FEAT: parallelize witness assignment of bytecode circuit using assign_regions api

[Velaciela]

Type of change

• Breaking change (fix or feature that would cause existing functionality to not work as expected)

Description

split and parallelize the time-consuming part of bytecode circuit assignment by using assign_regions()

---

uint test:

cargo test --package zkevm-circuits --lib bytecode_circuit::test::bytecode_circuit_parallel_assignment --features scroll,parallel_syn -- --nocapture

Feature: scroll and parallel_syn is required and cannot be executed together with regular unit tests,
This patch adds Run parallel assignment tests into CI workflow

---

super circuict benchmarking in scroll-zkevm:

Before: 21.000s

····Start: assign region: assign bytecode with poseidon hash extension
······Start: assign region 1st pass: assign bytecode with poseidon hash extension
······End: assign region 1st pass: assign bytecode with poseidon hash extension 10.638s
······Start: assign region 2nd pass: assign bytecode with poseidon hash extension
······End: assign region 2nd pass: assign bytecode with poseidon hash extension 10.361s
····End: assign region: assign bytecode with poseidon hash extension ...........21.000s

Now: 0.952s

1st pass: 771.092µs (calculate shape & return)
2nd pass: 0.951350769s (parallel assignment)

Speedup: 22x

Since the multi-phase num is 3, the above assignment will be executed three times,
so this patch reduce the time of witness generation(super circuit) by ~60s (20s x 3)

---

Notes

In super circuit (v0.4), bytecode circuit has 28 witness instances
When using assign_regions() for parallel execution, the load is unbalanced(min: 1, max: 24498)
but further optimization yields less, so here is just a log of profiling result

bytecode.len(): [2801, 17777, 24385, 46, 3943, 4936, 1322, 1822, 3177, 1360, 2164, 3060, 9268, 22143, 22143, 1614, 2152, 22143, 16663, 5333, 1, 16663, 24498, 16663, 1653, 10506, 6379, 12417]

assign bytecode with poseidon hash extension
 CS forked into 28 subCS took 65.971µs
 region assign(regions) bytecode with poseidon hash extension_20 2nd pass synthesis took 16.31µs
 region assign(regions) bytecode with poseidon hash extension_3 2nd pass synthesis took 1.809796ms
 region assign(regions) bytecode with poseidon hash extension_6 2nd pass synthesis took 49.93415ms
 region assign(regions) bytecode with poseidon hash extension_9 2nd pass synthesis took 51.426171ms
 region assign(regions) bytecode with poseidon hash extension_15 2nd pass synthesis took 60.946351ms
 region assign(regions) bytecode with poseidon hash extension_24 2nd pass synthesis took 75.451983ms
 region assign(regions) bytecode with poseidon hash extension_16 2nd pass synthesis took 81.610603ms
 region assign(regions) bytecode with poseidon hash extension_10 2nd pass synthesis took 94.575643ms
 region assign(regions) bytecode with poseidon hash extension_7 2nd pass synthesis took 99.249481ms
 region assign(regions) bytecode with poseidon hash extension_0 2nd pass synthesis took 105.976199ms
 region assign(regions) bytecode with poseidon hash extension_11 2nd pass synthesis took 115.976876ms
 region assign(regions) bytecode with poseidon hash extension_8 2nd pass synthesis took 120.148077ms
 region assign(regions) bytecode with poseidon hash extension_4 2nd pass synthesis took 149.78719ms
 region assign(regions) bytecode with poseidon hash extension_5 2nd pass synthesis took 186.413146ms
 region assign(regions) bytecode with poseidon hash extension_19 2nd pass synthesis took 201.365304ms
 region assign(regions) bytecode with poseidon hash extension_26 2nd pass synthesis took 288.769832ms
 region assign(regions) bytecode with poseidon hash extension_12 2nd pass synthesis took 350.249662ms
 region assign(regions) bytecode with poseidon hash extension_25 2nd pass synthesis took 397.082566ms
 region assign(regions) bytecode with poseidon hash extension_27 2nd pass synthesis took 502.59244ms
 region assign(regions) bytecode with poseidon hash extension_21 2nd pass synthesis took 630.980547ms
 region assign(regions) bytecode with poseidon hash extension_18 2nd pass synthesis took 631.629386ms
 region assign(regions) bytecode with poseidon hash extension_23 2nd pass synthesis took 631.664026ms
 region assign(regions) bytecode with poseidon hash extension_1 2nd pass synthesis took 722.695257ms
 region assign(regions) bytecode with poseidon hash extension_13 2nd pass synthesis took 839.83569ms
 region assign(regions) bytecode with poseidon hash extension_17 2nd pass synthesis took 840.486889ms
 region assign(regions) bytecode with poseidon hash extension_14 2nd pass synthesis took 841.069388ms
 region assign(regions) bytecode with poseidon hash extension_2 2nd pass synthesis took 926.174553ms
 region assign(regions) bytecode with poseidon hash extension_22 2nd pass synthesis took 930.444835ms
 Merge 28 subCS back took 6.24µs
 28 sub_regions of assign(regions) bytecode with poseidon hash extension 2nd pass synthesis took 932.071649ms

---

[Velaciela]

we need to merge scroll-tech/halo2#50 before we can pass all unit tests