[WIP]Optimization Bench

[ashWhiteHat]

Abstract

This branch checks only bench_evm_circuit_prover and how the optimization improves performance.

Performance

Degree: 19

Step	Default	Assembly	Rebase	Assembly And Rebase	Reduction Rate
Setup	318.167s	258.626s	317.871s	258.844s	18.64 %
Prove	5817.462s	5745.782s	1680.817s	1627.667s	72.02%
Verify	114.946ms	119.416ms	111.262ms	109.344ms	4.87%
Total	6156.35s	6022.38s	2129.42s	1904.50s	69.06%

Resource	Default	Assembly	Rebase	Assembly And Rebase	Reduction Rate
CPU	66.8%	33.1%	100.0%	100.0%	+49.70%
Memory	508.6Gb	442.8Gb	464.9Gb	464.1Gb	8.59%

Improvement

• Improve fft

Done

• Introduce flame graph (patch @lispc and Ho code)
• Introduce pprof (patch @lispc and Ho code)
• Introduce assembly
• Rebase zcash halo2 latest branch (actually not latest)

Note

I would like to get pprof file.

[barryWhiteHat]

Hmm seems like the acceleration from intel assembly seems to be so low. @AronisAt79 what CPU does the test server have ? Does virtualization block us from using cpu instruction based optimizations ?

[AronisAt79]

@barryWhiteHat i cant think of why virtualization might block optimizations. Here are the cpu details. We can repeat the specific test on other available architectures for similar instance size, if this would help.

Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Byte Order: Little Endian
Address sizes: 46 bits physical, 48 bits virtual
CPU(s): 96
On-line CPU(s) list: 0-95
Thread(s) per core: 2
Core(s) per socket: 24
Socket(s): 2
NUMA node(s): 2
Vendor ID: GenuineIntel
CPU family: 6
Model: 85
Model name: Intel(R) Xeon(R) Platinum 8259CL CPU @ 2.50GHz
Stepping: 7
CPU MHz: 2499.996
BogoMIPS: 4999.99
Hypervisor vendor: KVM
Virtualization type: full
L1d cache: 1.5 MiB
L1i cache: 1.5 MiB
L2 cache: 48 MiB
L3 cache: 71.5 MiB
NUMA node0 CPU(s): 0-23,48-71
NUMA node1 CPU(s): 24-47,72-95

[ashWhiteHat]

No worry.
I optimized and benched it again.
It reduces about 20% arithmetic time.
privacy-scaling-explorations/pairing#5 (comment)

I think if it's applied, the halo2 will be faster as well.