PR updates the pt LUT reading in disk/overlap TP, using ternary operators instead of a switch. This brings timing for overlap TPs more in line with barrel TPs.
HLS timing estimate for L1D1C: 4.596ns -> 3.465ns.
Edit:
Added an odd hack in L3L4 TP, to reduce LUT usage. stubptouter LUT is read in a similar manner to the overlap/disks, which greatly reduces LUTs for some reason.
Edit 2:
removed the hack and replaced with simple fix to the pt lut sizes - produces even better results in terms of resources and timing
table of resource/timing in affected modules post implementation:
Current:
PR updates the pt LUT reading in disk/overlap TP, using ternary operators instead of a switch. This brings timing for overlap TPs more in line with barrel TPs.
HLS timing estimate for L1D1C: 4.596ns -> 3.465ns.
Edit: Added an odd hack in L3L4 TP, to reduce LUT usage. stubptouter LUT is read in a similar manner to the overlap/disks, which greatly reduces LUTs for some reason.
Edit 2: removed the hack and replaced with simple fix to the pt lut sizes - produces even better results in terms of resources and timing