new api - wip

.github/workflows/build.yml

@@ -1,31 +1,20 @@
 name: Build
 
-on:
+on: 
   pull_request:
-    types: 
-      - ready_for_review
-      - opened
     branches:
-      - main
-      - dev
+      - "main"
+      - "dev"
     paths:
-      - icicle/**
-      - src/**
-      - Cargo.toml
-      - build.rs
-  push:
-    branches-ignore:
-      - main
-      - dev
-    paths:
-      - icicle/**
-      - src/**
-      - Cargo.toml
-      - build.rs
+      - "icicle/**"
+      - "src/**"
+      - "Cargo.toml"
+      - "build.rs"
 
 env:
   CARGO_TERM_COLOR: always
   ARCH_TYPE: sm_70
+  DEFAULT_STREAM: per-thread
 
 jobs:
   build-linux:

Cargo.toml

@@ -1,49 +1,9 @@
-[package]
-name = "icicle-utils"
+[workspace]
+name = "icicle"
 version = "0.1.0"
 edition = "2021"
-authors = [ "Ingonyama" ]
-description = "An implementation of the Ingonyama CUDA Library"
-homepage = "https://www.ingonyama.com"
-repository = "https://github.com/ingonyama-zk/icicle"
 
-[[bench]]
-name = "ntt"
-path = "benches/ntt.rs"
-harness = false
+# See more keys and their definitions at https://doc.rust-lang.org/cargo/reference/manifest.html
 
-[[bench]]
-name = "msm"
-path = "benches/msm.rs"
-harness = false
+members = ["icicle-core", "bls12-381", "bls12-377", "bn254"]
 
-[dependencies] 
-hex = "*"
-ark-std = "0.3.0"
-ark-ff = "0.3.0"
-ark-poly = "0.3.0"
-ark-ec = { version = "0.3.0", features = [ "parallel" ] }
-ark-bls12-381 = "0.3.0"
-ark-bls12-377 = "0.3.0"
-ark-bn254 = "0.3.0"
-
-serde = { version = "1.0", features = ["derive"] }
-serde_derive = "1.0"
-serde_cbor = "0.11.2"
-
-rustacuda = "0.1"
-rustacuda_core = "0.1"
-rustacuda_derive = "0.1"
-
-rand = "*" #TODO: move rand and ark dependencies to dev once random scalar/point generation is done "natively"
-
-[build-dependencies]
-cc = { version = "1.0", features = ["parallel"] }
-
-[dev-dependencies]
-"criterion" = "0.4.0"
-
-[features]
-default = ["bls12_381"]
-bls12_381 = ["ark-bls12-381/curve"]
-g2 = []

README.md

@@ -4,13 +4,6 @@
                   
 ![image (4)](https://user-images.githubusercontent.com/2446179/223707486-ed8eb5ab-0616-4601-8557-12050df8ccf7.png)
 
-
-<div align="center">
-![Build status](https://github.com/ingonyama-zk/icicle/actions/workflows/buil.yml/badge.svg)
-![Discord server](https://img.shields.io/discord/1063033227788423299?label=Discord&logo=Discord&logoColor=%23&style=plastic)
-![Follow us on twitter](https://img.shields.io/twitter/follow/Ingo_zk?style=social)
-</div>
-
 ## Background
 
 Zero Knowledge Proofs (ZKPs) are considered one of the greatest achievements of modern cryptography. Accordingly, ZKPs are expected to disrupt a number of industries and will usher in an era of trustless and privacy preserving services and infrastructure.
@@ -43,7 +36,7 @@ ICICLE is a CUDA implementation of general functions widely used in ZKP. ICICLE
 
 ```sh
 mkdir -p build
-nvcc -o build/<binary_name> ./icicle/curves/index.cu -lib -arch=native
+nvcc -o build/<ENTER_DIR_NAME> ./icicle/appUtils/ntt/ntt.cu ./icicle/appUtils/msm/msm.cu ./icicle/appUtils/vector_manipulation/ve_mod_mult.cu ./icicle/primitives/projective.cu -lib -arch=native
 ```
 
 ### Testing the CUDA code
@@ -102,64 +95,52 @@ Supporting additional curves can be done as follows:
 
 Create a JSON file with the curve parameters. The curve is defined by the following parameters: 
 - ``curve_name`` - e.g. ``bls12_381``.
-- ``modulus_p`` - scalar field modulus (in decimal).
-- ``bit_count_p`` - number of bits needed to represent `` modulus_p`` .
-- ``limb_p`` - number of bytes needed to represent `` modulus_p``  (rounded).
+- ``modolus_p`` - scalar field modolus (in decimal).
+- ``bit_count_p`` - number of bits needed to represent `` modolus_p`` .
+- ``limb_p`` - number of bytes needed to represent `` modolus_p``  (rounded).
 - ``ntt_size`` - log of the maximal size subgroup of the scalar field.    
-- ``modulus_q`` - base field modulus (in decimal).
-- ``bit_count_q`` - number of bits needed to represent `` modulus_q`` .
-- ``limb_q`` number of bytes needed to represent `` modulus_p``  (rounded).
+- ``modolus_q`` - base field modulus (in decimal).
+- ``bit_count_q`` - number of bits needed to represent `` modolus_q`` .
+- ``limb_q`` number of bytes needed to represent `` modolus_p``  (rounded).
 - ``weierstrass_b`` - Weierstrauss constant of the curve. 
-- ``weierstrass_b_g2_re`` - Weierstrauss real constant of the g2 curve. 
-- ``weierstrass_b_g2_im`` - Weierstrauss imaginary constant of the g2 curve. 
 - ``gen_x`` - x-value of a generator element for the curve. 
 - ``gen_y`` - y-value of a generator element for the curve.
-- ``gen_x_re`` - real x-value of a generator element for the g2 curve. 
-- ``gen_x_im`` - imaginary x-value of a generator element for the g2 curve. 
-- ``gen_y_re`` - real y-value of a generator element for the g2 curve. 
-- ``gen_y_im`` - imaginary y-value of a generator element for the g2 curve. 
 
 Here's an example for BLS12-381.
 ```
 {
     "curve_name" : "bls12_381", 
-    "modulus_p" : 52435875175126190479447740508185965837690552500527637822603658699938581184513,
+    "modolus_p" : 52435875175126190479447740508185965837690552500527637822603658699938581184513,
     "bit_count_p" : 255,
     "limb_p" :  8,
     "ntt_size" : 32,
-    "modulus_q" : 4002409555221667393417789825735904156556882819939007885332058136124031650490837864442687629129015664037894272559787,
+    "modolus_q" : 4002409555221667393417789825735904156556882819939007885332058136124031650490837864442687629129015664037894272559787,
     "bit_count_q" : 381,
     "limb_q" : 12,
     "weierstrass_b" : 4,
-    "weierstrass_b_g2_re":4,
-    "weierstrass_b_g2_im":4,
     "gen_x" : 3685416753713387016781088315183077757961620795782546409894578378688607592378376318836054947676345821548104185464507,
-    "gen_y" : 1339506544944476473020471379941921221584933875938349620426543736416511423956333506472724655353366534992391756441569,
-    "gen_x_re" : 352701069587466618187139116011060144890029952792775240219908644239793785735715026873347600343865175952761926303160,
-    "gen_x_im" : 3059144344244213709971259814753781636986470325476647558659373206291635324768958432433509563104347017837885763365758,
-    "gen_y_re" : 1985150602287291935568054521177171638300868978215655730859378665066344726373823718423869104263333984641494340347905,
-    "gen_y_im" : 927553665492332455747201965776037880757740193453592970025027978793976877002675564980949289727957565575433344219582
+    "gen_y" : 1339506544944476473020471379941921221584933875938349620426543736416511423956333506472724655353366534992391756441569
 }
 ```
 
-Save the parameters JSON file under the ``curve_parameters`` directory.
+Save the parameters JSON file in ``curve_parameters``.
 
-Then run the Python script ``new_curve_script.py `` from the root folder:
+Then run the Python script ``new_curve_script.py `` from the main icicle folder:
 
 ```
-python3 ./curve_parameters/new_curve_script.py ./curve_parameters/bls12_381.json
+python3 ./curve_parameters/new_curve_script_rust.py ./curve_parameters/bls12_381.json
 ```
 
 The script does the following:
 - Creates a folder in ``icicle/curves`` with the curve name, which contains all of the files needed for the supported operations in cuda.
-- Adds the curve's exported operations to ``icicle/curves/index.cu``. 
+- Adds the curve exported operations to ``icicle/curves/index.cu``. 
 - Creates a file with the curve name in ``src/curves`` with the relevant objects for the curve. 
 - Creates a test file with the curve name in ``src``. 
 
 Testing the new curve could be done by running the tests in ``tests_curve_name`` (e.g. ``tests_bls12_381``).
 ## Contributions
 
-Join our [Discord Server][DISCORD] and find us on the icicle channel. We will be happy to work together to support your use case and talk features, bugs and design.
+Join our [Discord Server](https://discord.gg/Y4SkbDf2Ff) and find us on the icicle channel. We will be happy to work together to support your use case and talk features, bugs and design.
 
 ### Hall of Fame
 
@@ -172,16 +153,13 @@ ICICLE is distributed under the terms of the MIT License.
 See [LICENSE-MIT][LMIT] for details.
 
 <!-- Begin Links -->
-[BLS12-381]: ./icicle/curves/bls12_381/supported_operations.cu
-[BLS12-377]: ./icicle/curves/bls12_377/supported_operations.cu
-[BN254]: ./icicle/curves/bn254/supported_operations.cu
+[BLS12-381]: ./icicle/curves/bls12_381.cuh
 [NVCC]: https://docs.nvidia.com/cuda/#installation-guides
 [CRV_TEMPLATE]: ./icicle/curves/curve_template.cuh
 [CRV_CONFIG]: ./icicle/curves/curve_config.cuh
 [B_SCRIPT]: ./build.rs
 [FDI]: https://github.com/ingonyama-zk/fast-danksharding
 [LMIT]: ./LICENSE
-[DISCORD]: https://discord.gg/Y4SkbDf2Ff
 [googletest]: https://github.com/google/googletest/
 
 <!-- End Links -->

benches/msm.rs

@@ -1,52 +0,0 @@
-extern crate criterion;
-
-use criterion::{criterion_group, criterion_main, Criterion};
-
-use icicle_utils::test_bls12_381::{
-    commit_batch_bls12_381, generate_random_points_bls12_381, set_up_scalars_bls12_381,
-};
-use icicle_utils::utils::*;
-#[cfg(feature = "g2")]
-use icicle_utils::{commit_batch_g2, field::ExtensionField};
-
-use rustacuda::prelude::*;
-
-const LOG_MSM_SIZES: [usize; 1] = [12];
-const BATCH_SIZES: [usize; 2] = [128, 256];
-
-fn bench_msm(c: &mut Criterion) {
-    let mut group = c.benchmark_group("MSM");
-    for log_msm_size in LOG_MSM_SIZES {
-        for batch_size in BATCH_SIZES {
-            let msm_size = 1 << log_msm_size;
-            let (scalars, _, _) = set_up_scalars_bls12_381(msm_size, 0, false);
-            let batch_scalars = vec![scalars; batch_size].concat();
-            let mut d_scalars = DeviceBuffer::from_slice(&batch_scalars[..]).unwrap();
-
-            let points = generate_random_points_bls12_381(msm_size, get_rng(None));
-            let batch_points = vec![points; batch_size].concat();
-            let mut d_points = DeviceBuffer::from_slice(&batch_points[..]).unwrap();
-
-            #[cfg(feature = "g2")]
-            let g2_points = generate_random_points::<ExtensionField>(msm_size, get_rng(None));
-            #[cfg(feature = "g2")]
-            let g2_batch_points = vec![g2_points; batch_size].concat();
-            #[cfg(feature = "g2")]
-            let mut d_g2_points = DeviceBuffer::from_slice(&g2_batch_points[..]).unwrap();
-
-            group.sample_size(30).bench_function(
-                &format!("MSM of size 2^{} in batch {}", log_msm_size, batch_size),
-                |b| b.iter(|| commit_batch_bls12_381(&mut d_points, &mut d_scalars, batch_size)),
-            );
-
-            #[cfg(feature = "g2")]
-            group.sample_size(10).bench_function(
-                &format!("G2 MSM of size 2^{} in batch {}", log_msm_size, batch_size),
-                |b| b.iter(|| commit_batch_g2(&mut d_g2_points, &mut d_scalars, batch_size))
-            );
-        }
-    }
-}
-
-criterion_group!(msm_benches, bench_msm);
-criterion_main!(msm_benches);

benches/ntt.rs

@@ -1,34 +0,0 @@
-extern crate criterion;
-
-use criterion::{criterion_group, criterion_main, Criterion};
-
-use icicle_utils::test_bls12_381::{interpolate_scalars_batch_bls12_381, interpolate_points_batch_bls12_381, set_up_scalars_bls12_381, set_up_points_bls12_381};
-
-
-const LOG_NTT_SIZES: [usize; 1] = [15];
-const BATCH_SIZES: [usize; 2] = [8, 16];
-
-fn bench_ntt(c: &mut Criterion) {
-    let mut group = c.benchmark_group("NTT");
-    for log_ntt_size in LOG_NTT_SIZES {
-        for batch_size in BATCH_SIZES {
-            let ntt_size = 1 << log_ntt_size;
-            let (_, mut d_evals, mut d_domain) = set_up_scalars_bls12_381(ntt_size * batch_size, log_ntt_size, true);
-            let (_, mut d_points_evals, _) = set_up_points_bls12_381(ntt_size * batch_size, log_ntt_size, true);
-
-            group.sample_size(100).bench_function(
-                &format!("Scalar NTT of size 2^{} in batch {}", log_ntt_size, batch_size),
-                |b| b.iter(|| interpolate_scalars_batch_bls12_381(&mut d_evals, &mut d_domain, batch_size))
-            );
-
-            group.sample_size(10).bench_function(
-                &format!("EC NTT of size 2^{} in batch {}", log_ntt_size, batch_size),
-                |b| b.iter(|| interpolate_points_batch_bls12_381(&mut d_points_evals, &mut d_domain, batch_size))
-            );
-        }
-    }
-}
-
-criterion_group!(ntt_benches, bench_ntt);
-criterion_main!(ntt_benches);
-

bls12-377/Cargo.toml

@@ -0,0 +1,34 @@
+[package]
+name = "bls12-377"
+version = "0.1.0"
+edition = "2021"
+authors = [ "Ingonyama" ]
+
+[dependencies]
+icicle-core = { path = "../icicle-core" }
+
+hex = "*"
+ark-std = "0.3.0"
+ark-ff = "0.3.0"
+ark-poly = "0.3.0"
+ark-ec = { version = "0.3.0", features = [ "parallel" ] }
+ark-bls12-377 = "0.3.0"
+
+serde = { version = "1.0", features = ["derive"] }
+serde_derive = "1.0"
+serde_cbor = "0.11.2"
+
+rustacuda = "0.1"
+rustacuda_core = "0.1"
+rustacuda_derive = "0.1"
+
+rand = "*" #TODO: move rand and ark dependencies to dev once random scalar/point generation is done "natively"
+
+[build-dependencies]
+cc = { version = "1.0", features = ["parallel"] }
+
+[dev-dependencies]
+"criterion" = "0.4.0"
+
+[features]
+g2 = []

bls12-377/build.rs

@@ -23,11 +23,12 @@ fn main() {
         nvcc.define("G2_DEFINED", None);
     }
     nvcc.cuda(true);
+    nvcc.define("FEATURE_BLS12_377", None);
     nvcc.debug(false);
     nvcc.flag(&arch);
     nvcc.flag(&stream);
     nvcc.files([
-        "./icicle/curves/index.cu",
+        "../icicle-cuda/curves/index.cu",
     ]);
     nvcc.compile("ingo_icicle"); //TODO: extension??
 }

bls12-377/src/basic_structs/field.rs

@@ -0,0 +1,4 @@
+pub trait Field<const NUM_LIMBS: usize> {
+    const MODOLUS: [u32;NUM_LIMBS];
+    const LIMBS: usize = NUM_LIMBS;
+}

bls12-377/src/basic_structs/mod.rs

@@ -0,0 +1,3 @@
+pub mod field; 
+pub mod scalar; 
+pub mod point; 

bls12-377/src/basic_structs/point.rs

@@ -0,0 +1,106 @@
+use std::ffi::c_uint;
+
+use ark_ec::AffineCurve;
+use ark_ff::{BigInteger256, PrimeField};
+use std::mem::transmute;
+use ark_ff::Field;
+use icicle_core::utils::{u32_vec_to_u64_vec, u64_vec_to_u32_vec};
+
+use rustacuda_core::DeviceCopy;
+use rustacuda_derive::DeviceCopy;
+
+use super::scalar::{get_fixed_limbs, self};
+
+
+#[derive(Debug, Clone, Copy, DeviceCopy)]
+#[repr(C)]
+pub struct PointT<BF: scalar::ScalarTrait> {
+    pub x: BF,
+    pub y: BF,
+    pub z: BF,
+}
+
+impl<BF: DeviceCopy + scalar::ScalarTrait> Default for PointT<BF> {
+    fn default() -> Self {
+        PointT::zero()
+    }
+}
+
+impl<BF: DeviceCopy + scalar::ScalarTrait> PointT<BF> {
+    pub fn zero() -> Self {
+        PointT {
+            x: BF::zero(),
+            y: BF::one(),
+            z: BF::zero(),
+        }
+    }
+
+    pub fn infinity() -> Self {
+        Self::zero()
+    }
+}
+
+#[derive(Debug, PartialEq, Clone, Copy, DeviceCopy)]
+#[repr(C)]
+pub struct PointAffineNoInfinityT<BF> {
+    pub x: BF,
+    pub y: BF,
+}
+
+impl<BF: scalar::ScalarTrait> Default for PointAffineNoInfinityT<BF> {
+    fn default() -> Self {
+        PointAffineNoInfinityT {
+            x: BF::zero(),
+            y: BF::zero(),
+        }
+    }
+}
+
+impl<BF: Copy + scalar::ScalarTrait> PointAffineNoInfinityT<BF> {
+    ///From u32 limbs x,y
+    pub fn from_limbs(x: &[u32], y: &[u32]) -> Self {
+        PointAffineNoInfinityT {
+            x: BF::from_limbs(x),
+            y: BF::from_limbs(y)
+        }
+    }
+
+    pub fn limbs(&self) -> Vec<u32> {
+        [self.x.limbs(), self.y.limbs()].concat()
+    }
+
+    pub fn to_projective(&self) -> PointT<BF> {
+        PointT {
+            x: self.x,
+            y: self.y,
+            z: BF::one(),
+        }
+    }
+}
+
+impl<BF: Copy + scalar::ScalarTrait> PointT<BF>  {
+    pub fn from_limbs(x: &[u32], y: &[u32], z: &[u32]) -> Self {
+        PointT {
+            x: BF::from_limbs(x),
+            y: BF::from_limbs(y),
+            z: BF::from_limbs(z)
+        }
+    }
+
+    pub fn from_xy_limbs(value: &[u32]) -> PointT<BF> {
+        let l = value.len();
+        assert_eq!(l, 3 * BF::base_limbs(), "length must be 3 * {}", BF::base_limbs());
+        PointT {
+            x: BF::from_limbs(value[..BF::base_limbs()].try_into().unwrap()),
+            y: BF::from_limbs(value[BF::base_limbs()..BF::base_limbs() * 2].try_into().unwrap()),
+            z: BF::from_limbs(value[BF::base_limbs() * 2..].try_into().unwrap())
+        }
+    }
+
+    pub fn to_xy_strip_z(&self) -> PointAffineNoInfinityT<BF> {
+        PointAffineNoInfinityT {
+            x: self.x,
+            y: self.y,
+        }
+    }
+}

bls12-377/src/basic_structs/scalar.rs

@@ -0,0 +1,102 @@
+use std::ffi::{c_int, c_uint};
+use rand::{rngs::StdRng, RngCore, SeedableRng};
+use rustacuda_core::DeviceCopy;
+use rustacuda_derive::DeviceCopy;
+use std::mem::transmute;
+use rustacuda::prelude::*;
+use rustacuda_core::DevicePointer;
+use rustacuda::memory::{DeviceBox, CopyDestination};
+
+use icicle_core::utils::{u32_vec_to_u64_vec, u64_vec_to_u32_vec};
+
+use std::marker::PhantomData;
+use std::convert::TryInto;
+
+use super::field::{Field, self};
+
+pub fn get_fixed_limbs<const NUM_LIMBS: usize>(val: &[u32]) -> [u32; NUM_LIMBS] {
+    match val.len() {
+        n if n < NUM_LIMBS => {
+            let mut padded: [u32; NUM_LIMBS] = [0; NUM_LIMBS];
+            padded[..val.len()].copy_from_slice(&val);
+            padded
+        }
+        n if n == NUM_LIMBS => val.try_into().unwrap(),
+        _ => panic!("slice has too many elements"),
+    }
+}
+
+pub trait ScalarTrait{
+    fn base_limbs() -> usize;
+    fn zero() -> Self;
+    fn from_limbs(value: &[u32]) -> Self;
+    fn one() -> Self;
+    fn to_bytes_le(&self) -> Vec<u8>;
+    fn limbs(&self) -> &[u32];
+}
+
+#[derive(Debug, PartialEq, Clone, Copy)]
+#[repr(C)]
+pub struct ScalarT<M, const NUM_LIMBS: usize> {
+    pub(crate) phantom: PhantomData<M>,
+    pub(crate) value : [u32; NUM_LIMBS]
+}
+
+impl<M, const NUM_LIMBS: usize> ScalarTrait for ScalarT<M, NUM_LIMBS>
+where
+    M: Field<NUM_LIMBS>,
+{
+
+    fn base_limbs() -> usize {
+        return NUM_LIMBS; 
+    }
+
+    fn zero() -> Self {
+        ScalarT {
+            value: [0u32; NUM_LIMBS],
+            phantom: PhantomData,
+        }
+    }
+
+    fn from_limbs(value: &[u32]) -> Self {
+        Self {
+            value: get_fixed_limbs(value),
+            phantom: PhantomData,
+        }
+    }
+
+    fn one() -> Self {
+        let mut s = [0u32; NUM_LIMBS];
+        s[0] = 1;
+        ScalarT { value: s, phantom: PhantomData }
+    }
+
+    fn to_bytes_le(&self) -> Vec<u8> {
+        self.value
+            .iter()
+            .map(|s| s.to_le_bytes().to_vec())
+            .flatten()
+            .collect::<Vec<_>>()
+    }
+
+    fn limbs(&self) -> &[u32] {
+        &self.value
+    }
+}
+
+impl<M, const NUM_LIMBS: usize> ScalarT<M, NUM_LIMBS> where M: field::Field<NUM_LIMBS>{
+    pub fn from_limbs_le(value: &[u32]) -> ScalarT<M,NUM_LIMBS> {
+        Self::from_limbs(value)
+     }
+ 
+    pub fn from_limbs_be(value: &[u32]) -> ScalarT<M,NUM_LIMBS> {
+         let mut value = value.to_vec();
+         value.reverse();
+         Self::from_limbs_le(&value)
+     }
+ 
+     // Additional Functions
+     pub fn add(&self, other:ScalarT<M, NUM_LIMBS>) -> ScalarT<M,NUM_LIMBS>{  // overload + 
+         return ScalarT{value: [self.value[0] + other.value[0];NUM_LIMBS], phantom: PhantomData }; 
+     }
+}

bls12-377/src/curve_structs.rs

@@ -0,0 +1,62 @@
+use std::ffi::{c_int, c_uint};
+use rand::{rngs::StdRng, RngCore, SeedableRng};
+use rustacuda_derive::DeviceCopy;
+use std::mem::transmute;
+use rustacuda::prelude::*;
+use rustacuda_core::DevicePointer;
+use rustacuda::memory::{DeviceBox, CopyDestination, DeviceCopy};
+
+use std::marker::PhantomData;
+use std::convert::TryInto;
+
+use crate::basic_structs::point::{PointT, PointAffineNoInfinityT};
+use crate::basic_structs::scalar::ScalarT;
+use crate::basic_structs::field::Field;
+
+
+#[derive(Debug, PartialEq, Clone, Copy,DeviceCopy)]
+#[repr(C)]
+pub struct ScalarField;
+impl Field<8> for ScalarField {
+    const MODOLUS: [u32; 8] = [0x0;8];
+}
+
+#[derive(Debug, PartialEq, Clone, Copy,DeviceCopy)]
+#[repr(C)]
+pub struct BaseField;
+impl Field<12> for BaseField {
+    const MODOLUS: [u32; 12] = [0x0;12];
+}
+
+
+pub type Scalar = ScalarT<ScalarField,8>;
+impl Default for Scalar {
+    fn default() -> Self {
+        Self{value: [0x0;ScalarField::LIMBS], phantom: PhantomData }
+    }
+}
+
+unsafe impl DeviceCopy for Scalar{}
+
+
+pub type Base = ScalarT<BaseField,12>;
+impl Default for Base {
+    fn default() -> Self {
+        Self{value: [0x0;BaseField::LIMBS], phantom: PhantomData }
+    }
+}
+
+unsafe impl DeviceCopy for Base{}
+
+pub type Point = PointT<Base>;
+pub type PointAffineNoInfinity = PointAffineNoInfinityT<Base>;
+
+extern "C" {
+    fn eq(point1: *const Point, point2: *const Point) -> c_uint;
+}
+
+impl PartialEq for Point {
+    fn eq(&self, other: &Self) -> bool {
+        unsafe { eq(self, other) != 0 }
+    }
+}

bls12-377/src/from_cuda.rs

@@ -0,0 +1,798 @@
+use std::ffi::{c_int, c_uint};
+use ark_std::UniformRand;
+use rand::{rngs::StdRng, RngCore, SeedableRng};
+use rustacuda::CudaFlags;
+use rustacuda::memory::DeviceBox;
+use rustacuda::prelude::{DeviceBuffer, Device, ContextFlags, Context};
+use rustacuda_core::DevicePointer;
+use std::mem::transmute;
+use crate::basic_structs::scalar::ScalarTrait;
+use crate::curve_structs::*;
+use icicle_core::utils::{u32_vec_to_u64_vec, u64_vec_to_u32_vec};
+use std::marker::PhantomData;
+use std::convert::TryInto;
+use ark_bls12_377::{Fq as Fq_BLS12_377, Fr as Fr_BLS12_377, G1Affine as G1Affine_BLS12_377, G1Projective as G1Projective_BLS12_377};
+use ark_ec::AffineCurve;
+use ark_ff::{BigInteger384, BigInteger256, PrimeField};
+use rustacuda::memory::{CopyDestination, DeviceCopy};
+
+extern "C" {
+    fn msm_cuda(
+        out: *mut Point,
+        points: *const PointAffineNoInfinity,
+        scalars: *const Scalar,
+        count: usize,
+        device_id: usize,
+    ) -> c_uint;
+
+    fn msm_batch_cuda(
+        out: *mut Point,
+        points: *const PointAffineNoInfinity,
+        scalars: *const Scalar,
+        batch_size: usize,
+        msm_size: usize,
+        device_id: usize,
+    ) -> c_uint;
+
+    fn commit_cuda(
+        d_out: DevicePointer<Point>,
+        d_scalars: DevicePointer<Scalar>,
+        d_points: DevicePointer<PointAffineNoInfinity>,
+        count: usize,
+        device_id: usize,
+    ) -> c_uint;
+
+    fn commit_batch_cuda(
+        d_out: DevicePointer<Point>,
+        d_scalars: DevicePointer<Scalar>,
+        d_points: DevicePointer<PointAffineNoInfinity>,
+        count: usize,
+        batch_size: usize,
+        device_id: usize,
+    ) -> c_uint;
+
+    fn build_domain_cuda(domain_size: usize, logn: usize, inverse: bool, device_id: usize) -> DevicePointer<Scalar>;
+
+    fn ntt_cuda(inout: *mut Scalar, n: usize, inverse: bool, device_id: usize) -> c_int;
+
+    fn ecntt_cuda(inout: *mut Point, n: usize, inverse: bool, device_id: usize) -> c_int;
+
+    fn ntt_batch_cuda(
+        inout: *mut Scalar,
+        arr_size: usize,
+        n: usize,
+        inverse: bool,
+    ) -> c_int;
+
+    fn ecntt_batch_cuda(inout: *mut Point, arr_size: usize, n: usize, inverse: bool) -> c_int;
+
+    fn interpolate_scalars_cuda(
+        d_out: DevicePointer<Scalar>,
+        d_evaluations: DevicePointer<Scalar>,
+        d_domain: DevicePointer<Scalar>, 
+        n: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn interpolate_scalars_batch_cuda(
+        d_out: DevicePointer<Scalar>,
+        d_evaluations: DevicePointer<Scalar>,
+        d_domain: DevicePointer<Scalar>,
+        n: usize,
+        batch_size: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn interpolate_points_cuda(
+        d_out: DevicePointer<Point>,
+        d_evaluations: DevicePointer<Point>,
+        d_domain: DevicePointer<Scalar>,
+        n: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn interpolate_points_batch_cuda(
+        d_out: DevicePointer<Point>,
+        d_evaluations: DevicePointer<Point>,
+        d_domain: DevicePointer<Scalar>,
+        n: usize,
+        batch_size: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn evaluate_scalars_cuda(
+        d_out: DevicePointer<Scalar>,
+        d_coefficients: DevicePointer<Scalar>,
+        d_domain: DevicePointer<Scalar>,
+        domain_size: usize,
+        n: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn evaluate_scalars_batch_cuda(
+        d_out: DevicePointer<Scalar>,
+        d_coefficients: DevicePointer<Scalar>,
+        d_domain: DevicePointer<Scalar>,
+        domain_size: usize,
+        n: usize,
+        batch_size: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn evaluate_points_cuda(
+        d_out: DevicePointer<Point>,
+        d_coefficients: DevicePointer<Point>,
+        d_domain: DevicePointer<Scalar>,
+        domain_size: usize,
+        n: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn evaluate_points_batch_cuda(
+        d_out: DevicePointer<Point>,
+        d_coefficients: DevicePointer<Point>,
+        d_domain: DevicePointer<Scalar>,
+        domain_size: usize,
+        n: usize,
+        batch_size: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn evaluate_scalars_on_coset_cuda(
+        d_out: DevicePointer<Scalar>,
+        d_coefficients: DevicePointer<Scalar>,
+        d_domain: DevicePointer<Scalar>,
+        domain_size: usize,
+        n: usize,
+        coset_powers: DevicePointer<Scalar>,
+        device_id: usize
+    ) -> c_int;
+
+    fn evaluate_scalars_on_coset_batch_cuda(
+        d_out: DevicePointer<Scalar>,
+        d_coefficients: DevicePointer<Scalar>,
+        d_domain: DevicePointer<Scalar>,
+        domain_size: usize,
+        n: usize,
+        batch_size: usize,
+        coset_powers: DevicePointer<Scalar>,
+        device_id: usize
+    ) -> c_int;
+
+    fn evaluate_points_on_coset_cuda(
+        d_out: DevicePointer<Point>,
+        d_coefficients: DevicePointer<Point>,
+        d_domain: DevicePointer<Scalar>,
+        domain_size: usize,
+        n: usize,
+        coset_powers: DevicePointer<Scalar>,
+        device_id: usize
+    ) -> c_int;
+
+    fn evaluate_points_on_coset_batch_cuda(
+        d_out: DevicePointer<Point>,
+        d_coefficients: DevicePointer<Point>,
+        d_domain: DevicePointer<Scalar>,
+        domain_size: usize,
+        n: usize,
+        batch_size: usize,
+        coset_powers: DevicePointer<Scalar>,
+        device_id: usize
+    ) -> c_int;
+
+    fn reverse_order_scalars_cuda(
+        d_arr: DevicePointer<Scalar>,
+        n: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn reverse_order_scalars_batch_cuda(
+        d_arr: DevicePointer<Scalar>,
+        n: usize,
+        batch_size: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn reverse_order_points_cuda(
+        d_arr: DevicePointer<Point>,
+        n: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn reverse_order_points_batch_cuda(
+        d_arr: DevicePointer<Point>,
+        n: usize,
+        batch_size: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn vec_mod_mult_point(
+        inout: *mut Point,
+        scalars: *const Scalar,
+        n_elements: usize,
+        device_id: usize,
+    ) -> c_int;
+
+    fn vec_mod_mult_scalar(
+        inout: *mut Scalar,
+        scalars: *const Scalar,
+        n_elements: usize,
+        device_id: usize,
+    ) -> c_int;
+
+    fn matrix_vec_mod_mult(
+        matrix_flattened: *const Scalar,
+        input: *const Scalar,
+        output: *mut Scalar,
+        n_elements: usize,
+        device_id: usize,
+    ) -> c_int;
+}
+
+pub fn msm(points: &[PointAffineNoInfinity], scalars: &[Scalar], device_id: usize) -> Point {
+    let count = points.len();
+    if count != scalars.len() {
+        todo!("variable length")
+    }
+
+    let mut ret = Point::zero();
+    unsafe {
+        msm_cuda(
+            &mut ret as *mut _ as *mut Point,
+            points as *const _ as *const PointAffineNoInfinity,
+            scalars as *const _ as *const Scalar,
+            scalars.len(),
+            device_id,
+        )
+    };
+
+    ret
+}
+
+pub fn msm_batch(
+    points: &[PointAffineNoInfinity],
+    scalars: &[Scalar],
+    batch_size: usize,
+    device_id: usize,
+) -> Vec<Point> {
+    let count = points.len();
+    if count != scalars.len() {
+        todo!("variable length")
+    }
+
+    let mut ret = vec![Point::zero(); batch_size];
+
+    unsafe {
+        msm_batch_cuda(
+            &mut ret[0] as *mut _ as *mut Point,
+            points as *const _ as *const PointAffineNoInfinity,
+            scalars as *const _ as *const Scalar,
+            batch_size,
+            count / batch_size,
+            device_id,
+        )
+    };
+
+    ret
+}
+
+pub fn commit(
+    points: &mut DeviceBuffer<PointAffineNoInfinity>,
+    scalars: &mut DeviceBuffer<Scalar>,
+) -> DeviceBox<Point> {
+    let mut res = DeviceBox::new(&Point::zero()).unwrap();
+    unsafe {
+        commit_cuda(
+            res.as_device_ptr(),
+            scalars.as_device_ptr(),
+            points.as_device_ptr(),
+            scalars.len(),
+            0,
+        );
+    }
+    return res;
+}
+
+pub fn commit_batch(
+    points: &mut DeviceBuffer<PointAffineNoInfinity>,
+    scalars: &mut DeviceBuffer<Scalar>,
+    batch_size: usize,
+) -> DeviceBuffer<Point> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(batch_size).unwrap() };
+    unsafe {
+        commit_batch_cuda(
+            res.as_device_ptr(),
+            scalars.as_device_ptr(),
+            points.as_device_ptr(),
+            scalars.len() / batch_size,
+            batch_size,
+            0,
+        );
+    }
+    return res;
+}
+
+/// Compute an in-place NTT on the input data.
+fn ntt_internal(values: &mut [Scalar], device_id: usize, inverse: bool) -> i32 {
+    let ret_code = unsafe {
+        ntt_cuda(
+            values as *mut _ as *mut Scalar,
+            values.len(),
+            inverse,
+            device_id,
+        )
+    };
+    ret_code
+}
+
+pub fn ntt(values: &mut [Scalar], device_id: usize) {
+    ntt_internal(values, device_id, false);
+}
+
+pub fn intt(values: &mut [Scalar], device_id: usize) {
+    ntt_internal(values, device_id, true);
+}
+
+/// Compute an in-place NTT on the input data.
+fn ntt_internal_batch(
+    values: &mut [Scalar],
+    device_id: usize,
+    batch_size: usize,
+    inverse: bool,
+) -> i32 {
+    unsafe {
+        ntt_batch_cuda(
+            values as *mut _ as *mut Scalar,
+            values.len(),
+            batch_size,
+            inverse,
+        )
+    }
+}
+
+pub fn ntt_batch(values: &mut [Scalar], batch_size: usize, device_id: usize) {
+    ntt_internal_batch(values, 0, batch_size, false);
+}
+
+pub fn intt_batch(values: &mut [Scalar], batch_size: usize, device_id: usize) {
+    ntt_internal_batch(values, 0, batch_size, true);
+}
+
+/// Compute an in-place ECNTT on the input data.
+fn ecntt_internal(values: &mut [Point], inverse: bool, device_id: usize) -> i32 {
+    unsafe {
+        ecntt_cuda(
+            values as *mut _ as *mut Point,
+            values.len(),
+            inverse,
+            device_id,
+        )
+    }
+}
+
+pub fn ecntt(values: &mut [Point], device_id: usize) {
+    ecntt_internal(values, false, device_id);
+}
+
+/// Compute an in-place iECNTT on the input data.
+pub fn iecntt(values: &mut [Point], device_id: usize) {
+    ecntt_internal(values, true, device_id);
+}
+
+/// Compute an in-place ECNTT on the input data.
+fn ecntt_internal_batch(
+    values: &mut [Point],
+    device_id: usize,
+    batch_size: usize,
+    inverse: bool,
+) -> i32 {
+    unsafe {
+        ecntt_batch_cuda(
+            values as *mut _ as *mut Point,
+            values.len(),
+            batch_size,
+            inverse,
+        )
+    }
+}
+
+pub fn ecntt_batch(values: &mut [Point], batch_size: usize, device_id: usize) {
+    ecntt_internal_batch(values, 0, batch_size, false);
+}
+
+/// Compute an in-place iECNTT on the input data.
+pub fn iecntt_batch(values: &mut [Point], batch_size: usize, device_id: usize) {
+    ecntt_internal_batch(values, 0, batch_size, true);
+}
+
+pub fn build_domain(domain_size: usize, logn: usize, inverse: bool) -> DeviceBuffer<Scalar> {
+    unsafe {
+        DeviceBuffer::from_raw_parts(build_domain_cuda(
+            domain_size,
+            logn,
+            inverse,
+            0
+        ), domain_size)
+    }
+}
+
+
+pub fn reverse_order_scalars(
+    d_scalars: &mut DeviceBuffer<Scalar>,
+) {
+    unsafe { reverse_order_scalars_cuda(
+        d_scalars.as_device_ptr(),
+        d_scalars.len(),
+        0
+    ); }
+}
+
+pub fn reverse_order_scalars_batch(
+    d_scalars: &mut DeviceBuffer<Scalar>,
+    batch_size: usize,
+) {
+    unsafe { reverse_order_scalars_batch_cuda(
+        d_scalars.as_device_ptr(),
+        d_scalars.len() / batch_size,
+        batch_size,
+        0
+    ); }
+}
+
+pub fn reverse_order_points(
+    d_points: &mut DeviceBuffer<Point>,
+) {
+    unsafe { reverse_order_points_cuda(
+        d_points.as_device_ptr(),
+        d_points.len(),
+        0
+    ); }
+}
+
+pub fn reverse_order_points_batch(
+    d_points: &mut DeviceBuffer<Point>,
+    batch_size: usize,
+) {
+    unsafe { reverse_order_points_batch_cuda(
+        d_points.as_device_ptr(),
+        d_points.len() / batch_size,
+        batch_size,
+        0
+    ); }
+}
+
+pub fn interpolate_scalars(
+    d_evaluations: &mut DeviceBuffer<Scalar>,
+    d_domain: &mut DeviceBuffer<Scalar>
+) -> DeviceBuffer<Scalar> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len()).unwrap() };
+    unsafe { interpolate_scalars_cuda(
+        res.as_device_ptr(),
+        d_evaluations.as_device_ptr(),
+        d_domain.as_device_ptr(),
+        d_domain.len(),
+        0
+    ) };
+    return res;
+}
+
+pub fn interpolate_scalars_batch(
+    d_evaluations: &mut DeviceBuffer<Scalar>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+    batch_size: usize,
+) -> DeviceBuffer<Scalar> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len() * batch_size).unwrap() };
+    unsafe { interpolate_scalars_batch_cuda(
+        res.as_device_ptr(),
+        d_evaluations.as_device_ptr(),
+        d_domain.as_device_ptr(),
+        d_domain.len(),
+        batch_size,
+        0
+    ) };
+    return res;
+}
+
+pub fn interpolate_points(
+    d_evaluations: &mut DeviceBuffer<Point>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+) -> DeviceBuffer<Point> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len()).unwrap() };
+    unsafe { interpolate_points_cuda(
+        res.as_device_ptr(),
+        d_evaluations.as_device_ptr(),
+        d_domain.as_device_ptr(),
+        d_domain.len(),
+        0
+    ) };
+    return res;
+}
+
+pub fn interpolate_points_batch(
+    d_evaluations: &mut DeviceBuffer<Point>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+    batch_size: usize,
+) -> DeviceBuffer<Point> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len() * batch_size).unwrap() };
+    unsafe { interpolate_points_batch_cuda(
+        res.as_device_ptr(),
+        d_evaluations.as_device_ptr(),
+        d_domain.as_device_ptr(),
+        d_domain.len(),
+        batch_size,
+        0
+    ) };
+    return res;
+}
+
+pub fn evaluate_scalars(
+    d_coefficients: &mut DeviceBuffer<Scalar>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+) -> DeviceBuffer<Scalar> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len()).unwrap() };
+    unsafe {
+        evaluate_scalars_cuda(
+            res.as_device_ptr(),
+            d_coefficients.as_device_ptr(),
+            d_domain.as_device_ptr(),
+            d_domain.len(),
+            d_coefficients.len(),
+            0
+        );
+    }
+    return res;
+}
+
+pub fn evaluate_scalars_batch(
+    d_coefficients: &mut DeviceBuffer<Scalar>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+    batch_size: usize,
+) -> DeviceBuffer<Scalar> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len() * batch_size).unwrap() };
+    unsafe {
+        evaluate_scalars_batch_cuda(
+            res.as_device_ptr(),
+            d_coefficients.as_device_ptr(),
+            d_domain.as_device_ptr(),
+            d_domain.len(),
+            d_coefficients.len() / batch_size,
+            batch_size,
+            0
+        );
+    }
+    return res;
+}
+
+pub fn evaluate_points(
+    d_coefficients: &mut DeviceBuffer<Point>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+) -> DeviceBuffer<Point> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len()).unwrap() };
+    unsafe {
+        evaluate_points_cuda(
+            res.as_device_ptr(),
+            d_coefficients.as_device_ptr(),
+            d_domain.as_device_ptr(),
+            d_domain.len(),
+            d_coefficients.len(),
+            0
+        );
+    }
+    return res;
+}
+
+pub fn evaluate_points_batch(
+    d_coefficients: &mut DeviceBuffer<Point>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+    batch_size: usize,
+) -> DeviceBuffer<Point> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len() * batch_size).unwrap() };
+    unsafe {
+        evaluate_points_batch_cuda(
+            res.as_device_ptr(),
+            d_coefficients.as_device_ptr(),
+            d_domain.as_device_ptr(),
+            d_domain.len(),
+            d_coefficients.len() / batch_size,
+            batch_size,
+            0
+        );
+    }
+    return res;
+}
+
+pub fn evaluate_scalars_on_coset(
+    d_coefficients: &mut DeviceBuffer<Scalar>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+    coset_powers: &mut DeviceBuffer<Scalar>,
+) -> DeviceBuffer<Scalar> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len()).unwrap() };
+    unsafe {
+        evaluate_scalars_on_coset_cuda(
+            res.as_device_ptr(),
+            d_coefficients.as_device_ptr(),
+            d_domain.as_device_ptr(),
+            d_domain.len(),
+            d_coefficients.len(),
+            coset_powers.as_device_ptr(),
+            0
+        );
+    }
+    return res;
+}
+
+pub fn evaluate_scalars_on_coset_batch(
+    d_coefficients: &mut DeviceBuffer<Scalar>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+    batch_size: usize,
+    coset_powers: &mut DeviceBuffer<Scalar>,
+) -> DeviceBuffer<Scalar> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len() * batch_size).unwrap() };
+    unsafe {
+        evaluate_scalars_on_coset_batch_cuda(
+            res.as_device_ptr(),
+            d_coefficients.as_device_ptr(),
+            d_domain.as_device_ptr(),
+            d_domain.len(),
+            d_coefficients.len() / batch_size,
+            batch_size,
+            coset_powers.as_device_ptr(),
+            0
+        );
+    }
+    return res;
+}
+
+pub fn evaluate_points_on_coset(
+    d_coefficients: &mut DeviceBuffer<Point>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+    coset_powers: &mut DeviceBuffer<Scalar>,
+) -> DeviceBuffer<Point> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len()).unwrap() };
+    unsafe {
+        evaluate_points_on_coset_cuda(
+            res.as_device_ptr(),
+            d_coefficients.as_device_ptr(),
+            d_domain.as_device_ptr(),
+            d_domain.len(),
+            d_coefficients.len(),
+            coset_powers.as_device_ptr(),
+            0
+        );
+    }
+    return res;
+}
+
+pub fn evaluate_points_on_coset_batch(
+    d_coefficients: &mut DeviceBuffer<Point>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+    batch_size: usize,
+    coset_powers: &mut DeviceBuffer<Scalar>,
+) -> DeviceBuffer<Point> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len() * batch_size).unwrap() };
+    unsafe {
+        evaluate_points_on_coset_batch_cuda(
+            res.as_device_ptr(),
+            d_coefficients.as_device_ptr(),
+            d_domain.as_device_ptr(),
+            d_domain.len(),
+            d_coefficients.len() / batch_size,
+            batch_size,
+            coset_powers.as_device_ptr(),
+            0
+        );
+    }
+    return res;
+}
+
+pub fn multp_vec(a: &mut [Point], b: &[Scalar], device_id: usize) {
+    assert_eq!(a.len(), b.len());
+    unsafe {
+        vec_mod_mult_point(
+            a as *mut _ as *mut Point,
+            b as *const _ as *const Scalar,
+            a.len(),
+            device_id,
+        );
+    }
+}
+
+pub fn mult_sc_vec(a: &mut [Scalar], b: &[Scalar], device_id: usize) {
+    assert_eq!(a.len(), b.len());
+    unsafe {
+        vec_mod_mult_scalar(
+            a as *mut _ as *mut Scalar,
+            b as *const _ as *const Scalar,
+            a.len(),
+            device_id,
+        );
+    }
+}
+
+// Multiply a matrix by a scalar:
+//  `a` - flattenned matrix;
+//  `b` - vector to multiply `a` by;
+pub fn mult_matrix_by_vec(a: &[Scalar], b: &[Scalar], device_id: usize) -> Vec<Scalar> {
+    let mut c = Vec::with_capacity(b.len());
+    for i in 0..b.len() {
+        c.push(Scalar::zero());
+    }
+    unsafe {
+        matrix_vec_mod_mult(
+            a as *const _ as *const Scalar,
+            b as *const _ as *const Scalar,
+            c.as_mut_slice() as *mut _ as *mut Scalar,
+            b.len(),
+            device_id,
+        );
+    }
+    c
+}
+
+pub fn clone_buffer<T: DeviceCopy>(buf: &mut DeviceBuffer<T>) -> DeviceBuffer<T> {
+    let mut buf_cpy = unsafe { DeviceBuffer::uninitialized(buf.len()).unwrap() };
+    unsafe { buf_cpy.copy_from(buf) };
+    return buf_cpy;
+}
+
+pub fn get_rng(seed: Option<u64>) -> Box<dyn RngCore> {
+    let rng: Box<dyn RngCore> = match seed {
+        Some(seed) => Box::new(StdRng::seed_from_u64(seed)),
+        None => Box::new(rand::thread_rng()),
+    };
+    rng
+}
+
+fn set_up_device() {
+    // Set up the context, load the module, and create a stream to run kernels in.
+    rustacuda::init(CudaFlags::empty()).unwrap();
+    let device = Device::get_device(0).unwrap();
+    let _ctx = Context::create_and_push(ContextFlags::MAP_HOST | ContextFlags::SCHED_AUTO, device).unwrap();
+}
+
+pub fn generate_random_points(
+    count: usize,
+    mut rng: Box<dyn RngCore>,
+) -> Vec<PointAffineNoInfinity> {
+    (0..count)
+        .map(|_| Point::from_ark(G1Projective_BLS12_377::rand(&mut rng)).to_xy_strip_z())
+        .collect()
+}
+
+pub fn generate_random_points_proj(count: usize, mut rng: Box<dyn RngCore>) -> Vec<Point> {
+    (0..count)
+        .map(|_| Point::from_ark(G1Projective_BLS12_377::rand(&mut rng)))
+        .collect()
+}
+
+pub fn generate_random_scalars(count: usize, mut rng: Box<dyn RngCore>) -> Vec<Scalar> {
+    (0..count)
+        .map(|_| Scalar::from_ark(Fr_BLS12_377::rand(&mut rng).into_repr()))
+        .collect()
+}
+
+pub fn set_up_points(test_size: usize, log_domain_size: usize, inverse: bool) -> (Vec<Point>, DeviceBuffer<Point>, DeviceBuffer<Scalar>) {
+    set_up_device();
+
+    let d_domain = build_domain(1 << log_domain_size, log_domain_size, inverse);
+
+    let seed = Some(0); // fix the rng to get two equal scalar 
+    let vector = generate_random_points_proj(test_size, get_rng(seed));
+    let mut vector_mut = vector.clone();
+
+    let mut d_vector = DeviceBuffer::from_slice(&vector[..]).unwrap();
+    (vector_mut, d_vector, d_domain)
+}
+
+pub fn set_up_scalars(test_size: usize, log_domain_size: usize, inverse: bool) -> (Vec<Scalar>, DeviceBuffer<Scalar>, DeviceBuffer<Scalar>) {
+    set_up_device();
+
+    let d_domain = build_domain(1 << log_domain_size, log_domain_size, inverse);
+
+    let seed = Some(0); // fix the rng to get two equal scalars
+    let mut vector_mut = generate_random_scalars(test_size, get_rng(seed));
+
+    let mut d_vector = DeviceBuffer::from_slice(&vector_mut[..]).unwrap();
+    (vector_mut, d_vector, d_domain)
+}
+

