Build with PGO #548
Description
Building with PGO speeds up loda benchmark by 0%-35%. However, this is quite difficult due to at least 3 profiles (cl, gcc, clang, maybe more because of x86_64/arm64?) and the dependency on the toolchain version. Example of changes here #547
Results from Apple M3, clang 17
Without PGO
| Sequence | Terms | Reg Eval | Inc Eval | Vir Eval |
|---|---|---|---|---|
| A000040 | 1000 | 3.48s | - | 0.43s |
| A000394 | 1000 | 1.72s | - | 0.22s |
| A000401 | 1000 | 2.98s | - | 0.22s |
| A000796 | 300 | 0.70s | - | - |
| A001041 | 300 | 0.73s | - | - |
| A001113 | 300 | 0.63s | - | - |
| A002110 | 300 | 0.77s | - | - |
| A002760 | 200 | 24.71s | - | 1.18s |
| A057552 | 300 | 2.81s | 0.03s | - |
| A079309 | 300 | 2.80s | 0.03s | - |
| A002193 | 400 | 0.56s | 0.23s | - |
| A035856 | 500 | 1.62s | - | - |
| A001609 | 1000 | 0.52s | 0.00s | - |
| A003411 | 1000 | 0.59s | 0.00s | - |
| A012866 | 1000 | 1.00s | 0.00s | - |
| A000045 | 2000 | 1.82s | 0.00s | - |
| A001304 | 3000 | 0.98s | 0.00s | - |
| A000005 | 5000 | 1.04s | - | - |
| A130487 | 5000 | 1.70s | 0.00s | - |
| A000030 | 500000 | 0.38s | - | - |
With PGO(instrumented profile, profile generated from loda mine -H 1)
| Sequence | Terms | Reg Eval | Inc Eval | Vir Eval |
|---|---|---|---|---|
| A000040 | 1000 | 2.22s | - | 0.32s |
| A000394 | 1000 | 1.41s | - | 0.17s |
| A000401 | 1000 | 2.34s | - | 0.17s |
| A000796 | 300 | 0.67s | - | - |
| A001041 | 300 | 0.72s | - | - |
| A001113 | 300 | 0.60s | - | - |
| A002110 | 300 | 0.73s | - | - |
| A002760 | 200 | 19.51s | - | 0.98s |
| A057552 | 300 | 2.79s | 0.03s | - |
| A079309 | 300 | 2.77s | 0.03s | - |
| A002193 | 400 | 0.56s | 0.21s | - |
| A035856 | 500 | 1.55s | - | - |
| A001609 | 1000 | 0.50s | 0.00s | - |
| A003411 | 1000 | 0.58s | 0.00s | - |
| A012866 | 1000 | 0.99s | 0.00s | - |
| A000045 | 2000 | 1.82s | 0.00s | - |
| A001304 | 3000 | 0.84s | 0.00s | - |
| A000005 | 5000 | 0.74s | - | - |
| A130487 | 5000 | 1.12s | 0.00s | - |
| A000030 | 500000 | 0.28s | - | - |