[Openmp-dev] PPC64 patch from Intel's fourth cmake patch

["C. Bergström"]

On 08/20/14 07:42 PM, Hal Finkel wrote:
> ----- Original Message -----
>> From: "James H Cownie" <james.h.cownie at intel.com>
>> To: "Hal Finkel" <hfinkel at anl.gov>, "Carlo Bertolli" <cbertol at us.ibm.com>
>> Cc: "Michael Wong" <michaelw at ca.ibm.com>, openmp-dev at dcs-maillist2.engr.illinois.edu, "C. Bergström"
>> <cbergstrom at pathscale.com>, "Alexey Bataev" <a.bataev at gmx.com>
>> Sent: Wednesday, August 20, 2014 5:48:38 AM
>> Subject: RE: [Openmp-dev] PPC64 patch from Intel's fourth cmake patch
>>
>>> As you've discovered, clang-omp currently does not use the
>>> arbitrary-length-parameter-list aspect of the microtasks.
>> Indeed, and I don't expect it too. IMO the way the Intel compiler
>> does it by passing a pointer argument for each
>> reference into the parent stack is unpleasant.
> Okay, this is all fine. I recommend that we do the following:
>
>   1. In the runtime source code (and its documentation), clearly document that the arbitrary-parameter-list aspect of microtasks is deprecated, and what the maximum number of arguments is that we expect from "new" frontends.
>
>   2. As I recall, we now have a libffi dependency to support arbitrary-length parameter lists on ARM. Is this really needed? Maybe they can use a switch statement just as I did on PPC?
I've got access to the target platforms and should be able to do 
validation with our compiler in the next 1-2 weeks. If you post a patch 
I'll test it.

Can someone write a testcase for this - either functional test and or 
even better would be a performance test which might be sensitive to this.

Thanks