You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Analysing the impact of using a large ambient trace (2^20) in the
ClientIVC bench, with no changes to the circuit, one culprit is
initalisation of polynomials defined over the full domain with 0. As
such, I parallelised the initialisation function inside the polynomial
class, which also brings improvement to the Client IVC bench as it is.
Default benchmark
NOW:
```
--------------------------------------------------------------------------------
Benchmark Time CPU
--------------------------------------------------------------------------------
ClientIVCBench/Full/6 29956 ms 28100 ms
```
BEFORE:
```
--------------------------------------------------------------------------------
Benchmark Time CPU
--------------------------------------------------------------------------------
ClientIVCBench/Full/6 32341 ms 30470 ms
```
Benchmark with 2^20 ambient trace
NOW:
```
--------------------------------------------------------------------------------
Benchmark Time CPU
--------------------------------------------------------------------------------
ClientIVCBench/Full/6 39013 ms 36526 ms
```
BEFORE:
```
--------------------------------------------------------------------------------
Benchmark Time CPU
--------------------------------------------------------------------------------
ClientIVCBench/Full/6 44346 ms 41778 ms
```
Note: this is disabled for AVM as they do parallel polynomial
construction and have smaller polynomials.
0 commit comments