Merge pull request #6466 from bosilca/backport/4.x/6029

Backport/4.x/6029

Author: hppritcha

opal/datatype/opal_convertor.c

@@ -12,8 +12,8 @@
  *                         All rights reserved.
  * Copyright (c) 2009      Oak Ridge National Labs.  All rights reserved.
  * Copyright (c) 2011      NVIDIA Corporation.  All rights reserved.
- * Copyright (c) 2013-2017 Research Organization for Information Science
- *                         and Technology (RIST). All rights reserved.
+ * Copyright (c) 2013-2018 Research Organization for Information Science
+ *                         and Technology (RIST).  All rights reserved.
  * Copyright (c) 2017      Intel, Inc. All rights reserved
  * $COPYRIGHT$
  *
@@ -330,7 +330,7 @@ static inline int opal_convertor_create_stack_with_pos_contig( opal_convertor_t*
     dt_stack_t* pStack;   /* pointer to the position on the stack */
     const opal_datatype_t* pData = pConvertor->pDesc;
     dt_elem_desc_t* pElems;
-    uint32_t count;
+    size_t count;
     ptrdiff_t extent;
 
     pStack = pConvertor->pStack;
@@ -340,7 +340,7 @@ static inline int opal_convertor_create_stack_with_pos_contig( opal_convertor_t*
      */
     pElems = pConvertor->use_desc->desc;
 
-    count = (uint32_t)(starting_point / pData->size);
+    count = starting_point / pData->size;
     extent = pData->ub - pData->lb;
 
     pStack[0].type     = OPAL_DATATYPE_LOOP;  /* the first one is always the loop */
@@ -349,7 +349,7 @@ static inline int opal_convertor_create_stack_with_pos_contig( opal_convertor_t*
     pStack[0].disp     = count * extent;
 
     /* now compute the number of pending bytes */
-    count = (uint32_t)(starting_point - count * pData->size);
+    count = starting_point - count * pData->size;
     /**
      * We save the current displacement starting from the begining
      * of this data.
@@ -563,7 +563,7 @@ size_t opal_convertor_compute_remote_size( opal_convertor_t* pConvertor )
 
 int32_t opal_convertor_prepare_for_recv( opal_convertor_t* convertor,
                                          const struct opal_datatype_t* datatype,
-                                         int32_t count,
+                                         size_t count,
                                          const void* pUserBuf )
 {
     /* Here I should check that the data is not overlapping */
@@ -605,7 +605,7 @@ int32_t opal_convertor_prepare_for_recv( opal_convertor_t* convertor,
 
 int32_t opal_convertor_prepare_for_send( opal_convertor_t* convertor,
                                          const struct opal_datatype_t* datatype,
-                                         int32_t count,
+                                         size_t count,
                                          const void* pUserBuf )
 {
     convertor->flags |= CONVERTOR_SEND;
@@ -699,11 +699,11 @@ int opal_convertor_clone( const opal_convertor_t* source,
 
 void opal_convertor_dump( opal_convertor_t* convertor )
 {
-    opal_output( 0, "Convertor %p count %d stack position %d bConverted %ld\n"
-                 "\tlocal_size %ld remote_size %ld flags %X stack_size %d pending_length %d\n"
+    opal_output( 0, "Convertor %p count %" PRIsize_t" stack position %d bConverted %" PRIsize_t "\n"
+                 "\tlocal_size %ld remote_size %ld flags %X stack_size %d pending_length %" PRIsize_t "\n"
                  "\tremote_arch %u local_arch %u\n",
                  (void*)convertor,
-                 convertor->count, convertor->stack_pos, (unsigned long)convertor->bConverted,
+                 convertor->count, convertor->stack_pos, convertor->bConverted,
                  (unsigned long)convertor->local_size, (unsigned long)convertor->remote_size,
                  convertor->flags, convertor->stack_size, convertor->partial_length,
                  convertor->remoteArch, opal_local_arch );
@@ -734,8 +734,8 @@ void opal_datatype_dump_stack( const dt_stack_t* pStack, int stack_pos,
 {
     opal_output( 0, "\nStack %p stack_pos %d name %s\n", (void*)pStack, stack_pos, name );
     for( ; stack_pos >= 0; stack_pos-- ) {
-        opal_output( 0, "%d: pos %d count %d disp %ld ", stack_pos, pStack[stack_pos].index,
-                     (int)pStack[stack_pos].count, (long)pStack[stack_pos].disp );
+        opal_output( 0, "%d: pos %d count %" PRIsize_t " disp %ld ", stack_pos, pStack[stack_pos].index,
+                     pStack[stack_pos].count, pStack[stack_pos].disp );
         if( pStack->index != -1 )
             opal_output( 0, "\t[desc count %lu disp %ld extent %ld]\n",
                          (unsigned long)pDesc[pStack[stack_pos].index].elem.count,

