Generate MTL and MPS structs and enums with Clang.jl

[christiangnrd]

Currently works for structs and enums in Metal.framework with a few hacks and this branch of Clang devved: https://github.com/christiangnrd/Clang.jl/tree/objectiveC

Any advice, suggestions, or help welcome!

[maleadt]

Even if only for enums etc, this would be pretty useful! I imagined having to write our own Clang.jl-based generator in order to generate ObjC calls, but this would probably be a good first step.

[christiangnrd]

I imagined having to write our own Clang.jl-based generator in order to generate ObjC calls, but this would probably be a good first step.

I think most of it can be upstreamed to Clang.jl. The changes I've done on my branch are either bug fixes or definitions of the Generators types for Objective-C constructs.

[github-actions]

Metal Benchmarks

Benchmark suite	Current: e47ff83	Previous: 5056e33	Ratio
private array/construct	26472.25 ns	25920.083333333336 ns	1.02
private array/broadcast	460958 ns	464792 ns	0.99
private array/random/randn/Float32	803541 ns	811812 ns	0.99
private array/random/randn!/Float32	660875 ns	673333 ns	0.98
private array/random/rand!/Int64	556041 ns	553250 ns	1.01
private array/random/rand!/Float32	591354.5 ns	601000 ns	0.98
private array/random/rand/Int64	761250 ns	763896 ns	1.00
private array/random/rand/Float32	636666.5 ns	616375 ns	1.03
private array/copyto!/gpu_to_gpu	670291 ns	660854.5 ns	1.01
private array/copyto!/cpu_to_gpu	822166 ns	619854 ns	1.33
private array/copyto!/gpu_to_cpu	591125 ns	834687.5 ns	0.71
private array/accumulate/1d	1345146.5 ns	1328542 ns	1.01
private array/accumulate/2d	1401125 ns	1387334 ns	1.01
private array/iteration/findall/int	2106875 ns	2070500 ns	1.02
private array/iteration/findall/bool	1843916 ns	1824916 ns	1.01
private array/iteration/findfirst/int	1700958 ns	1682146 ns	1.01
private array/iteration/findfirst/bool	1676417 ns	1637959 ns	1.02
private array/iteration/scalar	3408875 ns	3891833 ns	0.88
private array/iteration/logical	3199459 ns	3177708.5 ns	1.01
private array/iteration/findmin/1d	1775959 ns	1740229 ns	1.02
private array/iteration/findmin/2d	1346750 ns	1343375 ns	1.00
private array/reductions/reduce/1d	1044979.5 ns	1035708 ns	1.01
private array/reductions/reduce/2d	664542 ns	651667 ns	1.02
private array/reductions/mapreduce/1d	1039292 ns	1037167 ns	1.00
private array/reductions/mapreduce/2d	670541 ns	657917 ns	1.02
private array/permutedims/4d	2572375 ns	2537625 ns	1.01
private array/permutedims/2d	1021666 ns	1022708 ns	1.00
private array/permutedims/3d	1600749.5 ns	1577291.5 ns	1.01
private array/copy	582667 ns	621750 ns	0.94
latency/precompile	5770735750.5 ns	5243893542 ns	1.10
latency/ttfp	6564931562.5 ns	6538101604 ns	1.00
latency/import	1172940125 ns	1165440583 ns	1.01
integration/metaldevrt	707354.5 ns	705833 ns	1.00
integration/byval/slices=1	1581729.5 ns	1588833.5 ns	1.00
integration/byval/slices=3	11089208.5 ns	10079959 ns	1.10
integration/byval/reference	1576209 ns	1568145.5 ns	1.01
integration/byval/slices=2	2634791.5 ns	2643625 ns	1.00
kernel/indexing	456291 ns	444959 ns	1.03
kernel/indexing_checked	462834 ns	446541 ns	1.04
kernel/launch	8166 ns	10666.666666666666 ns	0.77
metal/synchronization/stream	15000 ns	15167 ns	0.99
metal/synchronization/context	15250 ns	15833 ns	0.96
shared array/construct	27764 ns	25503.5 ns	1.09
shared array/broadcast	475500 ns	475125 ns	1.00
shared array/random/randn/Float32	829791.5 ns	752770.5 ns	1.10
shared array/random/randn!/Float32	666417 ns	655458 ns	1.02
shared array/random/rand!/Int64	570645.5 ns	560916 ns	1.02
shared array/random/rand!/Float32	591542 ns	598459 ns	0.99
shared array/random/rand/Int64	783292 ns	774208 ns	1.01
shared array/random/rand/Float32	647541 ns	629583.5 ns	1.03
shared array/copyto!/gpu_to_gpu	88834 ns	85500 ns	1.04
shared array/copyto!/cpu_to_gpu	88875 ns	91792 ns	0.97
shared array/copyto!/gpu_to_cpu	78916 ns	77917 ns	1.01
shared array/accumulate/1d	1355104 ns	1347021 ns	1.01
shared array/accumulate/2d	1377334 ns	1383874.5 ns	1.00
shared array/iteration/findall/int	1815166 ns	1783167 ns	1.02
shared array/iteration/findall/bool	1583292 ns	1585458 ns	1.00
shared array/iteration/findfirst/int	1391896 ns	1392020.5 ns	1.00
shared array/iteration/findfirst/bool	1367375 ns	1354916.5 ns	1.01
shared array/iteration/scalar	159834 ns	162042 ns	0.99
shared array/iteration/logical	2976229 ns	2968958 ns	1.00
shared array/iteration/findmin/1d	1471125 ns	1461729.5 ns	1.01
shared array/iteration/findmin/2d	1367542 ns	1364792 ns	1.00
shared array/reductions/reduce/1d	729729.5 ns	730458 ns	1.00
shared array/reductions/reduce/2d	677979 ns	656958 ns	1.03
shared array/reductions/mapreduce/1d	744062.5 ns	746250 ns	1.00
shared array/reductions/mapreduce/2d	675271 ns	660562.5 ns	1.02
shared array/permutedims/4d	2566270.5 ns	2528541.5 ns	1.01
shared array/permutedims/2d	1011000 ns	1025917 ns	0.99
shared array/permutedims/3d	1598583 ns	1580896 ns	1.01
shared array/copy	244645.5 ns	242437.5 ns	1.01

This comment was automatically generated by workflow using github-action-benchmark.

[maleadt]

Please ping when this is ready for review; you've marked it as such, but more commits keep appearing 🙂

Also, were the necessary changes to Clang.jl merged?

[christiangnrd]

Please ping when this is ready for review; you've marked it as such, but more commits keep appearing 🙂

I do have a terrible habit of refining my PRs after marking ready. I'll be more mindful of that in the future.

Also, were the necessary changes to Clang.jl merged?

Not yet, JuliaInterop/Clang.jl#519 and JuliaInterop/Clang.jl#522 are general improvements/bug fixes to handle the types of enums that Apple ObjectiveC headers love so much (elaborated and attributes). I just created JuliaInterop/Clang.jl#524 which once the other 2 are merged, shouldn't be too hard to review.

I just rebased and any further changes other than review changes will be for a future PR.