Fix MPSNDArrayDescriptor wrapper

[christiangnrd]

I realized automatically reversing dimensions in the descriptor as part of the low-level wrappers is a terrible idea.

Also a few more improvements

Also using this as an opportunity to test the [skip benchmark] and benchmarking from a fork. Edit: It works

[codecov]

Codecov Report

Attention: Patch coverage is 41.17647% with 10 lines in your changes missing coverage. Please review.

Project coverage is 76.04%. Comparing base (52d7056) to head (2b16b45).
Report is 386 commits behind head on main.

Files with missing lines	Patch %	Lines
lib/mps/ndarray.jl	43.75%	9 Missing ⚠️
lib/mps/MPS.jl	0.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #502      +/-   ##
==========================================
+ Coverage   71.04%   76.04%   +5.00%     
==========================================
  Files          36       57      +21     
  Lines        1143     2793    +1650     
==========================================
+ Hits          812     2124    +1312     
- Misses        331      669     +338     

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

[christiangnrd]

@maleadt It seems like PR benchmark jobs are not commenting and are being uploaded because

Metal.jl/.github/workflows/Benchmark.yml

Line 45 in 5a8daaa

comment-always: ${{ github.event_name == 'pull_request' }}

and

Metal.jl/.github/workflows/Benchmark.yml

Line 49 in 5a8daaa

auto-push: ${{ github.event_name != 'pull_request' }}

were overlooked when changing pull_request to pull_request_target.

Should I add it to this PR or would it be better for you to commit the fix directly?

[maleadt]

Ah oops. Feel free to commit directly on master, because the modified Benchmark.yml on this PR wouldn't get loaded (at least I think it wouldn't, because of using the pull_request_base instead of pull_request).

We should probably also edit out or revert the upload on the gh-pages branch.

[christiangnrd]

We should probably also edit out or revert the upload on the gh-pages branch.

Done. I assume the benchmark page will update next time a benchmark action runs?

Feel free to commit directly on master

I don't think I have permission to commit directly to main. I tried and got the following error:

remote: error: GH006: Protected branch update failed for refs/heads/main.
remote: 
remote: - Changes must be made through a pull request.
To https://github.com/JuliaGPU/Metal.jl.git
 ! [remote rejected]   main -> main (protected branch hook declined)
error: failed to push some refs to 'https://github.com/JuliaGPU/Metal.jl.git'

[github-actions]

