Start optimizing the code.

Signed-off-by: George Bosilca <bosilca@icl.utk.edu>

Author: bosilca

opal/datatype/opal_datatype_pack.h

@@ -49,14 +49,14 @@ pack_predefined_data( opal_convertor_t* CONVERTOR,
     /* premptively update the number of COUNT we will return. */
     *(COUNT) -= cando_count;
 
-    if( 1 == _elem->count ) {  /* Everything is contiguous, handle it as a prologue */
+    if( 1 == _elem->count ) {  /* Everything is contiguous, handle it as the last contiguous piece */
         goto do_epilog;
     }
     if( 1 == _elem->blocklen ) { /* Do as many full blocklen as possible */
         for(; cando_count > 0; cando_count--) {
             OPAL_DATATYPE_SAFEGUARD_POINTER( _memory, blocklen_bytes, (CONVERTOR)->pBaseBuf,
                                              (CONVERTOR)->pDesc, (CONVERTOR)->count );
-            DO_DEBUG( opal_output( 0, "pack 2. memcpy( %p, %p, %lu ) => space %lu\n",
+            DO_DEBUG( opal_output( 0, "pack memcpy( %p, %p, %lu ) => space %lu [blen = 1]\n",
                                    (void*)_packed, (void*)_memory, (unsigned long)blocklen_bytes, (unsigned long)(*(SPACE) - (_packed - *(packed))) ); );
             MEMCPY_CSUM( _packed, _memory, blocklen_bytes, (CONVERTOR) );
             _packed     += blocklen_bytes;
@@ -75,23 +75,21 @@ pack_predefined_data( opal_convertor_t* CONVERTOR,
      * should be manipulated in the current call (this number is instead reflected on the
      * SPACE).
      */
-    do_now = do_now % _elem->blocklen;  /* any partial elements ? */
-
-    if( 0 != do_now ) {
+    if( 0 != (do_now = do_now % _elem->blocklen) ) {  /* any partial elements ? */
         size_t left_in_block = do_now;  /* left in the current blocklen */
-        do_now = (do_now > cando_count ) ? cando_count : do_now;
+        do_now = (do_now > cando_count) ? cando_count : do_now;
         do_now_bytes = do_now * opal_datatype_basicDatatypes[_elem->common.type]->size;
 
         OPAL_DATATYPE_SAFEGUARD_POINTER( _memory, do_now_bytes, (CONVERTOR)->pBaseBuf,
                                          (CONVERTOR)->pDesc, (CONVERTOR)->count );
-        DO_DEBUG( opal_output( 0, "pack 1. memcpy( %p, %p, %lu ) => space %lu [prolog]\n",
+        DO_DEBUG( opal_output( 0, "pack memcpy( %p, %p, %lu ) => space %lu [prolog]\n",
                                _packed, (void*)_memory, (unsigned long)do_now_bytes, (unsigned long)(*(SPACE)) ); );
         MEMCPY_CSUM( _packed, _memory, do_now_bytes, (CONVERTOR) );
+        _packed     += do_now_bytes;
         _memory     += (ptrdiff_t)do_now_bytes;
-        /* compensate if we just completed a blocklen */
-        if( do_now == left_in_block )
+
+        if( do_now == left_in_block )  /* compensate if completed a blocklen */
             _memory += _elem->extent - blocklen_bytes;
-        _packed     += do_now_bytes;
         cando_count -= do_now;
     }
 
