[libclc] [libclc] Produce subnormal results for exp/exp2 on FP32 (PR #212696)

[via cfe-commits]

Lurie97 wrote:

@arsenm @wenju-he

Could you take another look at the latest revision when you have a chance?

The previous review comments should all be addressed now. While working through CL_FP_DENORM enablement in Mesa, this patch turned out to be required for clean OpenCL CTS results, as exp/exp2 (and therefore erfc) still fail when subnormal results are flushed to zero.

Thanks!

https://github.com/llvm/llvm-project/pull/212696