Allo native builds on Mac OS M1 / Apple Silicon
rP26881
Actions

Description

Allo native builds on Mac OS M1 / Apple Silicon

Update GMP to a version that supports arm64 darwin
Patch spidermonkey 78 to build and run on arm64 darwin
Choose the correct architecture dynamically in build scripts
Include python workaround that fixes python related errors on spidermonkey build

Based on a patch by: kumikumi
Based on a patch by: Langbart

Fixes #6474

Differential Revision: https://code.wildfiregames.com/D4607

Details

Committed

wraitii

May 15 2022, 8:45 AM

Differential Revision

D4607: Fix build on MacOS Darwin arm64 / aarch64 (M1 / Apple Silicon)

Parents

rP26880: Fix rP26879

Branches

Unknown

Tags

Unknown

Build Status

Buildable 20317
Build 48883: Trigger Windows Autobuild
Build 48882: Post-Commit Build	Jenkins
Build 48881: Post-Commit Build (macOS)	Jenkins

Event Timeline

wraitii committed rP26881: Allo native builds on Mac OS M1 / Apple Silicon.May 15 2022, 8:45 AM

wraitii added an edge: D4607: Fix build on MacOS Darwin arm64 / aarch64 (M1 / Apple Silicon).

Harbormaster failed to build B20317: rP26881: Allo native builds on Mac OS M1 / Apple Silicon!May 15 2022, 9:00 AM

vladislavbelov added a subscriber: vladislavbelov.May 15 2022, 6:47 PM

vladislavbelov added inline comments.

/ps/trunk/source/graphics/Color.cpp
86–88	It doesn't look like a proper place to add the arch check.

wraitii mentioned this in D4647: Fix SM compilation on M1.May 15 2022, 6:47 PM

wraitii added inline comments.May 15 2022, 6:48 PM

/ps/trunk/source/graphics/Color.cpp
86–88	Well, the problem is that SSE is an x86 only thing, so the whole precompiled logic ought be improved.

Stan added a subscriber: Stan.May 15 2022, 6:51 PM

Stan added inline comments.

/ps/trunk/source/graphics/Color.cpp
86–88	Were you just fixing the tests ? In that case you might have just added a TestLogger in the relevant function. This message shows up on a Raspbery Pi as well.

wraitii added inline comments.May 15 2022, 6:54 PM

/ps/trunk/source/graphics/Color.cpp
86–88	Were you just fixing the tests ? In that case you might have just added a TestLogger in the relevant function. This message shows up on a Raspbery Pi as well. I don't remember, but the message is broken -> there's no SSE available, so it's not a 'slow fallback routine'. Likewise on raspberry PI I guess, though whether that victories anything is out of my knowledge.

vladislavbelov added inline comments.May 15 2022, 6:55 PM

/ps/trunk/source/graphics/Color.cpp
86–88	Nevertheless it doesn't make the code correct, with less error it could be added to `COMPILER_HAS_SSE`.

wraitii added inline comments.May 15 2022, 6:59 PM

/ps/trunk/source/graphics/Color.cpp
86–88	It could not because that branch uses x86 specific intrisics IIRC, and the whole define is meaningless. That being said, yes, this is not the best way to do this. I'm not up for refactoring the whole processor logic just yet.

vladislavbelov added inline comments.May 15 2022, 7:07 PM

/ps/trunk/source/graphics/Color.cpp
86–88	How does it contradicts? Only x86 (and related) supports SSE, so the sentence `No SSE available. Slow fallback routines will be used.` is true for all other archs. That's why the current approach is worse.

Question how do I build a X86 build on my M1 now ? I can pass arch to the libs, but not to updateworkspaces.sh, nor premake for the game

In rP26881#57875, @Stan wrote:

Question how do I build a X86 build on my M1 now ? I can pass arch to the libs, but not to updateworkspaces.sh, nor premake for the game

You can pass ARCH to update-workspaces.sh. That sets the workspaces for x86.

/ps/trunk/source/graphics/Color.cpp
86–88	I think it implies that the fallback routine is slow, when the 'fallback routine' can be vector optimised by the compiler (in the M1 case, is).