bls12-377/src/lib.rs

@@ -0,0 +1,4 @@
+pub mod test_bls12_377;
+pub mod basic_structs;
+pub mod from_cuda;
+pub mod curve_structs;

bls12-377/src/test_bls12_377.rs

@@ -0,0 +1,816 @@
+use std::ffi::{c_int, c_uint};
+use ark_std::UniformRand;
+use rand::{rngs::StdRng, RngCore, SeedableRng};
+use rustacuda::CudaFlags;
+use rustacuda::memory::DeviceBox;
+use rustacuda::prelude::{DeviceBuffer, Device, ContextFlags, Context};
+use rustacuda_core::DevicePointer;
+use std::mem::transmute;
+pub use crate::basic_structs::scalar::ScalarTrait;
+pub use crate::curve_structs::*;
+use icicle_core::utils::{u32_vec_to_u64_vec, u64_vec_to_u32_vec};
+use std::marker::PhantomData;
+use std::convert::TryInto;
+use ark_bls12_377::{Fq as Fq_BLS12_377, Fr as Fr_BLS12_377, G1Affine as G1Affine_BLS12_377, G1Projective as G1Projective_BLS12_377};
+use ark_ec::AffineCurve;
+use ark_ff::{BigInteger384, BigInteger256, PrimeField};
+use rustacuda::memory::{CopyDestination, DeviceCopy};
+
+
+impl Scalar {
+    pub fn to_biginteger254(&self) -> BigInteger256 {
+        BigInteger256::new(u32_vec_to_u64_vec(&self.limbs()).try_into().unwrap())
+    }
+
+    pub fn to_ark(&self) -> BigInteger256 {
+        BigInteger256::new(u32_vec_to_u64_vec(&self.limbs()).try_into().unwrap())
+    }
+
+    pub fn from_biginteger256(ark: BigInteger256) -> Self {
+        Self{ value: u64_vec_to_u32_vec(&ark.0).try_into().unwrap(), phantom : PhantomData}
+    }
+
+    pub fn to_biginteger256_transmute(&self) -> BigInteger256 {
+        unsafe { transmute(*self) }
+    }
+
+    pub fn from_biginteger_transmute(v: BigInteger256) -> Scalar {
+        Scalar{ value: unsafe{ transmute(v)}, phantom : PhantomData }
+    }
+
+    pub fn to_ark_transmute(&self) -> Fr_BLS12_377 {
+        unsafe { std::mem::transmute(*self) }
+    }
+
+    pub fn from_ark_transmute(v: &Fr_BLS12_377) -> Scalar {
+        unsafe { std::mem::transmute_copy(v) }
+    }
+
+    pub fn to_ark_mod_p(&self) -> Fr_BLS12_377 {
+        Fr_BLS12_377::new(BigInteger256::new(u32_vec_to_u64_vec(&self.limbs()).try_into().unwrap()))
+    }
+
+    pub fn to_ark_repr(&self) -> Fr_BLS12_377 {
+        Fr_BLS12_377::from_repr(BigInteger256::new(u32_vec_to_u64_vec(&self.limbs()).try_into().unwrap())).unwrap()
+    }
+
+    pub fn from_ark(v: BigInteger256) -> Scalar {
+        Self { value : u64_vec_to_u32_vec(&v.0).try_into().unwrap(), phantom: PhantomData}
+    }
+
+}
+
+impl Base {
+    pub fn to_ark(&self) -> BigInteger384 {
+        BigInteger384::new(u32_vec_to_u64_vec(&self.limbs()).try_into().unwrap())
+    }
+
+    pub fn from_ark(ark: BigInteger384) -> Self {
+        Self::from_limbs(&u64_vec_to_u32_vec(&ark.0))
+    }
+}
+
+
+impl Point {
+    pub fn to_ark(&self) -> G1Projective_BLS12_377 {
+        self.to_ark_affine().into_projective()
+    }
+
+    pub fn to_ark_affine(&self) -> G1Affine_BLS12_377 {
+        //TODO: generic conversion
+        use ark_ff::Field;
+        use std::ops::Mul;
+        let proj_x_field = Fq_BLS12_377::from_le_bytes_mod_order(&self.x.to_bytes_le());
+        let proj_y_field = Fq_BLS12_377::from_le_bytes_mod_order(&self.y.to_bytes_le());
+        let proj_z_field = Fq_BLS12_377::from_le_bytes_mod_order(&self.z.to_bytes_le());
+        let inverse_z = proj_z_field.inverse().unwrap();
+        let aff_x = proj_x_field.mul(inverse_z);
+        let aff_y = proj_y_field.mul(inverse_z);
+        G1Affine_BLS12_377::new(aff_x, aff_y, false)
+    }
+
+    pub fn from_ark(ark: G1Projective_BLS12_377) -> Point {
+        use ark_ff::Field;
+        let z_inv = ark.z.inverse().unwrap();
+        let z_invsq = z_inv * z_inv;
+        let z_invq3 = z_invsq * z_inv;
+        Point {
+            x: Base::from_ark((ark.x * z_invsq).into_repr()),
+            y: Base::from_ark((ark.y * z_invq3).into_repr()),
+            z: Base::one(),
+        }
+    }
+}
+
+impl PointAffineNoInfinity {
+
+    pub fn to_ark(&self) -> G1Affine_BLS12_377 {
+        G1Affine_BLS12_377::new(Fq_BLS12_377::new(self.x.to_ark()), Fq_BLS12_377::new(self.y.to_ark()), false)
+    }
+
+    pub fn to_ark_repr(&self) -> G1Affine_BLS12_377 {
+        G1Affine_BLS12_377::new(
+            Fq_BLS12_377::from_repr(self.x.to_ark()).unwrap(),
+            Fq_BLS12_377::from_repr(self.y.to_ark()).unwrap(),
+            false,
+        )
+    }
+
+    pub fn from_ark(p: &G1Affine_BLS12_377) -> Self {
+        PointAffineNoInfinity {
+            x: Base::from_ark(p.x.into_repr()),
+            y: Base::from_ark(p.y.into_repr()),
+        }
+    }
+}
+
+impl Point {
+    pub fn to_affine(&self) -> PointAffineNoInfinity {
+        let ark_affine = self.to_ark_affine();
+        PointAffineNoInfinity {
+            x: Base::from_ark(ark_affine.x.into_repr()),
+            y: Base::from_ark(ark_affine.y.into_repr()),
+        }
+    }
+}
+
+
+#[cfg(test)]
+pub(crate) mod tests_bls12_377 {
+    use std::ops::Add;
+    use ark_bls12_377::{Fr, G1Affine, G1Projective};
+    use ark_ec::{msm::VariableBaseMSM, AffineCurve, ProjectiveCurve};
+    use ark_ff::{FftField, Field, Zero, PrimeField};
+    use ark_std::UniformRand;
+    use rustacuda::prelude::{DeviceBuffer, CopyDestination};
+    use crate::curve_structs::{Point, Scalar, Base};
+    use crate::basic_structs::scalar::ScalarTrait;
+    use crate::from_cuda::{generate_random_points, get_rng, generate_random_scalars, msm, msm_batch, set_up_scalars, commit, commit_batch, ntt, intt, generate_random_points_proj, ecntt, iecntt, ntt_batch, ecntt_batch, iecntt_batch, intt_batch, reverse_order_scalars_batch, interpolate_scalars_batch, set_up_points, reverse_order_points, interpolate_points, reverse_order_points_batch, interpolate_points_batch, evaluate_scalars, interpolate_scalars, reverse_order_scalars, evaluate_points, build_domain, evaluate_scalars_on_coset, evaluate_points_on_coset, mult_matrix_by_vec, mult_sc_vec, multp_vec,evaluate_scalars_batch, evaluate_points_batch, evaluate_scalars_on_coset_batch, evaluate_points_on_coset_batch};
+
+    fn random_points_ark_proj(nof_elements: usize) -> Vec<G1Projective> {
+        let mut rng = ark_std::rand::thread_rng();
+        let mut points_ga: Vec<G1Projective> = Vec::new();
+        for _ in 0..nof_elements {
+            let aff = G1Projective::rand(&mut rng);
+            points_ga.push(aff);
+        }
+        points_ga
+    }
+
+    fn ecntt_arc_naive(
+        points: &Vec<G1Projective>,
+        size: usize,
+        inverse: bool,
+    ) -> Vec<G1Projective> {
+        let mut result: Vec<G1Projective> = Vec::new();
+        for _ in 0..size {
+            result.push(G1Projective::zero());
+        }
+        let rou: Fr;
+        if !inverse {
+            rou = Fr::get_root_of_unity(size).unwrap();
+        } else {
+            rou = Fr::inverse(&Fr::get_root_of_unity(size).unwrap()).unwrap();
+        }
+        for k in 0..size {
+            for l in 0..size {
+                let pow: [u64; 1] = [(l * k).try_into().unwrap()];
+                let mul_rou = Fr::pow(&rou, &pow);
+                result[k] = result[k].add(points[l].into_affine().mul(mul_rou));
+            }
+        }
+        if inverse {
+            let size2 = size as u64;
+            for k in 0..size {
+                let multfactor = Fr::inverse(&Fr::from(size2)).unwrap();
+                result[k] = result[k].into_affine().mul(multfactor);
+            }
+        }
+        return result;
+    }
+
+    fn check_eq(points: &Vec<G1Projective>, points2: &Vec<G1Projective>) -> bool {
+        let mut eq = true;
+        for i in 0..points.len() {
+            if points2[i].ne(&points[i]) {
+                eq = false;
+                break;
+            }
+        }
+        return eq;
+    }
+
+    fn test_naive_ark_ecntt(size: usize) {
+        let points = random_points_ark_proj(size);
+        let result1: Vec<G1Projective> = ecntt_arc_naive(&points, size, false);
+        let result2: Vec<G1Projective> = ecntt_arc_naive(&result1, size, true);
+        assert!(!check_eq(&result2, &result1));
+        assert!(check_eq(&result2, &points));
+    }
+
+    #[test]
+    fn test_msm() {
+        let test_sizes = [6, 9];
+
+        for pow2 in test_sizes {
+            let count = 1 << pow2;
+            let seed = None; // set Some to provide seed
+            let points = generate_random_points(count, get_rng(seed));
+            let scalars = generate_random_scalars(count, get_rng(seed));
+
+            let msm_result = msm(&points, &scalars, 0);
+
+            let point_r_ark: Vec<_> = points.iter().map(|x| x.to_ark_repr()).collect();
+            let scalars_r_ark: Vec<_> = scalars.iter().map(|x| x.to_ark()).collect();
+
+            let msm_result_ark = VariableBaseMSM::multi_scalar_mul(&point_r_ark, &scalars_r_ark);
+
+            assert_eq!(msm_result.to_ark_affine(), msm_result_ark);
+            assert_eq!(msm_result.to_ark(), msm_result_ark);
+            assert_eq!(
+                msm_result.to_ark_affine(),
+                Point::from_ark(msm_result_ark).to_ark_affine()
+            );
+        }
+    }
+
+    #[test]
+    fn test_batch_msm() {
+        for batch_pow2 in [2, 4] {
+            for pow2 in [4, 6] {
+                let msm_size = 1 << pow2;
+                let batch_size = 1 << batch_pow2;
+                let seed = None; // set Some to provide seed
+                let points_batch = generate_random_points(msm_size * batch_size, get_rng(seed));
+                let scalars_batch = generate_random_scalars(msm_size * batch_size, get_rng(seed));
+
+                let point_r_ark: Vec<_> = points_batch.iter().map(|x| x.to_ark_repr()).collect();
+                let scalars_r_ark: Vec<_> = scalars_batch.iter().map(|x| x.to_ark()).collect();
+
+                let expected: Vec<_> = point_r_ark
+                    .chunks(msm_size)
+                    .zip(scalars_r_ark.chunks(msm_size))
+                    .map(|p| Point::from_ark(VariableBaseMSM::multi_scalar_mul(p.0, p.1)))
+                    .collect();
+
+                let result = msm_batch(&points_batch, &scalars_batch, batch_size, 0);
+
+                assert_eq!(result, expected);
+            }
+        }
+    }
+
+    #[test]
+    fn test_commit() {
+        let test_size = 1 << 8;
+        let seed = Some(0);
+        let (mut scalars, mut d_scalars, _) = set_up_scalars(test_size, 0, false);
+        let mut points = generate_random_points(test_size, get_rng(seed));
+        let mut d_points = DeviceBuffer::from_slice(&points[..]).unwrap();
+
+        let msm_result = msm(&points, &scalars, 0);
+        let mut d_commit_result = commit(&mut d_points, &mut d_scalars);
+        let mut h_commit_result = Point::zero();
+        d_commit_result.copy_to(&mut h_commit_result).unwrap();
+
+        assert_eq!(msm_result, h_commit_result);
+        assert_ne!(msm_result, Point::zero());
+        assert_ne!(h_commit_result, Point::zero());
+    }
+
+    #[test]
+    fn test_batch_commit() {
+        let batch_size = 4;
+        let test_size = 1 << 12;
+        let seed = Some(0);
+        let (scalars, mut d_scalars, _) = set_up_scalars(test_size * batch_size, 0, false);
+        let points = generate_random_points(test_size * batch_size, get_rng(seed));
+        let mut d_points = DeviceBuffer::from_slice(&points[..]).unwrap();
+
+        let msm_result = msm_batch(&points, &scalars, batch_size, 0);
+        let mut d_commit_result = commit_batch(&mut d_points, &mut d_scalars, batch_size);
+        let mut h_commit_result: Vec<Point> = (0..batch_size).map(|_| Point::zero()).collect();
+        d_commit_result.copy_to(&mut h_commit_result[..]).unwrap();
+
+        assert_eq!(msm_result, h_commit_result);
+        for h in h_commit_result {
+            assert_ne!(h, Point::zero());
+        }
+    }
+
+    #[test]
+    fn test_ntt() {
+        //NTT
+        let seed = None; //some value to fix the rng
+        let test_size = 1 << 3;
+
+        let scalars = generate_random_scalars(test_size, get_rng(seed));
+
+        let mut ntt_result = scalars.clone();
+        ntt(&mut ntt_result, 0);
+
+        assert_ne!(ntt_result, scalars);
+
+        let mut intt_result = ntt_result.clone();
+
+        intt(&mut intt_result, 0);
+
+        assert_eq!(intt_result, scalars);
+
+        //ECNTT
+        let points_proj = generate_random_points_proj(test_size, get_rng(seed));
+
+        test_naive_ark_ecntt(test_size);
+
+        assert!(points_proj[0].to_ark().into_affine().is_on_curve());
+
+        //naive ark
+        let points_proj_ark = points_proj
+            .iter()
+            .map(|p| p.to_ark())
+            .collect::<Vec<G1Projective>>();
+
+        let ecntt_result_naive = ecntt_arc_naive(&points_proj_ark, points_proj_ark.len(), false);
+
+        let iecntt_result_naive = ecntt_arc_naive(&ecntt_result_naive, points_proj_ark.len(), true);
+
+        assert_eq!(points_proj_ark, iecntt_result_naive);
+
+        //ingo gpu
+        let mut ecntt_result = points_proj.to_vec();
+        ecntt(&mut ecntt_result, 0);
+
+        assert_ne!(ecntt_result, points_proj);
+
+        let mut iecntt_result = ecntt_result.clone();
+        iecntt(&mut iecntt_result, 0);
+
+        assert_eq!(
+            iecntt_result_naive,
+            points_proj
+                .iter()
+                .map(|p| p.to_ark_affine())
+                .collect::<Vec<G1Affine>>()
+        );
+        assert_eq!(
+            iecntt_result
+                .iter()
+                .map(|p| p.to_ark_affine())
+                .collect::<Vec<G1Affine>>(),
+            points_proj
+                .iter()
+                .map(|p| p.to_ark_affine())
+                .collect::<Vec<G1Affine>>()
+        );
+    }
+
+    #[test]
+    fn test_ntt_batch() {
+        //NTT
+        let seed = None; //some value to fix the rng
+        let test_size = 1 << 5;
+        let batches = 4;
+
+        let scalars_batch: Vec<Scalar> =
+            generate_random_scalars(test_size * batches, get_rng(seed));
+
+        let mut scalar_vec_of_vec: Vec<Vec<Scalar>> = Vec::new();
+
+        for i in 0..batches {
+            scalar_vec_of_vec.push(scalars_batch[i * test_size..(i + 1) * test_size].to_vec());
+        }
+
+        let mut ntt_result = scalars_batch.clone();
+
+        // do batch ntt
+        ntt_batch(&mut ntt_result, test_size, 0);
+
+        let mut ntt_result_vec_of_vec = Vec::new();
+
+        // do ntt for every chunk
+        for i in 0..batches {
+            ntt_result_vec_of_vec.push(scalar_vec_of_vec[i].clone());
+            ntt(&mut ntt_result_vec_of_vec[i], 0);
+        }
+
+        // check that the ntt of each vec of scalars is equal to the intt of the specific batch
+        for i in 0..batches {
+            assert_eq!(
+                ntt_result_vec_of_vec[i],
+                ntt_result[i * test_size..(i + 1) * test_size]
+            );
+        }
+
+        // check that ntt output is different from input
+        assert_ne!(ntt_result, scalars_batch);
+
+        let mut intt_result = ntt_result.clone();
+
+        // do batch intt
+        intt_batch(&mut intt_result, test_size, 0);
+
+        let mut intt_result_vec_of_vec = Vec::new();
+
+        // do intt for every chunk
+        for i in 0..batches {
+            intt_result_vec_of_vec.push(ntt_result_vec_of_vec[i].clone());
+            intt(&mut intt_result_vec_of_vec[i], 0);
+        }
+
+        // check that the intt of each vec of scalars is equal to the intt of the specific batch
+        for i in 0..batches {
+            assert_eq!(
+                intt_result_vec_of_vec[i],
+                intt_result[i * test_size..(i + 1) * test_size]
+            );
+        }
+
+        assert_eq!(intt_result, scalars_batch);
+
+        // //ECNTT
+        let points_proj = generate_random_points_proj(test_size * batches, get_rng(seed));
+
+        let mut points_vec_of_vec: Vec<Vec<Point>> = Vec::new();
+
+        for i in 0..batches {
+            points_vec_of_vec.push(points_proj[i * test_size..(i + 1) * test_size].to_vec());
+        }
+
+        let mut ntt_result_points = points_proj.clone();
+
+        // do batch ecintt
+        ecntt_batch(&mut ntt_result_points, test_size, 0);
+
+        let mut ntt_result_points_vec_of_vec = Vec::new();
+
+        for i in 0..batches {
+            ntt_result_points_vec_of_vec.push(points_vec_of_vec[i].clone());
+            ecntt(&mut ntt_result_points_vec_of_vec[i], 0);
+        }
+
+        for i in 0..batches {
+            assert_eq!(
+                ntt_result_points_vec_of_vec[i],
+                ntt_result_points[i * test_size..(i + 1) * test_size]
+            );
+        }
+
+        assert_ne!(ntt_result_points, points_proj);
+
+        let mut intt_result_points = ntt_result_points.clone();
+
+        // do batch ecintt
+        iecntt_batch(&mut intt_result_points, test_size, 0);
+
+        let mut intt_result_points_vec_of_vec = Vec::new();
+
+        // do ecintt for every chunk
+        for i in 0..batches {
+            intt_result_points_vec_of_vec.push(ntt_result_points_vec_of_vec[i].clone());
+            iecntt(&mut intt_result_points_vec_of_vec[i], 0);
+        }
+
+        // check that the ecintt of each vec of scalars is equal to the intt of the specific batch
+        for i in 0..batches {
+            assert_eq!(
+                intt_result_points_vec_of_vec[i],
+                intt_result_points[i * test_size..(i + 1) * test_size]
+            );
+        }
+
+        assert_eq!(intt_result_points, points_proj);
+    }
+
+    #[test]
+    fn test_scalar_interpolation() {
+        let log_test_size = 7;
+        let test_size = 1 << log_test_size;
+        let (mut evals_mut, mut d_evals, mut d_domain) = set_up_scalars(test_size, log_test_size, true);
+
+        reverse_order_scalars(&mut d_evals);
+        let mut d_coeffs = interpolate_scalars(&mut d_evals, &mut d_domain);
+        intt(&mut evals_mut, 0);
+        let mut h_coeffs: Vec<Scalar> = (0..test_size).map(|_| Scalar::zero()).collect();
+        d_coeffs.copy_to(&mut h_coeffs[..]).unwrap();
+
+        assert_eq!(h_coeffs, evals_mut);
+    }
+
+    #[test]
+    fn test_scalar_batch_interpolation() {
+        let batch_size = 4;
+        let log_test_size = 10;
+        let test_size = 1 << log_test_size;
+        let (mut evals_mut, mut d_evals, mut d_domain) = set_up_scalars(test_size * batch_size, log_test_size, true);
+
+        reverse_order_scalars_batch(&mut d_evals, batch_size);
+        let mut d_coeffs = interpolate_scalars_batch(&mut d_evals, &mut d_domain, batch_size);
+        intt_batch(&mut evals_mut, test_size, 0);
+        let mut h_coeffs: Vec<Scalar> = (0..test_size * batch_size).map(|_| Scalar::zero()).collect();
+        d_coeffs.copy_to(&mut h_coeffs[..]).unwrap();
+
+        assert_eq!(h_coeffs, evals_mut);
+    }
+
+    #[test]
+    fn test_point_interpolation() {
+        let log_test_size = 6;
+        let test_size = 1 << log_test_size;
+        let (mut evals_mut, mut d_evals, mut d_domain) = set_up_points(test_size, log_test_size, true);
+
+        reverse_order_points(&mut d_evals);
+        let mut d_coeffs = interpolate_points(&mut d_evals, &mut d_domain);
+        iecntt(&mut evals_mut[..], 0);
+        let mut h_coeffs: Vec<Point> = (0..test_size).map(|_| Point::zero()).collect();
+        d_coeffs.copy_to(&mut h_coeffs[..]).unwrap();
+        
+        assert_eq!(h_coeffs, *evals_mut);
+        for h in h_coeffs.iter() {
+            assert_ne!(*h, Point::zero());
+        }
+    }
+
+    #[test]
+    fn test_point_batch_interpolation() {
+        let batch_size = 4;
+        let log_test_size = 6;
+        let test_size = 1 << log_test_size;
+        let (mut evals_mut, mut d_evals, mut d_domain) = set_up_points(test_size * batch_size, log_test_size, true);
+
+        reverse_order_points_batch(&mut d_evals, batch_size);
+        let mut d_coeffs = interpolate_points_batch(&mut d_evals, &mut d_domain, batch_size);
+        iecntt_batch(&mut evals_mut[..], test_size, 0);
+        let mut h_coeffs: Vec<Point> = (0..test_size * batch_size).map(|_| Point::zero()).collect();
+        d_coeffs.copy_to(&mut h_coeffs[..]).unwrap();
+        
+        assert_eq!(h_coeffs, *evals_mut);
+        for h in h_coeffs.iter() {
+            assert_ne!(*h, Point::zero());
+        }
+    }
+
+    #[test]
+    fn test_scalar_evaluation() {
+        let log_test_domain_size = 8;
+        let coeff_size = 1 << 6;
+        let (h_coeffs, mut d_coeffs, mut d_domain) = set_up_scalars(coeff_size, log_test_domain_size, false);
+        let (_, _, mut d_domain_inv) = set_up_scalars(0, log_test_domain_size, true);
+
+        let mut d_evals = evaluate_scalars(&mut d_coeffs, &mut d_domain);
+        let mut d_coeffs_domain = interpolate_scalars(&mut d_evals, &mut d_domain_inv);
+        let mut h_coeffs_domain: Vec<Scalar> = (0..1 << log_test_domain_size).map(|_| Scalar::zero()).collect();
+        d_coeffs_domain.copy_to(&mut h_coeffs_domain[..]).unwrap();
+
+        assert_eq!(h_coeffs, h_coeffs_domain[..coeff_size]);
+        for i in coeff_size.. (1 << log_test_domain_size) {
+            assert_eq!(Scalar::zero(), h_coeffs_domain[i]);
+        }
+    }
+
+    #[test]
+    fn test_scalar_batch_evaluation() {
+        let batch_size = 6;
+        let log_test_domain_size = 8;
+        let domain_size = 1 << log_test_domain_size;
+        let coeff_size = 1 << 6;
+        let (h_coeffs, mut d_coeffs, mut d_domain) = set_up_scalars(coeff_size * batch_size, log_test_domain_size, false);
+        let (_, _, mut d_domain_inv) = set_up_scalars(0, log_test_domain_size, true);
+
+        let mut d_evals = evaluate_scalars_batch(&mut d_coeffs, &mut d_domain, batch_size);
+        let mut d_coeffs_domain = interpolate_scalars_batch(&mut d_evals, &mut d_domain_inv, batch_size);
+        let mut h_coeffs_domain: Vec<Scalar> = (0..domain_size * batch_size).map(|_| Scalar::zero()).collect();
+        d_coeffs_domain.copy_to(&mut h_coeffs_domain[..]).unwrap();
+
+        for j in 0..batch_size {
+            assert_eq!(h_coeffs[j * coeff_size..(j + 1) * coeff_size], h_coeffs_domain[j * domain_size..j * domain_size + coeff_size]);
+            for i in coeff_size..domain_size {
+                assert_eq!(Scalar::zero(), h_coeffs_domain[j * domain_size + i]);
+            }
+        }
+    }
+
+    #[test]
+    fn test_point_evaluation() {
+        let log_test_domain_size = 7;
+        let coeff_size = 1 << 7;
+        let (h_coeffs, mut d_coeffs, mut d_domain) = set_up_points(coeff_size, log_test_domain_size, false);
+        let (_, _, mut d_domain_inv) = set_up_points(0, log_test_domain_size, true);
+
+        let mut d_evals = evaluate_points(&mut d_coeffs, &mut d_domain);
+        let mut d_coeffs_domain = interpolate_points(&mut d_evals, &mut d_domain_inv);
+        let mut h_coeffs_domain: Vec<Point> = (0..1 << log_test_domain_size).map(|_| Point::zero()).collect();
+        d_coeffs_domain.copy_to(&mut h_coeffs_domain[..]).unwrap();
+
+        assert_eq!(h_coeffs[..], h_coeffs_domain[..coeff_size]);
+        for i in coeff_size..(1 << log_test_domain_size) {
+            assert_eq!(Point::zero(), h_coeffs_domain[i]);
+        }
+        for i in 0..coeff_size {
+            assert_ne!(h_coeffs_domain[i], Point::zero());
+        }
+    }
+
+    #[test]
+    fn test_point_batch_evaluation() {
+        let batch_size = 4;
+        let log_test_domain_size = 6;
+        let domain_size = 1 << log_test_domain_size;
+        let coeff_size = 1 << 5;
+        let (h_coeffs, mut d_coeffs, mut d_domain) = set_up_points(coeff_size * batch_size, log_test_domain_size, false);
+        let (_, _, mut d_domain_inv) = set_up_points(0, log_test_domain_size, true);
+
+        let mut d_evals = evaluate_points_batch(&mut d_coeffs, &mut d_domain, batch_size);
+        let mut d_coeffs_domain = interpolate_points_batch(&mut d_evals, &mut d_domain_inv, batch_size);
+        let mut h_coeffs_domain: Vec<Point> = (0..domain_size * batch_size).map(|_| Point::zero()).collect();
+        d_coeffs_domain.copy_to(&mut h_coeffs_domain[..]).unwrap();
+
+        for j in 0..batch_size {
+            assert_eq!(h_coeffs[j * coeff_size..(j + 1) * coeff_size], h_coeffs_domain[j * domain_size..(j * domain_size + coeff_size)]);
+            for i in coeff_size..domain_size {
+                assert_eq!(Point::zero(), h_coeffs_domain[j * domain_size + i]);
+            }
+            for i in j * domain_size..(j * domain_size + coeff_size) {
+                assert_ne!(h_coeffs_domain[i], Point::zero());
+            }
+        }
+    }
+
+    #[test]
+    fn test_scalar_evaluation_on_trivial_coset() {
+        // checks that the evaluations on the subgroup is the same as on the coset generated by 1
+        let log_test_domain_size = 8;
+        let coeff_size = 1 << 6;
+        let (_, mut d_coeffs, mut d_domain) = set_up_scalars(coeff_size, log_test_domain_size, false);
+        let (_, _, mut d_domain_inv) = set_up_scalars(coeff_size, log_test_domain_size, true);
+        let mut d_trivial_coset_powers = build_domain(1 << log_test_domain_size, 0, false);
+
+        let mut d_evals = evaluate_scalars(&mut d_coeffs, &mut d_domain);
+        let mut h_coeffs: Vec<Scalar> = (0..1 << log_test_domain_size).map(|_| Scalar::zero()).collect();
+        d_evals.copy_to(&mut h_coeffs[..]).unwrap();
+        let mut d_evals_coset = evaluate_scalars_on_coset(&mut d_coeffs, &mut d_domain, &mut d_trivial_coset_powers);
+        let mut h_evals_coset: Vec<Scalar> = (0..1 << log_test_domain_size).map(|_| Scalar::zero()).collect();
+        d_evals_coset.copy_to(&mut h_evals_coset[..]).unwrap();
+
+        assert_eq!(h_coeffs, h_evals_coset);
+    }
+
+    #[test]
+    fn test_scalar_evaluation_on_coset() {
+        // checks that evaluating a polynomial on a subgroup and its coset is the same as evaluating on a 2x larger subgroup 
+        let log_test_size = 8;
+        let test_size = 1 << log_test_size;
+        let (_, mut d_coeffs, mut d_domain) = set_up_scalars(test_size, log_test_size, false);
+        let (_, _, mut d_large_domain) = set_up_scalars(0, log_test_size + 1, false);
+        let mut d_coset_powers = build_domain(test_size, log_test_size + 1, false);
+
+        let mut d_evals_large = evaluate_scalars(&mut d_coeffs, &mut d_large_domain);
+        let mut h_evals_large: Vec<Scalar> = (0..2 * test_size).map(|_| Scalar::zero()).collect();
+        d_evals_large.copy_to(&mut h_evals_large[..]).unwrap();
+        let mut d_evals = evaluate_scalars(&mut d_coeffs, &mut d_domain);
+        let mut h_evals: Vec<Scalar> = (0..test_size).map(|_| Scalar::zero()).collect();
+        d_evals.copy_to(&mut h_evals[..]).unwrap();
+        let mut d_evals_coset = evaluate_scalars_on_coset(&mut d_coeffs, &mut d_domain, &mut d_coset_powers);
+        let mut h_evals_coset: Vec<Scalar> = (0..test_size).map(|_| Scalar::zero()).collect();
+        d_evals_coset.copy_to(&mut h_evals_coset[..]).unwrap();
+
+        assert_eq!(h_evals[..], h_evals_large[..test_size]);
+        assert_eq!(h_evals_coset[..], h_evals_large[test_size..2 * test_size]);
+    }
+
+    #[test]
+    fn test_scalar_batch_evaluation_on_coset() {
+        // checks that evaluating a polynomial on a subgroup and its coset is the same as evaluating on a 2x larger subgroup 
+        let batch_size = 4;
+        let log_test_size = 6;
+        let test_size = 1 << log_test_size;
+        let (_, mut d_coeffs, mut d_domain) = set_up_scalars(test_size * batch_size, log_test_size, false);
+        let (_, _, mut d_large_domain) = set_up_scalars(0, log_test_size + 1, false);
+        let mut d_coset_powers = build_domain(test_size, log_test_size + 1, false);
+
+        let mut d_evals_large = evaluate_scalars_batch(&mut d_coeffs, &mut d_large_domain, batch_size);
+        let mut h_evals_large: Vec<Scalar> = (0..2 * test_size * batch_size).map(|_| Scalar::zero()).collect();
+        d_evals_large.copy_to(&mut h_evals_large[..]).unwrap();
+        let mut d_evals = evaluate_scalars_batch(&mut d_coeffs, &mut d_domain, batch_size);
+        let mut h_evals: Vec<Scalar> = (0..test_size * batch_size).map(|_| Scalar::zero()).collect();
+        d_evals.copy_to(&mut h_evals[..]).unwrap();
+        let mut d_evals_coset = evaluate_scalars_on_coset_batch(&mut d_coeffs, &mut d_domain, batch_size, &mut d_coset_powers);
+        let mut h_evals_coset: Vec<Scalar> = (0..test_size * batch_size).map(|_| Scalar::zero()).collect();
+        d_evals_coset.copy_to(&mut h_evals_coset[..]).unwrap();
+
+        for i in 0..batch_size {
+            assert_eq!(h_evals_large[2 * i * test_size..(2 * i + 1) * test_size], h_evals[i * test_size..(i + 1) * test_size]);
+            assert_eq!(h_evals_large[(2 * i + 1) * test_size..(2 * i + 2) * test_size], h_evals_coset[i * test_size..(i + 1) * test_size]);
+        }
+    }
+
+    #[test]
+    fn test_point_evaluation_on_coset() {
+        // checks that evaluating a polynomial on a subgroup and its coset is the same as evaluating on a 2x larger subgroup 
+        let log_test_size = 8;
+        let test_size = 1 << log_test_size;
+        let (_, mut d_coeffs, mut d_domain) = set_up_points(test_size, log_test_size, false);
+        let (_, _, mut d_large_domain) = set_up_points(0, log_test_size + 1, false);
+        let mut d_coset_powers = build_domain(test_size, log_test_size + 1, false);
+
+        let mut d_evals_large = evaluate_points(&mut d_coeffs, &mut d_large_domain);
+        let mut h_evals_large: Vec<Point> = (0..2 * test_size).map(|_| Point::zero()).collect();
+        d_evals_large.copy_to(&mut h_evals_large[..]).unwrap();
+        let mut d_evals = evaluate_points(&mut d_coeffs, &mut d_domain);
+        let mut h_evals: Vec<Point> = (0..test_size).map(|_| Point::zero()).collect();
+        d_evals.copy_to(&mut h_evals[..]).unwrap();
+        let mut d_evals_coset = evaluate_points_on_coset(&mut d_coeffs, &mut d_domain, &mut d_coset_powers);
+        let mut h_evals_coset: Vec<Point> = (0..test_size).map(|_| Point::zero()).collect();
+        d_evals_coset.copy_to(&mut h_evals_coset[..]).unwrap();
+
+        assert_eq!(h_evals[..], h_evals_large[..test_size]);
+        assert_eq!(h_evals_coset[..], h_evals_large[test_size..2 * test_size]);
+        for i in 0..test_size {
+            assert_ne!(h_evals[i], Point::zero());
+            assert_ne!(h_evals_coset[i], Point::zero());
+            assert_ne!(h_evals_large[2 * i], Point::zero());
+            assert_ne!(h_evals_large[2 * i + 1], Point::zero());
+        }
+    }
+
+    #[test]
+    fn test_point_batch_evaluation_on_coset() {
+        // checks that evaluating a polynomial on a subgroup and its coset is the same as evaluating on a 2x larger subgroup 
+        let batch_size = 2;
+        let log_test_size = 6;
+        let test_size = 1 << log_test_size;
+        let (_, mut d_coeffs, mut d_domain) = set_up_points(test_size * batch_size, log_test_size, false);
+        let (_, _, mut d_large_domain) = set_up_points(0, log_test_size + 1, false);
+        let mut d_coset_powers = build_domain(test_size, log_test_size + 1, false);
+
+        let mut d_evals_large = evaluate_points_batch(&mut d_coeffs, &mut d_large_domain, batch_size);
+        let mut h_evals_large: Vec<Point> = (0..2 * test_size * batch_size).map(|_| Point::zero()).collect();
+        d_evals_large.copy_to(&mut h_evals_large[..]).unwrap();
+        let mut d_evals = evaluate_points_batch(&mut d_coeffs, &mut d_domain, batch_size);
+        let mut h_evals: Vec<Point> = (0..test_size * batch_size).map(|_| Point::zero()).collect();
+        d_evals.copy_to(&mut h_evals[..]).unwrap();
+        let mut d_evals_coset = evaluate_points_on_coset_batch(&mut d_coeffs, &mut d_domain, batch_size, &mut d_coset_powers);
+        let mut h_evals_coset: Vec<Point> = (0..test_size * batch_size).map(|_| Point::zero()).collect();
+        d_evals_coset.copy_to(&mut h_evals_coset[..]).unwrap();
+
+        for i in 0..batch_size {
+            assert_eq!(h_evals_large[2 * i * test_size..(2 * i + 1) * test_size], h_evals[i * test_size..(i + 1) * test_size]);
+            assert_eq!(h_evals_large[(2 * i + 1) * test_size..(2 * i + 2) * test_size], h_evals_coset[i * test_size..(i + 1) * test_size]);
+        }
+        for i in 0..test_size * batch_size {
+            assert_ne!(h_evals[i], Point::zero());
+            assert_ne!(h_evals_coset[i], Point::zero());
+            assert_ne!(h_evals_large[2 * i], Point::zero());
+            assert_ne!(h_evals_large[2 * i + 1], Point::zero());
+        }
+    }
+
+    // testing matrix multiplication by comparing the result of FFT with the naive multiplication by the DFT matrix
+    #[test]
+    fn test_matrix_multiplication() {
+        let seed = None; // some value to fix the rng
+        let test_size = 1 << 5;
+        let rou = Fr::get_root_of_unity(test_size).unwrap();
+        let matrix_flattened: Vec<Scalar> = (0..test_size).map(
+            |row_num| { (0..test_size).map( 
+                |col_num| {
+                    let pow: [u64; 1] = [(row_num * col_num).try_into().unwrap()];
+                    Scalar::from_ark(Fr::pow(&rou, &pow).into_repr())
+                }).collect::<Vec<Scalar>>()
+            }).flatten().collect::<Vec<_>>();
+        let vector: Vec<Scalar> = generate_random_scalars(test_size, get_rng(seed));
+
+        let result = mult_matrix_by_vec(&matrix_flattened, &vector, 0);
+        let mut ntt_result = vector.clone();
+        ntt(&mut ntt_result, 0);
+        
+        // we don't use the same roots of unity as arkworks, so the results are permutations
+        // of one another and the only guaranteed fixed scalars are the following ones:
+        assert_eq!(result[0], ntt_result[0]);
+        assert_eq!(result[test_size >> 1], ntt_result[test_size >> 1]);
+    }
+
+    #[test]
+    #[allow(non_snake_case)]
+    fn test_vec_scalar_mul() {
+        let mut intoo = [Scalar::one(), Scalar::one(), Scalar::zero()];
+        let expected = [Scalar::one(), Scalar::zero(), Scalar::zero()];
+        mult_sc_vec(&mut intoo, &expected, 0);
+        assert_eq!(intoo, expected);
+    }
+
+    #[test]
+    #[allow(non_snake_case)]
+    fn test_vec_point_mul() {
+        let dummy_one = Point {
+            x: Base::one(),
+            y: Base::one(),
+            z: Base::one(),
+        };
+
+        let mut inout = [dummy_one, dummy_one, Point::zero()];
+        let scalars = [Scalar::one(), Scalar::zero(), Scalar::zero()];
+        let expected = [dummy_one, Point::zero(), Point::zero()];
+        multp_vec(&mut inout, &scalars, 0);
+        assert_eq!(inout, expected);
+    }
+}

