Hi Jiangning, On 26 Aug 2013, at 11:13, Jiangning Liu <[email protected]> wrote: > int_arm_neon_vaddhn/int_arm_neon_vmulls and friends are all defined by ARM > target. I agree with your comments, but does it imply ARM back-end has > inappropriate implementation?
Certainly suboptimal, I'd say. The ARM implementation looks like it dates from at least 2009, when LLVM didn't have anything approaching auto-vectorisation, OpenCL was just a glint in Khronos's eye and the only real way to get vector code was via the NEON intrinsics. I'd be willing to work on the ARM side if you like; if we coordinate the changes then we could switch Clang over to using the more natural IR on both ARM and AArch64 simultaneously. Cheers. Tim. _______________________________________________ cfe-commits mailing list [email protected] http://lists.cs.uiuc.edu/mailman/listinfo/cfe-commits
