Add atomic fences for Metal 3.2+

[christiangnrd]

This is the fence part of #870.

Probably needs better tests since this only checks that things aren't loudly failing and atomic stuff doesn't seem to be validated anywhere other than the compiler we bypass

[github-actions]

Metal Benchmarks

Details

Benchmark suite	Current: 997fa62	Previous: ac0202d	Ratio
array/accumulate/Float32/1d	419125 ns	407375 ns	1.03
array/accumulate/Float32/dims=1	394208 ns	351791 ns	1.12
array/accumulate/Float32/dims=1L	8881334 ns	8796584 ns	1.01
array/accumulate/Float32/dims=2	433708 ns	439792 ns	0.99
array/accumulate/Float32/dims=2L	2377459 ns	2603208 ns	0.91
array/accumulate/Int64/1d	878291 ns	842333 ns	1.04
array/accumulate/Int64/dims=1	950833 ns	944125 ns	1.01
array/accumulate/Int64/dims=1L	9680167 ns	9521000 ns	1.02
array/accumulate/Int64/dims=2	948250 ns	1247208 ns	0.76
array/accumulate/Int64/dims=2L	6758667 ns	6499125 ns	1.04
array/broadcast	191000 ns	222000 ns	0.86
array/construct	2416 ns	2417 ns	1.00
array/permutedims/2d	461375 ns	449458 ns	1.03
array/permutedims/3d	701875 ns	1011000 ns	0.69
array/permutedims/4d	1034500 ns	1275416 ns	0.81
array/private/copy	244542 ns	229791 ns	1.06
array/private/copyto!/cpu_to_gpu	200875 ns	195833 ns	1.03
array/private/copyto!/gpu_to_cpu	206167 ns	194167 ns	1.06
array/private/copyto!/gpu_to_gpu	182709 ns	173708 ns	1.05
array/private/iteration/findall/bool	1126416 ns	1057250 ns	1.07
array/private/iteration/findall/int	1252666 ns	1228125 ns	1.02
array/private/iteration/findfirst/bool	1095042 ns	1111542 ns	0.99
array/private/iteration/findfirst/int	1112000 ns	1117792 ns	0.99
array/private/iteration/findmin/1d	1276542 ns	1229625 ns	1.04
array/private/iteration/findmin/2d	977541 ns	769709 ns	1.27
array/private/iteration/logical	1743166 ns	1688959 ns	1.03
array/private/iteration/scalar	1199125 ns	1216958 ns	0.99
array/random/rand/Float32	438875 ns	426125 ns	1.03
array/random/rand/Int64	543917 ns	513417 ns	1.06
array/random/rand!/Float32	371958 ns	313042 ns	1.19
array/random/rand!/Int64	327167 ns	406791 ns	0.80
array/random/randn/Float32	371583 ns	391833 ns	0.95
array/random/randn!/Float32	334250 ns	340208 ns	0.98
array/reductions/mapreduce/Float32/1d	391541 ns	383792 ns	1.02
array/reductions/mapreduce/Float32/dims=1	325667 ns	263000 ns	1.24
array/reductions/mapreduce/Float32/dims=1L	638792 ns	621791 ns	1.03
array/reductions/mapreduce/Float32/dims=2	328875 ns	275708 ns	1.19
array/reductions/mapreduce/Float32/dims=2L	781042 ns	841125 ns	0.93
array/reductions/mapreduce/Int64/1d	588958 ns	621125 ns	0.95
array/reductions/mapreduce/Int64/dims=1	547875 ns	450958 ns	1.21
array/reductions/mapreduce/Int64/dims=1L	1051917 ns	1026500 ns	1.02
array/reductions/mapreduce/Int64/dims=2	514333 ns	778292 ns	0.66
array/reductions/mapreduce/Int64/dims=2L	2162167 ns	2179166 ns	0.99
array/reductions/reduce/Float32/1d	387584 ns	412916 ns	0.94
array/reductions/reduce/Float32/dims=1	281125 ns	345084 ns	0.81
array/reductions/reduce/Float32/dims=1L	642250 ns	635167 ns	1.01
array/reductions/reduce/Float32/dims=2	202000 ns	228750 ns	0.88
array/reductions/reduce/Float32/dims=2L	442084 ns	448708 ns	0.99
array/reductions/reduce/Int64/1d	580667 ns	576542 ns	1.01
array/reductions/reduce/Int64/dims=1	543625 ns	614541 ns	0.88
array/reductions/reduce/Int64/dims=1L	1036042 ns	1030125 ns	1.01
array/reductions/reduce/Int64/dims=2	227916 ns	240250 ns	0.95
array/reductions/reduce/Int64/dims=2L	633084 ns	644833 ns	0.98
array/shared/copy	143250 ns	136583 ns	1.05
array/shared/copyto!/cpu_to_gpu	37625 ns	38292 ns	0.98
array/shared/copyto!/gpu_to_cpu	37083 ns	38167 ns	0.97
array/shared/copyto!/gpu_to_gpu	38208 ns	38917 ns	0.98
array/shared/iteration/findall/bool	1137542 ns	1076292 ns	1.06
array/shared/iteration/findall/int	1143958 ns	1220458 ns	0.94
array/shared/iteration/findfirst/bool	782166 ns	956459 ns	0.82
array/shared/iteration/findfirst/int	754625 ns	650041 ns	1.16
array/shared/iteration/findmin/1d	959375 ns	1089541 ns	0.88
array/shared/iteration/findmin/2d	1022458 ns	1075500 ns	0.95
array/shared/iteration/logical	1517292 ns	1553459 ns	0.98
array/shared/iteration/scalar	3953.125 ns	4023.714285714286 ns	0.98
array/sorting/1d	1858459 ns	1786083 ns	1.04
array/sorting/2d	8326292 ns	8305833 ns	1.00
integration/byval/reference	1103167 ns	1119042 ns	0.99
integration/byval/slices=1	1106625 ns	1109583 ns	1.00
integration/byval/slices=2	2039500 ns	2021291 ns	1.01
integration/byval/slices=3	14466792 ns	6610875 ns	2.19
integration/metaldevrt	309250 ns	379833 ns	0.81
kernel/indexing	173500 ns	208708 ns	0.83
kernel/indexing_checked	367375 ns	370667 ns	0.99
kernel/launch	1908.3 ns	1883.3 ns	1.01
kernel/rand	341709 ns	393792 ns	0.87
latency/import	2065032584 ns	2038510291 ns	1.01
latency/precompile	39245643708 ns	38792921333 ns	1.01
latency/ttfp	2405619625 ns	2354706125 ns	1.02
metal/synchronization/context	578.775956284153 ns	583.7912087912088 ns	0.99
metal/synchronization/stream	328.3333333333333 ns	332.58744394618833 ns	0.99

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

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 85.93%. Comparing base (ac0202d) to head (997fa62).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #875      +/-   ##
==========================================
- Coverage   86.21%   85.93%   -0.29%     
==========================================
  Files          76       76              
  Lines        5246     5246              
==========================================
- Hits         4523     4508      -15     
- Misses        723      738      +15     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.