Code in this repo generates precompile benchmarking data.
All vectors are subfolder of vectors. Each subfolder in turn contains two subfolders:
current- test vectors that have encoded precompile cost as of moment before EIP 2666proposed- test vectors that have encoded precompile cost as of moment after EIP 2666
Filenames are encoded as input_param_scalar_{param}_gas_{gas_value}.csv where param meaning varies between the precompiles and , and gas is an expected gas spent for evaluation of this test vector by the the corresponding precompile. Each csv file contains two columns and no header. First column is hex-encoded input bytestring, second column is hex encoded output bytestring.
param is an input length
param is an input length
param is always equal to 0 (no variety)
param is always equal to 0 (no variety)
param is number of pairs
param is number of rounds
Important: during benchmarking it's suggested NOT to compare output of the precompile to the expected output in a function being benchmarked.
Client developers are free to use any benchmarking harness to get a precompile running time (and expected gas spend). If measurements are performed in a simple loop then at least 1_000 repeats should be performed for each vector before averaging.