In D41599#1041200, @RKSimon wrote:

@uriel.k Sorry, this patch and D41168 fell off my radar for a while - are you still looking at this?

Simon, is there anything else you think that is needed to be changed before accepting the revision?
Thanks

In D41599#964739, @RKSimon wrote:

Won't this mean that explicit calls to the SSE sqrt intrinsics may be converted to the rsqrt+NR estimates in some cases?

now takes care of sqrt with round mode CURR_DIRECTION

Added the case of the pattern with X86::FAND. Hopefully this is the last update. Please let me know if there is something else I missed.

Added fast-isel tests.
Note that for avx512bw there is an open bug that responsible for that huge code size.

fixed requested changes by Simon and Craig and as you can see added the new test files to trunk as NFC commit and showed the diff.

In D38732#893224, @RKSimon wrote:

Add the new test files to trunk with current codegen and then rebase to show the diff from this patch.

Committed in rL315311. Sorry for the poor commit message, added a suitable message in comment in https://reviews.llvm.org/rL315311

Sorry for the poor commit message. Should have been:

Changed function to be DAG.getIntPtrConstant(0, dl) instead.

committed in 314195

change after Michael comment

committed in rL313134

Updated IR code in avx2-intrinsics-fast-isel.ll and ssse3-intrinsics-fast-isel.ll (figured this file should also be updated) to have the IR code generated from clang for the ABS intrinsics instead of the intrinsics' calls.

If I remove the builtins from the .def file this will not work, as I didn't replace the builtin call in the Header files but implemented the lowering by adding cases to the CGBuiltin.cpp and adding the IR by hand. Correct me if I'm wrong about this.