bls12-381/Cargo.toml

@@ -0,0 +1,34 @@
+[package]
+name = "bls12-381"
+version = "0.1.0"
+edition = "2021"
+authors = [ "Ingonyama" ]
+
+[dependencies]
+icicle-core = { path = "../icicle-core" }
+
+hex = "*"
+ark-std = "0.3.0"
+ark-ff = "0.3.0"
+ark-poly = "0.3.0"
+ark-ec = { version = "0.3.0", features = [ "parallel" ] }
+ark-bls12-381 = "0.3.0"
+
+serde = { version = "1.0", features = ["derive"] }
+serde_derive = "1.0"
+serde_cbor = "0.11.2"
+
+rustacuda = "0.1"
+rustacuda_core = "0.1"
+rustacuda_derive = "0.1"
+
+rand = "*" #TODO: move rand and ark dependencies to dev once random scalar/point generation is done "natively"
+
+[build-dependencies]
+cc = { version = "1.0", features = ["parallel"] }
+
+[dev-dependencies]
+"criterion" = "0.4.0"
+
+[features]
+g2 = []

bls12-381/build.rs

@@ -0,0 +1,36 @@
+use std::env;
+
+fn main() {
+    //TODO: check cargo features selected
+    //TODO: can conflict/duplicate with make ?
+
+    println!("cargo:rerun-if-env-changed=CXXFLAGS");
+    println!("cargo:rerun-if-changed=./icicle");
+
+    let arch_type = env::var("ARCH_TYPE").unwrap_or(String::from("native"));
+    let stream_type = env::var("DEFAULT_STREAM").unwrap_or(String::from("legacy"));
+
+    let mut arch = String::from("-arch=");
+    arch.push_str(&arch_type);
+    let mut stream = String::from("-default-stream=");
+    stream.push_str(&stream_type);
+
+    let mut nvcc = cc::Build::new();
+
+    println!("Compiling icicle library using arch: {}", &arch);
+
+    if cfg!(feature = "g2") {
+        nvcc.define("G2_DEFINED", None);
+    }
+    nvcc.cuda(true);
+    nvcc.define("FEATURE_BLS12_381", None);
+    nvcc.debug(false);
+    nvcc.flag(&arch);
+    nvcc.flag(&stream);
+    nvcc.shared_flag(false);
+    // nvcc.static_flag(true);
+    nvcc.files([
+        "../icicle-cuda/curves/index.cu",
+    ]);
+    nvcc.compile("ingo_icicle"); //TODO: extension??
+}

bls12-381/src/basic_structs/field.rs

@@ -0,0 +1,4 @@
+pub trait Field<const NUM_LIMBS: usize> {
+    const MODOLUS: [u32;NUM_LIMBS];
+    const LIMBS: usize = NUM_LIMBS;
+}

bls12-381/src/basic_structs/mod.rs

@@ -0,0 +1,3 @@
+pub mod field; 
+pub mod scalar; 
+pub mod point; 

bls12-381/src/basic_structs/point.rs

@@ -0,0 +1,106 @@
+use std::ffi::c_uint;
+
+use ark_ec::AffineCurve;
+use ark_ff::{BigInteger256, PrimeField};
+use std::mem::transmute;
+use ark_ff::Field;
+use icicle_core::utils::{u32_vec_to_u64_vec, u64_vec_to_u32_vec};
+
+use rustacuda_core::DeviceCopy;
+use rustacuda_derive::DeviceCopy;
+
+use super::scalar::{get_fixed_limbs, self};
+
+
+#[derive(Debug, Clone, Copy, DeviceCopy)]
+#[repr(C)]
+pub struct PointT<BF: scalar::ScalarTrait> {
+    pub x: BF,
+    pub y: BF,
+    pub z: BF,
+}
+
+impl<BF: DeviceCopy + scalar::ScalarTrait> Default for PointT<BF> {
+    fn default() -> Self {
+        PointT::zero()
+    }
+}
+
+impl<BF: DeviceCopy + scalar::ScalarTrait> PointT<BF> {
+    pub fn zero() -> Self {
+        PointT {
+            x: BF::zero(),
+            y: BF::one(),
+            z: BF::zero(),
+        }
+    }
+
+    pub fn infinity() -> Self {
+        Self::zero()
+    }
+}
+
+#[derive(Debug, PartialEq, Clone, Copy, DeviceCopy)]
+#[repr(C)]
+pub struct PointAffineNoInfinityT<BF> {
+    pub x: BF,
+    pub y: BF,
+}
+
+impl<BF: scalar::ScalarTrait> Default for PointAffineNoInfinityT<BF> {
+    fn default() -> Self {
+        PointAffineNoInfinityT {
+            x: BF::zero(),
+            y: BF::zero(),
+        }
+    }
+}
+
+impl<BF: Copy + scalar::ScalarTrait> PointAffineNoInfinityT<BF> {
+    ///From u32 limbs x,y
+    pub fn from_limbs(x: &[u32], y: &[u32]) -> Self {
+        PointAffineNoInfinityT {
+            x: BF::from_limbs(x),
+            y: BF::from_limbs(y)
+        }
+    }
+
+    pub fn limbs(&self) -> Vec<u32> {
+        [self.x.limbs(), self.y.limbs()].concat()
+    }
+
+    pub fn to_projective(&self) -> PointT<BF> {
+        PointT {
+            x: self.x,
+            y: self.y,
+            z: BF::one(),
+        }
+    }
+}
+
+impl<BF: Copy + scalar::ScalarTrait> PointT<BF>  {
+    pub fn from_limbs(x: &[u32], y: &[u32], z: &[u32]) -> Self {
+        PointT {
+            x: BF::from_limbs(x),
+            y: BF::from_limbs(y),
+            z: BF::from_limbs(z)
+        }
+    }
+
+    pub fn from_xy_limbs(value: &[u32]) -> PointT<BF> {
+        let l = value.len();
+        assert_eq!(l, 3 * BF::base_limbs(), "length must be 3 * {}", BF::base_limbs());
+        PointT {
+            x: BF::from_limbs(value[..BF::base_limbs()].try_into().unwrap()),
+            y: BF::from_limbs(value[BF::base_limbs()..BF::base_limbs() * 2].try_into().unwrap()),
+            z: BF::from_limbs(value[BF::base_limbs() * 2..].try_into().unwrap())
+        }
+    }
+
+    pub fn to_xy_strip_z(&self) -> PointAffineNoInfinityT<BF> {
+        PointAffineNoInfinityT {
+            x: self.x,
+            y: self.y,
+        }
+    }
+}

bls12-381/src/basic_structs/scalar.rs

@@ -0,0 +1,102 @@
+use std::ffi::{c_int, c_uint};
+use rand::{rngs::StdRng, RngCore, SeedableRng};
+use rustacuda_core::DeviceCopy;
+use rustacuda_derive::DeviceCopy;
+use std::mem::transmute;
+use rustacuda::prelude::*;
+use rustacuda_core::DevicePointer;
+use rustacuda::memory::{DeviceBox, CopyDestination};
+
+use icicle_core::utils::{u32_vec_to_u64_vec, u64_vec_to_u32_vec};
+
+use std::marker::PhantomData;
+use std::convert::TryInto;
+
+use super::field::{Field, self};
+
+pub fn get_fixed_limbs<const NUM_LIMBS: usize>(val: &[u32]) -> [u32; NUM_LIMBS] {
+    match val.len() {
+        n if n < NUM_LIMBS => {
+            let mut padded: [u32; NUM_LIMBS] = [0; NUM_LIMBS];
+            padded[..val.len()].copy_from_slice(&val);
+            padded
+        }
+        n if n == NUM_LIMBS => val.try_into().unwrap(),
+        _ => panic!("slice has too many elements"),
+    }
+}
+
+pub trait ScalarTrait{
+    fn base_limbs() -> usize;
+    fn zero() -> Self;
+    fn from_limbs(value: &[u32]) -> Self;
+    fn one() -> Self;
+    fn to_bytes_le(&self) -> Vec<u8>;
+    fn limbs(&self) -> &[u32];
+}
+
+#[derive(Debug, PartialEq, Clone, Copy)]
+#[repr(C)]
+pub struct ScalarT<M, const NUM_LIMBS: usize> {
+    pub(crate) phantom: PhantomData<M>,
+    pub(crate) value : [u32; NUM_LIMBS]
+}
+
+impl<M, const NUM_LIMBS: usize> ScalarTrait for ScalarT<M, NUM_LIMBS>
+where
+    M: Field<NUM_LIMBS>,
+{
+
+    fn base_limbs() -> usize {
+        return NUM_LIMBS; 
+    }
+
+    fn zero() -> Self {
+        ScalarT {
+            value: [0u32; NUM_LIMBS],
+            phantom: PhantomData,
+        }
+    }
+
+    fn from_limbs(value: &[u32]) -> Self {
+        Self {
+            value: get_fixed_limbs(value),
+            phantom: PhantomData,
+        }
+    }
+
+    fn one() -> Self {
+        let mut s = [0u32; NUM_LIMBS];
+        s[0] = 1;
+        ScalarT { value: s, phantom: PhantomData }
+    }
+
+    fn to_bytes_le(&self) -> Vec<u8> {
+        self.value
+            .iter()
+            .map(|s| s.to_le_bytes().to_vec())
+            .flatten()
+            .collect::<Vec<_>>()
+    }
+
+    fn limbs(&self) -> &[u32] {
+        &self.value
+    }
+}
+
+impl<M, const NUM_LIMBS: usize> ScalarT<M, NUM_LIMBS> where M: field::Field<NUM_LIMBS>{
+    pub fn from_limbs_le(value: &[u32]) -> ScalarT<M,NUM_LIMBS> {
+        Self::from_limbs(value)
+     }
+ 
+    pub fn from_limbs_be(value: &[u32]) -> ScalarT<M,NUM_LIMBS> {
+         let mut value = value.to_vec();
+         value.reverse();
+         Self::from_limbs_le(&value)
+     }
+ 
+     // Additional Functions
+     pub fn add(&self, other:ScalarT<M, NUM_LIMBS>) -> ScalarT<M,NUM_LIMBS>{  // overload + 
+         return ScalarT{value: [self.value[0] + other.value[0];NUM_LIMBS], phantom: PhantomData }; 
+     }
+}

bls12-381/src/curve_structs.rs

@@ -0,0 +1,62 @@
+use std::ffi::{c_int, c_uint};
+use rand::{rngs::StdRng, RngCore, SeedableRng};
+use rustacuda_derive::DeviceCopy;
+use std::mem::transmute;
+use rustacuda::prelude::*;
+use rustacuda_core::DevicePointer;
+use rustacuda::memory::{DeviceBox, CopyDestination, DeviceCopy};
+
+use std::marker::PhantomData;
+use std::convert::TryInto;
+
+use crate::basic_structs::point::{PointT, PointAffineNoInfinityT};
+use crate::basic_structs::scalar::ScalarT;
+use crate::basic_structs::field::Field;
+
+
+#[derive(Debug, PartialEq, Clone, Copy,DeviceCopy)]
+#[repr(C)]
+pub struct ScalarField;
+impl Field<8> for ScalarField {
+    const MODOLUS: [u32; 8] = [0x0;8];
+}
+
+#[derive(Debug, PartialEq, Clone, Copy,DeviceCopy)]
+#[repr(C)]
+pub struct BaseField;
+impl Field<12> for BaseField {
+    const MODOLUS: [u32; 12] = [0x0;12];
+}
+
+
+pub type Scalar = ScalarT<ScalarField,8>;
+impl Default for Scalar {
+    fn default() -> Self {
+        Self{value: [0x0;ScalarField::LIMBS], phantom: PhantomData }
+    }
+}
+
+unsafe impl DeviceCopy for Scalar{}
+
+
+pub type Base = ScalarT<BaseField,12>;
+impl Default for Base {
+    fn default() -> Self {
+        Self{value: [0x0;BaseField::LIMBS], phantom: PhantomData }
+    }
+}
+
+unsafe impl DeviceCopy for Base{}
+
+pub type Point = PointT<Base>;
+pub type PointAffineNoInfinity = PointAffineNoInfinityT<Base>;
+
+extern "C" {
+    fn eq(point1: *const Point, point2: *const Point) -> c_uint;
+}
+
+impl PartialEq for Point {
+    fn eq(&self, other: &Self) -> bool {
+        unsafe { eq(self, other) != 0 }
+    }
+}

bls12-381/src/from_cuda.rs

@@ -0,0 +1,798 @@
+use std::ffi::{c_int, c_uint};
+use ark_std::UniformRand;
+use rand::{rngs::StdRng, RngCore, SeedableRng};
+use rustacuda::CudaFlags;
+use rustacuda::memory::DeviceBox;
+use rustacuda::prelude::{DeviceBuffer, Device, ContextFlags, Context};
+use rustacuda_core::DevicePointer;
+use std::mem::transmute;
+use crate::basic_structs::scalar::ScalarTrait;
+use crate::curve_structs::*;
+use icicle_core::utils::{u32_vec_to_u64_vec, u64_vec_to_u32_vec};
+use std::marker::PhantomData;
+use std::convert::TryInto;
+use ark_bls12_381::{Fq as Fq_BLS12_381, Fr as Fr_BLS12_381, G1Affine as G1Affine_BLS12_381, G1Projective as G1Projective_BLS12_381};
+use ark_ec::AffineCurve;
+use ark_ff::{BigInteger384, BigInteger256, PrimeField};
+use rustacuda::memory::{CopyDestination, DeviceCopy};
+
+extern "C" {
+    fn msm_cuda(
+        out: *mut Point,
+        points: *const PointAffineNoInfinity,
+        scalars: *const Scalar,
+        count: usize,
+        device_id: usize,
+    ) -> c_uint;
+
+    fn msm_batch_cuda(
+        out: *mut Point,
+        points: *const PointAffineNoInfinity,
+        scalars: *const Scalar,
+        batch_size: usize,
+        msm_size: usize,
+        device_id: usize,
+    ) -> c_uint;
+
+    fn commit_cuda(
+        d_out: DevicePointer<Point>,
+        d_scalars: DevicePointer<Scalar>,
+        d_points: DevicePointer<PointAffineNoInfinity>,
+        count: usize,
+        device_id: usize,
+    ) -> c_uint;
+
+    fn commit_batch_cuda(
+        d_out: DevicePointer<Point>,
+        d_scalars: DevicePointer<Scalar>,
+        d_points: DevicePointer<PointAffineNoInfinity>,
+        count: usize,
+        batch_size: usize,
+        device_id: usize,
+    ) -> c_uint;
+
+    fn build_domain_cuda(domain_size: usize, logn: usize, inverse: bool, device_id: usize) -> DevicePointer<Scalar>;
+
+    fn ntt_cuda(inout: *mut Scalar, n: usize, inverse: bool, device_id: usize) -> c_int;
+
+    fn ecntt_cuda(inout: *mut Point, n: usize, inverse: bool, device_id: usize) -> c_int;
+
+    fn ntt_batch_cuda(
+        inout: *mut Scalar,
+        arr_size: usize,
+        n: usize,
+        inverse: bool,
+    ) -> c_int;
+
+    fn ecntt_batch_cuda(inout: *mut Point, arr_size: usize, n: usize, inverse: bool) -> c_int;
+
+    fn interpolate_scalars_cuda(
+        d_out: DevicePointer<Scalar>,
+        d_evaluations: DevicePointer<Scalar>,
+        d_domain: DevicePointer<Scalar>, 
+        n: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn interpolate_scalars_batch_cuda(
+        d_out: DevicePointer<Scalar>,
+        d_evaluations: DevicePointer<Scalar>,
+        d_domain: DevicePointer<Scalar>,
+        n: usize,
+        batch_size: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn interpolate_points_cuda(
+        d_out: DevicePointer<Point>,
+        d_evaluations: DevicePointer<Point>,
+        d_domain: DevicePointer<Scalar>,
+        n: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn interpolate_points_batch_cuda(
+        d_out: DevicePointer<Point>,
+        d_evaluations: DevicePointer<Point>,
+        d_domain: DevicePointer<Scalar>,
+        n: usize,
+        batch_size: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn evaluate_scalars_cuda(
+        d_out: DevicePointer<Scalar>,
+        d_coefficients: DevicePointer<Scalar>,
+        d_domain: DevicePointer<Scalar>,
+        domain_size: usize,
+        n: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn evaluate_scalars_batch_cuda(
+        d_out: DevicePointer<Scalar>,
+        d_coefficients: DevicePointer<Scalar>,
+        d_domain: DevicePointer<Scalar>,
+        domain_size: usize,
+        n: usize,
+        batch_size: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn evaluate_points_cuda(
+        d_out: DevicePointer<Point>,
+        d_coefficients: DevicePointer<Point>,
+        d_domain: DevicePointer<Scalar>,
+        domain_size: usize,
+        n: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn evaluate_points_batch_cuda(
+        d_out: DevicePointer<Point>,
+        d_coefficients: DevicePointer<Point>,
+        d_domain: DevicePointer<Scalar>,
+        domain_size: usize,
+        n: usize,
+        batch_size: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn evaluate_scalars_on_coset_cuda(
+        d_out: DevicePointer<Scalar>,
+        d_coefficients: DevicePointer<Scalar>,
+        d_domain: DevicePointer<Scalar>,
+        domain_size: usize,
+        n: usize,
+        coset_powers: DevicePointer<Scalar>,
+        device_id: usize
+    ) -> c_int;
+
+    fn evaluate_scalars_on_coset_batch_cuda(
+        d_out: DevicePointer<Scalar>,
+        d_coefficients: DevicePointer<Scalar>,
+        d_domain: DevicePointer<Scalar>,
+        domain_size: usize,
+        n: usize,
+        batch_size: usize,
+        coset_powers: DevicePointer<Scalar>,
+        device_id: usize
+    ) -> c_int;
+
+    fn evaluate_points_on_coset_cuda(
+        d_out: DevicePointer<Point>,
+        d_coefficients: DevicePointer<Point>,
+        d_domain: DevicePointer<Scalar>,
+        domain_size: usize,
+        n: usize,
+        coset_powers: DevicePointer<Scalar>,
+        device_id: usize
+    ) -> c_int;
+
+    fn evaluate_points_on_coset_batch_cuda(
+        d_out: DevicePointer<Point>,
+        d_coefficients: DevicePointer<Point>,
+        d_domain: DevicePointer<Scalar>,
+        domain_size: usize,
+        n: usize,
+        batch_size: usize,
+        coset_powers: DevicePointer<Scalar>,
+        device_id: usize
+    ) -> c_int;
+
+    fn reverse_order_scalars_cuda(
+        d_arr: DevicePointer<Scalar>,
+        n: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn reverse_order_scalars_batch_cuda(
+        d_arr: DevicePointer<Scalar>,
+        n: usize,
+        batch_size: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn reverse_order_points_cuda(
+        d_arr: DevicePointer<Point>,
+        n: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn reverse_order_points_batch_cuda(
+        d_arr: DevicePointer<Point>,
+        n: usize,
+        batch_size: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn vec_mod_mult_point(
+        inout: *mut Point,
+        scalars: *const Scalar,
+        n_elements: usize,
+        device_id: usize,
+    ) -> c_int;
+
+    fn vec_mod_mult_scalar(
+        inout: *mut Scalar,
+        scalars: *const Scalar,
+        n_elements: usize,
+        device_id: usize,
+    ) -> c_int;
+
+    fn matrix_vec_mod_mult(
+        matrix_flattened: *const Scalar,
+        input: *const Scalar,
+        output: *mut Scalar,
+        n_elements: usize,
+        device_id: usize,
+    ) -> c_int;
+}
+
+pub fn msm(points: &[PointAffineNoInfinity], scalars: &[Scalar], device_id: usize) -> Point {
+    let count = points.len();
+    if count != scalars.len() {
+        todo!("variable length")
+    }
+
+    let mut ret = Point::zero();
+    unsafe {
+        msm_cuda(
+            &mut ret as *mut _ as *mut Point,
+            points as *const _ as *const PointAffineNoInfinity,
+            scalars as *const _ as *const Scalar,
+            scalars.len(),
+            device_id,
+        )
+    };
+
+    ret
+}
+
+pub fn msm_batch(
+    points: &[PointAffineNoInfinity],
+    scalars: &[Scalar],
+    batch_size: usize,
+    device_id: usize,
+) -> Vec<Point> {
+    let count = points.len();
+    if count != scalars.len() {
+        todo!("variable length")
+    }
+
+    let mut ret = vec![Point::zero(); batch_size];
+
+    unsafe {
+        msm_batch_cuda(
+            &mut ret[0] as *mut _ as *mut Point,
+            points as *const _ as *const PointAffineNoInfinity,
+            scalars as *const _ as *const Scalar,
+            batch_size,
+            count / batch_size,
+            device_id,
+        )
+    };
+
+    ret
+}
+
+pub fn commit(
+    points: &mut DeviceBuffer<PointAffineNoInfinity>,
+    scalars: &mut DeviceBuffer<Scalar>,
+) -> DeviceBox<Point> {
+    let mut res = DeviceBox::new(&Point::zero()).unwrap();
+    unsafe {
+        commit_cuda(
+            res.as_device_ptr(),
+            scalars.as_device_ptr(),
+            points.as_device_ptr(),
+            scalars.len(),
+            0,
+        );
+    }
+    return res;
+}
+
+pub fn commit_batch(
+    points: &mut DeviceBuffer<PointAffineNoInfinity>,
+    scalars: &mut DeviceBuffer<Scalar>,
+    batch_size: usize,
+) -> DeviceBuffer<Point> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(batch_size).unwrap() };
+    unsafe {
+        commit_batch_cuda(
+            res.as_device_ptr(),
+            scalars.as_device_ptr(),
+            points.as_device_ptr(),
+            scalars.len() / batch_size,
+            batch_size,
+            0,
+        );
+    }
+    return res;
+}
+
+/// Compute an in-place NTT on the input data.
+fn ntt_internal(values: &mut [Scalar], device_id: usize, inverse: bool) -> i32 {
+    let ret_code = unsafe {
+        ntt_cuda(
+            values as *mut _ as *mut Scalar,
+            values.len(),
+            inverse,
+            device_id,
+        )
+    };
+    ret_code
+}
+
+pub fn ntt(values: &mut [Scalar], device_id: usize) {
+    ntt_internal(values, device_id, false);
+}
+
+pub fn intt(values: &mut [Scalar], device_id: usize) {
+    ntt_internal(values, device_id, true);
+}
+
+/// Compute an in-place NTT on the input data.
+fn ntt_internal_batch(
+    values: &mut [Scalar],
+    device_id: usize,
+    batch_size: usize,
+    inverse: bool,
+) -> i32 {
+    unsafe {
+        ntt_batch_cuda(
+            values as *mut _ as *mut Scalar,
+            values.len(),
+            batch_size,
+            inverse,
+        )
+    }
+}
+
+pub fn ntt_batch(values: &mut [Scalar], batch_size: usize, device_id: usize) {
+    ntt_internal_batch(values, 0, batch_size, false);
+}
+
+pub fn intt_batch(values: &mut [Scalar], batch_size: usize, device_id: usize) {
+    ntt_internal_batch(values, 0, batch_size, true);
+}
+
+/// Compute an in-place ECNTT on the input data.
+fn ecntt_internal(values: &mut [Point], inverse: bool, device_id: usize) -> i32 {
+    unsafe {
+        ecntt_cuda(
+            values as *mut _ as *mut Point,
+            values.len(),
+            inverse,
+            device_id,
+        )
+    }
+}
+
+pub fn ecntt(values: &mut [Point], device_id: usize) {
+    ecntt_internal(values, false, device_id);
+}
+
+/// Compute an in-place iECNTT on the input data.
+pub fn iecntt(values: &mut [Point], device_id: usize) {
+    ecntt_internal(values, true, device_id);
+}
+
+/// Compute an in-place ECNTT on the input data.
+fn ecntt_internal_batch(
+    values: &mut [Point],
+    device_id: usize,
+    batch_size: usize,
+    inverse: bool,
+) -> i32 {
+    unsafe {
+        ecntt_batch_cuda(
+            values as *mut _ as *mut Point,
+            values.len(),
+            batch_size,
+            inverse,
+        )
+    }
+}
+
+pub fn ecntt_batch(values: &mut [Point], batch_size: usize, device_id: usize) {
+    ecntt_internal_batch(values, 0, batch_size, false);
+}
+
+/// Compute an in-place iECNTT on the input data.
+pub fn iecntt_batch(values: &mut [Point], batch_size: usize, device_id: usize) {
+    ecntt_internal_batch(values, 0, batch_size, true);
+}
+
+pub fn build_domain(domain_size: usize, logn: usize, inverse: bool) -> DeviceBuffer<Scalar> {
+    unsafe {
+        DeviceBuffer::from_raw_parts(build_domain_cuda(
+            domain_size,
+            logn,
+            inverse,
+            0
+        ), domain_size)
+    }
+}
+
+
+pub fn reverse_order_scalars(
+    d_scalars: &mut DeviceBuffer<Scalar>,
+) {
+    unsafe { reverse_order_scalars_cuda(
+        d_scalars.as_device_ptr(),
+        d_scalars.len(),
+        0
+    ); }
+}
+
+pub fn reverse_order_scalars_batch(
+    d_scalars: &mut DeviceBuffer<Scalar>,
+    batch_size: usize,
+) {
+    unsafe { reverse_order_scalars_batch_cuda(
+        d_scalars.as_device_ptr(),
+        d_scalars.len() / batch_size,
+        batch_size,
+        0
+    ); }
+}
+
+pub fn reverse_order_points(
+    d_points: &mut DeviceBuffer<Point>,
+) {
+    unsafe { reverse_order_points_cuda(
+        d_points.as_device_ptr(),
+        d_points.len(),
+        0
+    ); }
+}
+
+pub fn reverse_order_points_batch(
+    d_points: &mut DeviceBuffer<Point>,
+    batch_size: usize,
+) {
+    unsafe { reverse_order_points_batch_cuda(
+        d_points.as_device_ptr(),
+        d_points.len() / batch_size,
+        batch_size,
+        0
+    ); }
+}
+
+pub fn interpolate_scalars(
+    d_evaluations: &mut DeviceBuffer<Scalar>,
+    d_domain: &mut DeviceBuffer<Scalar>
+) -> DeviceBuffer<Scalar> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len()).unwrap() };
+    unsafe { interpolate_scalars_cuda(
+        res.as_device_ptr(),
+        d_evaluations.as_device_ptr(),
+        d_domain.as_device_ptr(),
+        d_domain.len(),
+        0
+    ) };
+    return res;
+}
+
+pub fn interpolate_scalars_batch(
+    d_evaluations: &mut DeviceBuffer<Scalar>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+    batch_size: usize,
+) -> DeviceBuffer<Scalar> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len() * batch_size).unwrap() };
+    unsafe { interpolate_scalars_batch_cuda(
+        res.as_device_ptr(),
+        d_evaluations.as_device_ptr(),
+        d_domain.as_device_ptr(),
+        d_domain.len(),
+        batch_size,
+        0
+    ) };
+    return res;
+}
+
+pub fn interpolate_points(
+    d_evaluations: &mut DeviceBuffer<Point>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+) -> DeviceBuffer<Point> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len()).unwrap() };
+    unsafe { interpolate_points_cuda(
+        res.as_device_ptr(),
+        d_evaluations.as_device_ptr(),
+        d_domain.as_device_ptr(),
+        d_domain.len(),
+        0
+    ) };
+    return res;
+}
+
+pub fn interpolate_points_batch(
+    d_evaluations: &mut DeviceBuffer<Point>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+    batch_size: usize,
+) -> DeviceBuffer<Point> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len() * batch_size).unwrap() };
+    unsafe { interpolate_points_batch_cuda(
+        res.as_device_ptr(),
+        d_evaluations.as_device_ptr(),
+        d_domain.as_device_ptr(),
+        d_domain.len(),
+        batch_size,
+        0
+    ) };
+    return res;
+}
+
+pub fn evaluate_scalars(
+    d_coefficients: &mut DeviceBuffer<Scalar>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+) -> DeviceBuffer<Scalar> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len()).unwrap() };
+    unsafe {
+        evaluate_scalars_cuda(
+            res.as_device_ptr(),
+            d_coefficients.as_device_ptr(),
+            d_domain.as_device_ptr(),
+            d_domain.len(),
+            d_coefficients.len(),
+            0
+        );
+    }
+    return res;
+}
+
+pub fn evaluate_scalars_batch(
+    d_coefficients: &mut DeviceBuffer<Scalar>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+    batch_size: usize,
+) -> DeviceBuffer<Scalar> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len() * batch_size).unwrap() };
+    unsafe {
+        evaluate_scalars_batch_cuda(
+            res.as_device_ptr(),
+            d_coefficients.as_device_ptr(),
+            d_domain.as_device_ptr(),
+            d_domain.len(),
+            d_coefficients.len() / batch_size,
+            batch_size,
+            0
+        );
+    }
+    return res;
+}
+
+pub fn evaluate_points(
+    d_coefficients: &mut DeviceBuffer<Point>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+) -> DeviceBuffer<Point> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len()).unwrap() };
+    unsafe {
+        evaluate_points_cuda(
+            res.as_device_ptr(),
+            d_coefficients.as_device_ptr(),
+            d_domain.as_device_ptr(),
+            d_domain.len(),
+            d_coefficients.len(),
+            0
+        );
+    }
+    return res;
+}
+
+pub fn evaluate_points_batch(
+    d_coefficients: &mut DeviceBuffer<Point>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+    batch_size: usize,
+) -> DeviceBuffer<Point> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len() * batch_size).unwrap() };
+    unsafe {
+        evaluate_points_batch_cuda(
+            res.as_device_ptr(),
+            d_coefficients.as_device_ptr(),
+            d_domain.as_device_ptr(),
+            d_domain.len(),
+            d_coefficients.len() / batch_size,
+            batch_size,
+            0
+        );
+    }
+    return res;
+}
+
+pub fn evaluate_scalars_on_coset(
+    d_coefficients: &mut DeviceBuffer<Scalar>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+    coset_powers: &mut DeviceBuffer<Scalar>,
+) -> DeviceBuffer<Scalar> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len()).unwrap() };
+    unsafe {
+        evaluate_scalars_on_coset_cuda(
+            res.as_device_ptr(),
+            d_coefficients.as_device_ptr(),
+            d_domain.as_device_ptr(),
+            d_domain.len(),
+            d_coefficients.len(),
+            coset_powers.as_device_ptr(),
+            0
+        );
+    }
+    return res;
+}
+
+pub fn evaluate_scalars_on_coset_batch(
+    d_coefficients: &mut DeviceBuffer<Scalar>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+    batch_size: usize,
+    coset_powers: &mut DeviceBuffer<Scalar>,
+) -> DeviceBuffer<Scalar> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len() * batch_size).unwrap() };
+    unsafe {
+        evaluate_scalars_on_coset_batch_cuda(
+            res.as_device_ptr(),
+            d_coefficients.as_device_ptr(),
+            d_domain.as_device_ptr(),
+            d_domain.len(),
+            d_coefficients.len() / batch_size,
+            batch_size,
+            coset_powers.as_device_ptr(),
+            0
+        );
+    }
+    return res;
+}
+
+pub fn evaluate_points_on_coset(
+    d_coefficients: &mut DeviceBuffer<Point>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+    coset_powers: &mut DeviceBuffer<Scalar>,
+) -> DeviceBuffer<Point> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len()).unwrap() };
+    unsafe {
+        evaluate_points_on_coset_cuda(
+            res.as_device_ptr(),
+            d_coefficients.as_device_ptr(),
+            d_domain.as_device_ptr(),
+            d_domain.len(),
+            d_coefficients.len(),
+            coset_powers.as_device_ptr(),
+            0
+        );
+    }
+    return res;
+}
+
+pub fn evaluate_points_on_coset_batch(
+    d_coefficients: &mut DeviceBuffer<Point>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+    batch_size: usize,
+    coset_powers: &mut DeviceBuffer<Scalar>,
+) -> DeviceBuffer<Point> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len() * batch_size).unwrap() };
+    unsafe {
+        evaluate_points_on_coset_batch_cuda(
+            res.as_device_ptr(),
+            d_coefficients.as_device_ptr(),
+            d_domain.as_device_ptr(),
+            d_domain.len(),
+            d_coefficients.len() / batch_size,
+            batch_size,
+            coset_powers.as_device_ptr(),
+            0
+        );
+    }
+    return res;
+}
+
+pub fn multp_vec(a: &mut [Point], b: &[Scalar], device_id: usize) {
+    assert_eq!(a.len(), b.len());
+    unsafe {
+        vec_mod_mult_point(
+            a as *mut _ as *mut Point,
+            b as *const _ as *const Scalar,
+            a.len(),
+            device_id,
+        );
+    }
+}
+
+pub fn mult_sc_vec(a: &mut [Scalar], b: &[Scalar], device_id: usize) {
+    assert_eq!(a.len(), b.len());
+    unsafe {
+        vec_mod_mult_scalar(
+            a as *mut _ as *mut Scalar,
+            b as *const _ as *const Scalar,
+            a.len(),
+            device_id,
+        );
+    }
+}
+
+// Multiply a matrix by a scalar:
+//  `a` - flattenned matrix;
+//  `b` - vector to multiply `a` by;
+pub fn mult_matrix_by_vec(a: &[Scalar], b: &[Scalar], device_id: usize) -> Vec<Scalar> {
+    let mut c = Vec::with_capacity(b.len());
+    for i in 0..b.len() {
+        c.push(Scalar::zero());
+    }
+    unsafe {
+        matrix_vec_mod_mult(
+            a as *const _ as *const Scalar,
+            b as *const _ as *const Scalar,
+            c.as_mut_slice() as *mut _ as *mut Scalar,
+            b.len(),
+            device_id,
+        );
+    }
+    c
+}
+
+pub fn clone_buffer<T: DeviceCopy>(buf: &mut DeviceBuffer<T>) -> DeviceBuffer<T> {
+    let mut buf_cpy = unsafe { DeviceBuffer::uninitialized(buf.len()).unwrap() };
+    unsafe { buf_cpy.copy_from(buf) };
+    return buf_cpy;
+}
+
+pub fn get_rng(seed: Option<u64>) -> Box<dyn RngCore> {
+    let rng: Box<dyn RngCore> = match seed {
+        Some(seed) => Box::new(StdRng::seed_from_u64(seed)),
+        None => Box::new(rand::thread_rng()),
+    };
+    rng
+}
+
+fn set_up_device() {
+    // Set up the context, load the module, and create a stream to run kernels in.
+    rustacuda::init(CudaFlags::empty()).unwrap();
+    let device = Device::get_device(0).unwrap();
+    let _ctx = Context::create_and_push(ContextFlags::MAP_HOST | ContextFlags::SCHED_AUTO, device).unwrap();
+}
+
+pub fn generate_random_points(
+    count: usize,
+    mut rng: Box<dyn RngCore>,
+) -> Vec<PointAffineNoInfinity> {
+    (0..count)
+        .map(|_| Point::from_ark(G1Projective_BLS12_381::rand(&mut rng)).to_xy_strip_z())
+        .collect()
+}
+
+pub fn generate_random_points_proj(count: usize, mut rng: Box<dyn RngCore>) -> Vec<Point> {
+    (0..count)
+        .map(|_| Point::from_ark(G1Projective_BLS12_381::rand(&mut rng)))
+        .collect()
+}
+
+pub fn generate_random_scalars(count: usize, mut rng: Box<dyn RngCore>) -> Vec<Scalar> {
+    (0..count)
+        .map(|_| Scalar::from_ark(Fr_BLS12_381::rand(&mut rng).into_repr()))
+        .collect()
+}
+
+pub fn set_up_points(test_size: usize, log_domain_size: usize, inverse: bool) -> (Vec<Point>, DeviceBuffer<Point>, DeviceBuffer<Scalar>) {
+    set_up_device();
+
+    let d_domain = build_domain(1 << log_domain_size, log_domain_size, inverse);
+
+    let seed = Some(0); // fix the rng to get two equal scalar 
+    let vector = generate_random_points_proj(test_size, get_rng(seed));
+    let mut vector_mut = vector.clone();
+
+    let mut d_vector = DeviceBuffer::from_slice(&vector[..]).unwrap();
+    (vector_mut, d_vector, d_domain)
+}
+
+pub fn set_up_scalars(test_size: usize, log_domain_size: usize, inverse: bool) -> (Vec<Scalar>, DeviceBuffer<Scalar>, DeviceBuffer<Scalar>) {
+    set_up_device();
+
+    let d_domain = build_domain(1 << log_domain_size, log_domain_size, inverse);
+
+    let seed = Some(0); // fix the rng to get two equal scalars
+    let mut vector_mut = generate_random_scalars(test_size, get_rng(seed));
+
+    let mut d_vector = DeviceBuffer::from_slice(&vector_mut[..]).unwrap();
+    (vector_mut, d_vector, d_domain)
+}
+

