[MLIR][TPP] Enable --lower-pack-unpack-without-transpose on LoRA benchmark

[rolfmorel]

Only gets applied to MANUAL MLIR runs as for these the constant weights will remain packed (applying it to OV-derived IR would undo all packing, also on the weights that are passed as an argument).

Additionally, adds more sizes to test on.

[rolfmorel]

We need to wait for libxsmm/tpp-mlir#972 to be merged for this to be ready.

[rolfmorel]

Latest run:

$ python -m xonsh tools/mlir_bench/lora-runner.xsh 
lora_dim = 8
CONFIGS [[8], [16], [32], [64], [128], [256], [512], [1024], [2048], [4096], [8192]]
OV NO-MLIR [0.86, 1.41, 2.5, 4.68, 9.66, 18.86, 37.48, 74.49, 147.02, 310.32, 628.96]
OV MLIR [3.83, 4.37, 5.36, 8.01, 13.45, 23.76, 44.12, 84.9, 168.44, 336.46, 688.82]
NO-OV MLIR [3.93, 4.53, 5.62, 8.24, 13.43, 23.28, 43.59, 82.59, 163.77, 327.99, 659.49]
MANUAL MLIR [0.55, 1.1, 2.2, 4.36, 8.68, 17.41, 34.73, 69.25, 138.28, 279.09, 561.37]
lora_dim = 16
CONFIGS [[8], [16], [32], [64], [128], [256], [512], [1024], [2048], [4096], [8192]]
OV NO-MLIR [0.87, 1.41, 2.5, 4.68, 9.47, 18.93, 37.48, 74.43, 148.11, 310.44, 626.4]
OV MLIR [3.65, 4.39, 5.39, 8.01, 13.49, 23.78, 44.12, 85.64, 167.97, 339.72, 687.83]
NO-OV MLIR [3.97, 4.5, 5.65, 8.31, 13.11, 23.61, 43.56, 83.44, 163.44, 328.23, 669.44]
MANUAL MLIR [0.55, 1.1, 2.21, 4.43, 8.85, 17.78, 34.79, 69.5, 138.83, 280.01, 563.29]
lora_dim = 32
CONFIGS [[8], [16], [32], [64], [128], [256], [512], [1024], [2048], [4096], [8192]]
OV NO-MLIR [0.87, 1.42, 2.53, 4.74, 9.65, 19.2, 37.98, 75.31, 148.93, 313.33, 633.87]
OV MLIR [3.74, 4.3, 5.38, 8.24, 13.7, 23.98, 44.91, 86.28, 169.38, 345.83, 693.66]
NO-OV MLIR [4.14, 4.73, 5.74, 8.26, 13.41, 23.59, 43.98, 85.26, 165.78, 332.2, 668.01]
MANUAL MLIR [0.56, 1.12, 2.23, 4.5, 8.86, 17.93, 35.57, 70.39, 141.22, 283.93, 570.72]
lora_dim = 64
CONFIGS [[8], [16], [32], [64], [128], [256], [512], [1024], [2048], [4096], [8192]]
OV NO-MLIR [0.9, 1.45, 2.59, 4.87, 9.9, 19.9, 39.11, 76.66, 153.12, 322.33, 645.14]
OV MLIR [3.8, 4.57, 5.65, 8.44, 14.09, 24.99, 45.88, 89.65, 175.01, 356.92, 717.46]
NO-OV MLIR [4.27, 4.69, 5.95, 8.64, 13.97, 24.79, 45.6, 86.95, 171.35, 347.62, 689.98]
MANUAL MLIR [0.58, 1.16, 2.31, 4.64, 9.29, 18.64, 36.63, 72.51, 145.54, 292.52, 588.0]
lora_dim = 128
CONFIGS [[8], [16], [32], [64], [128], [256], [512], [1024], [2048], [4096], [8192]]
OV NO-MLIR [0.94, 1.54, 2.75, 5.16, 10.56, 20.95, 41.14, 81.0, 161.38, 339.54, 685.62]
OV MLIR [4.19, 4.89, 5.85, 9.18, 14.9, 26.23, 48.87, 94.78, 185.0, 373.16, 757.09]
NO-OV MLIR [4.48, 4.94, 6.26, 9.06, 14.72, 26.14, 47.85, 92.53, 181.66, 362.51, 734.62]
MANUAL MLIR [0.61, 1.22, 2.45, 4.82, 9.74, 19.72, 38.57, 76.8, 154.17, 309.47, 621.61]