nilfoundation / crypto3-multiprecision

Cryptography-Enhanced Multiprecision for =nil; Crypto3 C++ Cryptography Suite.

=nil; Foundation's Multiprecision Library

The Multiprecision Library provides optimal arithmetic operations over a finite field, using Montgomery and Barret reductions.

Support, bugs and feature requests

Bugs and feature requests can be reported through the Gitub issue tracker.

You can submit your changes through a pull request.

Development

Clone the module repository project:

git clone https://github.com/nilfoundation/crypto3-multiprecision
cd crypto3-multiprecision
git submodule update --init
mkdir build && cmake ..

Running tests

Dependencies

External

• Boost (>= 1.73). Because boost::config doesn't have BOOST_IF_CONSTEXPR definition before 1.73 version.

crypto3-multiprecision's Issues

Availability BOOST_MP_IS_CONST_EVALUATED check

Adjust constexpr_test_cpp_int* tests to be used with modular_cpp_int

Related to this comment.
For now we have multiple tests disabled after modular_cpp_int refactoring. Adjust them to test our modular int instead of upstream cpp_int and enable so they become included into CI

Add a new backend to multiprecision to be used exclusively with modular_adaptor, remove all code that belongs to boost.

Implement Joris van der Hoeven's multiplication

https://hal.archives-ouvertes.fr/hal-02070778/document

Implement modular adaptor inverse algorithms

Make function window_bits inline, so it will not create linking issues.

Generalize modular adaptor internals

Modular adaptor currently uses backend-dependent member functions https://github.com/NilFoundation/crypto3-multiprecision/blob/master/include/boost/multiprecision/modular/montgomery_params.hpp#L137, https://github.com/NilFoundation/crypto3-multiprecision/blob/master/include/boost/multiprecision/modular/montgomery_params.hpp#L142, https://github.com/NilFoundation/crypto3-multiprecision/blob/master/include/boost/multiprecision/modular/montgomery_params.hpp#L150, https://github.com/NilFoundation/crypto3-multiprecision/blob/master/include/boost/multiprecision/modular/montgomery_params.hpp#L164.

This results in GMP-backended tests failure and also makes the generalization useless. This has to be fixed.

Make modular adaptor hashable

Review of ressol

Ressol implementation seems to have a bug. Incorrect Shanks-Tonnelli algorithm implementation could be a reason.

Test data:
modulus = 0xFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFEFFFFFC2F
a = 0x6650aa44bf64a7e4a62402a88eed89b9c269af6839db7142cb40bb0b6f558063
expected_a_sqrt = 0xbb53444d14f93d9305e2e4d3e3a76639c20aa9f2a87483c6aba0ffb9bf31e48
produced_a_sqrt = 0xf44acbbb2eb06c26cfa1d1b2c1c5899c63df5560d578b7c39545f003640cdde7

Merge boost::multiprecision

Fix a few warnings, that can be interpreted as errors on mac.

Increase performance of modular adaptor

Asm subtract for modular adaptor with not trivial type and limbs size 1

Update CMake-based build system to support the C++03 deprecation

Remove ASM code from multiprecision.

We made a decision to remove ASM code from multiprecision. In some cases it was slightly faster than the C++ code, but in another machines it was significantly slower. We do not intend to support it on all compilers and environments for now, so removing it.

assertion fails on 0==0

Variable_base_scalar_mul test on data {0, 0}, -1 https://github.com/NilFoundation/zkllvm-blueprint/blob/ee0ed71c10775f2a249aecc42db53e55fbdac4d0/test/algebra/curves/plonk/variable_base_scalar_mul.cpp#L166 . Assertion assert(ydata_expected == ydata_real) fails, but should not, because both values are 0

Synchronize with Boost.Multiprecision

The synchronization with Boost.Multiprecision is required to handle low-priority updates and fixes with low effort amount.

Modular negation is broken

Current negate implementation is

constexpr void negate() {
    m_base.negate();
    eval_add(m_base, this->mod_data().get_mod().backend());
}

which breaks the invariant when m_base is zero by getting the value to be the modulus.
This should fix NilFoundation/crypto3-algebra#128.

Using virtual base class for dynamic modular adaptor

Add Checked template parameter to cpp_int_modular_backend

Related to this thread.

TLDR, we want to be able to check if there are any lossy conversions from longer type to shorter one

Dsiable non-working tests

Monty inverse

Review monty inverse algorithm and extended/inverse euclidean algorithm. Implement for different backends, not only for cpp_int_backend.