bls12-381/src/lib.rs

@@ -0,0 +1,4 @@
+pub mod test_bls12_381;
+pub mod basic_structs;
+pub mod from_cuda;
+pub mod curve_structs;

bls12-381/src/test_bls12_381.rs

@@ -0,0 +1,816 @@
+use std::ffi::{c_int, c_uint};
+use ark_std::UniformRand;
+use rand::{rngs::StdRng, RngCore, SeedableRng};
+use rustacuda::CudaFlags;
+use rustacuda::memory::DeviceBox;
+use rustacuda::prelude::{DeviceBuffer, Device, ContextFlags, Context};
+use rustacuda_core::DevicePointer;
+use std::mem::transmute;
+pub use crate::basic_structs::scalar::ScalarTrait;
+pub use crate::curve_structs::*;
+use icicle_core::utils::{u32_vec_to_u64_vec, u64_vec_to_u32_vec};
+use std::marker::PhantomData;
+use std::convert::TryInto;
+use ark_bls12_381::{Fq as Fq_BLS12_381, Fr as Fr_BLS12_381, G1Affine as G1Affine_BLS12_381, G1Projective as G1Projective_BLS12_381};
+use ark_ec::AffineCurve;
+use ark_ff::{BigInteger384, BigInteger256, PrimeField};
+use rustacuda::memory::{CopyDestination, DeviceCopy};
+
+
+impl Scalar {
+    pub fn to_biginteger254(&self) -> BigInteger256 {
+        BigInteger256::new(u32_vec_to_u64_vec(&self.limbs()).try_into().unwrap())
+    }
+
+    pub fn to_ark(&self) -> BigInteger256 {
+        BigInteger256::new(u32_vec_to_u64_vec(&self.limbs()).try_into().unwrap())
+    }
+
+    pub fn from_biginteger256(ark: BigInteger256) -> Self {
+        Self{ value: u64_vec_to_u32_vec(&ark.0).try_into().unwrap(), phantom : PhantomData}
+    }
+
+    pub fn to_biginteger256_transmute(&self) -> BigInteger256 {
+        unsafe { transmute(*self) }
+    }
+
+    pub fn from_biginteger_transmute(v: BigInteger256) -> Scalar {
+        Scalar{ value: unsafe{ transmute(v)}, phantom : PhantomData }
+    }
+
+    pub fn to_ark_transmute(&self) -> Fr_BLS12_381 {
+        unsafe { std::mem::transmute(*self) }
+    }
+
+    pub fn from_ark_transmute(v: &Fr_BLS12_381) -> Scalar {
+        unsafe { std::mem::transmute_copy(v) }
+    }
+
+    pub fn to_ark_mod_p(&self) -> Fr_BLS12_381 {
+        Fr_BLS12_381::new(BigInteger256::new(u32_vec_to_u64_vec(&self.limbs()).try_into().unwrap()))
+    }
+
+    pub fn to_ark_repr(&self) -> Fr_BLS12_381 {
+        Fr_BLS12_381::from_repr(BigInteger256::new(u32_vec_to_u64_vec(&self.limbs()).try_into().unwrap())).unwrap()
+    }
+
+    pub fn from_ark(v: BigInteger256) -> Scalar {
+        Self { value : u64_vec_to_u32_vec(&v.0).try_into().unwrap(), phantom: PhantomData}
+    }
+
+}
+
+impl Base {
+    pub fn to_ark(&self) -> BigInteger384 {
+        BigInteger384::new(u32_vec_to_u64_vec(&self.limbs()).try_into().unwrap())
+    }
+
+    pub fn from_ark(ark: BigInteger384) -> Self {
+        Self::from_limbs(&u64_vec_to_u32_vec(&ark.0))
+    }
+}
+
+
+impl Point {
+    pub fn to_ark(&self) -> G1Projective_BLS12_381 {
+        self.to_ark_affine().into_projective()
+    }
+
+    pub fn to_ark_affine(&self) -> G1Affine_BLS12_381 {
+        //TODO: generic conversion
+        use ark_ff::Field;
+        use std::ops::Mul;
+        let proj_x_field = Fq_BLS12_381::from_le_bytes_mod_order(&self.x.to_bytes_le());
+        let proj_y_field = Fq_BLS12_381::from_le_bytes_mod_order(&self.y.to_bytes_le());
+        let proj_z_field = Fq_BLS12_381::from_le_bytes_mod_order(&self.z.to_bytes_le());
+        let inverse_z = proj_z_field.inverse().unwrap();
+        let aff_x = proj_x_field.mul(inverse_z);
+        let aff_y = proj_y_field.mul(inverse_z);
+        G1Affine_BLS12_381::new(aff_x, aff_y, false)
+    }
+
+    pub fn from_ark(ark: G1Projective_BLS12_381) -> Point {
+        use ark_ff::Field;
+        let z_inv = ark.z.inverse().unwrap();
+        let z_invsq = z_inv * z_inv;
+        let z_invq3 = z_invsq * z_inv;
+        Point {
+            x: Base::from_ark((ark.x * z_invsq).into_repr()),
+            y: Base::from_ark((ark.y * z_invq3).into_repr()),
+            z: Base::one(),
+        }
+    }
+}
+
+impl PointAffineNoInfinity {
+
+    pub fn to_ark(&self) -> G1Affine_BLS12_381 {
+        G1Affine_BLS12_381::new(Fq_BLS12_381::new(self.x.to_ark()), Fq_BLS12_381::new(self.y.to_ark()), false)
+    }
+
+    pub fn to_ark_repr(&self) -> G1Affine_BLS12_381 {
+        G1Affine_BLS12_381::new(
+            Fq_BLS12_381::from_repr(self.x.to_ark()).unwrap(),
+            Fq_BLS12_381::from_repr(self.y.to_ark()).unwrap(),
+            false,
+        )
+    }
+
+    pub fn from_ark(p: &G1Affine_BLS12_381) -> Self {
+        PointAffineNoInfinity {
+            x: Base::from_ark(p.x.into_repr()),
+            y: Base::from_ark(p.y.into_repr()),
+        }
+    }
+}
+
+impl Point {
+    pub fn to_affine(&self) -> PointAffineNoInfinity {
+        let ark_affine = self.to_ark_affine();
+        PointAffineNoInfinity {
+            x: Base::from_ark(ark_affine.x.into_repr()),
+            y: Base::from_ark(ark_affine.y.into_repr()),
+        }
+    }
+}
+
+
+#[cfg(test)]
+pub(crate) mod tests_bls12_381 {
+    use std::ops::Add;
+    use ark_bls12_381::{Fr, G1Affine, G1Projective};
+    use ark_ec::{msm::VariableBaseMSM, AffineCurve, ProjectiveCurve};
+    use ark_ff::{FftField, Field, Zero, PrimeField};
+    use ark_std::UniformRand;
+    use rustacuda::prelude::{DeviceBuffer, CopyDestination};
+    use crate::curve_structs::{Point, Scalar, Base};
+    use crate::basic_structs::scalar::ScalarTrait;
+    use crate::from_cuda::{generate_random_points, get_rng, generate_random_scalars, msm, msm_batch, set_up_scalars, commit, commit_batch, ntt, intt, generate_random_points_proj, ecntt, iecntt, ntt_batch, ecntt_batch, iecntt_batch, intt_batch, reverse_order_scalars_batch, interpolate_scalars_batch, set_up_points, reverse_order_points, interpolate_points, reverse_order_points_batch, interpolate_points_batch, evaluate_scalars, interpolate_scalars, reverse_order_scalars, evaluate_points, build_domain, evaluate_scalars_on_coset, evaluate_points_on_coset, mult_matrix_by_vec, mult_sc_vec, multp_vec,evaluate_scalars_batch, evaluate_points_batch, evaluate_scalars_on_coset_batch, evaluate_points_on_coset_batch};
+
+    fn random_points_ark_proj(nof_elements: usize) -> Vec<G1Projective> {
+        let mut rng = ark_std::rand::thread_rng();
+        let mut points_ga: Vec<G1Projective> = Vec::new();
+        for _ in 0..nof_elements {
+            let aff = G1Projective::rand(&mut rng);
+            points_ga.push(aff);
+        }
+        points_ga
+    }
+
+    fn ecntt_arc_naive(
+        points: &Vec<G1Projective>,
+        size: usize,
+        inverse: bool,
+    ) -> Vec<G1Projective> {
+        let mut result: Vec<G1Projective> = Vec::new();
+        for _ in 0..size {
+            result.push(G1Projective::zero());
+        }
+        let rou: Fr;
+        if !inverse {
+            rou = Fr::get_root_of_unity(size).unwrap();
+        } else {
+            rou = Fr::inverse(&Fr::get_root_of_unity(size).unwrap()).unwrap();
+        }
+        for k in 0..size {
+            for l in 0..size {
+                let pow: [u64; 1] = [(l * k).try_into().unwrap()];
+                let mul_rou = Fr::pow(&rou, &pow);
+                result[k] = result[k].add(points[l].into_affine().mul(mul_rou));
+            }
+        }
+        if inverse {
+            let size2 = size as u64;
+            for k in 0..size {
+                let multfactor = Fr::inverse(&Fr::from(size2)).unwrap();
+                result[k] = result[k].into_affine().mul(multfactor);
+            }
+        }
+        return result;
+    }
+
+    fn check_eq(points: &Vec<G1Projective>, points2: &Vec<G1Projective>) -> bool {
+        let mut eq = true;
+        for i in 0..points.len() {
+            if points2[i].ne(&points[i]) {
+                eq = false;
+                break;
+            }
+        }
+        return eq;
+    }
+
+    fn test_naive_ark_ecntt(size: usize) {
+        let points = random_points_ark_proj(size);
+        let result1: Vec<G1Projective> = ecntt_arc_naive(&points, size, false);
+        let result2: Vec<G1Projective> = ecntt_arc_naive(&result1, size, true);
+        assert!(!check_eq(&result2, &result1));
+        assert!(check_eq(&result2, &points));
+    }
+
+    #[test]
+    fn test_msm() {
+        let test_sizes = [6, 9];
+
+        for pow2 in test_sizes {
+            let count = 1 << pow2;
+            let seed = None; // set Some to provide seed
+            let points = generate_random_points(count, get_rng(seed));
+            let scalars = generate_random_scalars(count, get_rng(seed));
+
+            let msm_result = msm(&points, &scalars, 0);
+
+            let point_r_ark: Vec<_> = points.iter().map(|x| x.to_ark_repr()).collect();
+            let scalars_r_ark: Vec<_> = scalars.iter().map(|x| x.to_ark()).collect();
+
+            let msm_result_ark = VariableBaseMSM::multi_scalar_mul(&point_r_ark, &scalars_r_ark);
+
+            assert_eq!(msm_result.to_ark_affine(), msm_result_ark);
+            assert_eq!(msm_result.to_ark(), msm_result_ark);
+            assert_eq!(
+                msm_result.to_ark_affine(),
+                Point::from_ark(msm_result_ark).to_ark_affine()
+            );
+        }
+    }
+
+    #[test]
+    fn test_batch_msm() {
+        for batch_pow2 in [2, 4] {
+            for pow2 in [4, 6] {
+                let msm_size = 1 << pow2;
+                let batch_size = 1 << batch_pow2;
+                let seed = None; // set Some to provide seed
+                let points_batch = generate_random_points(msm_size * batch_size, get_rng(seed));
+                let scalars_batch = generate_random_scalars(msm_size * batch_size, get_rng(seed));
+
+                let point_r_ark: Vec<_> = points_batch.iter().map(|x| x.to_ark_repr()).collect();
+                let scalars_r_ark: Vec<_> = scalars_batch.iter().map(|x| x.to_ark()).collect();
+
+                let expected: Vec<_> = point_r_ark
+                    .chunks(msm_size)
+                    .zip(scalars_r_ark.chunks(msm_size))
+                    .map(|p| Point::from_ark(VariableBaseMSM::multi_scalar_mul(p.0, p.1)))
+                    .collect();
+
+                let result = msm_batch(&points_batch, &scalars_batch, batch_size, 0);
+
+                assert_eq!(result, expected);
+            }
+        }
+    }
+
+    #[test]
+    fn test_commit() {
+        let test_size = 1 << 8;
+        let seed = Some(0);
+        let (mut scalars, mut d_scalars, _) = set_up_scalars(test_size, 0, false);
+        let mut points = generate_random_points(test_size, get_rng(seed));
+        let mut d_points = DeviceBuffer::from_slice(&points[..]).unwrap();
+
+        let msm_result = msm(&points, &scalars, 0);
+        let mut d_commit_result = commit(&mut d_points, &mut d_scalars);
+        let mut h_commit_result = Point::zero();
+        d_commit_result.copy_to(&mut h_commit_result).unwrap();
+
+        assert_eq!(msm_result, h_commit_result);
+        assert_ne!(msm_result, Point::zero());
+        assert_ne!(h_commit_result, Point::zero());
+    }
+
+    #[test]
+    fn test_batch_commit() {
+        let batch_size = 4;
+        let test_size = 1 << 12;
+        let seed = Some(0);
+        let (scalars, mut d_scalars, _) = set_up_scalars(test_size * batch_size, 0, false);
+        let points = generate_random_points(test_size * batch_size, get_rng(seed));
+        let mut d_points = DeviceBuffer::from_slice(&points[..]).unwrap();
+
+        let msm_result = msm_batch(&points, &scalars, batch_size, 0);
+        let mut d_commit_result = commit_batch(&mut d_points, &mut d_scalars, batch_size);
+        let mut h_commit_result: Vec<Point> = (0..batch_size).map(|_| Point::zero()).collect();
+        d_commit_result.copy_to(&mut h_commit_result[..]).unwrap();
+
+        assert_eq!(msm_result, h_commit_result);
+        for h in h_commit_result {
+            assert_ne!(h, Point::zero());
+        }
+    }
+
+    #[test]
+    fn test_ntt() {
+        //NTT
+        let seed = None; //some value to fix the rng
+        let test_size = 1 << 3;
+
+        let scalars = generate_random_scalars(test_size, get_rng(seed));
+
+        let mut ntt_result = scalars.clone();
+        ntt(&mut ntt_result, 0);
+
+        assert_ne!(ntt_result, scalars);
+
+        let mut intt_result = ntt_result.clone();
+
+        intt(&mut intt_result, 0);
+
+        assert_eq!(intt_result, scalars);
+
+        //ECNTT
+        let points_proj = generate_random_points_proj(test_size, get_rng(seed));
+
+        test_naive_ark_ecntt(test_size);
+
+        assert!(points_proj[0].to_ark().into_affine().is_on_curve());
+
+        //naive ark
+        let points_proj_ark = points_proj
+            .iter()
+            .map(|p| p.to_ark())
+            .collect::<Vec<G1Projective>>();
+
+        let ecntt_result_naive = ecntt_arc_naive(&points_proj_ark, points_proj_ark.len(), false);
+
+        let iecntt_result_naive = ecntt_arc_naive(&ecntt_result_naive, points_proj_ark.len(), true);
+
+        assert_eq!(points_proj_ark, iecntt_result_naive);
+
+        //ingo gpu
+        let mut ecntt_result = points_proj.to_vec();
+        ecntt(&mut ecntt_result, 0);
+
+        assert_ne!(ecntt_result, points_proj);
+
+        let mut iecntt_result = ecntt_result.clone();
+        iecntt(&mut iecntt_result, 0);
+
+        assert_eq!(
+            iecntt_result_naive,
+            points_proj
+                .iter()
+                .map(|p| p.to_ark_affine())
+                .collect::<Vec<G1Affine>>()
+        );
+        assert_eq!(
+            iecntt_result
+                .iter()
+                .map(|p| p.to_ark_affine())
+                .collect::<Vec<G1Affine>>(),
+            points_proj
+                .iter()
+                .map(|p| p.to_ark_affine())
+                .collect::<Vec<G1Affine>>()
+        );
+    }
+
+    #[test]
+    fn test_ntt_batch() {
+        //NTT
+        let seed = None; //some value to fix the rng
+        let test_size = 1 << 5;
+        let batches = 4;
+
+        let scalars_batch: Vec<Scalar> =
+            generate_random_scalars(test_size * batches, get_rng(seed));
+
+        let mut scalar_vec_of_vec: Vec<Vec<Scalar>> = Vec::new();
+
+        for i in 0..batches {
+            scalar_vec_of_vec.push(scalars_batch[i * test_size..(i + 1) * test_size].to_vec());
+        }
+
+        let mut ntt_result = scalars_batch.clone();
+
+        // do batch ntt
+        ntt_batch(&mut ntt_result, test_size, 0);
+
+        let mut ntt_result_vec_of_vec = Vec::new();
+
+        // do ntt for every chunk
+        for i in 0..batches {
+            ntt_result_vec_of_vec.push(scalar_vec_of_vec[i].clone());
+            ntt(&mut ntt_result_vec_of_vec[i], 0);
+        }
+
+        // check that the ntt of each vec of scalars is equal to the intt of the specific batch
+        for i in 0..batches {
+            assert_eq!(
+                ntt_result_vec_of_vec[i],
+                ntt_result[i * test_size..(i + 1) * test_size]
+            );
+        }
+
+        // check that ntt output is different from input
+        assert_ne!(ntt_result, scalars_batch);
+
+        let mut intt_result = ntt_result.clone();
+
+        // do batch intt
+        intt_batch(&mut intt_result, test_size, 0);
+
+        let mut intt_result_vec_of_vec = Vec::new();
+
+        // do intt for every chunk
+        for i in 0..batches {
+            intt_result_vec_of_vec.push(ntt_result_vec_of_vec[i].clone());
+            intt(&mut intt_result_vec_of_vec[i], 0);
+        }
+
+        // check that the intt of each vec of scalars is equal to the intt of the specific batch
+        for i in 0..batches {
+            assert_eq!(
+                intt_result_vec_of_vec[i],
+                intt_result[i * test_size..(i + 1) * test_size]
+            );
+        }
+
+        assert_eq!(intt_result, scalars_batch);
+
+        // //ECNTT
+        let points_proj = generate_random_points_proj(test_size * batches, get_rng(seed));
+
+        let mut points_vec_of_vec: Vec<Vec<Point>> = Vec::new();
+
+        for i in 0..batches {
+            points_vec_of_vec.push(points_proj[i * test_size..(i + 1) * test_size].to_vec());
+        }
+
+        let mut ntt_result_points = points_proj.clone();
+
+        // do batch ecintt
+        ecntt_batch(&mut ntt_result_points, test_size, 0);
+
+        let mut ntt_result_points_vec_of_vec = Vec::new();
+
+        for i in 0..batches {
+            ntt_result_points_vec_of_vec.push(points_vec_of_vec[i].clone());
+            ecntt(&mut ntt_result_points_vec_of_vec[i], 0);
+        }
+
+        for i in 0..batches {
+            assert_eq!(
+                ntt_result_points_vec_of_vec[i],
+                ntt_result_points[i * test_size..(i + 1) * test_size]
+            );
+        }
+
+        assert_ne!(ntt_result_points, points_proj);
+
+        let mut intt_result_points = ntt_result_points.clone();
+
+        // do batch ecintt
+        iecntt_batch(&mut intt_result_points, test_size, 0);
+
+        let mut intt_result_points_vec_of_vec = Vec::new();
+
+        // do ecintt for every chunk
+        for i in 0..batches {
+            intt_result_points_vec_of_vec.push(ntt_result_points_vec_of_vec[i].clone());
+            iecntt(&mut intt_result_points_vec_of_vec[i], 0);
+        }
+
+        // check that the ecintt of each vec of scalars is equal to the intt of the specific batch
+        for i in 0..batches {
+            assert_eq!(
+                intt_result_points_vec_of_vec[i],
+                intt_result_points[i * test_size..(i + 1) * test_size]
+            );
+        }
+
+        assert_eq!(intt_result_points, points_proj);
+    }
+
+    #[test]
+    fn test_scalar_interpolation() {
+        let log_test_size = 7;
+        let test_size = 1 << log_test_size;
+        let (mut evals_mut, mut d_evals, mut d_domain) = set_up_scalars(test_size, log_test_size, true);
+
+        reverse_order_scalars(&mut d_evals);
+        let mut d_coeffs = interpolate_scalars(&mut d_evals, &mut d_domain);
+        intt(&mut evals_mut, 0);
+        let mut h_coeffs: Vec<Scalar> = (0..test_size).map(|_| Scalar::zero()).collect();
+        d_coeffs.copy_to(&mut h_coeffs[..]).unwrap();
+
+        assert_eq!(h_coeffs, evals_mut);
+    }
+
+    #[test]
+    fn test_scalar_batch_interpolation() {
+        let batch_size = 4;
+        let log_test_size = 10;
+        let test_size = 1 << log_test_size;
+        let (mut evals_mut, mut d_evals, mut d_domain) = set_up_scalars(test_size * batch_size, log_test_size, true);
+
+        reverse_order_scalars_batch(&mut d_evals, batch_size);
+        let mut d_coeffs = interpolate_scalars_batch(&mut d_evals, &mut d_domain, batch_size);
+        intt_batch(&mut evals_mut, test_size, 0);
+        let mut h_coeffs: Vec<Scalar> = (0..test_size * batch_size).map(|_| Scalar::zero()).collect();
+        d_coeffs.copy_to(&mut h_coeffs[..]).unwrap();
+
+        assert_eq!(h_coeffs, evals_mut);
+    }
+
+    #[test]
+    fn test_point_interpolation() {
+        let log_test_size = 6;
+        let test_size = 1 << log_test_size;
+        let (mut evals_mut, mut d_evals, mut d_domain) = set_up_points(test_size, log_test_size, true);
+
+        reverse_order_points(&mut d_evals);
+        let mut d_coeffs = interpolate_points(&mut d_evals, &mut d_domain);
+        iecntt(&mut evals_mut[..], 0);
+        let mut h_coeffs: Vec<Point> = (0..test_size).map(|_| Point::zero()).collect();
+        d_coeffs.copy_to(&mut h_coeffs[..]).unwrap();
+        
+        assert_eq!(h_coeffs, *evals_mut);
+        for h in h_coeffs.iter() {
+            assert_ne!(*h, Point::zero());
+        }
+    }
+
+    #[test]
+    fn test_point_batch_interpolation() {
+        let batch_size = 4;
+        let log_test_size = 6;
+        let test_size = 1 << log_test_size;
+        let (mut evals_mut, mut d_evals, mut d_domain) = set_up_points(test_size * batch_size, log_test_size, true);
+
+        reverse_order_points_batch(&mut d_evals, batch_size);
+        let mut d_coeffs = interpolate_points_batch(&mut d_evals, &mut d_domain, batch_size);
+        iecntt_batch(&mut evals_mut[..], test_size, 0);
+        let mut h_coeffs: Vec<Point> = (0..test_size * batch_size).map(|_| Point::zero()).collect();
+        d_coeffs.copy_to(&mut h_coeffs[..]).unwrap();
+        
+        assert_eq!(h_coeffs, *evals_mut);
+        for h in h_coeffs.iter() {
+            assert_ne!(*h, Point::zero());
+        }
+    }
+
+    #[test]
+    fn test_scalar_evaluation() {
+        let log_test_domain_size = 8;
+        let coeff_size = 1 << 6;
+        let (h_coeffs, mut d_coeffs, mut d_domain) = set_up_scalars(coeff_size, log_test_domain_size, false);
+        let (_, _, mut d_domain_inv) = set_up_scalars(0, log_test_domain_size, true);
+
+        let mut d_evals = evaluate_scalars(&mut d_coeffs, &mut d_domain);
+        let mut d_coeffs_domain = interpolate_scalars(&mut d_evals, &mut d_domain_inv);
+        let mut h_coeffs_domain: Vec<Scalar> = (0..1 << log_test_domain_size).map(|_| Scalar::zero()).collect();
+        d_coeffs_domain.copy_to(&mut h_coeffs_domain[..]).unwrap();
+
+        assert_eq!(h_coeffs, h_coeffs_domain[..coeff_size]);
+        for i in coeff_size.. (1 << log_test_domain_size) {
+            assert_eq!(Scalar::zero(), h_coeffs_domain[i]);
+        }
+    }
+
+    #[test]
+    fn test_scalar_batch_evaluation() {
+        let batch_size = 6;
+        let log_test_domain_size = 8;
+        let domain_size = 1 << log_test_domain_size;
+        let coeff_size = 1 << 6;
+        let (h_coeffs, mut d_coeffs, mut d_domain) = set_up_scalars(coeff_size * batch_size, log_test_domain_size, false);
+        let (_, _, mut d_domain_inv) = set_up_scalars(0, log_test_domain_size, true);
+
+        let mut d_evals = evaluate_scalars_batch(&mut d_coeffs, &mut d_domain, batch_size);
+        let mut d_coeffs_domain = interpolate_scalars_batch(&mut d_evals, &mut d_domain_inv, batch_size);
+        let mut h_coeffs_domain: Vec<Scalar> = (0..domain_size * batch_size).map(|_| Scalar::zero()).collect();
+        d_coeffs_domain.copy_to(&mut h_coeffs_domain[..]).unwrap();
+
+        for j in 0..batch_size {
+            assert_eq!(h_coeffs[j * coeff_size..(j + 1) * coeff_size], h_coeffs_domain[j * domain_size..j * domain_size + coeff_size]);
+            for i in coeff_size..domain_size {
+                assert_eq!(Scalar::zero(), h_coeffs_domain[j * domain_size + i]);
+            }
+        }
+    }
+
+    #[test]
+    fn test_point_evaluation() {
+        let log_test_domain_size = 7;
+        let coeff_size = 1 << 7;
+        let (h_coeffs, mut d_coeffs, mut d_domain) = set_up_points(coeff_size, log_test_domain_size, false);
+        let (_, _, mut d_domain_inv) = set_up_points(0, log_test_domain_size, true);
+
+        let mut d_evals = evaluate_points(&mut d_coeffs, &mut d_domain);
+        let mut d_coeffs_domain = interpolate_points(&mut d_evals, &mut d_domain_inv);
+        let mut h_coeffs_domain: Vec<Point> = (0..1 << log_test_domain_size).map(|_| Point::zero()).collect();
+        d_coeffs_domain.copy_to(&mut h_coeffs_domain[..]).unwrap();
+
+        assert_eq!(h_coeffs[..], h_coeffs_domain[..coeff_size]);
+        for i in coeff_size..(1 << log_test_domain_size) {
+            assert_eq!(Point::zero(), h_coeffs_domain[i]);
+        }
+        for i in 0..coeff_size {
+            assert_ne!(h_coeffs_domain[i], Point::zero());
+        }
+    }
+
+    #[test]
+    fn test_point_batch_evaluation() {
+        let batch_size = 4;
+        let log_test_domain_size = 6;
+        let domain_size = 1 << log_test_domain_size;
+        let coeff_size = 1 << 5;
+        let (h_coeffs, mut d_coeffs, mut d_domain) = set_up_points(coeff_size * batch_size, log_test_domain_size, false);
+        let (_, _, mut d_domain_inv) = set_up_points(0, log_test_domain_size, true);
+
+        let mut d_evals = evaluate_points_batch(&mut d_coeffs, &mut d_domain, batch_size);
+        let mut d_coeffs_domain = interpolate_points_batch(&mut d_evals, &mut d_domain_inv, batch_size);
+        let mut h_coeffs_domain: Vec<Point> = (0..domain_size * batch_size).map(|_| Point::zero()).collect();
+        d_coeffs_domain.copy_to(&mut h_coeffs_domain[..]).unwrap();
+
+        for j in 0..batch_size {
+            assert_eq!(h_coeffs[j * coeff_size..(j + 1) * coeff_size], h_coeffs_domain[j * domain_size..(j * domain_size + coeff_size)]);
+            for i in coeff_size..domain_size {
+                assert_eq!(Point::zero(), h_coeffs_domain[j * domain_size + i]);
+            }
+            for i in j * domain_size..(j * domain_size + coeff_size) {
+                assert_ne!(h_coeffs_domain[i], Point::zero());
+            }
+        }
+    }
+
+    #[test]
+    fn test_scalar_evaluation_on_trivial_coset() {
+        // checks that the evaluations on the subgroup is the same as on the coset generated by 1
+        let log_test_domain_size = 8;
+        let coeff_size = 1 << 6;
+        let (_, mut d_coeffs, mut d_domain) = set_up_scalars(coeff_size, log_test_domain_size, false);
+        let (_, _, mut d_domain_inv) = set_up_scalars(coeff_size, log_test_domain_size, true);
+        let mut d_trivial_coset_powers = build_domain(1 << log_test_domain_size, 0, false);
+
+        let mut d_evals = evaluate_scalars(&mut d_coeffs, &mut d_domain);
+        let mut h_coeffs: Vec<Scalar> = (0..1 << log_test_domain_size).map(|_| Scalar::zero()).collect();
+        d_evals.copy_to(&mut h_coeffs[..]).unwrap();
+        let mut d_evals_coset = evaluate_scalars_on_coset(&mut d_coeffs, &mut d_domain, &mut d_trivial_coset_powers);
+        let mut h_evals_coset: Vec<Scalar> = (0..1 << log_test_domain_size).map(|_| Scalar::zero()).collect();
+        d_evals_coset.copy_to(&mut h_evals_coset[..]).unwrap();
+
+        assert_eq!(h_coeffs, h_evals_coset);
+    }
+
+    #[test]
+    fn test_scalar_evaluation_on_coset() {
+        // checks that evaluating a polynomial on a subgroup and its coset is the same as evaluating on a 2x larger subgroup 
+        let log_test_size = 8;
+        let test_size = 1 << log_test_size;
+        let (_, mut d_coeffs, mut d_domain) = set_up_scalars(test_size, log_test_size, false);
+        let (_, _, mut d_large_domain) = set_up_scalars(0, log_test_size + 1, false);
+        let mut d_coset_powers = build_domain(test_size, log_test_size + 1, false);
+
+        let mut d_evals_large = evaluate_scalars(&mut d_coeffs, &mut d_large_domain);
+        let mut h_evals_large: Vec<Scalar> = (0..2 * test_size).map(|_| Scalar::zero()).collect();
+        d_evals_large.copy_to(&mut h_evals_large[..]).unwrap();
+        let mut d_evals = evaluate_scalars(&mut d_coeffs, &mut d_domain);
+        let mut h_evals: Vec<Scalar> = (0..test_size).map(|_| Scalar::zero()).collect();
+        d_evals.copy_to(&mut h_evals[..]).unwrap();
+        let mut d_evals_coset = evaluate_scalars_on_coset(&mut d_coeffs, &mut d_domain, &mut d_coset_powers);
+        let mut h_evals_coset: Vec<Scalar> = (0..test_size).map(|_| Scalar::zero()).collect();
+        d_evals_coset.copy_to(&mut h_evals_coset[..]).unwrap();
+
+        assert_eq!(h_evals[..], h_evals_large[..test_size]);
+        assert_eq!(h_evals_coset[..], h_evals_large[test_size..2 * test_size]);
+    }
+
+    #[test]
+    fn test_scalar_batch_evaluation_on_coset() {
+        // checks that evaluating a polynomial on a subgroup and its coset is the same as evaluating on a 2x larger subgroup 
+        let batch_size = 4;
+        let log_test_size = 6;
+        let test_size = 1 << log_test_size;
+        let (_, mut d_coeffs, mut d_domain) = set_up_scalars(test_size * batch_size, log_test_size, false);
+        let (_, _, mut d_large_domain) = set_up_scalars(0, log_test_size + 1, false);
+        let mut d_coset_powers = build_domain(test_size, log_test_size + 1, false);
+
+        let mut d_evals_large = evaluate_scalars_batch(&mut d_coeffs, &mut d_large_domain, batch_size);
+        let mut h_evals_large: Vec<Scalar> = (0..2 * test_size * batch_size).map(|_| Scalar::zero()).collect();
+        d_evals_large.copy_to(&mut h_evals_large[..]).unwrap();
+        let mut d_evals = evaluate_scalars_batch(&mut d_coeffs, &mut d_domain, batch_size);
+        let mut h_evals: Vec<Scalar> = (0..test_size * batch_size).map(|_| Scalar::zero()).collect();
+        d_evals.copy_to(&mut h_evals[..]).unwrap();
+        let mut d_evals_coset = evaluate_scalars_on_coset_batch(&mut d_coeffs, &mut d_domain, batch_size, &mut d_coset_powers);
+        let mut h_evals_coset: Vec<Scalar> = (0..test_size * batch_size).map(|_| Scalar::zero()).collect();
+        d_evals_coset.copy_to(&mut h_evals_coset[..]).unwrap();
+
+        for i in 0..batch_size {
+            assert_eq!(h_evals_large[2 * i * test_size..(2 * i + 1) * test_size], h_evals[i * test_size..(i + 1) * test_size]);
+            assert_eq!(h_evals_large[(2 * i + 1) * test_size..(2 * i + 2) * test_size], h_evals_coset[i * test_size..(i + 1) * test_size]);
+        }
+    }
+
+    #[test]
+    fn test_point_evaluation_on_coset() {
+        // checks that evaluating a polynomial on a subgroup and its coset is the same as evaluating on a 2x larger subgroup 
+        let log_test_size = 8;
+        let test_size = 1 << log_test_size;
+        let (_, mut d_coeffs, mut d_domain) = set_up_points(test_size, log_test_size, false);
+        let (_, _, mut d_large_domain) = set_up_points(0, log_test_size + 1, false);
+        let mut d_coset_powers = build_domain(test_size, log_test_size + 1, false);
+
+        let mut d_evals_large = evaluate_points(&mut d_coeffs, &mut d_large_domain);
+        let mut h_evals_large: Vec<Point> = (0..2 * test_size).map(|_| Point::zero()).collect();
+        d_evals_large.copy_to(&mut h_evals_large[..]).unwrap();
+        let mut d_evals = evaluate_points(&mut d_coeffs, &mut d_domain);
+        let mut h_evals: Vec<Point> = (0..test_size).map(|_| Point::zero()).collect();
+        d_evals.copy_to(&mut h_evals[..]).unwrap();
+        let mut d_evals_coset = evaluate_points_on_coset(&mut d_coeffs, &mut d_domain, &mut d_coset_powers);
+        let mut h_evals_coset: Vec<Point> = (0..test_size).map(|_| Point::zero()).collect();
+        d_evals_coset.copy_to(&mut h_evals_coset[..]).unwrap();
+
+        assert_eq!(h_evals[..], h_evals_large[..test_size]);
+        assert_eq!(h_evals_coset[..], h_evals_large[test_size..2 * test_size]);
+        for i in 0..test_size {
+            assert_ne!(h_evals[i], Point::zero());
+            assert_ne!(h_evals_coset[i], Point::zero());
+            assert_ne!(h_evals_large[2 * i], Point::zero());
+            assert_ne!(h_evals_large[2 * i + 1], Point::zero());
+        }
+    }
+
+    #[test]
+    fn test_point_batch_evaluation_on_coset() {
+        // checks that evaluating a polynomial on a subgroup and its coset is the same as evaluating on a 2x larger subgroup 
+        let batch_size = 2;
+        let log_test_size = 6;
+        let test_size = 1 << log_test_size;
+        let (_, mut d_coeffs, mut d_domain) = set_up_points(test_size * batch_size, log_test_size, false);
+        let (_, _, mut d_large_domain) = set_up_points(0, log_test_size + 1, false);
+        let mut d_coset_powers = build_domain(test_size, log_test_size + 1, false);
+
+        let mut d_evals_large = evaluate_points_batch(&mut d_coeffs, &mut d_large_domain, batch_size);
+        let mut h_evals_large: Vec<Point> = (0..2 * test_size * batch_size).map(|_| Point::zero()).collect();
+        d_evals_large.copy_to(&mut h_evals_large[..]).unwrap();
+        let mut d_evals = evaluate_points_batch(&mut d_coeffs, &mut d_domain, batch_size);
+        let mut h_evals: Vec<Point> = (0..test_size * batch_size).map(|_| Point::zero()).collect();
+        d_evals.copy_to(&mut h_evals[..]).unwrap();
+        let mut d_evals_coset = evaluate_points_on_coset_batch(&mut d_coeffs, &mut d_domain, batch_size, &mut d_coset_powers);
+        let mut h_evals_coset: Vec<Point> = (0..test_size * batch_size).map(|_| Point::zero()).collect();
+        d_evals_coset.copy_to(&mut h_evals_coset[..]).unwrap();
+
+        for i in 0..batch_size {
+            assert_eq!(h_evals_large[2 * i * test_size..(2 * i + 1) * test_size], h_evals[i * test_size..(i + 1) * test_size]);
+            assert_eq!(h_evals_large[(2 * i + 1) * test_size..(2 * i + 2) * test_size], h_evals_coset[i * test_size..(i + 1) * test_size]);
+        }
+        for i in 0..test_size * batch_size {
+            assert_ne!(h_evals[i], Point::zero());
+            assert_ne!(h_evals_coset[i], Point::zero());
+            assert_ne!(h_evals_large[2 * i], Point::zero());
+            assert_ne!(h_evals_large[2 * i + 1], Point::zero());
+        }
+    }
+
+    // testing matrix multiplication by comparing the result of FFT with the naive multiplication by the DFT matrix
+    #[test]
+    fn test_matrix_multiplication() {
+        let seed = None; // some value to fix the rng
+        let test_size = 1 << 5;
+        let rou = Fr::get_root_of_unity(test_size).unwrap();
+        let matrix_flattened: Vec<Scalar> = (0..test_size).map(
+            |row_num| { (0..test_size).map( 
+                |col_num| {
+                    let pow: [u64; 1] = [(row_num * col_num).try_into().unwrap()];
+                    Scalar::from_ark(Fr::pow(&rou, &pow).into_repr())
+                }).collect::<Vec<Scalar>>()
+            }).flatten().collect::<Vec<_>>();
+        let vector: Vec<Scalar> = generate_random_scalars(test_size, get_rng(seed));
+
+        let result = mult_matrix_by_vec(&matrix_flattened, &vector, 0);
+        let mut ntt_result = vector.clone();
+        ntt(&mut ntt_result, 0);
+        
+        // we don't use the same roots of unity as arkworks, so the results are permutations
+        // of one another and the only guaranteed fixed scalars are the following ones:
+        assert_eq!(result[0], ntt_result[0]);
+        assert_eq!(result[test_size >> 1], ntt_result[test_size >> 1]);
+    }
+
+    #[test]
+    #[allow(non_snake_case)]
+    fn test_vec_scalar_mul() {
+        let mut intoo = [Scalar::one(), Scalar::one(), Scalar::zero()];
+        let expected = [Scalar::one(), Scalar::zero(), Scalar::zero()];
+        mult_sc_vec(&mut intoo, &expected, 0);
+        assert_eq!(intoo, expected);
+    }
+
+    #[test]
+    #[allow(non_snake_case)]
+    fn test_vec_point_mul() {
+        let dummy_one = Point {
+            x: Base::one(),
+            y: Base::one(),
+            z: Base::one(),
+        };
+
+        let mut inout = [dummy_one, dummy_one, Point::zero()];
+        let scalars = [Scalar::one(), Scalar::zero(), Scalar::zero()];
+        let expected = [dummy_one, Point::zero(), Point::zero()];
+        multp_vec(&mut inout, &scalars, 0);
+        assert_eq!(inout, expected);
+    }
+}

