[PATCH] AArch64: big endian constant vector pools

[Tim Northover]

Hi Jiangning,

> So now I'm confused and don't understand why solution 1 can be an option for
> big-endian?

My position has always been that both are viable, with different
trade-offs between performance, clarity and ease of implementation.
The one thing that won't work is mixing the two approaches.

Frankly, if I was doing the work I'd probably implement both of them
on different branches and then decide for myself which one turned out
better in the end. That's not something it's really fair to ask anyone
else to do though.

> Are you changing your mind and arguing the consistent in-register data
> layout for big-endian and little-endian? I'm personally happy with that, but
> it will not work together with GCC, and is violating aapcs64, isn't it?

To make it work, you have to fiddle the vectors into AAPCS-specified
form at every ABI-visible boundary (hence the changes to LowerCall
etc).

Cheers.

Tim.