Import_bits truncate most significant bits if cpp_int size is not a multiple of 8 (or of native word size)

gcc 12.x support

There is a compilation issue with the newer versions of gcc (tested 12.2).

➜  lorem-ipsum-cli git:(1-implement-data-transfer-workflow) ✗ make -C ${ZKLLVM_BUILD:-build} circuit_transpiler -j$(nproc) | tee build_log.log
make: Entering directory '/home/hgedia/Development/nil/lorem-ipsum-cli/build'
[ 50%] Building CXX object bin/circuit_transpiler/CMakeFiles/circuit_transpiler.dir/src/main.cpp.o
In file included from /home/hgedia/Development/nil/lorem-ipsum-cli/libs/crypto3/multiprecision/include/nil/crypto3/multiprecision/cpp_int.hpp:2725,
                 from /home/hgedia/Development/nil/lorem-ipsum-cli/libs/crypto3/algebra/include/nil/crypto3/algebra/fields/detail/exponentiation.hpp:33,
                 from /home/hgedia/Development/nil/lorem-ipsum-cli/libs/crypto3/algebra/include/nil/crypto3/algebra/fields/detail/element/fp.hpp:29,
                 from /home/hgedia/Development/nil/lorem-ipsum-cli/libs/crypto3/algebra/include/nil/crypto3/algebra/fields/pallas/base_field.hpp:29,
                 from /home/hgedia/Development/nil/lorem-ipsum-cli/libs/crypto3/algebra/include/nil/crypto3/algebra/curves/detail/pallas/types.hpp:29,
                 from /home/hgedia/Development/nil/lorem-ipsum-cli/libs/crypto3/algebra/include/nil/crypto3/algebra/curves/pallas.hpp:29,
                 from /home/hgedia/Development/nil/lorem-ipsum-cli/bin/circuit_transpiler/src/main.cpp:6:
/home/hgedia/Development/nil/lorem-ipsum-cli/libs/crypto3/algebra/include/nil/crypto3/algebra/fields/pallas/base_field.hpp:58:81:   in 'constexpr' expansion of 'nil::crypto3::multiprecision::modular_params<nil::crypto3::multiprecision::backends::cpp_int_backend<255, 255, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void> >(nil::crypto3::algebra::fields::pallas_base_field::modulus)'
/home/hgedia/Development/nil/lorem-ipsum-cli/libs/crypto3/multiprecision/include/nil/crypto3/multiprecision/modular/modular_params_fixed.hpp:62:68:   in 'constexpr' expansion of '((nil::crypto3::multiprecision::modular_params<nil::crypto3::multiprecision::backends::cpp_int_backend<255, 255, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void> >*)this)->nil::crypto3::multiprecision::modular_params<nil::crypto3::multiprecision::backends::cpp_int_backend<255, 255, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void> >::m_mod_obj.nil::crypto3::multiprecision::backends::modular_functions_fixed<nil::crypto3::multiprecision::backends::cpp_int_backend<255, 255, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void> >::modular_functions_fixed((* & m))'
/home/hgedia/Development/nil/lorem-ipsum-cli/libs/crypto3/multiprecision/include/nil/crypto3/multiprecision/modular/modular_functions_fixed.hpp:304:35:   in 'constexpr' expansion of '((nil::crypto3::multiprecision::backends::modular_functions_fixed<nil::crypto3::multiprecision::backends::cpp_int_backend<255, 255, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void> >*)this)->nil::crypto3::multiprecision::backends::modular_functions_fixed<nil::crypto3::multiprecision::backends::cpp_int_backend<255, 255, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void> >::initialize(nil::crypto3::multiprecision::number<nil::crypto3::multiprecision::backends::cpp_int_backend<255, 255, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void> >((* & m), 0))'
/home/hgedia/Development/nil/lorem-ipsum-cli/libs/crypto3/multiprecision/include/nil/crypto3/multiprecision/modular/modular_functions_fixed.hpp:269:50:   in 'constexpr' expansion of '((nil::crypto3::multiprecision::backends::modular_functions_fixed<nil::crypto3::multiprecision::backends::cpp_int_backend<255, 255, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void> >*)this)->nil::crypto3::multiprecision::backends::modular_functions_fixed<nil::crypto3::multiprecision::backends::cpp_int_backend<255, 255, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void> >::initialize_barrett_params()'