opal/datatype/opal_convertor.h

@@ -12,8 +12,8 @@
  *                         All rights reserved.
  * Copyright (c) 2009      Oak Ridge National Labs.  All rights reserved.
  * Copyright (c) 2014      NVIDIA Corporation.  All rights reserved.
- * Copyright (c) 2017      Research Organization for Information Science
- *                         and Technology (RIST). All rights reserved.
+ * Copyright (c) 2017-2018 Research Organization for Information Science
+ *                         and Technology (RIST).  All rights reserved.
  * Copyright (c) 2017      Intel, Inc. All rights reserved
  * $COPYRIGHT$
  *
@@ -74,6 +74,7 @@ struct opal_convertor_master_t;
 struct dt_stack_t {
     int32_t           index;    /**< index in the element description */
     int16_t           type;     /**< the type used for the last pack/unpack (original or OPAL_DATATYPE_UINT1) */
+    int16_t           padding;
     size_t            count;    /**< number of times we still have to do it */
     ptrdiff_t         disp;     /**< actual displacement depending on the count field */
 };
@@ -93,30 +94,33 @@ struct opal_convertor_t {
     const opal_datatype_t*        pDesc;          /**< the datatype description associated with the convertor */
     const dt_type_desc_t*         use_desc;       /**< the version used by the convertor (normal or optimized) */
     opal_datatype_count_t         count;          /**< the total number of full datatype elements */
+
+    /* --- cacheline boundary (64 bytes - if 64bits arch and !OPAL_ENABLE_DEBUG) --- */
     uint32_t                      stack_size;     /**< size of the allocated stack */
-    /* --- cacheline 1 boundary (64 bytes) --- */
     unsigned char*                pBaseBuf;       /**< initial buffer as supplied by the user */
     dt_stack_t*                   pStack;         /**< the local stack for the actual conversion */
     convertor_advance_fct_t       fAdvance;       /**< pointer to the pack/unpack functions */
+
+    /* --- cacheline boundary (96 bytes - if 64bits arch and !OPAL_ENABLE_DEBUG) --- */
     struct opal_convertor_master_t* master;       /**< the master convertor */
 
     /* All others fields get modified for every call to pack/unpack functions */
     uint32_t                      stack_pos;      /**< the actual position on the stack */
-    uint32_t                      partial_length; /**< amount of data left over from the last unpack */
+    size_t                        partial_length; /**< amount of data left over from the last unpack */
     size_t                        bConverted;     /**< # of bytes already converted */
+
+    /* --- cacheline boundary (128 bytes - if 64bits arch and !OPAL_ENABLE_DEBUG) --- */
     uint32_t                      checksum;       /**< checksum computed by pack/unpack operation */
     uint32_t                      csum_ui1;       /**< partial checksum computed by pack/unpack operation */
     size_t                        csum_ui2;       /**< partial checksum computed by pack/unpack operation */
-     /* --- cacheline 2 boundary (128 bytes) --- */
+
+    /* --- fields are no more aligned on cacheline --- */
     dt_stack_t                    static_stack[DT_STATIC_STACK_SIZE];  /**< local stack for small datatypes */
-    /* --- cacheline 3 boundary (192 bytes) was 56 bytes ago --- */
 
