag = ORTE_RML_TAG_NTOH((h)->tag);     \
     (h)->nbytes = ntohl((h)->nbytes);
 
@@ -81,7 +82,6 @@ typedef struct {
 #define MCA_OOB_TCP_HDR_HTON(h)                \
     ORTE_PROCESS_NAME_HTON((h)->origin);        \
     ORTE_PROCESS_NAME_HTON((h)->dst);           \
-    (h)->type = htonl((h)->type);               \
     (h)->tag = ORTE_RML_TAG_HTON((h)->tag);     \
     (h)->nbytes = htonl((h)->nbytes);