bn254/Cargo.toml

@@ -0,0 +1,34 @@
+[package]
+name = "bn254"
+version = "0.1.0"
+edition = "2021"
+authors = [ "Ingonyama" ]
+
+[dependencies]
+icicle-core = { path = "../icicle-core" }
+
+hex = "*"
+ark-std = "0.3.0"
+ark-ff = "0.3.0"
+ark-poly = "0.3.0"
+ark-ec = { version = "0.3.0", features = [ "parallel" ] }
+ark-bn254 = "0.3.0"
+
+serde = { version = "1.0", features = ["derive"] }
+serde_derive = "1.0"
+serde_cbor = "0.11.2"
+
+rustacuda = "0.1"
+rustacuda_core = "0.1"
+rustacuda_derive = "0.1"
+
+rand = "*" #TODO: move rand and ark dependencies to dev once random scalar/point generation is done "natively"
+
+[build-dependencies]
+cc = { version = "1.0", features = ["parallel"] }
+
+[dev-dependencies]
+"criterion" = "0.4.0"
+
+[features]
+g2 = []

bn254/build.rs

@@ -0,0 +1,36 @@
+use std::env;
+
+fn main() {
+    //TODO: check cargo features selected
+    //TODO: can conflict/duplicate with make ?
+
+    println!("cargo:rerun-if-env-changed=CXXFLAGS");
+    println!("cargo:rerun-if-changed=./icicle");
+
+    let arch_type = env::var("ARCH_TYPE").unwrap_or(String::from("native"));
+    let stream_type = env::var("DEFAULT_STREAM").unwrap_or(String::from("legacy"));
+
+    let mut arch = String::from("-arch=");
+    arch.push_str(&arch_type);
+    let mut stream = String::from("-default-stream=");
+    stream.push_str(&stream_type);
+
+    let mut nvcc = cc::Build::new();
+
+    println!("Compiling icicle library using arch: {}", &arch);
+
+    if cfg!(feature = "g2") {
+        nvcc.define("G2_DEFINED", None);
+    }
+    nvcc.cuda(true);
+    nvcc.define("FEATURE_BN254", None);
+    nvcc.debug(false);
+    nvcc.flag(&arch);
+    nvcc.flag(&stream);
+    nvcc.shared_flag(false);
+    // nvcc.static_flag(true);
+    nvcc.files([
+        "../icicle-cuda/curves/index.cu",
+    ]);
+    nvcc.compile("ingo_icicle"); //TODO: extension??
+}

bn254/src/basic_structs/field.rs

@@ -0,0 +1,4 @@
+pub trait Field<const NUM_LIMBS: usize> {
+    const MODOLUS: [u32;NUM_LIMBS];
+    const LIMBS: usize = NUM_LIMBS;
+}

bn254/src/basic_structs/mod.rs

@@ -0,0 +1,3 @@
+pub mod field; 
+pub mod scalar; 
+pub mod point; 

bn254/src/basic_structs/point.rs

@@ -0,0 +1,108 @@
+use std::ffi::c_uint;
+
+use ark_bn254::{Fq as Fq_BN254, Fr as Fr_BN254, G1Affine as G1Affine_BN254, G1Projective as G1Projective_BN254};
+
+use ark_ec::AffineCurve;
+use ark_ff::{BigInteger256, PrimeField};
+use std::mem::transmute;
+use ark_ff::Field;
+use icicle_core::utils::{u32_vec_to_u64_vec, u64_vec_to_u32_vec};
+
+use rustacuda_core::DeviceCopy;
+use rustacuda_derive::DeviceCopy;
+
+use super::scalar::{get_fixed_limbs, self};
+
+
+#[derive(Debug, Clone, Copy, DeviceCopy)]
+#[repr(C)]
+pub struct PointT<BF: scalar::ScalarTrait> {
+    pub x: BF,
+    pub y: BF,
+    pub z: BF,
+}
+
+impl<BF: DeviceCopy + scalar::ScalarTrait> Default for PointT<BF> {
+    fn default() -> Self {
+        PointT::zero()
+    }
+}
+
+impl<BF: DeviceCopy + scalar::ScalarTrait> PointT<BF> {
+    pub fn zero() -> Self {
+        PointT {
+            x: BF::zero(),
+            y: BF::one(),
+            z: BF::zero(),
+        }
+    }
+
+    pub fn infinity() -> Self {
+        Self::zero()
+    }
+}
+
+#[derive(Debug, PartialEq, Clone, Copy, DeviceCopy)]
+#[repr(C)]
+pub struct PointAffineNoInfinityT<BF> {
+    pub x: BF,
+    pub y: BF,
+}
+
+impl<BF: scalar::ScalarTrait> Default for PointAffineNoInfinityT<BF> {
+    fn default() -> Self {
+        PointAffineNoInfinityT {
+            x: BF::zero(),
+            y: BF::zero(),
+        }
+    }
+}
+
+impl<BF: Copy + scalar::ScalarTrait> PointAffineNoInfinityT<BF> {
+    ///From u32 limbs x,y
+    pub fn from_limbs(x: &[u32], y: &[u32]) -> Self {
+        PointAffineNoInfinityT {
+            x: BF::from_limbs(x),
+            y: BF::from_limbs(y)
+        }
+    }
+
+    pub fn limbs(&self) -> Vec<u32> {
+        [self.x.limbs(), self.y.limbs()].concat()
+    }
+
+    pub fn to_projective(&self) -> PointT<BF> {
+        PointT {
+            x: self.x,
+            y: self.y,
+            z: BF::one(),
+        }
+    }
+}
+
+impl<BF: Copy + scalar::ScalarTrait> PointT<BF>  {
+    pub fn from_limbs(x: &[u32], y: &[u32], z: &[u32]) -> Self {
+        PointT {
+            x: BF::from_limbs(x),
+            y: BF::from_limbs(y),
+            z: BF::from_limbs(z)
+        }
+    }
+
+    pub fn from_xy_limbs(value: &[u32]) -> PointT<BF> {
+        let l = value.len();
+        assert_eq!(l, 3 * BF::base_limbs(), "length must be 3 * {}", BF::base_limbs());
+        PointT {
+            x: BF::from_limbs(value[..BF::base_limbs()].try_into().unwrap()),
+            y: BF::from_limbs(value[BF::base_limbs()..BF::base_limbs() * 2].try_into().unwrap()),
+            z: BF::from_limbs(value[BF::base_limbs() * 2..].try_into().unwrap())
+        }
+    }
+
+    pub fn to_xy_strip_z(&self) -> PointAffineNoInfinityT<BF> {
+        PointAffineNoInfinityT {
+            x: self.x,
+            y: self.y,
+        }
+    }
+}

bn254/src/basic_structs/scalar.rs

@@ -0,0 +1,102 @@
+use std::ffi::{c_int, c_uint};
+use rand::{rngs::StdRng, RngCore, SeedableRng};
+use rustacuda_core::DeviceCopy;
+use rustacuda_derive::DeviceCopy;
+use std::mem::transmute;
+use rustacuda::prelude::*;
+use rustacuda_core::DevicePointer;
+use rustacuda::memory::{DeviceBox, CopyDestination};
+
+use icicle_core::utils::{u32_vec_to_u64_vec, u64_vec_to_u32_vec};
+
+use std::marker::PhantomData;
+use std::convert::TryInto;
+
+use super::field::{Field, self};
+
+pub fn get_fixed_limbs<const NUM_LIMBS: usize>(val: &[u32]) -> [u32; NUM_LIMBS] {
+    match val.len() {
+        n if n < NUM_LIMBS => {
+            let mut padded: [u32; NUM_LIMBS] = [0; NUM_LIMBS];
+            padded[..val.len()].copy_from_slice(&val);
+            padded
+        }
+        n if n == NUM_LIMBS => val.try_into().unwrap(),
+        _ => panic!("slice has too many elements"),
+    }
+}
+
+pub trait ScalarTrait{
+    fn base_limbs() -> usize;
+    fn zero() -> Self;
+    fn from_limbs(value: &[u32]) -> Self;
+    fn one() -> Self;
+    fn to_bytes_le(&self) -> Vec<u8>;
+    fn limbs(&self) -> &[u32];
+}
+
+#[derive(Debug, PartialEq, Clone, Copy)]
+#[repr(C)]
+pub struct ScalarT<M, const NUM_LIMBS: usize> {
+    pub(crate) phantom: PhantomData<M>,
+    pub(crate) value : [u32; NUM_LIMBS]
+}
+
+impl<M, const NUM_LIMBS: usize> ScalarTrait for ScalarT<M, NUM_LIMBS>
+where
+    M: Field<NUM_LIMBS>,
+{
+
+    fn base_limbs() -> usize {
+        return NUM_LIMBS; 
+    }
+
+    fn zero() -> Self {
+        ScalarT {
+            value: [0u32; NUM_LIMBS],
+            phantom: PhantomData,
+        }
+    }
+
+    fn from_limbs(value: &[u32]) -> Self {
+        Self {
+            value: get_fixed_limbs(value),
+            phantom: PhantomData,
+        }
+    }
+
+    fn one() -> Self {
+        let mut s = [0u32; NUM_LIMBS];
+        s[0] = 1;
+        ScalarT { value: s, phantom: PhantomData }
+    }
+
+    fn to_bytes_le(&self) -> Vec<u8> {
+        self.value
+            .iter()
+            .map(|s| s.to_le_bytes().to_vec())
+            .flatten()
+            .collect::<Vec<_>>()
+    }
+
+    fn limbs(&self) -> &[u32] {
+        &self.value
+    }
+}
+
+impl<M, const NUM_LIMBS: usize> ScalarT<M, NUM_LIMBS> where M: field::Field<NUM_LIMBS>{
+    pub fn from_limbs_le(value: &[u32]) -> ScalarT<M,NUM_LIMBS> {
+        Self::from_limbs(value)
+     }
+ 
+    pub fn from_limbs_be(value: &[u32]) -> ScalarT<M,NUM_LIMBS> {
+         let mut value = value.to_vec();
+         value.reverse();
+         Self::from_limbs_le(&value)
+     }
+ 
+     // Additional Functions
+     pub fn add(&self, other:ScalarT<M, NUM_LIMBS>) -> ScalarT<M,NUM_LIMBS>{  // overload + 
+         return ScalarT{value: [self.value[0] + other.value[0];NUM_LIMBS], phantom: PhantomData }; 
+     }
+}

bn254/src/curve_structs.rs

@@ -0,0 +1,62 @@
+use std::ffi::{c_int, c_uint};
+use rand::{rngs::StdRng, RngCore, SeedableRng};
+use rustacuda_derive::DeviceCopy;
+use std::mem::transmute;
+use rustacuda::prelude::*;
+use rustacuda_core::DevicePointer;
+use rustacuda::memory::{DeviceBox, CopyDestination, DeviceCopy};
+
+use std::marker::PhantomData;
+use std::convert::TryInto;
+
+use crate::basic_structs::point::{PointT, PointAffineNoInfinityT};
+use crate::basic_structs::scalar::ScalarT;
+use crate::basic_structs::field::Field;
+
+
+#[derive(Debug, PartialEq, Clone, Copy,DeviceCopy)]
+#[repr(C)]
+pub struct ScalarField;
+impl Field<8> for ScalarField {
+    const MODOLUS: [u32; 8] = [0x0;8];
+}
+
+#[derive(Debug, PartialEq, Clone, Copy,DeviceCopy)]
+#[repr(C)]
+pub struct BaseField;
+impl Field<8> for BaseField {
+    const MODOLUS: [u32; 8] = [0x0;8];
+}
+
+
+pub type Scalar = ScalarT<ScalarField,8>;
+impl Default for Scalar {
+    fn default() -> Self {
+        Self{value: [0x0;ScalarField::LIMBS], phantom: PhantomData }
+    }
+}
+
+unsafe impl DeviceCopy for Scalar{}
+
+
+pub type Base = ScalarT<BaseField,8>;
+impl Default for Base {
+    fn default() -> Self {
+        Self{value: [0x0;BaseField::LIMBS], phantom: PhantomData }
+    }
+}
+
+unsafe impl DeviceCopy for Base{}
+
+pub type Point = PointT<Base>;
+pub type PointAffineNoInfinity = PointAffineNoInfinityT<Base>;
+
+extern "C" {
+    fn eq(point1: *const Point, point2: *const Point) -> c_uint;
+}
+
+impl PartialEq for Point {
+    fn eq(&self, other: &Self) -> bool {
+        unsafe { eq(self, other) != 0 }
+    }
+}

bn254/src/from_cuda.rs

@@ -0,0 +1,797 @@
+use std::ffi::{c_int, c_uint};
+use ark_std::UniformRand;
+use rand::{rngs::StdRng, RngCore, SeedableRng};
+use rustacuda::CudaFlags;
+use rustacuda::memory::DeviceBox;
+use rustacuda::prelude::{DeviceBuffer, Device, ContextFlags, Context};
+use rustacuda_core::DevicePointer;
+use std::mem::transmute;
+use crate::basic_structs::scalar::ScalarTrait;
+use crate::curve_structs::*;
+use icicle_core::utils::{u32_vec_to_u64_vec, u64_vec_to_u32_vec};
+use std::marker::PhantomData;
+use std::convert::TryInto;
+use ark_bn254::{Fq as Fq_BN254, Fr as Fr_BN254, G1Affine as G1Affine_BN254, G1Projective as G1Projective_BN254};
+use ark_ec::AffineCurve;
+use ark_ff::{BigInteger384, BigInteger256, PrimeField};
+use rustacuda::memory::{CopyDestination, DeviceCopy};
+
+extern "C" {
+    fn msm_cuda(
+        out: *mut Point,
+        points: *const PointAffineNoInfinity,
+        scalars: *const Scalar,
+        count: usize,
+        device_id: usize,
+    ) -> c_uint;
+
+    fn msm_batch_cuda(
+        out: *mut Point,
+        points: *const PointAffineNoInfinity,
+        scalars: *const Scalar,
+        batch_size: usize,
+        msm_size: usize,
+        device_id: usize,
+    ) -> c_uint;
+
+    fn commit_cuda(
+        d_out: DevicePointer<Point>,
+        d_scalars: DevicePointer<Scalar>,
+        d_points: DevicePointer<PointAffineNoInfinity>,
+        count: usize,
+        device_id: usize,
+    ) -> c_uint;
+
+    fn commit_batch_cuda(
+        d_out: DevicePointer<Point>,
+        d_scalars: DevicePointer<Scalar>,
+        d_points: DevicePointer<PointAffineNoInfinity>,
+        count: usize,
+        batch_size: usize,
+        device_id: usize,
+    ) -> c_uint;
+
+    fn build_domain_cuda(domain_size: usize, logn: usize, inverse: bool, device_id: usize) -> DevicePointer<Scalar>;
+
+    fn ntt_cuda(inout: *mut Scalar, n: usize, inverse: bool, device_id: usize) -> c_int;
+
+    fn ecntt_cuda(inout: *mut Point, n: usize, inverse: bool, device_id: usize) -> c_int;
+
+    fn ntt_batch_cuda(
+        inout: *mut Scalar,
+        arr_size: usize,
+        n: usize,
+        inverse: bool,
+    ) -> c_int;
+
+    fn ecntt_batch_cuda(inout: *mut Point, arr_size: usize, n: usize, inverse: bool) -> c_int;
+
+    fn interpolate_scalars_cuda(
+        d_out: DevicePointer<Scalar>,
+        d_evaluations: DevicePointer<Scalar>,
+        d_domain: DevicePointer<Scalar>, 
+        n: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn interpolate_scalars_batch_cuda(
+        d_out: DevicePointer<Scalar>,
+        d_evaluations: DevicePointer<Scalar>,
+        d_domain: DevicePointer<Scalar>,
+        n: usize,
+        batch_size: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn interpolate_points_cuda(
+        d_out: DevicePointer<Point>,
+        d_evaluations: DevicePointer<Point>,
+        d_domain: DevicePointer<Scalar>,
+        n: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn interpolate_points_batch_cuda(
+        d_out: DevicePointer<Point>,
+        d_evaluations: DevicePointer<Point>,
+        d_domain: DevicePointer<Scalar>,
+        n: usize,
+        batch_size: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn evaluate_scalars_cuda(
+        d_out: DevicePointer<Scalar>,
+        d_coefficients: DevicePointer<Scalar>,
+        d_domain: DevicePointer<Scalar>,
+        domain_size: usize,
+        n: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn evaluate_scalars_batch_cuda(
+        d_out: DevicePointer<Scalar>,
+        d_coefficients: DevicePointer<Scalar>,
+        d_domain: DevicePointer<Scalar>,
+        domain_size: usize,
+        n: usize,
+        batch_size: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn evaluate_points_cuda(
+        d_out: DevicePointer<Point>,
+        d_coefficients: DevicePointer<Point>,
+        d_domain: DevicePointer<Scalar>,
+        domain_size: usize,
+        n: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn evaluate_points_batch_cuda(
+        d_out: DevicePointer<Point>,
+        d_coefficients: DevicePointer<Point>,
+        d_domain: DevicePointer<Scalar>,
+        domain_size: usize,
+        n: usize,
+        batch_size: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn evaluate_scalars_on_coset_cuda(
+        d_out: DevicePointer<Scalar>,
+        d_coefficients: DevicePointer<Scalar>,
+        d_domain: DevicePointer<Scalar>,
+        domain_size: usize,
+        n: usize,
+        coset_powers: DevicePointer<Scalar>,
+        device_id: usize
+    ) -> c_int;
+
+    fn evaluate_scalars_on_coset_batch_cuda(
+        d_out: DevicePointer<Scalar>,
+        d_coefficients: DevicePointer<Scalar>,
+        d_domain: DevicePointer<Scalar>,
+        domain_size: usize,
+        n: usize,
+        batch_size: usize,
+        coset_powers: DevicePointer<Scalar>,
+        device_id: usize
+    ) -> c_int;
+
+    fn evaluate_points_on_coset_cuda(
+        d_out: DevicePointer<Point>,
+        d_coefficients: DevicePointer<Point>,
+        d_domain: DevicePointer<Scalar>,
+        domain_size: usize,
+        n: usize,
+        coset_powers: DevicePointer<Scalar>,
+        device_id: usize
+    ) -> c_int;
+
+    fn evaluate_points_on_coset_batch_cuda(
+        d_out: DevicePointer<Point>,
+        d_coefficients: DevicePointer<Point>,
+        d_domain: DevicePointer<Scalar>,
+        domain_size: usize,
+        n: usize,
+        batch_size: usize,
+        coset_powers: DevicePointer<Scalar>,
+        device_id: usize
+    ) -> c_int;
+
+    fn reverse_order_scalars_cuda(
+        d_arr: DevicePointer<Scalar>,
+        n: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn reverse_order_scalars_batch_cuda(
+        d_arr: DevicePointer<Scalar>,
+        n: usize,
+        batch_size: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn reverse_order_points_cuda(
+        d_arr: DevicePointer<Point>,
+        n: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn reverse_order_points_batch_cuda(
+        d_arr: DevicePointer<Point>,
+        n: usize,
+        batch_size: usize,
+        device_id: usize
+    ) -> c_int;
+
+    fn vec_mod_mult_point(
+        inout: *mut Point,
+        scalars: *const Scalar,
+        n_elements: usize,
+        device_id: usize,
+    ) -> c_int;
+
+    fn vec_mod_mult_scalar(
+        inout: *mut Scalar,
+        scalars: *const Scalar,
+        n_elements: usize,
+        device_id: usize,
+    ) -> c_int;
+
+    fn matrix_vec_mod_mult(
+        matrix_flattened: *const Scalar,
+        input: *const Scalar,
+        output: *mut Scalar,
+        n_elements: usize,
+        device_id: usize,
+    ) -> c_int;
+}
+
+pub fn msm(points: &[PointAffineNoInfinity], scalars: &[Scalar], device_id: usize) -> Point {
+    let count = points.len();
+    if count != scalars.len() {
+        todo!("variable length")
+    }
+    let mut ret = Point::zero();
+    unsafe {
+        msm_cuda(
+            &mut ret as *mut _ as *mut Point,
+            points as *const _ as *const PointAffineNoInfinity,
+            scalars as *const _ as *const Scalar,
+            scalars.len(),
+            device_id,
+        )
+    };
+
+    ret
+}
+
+pub fn msm_batch(
+    points: &[PointAffineNoInfinity],
+    scalars: &[Scalar],
+    batch_size: usize,
+    device_id: usize,
+) -> Vec<Point> {
+    let count = points.len();
+    if count != scalars.len() {
+        todo!("variable length")
+    }
+
+    let mut ret = vec![Point::zero(); batch_size];
+
+    unsafe {
+        msm_batch_cuda(
+            &mut ret[0] as *mut _ as *mut Point,
+            points as *const _ as *const PointAffineNoInfinity,
+            scalars as *const _ as *const Scalar,
+            batch_size,
+            count / batch_size,
+            device_id,
+        )
+    };
+
+    ret
+}
+
+pub fn commit(
+    points: &mut DeviceBuffer<PointAffineNoInfinity>,
+    scalars: &mut DeviceBuffer<Scalar>,
+) -> DeviceBox<Point> {
+    let mut res = DeviceBox::new(&Point::zero()).unwrap();
+    unsafe {
+        commit_cuda(
+            res.as_device_ptr(),
+            scalars.as_device_ptr(),
+            points.as_device_ptr(),
+            scalars.len(),
+            0,
+        );
+    }
+    return res;
+}
+
+pub fn commit_batch(
+    points: &mut DeviceBuffer<PointAffineNoInfinity>,
+    scalars: &mut DeviceBuffer<Scalar>,
+    batch_size: usize,
+) -> DeviceBuffer<Point> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(batch_size).unwrap() };
+    unsafe {
+        commit_batch_cuda(
+            res.as_device_ptr(),
+            scalars.as_device_ptr(),
+            points.as_device_ptr(),
+            scalars.len() / batch_size,
+            batch_size,
+            0,
+        );
+    }
+    return res;
+}
+
+/// Compute an in-place NTT on the input data.
+fn ntt_internal(values: &mut [Scalar], device_id: usize, inverse: bool) -> i32 {
+    let ret_code = unsafe {
+        ntt_cuda(
+            values as *mut _ as *mut Scalar,
+            values.len(),
+            inverse,
+            device_id,
+        )
+    };
+    ret_code
+}
+
+pub fn ntt(values: &mut [Scalar], device_id: usize) {
+    ntt_internal(values, device_id, false);
+}
+
+pub fn intt(values: &mut [Scalar], device_id: usize) {
+    ntt_internal(values, device_id, true);
+}
+
+/// Compute an in-place NTT on the input data.
+fn ntt_internal_batch(
+    values: &mut [Scalar],
+    device_id: usize,
+    batch_size: usize,
+    inverse: bool,
+) -> i32 {
+    unsafe {
+        ntt_batch_cuda(
+            values as *mut _ as *mut Scalar,
+            values.len(),
+            batch_size,
+            inverse,
+        )
+    }
+}
+
+pub fn ntt_batch(values: &mut [Scalar], batch_size: usize, device_id: usize) {
+    ntt_internal_batch(values, 0, batch_size, false);
+}
+
+pub fn intt_batch(values: &mut [Scalar], batch_size: usize, device_id: usize) {
+    ntt_internal_batch(values, 0, batch_size, true);
+}
+
+/// Compute an in-place ECNTT on the input data.
+fn ecntt_internal(values: &mut [Point], inverse: bool, device_id: usize) -> i32 {
+    unsafe {
+        ecntt_cuda(
+            values as *mut _ as *mut Point,
+            values.len(),
+            inverse,
+            device_id,
+        )
+    }
+}
+
+pub fn ecntt(values: &mut [Point], device_id: usize) {
+    ecntt_internal(values, false, device_id);
+}
+
+/// Compute an in-place iECNTT on the input data.
+pub fn iecntt(values: &mut [Point], device_id: usize) {
+    ecntt_internal(values, true, device_id);
+}
+
+/// Compute an in-place ECNTT on the input data.
+fn ecntt_internal_batch(
+    values: &mut [Point],
+    device_id: usize,
+    batch_size: usize,
+    inverse: bool,
+) -> i32 {
+    unsafe {
+        ecntt_batch_cuda(
+            values as *mut _ as *mut Point,
+            values.len(),
+            batch_size,
+            inverse,
+        )
+    }
+}
+
+pub fn ecntt_batch(values: &mut [Point], batch_size: usize, device_id: usize) {
+    ecntt_internal_batch(values, 0, batch_size, false);
+}
+
+/// Compute an in-place iECNTT on the input data.
+pub fn iecntt_batch(values: &mut [Point], batch_size: usize, device_id: usize) {
+    ecntt_internal_batch(values, 0, batch_size, true);
+}
+
+pub fn build_domain(domain_size: usize, logn: usize, inverse: bool) -> DeviceBuffer<Scalar> {
+    unsafe {
+        DeviceBuffer::from_raw_parts(build_domain_cuda(
+            domain_size,
+            logn,
+            inverse,
+            0
+        ), domain_size)
+    }
+}
+
+
+pub fn reverse_order_scalars(
+    d_scalars: &mut DeviceBuffer<Scalar>,
+) {
+    unsafe { reverse_order_scalars_cuda(
+        d_scalars.as_device_ptr(),
+        d_scalars.len(),
+        0
+    ); }
+}
+
+pub fn reverse_order_scalars_batch(
+    d_scalars: &mut DeviceBuffer<Scalar>,
+    batch_size: usize,
+) {
+    unsafe { reverse_order_scalars_batch_cuda(
+        d_scalars.as_device_ptr(),
+        d_scalars.len() / batch_size,
+        batch_size,
+        0
+    ); }
+}
+
+pub fn reverse_order_points(
+    d_points: &mut DeviceBuffer<Point>,
+) {
+    unsafe { reverse_order_points_cuda(
+        d_points.as_device_ptr(),
+        d_points.len(),
+        0
+    ); }
+}
+
+pub fn reverse_order_points_batch(
+    d_points: &mut DeviceBuffer<Point>,
+    batch_size: usize,
+) {
+    unsafe { reverse_order_points_batch_cuda(
+        d_points.as_device_ptr(),
+        d_points.len() / batch_size,
+        batch_size,
+        0
+    ); }
+}
+
+pub fn interpolate_scalars(
+    d_evaluations: &mut DeviceBuffer<Scalar>,
+    d_domain: &mut DeviceBuffer<Scalar>
+) -> DeviceBuffer<Scalar> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len()).unwrap() };
+    unsafe { interpolate_scalars_cuda(
+        res.as_device_ptr(),
+        d_evaluations.as_device_ptr(),
+        d_domain.as_device_ptr(),
+        d_domain.len(),
+        0
+    ) };
+    return res;
+}
+
+pub fn interpolate_scalars_batch(
+    d_evaluations: &mut DeviceBuffer<Scalar>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+    batch_size: usize,
+) -> DeviceBuffer<Scalar> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len() * batch_size).unwrap() };
+    unsafe { interpolate_scalars_batch_cuda(
+        res.as_device_ptr(),
+        d_evaluations.as_device_ptr(),
+        d_domain.as_device_ptr(),
+        d_domain.len(),
+        batch_size,
+        0
+    ) };
+    return res;
+}
+
+pub fn interpolate_points(
+    d_evaluations: &mut DeviceBuffer<Point>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+) -> DeviceBuffer<Point> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len()).unwrap() };
+    unsafe { interpolate_points_cuda(
+        res.as_device_ptr(),
+        d_evaluations.as_device_ptr(),
+        d_domain.as_device_ptr(),
+        d_domain.len(),
+        0
+    ) };
+    return res;
+}
+
+pub fn interpolate_points_batch(
+    d_evaluations: &mut DeviceBuffer<Point>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+    batch_size: usize,
+) -> DeviceBuffer<Point> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len() * batch_size).unwrap() };
+    unsafe { interpolate_points_batch_cuda(
+        res.as_device_ptr(),
+        d_evaluations.as_device_ptr(),
+        d_domain.as_device_ptr(),
+        d_domain.len(),
+        batch_size,
+        0
+    ) };
+    return res;
+}
+
+pub fn evaluate_scalars(
+    d_coefficients: &mut DeviceBuffer<Scalar>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+) -> DeviceBuffer<Scalar> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len()).unwrap() };
+    unsafe {
+        evaluate_scalars_cuda(
+            res.as_device_ptr(),
+            d_coefficients.as_device_ptr(),
+            d_domain.as_device_ptr(),
+            d_domain.len(),
+            d_coefficients.len(),
+            0
+        );
+    }
+    return res;
+}
+
+pub fn evaluate_scalars_batch(
+    d_coefficients: &mut DeviceBuffer<Scalar>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+    batch_size: usize,
+) -> DeviceBuffer<Scalar> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len() * batch_size).unwrap() };
+    unsafe {
+        evaluate_scalars_batch_cuda(
+            res.as_device_ptr(),
+            d_coefficients.as_device_ptr(),
+            d_domain.as_device_ptr(),
+            d_domain.len(),
+            d_coefficients.len() / batch_size,
+            batch_size,
+            0
+        );
+    }
+    return res;
+}
+
+pub fn evaluate_points(
+    d_coefficients: &mut DeviceBuffer<Point>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+) -> DeviceBuffer<Point> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len()).unwrap() };
+    unsafe {
+        evaluate_points_cuda(
+            res.as_device_ptr(),
+            d_coefficients.as_device_ptr(),
+            d_domain.as_device_ptr(),
+            d_domain.len(),
+            d_coefficients.len(),
+            0
+        );
+    }
+    return res;
+}
+
+pub fn evaluate_points_batch(
+    d_coefficients: &mut DeviceBuffer<Point>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+    batch_size: usize,
+) -> DeviceBuffer<Point> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len() * batch_size).unwrap() };
+    unsafe {
+        evaluate_points_batch_cuda(
+            res.as_device_ptr(),
+            d_coefficients.as_device_ptr(),
+            d_domain.as_device_ptr(),
+            d_domain.len(),
+            d_coefficients.len() / batch_size,
+            batch_size,
+            0
+        );
+    }
+    return res;
+}
+
+pub fn evaluate_scalars_on_coset(
+    d_coefficients: &mut DeviceBuffer<Scalar>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+    coset_powers: &mut DeviceBuffer<Scalar>,
+) -> DeviceBuffer<Scalar> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len()).unwrap() };
+    unsafe {
+        evaluate_scalars_on_coset_cuda(
+            res.as_device_ptr(),
+            d_coefficients.as_device_ptr(),
+            d_domain.as_device_ptr(),
+            d_domain.len(),
+            d_coefficients.len(),
+            coset_powers.as_device_ptr(),
+            0
+        );
+    }
+    return res;
+}
+
+pub fn evaluate_scalars_on_coset_batch(
+    d_coefficients: &mut DeviceBuffer<Scalar>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+    batch_size: usize,
+    coset_powers: &mut DeviceBuffer<Scalar>,
+) -> DeviceBuffer<Scalar> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len() * batch_size).unwrap() };
+    unsafe {
+        evaluate_scalars_on_coset_batch_cuda(
+            res.as_device_ptr(),
+            d_coefficients.as_device_ptr(),
+            d_domain.as_device_ptr(),
+            d_domain.len(),
+            d_coefficients.len() / batch_size,
+            batch_size,
+            coset_powers.as_device_ptr(),
+            0
+        );
+    }
+    return res;
+}
+
+pub fn evaluate_points_on_coset(
+    d_coefficients: &mut DeviceBuffer<Point>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+    coset_powers: &mut DeviceBuffer<Scalar>,
+) -> DeviceBuffer<Point> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len()).unwrap() };
+    unsafe {
+        evaluate_points_on_coset_cuda(
+            res.as_device_ptr(),
+            d_coefficients.as_device_ptr(),
+            d_domain.as_device_ptr(),
+            d_domain.len(),
+            d_coefficients.len(),
+            coset_powers.as_device_ptr(),
+            0
+        );
+    }
+    return res;
+}
+
+pub fn evaluate_points_on_coset_batch(
+    d_coefficients: &mut DeviceBuffer<Point>,
+    d_domain: &mut DeviceBuffer<Scalar>,
+    batch_size: usize,
+    coset_powers: &mut DeviceBuffer<Scalar>,
+) -> DeviceBuffer<Point> {
+    let mut res = unsafe { DeviceBuffer::uninitialized(d_domain.len() * batch_size).unwrap() };
+    unsafe {
+        evaluate_points_on_coset_batch_cuda(
+            res.as_device_ptr(),
+            d_coefficients.as_device_ptr(),
+            d_domain.as_device_ptr(),
+            d_domain.len(),
+            d_coefficients.len() / batch_size,
+            batch_size,
+            coset_powers.as_device_ptr(),
+            0
+        );
+    }
+    return res;
+}
+
+pub fn multp_vec(a: &mut [Point], b: &[Scalar], device_id: usize) {
+    assert_eq!(a.len(), b.len());
+    unsafe {
+        vec_mod_mult_point(
+            a as *mut _ as *mut Point,
+            b as *const _ as *const Scalar,
+            a.len(),
+            device_id,
+        );
+    }
+}
+
+pub fn mult_sc_vec(a: &mut [Scalar], b: &[Scalar], device_id: usize) {
+    assert_eq!(a.len(), b.len());
+    unsafe {
+        vec_mod_mult_scalar(
+            a as *mut _ as *mut Scalar,
+            b as *const _ as *const Scalar,
+            a.len(),
+            device_id,
+        );
+    }
+}
+
+// Multiply a matrix by a scalar:
+//  `a` - flattenned matrix;
+//  `b` - vector to multiply `a` by;
+pub fn mult_matrix_by_vec(a: &[Scalar], b: &[Scalar], device_id: usize) -> Vec<Scalar> {
+    let mut c = Vec::with_capacity(b.len());
+    for i in 0..b.len() {
+        c.push(Scalar::zero());
+    }
+    unsafe {
+        matrix_vec_mod_mult(
+            a as *const _ as *const Scalar,
+            b as *const _ as *const Scalar,
+            c.as_mut_slice() as *mut _ as *mut Scalar,
+            b.len(),
+            device_id,
+        );
+    }
+    c
+}
+
+pub fn clone_buffer<T: DeviceCopy>(buf: &mut DeviceBuffer<T>) -> DeviceBuffer<T> {
+    let mut buf_cpy = unsafe { DeviceBuffer::uninitialized(buf.len()).unwrap() };
+    unsafe { buf_cpy.copy_from(buf) };
+    return buf_cpy;
+}
+
+pub fn get_rng(seed: Option<u64>) -> Box<dyn RngCore> {
+    let rng: Box<dyn RngCore> = match seed {
+        Some(seed) => Box::new(StdRng::seed_from_u64(seed)),
+        None => Box::new(rand::thread_rng()),
+    };
+    rng
+}
+
+fn set_up_device() {
+    // Set up the context, load the module, and create a stream to run kernels in.
+    rustacuda::init(CudaFlags::empty()).unwrap();
+    let device = Device::get_device(0).unwrap();
+    let _ctx = Context::create_and_push(ContextFlags::MAP_HOST | ContextFlags::SCHED_AUTO, device).unwrap();
+}
+
+pub fn generate_random_points(
+    count: usize,
+    mut rng: Box<dyn RngCore>,
+) -> Vec<PointAffineNoInfinity> {
+    (0..count)
+        .map(|_| Point::from_ark(G1Projective_BN254::rand(&mut rng)).to_xy_strip_z())
+        .collect()
+}
+
+pub fn generate_random_points_proj(count: usize, mut rng: Box<dyn RngCore>) -> Vec<Point> {
+    (0..count)
+        .map(|_| Point::from_ark(G1Projective_BN254::rand(&mut rng)))
+        .collect()
+}
+
+pub fn generate_random_scalars(count: usize, mut rng: Box<dyn RngCore>) -> Vec<Scalar> {
+    (0..count)
+        .map(|_| Scalar::from_ark(Fr_BN254::rand(&mut rng).into_repr()))
+        .collect()
+}
+
+pub fn set_up_points(test_size: usize, log_domain_size: usize, inverse: bool) -> (Vec<Point>, DeviceBuffer<Point>, DeviceBuffer<Scalar>) {
+    set_up_device();
+
+    let d_domain = build_domain(1 << log_domain_size, log_domain_size, inverse);
+
+    let seed = Some(0); // fix the rng to get two equal scalar 
+    let vector = generate_random_points_proj(test_size, get_rng(seed));
+    let mut vector_mut = vector.clone();
+
+    let mut d_vector = DeviceBuffer::from_slice(&vector[..]).unwrap();
+    (vector_mut, d_vector, d_domain)
+}
+
+pub fn set_up_scalars(test_size: usize, log_domain_size: usize, inverse: bool) -> (Vec<Scalar>, DeviceBuffer<Scalar>, DeviceBuffer<Scalar>) {
+    set_up_device();
+
+    let d_domain = build_domain(1 << log_domain_size, log_domain_size, inverse);
+
+    let seed = Some(0); // fix the rng to get two equal scalars
+    let mut vector_mut = generate_random_scalars(test_size, get_rng(seed));
+
+    let mut d_vector = DeviceBuffer::from_slice(&vector_mut[..]).unwrap();
+    (vector_mut, d_vector, d_domain)
+}
+

