Re: [PATCH] Reduction support for parloop, OMP_ATOMIC Changes

[Razya Ladelsky <RAZYA at il dot ibm dot com>]

Now the testsuite zipped file was corrupted...
This is the final patch and testsuite directory :)
Thanks, 
Razya
 
Razya Ladelsky/Haifa/IBM wrote on 23/09/2007 21:36:14:

> Forgot the testsuite. Here's the patch again....
> __________________
> 
> Hi,
> 
> This is a follow-up patch to 
> http://gcc.gnu.org/ml/gcc-patches/2007-09/msg00891.html
> The prior patch failed for a few testcases of libgomp.
> This patch passes successfully.
> The main problem was lowering the OMP_ATOMIC.
> Although it should be a simple statement, sometimes implicit 
> conversions make the OMP_ATOMIC expression more complex, so it does not
> fit the gimple form well.
> To overcome this problem, OMP_ATOMIC was split into two codes: 
> OMP_ATOMIC_LOAD and OMP_ATOMIC_ATORE.
> 
> the sequence will now be :
> 
>    OMP_LOAD (tmp, mem)
>    val = some computations involving tmp;
>    OMP_STORE (val) 
> 
> (these two new codes are described in tree.def)
> 
> The reduction code was also adapted to use these codes, 
> and I added a new testsuite directory autopar including a few 
> reduction tests.
> I also moved the parallelization test that Zdenek provided with 
> the autopar patch from tree-ssa directory to autopar.
> 
> Thanks to Zdenek who provided the bulk part of the omp_atomic 
> gimplify/expand code.
> 
> Bootstrapped & regtested on x86.
> O.K. for mainline?
>

Attachment: autopar.tgz
Description: Binary data

Attachment: auto_reductions.diff
Description: Binary data

Attachment: ChangeLog_23_09
Description: Binary data