@@ -119,7 +117,7 @@ pack_predefined_data( opal_convertor_t* CONVERTOR,
         OPAL_DATATYPE_SAFEGUARD_POINTER( _memory, do_now_bytes, (CONVERTOR)->pBaseBuf,
                                          (CONVERTOR)->pDesc, (CONVERTOR)->count );
         DO_DEBUG( opal_output( 0, "pack 3. memcpy( %p, %p, %lu ) => space %lu [epilog]\n",
-                               (void*)_packed, (void*)_memory, (unsigned long)do_now_bytes, (unsigned long)(*(SPACE)) ); );
+                               (void*)_packed, (void*)_memory, (unsigned long)do_now_bytes, (unsigned long)(*(SPACE) - (_packed - *(packed))) ); );
         MEMCPY_CSUM( _packed, _memory, do_now_bytes, (CONVERTOR) );
         _memory   += do_now_bytes;
         _packed   += do_now_bytes;

opal/datatype/opal_datatype_unpack.h

@@ -49,14 +49,14 @@ unpack_predefined_data( opal_convertor_t* CONVERTOR,
     /* premptively update the number of COUNT we will return. */
     *(COUNT) -= cando_count;
 
-    if( 1 == _elem->count ) {  /* Everything is contiguous, handle it as a prologue */
+    if( 1 == _elem->count ) {  /* Everything is contiguous, handle it as the last contiguous piece */
         goto do_epilog;
     }
     if( 1 == _elem->blocklen ) {  /* Do as many full blocklen as possible */
         for(; cando_count > 0; cando_count--) {
             OPAL_DATATYPE_SAFEGUARD_POINTER( _memory, blocklen_bytes, (CONVERTOR)->pBaseBuf,
                                              (CONVERTOR)->pDesc, (CONVERTOR)->count );
-            DO_DEBUG( opal_output( 0, "unpack 2. memcpy( %p, %p, %lu ) => space %lu\n",
+            DO_DEBUG( opal_output( 0, "unpack memcpy( %p, %p, %lu ) => space %lu [blen = 1]\n",
                                    (void*)_memory, (void*)_packed, (unsigned long)blocklen_bytes, (unsigned long)(*(SPACE) - (_packed - *(packed))) ); );
             MEMCPY_CSUM( _memory, _packed, blocklen_bytes, (CONVERTOR) );
             _packed     += blocklen_bytes;
@@ -76,23 +76,21 @@ unpack_predefined_data( opal_convertor_t* CONVERTOR,
      * should be manipulated in the current call (this number is instead reflected on the
      * SPACE).
      */
-    do_now = do_now % _elem->blocklen;  /* any partial elements ? */
-
-    if( 0 != do_now ) {
+    if( 0 != (do_now = do_now % _elem->blocklen) ) {  /* any partial elements ? */
         size_t left_in_block = do_now;  /* left in the current blocklen */
-        do_now = (do_now > cando_count ) ? cando_count : do_now;
+        do_now = (do_now > cando_count) ? cando_count : do_now;
         do_now_bytes = do_now * opal_datatype_basicDatatypes[_elem->common.type]->size;
 
         OPAL_DATATYPE_SAFEGUARD_POINTER( _memory, do_now_bytes, (CONVERTOR)->pBaseBuf,
                                          (CONVERTOR)->pDesc, (CONVERTOR)->count );
-        DO_DEBUG( opal_output( 0, "unpack 1. memcpy( %p, %p, %lu ) => space %lu [prolog]\n",
+        DO_DEBUG( opal_output( 0, "unpack memcpy( %p, %p, %lu ) => space %lu [prolog]\n",
                                (void*)_memory, (void*)_packed, (unsigned long)do_now_bytes, (unsigned long)(*(SPACE)) ); );
         MEMCPY_CSUM( _memory, _packed, do_now_bytes, (CONVERTOR) );
-        _memory      += (ptrdiff_t)do_now_bytes;
-        /* compensate if we just completed a blocklen */
-        if( do_now == left_in_block )
-            _memory += _elem->extent - blocklen_bytes;
         _packed     += do_now_bytes;
+        _memory     += (ptrdiff_t)do_now_bytes;
+
+        if( do_now == left_in_block )  /* compensate if completed a blocklen */
+            _memory += _elem->extent - blocklen_bytes;
         cando_count -= do_now;
     }