[Beignet] [PATCH 0/3] Generate as_ and convert_ functions

[Zhigang Gong]

I just pushed this patchset. Thanks for the contribution.

On Fri, Jun 21, 2013 at 01:11:57PM +0800, He Junyan wrote:
> On 06/20/2013 07:05 PM, Zhigang Gong wrote:
> >On Tue, Jun 18, 2013 at 03:58:10PM +0800, He Junyan wrote:
> >>hi Simon:
> >>
> >>Very thanks for your patch.
> >>There still some issues after apply your patch:
> >>1.  Some convert function like convert_int8_rtp, convert_int8_rtn
> >>and convert_int8_rte
> >>      seems be ignored, you can refer to the test case of
> >>vector-conversion.cl in piglit CL test.
> >>2.  The test case in piglit named gegl-rgb-gamma-u8-to-ragabaf.cl
> >>report a precision problem.
> >>      This case convert a  int vect to a float vect and then do the
> >>calculation. The precision may caused
> >>      by conversion and may not, I need to check further.
> >Junyan, does the patch cause any regression? Are you ok to accept this
> >patch based on your test result? I think the issue 1 should be fixed
> >in a new patch, and the issue 2 may be another independent issue. Or do
> >you have any update for issue 3.
> I agree, we can send new patch to fix these issues.
> >
> >>3.  The ocl_stdlib.h after the patch seems huge and I can feel it
> >>makes the runtime compile time
> >>      a little longer when I run all the test cases one by one. I
> >>think we should seek some solution
> >>      such as external AST cache file to avoid these code to be built
> >>every time by Clang for every
> >>      kernel file.
> >I raised the same issue when Simon submitted the first version of patch.
> >And I know Simon have been working on fix this issue about one month ago.
> >As now the highest priority is to meet the piglit pass rate, I prefer to
> >ignore this issue currently.
> >
> >What's your opinion?
> Really no need to do it so hurry.  Just a small performance side
> effect. If Simon have no
> time, I will research for it, too.
> >>
> >>On 06/13/2013 07:15 PM, Simon Richter wrote:
> >>>This replaces the incomplete list of conversion functions with a generated
> >>>one that can be updated by a script.
> >>>
> >>>In the long run, this should be generated on-the-fly via an ExternalASTSource
> >>>
> >>>Simon Richter (3):
> >>>   Generate all supported as_* functions
> >>>   Define all convert_* functions.
> >>>   Add long and ulong types to generated functions.
> >>>
> >>>  backend/src/gen_as.sh         |   83 ++
> >>>  backend/src/gen_convert.sh    |   52 +
> >>>  backend/src/genconfig.sh      |   11 +
> >>>  backend/src/ocl_stdlib.h      | 3223 ++++++++++++++++++++++++++++++++++++++++-
> >>>  backend/src/update.sh         |    3 +
> >>>  backend/src/update_as.sh      |   11 +
> >>>  backend/src/update_convert.sh |   11 +
> >>>  7 files changed, 3363 insertions(+), 31 deletions(-)
> >>>  create mode 100755 backend/src/gen_as.sh
> >>>  create mode 100755 backend/src/gen_convert.sh
> >>>  create mode 100644 backend/src/genconfig.sh
> >>>  create mode 100755 backend/src/update.sh
> >>>  create mode 100755 backend/src/update_as.sh
> >>>  create mode 100755 backend/src/update_convert.sh
> >>>
> >>_______________________________________________
> >>Beignet mailing list
> >>Beignet at lists.freedesktop.org
> >>http://lists.freedesktop.org/mailman/listinfo/beignet
> >_______________________________________________
> >Beignet mailing list
> >Beignet at lists.freedesktop.org
> >http://lists.freedesktop.org/mailman/listinfo/beignet
> >
>