[LLVMdev] RFC: Loop distribution/Partial vectorization

Krzysztof Parzyszek kparzysz at codeaurora.org
Wed Jan 14 08:25:05 PST 2015


On 1/13/2015 1:05 PM, Adam Nemet wrote:
>
> This would take some improvements to the current memory dependence checker which handles a single loop right now.  Do you have some standard algorithms in mind that would benefit from this?

The first candidate would be the loop distribution itself---it would be 
able to handle more cases, even if its purpose is vectorization.

Overall, any loop nest transformation would be a consumer of such 
information, but that probably wouldn't get past polly advocates. :)

If I come up with another motivating example other than the above, I'll 
let you know.

-Krzysztof

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



More information about the llvm-dev mailing list