[clang] [RISCV] Disable rounding of aggregate return/arguments to iXLen. (PR #184736)

[Craig Topper via cfe-commits]

topperc wrote:

> I checked several cases in GCC, it seems to match GCC's behavior after this change.
> 
> And I am curious about how many performance gain we can have.

I forgot to say that I found this while exploring 32-bit vector types on RV64 for the P extension intrinsics.

https://github.com/llvm/llvm-project/pull/184736