gcolvin / evm-drag-race

EVM performance tests for a range of clients

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

This directory contains a small collection of EVM assembly and Solidity performance tests originally written for tuning the cpp-ethereum interpreter and for presentation at DevCon3. They are out of date with the current clients, which I'm cleaning up as a low-priority task, but I'm happy to help get them working for your needs if you ask me.

The .asm tests are meant to isolate individual opcodes. The .sol tests are meant to exercise larger units, like kernels for random number generation and message encryption. For many of the tests a corresponding C or C++ version is provided as well.

Running the tests can be handled indiviually at the command line, or with the rather curious tests.mk.

make -f tests.mk [ETHVM=ethvm] [EVM=evm] [PARITY=parity-evm] \
                [all | ops | programs | mul64 | <test>.bin | <test>.ran]

Runs only the programs for which a path is provided on the command line to make the given targets. There is further documentation in tests.mk.

We also provide a few python scripts to help make sense of the output.

log2csv.py

Converts a log file from tests.mk output from one or more clients into a .csv file suitable for use by spreadsheets.

csv2nano.py

Converts a .csv file from log2csv into another .csv file with the times scaled to nanoseconds per test.

csv2gas.py

Converts a .csv file from log2csv into another .csv file with the times scaled to nanoseconds per unit of gas.

csv2ops.py

Converts a .csv file from log2csv into another .csv file with the times scaled to nanoseconds per individual operation, with the estimated overhead from the interpreter and the test harness removed.

An example of a complete run follows.

# compile, assemble, and run all the tests and log their output
$ make -f tests.mk ETHVM=ethvm EVM=evm PARITY=parity-evm all >& times.log

# scrape log to a CSV file of gas and seconds per test run for each client
$ python log2csv.py < times.log > times.csv
$ cat times.csv
(sec/run), gas, ethvm, evm, parity-evm
nop, 378535997, 4.88, 19.41, 14.62
pop, 781189181, 5.38, 23.92, 19.85
add64, 915406909, 7.98, 35.11, 21.49
add128, 915406909, 10.12, 32.64, 21.58
add256, 915406909, 11.20, 34.68, 21.79
sub64, 915406909, 8.24, 32.96, 21.44
sub128, 915406909, 9.76, 33.78, 21.50
sub256, 915406909, 10.31, 35.62, 21.95
mul64, 1183842365, 8.40, 33.12, 24.64
mul128, 1183842365, 8.96, 33.83, 26.32
mul256, 1183842365, 15.16, 69.21, 50.19
div64, 1183842365, 15.12, 38.88, 122.54
div128, 1183842365, 25.28, 46.76, 191.37
div256, 1183842365, 39.14, 108.77, 350.33
exp, 1692213309, 34.64, 153.59, 73.22
loop, 53497491, 0.90, 1.94, 1.24
fun, 1060110591, 13.38, 36.19, 29.22
rc5, 2304011914, 27.55, 99.50, 52.16
mix, 1074334585, 14.32, 53.56, 35.23
rng, 2081599617, 27.39, 106.10, 50.99

# calculate nanoseconds per test
$ python csv2nano.py < times.csv
(ns/test), gas, ethvm, evm, parity-evm
nop, 3, 36, 145, 109
pop, 6, 40, 178, 148
add64, 7, 59, 262, 160
add128, 7, 75, 243, 161
add256, 7, 83, 258, 162
sub64, 7, 61, 246, 160
sub128, 7, 73, 252, 160
sub256, 7, 77, 265, 164
mul64, 9, 63, 247, 184
mul128, 9, 67, 252, 196
mul256, 9, 113, 516, 374
div64, 9, 113, 290, 913
div128, 9, 188, 348, 1426
div256, 9, 292, 810, 2610
exp, 1614, 33035, 146475, 69828
loop, 51, 858, 1850, 1183
fun, 1011, 12760, 34513, 27866
rc5, 2197, 26274, 94891, 49744
mix, 1025, 13657, 51079, 33598
rng, 1985, 26121, 101185, 48628

# calculate nanoseconds per unit of gas
$ python csv2gas.py < times.csv
(ns/gas), ethvm, evm, parity-evm
nop, 13, 51, 39
pop, 7, 31, 25
add64, 9, 38, 23
add128, 11, 36, 24
add256, 12, 38, 24
sub64, 9, 36, 23
sub128, 11, 37, 23
sub256, 11, 39, 24
mul64, 7, 28, 21
mul128, 8, 29, 22
mul256, 13, 58, 42
div64, 13, 33, 104
div128, 21, 39, 162
div256, 33, 92, 296
exp, 20, 91, 43
loop, 17, 36, 23
fun, 13, 34, 28
rc5, 12, 43, 23
mix, 13, 50, 33
rng, 13, 51, 24

# estimate nanoseconds per opcode
$ python csv2ops.py < times.csv
(ns/OP), ethvm, evm, parity-evm
add64, 21, 101, 32
add128, 37, 82, 33
add256, 45, 97, 34
sub64, 23, 85, 32
sub128, 35, 91, 32
sub256, 39, 104, 36
mul64, 25, 86, 56
mul128, 29, 91, 68
mul256, 75, 355, 246
div64, 75, 129, 785
div128, 150, 187, 1298
div256, 254, 649, 2482

About

EVM performance tests for a range of clients


Languages

Language:C++ 90.0%Language:Assembly 6.2%Language:C 2.2%Language:Python 1.0%Language:Makefile 0.7%