cbalint13 commented on code in PR #16425:
URL: https://github.com/apache/tvm/pull/16425#discussion_r1457343144
##########
src/target/llvm/llvm_instance.cc:
##########
@@ -845,10 +847,11 @@ const Array<String>
LLVMTargetInfo::GetAllLLVMCpuFeatures() const {
#else
MCInfo->getAllProcessorFeatures();
#endif
- Array<String> cpu_features;
+ // TVM doesn't have an FFI friendly Set, so use a Map instead for now
+ Map<String, String> cpu_features;
Review Comment:
> Good point thanks,
However I think my request concerns the pedantic side and ignores the
performance gains.
>I'd forgotten to check this. Due to the restriction on the FFI, perhaps we
can fix this on the python side? For example:
>
> ```
> feature_map = _ffi_api.llvm_get_cpu_features(target)
> return set(feature_map.keys())
> ```
Super !
I think this would be enough for python side.
>
> We could also use std::set for the c++ API if desired, we'd just have to
make the conversion to/from Map before/after crossing the FFI boundary.
See, well I think ```Map``` is fine and it may be the fastest option
concerning lookup speeds in hot-paths.
Thanks @lhutton1 !
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: commits-unsubscr...@tvm.apache.org
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
