A blog entry on Research@Intel - Parallel Programming with Transactions:
http://blogs.intel.com/research/2008/02/parallel_programming_with_tran.php

I wonder if one day D's `synchronized` could be used the same way as proposed in article `atomic` statement?

I understand that this is rather runtime library feature but still, it would be nice to know that today's language constructs semantically ready to the point then compiler/lib could automagically parallelize code written today.

-- serg

serg kovrov wrote:
> A blog entry on Research@Intel - Parallel Programming with Transactions:
> http://blogs.intel.com/research/2008/02/parallel_programming_with_tran.php
> 
> I wonder if one day D's `synchronized` could be used the same way as proposed in article `atomic` statement?
> 
> I understand that this is rather runtime library feature but still, it would be nice to know that today's language constructs semantically ready to the point then compiler/lib could automagically parallelize code written today.
> 
> -- serg

It was discussed at the conference, all that time ago:
http://s3.amazonaws.com/dconf2007/DSTM.ppt

serg kovrov Wrote:

> A blog entry on Research@Intel - Parallel Programming with Transactions: http://blogs.intel.com/research/2008/02/parallel_programming_with_tran.php
> 
> I wonder if one day D's `synchronized` could be used the same way as proposed in article `atomic` statement?
> 
> I understand that this is rather runtime library feature but still, it would be nice to know that today's language constructs semantically ready to the point then compiler/lib could automagically parallelize code written today.
> 
> -- serg

The problem with parallel programming in it's current incarnation is that most of the algorithms were designed by Java Programmers, which is only a slight improvement on HTML Web Designers - the point being that they've never read a hex number in their life let alone seen machine code.

I like locking only for allocate, free, and share; and then having each thread handle it's own stuff otherwise.

Done properly, spin locks will only loop if another thread has successfully passed the gate first - removing deadlock and naturally resolving most livelock conditions.

They're also markedly easier to implement with the device only weighing about 40-100 bytes.

Regards,
Dan