Audio: Crossover / Multiband DRC: Change to HiFi4 and HiFi5 optimized IIR DF1 core type

[singalsu]

No description provided.

[singalsu]

Note: The optimization will continue with the IIR DF1 core. I'm thinking to add a conversion for the coefficients blob to 128 bits load compatible to make it more efficient. The blobs in user space would remain the same.

[johnylin76]

LGTM, thanks

[singalsu]

Note: The optimization will continue with the IIR DF1 core. I'm thinking to add a conversion for the coefficients blob to 128 bits load compatible to make it more efficient. The blobs in user space would remain the same.

Here's the top 10 results from profiling before/after. The IIR still remains the top function:

Pipeline MCPS:  96.50

Flat profile:
                                           self      total          
       cumulative       self             cycles     cycles          
  %        cycles     cycles    calls     /call      /call  name    
             (K)        (K)                (K)        (K)           
39.01     54861.72   54861.72  1096720      0.05       0.05  iir_df2t
10.12     69090.62   14228.90     1430      9.95      94.57  multiband_drc_s32_default
 7.51     79653.12   10562.50     6426      1.64       2.71  drc_update_detector_average
 7.42     90084.45   10431.32     6429      1.62       2.65  drc_compress_output
 7.28    100327.70   10243.26   205635      0.05       0.24  multiband_drc_s32_process_drc
 5.41    107936.20    7608.49    68545      0.11       0.91  multiband_drc_process_emp_crossover
 4.97    114927.79    6991.59   137090      0.05       0.35  crossover_generic_split_3way
 4.68    121511.08    6583.30   205728      0.03       0.03  sofm_lut_sin_fixed_16b
 3.56    126514.62    5003.54    35834      0.14       0.14  sofm_exp_int32
 2.24    129670.89    3156.27   139259      0.02       0.02  memcpy

Pipeline MCPS:  90.36

Flat profile:
                                           self      total          
       cumulative       self             cycles     cycles          
  %        cycles     cycles    calls     /call      /call  name    
             (K)        (K)                (K)        (K)           
34.95     46087.96   46087.96  1096720      0.04       0.04  iir_df1
10.79     60316.86   14228.90     1430      9.95      88.44  multiband_drc_s32_default
 8.01     70879.36   10562.50     6426      1.64       2.71  drc_update_detector_average
 7.91     81310.68   10431.32     6429      1.62       2.65  drc_compress_output
 7.77     91553.94   10243.26   205635      0.05       0.24  multiband_drc_s32_process_drc
 5.77     99162.44    7608.49    68545      0.11       0.80  multiband_drc_process_emp_crossover
 5.30    106154.03    6991.59   137090      0.05       0.30  crossover_generic_split_3way
 4.99    112737.32    6583.30   205728      0.03       0.03  sofm_lut_sin_fixed_16b
 3.79    117740.86    5003.54    35834      0.14       0.14  sofm_exp_int32
 2.39    120897.13    3156.27   139259      0.02       0.02  memcpy

This was done with script run scripts/sof-testbench-helper.sh -x -m drc_multiband -p profile-drc32_multiband.txt

Edit: Next step is simplified IIR core for crossover_generic_process_lr4() function, the checks and outer loop can be removed for a fixed 4th order (2 biquads) calculate.