[llvm-commits] [llvm] r92939 - /llvm/trunk/include/llvm/IntrinsicsX86.td

David Greene dag at cray.com
Thu Jan 7 14:20:19 PST 2010


On Thursday 07 January 2010 15:34, Dan Gohman wrote:

> Almost, but unfortunately not quite. The problem with the "scalar"
> intrinsics is that they are still defined to have vector operands.
> Theoretically the addss intrinsic could be lowered to
> extractelement+fadd+insertelement. Theoretically it could optimize down to
> the same eventual result; in practice, it'll probably take someone to look
> at it.

Interestingly enough, this would be almost trivial to do in the New SIMD Order 
(tm).  We could write this pattern once and have it apply to all relevant 
instructions (ADD, SUB, MUL, etc.).  Then we could dump the scalar intrinsics.

It's not a use I had anticipated, but it could work with some very minor
enhancements.  Kinda neat.  :)

                               -Dave



More information about the llvm-commits mailing list