Skip to content

Cuda implementation of EVM bytecode executor

License

Notifications You must be signed in to change notification settings

PreResearch-Labs/CuEVM-cuda-evm

 
 

Repository files navigation

CuEVM

Cuda implementation of EVM bytecode executor

Prerequisites

  • CUDA Toolkit (Version 12.0+, because we use --std c++20)
  • A CUDA-capable GPU (CUDA compute capabilily 7+ other older GPUs compability are not tested fully)
  • A C++ compiler compatible with the CUDA Toolkit (gcc/g++ version 10+)
  • For docker image, you don't need the above but the system with docker installed

Compile and Build binary

There are two methods, one requires installing all prequisited in the system, the second one use docker image:

On your own system

Building on Ubuntu (with sudo):

  • Setup required libraries: sudo apt install libgmp-dev
  • Setup cJSON: sudo apt install libcjson-dev
  • Use cmake to build the binary (Adjust -DCUDA_COMPUTE_CAPABILITY=86 according to your GPU compute capability number):
cmake -S . -B build -DTESTS=OFF -DGPU=ON -DCPU=OFF \
    -DCUDA_COMPUTE_CAPABILITY=86
    -DENABLE_EIP_3155_OPTIONAL=OFF \
    -DENABLE_EIP_3155=ON \
    -DENABLE_PAIRING_CODE=ON

cmake --build build

Building without sudo is also possible with extra configuration and modification on the Makefile or evironment variables, please refer to online tutorials on how to build and use libraries in custom directory

Building using docker image

  • Pull the docker image first: docker pull augustus/goevmlab-cuevm:20241008
  • Run and mount the current code folder docker run -it -v $(pwd):/workspaces/CuEVM augustus/goevmlab-cuevm:20241008
  • Inside the docker container, you can build the code using the same commands as above (Adjust -DCUDA_COMPUTE_CAPABILITY=86 according to your GPU compute capability number):
cmake -S . -B build -DTESTS=OFF -DGPU=ON -DCPU=OFF \
    -DCUDA_COMPUTE_CAPABILITY=86 \
    -DENABLE_EIP_3155_OPTIONAL=OFF \
    -DENABLE_EIP_3155=ON \
    -DENABLE_PAIRING_CODE=ON

cmake --build build

Usage

After building the binary, a binary named cuevm_GPU will be created in the build folder. You can run the binary using the following command:

./build/cuevmGPU --input [input_json_file]

For example:

./build/cuevm_GPU --input samples/underflowTest.json

The execution trace and output state will be printed to the stdout, you can use jq to keep only the Json output:

./build/cuevm_GPU --input samples/underflowTest.json \
    | jq -R 'select(try fromjson? | . != null) | fromjson'
{
  "pc": 0,
  "op": 97,
  "gas": "0x79bf20",
  "gasCost": "0x3",
  "memSize": 0,
  "stack": [],
  "depth": 1,
  "refund": 0
}
...

Demo of the GPU-accelerated fuzzer and CuEVM library mode

Run Google Colab demo using free GPU

Testing using ethtest

The script scripts/run-ethtest-by-fork can be used to run the tests from the ethereum/tests. It compares the traces from the outputs of CuEVM and geth without stateRoot.

Requirements:

The following will run all the tests in ethereum/tests/GeneralStateTests, note that this may take a few hours:

git clone --depth=1 --branch shanghai https://github.com/ethereum/tests.git ethereum/tests
python scripts/run-ethtest-by-fork.py --without-state-root --ignore-errors --microtests \
  --input ethereum/tests/GeneralStateTests/ \
  -t /tmp/ \
  --runtest-bin runtest \
  --geth geth \
  --cuevm build/cuevm_GPU

To run a single json test case in ethereum/tests, create a new folder and copy the json file to the new folder, for example,

mkdir my-test-folder
cp ethereum/tests/GeneralStateTests/Shanghai/stEIP3651-warmcoinbase/coinbaseWarmAccountCallGas.json my-test-folder/
python scripts/run-ethtest-by-fork.py --without-state-root --ignore-errors --microtests \
  --input my-test-folder/ \
  -t /tmp/ \
  --runtest-bin runtest \
  --geth geth \
  --cuevm build/cuevm_GPU

To inpsect the output from each EVM exeuctor, specify the path of the generated test file:

