os/mac/hardware: use Westmere on >= Ventura #16392
Conversation
| nehalem: "-march=nehalem", | ||
| core2: "-march=core2", | ||
| core: "-march=prescott", | ||
| arm_vortex_tempest: "", | ||
| arm_vortex_tempest: "", # TODO: -mcpu=apple-m1 when we've patched all our GCCs to support it |
There was a problem hiding this comment.
Some earlier versions of Xcode may barf on this too.
There was a problem hiding this comment.
In terms of LLVM: added in LLVM 13, though we don't support versioned LLVM as a compiler yet.
That would equate to Xcode 13.3 but it's very possible Apple had it in their internal Clang earlier. If we find any of Xcode 12.2-13.2 don't support it, we could probably swap it on shim level to -mcpu=apple-a13 or something (not quite the same but close enough). We don't support 12.2-12.4 in the CLT so we could drop that on the Xcode side should -mcpu=apple-m1 only be a problem there.
There was a problem hiding this comment.
Yea, Apple clang supported the flag earlier than upstream LLVM did IIRC. ("earlier" in a corresponding upstream LLVM version sense -- I think it was the latest released version of Xcode when upstream LLVM 13 dropped.)
NBD to enable this flag as-is as soon as we have sufficient GCC support -- we can fix any problems that come up as they emerge.
There was a problem hiding this comment.
I checked Apple's OSS LLVM fork but it wasn't there prior to it landing in LLVM so it seems they kept it in their private builds initially.
I can easily test with some CLT versions when the time arrives. The ones to test really are Xcode/CLT 12.2, 12.5 and 13.0 as those are the main big Clang branches, narrowing down further if necessary.
|
Thanks again @Bo98! |
Although we can't use anything with AVX in order to support Rosetta (similar to how Apple also avoid AVX in their own tools), we can do a minor bump to Westmere which adds
CLMULinstructions that Rosetta 2 supports.Not much of a difference to Nehalem, but it's something.