bn254/src/lib.rs

@@ -0,0 +1,4 @@
+pub mod test_bn254;
+pub mod basic_structs;
+pub mod from_cuda;
+pub mod curve_structs;

bn254/src/test_bn254.rs

@@ -0,0 +1,816 @@
+use std::ffi::{c_int, c_uint};
+use ark_std::UniformRand;
+use rand::{rngs::StdRng, RngCore, SeedableRng};
+use rustacuda::CudaFlags;
+use rustacuda::memory::DeviceBox;
+use rustacuda::prelude::{DeviceBuffer, Device, ContextFlags, Context};
+use rustacuda_core::DevicePointer;
+use std::mem::transmute;
+pub use crate::basic_structs::scalar::ScalarTrait;
+pub use crate::curve_structs::*;
+use icicle_core::utils::{u32_vec_to_u64_vec, u64_vec_to_u32_vec};
+use std::marker::PhantomData;
+use std::convert::TryInto;
+use ark_bn254::{Fq as Fq_BN254, Fr as Fr_BN254, G1Affine as G1Affine_BN254, G1Projective as G1Projective_BN254};
+use ark_ec::AffineCurve;
+use ark_ff::{BigInteger384, BigInteger256, PrimeField};
+use rustacuda::memory::{CopyDestination, DeviceCopy};
+
+
+impl Scalar {
+    pub fn to_biginteger254(&self) -> BigInteger256 {
+        BigInteger256::new(u32_vec_to_u64_vec(&self.limbs()).try_into().unwrap())
+    }
+
+    pub fn to_ark(&self) -> BigInteger256 {
+        BigInteger256::new(u32_vec_to_u64_vec(&self.limbs()).try_into().unwrap())
+    }
+
+    pub fn from_biginteger256(ark: BigInteger256) -> Self {
+        Self{ value: u64_vec_to_u32_vec(&ark.0).try_into().unwrap(), phantom : PhantomData}
+    }
+
+    pub fn to_biginteger256_transmute(&self) -> BigInteger256 {
+        unsafe { transmute(*self) }
+    }
+
+    pub fn from_biginteger_transmute(v: BigInteger256) -> Scalar {
+        Scalar{ value: unsafe{ transmute(v)}, phantom : PhantomData }
+    }
+
+    pub fn to_ark_transmute(&self) -> Fr_BN254 {
+        unsafe { std::mem::transmute(*self) }
+    }
+
+    pub fn from_ark_transmute(v: &Fr_BN254) -> Scalar {
+        unsafe { std::mem::transmute_copy(v) }
+    }
+
+    pub fn to_ark_mod_p(&self) -> Fr_BN254 {
+        Fr_BN254::new(BigInteger256::new(u32_vec_to_u64_vec(&self.limbs()).try_into().unwrap()))
+    }
+
+    pub fn to_ark_repr(&self) -> Fr_BN254 {
+        Fr_BN254::from_repr(BigInteger256::new(u32_vec_to_u64_vec(&self.limbs()).try_into().unwrap())).unwrap()
+    }
+
+    pub fn from_ark(v: BigInteger256) -> Scalar {
+        Self { value : u64_vec_to_u32_vec(&v.0).try_into().unwrap(), phantom: PhantomData}
+    }
+
+}
+
+impl Base {
+    pub fn to_ark(&self) -> BigInteger256 {
+        BigInteger256::new(u32_vec_to_u64_vec(&self.limbs()).try_into().unwrap())
+    }
+
+    pub fn from_ark(ark: BigInteger256) -> Self {
+        Self::from_limbs(&u64_vec_to_u32_vec(&ark.0))
+    }
+}
+
+
+impl Point {
+    pub fn to_ark(&self) -> G1Projective_BN254 {
+        self.to_ark_affine().into_projective()
+    }
+
+    pub fn to_ark_affine(&self) -> G1Affine_BN254 {
+        //TODO: generic conversion
+        use ark_ff::Field;
+        use std::ops::Mul;
+        let proj_x_field = Fq_BN254::from_le_bytes_mod_order(&self.x.to_bytes_le());
+        let proj_y_field = Fq_BN254::from_le_bytes_mod_order(&self.y.to_bytes_le());
+        let proj_z_field = Fq_BN254::from_le_bytes_mod_order(&self.z.to_bytes_le());
+        let inverse_z = proj_z_field.inverse().unwrap();
+        let aff_x = proj_x_field.mul(inverse_z);
+        let aff_y = proj_y_field.mul(inverse_z);
+        G1Affine_BN254::new(aff_x, aff_y, false)
+    }
+
+    pub fn from_ark(ark: G1Projective_BN254) -> Point {
+        use ark_ff::Field;
+        let z_inv = ark.z.inverse().unwrap();
+        let z_invsq = z_inv * z_inv;
+        let z_invq3 = z_invsq * z_inv;
+        Point {
+            x: Base::from_ark((ark.x * z_invsq).into_repr()),
+            y: Base::from_ark((ark.y * z_invq3).into_repr()),
+            z: Base::one(),
+        }
+    }
+}
+
+impl PointAffineNoInfinity {
+
+    pub fn to_ark(&self) -> G1Affine_BN254 {
+        G1Affine_BN254::new(Fq_BN254::new(self.x.to_ark()), Fq_BN254::new(self.y.to_ark()), false)
+    }
+
+    pub fn to_ark_repr(&self) -> G1Affine_BN254 {
+        G1Affine_BN254::new(
+            Fq_BN254::from_repr(self.x.to_ark()).unwrap(),
+            Fq_BN254::from_repr(self.y.to_ark()).unwrap(),
+            false,
+        )
+    }
+
+    pub fn from_ark(p: &G1Affine_BN254) -> Self {
+        PointAffineNoInfinity {
+            x: Base::from_ark(p.x.into_repr()),
+            y: Base::from_ark(p.y.into_repr()),
+        }
+    }
+}
+
+impl Point {
+    pub fn to_affine(&self) -> PointAffineNoInfinity {
+        let ark_affine = self.to_ark_affine();
+        PointAffineNoInfinity {
+            x: Base::from_ark(ark_affine.x.into_repr()),
+            y: Base::from_ark(ark_affine.y.into_repr()),
+        }
+    }
+}
+
+
+#[cfg(test)]
+pub(crate) mod tests_bn254 {
+    use std::ops::Add;
+    use ark_bn254::{Fr, G1Affine, G1Projective};
+    use ark_ec::{msm::VariableBaseMSM, AffineCurve, ProjectiveCurve};
+    use ark_ff::{FftField, Field, Zero, PrimeField};
+    use ark_std::UniformRand;
+    use rustacuda::prelude::{DeviceBuffer, CopyDestination};
+    use crate::curve_structs::{Point, Scalar, Base};
+    use crate::basic_structs::scalar::ScalarTrait;
+    use crate::from_cuda::{generate_random_points, get_rng, generate_random_scalars, msm, msm_batch, set_up_scalars, commit, commit_batch, ntt, intt, generate_random_points_proj, ecntt, iecntt, ntt_batch, ecntt_batch, iecntt_batch, intt_batch, reverse_order_scalars_batch, interpolate_scalars_batch, set_up_points, reverse_order_points, interpolate_points, reverse_order_points_batch, interpolate_points_batch, evaluate_scalars, interpolate_scalars, reverse_order_scalars, evaluate_points, build_domain, evaluate_scalars_on_coset, evaluate_points_on_coset, mult_matrix_by_vec, mult_sc_vec, multp_vec,evaluate_scalars_batch, evaluate_points_batch, evaluate_scalars_on_coset_batch, evaluate_points_on_coset_batch};
+
+    fn random_points_ark_proj(nof_elements: usize) -> Vec<G1Projective> {
+        let mut rng = ark_std::rand::thread_rng();
+        let mut points_ga: Vec<G1Projective> = Vec::new();
+        for _ in 0..nof_elements {
+            let aff = G1Projective::rand(&mut rng);
+            points_ga.push(aff);
+        }
+        points_ga
+    }
+
+    fn ecntt_arc_naive(
+        points: &Vec<G1Projective>,
+        size: usize,
+        inverse: bool,
+    ) -> Vec<G1Projective> {
+        let mut result: Vec<G1Projective> = Vec::new();
+        for _ in 0..size {
+            result.push(G1Projective::zero());
+        }
+        let rou: Fr;
+        if !inverse {
+            rou = Fr::get_root_of_unity(size).unwrap();
+        } else {
+            rou = Fr::inverse(&Fr::get_root_of_unity(size).unwrap()).unwrap();
+        }
+        for k in 0..size {
+            for l in 0..size {
+                let pow: [u64; 1] = [(l * k).try_into().unwrap()];
+                let mul_rou = Fr::pow(&rou, &pow);
+                result[k] = result[k].add(points[l].into_affine().mul(mul_rou));
+            }
+        }
+        if inverse {
+            let size2 = size as u64;
+            for k in 0..size {
+                let multfactor = Fr::inverse(&Fr::from(size2)).unwrap();
+                result[k] = result[k].into_affine().mul(multfactor);
+            }
+        }
+        return result;
+    }
+
+    fn check_eq(points: &Vec<G1Projective>, points2: &Vec<G1Projective>) -> bool {
+        let mut eq = true;
+        for i in 0..points.len() {
+            if points2[i].ne(&points[i]) {
+                eq = false;
+                break;
+            }
+        }
+        return eq;
+    }
+
+    fn test_naive_ark_ecntt(size: usize) {
+        let points = random_points_ark_proj(size);
+        let result1: Vec<G1Projective> = ecntt_arc_naive(&points, size, false);
+        let result2: Vec<G1Projective> = ecntt_arc_naive(&result1, size, true);
+        assert!(!check_eq(&result2, &result1));
+        assert!(check_eq(&result2, &points));
+    }
+
+    #[test]
+    fn test_msm() {
+        let test_sizes = [6, 9];
+
+        for pow2 in test_sizes {
+            let count = 1 << pow2;
+            let seed = None; // set Some to provide seed
+            let points = generate_random_points(count, get_rng(seed));
+            let scalars = generate_random_scalars(count, get_rng(seed));
+
+            let msm_result = msm(&points, &scalars, 0);
+
+            let point_r_ark: Vec<_> = points.iter().map(|x| x.to_ark_repr()).collect();
+            let scalars_r_ark: Vec<_> = scalars.iter().map(|x| x.to_ark()).collect();
+
+            let msm_result_ark = VariableBaseMSM::multi_scalar_mul(&point_r_ark, &scalars_r_ark);
+
+            assert_eq!(msm_result.to_ark_affine(), msm_result_ark);
+            assert_eq!(msm_result.to_ark(), msm_result_ark);
+            assert_eq!(
+                msm_result.to_ark_affine(),
+                Point::from_ark(msm_result_ark).to_ark_affine()
+            );
+        }
+    }
+
+    #[test]
+    fn test_batch_msm() {
+        for batch_pow2 in [2, 4] {
+            for pow2 in [4, 6] {
+                let msm_size = 1 << pow2;
+                let batch_size = 1 << batch_pow2;
+                let seed = None; // set Some to provide seed
+                let points_batch = generate_random_points(msm_size * batch_size, get_rng(seed));
+                let scalars_batch = generate_random_scalars(msm_size * batch_size, get_rng(seed));
+
+                let point_r_ark: Vec<_> = points_batch.iter().map(|x| x.to_ark_repr()).collect();
+                let scalars_r_ark: Vec<_> = scalars_batch.iter().map(|x| x.to_ark()).collect();
+
+                let expected: Vec<_> = point_r_ark
+                    .chunks(msm_size)
+                    .zip(scalars_r_ark.chunks(msm_size))
+                    .map(|p| Point::from_ark(VariableBaseMSM::multi_scalar_mul(p.0, p.1)))
+                    .collect();
+
+                let result = msm_batch(&points_batch, &scalars_batch, batch_size, 0);
+
+                assert_eq!(result, expected);
+            }
+        }
+    }
+
+    #[test]
+    fn test_commit() {
+        let test_size = 1 << 8;
+        let seed = Some(0);
+        let (mut scalars, mut d_scalars, _) = set_up_scalars(test_size, 0, false);
+        let mut points = generate_random_points(test_size, get_rng(seed));
+        let mut d_points = DeviceBuffer::from_slice(&points[..]).unwrap();
+
+        let msm_result = msm(&points, &scalars, 0);
+        let mut d_commit_result = commit(&mut d_points, &mut d_scalars);
+        let mut h_commit_result = Point::zero();
+        d_commit_result.copy_to(&mut h_commit_result).unwrap();
+
+        assert_eq!(msm_result, h_commit_result);
+        assert_ne!(msm_result, Point::zero());
+        assert_ne!(h_commit_result, Point::zero());
+    }
+
+    #[test]
+    fn test_batch_commit() {
+        let batch_size = 4;
+        let test_size = 1 << 12;
+        let seed = Some(0);
+        let (scalars, mut d_scalars, _) = set_up_scalars(test_size * batch_size, 0, false);
+        let points = generate_random_points(test_size * batch_size, get_rng(seed));
+        let mut d_points = DeviceBuffer::from_slice(&points[..]).unwrap();
+
+        let msm_result = msm_batch(&points, &scalars, batch_size, 0);
+        let mut d_commit_result = commit_batch(&mut d_points, &mut d_scalars, batch_size);
+        let mut h_commit_result: Vec<Point> = (0..batch_size).map(|_| Point::zero()).collect();
+        d_commit_result.copy_to(&mut h_commit_result[..]).unwrap();
+
+        assert_eq!(msm_result, h_commit_result);
+        for h in h_commit_result {
+            assert_ne!(h, Point::zero());
+        }
+    }
+
+    #[test]
+    fn test_ntt() {
+        //NTT
+        let seed = None; //some value to fix the rng
+        let test_size = 1 << 3;
+
+        let scalars = generate_random_scalars(test_size, get_rng(seed));
+
+        let mut ntt_result = scalars.clone();
+        ntt(&mut ntt_result, 0);
+
+        assert_ne!(ntt_result, scalars);
+
+        let mut intt_result = ntt_result.clone();
+
+        intt(&mut intt_result, 0);
+
+        assert_eq!(intt_result, scalars);
+
+        //ECNTT
+        let points_proj = generate_random_points_proj(test_size, get_rng(seed));
+
+        test_naive_ark_ecntt(test_size);
+
+        assert!(points_proj[0].to_ark().into_affine().is_on_curve());
+
+        //naive ark
+        let points_proj_ark = points_proj
+            .iter()
+            .map(|p| p.to_ark())
+            .collect::<Vec<G1Projective>>();
+
+        let ecntt_result_naive = ecntt_arc_naive(&points_proj_ark, points_proj_ark.len(), false);
+
+        let iecntt_result_naive = ecntt_arc_naive(&ecntt_result_naive, points_proj_ark.len(), true);
+
+        assert_eq!(points_proj_ark, iecntt_result_naive);
+
+        //ingo gpu
+        let mut ecntt_result = points_proj.to_vec();
+        ecntt(&mut ecntt_result, 0);
+
+        assert_ne!(ecntt_result, points_proj);
+
+        let mut iecntt_result = ecntt_result.clone();
+        iecntt(&mut iecntt_result, 0);
+
+        assert_eq!(
+            iecntt_result_naive,
+            points_proj
+                .iter()
+                .map(|p| p.to_ark_affine())
+                .collect::<Vec<G1Affine>>()
+        );
+        assert_eq!(
+            iecntt_result
+                .iter()
+                .map(|p| p.to_ark_affine())
+                .collect::<Vec<G1Affine>>(),
+            points_proj
+                .iter()
+                .map(|p| p.to_ark_affine())
+                .collect::<Vec<G1Affine>>()
+        );
+    }
+
+    #[test]
+    fn test_ntt_batch() {
+        //NTT
+        let seed = None; //some value to fix the rng
+        let test_size = 1 << 5;
+        let batches = 4;
+
+        let scalars_batch: Vec<Scalar> =
+            generate_random_scalars(test_size * batches, get_rng(seed));
+
+        let mut scalar_vec_of_vec: Vec<Vec<Scalar>> = Vec::new();
+
+        for i in 0..batches {
+            scalar_vec_of_vec.push(scalars_batch[i * test_size..(i + 1) * test_size].to_vec());
+        }
+
+        let mut ntt_result = scalars_batch.clone();
+
+        // do batch ntt
+        ntt_batch(&mut ntt_result, test_size, 0);
+
+        let mut ntt_result_vec_of_vec = Vec::new();
+
+        // do ntt for every chunk
+        for i in 0..batches {
+            ntt_result_vec_of_vec.push(scalar_vec_of_vec[i].clone());
+            ntt(&mut ntt_result_vec_of_vec[i], 0);
+        }
+
+        // check that the ntt of each vec of scalars is equal to the intt of the specific batch
+        for i in 0..batches {
+            assert_eq!(
+                ntt_result_vec_of_vec[i],
+                ntt_result[i * test_size..(i + 1) * test_size]
+            );
+        }
+
+        // check that ntt output is different from input
+        assert_ne!(ntt_result, scalars_batch);
+
+        let mut intt_result = ntt_result.clone();
+
+        // do batch intt
+        intt_batch(&mut intt_result, test_size, 0);
+
+        let mut intt_result_vec_of_vec = Vec::new();
+
+        // do intt for every chunk
+        for i in 0..batches {
+            intt_result_vec_of_vec.push(ntt_result_vec_of_vec[i].clone());
+            intt(&mut intt_result_vec_of_vec[i], 0);
+        }
+
+        // check that the intt of each vec of scalars is equal to the intt of the specific batch
+        for i in 0..batches {
+            assert_eq!(
+                intt_result_vec_of_vec[i],
+                intt_result[i * test_size..(i + 1) * test_size]
+            );
+        }
+
+        assert_eq!(intt_result, scalars_batch);
+
+        // //ECNTT
+        let points_proj = generate_random_points_proj(test_size * batches, get_rng(seed));
+
+        let mut points_vec_of_vec: Vec<Vec<Point>> = Vec::new();
+
+        for i in 0..batches {
+            points_vec_of_vec.push(points_proj[i * test_size..(i + 1) * test_size].to_vec());
+        }
+
+        let mut ntt_result_points = points_proj.clone();
+
+        // do batch ecintt
+        ecntt_batch(&mut ntt_result_points, test_size, 0);
+
+        let mut ntt_result_points_vec_of_vec = Vec::new();
+
+        for i in 0..batches {
+            ntt_result_points_vec_of_vec.push(points_vec_of_vec[i].clone());
+            ecntt(&mut ntt_result_points_vec_of_vec[i], 0);
+        }
+
+        for i in 0..batches {
+            assert_eq!(
+                ntt_result_points_vec_of_vec[i],
+                ntt_result_points[i * test_size..(i + 1) * test_size]
+            );
+        }
+
+        assert_ne!(ntt_result_points, points_proj);
+
+        let mut intt_result_points = ntt_result_points.clone();
+
+        // do batch ecintt
+        iecntt_batch(&mut intt_result_points, test_size, 0);
+
+        let mut intt_result_points_vec_of_vec = Vec::new();
+
+        // do ecintt for every chunk
+        for i in 0..batches {
+            intt_result_points_vec_of_vec.push(ntt_result_points_vec_of_vec[i].clone());
+            iecntt(&mut intt_result_points_vec_of_vec[i], 0);
+        }
+
+        // check that the ecintt of each vec of scalars is equal to the intt of the specific batch
+        for i in 0..batches {
+            assert_eq!(
+                intt_result_points_vec_of_vec[i],
+                intt_result_points[i * test_size..(i + 1) * test_size]
+            );
+        }
+
+        assert_eq!(intt_result_points, points_proj);
+    }
+
+    #[test]
+    fn test_scalar_interpolation() {
+        let log_test_size = 7;
+        let test_size = 1 << log_test_size;
+        let (mut evals_mut, mut d_evals, mut d_domain) = set_up_scalars(test_size, log_test_size, true);
+
+        reverse_order_scalars(&mut d_evals);
+        let mut d_coeffs = interpolate_scalars(&mut d_evals, &mut d_domain);
+        intt(&mut evals_mut, 0);
+        let mut h_coeffs: Vec<Scalar> = (0..test_size).map(|_| Scalar::zero()).collect();
+        d_coeffs.copy_to(&mut h_coeffs[..]).unwrap();
+
+        assert_eq!(h_coeffs, evals_mut);
+    }
+
+    #[test]
+    fn test_scalar_batch_interpolation() {
+        let batch_size = 4;
+        let log_test_size = 10;
+        let test_size = 1 << log_test_size;
+        let (mut evals_mut, mut d_evals, mut d_domain) = set_up_scalars(test_size * batch_size, log_test_size, true);
+
+        reverse_order_scalars_batch(&mut d_evals, batch_size);
+        let mut d_coeffs = interpolate_scalars_batch(&mut d_evals, &mut d_domain, batch_size);
+        intt_batch(&mut evals_mut, test_size, 0);
+        let mut h_coeffs: Vec<Scalar> = (0..test_size * batch_size).map(|_| Scalar::zero()).collect();
+        d_coeffs.copy_to(&mut h_coeffs[..]).unwrap();
+
+        assert_eq!(h_coeffs, evals_mut);
+    }
+
+    #[test]
+    fn test_point_interpolation() {
+        let log_test_size = 6;
+        let test_size = 1 << log_test_size;
+        let (mut evals_mut, mut d_evals, mut d_domain) = set_up_points(test_size, log_test_size, true);
+
+        reverse_order_points(&mut d_evals);
+        let mut d_coeffs = interpolate_points(&mut d_evals, &mut d_domain);
+        iecntt(&mut evals_mut[..], 0);
+        let mut h_coeffs: Vec<Point> = (0..test_size).map(|_| Point::zero()).collect();
+        d_coeffs.copy_to(&mut h_coeffs[..]).unwrap();
+        
+        assert_eq!(h_coeffs, *evals_mut);
+        for h in h_coeffs.iter() {
+            assert_ne!(*h, Point::zero());
+        }
+    }
+
+    #[test]
+    fn test_point_batch_interpolation() {
+        let batch_size = 4;
+        let log_test_size = 6;
+        let test_size = 1 << log_test_size;
+        let (mut evals_mut, mut d_evals, mut d_domain) = set_up_points(test_size * batch_size, log_test_size, true);
+
+        reverse_order_points_batch(&mut d_evals, batch_size);
+        let mut d_coeffs = interpolate_points_batch(&mut d_evals, &mut d_domain, batch_size);
+        iecntt_batch(&mut evals_mut[..], test_size, 0);
+        let mut h_coeffs: Vec<Point> = (0..test_size * batch_size).map(|_| Point::zero()).collect();
+        d_coeffs.copy_to(&mut h_coeffs[..]).unwrap();
+        
+        assert_eq!(h_coeffs, *evals_mut);
+        for h in h_coeffs.iter() {
+            assert_ne!(*h, Point::zero());
+        }
+    }
+
+    #[test]
+    fn test_scalar_evaluation() {
+        let log_test_domain_size = 8;
+        let coeff_size = 1 << 6;
+        let (h_coeffs, mut d_coeffs, mut d_domain) = set_up_scalars(coeff_size, log_test_domain_size, false);
+        let (_, _, mut d_domain_inv) = set_up_scalars(0, log_test_domain_size, true);
+
+        let mut d_evals = evaluate_scalars(&mut d_coeffs, &mut d_domain);
+        let mut d_coeffs_domain = interpolate_scalars(&mut d_evals, &mut d_domain_inv);
+        let mut h_coeffs_domain: Vec<Scalar> = (0..1 << log_test_domain_size).map(|_| Scalar::zero()).collect();
+        d_coeffs_domain.copy_to(&mut h_coeffs_domain[..]).unwrap();
+
+        assert_eq!(h_coeffs, h_coeffs_domain[..coeff_size]);
+        for i in coeff_size.. (1 << log_test_domain_size) {
+            assert_eq!(Scalar::zero(), h_coeffs_domain[i]);
+        }
+    }
+
+    #[test]
+    fn test_scalar_batch_evaluation() {
+        let batch_size = 6;
+        let log_test_domain_size = 8;
+        let domain_size = 1 << log_test_domain_size;
+        let coeff_size = 1 << 6;
+        let (h_coeffs, mut d_coeffs, mut d_domain) = set_up_scalars(coeff_size * batch_size, log_test_domain_size, false);
+        let (_, _, mut d_domain_inv) = set_up_scalars(0, log_test_domain_size, true);
+
+        let mut d_evals = evaluate_scalars_batch(&mut d_coeffs, &mut d_domain, batch_size);
+        let mut d_coeffs_domain = interpolate_scalars_batch(&mut d_evals, &mut d_domain_inv, batch_size);
+        let mut h_coeffs_domain: Vec<Scalar> = (0..domain_size * batch_size).map(|_| Scalar::zero()).collect();
+        d_coeffs_domain.copy_to(&mut h_coeffs_domain[..]).unwrap();
+
+        for j in 0..batch_size {
+            assert_eq!(h_coeffs[j * coeff_size..(j + 1) * coeff_size], h_coeffs_domain[j * domain_size..j * domain_size + coeff_size]);
+            for i in coeff_size..domain_size {
+                assert_eq!(Scalar::zero(), h_coeffs_domain[j * domain_size + i]);
+            }
+        }
+    }
+
+    #[test]
+    fn test_point_evaluation() {
+        let log_test_domain_size = 7;
+        let coeff_size = 1 << 7;
+        let (h_coeffs, mut d_coeffs, mut d_domain) = set_up_points(coeff_size, log_test_domain_size, false);
+        let (_, _, mut d_domain_inv) = set_up_points(0, log_test_domain_size, true);
+
+        let mut d_evals = evaluate_points(&mut d_coeffs, &mut d_domain);
+        let mut d_coeffs_domain = interpolate_points(&mut d_evals, &mut d_domain_inv);
+        let mut h_coeffs_domain: Vec<Point> = (0..1 << log_test_domain_size).map(|_| Point::zero()).collect();
+        d_coeffs_domain.copy_to(&mut h_coeffs_domain[..]).unwrap();
+
+        assert_eq!(h_coeffs[..], h_coeffs_domain[..coeff_size]);
+        for i in coeff_size..(1 << log_test_domain_size) {
+            assert_eq!(Point::zero(), h_coeffs_domain[i]);
+        }
+        for i in 0..coeff_size {
+            assert_ne!(h_coeffs_domain[i], Point::zero());
+        }
+    }
+
+    #[test]
+    fn test_point_batch_evaluation() {
+        let batch_size = 4;
+        let log_test_domain_size = 6;
+        let domain_size = 1 << log_test_domain_size;
+        let coeff_size = 1 << 5;
+        let (h_coeffs, mut d_coeffs, mut d_domain) = set_up_points(coeff_size * batch_size, log_test_domain_size, false);
+        let (_, _, mut d_domain_inv) = set_up_points(0, log_test_domain_size, true);
+
+        let mut d_evals = evaluate_points_batch(&mut d_coeffs, &mut d_domain, batch_size);
+        let mut d_coeffs_domain = interpolate_points_batch(&mut d_evals, &mut d_domain_inv, batch_size);
+        let mut h_coeffs_domain: Vec<Point> = (0..domain_size * batch_size).map(|_| Point::zero()).collect();
+        d_coeffs_domain.copy_to(&mut h_coeffs_domain[..]).unwrap();
+
+        for j in 0..batch_size {
+            assert_eq!(h_coeffs[j * coeff_size..(j + 1) * coeff_size], h_coeffs_domain[j * domain_size..(j * domain_size + coeff_size)]);
+            for i in coeff_size..domain_size {
+                assert_eq!(Point::zero(), h_coeffs_domain[j * domain_size + i]);
+            }
+            for i in j * domain_size..(j * domain_size + coeff_size) {
+                assert_ne!(h_coeffs_domain[i], Point::zero());
+            }
+        }
+    }
+
+    #[test]
+    fn test_scalar_evaluation_on_trivial_coset() {
+        // checks that the evaluations on the subgroup is the same as on the coset generated by 1
+        let log_test_domain_size = 8;
+        let coeff_size = 1 << 6;
+        let (_, mut d_coeffs, mut d_domain) = set_up_scalars(coeff_size, log_test_domain_size, false);
+        let (_, _, mut d_domain_inv) = set_up_scalars(coeff_size, log_test_domain_size, true);
+        let mut d_trivial_coset_powers = build_domain(1 << log_test_domain_size, 0, false);
+
+        let mut d_evals = evaluate_scalars(&mut d_coeffs, &mut d_domain);
+        let mut h_coeffs: Vec<Scalar> = (0..1 << log_test_domain_size).map(|_| Scalar::zero()).collect();
+        d_evals.copy_to(&mut h_coeffs[..]).unwrap();
+        let mut d_evals_coset = evaluate_scalars_on_coset(&mut d_coeffs, &mut d_domain, &mut d_trivial_coset_powers);
+        let mut h_evals_coset: Vec<Scalar> = (0..1 << log_test_domain_size).map(|_| Scalar::zero()).collect();
+        d_evals_coset.copy_to(&mut h_evals_coset[..]).unwrap();
+
+        assert_eq!(h_coeffs, h_evals_coset);
+    }
+
+    #[test]
+    fn test_scalar_evaluation_on_coset() {
+        // checks that evaluating a polynomial on a subgroup and its coset is the same as evaluating on a 2x larger subgroup 
+        let log_test_size = 8;
+        let test_size = 1 << log_test_size;
+        let (_, mut d_coeffs, mut d_domain) = set_up_scalars(test_size, log_test_size, false);
+        let (_, _, mut d_large_domain) = set_up_scalars(0, log_test_size + 1, false);
+        let mut d_coset_powers = build_domain(test_size, log_test_size + 1, false);
+
+        let mut d_evals_large = evaluate_scalars(&mut d_coeffs, &mut d_large_domain);
+        let mut h_evals_large: Vec<Scalar> = (0..2 * test_size).map(|_| Scalar::zero()).collect();
+        d_evals_large.copy_to(&mut h_evals_large[..]).unwrap();
+        let mut d_evals = evaluate_scalars(&mut d_coeffs, &mut d_domain);
+        let mut h_evals: Vec<Scalar> = (0..test_size).map(|_| Scalar::zero()).collect();
+        d_evals.copy_to(&mut h_evals[..]).unwrap();
+        let mut d_evals_coset = evaluate_scalars_on_coset(&mut d_coeffs, &mut d_domain, &mut d_coset_powers);
+        let mut h_evals_coset: Vec<Scalar> = (0..test_size).map(|_| Scalar::zero()).collect();
+        d_evals_coset.copy_to(&mut h_evals_coset[..]).unwrap();
+
+        assert_eq!(h_evals[..], h_evals_large[..test_size]);
+        assert_eq!(h_evals_coset[..], h_evals_large[test_size..2 * test_size]);
+    }
+
+    #[test]
+    fn test_scalar_batch_evaluation_on_coset() {
+        // checks that evaluating a polynomial on a subgroup and its coset is the same as evaluating on a 2x larger subgroup 
+        let batch_size = 4;
+        let log_test_size = 6;
+        let test_size = 1 << log_test_size;
+        let (_, mut d_coeffs, mut d_domain) = set_up_scalars(test_size * batch_size, log_test_size, false);
+        let (_, _, mut d_large_domain) = set_up_scalars(0, log_test_size + 1, false);
+        let mut d_coset_powers = build_domain(test_size, log_test_size + 1, false);
+
+        let mut d_evals_large = evaluate_scalars_batch(&mut d_coeffs, &mut d_large_domain, batch_size);
+        let mut h_evals_large: Vec<Scalar> = (0..2 * test_size * batch_size).map(|_| Scalar::zero()).collect();
+        d_evals_large.copy_to(&mut h_evals_large[..]).unwrap();
+        let mut d_evals = evaluate_scalars_batch(&mut d_coeffs, &mut d_domain, batch_size);
+        let mut h_evals: Vec<Scalar> = (0..test_size * batch_size).map(|_| Scalar::zero()).collect();
+        d_evals.copy_to(&mut h_evals[..]).unwrap();
+        let mut d_evals_coset = evaluate_scalars_on_coset_batch(&mut d_coeffs, &mut d_domain, batch_size, &mut d_coset_powers);
+        let mut h_evals_coset: Vec<Scalar> = (0..test_size * batch_size).map(|_| Scalar::zero()).collect();
+        d_evals_coset.copy_to(&mut h_evals_coset[..]).unwrap();
+
+        for i in 0..batch_size {
+            assert_eq!(h_evals_large[2 * i * test_size..(2 * i + 1) * test_size], h_evals[i * test_size..(i + 1) * test_size]);
+            assert_eq!(h_evals_large[(2 * i + 1) * test_size..(2 * i + 2) * test_size], h_evals_coset[i * test_size..(i + 1) * test_size]);
+        }
+    }
+
+    #[test]
+    fn test_point_evaluation_on_coset() {
+        // checks that evaluating a polynomial on a subgroup and its coset is the same as evaluating on a 2x larger subgroup 
+        let log_test_size = 8;
+        let test_size = 1 << log_test_size;
+        let (_, mut d_coeffs, mut d_domain) = set_up_points(test_size, log_test_size, false);
+        let (_, _, mut d_large_domain) = set_up_points(0, log_test_size + 1, false);
+        let mut d_coset_powers = build_domain(test_size, log_test_size + 1, false);
+
+        let mut d_evals_large = evaluate_points(&mut d_coeffs, &mut d_large_domain);
+        let mut h_evals_large: Vec<Point> = (0..2 * test_size).map(|_| Point::zero()).collect();
+        d_evals_large.copy_to(&mut h_evals_large[..]).unwrap();
+        let mut d_evals = evaluate_points(&mut d_coeffs, &mut d_domain);
+        let mut h_evals: Vec<Point> = (0..test_size).map(|_| Point::zero()).collect();
+        d_evals.copy_to(&mut h_evals[..]).unwrap();
+        let mut d_evals_coset = evaluate_points_on_coset(&mut d_coeffs, &mut d_domain, &mut d_coset_powers);
+        let mut h_evals_coset: Vec<Point> = (0..test_size).map(|_| Point::zero()).collect();
+        d_evals_coset.copy_to(&mut h_evals_coset[..]).unwrap();
+
+        assert_eq!(h_evals[..], h_evals_large[..test_size]);
+        assert_eq!(h_evals_coset[..], h_evals_large[test_size..2 * test_size]);
+        for i in 0..test_size {
+            assert_ne!(h_evals[i], Point::zero());
+            assert_ne!(h_evals_coset[i], Point::zero());
+            assert_ne!(h_evals_large[2 * i], Point::zero());
+            assert_ne!(h_evals_large[2 * i + 1], Point::zero());
+        }
+    }
+
+    #[test]
+    fn test_point_batch_evaluation_on_coset() {
+        // checks that evaluating a polynomial on a subgroup and its coset is the same as evaluating on a 2x larger subgroup 
+        let batch_size = 2;
+        let log_test_size = 6;
+        let test_size = 1 << log_test_size;
+        let (_, mut d_coeffs, mut d_domain) = set_up_points(test_size * batch_size, log_test_size, false);
+        let (_, _, mut d_large_domain) = set_up_points(0, log_test_size + 1, false);
+        let mut d_coset_powers = build_domain(test_size, log_test_size + 1, false);
+
+        let mut d_evals_large = evaluate_points_batch(&mut d_coeffs, &mut d_large_domain, batch_size);
+        let mut h_evals_large: Vec<Point> = (0..2 * test_size * batch_size).map(|_| Point::zero()).collect();
+        d_evals_large.copy_to(&mut h_evals_large[..]).unwrap();
+        let mut d_evals = evaluate_points_batch(&mut d_coeffs, &mut d_domain, batch_size);
+        let mut h_evals: Vec<Point> = (0..test_size * batch_size).map(|_| Point::zero()).collect();
+        d_evals.copy_to(&mut h_evals[..]).unwrap();
+        let mut d_evals_coset = evaluate_points_on_coset_batch(&mut d_coeffs, &mut d_domain, batch_size, &mut d_coset_powers);
+        let mut h_evals_coset: Vec<Point> = (0..test_size * batch_size).map(|_| Point::zero()).collect();
+        d_evals_coset.copy_to(&mut h_evals_coset[..]).unwrap();
+
+        for i in 0..batch_size {
+            assert_eq!(h_evals_large[2 * i * test_size..(2 * i + 1) * test_size], h_evals[i * test_size..(i + 1) * test_size]);
+            assert_eq!(h_evals_large[(2 * i + 1) * test_size..(2 * i + 2) * test_size], h_evals_coset[i * test_size..(i + 1) * test_size]);
+        }
+        for i in 0..test_size * batch_size {
+            assert_ne!(h_evals[i], Point::zero());
+            assert_ne!(h_evals_coset[i], Point::zero());
+            assert_ne!(h_evals_large[2 * i], Point::zero());
+            assert_ne!(h_evals_large[2 * i + 1], Point::zero());
+        }
+    }
+
+    // testing matrix multiplication by comparing the result of FFT with the naive multiplication by the DFT matrix
+    #[test]
+    fn test_matrix_multiplication() {
+        let seed = None; // some value to fix the rng
+        let test_size = 1 << 5;
+        let rou = Fr::get_root_of_unity(test_size).unwrap();
+        let matrix_flattened: Vec<Scalar> = (0..test_size).map(
+            |row_num| { (0..test_size).map( 
+                |col_num| {
+                    let pow: [u64; 1] = [(row_num * col_num).try_into().unwrap()];
+                    Scalar::from_ark(Fr::pow(&rou, &pow).into_repr())
+                }).collect::<Vec<Scalar>>()
+            }).flatten().collect::<Vec<_>>();
+        let vector: Vec<Scalar> = generate_random_scalars(test_size, get_rng(seed));
+
+        let result = mult_matrix_by_vec(&matrix_flattened, &vector, 0);
+        let mut ntt_result = vector.clone();
+        ntt(&mut ntt_result, 0);
+        
+        // we don't use the same roots of unity as arkworks, so the results are permutations
+        // of one another and the only guaranteed fixed scalars are the following ones:
+        assert_eq!(result[0], ntt_result[0]);
+        assert_eq!(result[test_size >> 1], ntt_result[test_size >> 1]);
+    }
+
+    #[test]
+    #[allow(non_snake_case)]
+    fn test_vec_scalar_mul() {
+        let mut intoo = [Scalar::one(), Scalar::one(), Scalar::zero()];
+        let expected = [Scalar::one(), Scalar::zero(), Scalar::zero()];
+        mult_sc_vec(&mut intoo, &expected, 0);
+        assert_eq!(intoo, expected);
+    }
+
+    #[test]
+    #[allow(non_snake_case)]
+    fn test_vec_point_mul() {
+        let dummy_one = Point {
+            x: Base::one(),
+            y: Base::one(),
+            z: Base::one(),
+        };
+
+        let mut inout = [dummy_one, dummy_one, Point::zero()];
+        let scalars = [Scalar::one(), Scalar::zero(), Scalar::zero()];
+        let expected = [dummy_one, Point::zero(), Point::zero()];
+        multp_vec(&mut inout, &scalars, 0);
+        assert_eq!(inout, expected);
+    }
+}

