[PATCH] D50791: [SelectionDAG] unroll unsupported vector FP ops earlier to avoid libcalls on undef elements (PR38527)

[Sanjay Patel via Phabricator via llvm-commits]

spatel added a comment.

In https://reviews.llvm.org/D50791#1201420, @craig.topper wrote:

> Thinking forward to a future where we have SVML library support for v4f32 and a v8f32 sin/cos. What should we do for v5f32?


Not sure. Over in:
https://bugs.llvm.org/show_bug.cgi?id=38528
...we discussed transforming to veclib calls late in IR, so any special lib support would already be taken care of before we reach this point?

Another note about the current patch: I did try sinking the expansion check directly into WidenVecRes_Unary(N), so it would apply to all of the unary opcodes in the switch rather than just the FP libcall opcodes. But that results in a regression in test/CodeGen/X86/bswap-vector.ll because we scalarize something that currently gets turned into a shuffle.


https://reviews.llvm.org/D50791