Use im2row/im2col as expansion of convolution node

[Moehre2]

No description provided.

[daisytuner]

Daisytuner Report - mlir_torch_models (chamomile)

@@                                   Benchmarks                                   @@
=====================================================================================
  Benchmark              Time        ΔTime       Thr         Energy      ΔEnergy     
=====================================================================================
# bn_conv_bn_relu_maxpool_torch18.55 s     +0.43%      N/A         3466.93 J   +0.32%      
+ bn_conv_bn_relu_maxpool_run_none3.27 s      -28.80%     N/A         630.03 J    -22.62%     
+ bn_conv_bn_relu_maxpool_run_sequential3.31 s      -27.89%     N/A         641.00 J    -17.45%     
# bn_conv_bn_relu_maxpool_run_openmp3.41 s      +0.51%      N/A         675.79 J    +1.08%      
# bn_conv_bn_relu_maxpool_run_cuda3.83 s      +4.42%      N/A         718.79 J    +4.12%      

[daisytuner]

Daisytuner Report - python_npbench (zinnia)

@@                                   Benchmarks                                   @@
=====================================================================================
  Benchmark              Time        ΔTime       Thr         Energy      ΔEnergy     
=====================================================================================
# adi_numpy              1.34 s      +1.82%      N/A         134.11 J    +1.74%      
# adi_omp                15.26 s     +0.07%      N/A         1501.86 J   -1.29%      
# adi_cuda               4.71 s      -0.53%      N/A         457.11 J    -0.59%      
# adi_seq_tuning         16.10 s     +0.23%      N/A         1520.52 J   +0.26%      
# atax_numpy             2.15 s      -0.12%      N/A         223.07 J    -0.13%      
# atax_omp               2.98 s      +0.48%      N/A         375.51 J    +0.32%      
# atax_cuda              4.11 s      -0.13%      N/A         423.39 J    -0.73%      
# atax_seq_tuning        3.73 s      -0.43%      N/A         377.46 J    -0.35%      
# gemm_numpy             1.23 s      -0.59%      N/A         198.34 J    -0.54%      
# gemm_omp               1.12 s      +0.20%      N/A         162.68 J    +0.11%      
# gemm_cuda              10.63 s     +0.39%      N/A         1012.41 J   +0.56%      
# gemm_seq_tuning        1.12 s      +0.17%      N/A         162.34 J    +0.31%      
# gesummv_numpy          1.75 s      +0.56%      N/A         250.37 J    +0.70%      
# gesummv_omp            2.02 s      +3.68%      N/A         319.44 J    +4.14%      
# gesummv_cuda           8.34 s      -0.45%      N/A         1002.55 J   -0.28%      
# gesummv_seq_tuning     6.66 s      -0.22%      N/A         813.28 J    -0.39%      
# gemver_numpy           1.08 s      +1.81%      N/A         166.98 J    +1.77%      
# gemver_omp             853.87 ms   -0.08%      N/A         111.19 J    +0.31%      
# gemver_cuda            3.90 s      +0.38%      N/A         390.01 J    -0.07%      
# gemver_seq_tuning      4.49 s      -0.63%      N/A         434.62 J    -0.50%      
# k2mm_numpy             1.20 s      -0.46%      N/A         197.38 J    -0.76%      
# k2mm_omp               3.56 s      +1.08%      N/A         661.74 J    +0.53%      
# k2mm_cuda              13.59 s     -0.03%      N/A         1289.56 J   +0.12%      
# k2mm_seq_tuning        3.64 s      +0.48%      N/A         469.41 J    +0.57%      
# k3mm_numpy             1.03 s      +0.19%      N/A         184.37 J    +0.18%      
# k3mm_omp               5.67 s      +1.11%      N/A         949.38 J    +0.17%      
# k3mm_cuda              19.84 s     -0.06%      N/A         1871.39 J   -0.05%      
# k3mm_seq_tuning        5.73 s      +0.12%      N/A         791.99 J    -0.12%      
# mvt_numpy              2.43 s      +0.04%      N/A         249.30 J    -0.06%      
# mvt_omp                2.75 s      -0.07%      N/A         285.12 J    -0.15%      
# mvt_cuda               3.35 s      +0.02%      N/A         342.36 J    +0.06%      
# mvt_seq_tuning         2.74 s      -0.07%      N/A         284.97 J    -0.02%      
# symm_numpy             780.33 ms   -1.26%      N/A         80.49 J     -1.20%      
# symm_omp               6.20 s      +0.65%      N/A         607.45 J    +0.14%      
# symm_seq_tuning        8.49 s      -0.06%      N/A         809.86 J    +0.02%      
# syr2k_numpy            879.11 ms   -0.92%      N/A         89.59 J     -0.83%      
# syr2k_omp              9.92 s      +0.09%      N/A         943.76 J    +0.13%      
# syr2k_cuda             1.65 s      +0.40%      N/A         171.44 J    +0.56%      
# syr2k_seq_tuning       9.95 s      +0.86%      N/A         946.11 J    +0.79%      
# syrk_numpy             770.50 ms   -0.72%      N/A         79.44 J     -0.86%      
# syrk_omp               5.97 s      -0.35%      N/A         574.58 J    -0.32%      
# syrk_cuda              1.53 s      -0.78%      N/A         159.97 J    -0.78%      
# syrk_seq_tuning        5.96 s      +0.10%      N/A         573.76 J    +0.21%      
# trmm_numpy             874.55 ms   -1.10%      N/A         89.13 J     -0.97%      
# trmm_omp               708.47 ms   -2.14%      N/A         90.95 J     -0.38%      
# trmm_seq_tuning        3.41 s      +1.05%      N/A         325.21 J    +0.67%      