curve_parameters/bls12_377.json

@@ -1,20 +1,13 @@
 {
     "curve_name" : "bls12_377",
-    "modulus_p" : 8444461749428370424248824938781546531375899335154063827935233455917409239041,
+    "modolus_p" : 8444461749428370424248824938781546531375899335154063827935233455917409239041,
     "bit_count_p" : 253,
     "limb_p" :  8,
     "ntt_size" : 32,
-    "modulus_q" : 258664426012969094010652733694893533536393512754914660539884262666720468348340822774968888139573360124440321458177,
+    "modolus_q" : 258664426012969094010652733694893533536393512754914660539884262666720468348340822774968888139573360124440321458177,
     "bit_count_q" : 377,
     "limb_q" : 12,
-    "root_of_unity" : 5928890464389279575069867463136436689218492512582288454256978381122364252082,
     "weierstrass_b" : 1,
-    "weierstrass_b_g2_re" : 0,
-    "weierstrass_b_g2_im" : 155198655607781456406391640216936120121836107652948796323930557600032281009004493664981332883744016074664192874906,
-    "g1_gen_x" : 81937999373150964239938255573465948239988671502647976594219695644855304257327692006745978603320413799295628339695,
-    "g1_gen_y" : 241266749859715473739788878240585681733927191168601896383759122102112907357779751001206799952863815012735208165030,
-    "g2_gen_x_re" : 233578398248691099356572568220835526895379068987715365179118596935057653620464273615301663571204657964920925606294,
-    "g2_gen_x_im" : 140913150380207355837477652521042157274541796891053068589147167627541651775299824604154852141315666357241556069118,
-    "g2_gen_y_re" : 63160294768292073209381361943935198908131692476676907196754037919244929611450776219210369229519898517858833747423,
-    "g2_gen_y_im" : 149157405641012693445398062341192467754805999074082136895788947234480009303640899064710353187729182149407503257491
+    "gen_x" : 81937999373150964239938255573465948239988671502647976594219695644855304257327692006745978603320413799295628339695,
+    "gen_y" : 241266749859715473739788878240585681733927191168601896383759122102112907357779751001206799952863815012735208165030
 }

curve_parameters/bls12_381.json

@@ -1,20 +1,13 @@
 {
     "curve_name" : "bls12_381",
-    "modulus_p" : 52435875175126190479447740508185965837690552500527637822603658699938581184513,
+    "modolus_p" : 52435875175126190479447740508185965837690552500527637822603658699938581184513,
     "bit_count_p" : 255,
     "limb_p" :  8,
     "ntt_size" : 32,
-    "modulus_q" : 4002409555221667393417789825735904156556882819939007885332058136124031650490837864442687629129015664037894272559787,
+    "modolus_q" : 4002409555221667393417789825735904156556882819939007885332058136124031650490837864442687629129015664037894272559787,
     "bit_count_q" : 381,
     "limb_q" : 12,
-    "root_of_unity" : 937917089079007706106976984802249742464848817460758522850752807661925904159,
     "weierstrass_b" : 4,
-    "weierstrass_b_g2_re":4,
-    "weierstrass_b_g2_im":4,
-    "g1_gen_x" : 3685416753713387016781088315183077757961620795782546409894578378688607592378376318836054947676345821548104185464507,
-    "g1_gen_y" : 1339506544944476473020471379941921221584933875938349620426543736416511423956333506472724655353366534992391756441569,
-    "g2_gen_x_re" : 352701069587466618187139116011060144890029952792775240219908644239793785735715026873347600343865175952761926303160,
-    "g2_gen_x_im" : 3059144344244213709971259814753781636986470325476647558659373206291635324768958432433509563104347017837885763365758,
-    "g2_gen_y_re" : 1985150602287291935568054521177171638300868978215655730859378665066344726373823718423869104263333984641494340347905,
-    "g2_gen_y_im" : 927553665492332455747201965776037880757740193453592970025027978793976877002675564980949289727957565575433344219582
+    "gen_x" : 3685416753713387016781088315183077757961620795782546409894578378688607592378376318836054947676345821548104185464507,
+    "gen_y" : 1339506544944476473020471379941921221584933875938349620426543736416511423956333506472724655353366534992391756441569
 }

curve_parameters/bn254.json

@@ -1,20 +1,13 @@
 {
     "curve_name" : "bn254",
-    "modulus_p" : 21888242871839275222246405745257275088548364400416034343698204186575808495617,
+    "modolus_p" : 21888242871839275222246405745257275088548364400416034343698204186575808495617,
     "bit_count_p" : 254,
     "limb_p" :  8,
-    "ntt_size" : 28,
-    "modulus_q" : 21888242871839275222246405745257275088696311157297823662689037894645226208583,
+    "ntt_size" : 16,
+    "modolus_q" : 21888242871839275222246405745257275088696311157297823662689037894645226208583,
     "bit_count_q" : 254,
     "limb_q" : 8,
-    "root_of_unity" : 19103219067921713944291392827692070036145651957329286315305642004821462161904,
     "weierstrass_b" : 3,
-    "weierstrass_b_g2_re" : 19485874751759354771024239261021720505790618469301721065564631296452457478373,
-    "weierstrass_b_g2_im" : 266929791119991161246907387137283842545076965332900288569378510910307636690,
-    "g1_gen_x" : 1,
-    "g1_gen_y" : 2,
-    "g2_gen_x_re" : 10857046999023057135944570762232829481370756359578518086990519993285655852781,
-    "g2_gen_x_im" : 11559732032986387107991004021392285783925812861821192530917403151452391805634,
-    "g2_gen_y_re" : 8495653923123431417604973247489272438418190587263600148770280649306958101930,
-    "g2_gen_y_im" : 4082367875863433681332203403145435568316851327593401208105741076214120093531
+    "gen_x" : 1,
+    "gen_y" : 2
 }

curve_parameters/new_curve_script.py

@@ -1,12 +1,30 @@
 import json
 import math
 import os
-from string import Template
+from sympy.ntheory import isprime, primitive_root
+import subprocess
+import random 
 import sys
 
+data = None
+with open(sys.argv[1]) as json_file:
+    data = json.load(json_file)
 
-def to_hex(val: int, length):
-    x = hex(val)[2:]
+curve_name = data["curve_name"]
+modolus_p = data["modolus_p"]
+bit_count_p = data["bit_count_p"]
+limb_p =  data["limb_p"]
+ntt_size = data["ntt_size"]
+modolus_q = data["modolus_q"]
+bit_count_q = data["bit_count_q"] 
+limb_q = data["limb_q"]
+weierstrass_b = data["weierstrass_b"]
+gen_x = data["gen_x"]
+gen_y = data["gen_y"]
+
+
+def to_hex(val, length):
+    x = str(hex(val))[2:]
     if len(x) % 8 != 0:
         x = "0" * (8-len(x) % 8) + x
     if len(x) != length:
@@ -15,260 +33,133 @@ def to_hex(val: int, length):
     chunks = [x[i:i+n] for i in range(0, len(x), n)][::-1]
     s = ""
     for c in chunks:
-        s += f'0x{c}, '
-        
-    return s[:-2]
+        s += "0x" + c + ", "
+    return s
 
 
-def compute_values(modulus, modulus_bit_count, limbs):
-    limb_size = 8*limbs
-    modulus_ = to_hex(modulus,limb_size)
-    modulus_2 = to_hex(modulus*2,limb_size)
-    modulus_4 = to_hex(modulus*4,limb_size)
-    modulus_wide = to_hex(modulus,limb_size*2)
-    modulus_squared = to_hex(modulus*modulus,limb_size)
-    modulus_squared_2 = to_hex(modulus*modulus*2,limb_size)
-    modulus_squared_4 = to_hex(modulus*modulus*4,limb_size)
-    m_raw = int(math.floor(int(pow(2,2*modulus_bit_count) // modulus)))
-    m = to_hex(m_raw,limb_size)
-    one = to_hex(1,limb_size)
-    zero = to_hex(0,limb_size)
+def get_root_of_unity(order: int) -> int:
+    assert (modolus_p - 1) % order == 0
+    return pow(5, (modolus_p - 1) // order, modolus_p)
 
-    return (
-        modulus_,
-        modulus_2,
-        modulus_4,
-        modulus_wide,
-        modulus_squared,
-        modulus_squared_2,
-        modulus_squared_4,
-        m,
-        one,
-        zero
-    )
+def create_field_parameters_struct(modulus, modulus_bits_count,limbs,ntt,size,name):
+    s = " struct "+name+"{\n"
+    s += "   static constexpr unsigned limbs_count = " + str(limbs)+";\n"
+    s += "   static constexpr storage<limbs_count> modulus = {"+to_hex(modulus,8*limbs)[:-2]+"};\n"
+    s += "   static constexpr storage<limbs_count> modulus_2 = {"+to_hex(modulus*2,8*limbs)[:-2]+"};\n"   
+    s += "   static constexpr storage<limbs_count> modulus_4 = {"+to_hex(modulus*4,8*limbs)[:-2]+"};\n"
+    s += "   static constexpr storage<2*limbs_count> modulus_wide = {"+to_hex(modulus,8*limbs*2)[:-2]+"};\n"
+    s += "   static constexpr storage<2*limbs_count> modulus_sqared = {"+to_hex(modulus*modulus,8*limbs)[:-2]+"};\n"  
+    s += "   static constexpr storage<2*limbs_count> modulus_sqared_2 = {"+to_hex(modulus*modulus*2,8*limbs)[:-2]+"};\n"   
+    s += "   static constexpr storage<2*limbs_count> modulus_sqared_4 = {"+to_hex(modulus*modulus*2*2,8*limbs)[:-2]+"};\n"   
+    s += "   static constexpr unsigned modulus_bits_count = "+str(modulus_bits_count)+";\n"
+    m = int(math.floor(int(pow(2,2*modulus_bits_count) // modulus)))
+    s += "   static constexpr storage<limbs_count> m = {"+ to_hex(m,8*limbs)[:-2] +"};\n"
+    s += "   static constexpr storage<limbs_count> one = {"+ to_hex(1,8*limbs)[:-2] +"};\n"
+    s += "   static constexpr storage<limbs_count> zero = {"+ to_hex(0,8*limbs)[:-2] +"};\n"
 
-
-def get_fq_params(modulus, modulus_bit_count, limbs, g1_gen_x, g1_gen_y, g2_gen_x_re, g2_gen_x_im, g2_gen_y_re, g2_gen_y_im):
-    (
-        modulus,
-        modulus_2,
-        modulus_4,
-        modulus_wide,
-        modulus_squared,
-        modulus_squared_2,
-        modulus_squared_4,
-        m,
-        one,
-        zero
-    ) = compute_values(modulus, modulus_bit_count, limbs)
-
-    limb_size = 8*limbs
-    return {
-        'fq_modulus': modulus,
-        'fq_modulus_2': modulus_2,
-        'fq_modulus_4': modulus_4,
-        'fq_modulus_wide': modulus_wide,
-        'fq_modulus_squared': modulus_squared,
-        'fq_modulus_squared_2': modulus_squared_2,
-        'fq_modulus_squared_4': modulus_squared_4,
-        'fq_m': m,
-        'fq_one': one,
-        'fq_zero': zero,
-        'fq_gen_x': to_hex(g1_gen_x, limb_size),
-        'fq_gen_y': to_hex(g1_gen_y, limb_size),
-        'fq_gen_x_re': to_hex(g2_gen_x_re, limb_size),
-        'fq_gen_x_im': to_hex(g2_gen_x_im, limb_size),
-        'fq_gen_y_re': to_hex(g2_gen_y_re, limb_size),
-        'fq_gen_y_im': to_hex(g2_gen_y_im, limb_size)
-    }
-
-
-def get_fp_params(modulus, modulus_bit_count, limbs, root_of_unity, size=0):
-    (
-        modulus_,
-        modulus_2,
-        modulus_4,
-        modulus_wide,
-        modulus_squared,
-        modulus_squared_2,
-        modulus_squared_4,
-        m,
-        one,
-        zero
-    ) = compute_values(modulus, modulus_bit_count, limbs)
-    limb_size = 8*limbs
-    if size > 0:
-        omega = ''
-        omega_inv = ''
-        inv = ''
-        omegas = []
-        omegas_inv = []
+    if ntt:
         for k in range(size):
-            if k == 0:
-                om = root_of_unity
-            else:
-                om = pow(om, 2, modulus)
-            omegas.append(om)
-            omegas_inv.append(pow(om, -1, modulus))
-        omegas.reverse()
-        omegas_inv.reverse()
+            omega = get_root_of_unity(int(pow(2,k+1)))
+            s += "   static constexpr storage<limbs_count> omega"+str(k+1)+"= {"+ to_hex(omega,8*limbs)[:-2]+"};\n"
         for k in range(size):
-            omega += "\n              {"+ to_hex(omegas[k],limb_size)+"}," if k>0 else "      {"+ to_hex(omegas[k],limb_size)+"},"
-            omega_inv += "\n              {"+ to_hex(omegas_inv[k],limb_size)+"}," if k>0 else "      {"+ to_hex(omegas_inv[k],limb_size)+"},"
-            inv += "\n              {"+ to_hex(pow(int(pow(2,k+1)), -1, modulus),limb_size)+"}," if k>0 else "      {"+ to_hex(pow(int(pow(2,k+1)), -1, modulus),limb_size)+"},"
-  
-  
-    return {
-        'fp_modulus': modulus_,
-        'fp_modulus_2': modulus_2,
-        'fp_modulus_4': modulus_4,
-        'fp_modulus_wide': modulus_wide,
-        'fp_modulus_squared': modulus_squared,
-        'fp_modulus_squared_2': modulus_squared_2,
-        'fp_modulus_squared_4': modulus_squared_4,
-        'fp_m': m,
-        'fp_one': one,
-        'fp_zero': zero,
-        'omega': omega[:-1],
-        'omega_inv': omega_inv[:-1],
-        'inv': inv[:-1],
-    }
+            omega = get_root_of_unity(int(pow(2,k+1)))
+            s += "   static constexpr storage<limbs_count> omega_inv"+str(k+1)+"= {"+ to_hex(pow(omega, -1, modulus),8*limbs)[:-2]+"};\n"
+        for k in range(size):
+            s += "   static constexpr storage<limbs_count> inv"+str(k+1)+"= {"+ to_hex(pow(int(pow(2,k+1)), -1, modulus),8*limbs)[:-2]+"};\n"  
+    s+=" };\n"   
+    return s
 
+def create_gen():
+    s = " struct group_generator {\n"
+    s += "  static constexpr storage<fq_config::limbs_count> generator_x = {"+to_hex(gen_x,8*limb_q)[:-2]+ "};\n"
+    s += "  static constexpr storage<fq_config::limbs_count> generator_y = {"+to_hex(gen_y,8*limb_q)[:-2]+ "};\n"
+    s+=" };\n" 
+    return s
 
-def get_weier_params(weierstrass_b, weierstrass_b_g2_re, weierstrass_b_g2_im, size):
-    
-    return {
-        'weier_b': to_hex(weierstrass_b, size),
-        'weier_b_g2_re': to_hex(weierstrass_b_g2_re, size),
-        'weier_b_g2_im': to_hex(weierstrass_b_g2_im, size),
-    }
+def get_config_file_content(modolus_p, bit_count_p, limb_p, ntt_size, modolus_q, bit_count_q, limb_q, weierstrass_b):
+    file_content = ""
+    file_content += "#pragma once\n#include \"../../utils/storage.cuh\"\n"
+    file_content += "namespace PARAMS_"+curve_name.upper()+"{\n"
+    file_content += create_field_parameters_struct(modolus_p,bit_count_p,limb_p,True,ntt_size,"fp_config")
+    file_content += create_field_parameters_struct(modolus_q,bit_count_q,limb_q,False,0,"fq_config")
+    file_content += " static constexpr unsigned weierstrass_b = " + str(weierstrass_b)+ ";\n"
+    file_content += create_gen()
+    file_content+="}\n"
+    return file_content
 
 
-def get_params(config):
-    global ntt_size
-    curve_name = config["curve_name"]
-    modulus_p = config["modulus_p"]
-    bit_count_p = config["bit_count_p"]
-    limb_p =  config["limb_p"]
-    ntt_size = config["ntt_size"]
-    modulus_q = config["modulus_q"]
-    bit_count_q = config["bit_count_q"] 
-    limb_q = config["limb_q"]
-    root_of_unity = config["root_of_unity"]
-    if root_of_unity == modulus_p:
-        sys.exit("Invalid root_of_unity value; please update in curve parameters")
-
-    weierstrass_b = config["weierstrass_b"]
-    weierstrass_b_g2_re = config["weierstrass_b_g2_re"]
-    weierstrass_b_g2_im = config["weierstrass_b_g2_im"]
-    g1_gen_x = config["g1_gen_x"]
-    g1_gen_y = config["g1_gen_y"]
-    g2_generator_x_re = config["g2_gen_x_re"]
-    g2_generator_x_im = config["g2_gen_x_im"]
-    g2_generator_y_re = config["g2_gen_y_re"]
-    g2_generator_y_im = config["g2_gen_y_im"]
-
-    params = {
-        'curve_name_U': curve_name.upper(),
-        'fp_num_limbs': limb_p,
-        'fq_num_limbs': limb_q,
-        'fp_modulus_bit_count': bit_count_p,
-        'fq_modulus_bit_count': bit_count_q,
-        'num_omegas': ntt_size
-    }
-    
-    fp_params = get_fp_params(modulus_p, bit_count_p, limb_p, root_of_unity, ntt_size)
-    fq_params = get_fq_params(modulus_q, bit_count_q, limb_q, g1_gen_x, g1_gen_y, g2_generator_x_re, g2_generator_x_im, g2_generator_y_re, g2_generator_y_im)
-    weier_params = get_weier_params(weierstrass_b, weierstrass_b_g2_re, weierstrass_b_g2_im, 8*limb_q)
-
-    return {
-        **params,
-        **fp_params,
-        **fq_params,
-        **weier_params
-    }
-
-
-config = None
-with open(sys.argv[1]) as json_file:
-    config = json.load(json_file)
-
-curve_name_lower = config["curve_name"].lower()
-curve_name_upper = config["curve_name"].upper()
-limb_q = config["limb_q"]
-limb_p = config["limb_p"]
-
 # Create Cuda interface
 
-newpath = f'./icicle/curves/{curve_name_lower}'
+newpath = "./icicle-cuda/curves/"+curve_name 
 if not os.path.exists(newpath):
     os.makedirs(newpath)
 
-with open("./icicle/curves/curve_template/params.cuh", "r") as params_file:
-    params_file_template = Template(params_file.read())
-    params = get_params(config)
-    params_content = params_file_template.safe_substitute(params)
-    with open(f'./icicle/curves/{curve_name_lower}/params.cuh', 'w') as f:
-        f.write(params_content)
+fc = get_config_file_content(modolus_p, bit_count_p, limb_p, ntt_size, modolus_q, bit_count_q, limb_q, weierstrass_b)
+text_file = open("./icicle-cuda/curves/"+curve_name+"/params.cuh", "w")
+n = text_file.write(fc)
+text_file.close()
 
-with open("./icicle/curves/curve_template/lde.cu", "r") as lde_file:
-    template_content = Template(lde_file.read())
-    lde_content = template_content.safe_substitute(
-        CURVE_NAME_U=curve_name_upper, 
-        CURVE_NAME_L=curve_name_lower
-    )
-    with open(f'./icicle/curves/{curve_name_lower}/lde.cu', 'w') as f:
-        f.write(lde_content)
+with open("./icicle-cuda/curves/curve_template/lde.cu", "r") as lde_file:
+    content = lde_file.read()
+    content = content.replace("CURVE_NAME_U",curve_name.upper())
+    content = content.replace("CURVE_NAME_L",curve_name.lower())
+    text_file = open("./icicle-cuda/curves/"+curve_name+"/lde.cu", "w")
+    n = text_file.write(content)
+    text_file.close()
     
-with open("./icicle/curves/curve_template/msm.cu", "r") as msm_file:
-    template_content = Template(msm_file.read())
-    msm_content = template_content.safe_substitute(
-        CURVE_NAME_U=curve_name_upper, 
-        CURVE_NAME_L=curve_name_lower
-    )
-    with open(f'./icicle/curves/{curve_name_lower}/msm.cu', 'w') as f:
-        f.write(msm_content)
+with open("./icicle-cuda/curves/curve_template/msm.cu", "r") as msm_file:
+    content = msm_file.read()
+    content = content.replace("CURVE_NAME_U",curve_name.upper())
+    content = content.replace("CURVE_NAME_L",curve_name.lower())
+    text_file = open("./icicle-cuda/curves/"+curve_name+"/msm.cu", "w")
+    n = text_file.write(content)
+    text_file.close()
 
-with open("./icicle/curves/curve_template/ve_mod_mult.cu", "r") as ve_mod_mult_file:
-    template_content = Template(ve_mod_mult_file.read())
-    ve_mod_mult_content = template_content.safe_substitute(
-        CURVE_NAME_U=curve_name_upper, 
-        CURVE_NAME_L=curve_name_lower
-    )
-    with open(f'./icicle/curves/{curve_name_lower}/ve_mod_mult.cu', 'w') as f:
-        f.write(ve_mod_mult_content)
+with open("./icicle-cuda/curves/curve_template/ve_mod_mult.cu", "r") as ve_mod_mult_file:
+    content = ve_mod_mult_file.read()
+    content = content.replace("CURVE_NAME_U",curve_name.upper())
+    content = content.replace("CURVE_NAME_L",curve_name.lower())
+    text_file = open("./icicle-cuda/curves/"+curve_name+"/ve_mod_mult.cu", "w")
+    n = text_file.write(content)
+    text_file.close()
     
 
-with open(f'./icicle/curves/curve_template/curve_config.cuh', 'r') as cc:
-    template_content = Template(cc.read())
-    cc_content = template_content.safe_substitute(
-        CURVE_NAME_U=curve_name_upper,
-    )
-    with open(f'./icicle/curves/{curve_name_lower}/curve_config.cuh', 'w') as f:
-        f.write(cc_content)
+namespace = '#include "params.cuh"\n'+'''namespace CURVE_NAME_U {
+    typedef Field<PARAMS_CURVE_NAME_U::fp_config> scalar_field_t;\
+    typedef scalar_field_t scalar_t;\
+    typedef Field<PARAMS_CURVE_NAME_U::fq_config> point_field_t;
+    typedef Projective<point_field_t, scalar_field_t, PARAMS_CURVE_NAME_U::group_generator, PARAMS_CURVE_NAME_U::weierstrass_b> projective_t;
+    typedef Affine<point_field_t> affine_t;
+}'''
+
+with open('./icicle-cuda/curves/'+curve_name+'/curve_config.cuh', 'w') as f:
+    f.write(namespace.replace("CURVE_NAME_U",curve_name.upper()))
     
+    
+eq = '''
+#include <cuda.h>\n
+#include "curve_config.cuh"\n
+#include "../../primitives/projective.cuh"\n
+extern "C" bool eq_CURVE_NAME_L(CURVE_NAME_U::projective_t *point1, CURVE_NAME_U::projective_t *point2)
+{
+    return (*point1 == *point2);
+}'''
 
-with open(f'./icicle/curves/curve_template/projective.cu', 'r') as proj:
-    template_content = Template(proj.read())
-    proj_content = template_content.safe_substitute(
-        CURVE_NAME_U=curve_name_upper, 
-        CURVE_NAME_L=curve_name_lower
-    )
-    with open(f'./icicle/curves/{curve_name_lower}/projective.cu', 'w') as f:
-        f.write(proj_content)
+with open('./icicle-cuda/curves/'+curve_name+'/projective.cu', 'w') as f:
+    f.write(eq.replace("CURVE_NAME_U",curve_name.upper()).replace("CURVE_NAME_L",curve_name.lower()))
 
+supported_operations = '''
+#include "projective.cu"
+#include "lde.cu"
+#include "msm.cu"
+#include "ve_mod_mult.cu"
+'''
 
-with open(f'./icicle/curves/curve_template/supported_operations.cu', 'r') as supp_ops:
-    template_content = Template(supp_ops.read())
-    supp_ops_content = template_content.safe_substitute()
-    with open(f'./icicle/curves/{curve_name_lower}/supported_operations.cu', 'w') as f:
-        f.write(supp_ops_content)
-
-
-with open('./icicle/curves/index.cu', 'r+') as f:
-    index_text = f.read()
-    if index_text.find(curve_name_lower) == -1:
-        f.write(f'\n#include "{curve_name_lower}/supported_operations.cu"')
+with open('./icicle-cuda/curves/'+curve_name+'/supported_operations.cu', 'w') as f:
+    f.write(supported_operations.replace("CURVE_NAME_U",curve_name.upper()).replace("CURVE_NAME_L",curve_name.lower()))
+    
+with open('./icicle-cuda/curves/index.cu', 'a') as f:
+    f.write('\n#include "'+curve_name.lower()+'/supported_operations.cu"')
     
 
 