/home/hgedia/Development/nil/lorem-ipsum-cli/libs/crypto3/multiprecision/include/nil/crypto3/multiprecision/modular/modular_functions_fixed.hpp:219:36:   in 'constexpr' expansion of 'nil::crypto3::multiprecision::backends::eval_divide<511, 511, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void, 255, 255, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void>(((nil::crypto3::multiprecision::backends::modular_functions_fixed<nil::crypto3::multiprecision::backends::cpp_int_backend<255, 255, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void> >*)this)->nil::crypto3::multiprecision::backends::modular_functions_fixed<nil::crypto3::multiprecision::backends::cpp_int_backend<255, 255, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void> >::m_barrett_mu, (*(const nil::crypto3::multiprecision::backends::cpp_int_backend<255, 255, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void>*)(&((nil::crypto3::multiprecision::backends::modular_functions_fixed<nil::crypto3::multiprecision::backends::cpp_int_backend<255, 255, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void> >*)this)->nil::crypto3::multiprecision::backends::modular_functions_fixed<nil::crypto3::multiprecision::backends::cpp_int_backend<255, 255, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void> >::m_mod.nil::crypto3::multiprecision::number<nil::crypto3::multiprecision::backends::cpp_int_backend<255, 255, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void> >::backend())))'
/home/hgedia/Development/nil/lorem-ipsum-cli/libs/crypto3/multiprecision/include/nil/crypto3/multiprecision/cpp_int/divide.hpp:494:32:   in 'constexpr' expansion of 'nil::crypto3::multiprecision::backends::eval_divide<511, 511, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void, 511, 511, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void, 255, 255, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void>((* & result), a, (* & b))'
/home/hgedia/Development/nil/lorem-ipsum-cli/libs/crypto3/multiprecision/include/nil/crypto3/multiprecision/cpp_int/divide.hpp:425:43:   in 'constexpr' expansion of 'nil::crypto3::multiprecision::backends::divide_unsigned_helper<cpp_int_backend<511, 511, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void>, cpp_int_backend<511, 511, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void>, cpp_int_backend<255, 255, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void> >((& result), (* & a), (* & b), r)'
/home/hgedia/Development/nil/lorem-ipsum-cli/libs/crypto3/multiprecision/include/nil/crypto3/multiprecision/cpp_int/divide.hpp:19:40: error: '(((void*)(& a)) == ((void*)(& nil::crypto3::algebra::fields::pallas_base_field::modulus_params.nil::crypto3::multiprecision::modular_params<nil::crypto3::multiprecision::backends::cpp_int_backend<255, 255, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void> >::m_mod_obj.nil::crypto3::multiprecision::backends::modular_functions_fixed<nil::crypto3::multiprecision::backends::cpp_int_backend<255, 255, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void> >::m_barrett_mu)))' is not a constant expression
   19 |                     if (((void*)result == (void*)&x) || ((void*)&r == (void*)&x)) {
      |                         ~~~~~~~~~~~~~~~^~~~~~~~~~~~~
In file included from /home/hgedia/Development/nil/lorem-ipsum-cli/libs/crypto3/hash/include/nil/crypto3/hash/detail/keccak/keccak_functions.hpp:38,
                 from /home/hgedia/Development/nil/lorem-ipsum-cli/libs/crypto3/hash/include/nil/crypto3/hash/keccak.hpp:32,
                 from /home/hgedia/Development/nil/lorem-ipsum-cli/libs/crypto3/zk/include/nil/crypto3/zk/transcript/fiat_shamir.hpp:35,
                 from /home/hgedia/Development/nil/lorem-ipsum-cli/libs/crypto3/zk/include/nil/crypto3/zk/commitments/detail/polynomial/basic_fri.hpp:43,
                 from /home/hgedia/Development/nil/lorem-ipsum-cli/libs/crypto3/marshalling/zk/include/nil/crypto3/marshalling/zk/types/commitments/fri.hpp:45,
                 from /home/hgedia/Development/nil/lorem-ipsum-cli/libs/crypto3/marshalling/zk/include/nil/crypto3/marshalling/zk/types/commitments/lpc.hpp:44,
                 from /home/hgedia/Development/nil/lorem-ipsum-cli/libs/crypto3/marshalling/zk/include/nil/crypto3/marshalling/zk/types/placeholder/proof.hpp:43,
                 from /home/hgedia/Development/nil/lorem-ipsum-cli/bin/circuit_transpiler/src/main.cpp:15:

