MetalExt: Support Metal 1.10

[jpsamaroo]

Written by Cursor Grok

[github-actions]

Dagger benchmarks: dirty vs master

	master	dirty	master / dirty
array/dagger/N=1024 (block 512)/add (X + X)	3.78 ± 0.51 ms	3.73 ± 0.47 ms	1.01 ± 0.19
array/dagger/N=1024 (block 512)/alloc (rand)	2.74 ± 0.066 ms	2.81 ± 0.13 ms	0.973 ± 0.051
array/dagger/N=1024 (block 512)/broadcast (X .+ 1)	2.54 ± 0.34 ms	2.58 ± 0.29 ms	0.983 ± 0.17
array/dagger/N=1024 (block 512)/map (sin.(X))	5.69 ± 0.28 ms	5.73 ± 0.0049 ms	0.993 ± 0.049
array/dagger/N=1024 (block 512)/norm	1.21 ± 0.26 ms	1.37 ± 0.16 ms	0.885 ± 0.21
array/dagger/N=1024 (block 512)/reduce (sum)	2.24 ± 0.17 ms	2.44 ± 0.43 ms	0.916 ± 0.18
array/dagger/N=1024 (block 512)/transpose (permutedims)	3.96 ± 0.17 ms	4.03 ± 0.048 ms	0.981 ± 0.043
array/dagger/N=256 (block 256)/add (X + X)	1.24 ± 0.11 ms	1.27 ± 0.22 ms	0.977 ± 0.19
array/dagger/N=256 (block 256)/alloc (rand)	1.02 ± 0.0052 ms	1.09 ± 0.029 ms	0.934 ± 0.025
array/dagger/N=256 (block 256)/broadcast (X .+ 1)	0.735 ± 2 ms	0.651 ± 0.21 ms	1.13 ± 3.1
array/dagger/N=256 (block 256)/map (sin.(X))	0.961 ± 0.053 ms	1.01 ± 0.12 ms	0.951 ± 0.13
array/dagger/N=256 (block 256)/norm	0.526 ± 0.067 ms	0.427 ± 0.17 ms	1.23 ± 0.51
array/dagger/N=256 (block 256)/reduce (sum)	1.09 ± 0.046 ms	0.923 ± 0.13 ms	1.18 ± 0.17
array/dagger/N=256 (block 256)/transpose (permutedims)	1.08 ± 0.29 ms	0.872 ± 0.2 ms	1.24 ± 0.44
linalg/dagger/N=1024 (block 512)/cholesky	19.9 ± 1.8 ms	19.4 ± 2.6 ms	1.02 ± 0.17
linalg/dagger/N=1024 (block 512)/lu	0.0326 ± 0.0018 s	0.0344 ± 0.0032 s	0.948 ± 0.1
linalg/dagger/N=1024 (block 512)/matmul (A*A)	0.0355 ± 0.0065 s	27.9 ± 0.58 ms	1.27 ± 0.23
linalg/dagger/N=1024 (block 512)/matvec (A*x)	3.3 ± 0.71 ms	2.99 ± 0.27 ms	1.1 ± 0.26
linalg/dagger/N=1024 (block 512)/qr	0.092 ± 0.0034 s	0.0942 ± 0.011 s	0.976 ± 0.12
linalg/dagger/N=1024 (block 512)/solve (A\b via lu)	0.0411 ± 0.0017 s	0.0426 ± 0.0053 s	0.965 ± 0.13
linalg/dagger/N=1024 (block 512)/syrk (A'*A)	23.2 ± 1.5 ms	22.8 ± 1.4 ms	1.02 ± 0.093
linalg/dagger/N=256 (block 256)/cholesky	2.75 ± 0.21 ms	2.96 ± 0.8 ms	0.931 ± 0.26
linalg/dagger/N=256 (block 256)/lu	4.42 ± 0.51 ms	4.76 ± 0.5 ms	0.929 ± 0.15