[daisytuner]

Daisytuner Report - mlir_torch_layers (chamomile)

@@                                   Benchmarks                                   @@
=====================================================================================
  Benchmark              Time        ΔTime       Thr         Energy      ΔEnergy     
=====================================================================================
# batchnorm_torch        18.86 s     -0.47%      N/A         3650.38 J   -1.46%      
# batchnorm_run_none     7.71 s      +1.24%      N/A         1473.06 J   +0.10%      
# batchnorm_run_sequential8.59 s      +0.63%      N/A         1634.17 J   -0.82%      
# batchnorm_run_openmp   4.72 s      +0.75%      N/A         1054.13 J   +0.25%      
# batchnorm_run_cuda     12.87 s     +1.29%      N/A         2463.58 J   +0.09%      
# conv2d_torch           18.50 s     -0.24%      N/A         3586.03 J   -1.18%      
# conv2d_run_openmp      4.31 s      -7.57%      N/A         995.61 J    -7.03%      
- conv2d_run_cuda        6.51 s      +32.76%     N/A         1257.76 J   +30.48%     
# linear_torch           6.19 s      +0.36%      N/A         1468.36 J   -0.92%      
# linear_run_none        10.06 s     +0.16%      N/A         2710.41 J   -0.57%      
# linear_run_sequential  8.61 s      +0.65%      N/A         2473.12 J   +0.12%      
# linear_run_openmp      8.12 s      +0.65%      N/A         2332.28 J   +0.06%      
# linear_run_cuda        7.40 s      +0.37%      N/A         1432.40 J   -0.03%      
# matmul_torch           6.24 s      +1.09%      N/A         1486.14 J   -0.13%      
# matmul_run_none        10.10 s     +0.69%      N/A         2727.01 J   +0.35%      
# matmul_run_sequential  8.53 s      -0.04%      N/A         2451.84 J   -0.53%      
# matmul_run_openmp      8.05 s      -1.26%      N/A         2316.29 J   -2.63%      
# matmul_run_cuda        7.31 s      +0.72%      N/A         1407.92 J   -0.68%      
# pooling_torch          25.69 s     +1.55%      N/A         5009.32 J   +0.52%      
# pooling_run_none       17.50 s     +0.86%      N/A         3312.98 J   -0.24%      
# pooling_run_sequential 17.56 s     +1.37%      N/A         3320.70 J   +0.18%      
# pooling_run_openmp     9.95 s      +2.27%      N/A         2017.41 J   +1.08%      
# pooling_run_cuda       23.66 s     +0.52%      N/A         4542.27 J   -0.59%      
# relu_torch             19.11 s     +1.61%      N/A         3700.58 J   +0.49%      
# relu_run_none          4.43 s      +1.21%      N/A         856.02 J    +0.14%      
# relu_run_sequential    4.48 s      +2.14%      N/A         866.97 J    +1.18%      
# relu_run_openmp        3.75 s      +1.22%      N/A         771.21 J    -0.27%      
# relu_run_cuda          6.04 s      +0.75%      N/A         1169.75 J   -0.77%