Vectorized AVX-512 256-bit arithmetic library focused on BN254 field vector operations with Rust bindings. Boasts 1-3x speedups over plain CPU arithmetic.
The library performs elementwise operations on vectors:
This library provides optimized vector operations for BN254 field arithmetic, leveraging AVX-512 instructions. The following operations are supported:
- Vector operations:
add
,sub
,mul
,sum
,inner_product
- Parallel versions:
add_par
,sub_par
,mul_par
,sum_par
,inner_product_par
- In-place operations:
add_inplace
,sub_inplace
,mul_inplace
All operations are suffixed with _bn254
.
These operations use AVX-512 instructions for efficiency.
Note
Benchmarks performed on an AWS c7a.8xlarge instance.
sragss/gpu-sumcheck at 6971f52
Please get in touch with @samrags_ if you'd like specific changes or are curious if this works with your workload.