[llvm] [TargetParser][AArch64] Believe runtime feature detection (PR #95694)

[David Green via llvm-commits]

https://github.com/davemgreen commented:

Thanks for working on adding this. "crypto" is an old feature-name now, which has been split-up into aes+sha2, and the newer sha3+sm4 features in certain situations. It would technically be incorrect to turn "aes+sha2" into "crypto" if "crypto" is turned back into "sha2+aes+sha3+sm4". I think it might be aes = CAP_AES&CAP_PMULL and sha2 = CAP_SHA1&CAP_SHA2, so they don't necessarily fit into the LLVMFeatureStr map unfortunately.

Can we split this into individual sha2 and aes features? And can we add a test to TargetParserTest.cpp?

https://github.com/llvm/llvm-project/pull/95694