[libc] [mlir] [flang] [clang] [openmp] [clang-tools-extra] [libcxxabi] [libcxx] [compiler-rt] [llvm] [AArch64] Add custom lowering for load <3 x i8>. (PR #78632)

[Florian Hahn via llvm-commits]

fhahn wrote:

Thanks for taking a look @TNorthover! I tried to address the comments, but with them addressed it turned out to not really be feasible to go down that path. I changed the codegen back to use the slightly longer (but using instructions that are cheaper/less complex usually) below, with a comment about the alternative sequence using `ld1` and why it is not used at the moment. WDYT?
```
ldrb wX, [x0, #2]
ldrh wY, [x0]
orr wX, wY, wX, lsl #16
fmov s0, wX
```

https://github.com/llvm/llvm-project/pull/78632