VittorioBurani wrote:

Update: reworked per @lenary's point on intrinsics vs IR types.

The element-wise XCVsimd builtins that map to plain vector ops are removed: 
min/minu/max/maxu/and/or/xor/abs (and add_b/sub_b). These are now expressed as 
native operations on `4 x i8` / `2 x i16` vector types and lowered by the 
backend (#204880). The scalar-broadcast (sc) builtins are kept. The 
shift/avg/compare builtins remain for now, pending the matching backend change.

Depends on #204880.


https://github.com/llvm/llvm-project/pull/204881
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to