 #if OPAL_CUDA_SUPPORT
     memcpy_fct_t                  cbmemcpy;       /**< memcpy or cuMemcpy */
     void *                        stream;         /**< CUstream for async copy */
 #endif
-    /* size: 248, cachelines: 4, members: 20 */
-    /* last cacheline: 56 bytes */
 };
 OPAL_DECLSPEC OBJ_CLASS_DECLARATION( opal_convertor_t );
 
@@ -251,12 +255,12 @@ static inline void opal_convertor_get_offset_pointer( const opal_convertor_t* pC
  */
 OPAL_DECLSPEC int32_t opal_convertor_prepare_for_send( opal_convertor_t* convertor,
                                                        const struct opal_datatype_t* datatype,
-                                                       int32_t count,
+                                                       size_t count,
                                                        const void* pUserBuf);
 
 static inline int32_t opal_convertor_copy_and_prepare_for_send( const opal_convertor_t* pSrcConv,
                                                                 const struct opal_datatype_t* datatype,
-                                                                int32_t count,
+                                                                size_t count,
                                                                 const void* pUserBuf,
                                                                 int32_t flags,
                                                                 opal_convertor_t* convertor )
@@ -273,11 +277,11 @@ static inline int32_t opal_convertor_copy_and_prepare_for_send( const opal_conve
  */
 OPAL_DECLSPEC int32_t opal_convertor_prepare_for_recv( opal_convertor_t* convertor,
                                                        const struct opal_datatype_t* datatype,
-                                                       int32_t count,
+                                                       size_t count,
                                                        const void* pUserBuf );
 static inline int32_t opal_convertor_copy_and_prepare_for_recv( const opal_convertor_t* pSrcConv,
                                                                 const struct opal_datatype_t* datatype,
-                                                                int32_t count,
+                                                                size_t count,
                                                                 const void* pUserBuf,
                                                                 int32_t flags,
                                                                 opal_convertor_t* convertor )

opal/datatype/opal_convertor_raw.c

@@ -43,11 +43,11 @@ opal_convertor_raw( opal_convertor_t* pConvertor,
     const opal_datatype_t *pData = pConvertor->pDesc;
     dt_stack_t* pStack;       /* pointer to the position on the stack */
     uint32_t pos_desc;        /* actual position in the description of the derived datatype */
-    uint32_t count_desc;      /* the number of items already done in the actual pos_desc */
+    size_t count_desc;        /* the number of items already done in the actual pos_desc */
     dt_elem_desc_t* description, *pElem;
     unsigned char *source_base;  /* origin of the data */
     size_t raw_data = 0;      /* sum of raw data lengths in the iov_len fields */
-    uint32_t index = 0, i;    /* the iov index and a simple counter */
+    uint32_t index = 0;       /* the iov index and a simple counter */
 