linalg/dagger/N=256 (block 256)/matmul (A*A)	2.15 ± 0.25 ms	2.15 ± 0.14 ms	1 ± 0.13
linalg/dagger/N=256 (block 256)/matvec (A*x)	1.41 ± 0.05 ms	1.49 ± 0.29 ms	0.945 ± 0.19
linalg/dagger/N=256 (block 256)/qr	5.07 ± 1.3 ms	5 ± 0.15 ms	1.01 ± 0.27
linalg/dagger/N=256 (block 256)/solve (A\b via lu)	7.53 ± 0.13 ms	7.72 ± 0.32 ms	0.976 ± 0.045
linalg/dagger/N=256 (block 256)/syrk (A'*A)	3.67 ± 0.64 ms	3.41 ± 0.31 ms	1.07 ± 0.21
stencil/dagger/N=1024 (block 512)/alloc (neighbors Wrap)	21.3 ± 0.57 ms	21.3 ± 1.1 ms	1 ± 0.059
stencil/dagger/N=1024 (block 512)/assign (const)	1.63 ± 0.018 ms	1.59 ± 0.035 ms	1.03 ± 0.025
stencil/dagger/N=1024 (block 512)/multi-expr	3.86 ± 0.18 ms	4.01 ± 0.35 ms	0.964 ± 0.095
stencil/dagger/N=1024 (block 512)/neighbors (Clamp)	20.2 ± 0.31 ms	19.5 ± 1 ms	1.04 ± 0.057
stencil/dagger/N=1024 (block 512)/neighbors (Pad)	19.8 ± 0.98 ms	20.6 ± 0.39 ms	0.963 ± 0.051
stencil/dagger/N=1024 (block 512)/neighbors (Reflect)	18.9 ± 0.69 ms	21.2 ± 3.4 ms	0.894 ± 0.15
stencil/dagger/N=1024 (block 512)/neighbors (Wrap)	0.0513 ± 0.0045 s	20.1 ± 0.12 ms	2.55 ± 0.22
stencil/dagger/N=1024 (block 512)/update (+)	2.55 ± 0.58 ms	2.24 ± 0.11 ms	1.14 ± 0.27
stencil/dagger/N=256 (block 256)/alloc (neighbors Wrap)	4.3 ± 0.5 ms	4.21 ± 0.26 ms	1.02 ± 0.14
stencil/dagger/N=256 (block 256)/assign (const)	0.822 ± 0.059 ms	0.654 ± 0.037 ms	1.26 ± 0.11
stencil/dagger/N=256 (block 256)/multi-expr	1.53 ± 0.036 ms	1.54 ± 0.066 ms	0.991 ± 0.049
stencil/dagger/N=256 (block 256)/neighbors (Clamp)	4.13 ± 0.17 ms	4.09 ± 0.046 ms	1.01 ± 0.043
stencil/dagger/N=256 (block 256)/neighbors (Pad)	4.03 ± 0.12 ms	3.83 ± 0.12 ms	1.05 ± 0.044
stencil/dagger/N=256 (block 256)/neighbors (Reflect)	4.44 ± 1.8 ms	4 ± 0.12 ms	1.11 ± 0.44
stencil/dagger/N=256 (block 256)/neighbors (Wrap)	5.63 ± 0.41 ms	4.03 ± 0.035 ms	1.4 ± 0.1
stencil/dagger/N=256 (block 256)/update (+)	0.921 ± 0.091 ms	0.837 ± 0.092 ms	1.1 ± 0.16
time_to_load	0.957 ± 0.0052 s	0.959 ± 0.0053 s	0.998 ± 0.0077

Plots

No regressions beyond 25.0% 🎉

Improvements (> 25.0% faster)

• stencil/dagger/N=1024 (block 512)/neighbors (Wrap): -60.8%
• stencil/dagger/N=256 (block 256)/neighbors (Wrap): -28.5%

Full results and plots (download the benchmark-results artifact).