Double precision not working with -cl-std=c++

[davidrohr]

Bugzilla Link	40603
Resolution	FIXED
Resolved on	Feb 12, 2019 03:03
Version	8.0
OS	Linux
Blocks	#39678
CC	@AnastasiaStulova,@zmodem

Extended Description

Usage of double precision floating point works with command line argument -cl-std=CL2.0 but not with -cl-std=c++.
Compiling with -cl-std=c++ yields the following error:
error: use of type 'double' requires cl_khr_fp64 extension to be enabled
However, enabling the cl_khr_fp64 extensions yields:
warning: unsupported OpenCL extension 'cl_khr_fp64' - ignoring [-Wignored-pragmas]
For reference: the cl_khr_fp64 extensions works when I set a different -cl-std, e.g. CL2.0

[AnastasiaStulova]

Thanks for reporting this bug!

The fix for this issue is in the review:
https://reviews.llvm.org/D57824

I will commit to master that will end up in the future release of clang 9.0.

Do you need this to be back ported to clang 8.0 as well? If yes, I will have to request this asap.

[davidrohr]

Hi, yes, if you could backport to clang 8, that would be very usefaul.
Thanks.

[AnastasiaStulova]

The fix has been committed in Clang@r353431. This issue now blocks the release 8.0.0, so it should be merged into the release branch soon.

Keeping this open until it's merged into the release branch.

[zmodem]

Merged in r353826.