datadeps: Make some Array allocations eagerly freeable

[jpsamaroo]

Allows freeing Array allocations within Datadeps when no longer used, by allocating temporary slots from Libc.malloc (which we can then Libc.free). Should greatly reduce memory usage for CPU multi-processor execution.

Written by Claude

[github-actions]

Dagger benchmarks: dirty vs master

	master	dirty	master / dirty
array/dagger/N=1024 (block 512)/add (X + X)	4.21 ± 0.09 ms	4.3 ± 0.15 ms	0.979 ± 0.041
array/dagger/N=1024 (block 512)/alloc (rand)	3.42 ± 0.24 ms	3.73 ± 0.092 ms	0.918 ± 0.068
array/dagger/N=1024 (block 512)/broadcast (X .+ 1)	2.83 ± 0.063 ms	1.97 ± 0.17 ms	1.44 ± 0.13
array/dagger/N=1024 (block 512)/map (sin.(X))	7.23 ± 0.023 ms	5.44 ± 1.7 ms	1.33 ± 0.41
array/dagger/N=1024 (block 512)/norm	1.2 ± 0.071 ms	1.31 ± 0.025 ms	0.916 ± 0.057
array/dagger/N=1024 (block 512)/reduce (sum)	2.61 ± 0.28 ms	2.52 ± 1.3 ms	1.04 ± 0.54
array/dagger/N=1024 (block 512)/transpose (permutedims)	6.74 ± 1.4 ms	7.22 ± 1.4 ms	0.934 ± 0.26
array/dagger/N=256 (block 256)/add (X + X)	1.31 ± 0.36 ms	1.38 ± 0.12 ms	0.946 ± 0.27
array/dagger/N=256 (block 256)/alloc (rand)	1.14 ± 0.11 ms	1.08 ± 0.036 ms	1.05 ± 0.11
array/dagger/N=256 (block 256)/broadcast (X .+ 1)	0.558 ± 0.17 ms	0.655 ± 0.25 ms	0.852 ± 0.41
array/dagger/N=256 (block 256)/map (sin.(X))	1.11 ± 0.11 ms	1.27 ± 0.021 ms	0.877 ± 0.088
array/dagger/N=256 (block 256)/norm	0.537 ± 0.012 ms	0.527 ± 0.039 ms	1.02 ± 0.079
array/dagger/N=256 (block 256)/reduce (sum)	1.05 ± 0.21 ms	1.16 ± 0.27 ms	0.906 ± 0.27
array/dagger/N=256 (block 256)/transpose (permutedims)	1.14 ± 0.12 ms	1.04 ± 0.49 ms	1.1 ± 0.53
linalg/dagger/N=1024 (block 512)/cholesky	22.4 ± 4.6 ms	19.5 ± 1.7 ms	1.15 ± 0.26
linalg/dagger/N=1024 (block 512)/lu	0.0456 ± 0.0036 s	0.0432 ± 0.0046 s	1.06 ± 0.14
linalg/dagger/N=1024 (block 512)/matmul (A*A)	0.0517 ± 0.0084 s	0.047 ± 0.0023 s	1.1 ± 0.19
linalg/dagger/N=1024 (block 512)/matvec (A*x)	3.13 ± 0.16 ms	3.02 ± 0.13 ms	1.04 ± 0.068
linalg/dagger/N=1024 (block 512)/qr	0.116 ± 0.0044 s	0.118 ± 0.0031 s	0.981 ± 0.045
linalg/dagger/N=1024 (block 512)/solve (A\b via lu)	0.0517 ± 0.0031 s	0.0534 ± 0.0035 s	0.968 ± 0.086
linalg/dagger/N=1024 (block 512)/syrk (A'*A)	0.0359 ± 0.0017 s	0.0357 ± 0.0055 s	1.01 ± 0.16
