introduces standalone Tile API

[lukastruemper]

Refactors most of the LocalStorage additions/experiments into a standalone API which can be easily tested.

Documentation: https://github.com/daisytuner/docc/tree/tile-algebra/opt#tile-algebra-reasoning-about-memory-levels

[daisytuner]

Daisytuner Report - python_npbench (zinnia)

@@                              Benchmarks                              @@
===========================================================================
  Benchmark              Speedup (Time) ±%           Energy      Regions   
===========================================================================
# adi                    
#   numpy                114.55 ms -1.8%             10.96 J     —         
#   docc-sequential      2.51x (45.64 ms) +0.5%      8.91 J      2         
#   docc-omp             2.06x (55.47 ms) -0.8%      9.31 J      2         
#   docc-cuda            2.46x (46.56 ms) -0.3%      4.51 J      6         
# atax                   
#   numpy                55.91 ms +8.7%              9.87 J      —         
#   docc-sequential      3.22x (17.37 ms) -0.0%      3.62 J      3         
#   docc-omp             2.53x (22.09 ms) +4.3%      4.74 J      3         
#   docc-cuda            3.03x (18.48 ms) -0.1%      1.83 J      7         
# gemm                   
#   numpy                93.54 ms +7.4%              18.68 J     —         
#   docc-sequential      3.71x (25.23 ms) +0.4%      7.20 J      1         
#   docc-omp             3.72x (25.15 ms) +0.6%      7.22 J      1         
#   docc-cuda            0.12x (751.97 ms) -0.0%     71.94 J     1         
# gesummv                
#   numpy                126.51 ms +1.5%             22.01 J     —         
#   docc-sequential      0.71x (178.25 ms) +0.4%     33.27 J     5         
#   docc-omp             1.70x (74.47 ms) -0.6%      15.84 J     5         
#   docc-cuda            10.37x (12.21 ms) -0.2%     1.25 J      11        
# gemver                 
-   numpy                99.43 ms +12.8%             17.11 J     —         
#   docc-sequential      11.61x (8.56 ms) +6.7%      1.77 J      7         
#   docc-omp             7.55x (13.17 ms) -1.7%      2.79 J      7         
#   docc-cuda            7.89x (12.61 ms) -0.6%      1.25 J      11        
# k2mm                   
-   numpy                94.10 ms +17.4%             19.66 J     —         
#   docc-sequential      0.70x (135.32 ms) +0.3%     24.32 J     3         
#   docc-omp             0.69x (136.43 ms) -0.5%     26.35 J     3         
#   docc-cuda            0.10x (940.63 ms) -0.0%     90.14 J     7         
# k3mm                   
#   numpy                73.63 ms -3.0%              18.51 J     —         
#   docc-sequential      0.24x (308.27 ms) -0.1%     50.22 J     4         
#   docc-omp             0.22x (332.53 ms) +0.1%     56.84 J     4         
#   docc-cuda            0.05x (1.53 s) -0.2%        146.83 J    8         
# mvt                    
#   numpy                46.16 ms +2.6%              8.16 J      —         
#   docc-sequential      2.55x (18.13 ms) +0.0%      3.79 J      2         
#   docc-omp             2.54x (18.14 ms) +0.3%      3.83 J      2         
#   docc-cuda            0.13x (361.29 ms) -0.0%     35.47 J     2         
# symm                   
#   numpy                63.83 ms -0.3%              6.15 J      —         
#   docc-sequential      4.29x (14.88 ms) +0.8%      3.00 J      3         
#   docc-omp             3.55x (17.99 ms) +2.0%      3.83 J      3         
# syr2k                  
#   numpy                74.05 ms -1.9%              7.13 J      —         
#   docc-sequential      2.09x (35.50 ms) +1.0%      6.83 J      1         
#   docc-omp             2.21x (33.48 ms) -1.9%      6.19 J      1         
#   docc-cuda            2.41x (30.72 ms) +4.2%      2.97 J      1         
# syrk                   
#   numpy                63.15 ms -0.8%              6.06 J      —         
#   docc-sequential      2.41x (26.18 ms) +0.4%      4.95 J      1         
#   docc-omp             2.40x (26.32 ms) -1.5%      5.01 J      1         
#   docc-cuda            2.74x (23.06 ms) +0.1%      2.21 J      1         
# trmm                   
#   numpy                70.44 ms +0.6%              6.83 J      —         
#   docc-sequential      4.96x (14.21 ms) -1.5%      2.76 J      3         
#   docc-omp             4.46x (15.80 ms) -0.9%      3.44 J      3         

[daisytuner]

Daisytuner Report - pytorch_models (chamomile)

@@                              Benchmarks                              @@
===========================================================================
  Benchmark              Speedup (Time) ±%           Energy      Regions   
===========================================================================
# resnet18               
#   torch                75.87 ms -4.0%              20.13 J     —         
#   torch-cuda           19.75 ms +1.9%              3.82 J      —         
#   docc-omp             0.07x (1.02 s) -3.1%        312.23 J    45        
#   docc-cuda            0.00x (8.07 s) -0.8%        1654.35 J   148       
# segformer              
#   torch-cuda           37.50 ms -0.8%              7.39 J      —         
#   docc-cuda            0.30x (124.94 ms) +0.4%     21.70 J     266       

[daisytuner]

Daisytuner Report - mlir_torch_models (chamomile)

@@                              Benchmarks                              @@
===========================================================================
  Benchmark              Speedup (Time) ±%           Energy      Regions   
===========================================================================
# resnet18               
#   torch                75.69 ms -2.5%              19.42 J     —         
#   torch-cuda           18.34 ms -0.6%              3.47 J      —         
#   docc-omp             0.08x (999.33 ms) -6.6%     300.15 J    53        
#   docc-cuda            0.00x (8.63 s) +2.4%        1679.30 J   159       
# segformer              
#   torch-cuda           38.51 ms +0.7%              7.09 J      —         
#   docc-cuda            0.22x (174.71 ms) -0.0%     29.59 J     402