[PATCH] D27738: [AArch64] Code generalization/clean-up for replacing indexed vector instructions for Exynos

[Abderrazek Zaafrani via Phabricator via llvm-commits]

az added a comment.

I agree that changing hard-coded switches by hard coded tables does not bring anything. However, the section of code that I was hoping to improve is how to build the operands of the two new instructions from the operands of the original instruction. That section of code gets more complicated if we add more instructions that do not follow exactly the fmla instruction format. I submitted a new patch to add more instructions to the pass and that patch shows in a small way this extra inconvenience (https://reviews.llvm.org/D28069).

I will abandon this changes for now and I will bring it back one day if improvement are more obvious.


https://reviews.llvm.org/D27738