No optimizations when building on Apple arm64

[katjav]

Makefile.pdlibbuilder defines target architecture according to the (native or cross-) compiler's build triplet and specifies optimization options per architecture.

As I understand (but can not verify myself) "Apple Silicon" has architecture arm64. This architecture is not defined in Makefile.pdlibbuilder, only arm and aarch64. So I guess that externals built for "Apple Silicon" run unoptimized currently.

Can anyone confirm? To check the makefile definitions: in a project with Makefile.pdlibbuilder do make vars and check variables dumpmachine.cpu and optimization.flags.

[katjav]

It seems that arm64 and aarch64 are synonyms for the large arm 64 bit processor family. Raspberry Pi 4 identifies as aarch64 (as I've checked withgcc -dumpmachineand uname -m) and "Apple silicon" as arm64 (as I could read online but not verify). Fortunately not the same, so they can be distinguished for platform detection. But apart from how they identify, it may still be the case that a build without options for MacOS arm64 uses the same instruction set as a build for Raspberry Pi aarch64 when a compiler is conservative. I don't know how to find out. In any case here is a somewhat comprehensive blog article about architecture options for ARM that may be useful for future reference:

https://community.arm.com/arm-community-blogs/b/tools-software-ides-blog/posts/compiler-flags-across-architectures-march-mtune-and-mcpu

[katjav]

I come to realize that passing appropriate architecture options seems most problematic for a fat binary to be built on "Apple Silicon". Current Makefile.pdlibbuilder will define no optimizations at all. When optimizations would be defined for arm64, the x86_64 slice will remain unoptimized.

For Intel 32/64 fat builds this was never a problem. With a build (fat or not) done on x86_64 you get these options:

arch.c.flags = -march=core2 -mfpmath=sse -msse -msse2 -msse3

and all of these happen to be valid for the i386 slice too because the first Intel Macs were core2 (edit: core duo, supporting the sse* options as defined).

Edit: we never passed -m options for OSX/MacOS but only the -arch option and general optimization flags.**

But building fat on "Apple Silicon" the situation is very different. We can no longer define a single set of optimization flags for the compiler and just add the target architectures when building fat. Apple has the following recommendation:

For makefiles you create outside of Xcode, pass the appropriate architecture values to the compiler using the -target option. The following example shows a makefile that compiles a single-source file twice—once for each architecture. It then creates a universal binary by merging the resulting executable files together with the lipo tool.

x86_app: main.c
    $(CC) main.c -o x86_app -target x86_64-apple-macos10.12
arm_app: main.c
    $(CC) main.c -o arm_app -target arm64-apple-macos11
universal_app: x86_app arm_app
    lipo -create -output universal_app x86_app arm_app

(from https://developer.apple.com/documentation/apple-silicon/building-a-universal-macos-binary)

[umlaeute]

as I could read online but not verify

macOS/Apple Silicon (M1)

me@mymac ~ % uname -a
Darwin mymac.local 24.6.0 Darwin Kernel Version 24.6.0: Mon Jul 14 11:30:29 PDT 2025; root:xnu-11417.140.69~1/RELEASE_ARM64_T6000 arm64
me@mymac ~ % gcc -dumpmachine
arm64-apple-darwin24.6.0
zmoelnig@tyndall ~ % 

Linux/arm64

(this is on a Debian/porterbox rather than an RPi4, but since Raspbian is based on Debian i guess they are pretty much the same)

me@amdahl:~$ uname -a
Linux amdahl 6.12.63+deb13-arm64 #1 SMP Debian 6.12.63-1 (2025-12-30) aarch64 GNU/Linux
me@amdahl:~$ gcc -dumpmachine
aarch64-linux-gnu

Windows/arm64

• see Windows arm64 support #83

$ cc -dumpmachine
aarch64-w64-windows-gnu