| 1 |
track history more precisely in documentation for individual source files |
| 2 |
allow shared-* for API functions (requires tweaking dispatch) |
| 3 |
speedups for more architectures |
| 4 |
speedups for more microarchitectures |
| 5 |
consider faster PRF for the keyed hash giving the nonce |
| 6 |
merge subroutines in source to the extent possible |
| 7 |
scan for and remove any unused functions and files |
| 8 |
restructure for more merging at object-code level |
| 9 |
sort object files (for, e.g., improved cache utilization) |
| 10 |
optionally allow post-installation patching of current cpu as an exceptional cpuid |
| 11 |
(based on benchmarks and, with more CPU time, full functionality tests) |
| 12 |
dispatch: eliminate, e.g., avx2 if avx is higher priority |
| 13 |
speed up dispatch cpuid tests (lazy evaluation, merging cpuid calls) |
| 14 |
powbatch: adjust MAXTODO considering time vs. memory usage |
| 15 |
powbatch: use inversion tree, and try parallel mults |
| 16 |
nPbatch: use larger batches to allow faster powbatch |
| 17 |
nPbatch: use 0-padding of leftovers when this is faster than individual nP |
| 18 |
multiscalar: adjust MAXTODO |
| 19 |
multiscalar: vectorize across inputs |
| 20 |
nG etc.: vectorize across points being added |
| 21 |
batch versions of all other operations |
| 22 |
vectorize all other parallelizable operations |
| 23 |
verify constbranch, constindex |
| 24 |
full functional verification |