This code compiles on version 11.3 gcc and seems related to constant expressions. This was found on lorem-ipsum-cli repo.

Steps to re-create

1. install 12.3 version of gcc
2. clone lorem-ipsum repository
3. cmake with compiler path if multiple cmake -G "Unix Makefiles" -B ${ZKLLVM_BUILD:-build} -DCMAKE_BUILD_TYPE=Release -DCMAKE_C_COMPILER=/usr/bin/gcc -DCMAKE_CXX_COMPILER=/usr/bin/g++ .
4. build transpiler make -C ${ZKLLVM_BUILD:-build} circuit_transpiler -j$(nproc)

Implement assembly-based backend

Boost/MIT/Apache2-licensed assembly-based backend might be useful for speed and reliability improvements.

Move import_bits and export_bits from crypto3::multiprecision to crypto3::marshalling

Paired with the crypto3::marshalling issue.

Curve element constructor build fails

Curve constructors build fails
https://github.com/NilFoundation/zkllvm/actions/runs/6316232423/job/17150400010?pr=237#step:14:22046

• turn off warnings
• run benchmarks

Remove support of all the backends, except cpp_int

Warning: unsupported size for integer register

Step

git clone --recursive [email protected]:NilFoundation/ton-voting-protocol.git ever-voting-protocol
cd ever-voting-protocol
mkdir build && cd build
cmake -DCMAKE_BUILD_TYPE=Release ..
make cli

Out

[100%] Building CXX object bin/cli/CMakeFiles/cli.dir/src/main.cpp.o
In file included from /src/ever-voting-protocol/libs/multiprecision/include/nil/crypto3/multiprecision/modular/modular_functions_fixed.hpp:14,
                 from /src/ever-voting-protocol/libs/multiprecision/include/nil/crypto3/multiprecision/cpp_int/eval_jacobi.hpp:15,
                 from /src/ever-voting-protocol/libs/multiprecision/include/nil/crypto3/multiprecision/cpp_int.hpp:2660,
                 from /src/ever-voting-protocol/libs/multiprecision/include/nil/crypto3/multiprecision/cpp_bin_float.hpp:9,
                 from /src/ever-voting-protocol/libs/algebra/include/nil/crypto3/algebra/random_element.hpp:32,
                 from /src/ever-voting-protocol/bin/cli/src/detail/r1cs_examples.hpp:34,
                 from /src/ever-voting-protocol/bin/cli/src/main.cpp:31:
/src/ever-voting-protocol/libs/multiprecision/include/nil/crypto3/multiprecision/modular/asm_functions.hpp: In function ‘bool nil::crypto3::multiprecision::backends::reduce_limb_asm(const size_t&, Limb1*, const Limb2*, const Limb3&) [with Limb1 = long long unsigned int; Limb2 = long long unsigned int; Limb3 = __int128 unsigned]’:
/src/ever-voting-protocol/libs/multiprecision/include/nil/crypto3/multiprecision/modular/asm_functions.hpp:95:17: warning: unsupported size for integer register
   95 |                 }
      |                 ^
[100%] Linking CXX executable cli
[100%] Built target cli

Modular GCD

Fix the constexpr compile issue on GCC 13

When compiling zk on CGG 13 I get the following error:

/home/martun/nil/solana-consensus-proof/libs/multiprecision/include/nil/crypto3/multiprecision/cpp_int/divide.hpp:19:40: error: ‘(((void*)(& a)) == ((void*)(& nil::crypto3::algebra::fields::bls12_base_field<381>::modulus_params.nil::crypto3::multiprecision::modular_params<nil::crypto3::multiprecision::backends::cpp_int_backend<381, 381, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void> >::m_mod_obj.nil::crypto3::multiprecision::backends::modular_functions_fixed<nil::crypto3::multiprecision::backends::cpp_int_backend<381, 381, nil::crypto3::multiprecision::signed_magnitude, nil::crypto3::multiprecision::unchecked, void> >::m_barrett_mu)))’ is not a constant expression
   19 |                     if (((void*)result == (void*)&x) || ((void*)&r == (void*)&x)) {
      |                         ~~~~~~~~~~~~~~~^~~~~~~~~~~~~

Introduce zkLLVM circuit backend

Introduce zkLLVM circuit backend-related definitions and optimizations.

Implement Fürer's multiplication algorithm

https://arxiv.org/abs/1902.10935 or http://www.cse.psu.edu/~furer/Papers/mult.pdf techniques would be great to have implemented.

Slightly fix usage of BOOST_MP_ASSERT

Rename targets and include directories

Since the fork of Boost.Multiprecision has officially happened, it is required to rename directories and change target names to nil:: and nil/-prefixed ones to avoid target selection confusion.

Fix hashing of modular adaptor

Introduce constexpr modular_adaptor

Constexpr modular_adaptor is highly demanded by algebra.

Implement constexpr modular adaptor