momchil-velikov wrote: I've implemented all the relevant Neon FP8 intrinsics on top of two alternative implementations for Neon FP8 vector types: * builtin vector types: https://github.com/llvm/llvm-project/pull/121804 (top of PR stack, this PR right here is a part of it) * using `neon_vector_type` attribute: https://github.com/llvm/llvm-project/pull/120476 (a single PR, multiple commits, will spin off individual PRs)
Comparing the two solutions, I would conclude that the approach using `neon_vector_type` fits better with and can take advantage of the existing Clang types framework, thus requiring less amount of special handling of the new types. One drawback is that the AAPCS64 required types `__Mfloat8x8_t` and `__Mfloat8x16_t` would not be recognised, just like all the others: https://github.com/llvm/llvm-project/issues/113297 https://github.com/llvm/llvm-project/pull/119031 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