     assert( (*iov_count) > 0 );
     if( OPAL_LIKELY(pConvertor->flags & CONVERTOR_COMPLETED) ) {
@@ -83,15 +83,15 @@ opal_convertor_raw( opal_convertor_t* pConvertor,
     pStack = pConvertor->pStack + pConvertor->stack_pos;
     pos_desc     = pStack->index;
     source_base  = pConvertor->pBaseBuf + pStack->disp;
-    count_desc   = (uint32_t)pStack->count;
+    count_desc   = pStack->count;
     pStack--;
     pConvertor->stack_pos--;
     pElem = &(description[pos_desc]);
     source_base += pStack->disp;
-    DO_DEBUG( opal_output( 0, "raw start pos_desc %d count_desc %d disp %ld\n"
-                           "stack_pos %d pos_desc %d count_desc %d disp %ld\n",
+    DO_DEBUG( opal_output( 0, "raw start pos_desc %d count_desc %" PRIsize_t " disp %ld\n"
+                           "stack_pos %d pos_desc %d count_desc %" PRIsize_t " disp %ld\n",
                            pos_desc, count_desc, (long)(source_base - pConvertor->pBaseBuf),
-                           pConvertor->stack_pos, pStack->index, (int)pStack->count, (long)pStack->disp ); );
+                           pConvertor->stack_pos, pStack->index, pStack->count, (long)pStack->disp ); );
     while( 1 ) {
         while( pElem->elem.common.flags & OPAL_DATATYPE_FLAG_DATA ) {
             size_t blength = opal_datatype_basicDatatypes[pElem->elem.common.type]->size;
@@ -112,7 +112,7 @@ opal_convertor_raw( opal_convertor_t* pConvertor,
                     count_desc = 0;
                 }
             } else {
-                for( i = count_desc; (i > 0) && (index < *iov_count); i--, index++ ) {
+                for(size_t i = count_desc; (i > 0) && (index < *iov_count); i--, index++ ) {
                     OPAL_DATATYPE_SAFEGUARD_POINTER( source_base, blength, pConvertor->pBaseBuf,
                                                 pConvertor->pDesc, pConvertor->count );
                     DO_DEBUG( opal_output( 0, "raw 2. iov[%d] = {base %p, length %" PRIsize_t "}\n",
@@ -134,9 +134,9 @@ opal_convertor_raw( opal_convertor_t* pConvertor,
             goto complete_loop;
         }
         if( OPAL_DATATYPE_END_LOOP == pElem->elem.common.type ) { /* end of the current loop */
-            DO_DEBUG( opal_output( 0, "raw end_loop count %d stack_pos %d"
+            DO_DEBUG( opal_output( 0, "raw end_loop count %" PRIsize_t " stack_pos %d"
                                    " pos_desc %d disp %ld space %lu\n",
-                                   (int)pStack->count, pConvertor->stack_pos,
+                                   pStack->count, pConvertor->stack_pos,
                                    pos_desc, (long)pStack->disp, (unsigned long)raw_data ); );
             if( --(pStack->count) == 0 ) { /* end of loop */
                 if( pConvertor->stack_pos == 0 ) {
@@ -160,9 +160,9 @@ opal_convertor_raw( opal_convertor_t* pConvertor,
             }
             source_base = pConvertor->pBaseBuf + pStack->disp;
             UPDATE_INTERNAL_COUNTERS( description, pos_desc, pElem, count_desc );
-            DO_DEBUG( opal_output( 0, "raw new_loop count %d stack_pos %d "
+            DO_DEBUG( opal_output( 0, "raw new_loop count %" PRIsize_t " stack_pos %d "
                                    "pos_desc %d disp %ld space %lu\n",
-                                   (int)pStack->count, pConvertor->stack_pos,
+                                   pStack->count, pConvertor->stack_pos,
                                    pos_desc, (long)pStack->disp, (unsigned long)raw_data ); );
         }
         if( OPAL_DATATYPE_LOOP == pElem->elem.common.type ) {
@@ -172,7 +172,7 @@ opal_convertor_raw( opal_convertor_t* pConvertor,
             if( pElem->loop.common.flags & OPAL_DATATYPE_FLAG_CONTIGUOUS ) {
                 ptrdiff_t offset = end_loop->first_elem_disp;
                 source_base += offset;
-                for(uint32_t i = MIN(count_desc, *iov_count - index); i > 0; i--, index++ ) {
+                for(size_t i = MIN(count_desc, *iov_count - index); i > 0; i--, index++ ) {
                     OPAL_DATATYPE_SAFEGUARD_POINTER( source_base, end_loop->size, pConvertor->pBaseBuf,
                                                 pConvertor->pDesc, pConvertor->count );
                     iov[index].iov_base = (IOVBASE_TYPE *) source_base;
@@ -216,7 +216,7 @@ opal_convertor_raw( opal_convertor_t* pConvertor,
     /* I complete an element, next step I should go to the next one */
     PUSH_STACK( pStack, pConvertor->stack_pos, pos_desc, OPAL_DATATYPE_UINT1, count_desc,
                 source_base - pStack->disp - pConvertor->pBaseBuf );
-    DO_DEBUG( opal_output( 0, "raw save stack stack_pos %d pos_desc %d count_desc %d disp %ld\n",
-                           pConvertor->stack_pos, pStack->index, (int)pStack->count, (long)pStack->disp ); );
+    DO_DEBUG( opal_output( 0, "raw save stack stack_pos %d pos_desc %d count_desc %" PRIsize_t " disp %ld\n",
+                           pConvertor->stack_pos, pStack->index, pStack->count, (long)pStack->disp ); );
     return 0;
 }