[llvm-dev] Rotates, once again

[Krzysztof Parzyszek via llvm-dev]

On 5/15/2018 5:34 PM, Sanjay Patel via llvm-dev wrote:
> Thanks for writing this up. I'd like to have this intrinsic too.

I'd vote for the "valign" variant that David proposed. It becomes a 
rotate when both inputs are the same.

<ty> %result = @llvm.valign.<ty>(<ty> %a0, <ty> %a1, i32 s)
result = truncate (concat(a1,a0) >> s)

Where "concat" concatenates the bits of both inputs to form a value of a 
twice as long type, and "truncate" takes the lower half of the bits of 
its input.

-Krzysztof

-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum, 
hosted by The Linux Foundation