Try to support virtual devices

[maleadt]

Another stab at #309 -- things seem to work locally at least on a virtual macOS 15 on my M3 Pro.

[codecov]

Codecov Report

❌ Patch coverage is 88.23529% with 4 lines in your changes missing coverage. Please review.
✅ Project coverage is 81.05%. Comparing base (25dcb95) to head (3ebe242).
⚠️ Report is 15 commits behind head on main.

Files with missing lines	Patch %	Lines
lib/mtl/device.jl	80.00%	2 Missing ⚠️
src/state.jl	50.00%	2 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #789      +/-   ##
==========================================
+ Coverage   80.96%   81.05%   +0.09%     
==========================================
  Files          64       64              
  Lines        3057     3083      +26     
==========================================
+ Hits         2475     2499      +24     
- Misses        582      584       +2     

☔ View full report in Codecov by Sentry.
📢 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.

[github-actions]

Metal Benchmarks

Details

Benchmark suite	Current: 6fe48f0	Previous: 3b6c586	Ratio
array/accumulate/Float32/1d	759875 ns	820187.5 ns	0.93
array/accumulate/Float32/dims=1	978875 ns	1001416.5 ns	0.98
array/accumulate/Float32/dims=1L	10410916 ns	10577334 ns	0.98
array/accumulate/Float32/dims=2	1266458.5 ns	1279958 ns	0.99
array/accumulate/Float32/dims=2L	4744958.5 ns	4806791.5 ns	0.99
array/accumulate/Int64/1d	960583.5 ns	982083 ns	0.98
array/accumulate/Int64/dims=1	1109792 ns	1116021 ns	0.99
array/accumulate/Int64/dims=1L	12286895.5 ns	12565083 ns	0.98
array/accumulate/Int64/dims=2	1463709 ns	1466354.5 ns	1.00
array/accumulate/Int64/dims=2L	9376812.5 ns	8937708 ns	1.05
array/broadcast	331833 ns	329583 ns	1.01
array/construct	6041 ns	5625 ns	1.07
array/permutedims/2d	624937.5 ns	628208.5 ns	0.99
array/permutedims/3d	1127125 ns	1128667 ns	1.00
array/permutedims/4d	1359917 ns	1358834 ns	1.00
array/private/copy	352229.5 ns	365313 ns	0.96
array/private/copyto!/cpu_to_gpu	244458 ns	240750.5 ns	1.02
array/private/copyto!/gpu_to_cpu	283458.5 ns	234125 ns	1.21
array/private/copyto!/gpu_to_gpu	264937.5 ns	255375 ns	1.04
array/private/iteration/findall/bool	1105125 ns	1128729 ns	0.98
array/private/iteration/findall/int	1218729 ns	1291770.5 ns	0.94
array/private/iteration/findfirst/bool	1258750 ns	1245584 ns	1.01
array/private/iteration/findfirst/int	1298542 ns	1276500 ns	1.02
array/private/iteration/findmin/1d	1381125 ns	1432875 ns	0.96
array/private/iteration/findmin/2d	1172166.5 ns	1181438 ns	0.99
array/private/iteration/logical	1727209 ns	1770770.5 ns	0.98
array/private/iteration/scalar	1715875 ns	1476917 ns	1.16
array/random/rand/Float32	564792 ns	600333 ns	0.94
array/random/rand/Int64	660000 ns	648792 ns	1.02
array/random/rand!/Float32	507292 ns	521666 ns	0.97
array/random/rand!/Int64	486541 ns	481750 ns	1.01
array/random/randn/Float32	565333 ns	568458 ns	0.99
array/random/randn!/Float32	478875 ns	484104 ns	0.99
array/reductions/mapreduce/Float32/1d	535625 ns	498833 ns	1.07
array/reductions/mapreduce/Float32/dims=1	459750 ns	461979.5 ns	1.00
array/reductions/mapreduce/Float32/dims=1L	716333 ns	721625 ns	0.99
array/reductions/mapreduce/Float32/dims=2	461750 ns	466083.5 ns	0.99
array/reductions/mapreduce/Float32/dims=2L	1120458 ns	1094312.5 ns	1.02
array/reductions/mapreduce/Int64/1d	777750 ns	795687 ns	0.98
array/reductions/mapreduce/Int64/dims=1	770875 ns	773437.5 ns	1.00
array/reductions/mapreduce/Int64/dims=1L	1211083 ns	1222229.5 ns	0.99
array/reductions/mapreduce/Int64/dims=2	967667 ns	925458 ns	1.05
array/reductions/mapreduce/Int64/dims=2L	2260375 ns	2261708 ns	1.00
array/reductions/reduce/Float32/1d	494583 ns	496459 ns	1.00
array/reductions/reduce/Float32/dims=1	457854.5 ns	463333 ns	0.99
array/reductions/reduce/Float32/dims=1L	715792 ns	719875 ns	0.99
array/reductions/reduce/Float32/dims=2	463229.5 ns	464875 ns	1.00
array/reductions/reduce/Float32/dims=2L	1116083 ns	1096729.5 ns	1.02
array/reductions/reduce/Int64/1d	804250 ns	787958 ns	1.02
array/reductions/reduce/Int64/dims=1	763791.5 ns	767125 ns	1.00
array/reductions/reduce/Int64/dims=1L	1250625 ns	1224375 ns	1.02
array/reductions/reduce/Int64/dims=2	953667 ns	961167 ns	0.99
array/reductions/reduce/Int64/dims=2L	2262625 ns	2264104.5 ns	1.00
array/shared/copy	167750 ns	166458 ns	1.01
array/shared/copyto!/cpu_to_gpu	48000 ns	47709 ns	1.01
array/shared/copyto!/gpu_to_cpu	41166 ns	43500 ns	0.95
array/shared/copyto!/gpu_to_gpu	48708 ns	48333 ns	1.01
array/shared/iteration/findall/bool	1110625 ns	1134875 ns	0.98
array/shared/iteration/findall/int	1219000 ns	1293166.5 ns	0.94
array/shared/iteration/findfirst/bool	1026792 ns	1033333 ns	0.99
array/shared/iteration/findfirst/int	1070688 ns	1081708.5 ns	0.99
array/shared/iteration/findmin/1d	1196875 ns	1201542 ns	1.00
array/shared/iteration/findmin/2d	1174333 ns	1177146 ns	1.00
array/shared/iteration/logical	1577646 ns	1625834 ns	0.97
array/shared/iteration/scalar	8375 ns	5951.333333333333 ns	1.41
integration/byval/reference	1175041.5 ns	1174791 ns	1.00
integration/byval/slices=1	1177458 ns	1178750 ns	1.00
integration/byval/slices=2	2122833 ns	2125604 ns	1.00
integration/byval/slices=3	21131854.5 ns	19836667 ns	1.07
integration/metaldevrt	458041.5 ns	458584 ns	1.00
kernel/indexing	325666 ns	302000 ns	1.08
kernel/indexing_checked	313542 ns	325084 ns	0.96
kernel/launch	13208 ns	14125 ns	0.94
kernel/rand	271250 ns	347958 ns	0.78
latency/import	1403534666.5 ns	1405863458.5 ns	1.00
latency/precompile	29725972833 ns	30266521250 ns	0.98
latency/ttfp	1677229520.5 ns	1681368375.5 ns	1.00
metal/synchronization/context	1037.5 ns	837.6436781609195 ns	1.24
metal/synchronization/stream	665.6335403726708 ns	427.5527638190955 ns	1.56

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