Enable SHA assembly on macosx-aarch64#1226
Conversation
| echo copying libcrypto source | ||
| rm -f crypto/*.c crypto/*.h | ||
| touch crypto/empty.c | ||
| git checkout crypto/crypto_assembly.h |
There was a problem hiding this comment.
This seems weird. Why is this needed?
There was a problem hiding this comment.
For reasons unknown we remove all *.h files two lines earlier... we won't need this once crypto_assembly.h is upstream though.
There was a problem hiding this comment.
The rm above could be replaced with git clean -f -- crypto/*.c crypto/*.h - the original intent was just to delete untracked files that had been copied in from a previous build, but adding a tracked .h file changes that assumption
| @@ -0,0 +1,299 @@ | |||
| Index: crypto/sha/sha1_aarch64_ce.S | |||
There was a problem hiding this comment.
Why not just check in a standalone version of sha1_aarch64_ce_darwin.S at this point? Is it necessary to have all of this indirection for the sake of having one file instead of two? How often is this implementation going to change anyway?
Things only gets worse to impossible with armasm64 from VS, and once we start touching the stack like in sha512 and handling SEH. The complexity of trying to reinvent perlasm in cpp seems it would dwarf any benefits.
There was a problem hiding this comment.
I could get mingw64 sort of working with this macro approach, but that's about it.
There was a problem hiding this comment.
I'm not against having separate sha1_aarch64_ce_darwin.S and sha1_aarch64_ce_windows.S or whatever, when there is significant divergence - there are a few options, including some form of templating/generation that we then use to create the .S and check them in (as opposed to perlasm being very opaque and part of the build process).
This branch was really just a proof-of-concept - for darwin I think the differences are resolvable and I was planning on upstreaming the majority of this diff. The mingw64/windows side is the one that I've not yet looked in to.
|
I rebased this on master and applied a few fixes; would you like me to just push to this PR, open a new one, etc.? This needs a rebase since it no longer works on the latest openbsd, or I'd just PR diffs alone. Changes here: master...busterb:portable:pr-1226 |
| endif() | ||
|
|
||
| if(HOST_ASM_MACOSX_AARCH64) | ||
| set(CRYPTO_SRC ${CRYPTO_SRC} sha/sha1_amd64.c sha/sha256_amd64.c sha/sha512_amd64.c) |
3 participants
No description provided.