Metal Benchmarks

Benchmark suite	Current: 2b16b45	Previous: 5a8daaa	Ratio
private array/construct	26680.5 ns	26673.5 ns	1.00
private array/broadcast	462104.5 ns	458208 ns	1.01
private array/random/randn/Float32	745125 ns	852625 ns	0.87
private array/random/randn!/Float32	631375 ns	658334 ns	0.96
private array/random/rand!/Int64	538625 ns	555625 ns	0.97
private array/random/rand!/Float32	577333 ns	589208 ns	0.98
private array/random/rand/Int64	772188 ns	802125 ns	0.96
private array/random/rand/Float32	621291.5 ns	595625 ns	1.04
private array/copyto!/gpu_to_gpu	643020.5 ns	679834 ns	0.95
private array/copyto!/cpu_to_gpu	604041.5 ns	666958.5 ns	0.91
private array/copyto!/gpu_to_cpu	819812 ns	518021 ns	1.58
private array/accumulate/1d	1316917 ns	1339437.5 ns	0.98
private array/accumulate/2d	1371666 ns	1374708 ns	1.00
private array/iteration/findall/int	2087834 ns	2065062.5 ns	1.01
private array/iteration/findall/bool	1799333.5 ns	1816354 ns	0.99
private array/iteration/findfirst/int	1689500 ns	1690417 ns	1.00
private array/iteration/findfirst/bool	1664583 ns	1658583 ns	1.00
private array/iteration/scalar	3601875 ns	3599687.5 ns	1.00
private array/iteration/logical	3197625 ns	3169854 ns	1.01
private array/iteration/findmin/1d	1753916 ns	1746000 ns	1.00
private array/iteration/findmin/2d	1340791 ns	1351313 ns	0.99
private array/reductions/reduce/1d	1046833 ns	1041021 ns	1.01
private array/reductions/reduce/2d	655875 ns	655041 ns	1.00
private array/reductions/mapreduce/1d	1035812.5 ns	1040833 ns	1.00
private array/reductions/mapreduce/2d	660417 ns	660979.5 ns	1.00
private array/permutedims/4d	2540625 ns	2523625 ns	1.01
private array/permutedims/2d	1017917 ns	1020958 ns	1.00
private array/permutedims/3d	1584167 ns	1584104.5 ns	1.00
private array/copy	560166 ns	580395.5 ns	0.97
latency/precompile	5263931459 ns	5246987875 ns	1.00
latency/ttfp	6619689354 ns	6666328874.5 ns	0.99
latency/import	1164853499.5 ns	1165891041.5 ns	1.00
integration/metaldevrt	684146 ns	718542 ns	0.95
integration/byval/slices=1	1520958 ns	1537583 ns	0.99
integration/byval/slices=3	10693916.5 ns	8342167 ns	1.28
integration/byval/reference	1622854.5 ns	1552229 ns	1.05
integration/byval/slices=2	2617791.5 ns	2618584 ns	1.00
kernel/indexing	443583 ns	458000 ns	0.97
kernel/indexing_checked	442916.5 ns	473584 ns	0.94
kernel/launch	9513.833333333332 ns	10069.5 ns	0.94
metal/synchronization/stream	14417 ns	15479.5 ns	0.93
metal/synchronization/context	14833 ns	15084 ns	0.98
shared array/construct	27600.75 ns	26243.083333333336 ns	1.05
shared array/broadcast	448500 ns	473062.5 ns	0.95
shared array/random/randn/Float32	615624.5 ns	806375 ns	0.76
shared array/random/randn!/Float32	459250 ns	662875 ns	0.69
shared array/random/rand!/Int64	370083 ns	555333 ns	0.67
shared array/random/rand!/Float32	412667 ns	589000 ns	0.70
shared array/random/rand/Int64	670812.5 ns	737791.5 ns	0.91
shared array/random/rand/Float32	558104 ns	636875 ns	0.88
shared array/copyto!/gpu_to_gpu	87708 ns	89084 ns	0.98
shared array/copyto!/cpu_to_gpu	87417 ns	84916 ns	1.03
shared array/copyto!/gpu_to_cpu	83375 ns	77708 ns	1.07
shared array/accumulate/1d	1385083.5 ns	1339729 ns	1.03
shared array/accumulate/2d	1386625 ns	1387291 ns	1.00
shared array/iteration/findall/int	1960958 ns	1813291 ns	1.08
shared array/iteration/findall/bool	1680125 ns	1572375 ns	1.07
shared array/iteration/findfirst/int	1405875 ns	1385750 ns	1.01
shared array/iteration/findfirst/bool	1377145.5 ns	1363396 ns	1.01
shared array/iteration/scalar	157167 ns	158084 ns	0.99
shared array/iteration/logical	2981958 ns	2988208 ns	1.00
shared array/iteration/findmin/1d	1450020.5 ns	1467875 ns	0.99
shared array/iteration/findmin/2d	1361562.5 ns	1368229 ns	1.00
shared array/reductions/reduce/1d	724500 ns	731459 ns	0.99
shared array/reductions/reduce/2d	654771 ns	669917 ns	0.98
shared array/reductions/mapreduce/1d	731875 ns	733042 ns	1.00
shared array/reductions/mapreduce/2d	654167 ns	671958 ns	0.97
shared array/permutedims/4d	2332875 ns	2426458 ns	0.96
shared array/permutedims/2d	1014874.5 ns	1015708 ns	1.00
shared array/permutedims/3d	1573062.5 ns	1591625 ns	0.99
shared array/copy	223667 ns	250687 ns	0.89

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