[Gomp-discuss] The draft ... issues

[Lars Segerlund]

 Now here follows my opinions on this, and are to be treated as such.

 In the introduction I would like an additional statement right after 
'introducing openMP directives in the compiler' along the lines of, ' 
and provide a general concurrency aware framework for gcc internals, in 
order to handle all forms of concurrency and memory models'.
 I believed that this was what we strived for as related to the IR, ( 
trees and such ). It might also better be placed in the 'GCC and 
concurrency' section, but I'm not sure about this.

 In the from directives to IR section I would like to se the openMP 
fortran spec included, to single out C/C++ seem's a bit narrow. I do 
however agree on first targetting C/C++. ( not much difference anyway ).

 A bit off topic now, I think we could start with implementing the 
framework from Diegos paper, and work from there, we should ba able to 
do atleast as well. ( he has some performace comparisons ).

 In the thread creation part, I think we can safely say that the Posix 
thread model is a good starting point, ( whatever underlying library 
used this is a standard model with standard semantics, and nobody has 
objected to it on the list ).

 I have some issues with the last part, since locks and 
synchronizations are most often implemented ontop of mutexes, semaphores 
and other 'more' primitive elements. I also believe that the term 
threading model here should be exchanged with threading implementation 
in this context.

 Very nice work, now were starting to get a target to work against.

 / Lars Segerlund.