adds software pipelining implementation for CDNA

[lukastruemper]

Adds dispatchers for ROCm CDNA path

[daisytuner]

Daisytuner Report - python_npbench (zinnia)

@@                              Benchmarks                              @@
===========================================================================
  Benchmark              Speedup (Time) ±%           Energy      Regions   
===========================================================================
# adi                    
#   numpy                116.21 ms -0.2%             11.52 J     —         
#   docc-sequential      2.57x (45.19 ms) -0.6%      8.95 J      2         
#   docc-omp             2.06x (56.32 ms) +4.3%      9.35 J      2         
#   docc-cuda            2.47x (47.09 ms) -2.8%      4.70 J      6         
# atax                   
-   numpy                58.72 ms +18.1%             11.17 J     —         
#   docc-sequential      3.38x (17.36 ms) +0.1%      3.66 J      3         
#   docc-omp             2.66x (22.09 ms) +2.4%      4.71 J      3         
#   docc-cuda            3.17x (18.55 ms) +0.1%      1.88 J      7         
# gemm                   
#   numpy                93.71 ms +4.9%              18.80 J     —         
#   docc-sequential      3.72x (25.18 ms) +0.1%      7.26 J      1         
#   docc-omp             3.72x (25.21 ms) +0.5%      7.16 J      1         
#   docc-cuda            0.12x (751.98 ms) +0.0%     74.00 J     1         
# gesummv                
#   numpy                128.88 ms +5.6%             22.83 J     —         
#   docc-sequential      0.74x (175.26 ms) +0.1%     32.96 J     5         
#   docc-omp             1.80x (71.62 ms) +1.2%      14.86 J     5         
#   docc-cuda            10.52x (12.24 ms) -0.1%     1.29 J      11        
# gemver                 
#   numpy                87.92 ms -8.6%              16.07 J     —         
#   docc-sequential      10.53x (8.35 ms) +0.7%      1.71 J      7         
#   docc-omp             6.85x (12.84 ms) -9.4%      2.71 J      7         
#   docc-cuda            6.94x (12.68 ms) +0.1%      1.34 J      11        
# k2mm                   
#   numpy                84.84 ms -5.3%              18.36 J     —         
#   docc-sequential      0.63x (135.02 ms) -0.5%     24.55 J     3         
#   docc-omp             0.63x (135.57 ms) -0.0%     26.46 J     3         
#   docc-cuda            0.09x (941.12 ms) -0.0%     92.63 J     7         
# k3mm                   
#   numpy                65.80 ms -3.4%              17.26 J     —         
#   docc-sequential      0.21x (307.94 ms) -0.0%     50.45 J     4         
#   docc-omp             0.20x (331.03 ms) -0.9%     56.73 J     4         
#   docc-cuda            0.04x (1.53 s) -0.1%        150.89 J    8         
# mvt                    
-   numpy                49.62 ms +38.1%             9.11 J      —         
#   docc-sequential      2.74x (18.10 ms) -0.8%      3.85 J      2         
#   docc-omp             2.75x (18.05 ms) -0.1%      3.80 J      2         
#   docc-cuda            0.14x (361.08 ms) +0.0%     36.40 J     2         
# symm                   
#   numpy                63.78 ms +0.1%              6.32 J      —         
#   docc-sequential      4.28x (14.92 ms) +0.8%      2.96 J      3         
#   docc-omp             3.70x (17.23 ms) -4.2%      3.55 J      3         
# syr2k                  
#   numpy                74.47 ms -0.2%              7.41 J      —         
#   docc-sequential      2.08x (35.82 ms) +6.3%      7.00 J      1         
#   docc-omp             2.17x (34.40 ms) -3.7%      6.63 J      1         
#   docc-cuda            2.52x (29.50 ms) +0.0%      2.96 J      1         
# syrk                   
#   numpy                62.64 ms -0.7%              6.22 J      —         
#   docc-sequential      2.42x (25.90 ms) -2.7%      5.07 J      1         
#   docc-omp             2.38x (26.29 ms) -0.4%      5.00 J      1         
#   docc-cuda            2.71x (23.09 ms) +0.5%      2.27 J      1         
# trmm                   
#   numpy                70.25 ms +0.6%              6.97 J      —         
#   docc-sequential      4.89x (14.37 ms) -0.2%      2.82 J      3         
#   docc-omp             4.35x (16.13 ms) -2.5%      3.38 J      3         

[daisytuner]

Daisytuner Report - pytorch_models (chamomile)

@@                              Benchmarks                              @@
===========================================================================
  Benchmark              Speedup (Time) ±%           Energy      Regions   
===========================================================================
# resnet18               
#   torch                77.96 ms +4.5%              20.61 J     —         
#   torch-cuda           18.69 ms -9.2%              3.81 J      —         
#   docc-omp             0.07x (1.04 s) -6.5%        317.88 J    45        
#   docc-cuda            0.00x (8.11 s) -0.4%        1647.81 J   148       
# segformer              
#   torch-cuda           37.08 ms -2.8%              7.35 J      —         
#   docc-cuda            0.30x (124.76 ms) +0.0%     18.04 J     266       

[daisytuner]

Daisytuner Report - mlir_torch_models (chamomile)

@@                              Benchmarks                              @@
===========================================================================
  Benchmark              Speedup (Time) ±%           Energy      Regions   
===========================================================================
# resnet18               
#   torch                80.11 ms +4.0%              19.68 J     —         
#   torch-cuda           18.28 ms +2.6%              3.52 J      —         
#   docc-omp             0.07x (1.11 s) +5.0%        329.20 J    53        
#   docc-cuda            0.00x (8.60 s) -0.7%        1674.22 J   159       
# segformer              
#   torch-cuda           38.86 ms +1.4%              7.35 J      —         
#   docc-cuda            0.22x (174.74 ms) +0.3%     23.20 J     402