linalg/dagger/N=256 (block 256)/cholesky	2.73 ± 0.27 ms	3.23 ± 0.68 ms	0.845 ± 0.2
linalg/dagger/N=256 (block 256)/lu	4.52 ± 0.19 ms	4.53 ± 2.1 ms	0.998 ± 0.47
linalg/dagger/N=256 (block 256)/matmul (A*A)	2.74 ± 0.62 ms	2.19 ± 0.4 ms	1.25 ± 0.36
linalg/dagger/N=256 (block 256)/matvec (A*x)	1.37 ± 0.067 ms	1.66 ± 0.26 ms	0.824 ± 0.13
linalg/dagger/N=256 (block 256)/qr	5.31 ± 0.35 ms	5.62 ± 0.79 ms	0.945 ± 0.15
linalg/dagger/N=256 (block 256)/solve (A\b via lu)	8.43 ± 0.086 ms	8.89 ± 0.46 ms	0.948 ± 0.05
linalg/dagger/N=256 (block 256)/syrk (A'*A)	3.28 ± 0.59 ms	3.84 ± 0.65 ms	0.854 ± 0.21
stencil/dagger/N=1024 (block 512)/alloc (neighbors Wrap)	26.7 ± 0.3 ms	26 ± 3 ms	1.02 ± 0.12
stencil/dagger/N=1024 (block 512)/assign (const)	1.44 ± 0.029 ms	1.65 ± 0.17 ms	0.876 ± 0.091
stencil/dagger/N=1024 (block 512)/multi-expr	3.87 ± 0.13 ms	4 ± 0.046 ms	0.966 ± 0.034
stencil/dagger/N=1024 (block 512)/neighbors (Clamp)	23.9 ± 1.5 ms	25.4 ± 1.3 ms	0.94 ± 0.077
stencil/dagger/N=1024 (block 512)/neighbors (Pad)	24.2 ± 2.4 ms	23.9 ± 0.7 ms	1.01 ± 0.11
stencil/dagger/N=1024 (block 512)/neighbors (Reflect)	24.8 ± 0.21 ms	25 ± 1.4 ms	0.991 ± 0.055
stencil/dagger/N=1024 (block 512)/neighbors (Wrap)	24.8 ± 0.53 ms	26 ± 6.6 ms	0.954 ± 0.24
stencil/dagger/N=1024 (block 512)/update (+)	2.93 ± 0.26 ms	3.42 ± 0.36 ms	0.857 ± 0.12
stencil/dagger/N=256 (block 256)/alloc (neighbors Wrap)	4.95 ± 0.13 ms	5.12 ± 0.14 ms	0.966 ± 0.037
stencil/dagger/N=256 (block 256)/assign (const)	0.72 ± 0.096 ms	0.785 ± 0.052 ms	0.917 ± 0.14
stencil/dagger/N=256 (block 256)/multi-expr	1.5 ± 0.2 ms	1.53 ± 0.029 ms	0.983 ± 0.13
stencil/dagger/N=256 (block 256)/neighbors (Clamp)	4.81 ± 0.21 ms	4.65 ± 0.5 ms	1.03 ± 0.12
stencil/dagger/N=256 (block 256)/neighbors (Pad)	4.71 ± 0.36 ms	4.97 ± 0.95 ms	0.947 ± 0.2
stencil/dagger/N=256 (block 256)/neighbors (Reflect)	4.91 ± 0.17 ms	4.74 ± 0.18 ms	1.04 ± 0.053
stencil/dagger/N=256 (block 256)/neighbors (Wrap)	4.81 ± 0.35 ms	4.97 ± 0.71 ms	0.967 ± 0.15
stencil/dagger/N=256 (block 256)/update (+)	0.884 ± 0.0071 ms	0.865 ± 0.025 ms	1.02 ± 0.031
time_to_load	1.04 ± 0.011 s	1.09 ± 0.04 s	0.957 ± 0.037

Plots

No regressions beyond 25.0% 🎉

Improvements (> 25.0% faster)

• array/dagger/N=1024 (block 512)/broadcast (X .+ 1): -30.5%

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