osx-arm64 optimal code generation

[sdmaclea]

The Apple Silicon dev kit reports the following hardware features.

$ sysctl-a | grep hw.optional
hw.optional.floatingpoint: 1                                                   
hw.optional.watchpoint: 4           
hw.optional.breakpoint: 6                                                      
hw.optional.neon: 1                
hw.optional.neon_hpfp: 1              
hw.optional.neon_fp16: 1                                                       
hw.optional.armv8_1_atomics: 1  
hw.optional.armv8_crc32: 1                                                     
hw.optional.armv8_2_fhm: 0                                                     
hw.optional.amx_version: 0           
hw.optional.ucnormal_mem: 0                                                    
hw.optional.arm64: 1                                                           

I believe there is at least draft support for armv8_1_atomics, but given these are performance critical we should try to make sure we have used them in any perf critical code.

I believe we have enabled armv8_crc32 intrinsics

@tannergooding was looking for half precision floating point for AI work. Given that it is supported here it might be good to at least add the intrinsics. Maybe consider higher level support too.

category:cq
theme:vector-codegen
skill-level:expert
cost:large
impact:large

[BruceForstall]

@echesakovMSFT Arm64 intrinsics and Apple silicon

[echesakov]

@sdmaclea Does it support ARMv8.3-CompNum ? I was planning to work on these too.

[sdmaclea]

The above list was unabridged. It is not on the list on the Apple Silicon prototype. We will have to see what is on the commercial hardware.

[JulieLeeMSFT]

@kunalspathak this is related to atomics. Please feel free to move to .NET 8.

[kunalspathak]

Yes. I will update this with my work.

[kunalspathak]

We have handled atomics in #71512. But I would like to keep this issue open for more features enable. I will move this to 8.0

[neon-sunset]

We have handled atomics in #71512. But I would like to keep this issue open for more features enable. I will move this to 8.0

If it helps, other features are detected on osx-arm64 since 3580ba7
While it doesn't cover everything reported like Sha3 and Sha512, which might get support in JIT similar to other cryptography intrinsics, I guess all existing ARM-specific features are taken advantage of more or less since xplat Vector64/128/256 changes.

p.s.: Since 12.0, now in 13.0 Beta 3, the only new sysctl -a hw.optional line is hw.optional.arm.FEAT_DIT: 1 which is security specific and seems to be untouched for now (see https://lore.kernel.org/all/20210211125900.22777-5-peter.maydell@linaro.org/ and golang/go#49702)

[kunalspathak]

Don't think anything actionable will be done in .NET 8.