  • for go-ethereum:
geth --json --noreturndata --dump statetest {input_test_file}
# eg., geth --json --noreturndata --dump statetest /tmp/./coinbaseWarmAccountCallGas-0-0-0/coinbaseWarmAccountCallGas.json
  • for CuEVM
./build/cuevm_GPU --input {input_test_file}
# eg., ./build/cuevm_GPU --input /tmp/./coinbaseWarmAccountCallGas-0-0-0/coinbaseWarmAccountCallGas.json
  • to compare results between CuEVM and go-ethereum:
runtest --outdir=./ --geth=geth --cuevm=./build/cuevm_GPU /tmp/./coinbaseWarmAccountCallGas-0-0-0/coinbaseWarmAccountCallGas.json

Test results

Test results by comparing the traces between geth and cuevm without stateRoot comparison

We use the test files in Shanghai branch of ethereum/tests/GeneralStateTests to test whether we can get the same traces with the go-ethereum. To run the tests,

  • Build the cuevm_GPU binary
  • Either download or build the geth binary, in this test the version v1.13.14 is used, later version will probably produce the same result
  • Get the test json files from ethereum/tests: git clone --depth=1 --branch shanghai https://github.com/ethereum/tests.git ethereum/tests
  • Build the runtest binary from cassc/goevmlab which adds support for CuEVM
  • Run goevmlab/runtest to compare the results between geth and cuevm
  • The tests results are collected by running the Python script from the ethereum/tests root folder:
python ./run-ethtest-without-stateroot-comparison.py -t /tmp/out --runtest-bin runtest --geth geth --cuevm cuevm  --ignore-errors
  • Note that there can be multiple tests in one input json, the number of tests shown below can be larger than number of input json files.
  • Some tests are skipped if they do not support Shanghai VM.
Test folder Passed Failed Skipped Timeout
stNonZeroCallsTest 24 0 0 0
stEIP3607 12 0 0 0
stEIP150singleCodeGasPrices 340 0 1 0
stCallDelegateCodesCallCodeHomestead 58 0 0 0
stArgsZeroOneBalance 96 0 0 0
stStaticFlagEnabled 25 9 0 0
stShift 40 0 0 2
stEIP158Specific 6 1 0 0
stMemoryTest 569 0 0 9
stZeroKnowledge2 519 0 0 0
stEIP1559 1824 19 0 2
stReturnDataTest 223 50 0 0
stCodeCopyTest 2 0 0 0
stMemoryStressTest 80 2 0 0
stInitCodeTest 22 0 0 0
stMemExpandingEIP150Calls 10 0 0 0
stWalletTest 20 26 0 0
stSpecialTest 18 2 0 2
stExtCodeHash 64 1 0 0
stTimeConsuming 5187 0 0 3
stCreateTest 166 34 0 3
stRecursiveCreate 1 0 0 1
stCallDelegateCodesHomestead 58 0 0 0
stZeroKnowledge 751 49 0 0
stTransitionTest 6 0 0 0
stCallCodes 87 0 0 0
stHomesteadSpecific 5 0 0 0
stCallCreateCallCodeTest 40 9 0 6
stSolidityTest 13 8 0 2
stExample 33 6 0 0
stSStoreTest 468 7 0 0
stZeroCallsTest 24 0 0 0
stSelfBalance 32 9 0 1
stDelegatecallTestHomestead 23 3 0 5
stQuadraticComplexityTest 13 1 0 18
stEIP150Specific 25 0 0 0
stStackTests 247 118 0 10
stChainId 2 0 0 0
stAttackTest 0 1 0 1
stBugs 9 0 0 0
stBadOpcode 3651 448 1 116
stTransactionTest 164 0 0 0
stCreate2 183 3 0 4
stPreCompiledContracts2 248 0 0 0
stRevertTest 259 5 0 7
stLogTests 46 0 0 0
stRandom 303 4 0 7
stRefundTest 26 0 1 0
stStaticCall 415 5 0 58
stRandom2 214 7 0 5
Shanghai 25 1 0 1
stCodeSizeLimit 7 0 0 0
stZeroCallsRevert 16 0 0 0
stPreCompiledContracts 791 169 0 0
stSystemOperationsTest 79 1 0 3
stEIP2930 15 125 0 0
VMTests 628 0 3 0
stSLoadTest 1 0 0 0

Citation

If you use our code in your research, please kindly cite:

To be updated.

Contributors

minhhn2910 Nhut-Minh Ho sdcioc Stefan-Dan Ciocirlan cassc Chen Li

This project is part of the Singapore Blockchain Innovation Programme (SBIP). We extend our gratitude to the programme and its team members for their expertise and dedication to the development of this project.

Documentation

An auto generated source code documentation is available at https://sbip-sg.github.io/CuEVM-internal/files.html

About

Cuda implementation of EVM bytecode executor

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages

  • Cuda 96.8%
  • Python 2.0%
  • CMake 1.2%