@@ -277,40 +168,36 @@ with open('./icicle/curves/index.cu', 'r+') as f:
 if limb_p == limb_q: 
     with open("./src/curve_templates/curve_same_limbs.rs", "r") as curve_file:
         content = curve_file.read()
-        content = content.replace("CURVE_NAME_U",curve_name_upper)
-        content = content.replace("CURVE_NAME_L",curve_name_lower)
+        content = content.replace("CURVE_NAME_U",curve_name.upper())
+        content = content.replace("CURVE_NAME_L",curve_name.lower())
         content = content.replace("_limbs_p",str(limb_p * 8 * 4))
         content = content.replace("limbs_p",str(limb_p))
-        text_file = open("./src/curves/"+curve_name_lower+".rs", "w")
+        text_file = open("./src/curves/"+curve_name+".rs", "w")
         n = text_file.write(content)
         text_file.close()
 else:
     with open("./src/curve_templates/curve_different_limbs.rs", "r") as curve_file:
         content = curve_file.read()
-        content = content.replace("CURVE_NAME_U",curve_name_upper)
-        content = content.replace("CURVE_NAME_L",curve_name_lower)
+        content = content.replace("CURVE_NAME_U",curve_name.upper())
+        content = content.replace("CURVE_NAME_L",curve_name.lower())
         content = content.replace("_limbs_p",str(limb_p * 8 * 4))
         content = content.replace("limbs_p",str(limb_p))
         content = content.replace("_limbs_q",str(limb_q * 8 * 4))
         content = content.replace("limbs_q",str(limb_q))
-        text_file = open("./src/curves/"+curve_name_lower+".rs", "w")
+        text_file = open("./src/curves/"+curve_name+".rs", "w")
         n = text_file.write(content)
         text_file.close()
 
 with open("./src/curve_templates/test.rs", "r") as test_file:
     content = test_file.read()
-    content = content.replace("CURVE_NAME_U",curve_name_upper)
-    content = content.replace("CURVE_NAME_L",curve_name_lower)
-    text_file = open("./src/test_"+curve_name_lower+".rs", "w")
+    content = content.replace("CURVE_NAME_U",curve_name.upper())
+    content = content.replace("CURVE_NAME_L",curve_name.lower())
+    text_file = open("./src/test_"+curve_name+".rs", "w")
     n = text_file.write(content)
     text_file.close()
     
-with open('./src/curves/mod.rs', 'r+') as f:
-    mod_text = f.read()
-    if mod_text.find(curve_name_lower) == -1:
-        f.write('\npub mod ' + curve_name_lower + ';')
+with open('./src/curves/mod.rs', 'a') as f:
+    f.write('\n pub mod ' + curve_name + ';')
 
-with open('./src/lib.rs', 'r+') as f:
-    lib_text = f.read()
-    if lib_text.find(curve_name_lower) == -1:
-        f.write('\npub mod ' + curve_name_lower + ';')
+with open('./src/lib.rs', 'a') as f:
+    f.write('\npub mod ' + curve_name + ';')

icicle-core/Cargo.toml

@@ -0,0 +1,49 @@
+[package]
+name = "icicle-core"
+version = "0.1.0"
+edition = "2021"
+authors = [ "Ingonyama" ]
+description = "An implementation of the Ingonyama CUDA Library"
+homepage = "https://www.ingonyama.com"
+repository = "https://github.com/ingonyama-zk/icicle"
+
+[[bench]]
+name = "ntt"
+path = "benches/ntt.rs"
+harness = false
+
+[[bench]]
+name = "msm"
+path = "benches/msm.rs"
+harness = false
+
+[dependencies] 
+hex = "*"
+ark-std = "0.3.0"
+ark-ff = "0.3.0"
+ark-poly = "0.3.0"
+ark-ec = { version = "0.3.0", features = [ "parallel" ] }
+ark-bls12-381 = "0.3.0"
+ark-bls12-377 = "0.3.0"
+ark-bn254 = "0.3.0"
+
+serde = { version = "1.0", features = ["derive"] }
+serde_derive = "1.0"
+serde_cbor = "0.11.2"
+
+rustacuda = "0.1"
+rustacuda_core = "0.1"
+rustacuda_derive = "0.1"
+
+rand = "*" #TODO: move rand and ark dependencies to dev once random scalar/point generation is done "natively"
+
+[build-dependencies]
+cc = { version = "1.0", features = ["parallel"] }
+
+[dev-dependencies]
+"criterion" = "0.4.0"
+
+[features]
+default = ["bls12-381"]
+bls12-381 = ["ark-bls12-381/curve"]
+g2 = []

icicle-core/src/basic_structs/field.rs

@@ -0,0 +1,4 @@
+pub trait Field<const NUM_LIMBS: usize> {
+    const MODOLUS: [u32;NUM_LIMBS];
+    const LIMBS: usize = NUM_LIMBS;
+}

icicle-core/src/basic_structs/mod.rs

@@ -0,0 +1,3 @@
+pub mod field; 
+pub mod scalar; 
+pub mod point; 

icicle-core/src/basic_structs/point.rs

@@ -0,0 +1,108 @@
+use std::ffi::c_uint;
+
+use ark_bn254::{Fq as Fq_BN254, Fr as Fr_BN254, G1Affine as G1Affine_BN254, G1Projective as G1Projective_BN254};
+
+use ark_ec::AffineCurve;
+use ark_ff::{BigInteger256, PrimeField};
+use std::mem::transmute;
+use ark_ff::Field;
+use crate::utils::{u32_vec_to_u64_vec, u64_vec_to_u32_vec};
+
+use rustacuda_core::DeviceCopy;
+use rustacuda_derive::DeviceCopy;
+
+use super::scalar::{get_fixed_limbs, self};
+
+
+#[derive(Debug, Clone, Copy, DeviceCopy)]
+#[repr(C)]
+pub struct PointT<BF: scalar::ScalarTrait> {
+    pub x: BF,
+    pub y: BF,
+    pub z: BF,
+}
+
+impl<BF: DeviceCopy + scalar::ScalarTrait> Default for PointT<BF> {
+    fn default() -> Self {
+        PointT::zero()
+    }
+}
+
+impl<BF: DeviceCopy + scalar::ScalarTrait> PointT<BF> {
+    pub fn zero() -> Self {
+        PointT {
+            x: BF::zero(),
+            y: BF::one(),
+            z: BF::zero(),
+        }
+    }
+
+    pub fn infinity() -> Self {
+        Self::zero()
+    }
+}
+
+#[derive(Debug, PartialEq, Clone, Copy, DeviceCopy)]
+#[repr(C)]
+pub struct PointAffineNoInfinityT<BF> {
+    pub x: BF,
+    pub y: BF,
+}
+
+impl<BF: scalar::ScalarTrait> Default for PointAffineNoInfinityT<BF> {
+    fn default() -> Self {
+        PointAffineNoInfinityT {
+            x: BF::zero(),
+            y: BF::zero(),
+        }
+    }
+}
+
+impl<BF: Copy + scalar::ScalarTrait> PointAffineNoInfinityT<BF> {
+    ///From u32 limbs x,y
+    pub fn from_limbs(x: &[u32], y: &[u32]) -> Self {
+        PointAffineNoInfinityT {
+            x: BF::from_limbs(x),
+            y: BF::from_limbs(y)
+        }
+    }
+
+    pub fn limbs(&self) -> Vec<u32> {
+        [self.x.limbs(), self.y.limbs()].concat()
+    }
+
+    pub fn to_projective(&self) -> PointT<BF> {
+        PointT {
+            x: self.x,
+            y: self.y,
+            z: BF::one(),
+        }
+    }
+}
+
+impl<BF: Copy + scalar::ScalarTrait> PointT<BF>  {
+    pub fn from_limbs(x: &[u32], y: &[u32], z: &[u32]) -> Self {
+        PointT {
+            x: BF::from_limbs(x),
+            y: BF::from_limbs(y),
+            z: BF::from_limbs(z)
+        }
+    }
+
+    pub fn from_xy_limbs(value: &[u32]) -> PointT<BF> {
+        let l = value.len();
+        assert_eq!(l, 3 * BF::base_limbs(), "length must be 3 * {}", BF::base_limbs());
+        PointT {
+            x: BF::from_limbs(value[..BF::base_limbs()].try_into().unwrap()),
+            y: BF::from_limbs(value[BF::base_limbs()..BF::base_limbs() * 2].try_into().unwrap()),
+            z: BF::from_limbs(value[BF::base_limbs() * 2..].try_into().unwrap())
+        }
+    }
+
+    pub fn to_xy_strip_z(&self) -> PointAffineNoInfinityT<BF> {
+        PointAffineNoInfinityT {
+            x: self.x,
+            y: self.y,
+        }
+    }
+}

icicle-core/src/basic_structs/scalar.rs

@@ -0,0 +1,102 @@
+use std::ffi::{c_int, c_uint};
+use rand::{rngs::StdRng, RngCore, SeedableRng};
+use rustacuda_core::DeviceCopy;
+use rustacuda_derive::DeviceCopy;
+use std::mem::transmute;
+use rustacuda::prelude::*;
+use rustacuda_core::DevicePointer;
+use rustacuda::memory::{DeviceBox, CopyDestination};
+
+use crate::utils::{u32_vec_to_u64_vec, u64_vec_to_u32_vec};
+
+use std::marker::PhantomData;
+use std::convert::TryInto;
+
+use super::field::{Field, self};
+
+pub fn get_fixed_limbs<const NUM_LIMBS: usize>(val: &[u32]) -> [u32; NUM_LIMBS] {
+    match val.len() {
+        n if n < NUM_LIMBS => {
+            let mut padded: [u32; NUM_LIMBS] = [0; NUM_LIMBS];
+            padded[..val.len()].copy_from_slice(&val);
+            padded
+        }
+        n if n == NUM_LIMBS => val.try_into().unwrap(),
+        _ => panic!("slice has too many elements"),
+    }
+}
+
+pub trait ScalarTrait{
+    fn base_limbs() -> usize;
+    fn zero() -> Self;
+    fn from_limbs(value: &[u32]) -> Self;
+    fn one() -> Self;
+    fn to_bytes_le(&self) -> Vec<u8>;
+    fn limbs(&self) -> &[u32];
+}
+
+#[derive(Debug, PartialEq, Clone, Copy)]
+#[repr(C)]
+pub struct ScalarT<M, const NUM_LIMBS: usize> {
+    pub(crate) phantom: PhantomData<M>,
+    pub(crate) value : [u32; NUM_LIMBS]
+}
+
+impl<M, const NUM_LIMBS: usize> ScalarTrait for ScalarT<M, NUM_LIMBS>
+where
+    M: Field<NUM_LIMBS>,
+{
+
+    fn base_limbs() -> usize {
+        return NUM_LIMBS; 
+    }
+
+    fn zero() -> Self {
+        ScalarT {
+            value: [0u32; NUM_LIMBS],
+            phantom: PhantomData,
+        }
+    }
+
+    fn from_limbs(value: &[u32]) -> Self {
+        Self {
+            value: get_fixed_limbs(value),
+            phantom: PhantomData,
+        }
+    }
+
+    fn one() -> Self {
+        let mut s = [0u32; NUM_LIMBS];
+        s[0] = 1;
+        ScalarT { value: s, phantom: PhantomData }
+    }
+
+    fn to_bytes_le(&self) -> Vec<u8> {
+        self.value
+            .iter()
+            .map(|s| s.to_le_bytes().to_vec())
+            .flatten()
+            .collect::<Vec<_>>()
+    }
+
+    fn limbs(&self) -> &[u32] {
+        &self.value
+    }
+}
+
+impl<M, const NUM_LIMBS: usize> ScalarT<M, NUM_LIMBS> where M: field::Field<NUM_LIMBS>{
+    pub fn from_limbs_le(value: &[u32]) -> ScalarT<M,NUM_LIMBS> {
+        Self::from_limbs(value)
+     }
+ 
+    pub fn from_limbs_be(value: &[u32]) -> ScalarT<M,NUM_LIMBS> {
+         let mut value = value.to_vec();
+         value.reverse();
+         Self::from_limbs_le(&value)
+     }
+ 
+     // Additional Functions
+     pub fn add(&self, other:ScalarT<M, NUM_LIMBS>) -> ScalarT<M,NUM_LIMBS>{  // overload + 
+         return ScalarT{value: [self.value[0] + other.value[0];NUM_LIMBS], phantom: PhantomData }; 
+     }
+}

icicle-core/src/lib.rs

@@ -0,0 +1,2 @@
+pub mod utils;
+pub mod basic_structs;

icicle-core/src/utils.rs

@@ -0,0 +1,42 @@
+use rand::RngCore;
+use rand::rngs::StdRng;
+use rand::SeedableRng;
+
+pub fn from_limbs<T>(limbs: Vec<u32>, chunk_size: usize, f: fn(&[u32]) -> T) -> Vec<T> {
+    let points = limbs
+        .chunks(chunk_size)
+        .map(|lmbs| f(lmbs))
+        .collect::<Vec<T>>();
+    points
+}
+
+pub fn u32_vec_to_u64_vec(arr_u32: &[u32]) -> Vec<u64> {
+    let len = (arr_u32.len() / 2) as usize;
+    let mut arr_u64 = vec![0u64; len];
+
+    for i in 0..len {
+        arr_u64[i] = u64::from(arr_u32[i * 2]) | (u64::from(arr_u32[i * 2 + 1]) << 32);
+    }
+
+    arr_u64
+}
+
+pub fn u64_vec_to_u32_vec(arr_u64: &[u64]) -> Vec<u32> {
+    let len = arr_u64.len() * 2;
+    let mut arr_u32 = vec![0u32; len];
+
+    for i in 0..arr_u64.len() {
+        arr_u32[i * 2] = arr_u64[i] as u32;
+        arr_u32[i * 2 + 1] = (arr_u64[i] >> 32) as u32;
+    }
+
+    arr_u32
+}
+
+pub fn get_rng(seed: Option<u64>) -> Box<dyn RngCore> { //TOOD: this func is universal
+    let rng: Box<dyn RngCore> = match seed {
+        Some(seed) => Box::new(StdRng::seed_from_u64(seed)),
+        None => Box::new(rand::thread_rng()),
+    };
+    rng
+}

icicle-cuda/appUtils/msm/msm.cu

@@ -0,0 +1,558 @@
+#ifndef MSM
+#define MSM
+#pragma once
+#include <stdexcept>
+#include <cuda.h>
+#include "../../primitives/affine.cuh"
+#include <iostream>
+#include <vector>
+#include <cub/device/device_radix_sort.cuh>
+#include <cub/device/device_run_length_encode.cuh>
+#include <cub/device/device_scan.cuh>
+#include "../../utils/cuda_utils.cuh"
+#include "../../primitives/projective.cuh"
+#include "../../primitives/field.cuh"
+#include "msm.cuh"
+
+
+#define BIG_TRIANGLE
+// #define SSM_SUM  //WIP
+
+//this kernel performs single scalar multiplication
+//each thread multilies a single scalar and point
+template <typename P, typename S>
+__global__ void ssm_kernel(S *scalars, P *points, P *results, unsigned N) {
+
+  unsigned tid = (blockIdx.x * blockDim.x) + threadIdx.x;
+  if (tid < N) results[tid] = scalars[tid]*points[tid];
+
+}
+
+//this kernel sums all the elements in a given vector using multiple threads
+template <typename P>
+__global__ void sum_reduction_kernel(P *v, P* v_r) {
+
+	unsigned tid = blockIdx.x * blockDim.x + threadIdx.x;
+
+	// Start at 1/2 block stride and divide by two each iteration
+	for (unsigned s = blockDim.x / 2; s > 0; s >>= 1) {
+		// Each thread does work unless it is further than the stride
+		if (threadIdx.x < s) {
+			v[tid] = v[tid] + v[tid + s];
+		}
+    __syncthreads();
+	}
+
+	// Let the thread 0 for this block write the final result
+	if (threadIdx.x == 0) {
+		v_r[blockIdx.x] = v[tid];
+	}
+}
+
+//this kernel initializes the buckets with zero points
+//each thread initializes a different bucket
+template <typename P>
+__global__ void initialize_buckets_kernel(P *buckets, unsigned N) {
+  
+  unsigned tid = (blockIdx.x * blockDim.x) + threadIdx.x;
+  if (tid < N) buckets[tid] = P::zero(); //zero point
+
+}
+
+//this kernel splits the scalars into digits of size c
+//each thread splits a single scalar into nof_bms digits
+template <typename S>
+__global__ void split_scalars_kernel(unsigned *buckets_indices, unsigned *point_indices, S *scalars, unsigned total_size, unsigned msm_log_size, unsigned nof_bms, unsigned bm_bitsize, unsigned c){
+  
+  unsigned tid = (blockIdx.x * blockDim.x) + threadIdx.x;
+  unsigned bucket_index;
+  unsigned current_index;
+  unsigned msm_index = tid >> msm_log_size;
+  if (tid < total_size){
+    S scalar = scalars[tid];
+
+    for (unsigned bm = 0; bm < nof_bms; bm++)
+    {
+      bucket_index = scalar.get_scalar_digit(bm, c);
+      current_index = bm * total_size + tid;
+      buckets_indices[current_index] = (msm_index<<(c+bm_bitsize)) | (bm<<c) | bucket_index;  //the bucket module number and the msm number are appended at the msbs
+      point_indices[current_index] = tid; //the point index is saved for later
+    }
+  }
+}
+
+//this kernel adds up the points in each bucket
+template <typename P, typename A>
+// __global__ void accumulate_buckets_kernel(P *__restrict__ buckets, unsigned *__restrict__ bucket_offsets,
+              //  unsigned *__restrict__ bucket_sizes, unsigned *__restrict__ single_bucket_indices, unsigned *__restrict__ point_indices, A *__restrict__ points, unsigned nof_buckets, unsigned batch_size, unsigned msm_idx_shift){
+__global__ void accumulate_buckets_kernel(P *buckets, unsigned *bucket_offsets, unsigned *bucket_sizes, unsigned *single_bucket_indices, unsigned *point_indices, A *points, unsigned nof_buckets, unsigned *nof_buckets_to_compute, unsigned msm_idx_shift){
+  
+  unsigned tid = (blockIdx.x * blockDim.x) + threadIdx.x;
+  if (tid >= *nof_buckets_to_compute){ 
+    return;
+  }
+  unsigned msm_index = single_bucket_indices[tid]>>msm_idx_shift;
+  unsigned bucket_index = msm_index * nof_buckets + (single_bucket_indices[tid]&((1<<msm_idx_shift)-1));
+  unsigned bucket_offset = bucket_offsets[tid];
+  for (unsigned i = 0; i < bucket_sizes[tid]; i++)  //add the relevant points starting from the relevant offset up to the bucket size
+  {
+    buckets[bucket_index] = buckets[bucket_index] + points[point_indices[bucket_offset+i]];
+  }
+}
+
+//this kernel sums the entire bucket module
+//each thread deals with a single bucket module
+template <typename P>
+__global__ void big_triangle_sum_kernel(P* buckets, P* final_sums, unsigned nof_bms, unsigned c){
+
+  unsigned tid = (blockIdx.x * blockDim.x) + threadIdx.x;
+  if (tid >= nof_bms) return;
+  P line_sum = buckets[(tid+1)*(1<<c)-1];
+  final_sums[tid] = line_sum;
+  for (unsigned i = (1<<c)-2; i >0; i--)
+  {
+    line_sum = line_sum + buckets[tid*(1<<c) + i];  //using the running sum method
+    final_sums[tid] = final_sums[tid] + line_sum;
+  }
+}
+
+//this kernel uses single scalar multiplication to multiply each bucket by its index
+//each thread deals with a single bucket
+template <typename P, typename S>
+__global__ void ssm_buckets_kernel(P* buckets, unsigned* single_bucket_indices, unsigned nof_buckets, unsigned c){
+  
+  unsigned tid = (blockIdx.x * blockDim.x) + threadIdx.x;
+  if (tid>nof_buckets) return;
+  unsigned bucket_index = single_bucket_indices[tid];
+  S scalar_bucket_multiplier;
+  scalar_bucket_multiplier = {bucket_index&((1<<c)-1), 0, 0, 0, 0, 0, 0, 0}; //the index without the bucket module index
+  buckets[bucket_index] = scalar_bucket_multiplier*buckets[bucket_index];
+
+}
+
+//this kernel computes the final result using the double and add algorithm
+//it is done by a single thread
+template <typename P, typename S>
+__global__ void final_accumulation_kernel(P* final_sums, P* final_results, unsigned nof_msms, unsigned nof_bms, unsigned c){
+  
+  unsigned tid = (blockIdx.x * blockDim.x) + threadIdx.x;
+  if (tid>nof_msms) return;
+  P final_result = P::zero();
+  for (unsigned i = nof_bms; i >1; i--)
+  {
+    final_result = final_result + final_sums[i-1 + tid*nof_bms];  //add
+    for (unsigned j=0; j<c; j++)  //double
+    {
+      final_result = final_result + final_result;
+    }
+  }
+  final_results[tid] = final_result + final_sums[tid*nof_bms];
+
+}
+
+//this function computes msm using the bucket method
+template <typename S, typename P, typename A>
+void bucket_method_msm(unsigned bitsize, unsigned c, S *scalars, A *points, unsigned size, P* final_result, bool on_device, cudaStream_t stream) {
+  
+  S *d_scalars;
+  A *d_points;
+  if (!on_device) {
+    //copy scalars and point to gpu
+    cudaMallocAsync(&d_scalars, sizeof(S) * size, stream);
+    cudaMallocAsync(&d_points, sizeof(A) * size, stream);
+    cudaMemcpyAsync(d_scalars, scalars, sizeof(S) * size, cudaMemcpyHostToDevice, stream);
+    cudaMemcpyAsync(d_points, points, sizeof(A) * size, cudaMemcpyHostToDevice, stream);
+  }
+  else {
+    d_scalars = scalars;
+    d_points = points;
+  }
+
+  P *buckets;
+  //compute number of bucket modules and number of buckets in each module
+  unsigned nof_bms = bitsize/c;
+  unsigned msm_log_size = ceil(log2(size));
+  unsigned bm_bitsize = ceil(log2(nof_bms));
+
+  if (bitsize%c){
+    nof_bms++;
+  }
+  unsigned nof_buckets = nof_bms<<c;
+  cudaMallocAsync(&buckets, sizeof(P) * nof_buckets, stream);
+
+  // launch the bucket initialization kernel with maximum threads
+  unsigned NUM_THREADS = 1 << 10;
+  unsigned NUM_BLOCKS = (nof_buckets + NUM_THREADS - 1) / NUM_THREADS;
+  initialize_buckets_kernel<<<NUM_BLOCKS, NUM_THREADS, 0, stream>>>(buckets, nof_buckets);
+
+  unsigned *bucket_indices;
+  unsigned *point_indices;
+  cudaMallocAsync(&bucket_indices, sizeof(unsigned) * size * (nof_bms+1), stream);
+  cudaMallocAsync(&point_indices, sizeof(unsigned) * size * (nof_bms+1), stream);
+
+  //split scalars into digits
+  NUM_THREADS = 1 << 10;
+  NUM_BLOCKS = (size * (nof_bms+1) + NUM_THREADS - 1) / NUM_THREADS;
+  split_scalars_kernel<<<NUM_BLOCKS, NUM_THREADS, 0, stream>>>(bucket_indices + size, point_indices + size, d_scalars, size, msm_log_size, 
+                                                    nof_bms, bm_bitsize, c); //+size - leaving the first bm free for the out of place sort later
+  
+  //sort indices - the indices are sorted from smallest to largest in order to group together the points that belong to each bucket
+  unsigned *sort_indices_temp_storage{};
+  size_t sort_indices_temp_storage_bytes;
+  // The second to last parameter is the default value supplied explicitly to allow passing the stream
+  // See https://nvlabs.github.io/cub/structcub_1_1_device_radix_sort.html#a65e82152de448c6373ed9563aaf8af7e for more info
+  cub::DeviceRadixSort::SortPairs(sort_indices_temp_storage, sort_indices_temp_storage_bytes, bucket_indices + size, bucket_indices,
+                                 point_indices + size, point_indices, size, 0, sizeof(unsigned) * 8, stream);
+  cudaMallocAsync(&sort_indices_temp_storage, sort_indices_temp_storage_bytes, stream);
+  for (unsigned i = 0; i < nof_bms; i++) {
+    unsigned offset_out = i * size;
+    unsigned offset_in = offset_out + size;
+    // The second to last parameter is the default value supplied explicitly to allow passing the stream
+    // See https://nvlabs.github.io/cub/structcub_1_1_device_radix_sort.html#a65e82152de448c6373ed9563aaf8af7e for more info
+    cub::DeviceRadixSort::SortPairs(sort_indices_temp_storage, sort_indices_temp_storage_bytes, bucket_indices + offset_in, bucket_indices + offset_out,
+                                 point_indices + offset_in, point_indices + offset_out, size, 0, sizeof(unsigned) * 8, stream);
+  }
+  cudaFreeAsync(sort_indices_temp_storage, stream);
+
+  //find bucket_sizes
+  unsigned *single_bucket_indices;
+  unsigned *bucket_sizes;
+  unsigned *nof_buckets_to_compute;
+  cudaMallocAsync(&single_bucket_indices, sizeof(unsigned)*nof_buckets, stream);
+  cudaMallocAsync(&bucket_sizes, sizeof(unsigned)*nof_buckets, stream);
+  cudaMallocAsync(&nof_buckets_to_compute, sizeof(unsigned), stream);
+  unsigned *encode_temp_storage{};
+  size_t encode_temp_storage_bytes = 0;
+  cub::DeviceRunLengthEncode::Encode(encode_temp_storage, encode_temp_storage_bytes, bucket_indices, single_bucket_indices, bucket_sizes,
+                                        nof_buckets_to_compute, nof_bms*size, stream);
+  cudaMallocAsync(&encode_temp_storage, encode_temp_storage_bytes, stream);
+  cub::DeviceRunLengthEncode::Encode(encode_temp_storage, encode_temp_storage_bytes, bucket_indices, single_bucket_indices, bucket_sizes,
+                                        nof_buckets_to_compute, nof_bms*size, stream);
+  cudaFreeAsync(encode_temp_storage, stream);
+
+  //get offsets - where does each new bucket begin
+  unsigned* bucket_offsets;
+  cudaMallocAsync(&bucket_offsets, sizeof(unsigned)*nof_buckets, stream);
+  unsigned* offsets_temp_storage{};
+  size_t offsets_temp_storage_bytes = 0;
+  cub::DeviceScan::ExclusiveSum(offsets_temp_storage, offsets_temp_storage_bytes, bucket_sizes, bucket_offsets, nof_buckets, stream);
+  cudaMallocAsync(&offsets_temp_storage, offsets_temp_storage_bytes, stream);
+  cub::DeviceScan::ExclusiveSum(offsets_temp_storage, offsets_temp_storage_bytes, bucket_sizes, bucket_offsets, nof_buckets, stream);
+  cudaFreeAsync(offsets_temp_storage, stream);
+
+  //launch the accumulation kernel with maximum threads
+  NUM_THREADS = 1 << 8;
+  NUM_BLOCKS = (nof_buckets + NUM_THREADS - 1) / NUM_THREADS;
+  accumulate_buckets_kernel<<<NUM_BLOCKS, NUM_THREADS, 0, stream>>>(buckets, bucket_offsets, bucket_sizes, single_bucket_indices, point_indices, 
+                                                         d_points, nof_buckets, nof_buckets_to_compute, c+bm_bitsize);
+
+  #ifdef SSM_SUM
+    //sum each bucket
+    NUM_THREADS = 1 << 10;
+    NUM_BLOCKS = (nof_buckets + NUM_THREADS - 1) / NUM_THREADS;
+    ssm_buckets_kernel<fake_point, fake_scalar><<<NUM_BLOCKS, NUM_THREADS, 0, stream>>>(buckets, single_bucket_indices, nof_buckets, c);
+   
+    //sum each bucket module
+    P* final_results;
+    cudaMallocAsync(&final_results, sizeof(P) * nof_bms, stream);
+    NUM_THREADS = 1<<c;
+    NUM_BLOCKS = nof_bms;
+    sum_reduction_kernel<<<NUM_BLOCKS,NUM_THREADS, 0, stream>>>(buckets, final_results);
+  #endif
+
+  #ifdef BIG_TRIANGLE
+    P* final_results;
+    cudaMallocAsync(&final_results, sizeof(P) * nof_bms, stream);
+    //launch the bucket module sum kernel - a thread for each bucket module
+    NUM_THREADS = nof_bms;
+    NUM_BLOCKS = 1;
+    big_triangle_sum_kernel<<<NUM_BLOCKS, NUM_THREADS, 0, stream>>>(buckets, final_results, nof_bms, c);
+  #endif
+
+  P* d_final_result;
+  if (!on_device)
+    cudaMallocAsync(&d_final_result, sizeof(P), stream);
+
+  //launch the double and add kernel, a single thread
+  final_accumulation_kernel<P, S><<<1,1,0,stream>>>(final_results, on_device ? final_result : d_final_result, 1, nof_bms, c);
+  
+  //copy final result to host
+  cudaStreamSynchronize(stream);
+  if (!on_device)
+    cudaMemcpyAsync(final_result, d_final_result, sizeof(P), cudaMemcpyDeviceToHost, stream);
+
+  //free memory
+  if (!on_device) {
+    cudaFreeAsync(d_points, stream);
+    cudaFreeAsync(d_scalars, stream);
+    cudaFreeAsync(d_final_result, stream);
+  }
+  cudaFreeAsync(buckets, stream);
+  cudaFreeAsync(bucket_indices, stream);
+  cudaFreeAsync(point_indices, stream);
+  cudaFreeAsync(single_bucket_indices, stream);
+  cudaFreeAsync(bucket_sizes, stream);
+  cudaFreeAsync(nof_buckets_to_compute, stream);
+  cudaFreeAsync(bucket_offsets, stream);
+  cudaFreeAsync(final_results, stream);
+
+  cudaStreamSynchronize(stream);
+}
+
+//this function computes msm using the bucket method
+template <typename S, typename P, typename A>
+void batched_bucket_method_msm(unsigned bitsize, unsigned c, S *scalars, A *points, unsigned batch_size, unsigned msm_size, P* final_results, bool on_device, cudaStream_t stream){
+
+  unsigned total_size = batch_size * msm_size;
+  S *d_scalars;
+  A *d_points;
+  if (!on_device) {
+    //copy scalars and point to gpu
+    cudaMallocAsync(&d_scalars, sizeof(S) * total_size, stream);
+    cudaMallocAsync(&d_points, sizeof(A) * total_size, stream);
+    cudaMemcpyAsync(d_scalars, scalars, sizeof(S) * total_size, cudaMemcpyHostToDevice, stream);
+    cudaMemcpyAsync(d_points, points, sizeof(A) * total_size, cudaMemcpyHostToDevice, stream);
+  }
+  else {
+    d_scalars = scalars;
+    d_points = points;
+  }
+
+  P *buckets;
+  //compute number of bucket modules and number of buckets in each module
+  unsigned nof_bms = bitsize/c;
+  if (bitsize%c){
+    nof_bms++;
+  }
+  unsigned msm_log_size = ceil(log2(msm_size));
+  unsigned bm_bitsize = ceil(log2(nof_bms));
+  unsigned nof_buckets = (nof_bms<<c);
+  unsigned total_nof_buckets = nof_buckets*batch_size;
+  cudaMallocAsync(&buckets, sizeof(P) * total_nof_buckets, stream); 
+
+  //lanch the bucket initialization kernel with maximum threads
+  unsigned NUM_THREADS = 1 << 10;
+  unsigned NUM_BLOCKS = (total_nof_buckets + NUM_THREADS - 1) / NUM_THREADS;
+  initialize_buckets_kernel<<<NUM_BLOCKS, NUM_THREADS, 0, stream>>>(buckets, total_nof_buckets); 
+
+  unsigned *bucket_indices;
+  unsigned *point_indices;
+  cudaMallocAsync(&bucket_indices, sizeof(unsigned) * (total_size * nof_bms + msm_size), stream);
+  cudaMallocAsync(&point_indices, sizeof(unsigned) * (total_size * nof_bms + msm_size), stream);
+
+  //split scalars into digits
+  NUM_THREADS = 1 << 8;
+  NUM_BLOCKS = (total_size * nof_bms + msm_size + NUM_THREADS - 1) / NUM_THREADS;
+  split_scalars_kernel<<<NUM_BLOCKS, NUM_THREADS, 0, stream>>>(bucket_indices + msm_size, point_indices + msm_size, d_scalars, total_size, 
+                                                    msm_log_size, nof_bms, bm_bitsize, c); //+size - leaving the first bm free for the out of place sort later
+
+  //sort indices - the indices are sorted from smallest to largest in order to group together the points that belong to each bucket
+  unsigned *sorted_bucket_indices;
+  unsigned *sorted_point_indices;
+  cudaMallocAsync(&sorted_bucket_indices, sizeof(unsigned) * (total_size * nof_bms), stream);
+  cudaMallocAsync(&sorted_point_indices, sizeof(unsigned) * (total_size * nof_bms), stream);
+
+  unsigned *sort_indices_temp_storage{};
+  size_t sort_indices_temp_storage_bytes;
+  // The second to last parameter is the default value supplied explicitly to allow passing the stream
+  // See https://nvlabs.github.io/cub/structcub_1_1_device_radix_sort.html#a65e82152de448c6373ed9563aaf8af7e for more info
+  cub::DeviceRadixSort::SortPairs(sort_indices_temp_storage, sort_indices_temp_storage_bytes, bucket_indices + msm_size, sorted_bucket_indices,
+                                 point_indices + msm_size, sorted_point_indices, total_size * nof_bms, 0, sizeof(unsigned)*8, stream);
+  cudaMallocAsync(&sort_indices_temp_storage, sort_indices_temp_storage_bytes, stream);
+  // The second to last parameter is the default value supplied explicitly to allow passing the stream
+  // See https://nvlabs.github.io/cub/structcub_1_1_device_radix_sort.html#a65e82152de448c6373ed9563aaf8af7e for more info
+  cub::DeviceRadixSort::SortPairs(sort_indices_temp_storage, sort_indices_temp_storage_bytes, bucket_indices + msm_size, sorted_bucket_indices,
+                                 point_indices + msm_size, sorted_point_indices, total_size * nof_bms, 0, sizeof(unsigned)*8, stream);
+  cudaFreeAsync(sort_indices_temp_storage, stream);
+
+  //find bucket_sizes
+  unsigned *single_bucket_indices;
+  unsigned *bucket_sizes;
+  unsigned *total_nof_buckets_to_compute;
+  cudaMallocAsync(&single_bucket_indices, sizeof(unsigned)*total_nof_buckets, stream);
+  cudaMallocAsync(&bucket_sizes, sizeof(unsigned)*total_nof_buckets, stream);
+  cudaMallocAsync(&total_nof_buckets_to_compute, sizeof(unsigned), stream);
+  unsigned *encode_temp_storage{};
+  size_t encode_temp_storage_bytes = 0;
+  cub::DeviceRunLengthEncode::Encode(encode_temp_storage, encode_temp_storage_bytes, sorted_bucket_indices, single_bucket_indices, bucket_sizes,
+                                        total_nof_buckets_to_compute, nof_bms*total_size, stream);  
+  cudaMallocAsync(&encode_temp_storage, encode_temp_storage_bytes, stream);
+  cub::DeviceRunLengthEncode::Encode(encode_temp_storage, encode_temp_storage_bytes, sorted_bucket_indices, single_bucket_indices, bucket_sizes,
+                                        total_nof_buckets_to_compute, nof_bms*total_size, stream);
+  cudaFreeAsync(encode_temp_storage, stream);
+
+  //get offsets - where does each new bucket begin
+  unsigned* bucket_offsets;
+  cudaMallocAsync(&bucket_offsets, sizeof(unsigned)*total_nof_buckets, stream);
+  unsigned* offsets_temp_storage{};
+  size_t offsets_temp_storage_bytes = 0;
+  cub::DeviceScan::ExclusiveSum(offsets_temp_storage, offsets_temp_storage_bytes, bucket_sizes, bucket_offsets, total_nof_buckets, stream);
+  cudaMallocAsync(&offsets_temp_storage, offsets_temp_storage_bytes, stream);
+  cub::DeviceScan::ExclusiveSum(offsets_temp_storage, offsets_temp_storage_bytes, bucket_sizes, bucket_offsets, total_nof_buckets, stream);
+  cudaFreeAsync(offsets_temp_storage, stream);
+
+  //launch the accumulation kernel with maximum threads
+  NUM_THREADS = 1 << 8;
+  NUM_BLOCKS = (total_nof_buckets + NUM_THREADS - 1) / NUM_THREADS;
+  accumulate_buckets_kernel<<<NUM_BLOCKS, NUM_THREADS, 0, stream>>>(buckets, bucket_offsets, bucket_sizes, single_bucket_indices, sorted_point_indices,
+                                                        d_points, nof_buckets, total_nof_buckets_to_compute, c+bm_bitsize);
+
+  #ifdef SSM_SUM
+    //sum each bucket
+    NUM_THREADS = 1 << 10;
+    NUM_BLOCKS = (nof_buckets + NUM_THREADS - 1) / NUM_THREADS;
+    ssm_buckets_kernel<P, S><<<NUM_BLOCKS, NUM_THREADS, 0, stream>>>(buckets, single_bucket_indices, nof_buckets, c);
+   
+    //sum each bucket module
+    P* final_results;
+    cudaMallocAsync(&final_results, sizeof(P) * nof_bms, stream);
+    NUM_THREADS = 1<<c;
+    NUM_BLOCKS = nof_bms;
+    sum_reduction_kernel<<<NUM_BLOCKS,NUM_THREADS, 0, stream>>>(buckets, final_results);
+  #endif
+
+  #ifdef BIG_TRIANGLE
+    P* bm_sums;
+    cudaMallocAsync(&bm_sums, sizeof(P) * nof_bms * batch_size, stream);
+    //launch the bucket module sum kernel - a thread for each bucket module
+    NUM_THREADS = 1<<8;
+    NUM_BLOCKS = (nof_bms*batch_size + NUM_THREADS - 1) / NUM_THREADS;
+    big_triangle_sum_kernel<<<NUM_BLOCKS, NUM_THREADS, 0, stream>>>(buckets, bm_sums, nof_bms*batch_size, c);
+  #endif
+
+  P* d_final_results;
+  if (!on_device)
+    cudaMallocAsync(&d_final_results, sizeof(P)*batch_size, stream);
+
+  //launch the double and add kernel, a single thread for each msm
+  NUM_THREADS = 1<<8;
+  NUM_BLOCKS = (batch_size + NUM_THREADS - 1) / NUM_THREADS;
+  final_accumulation_kernel<P, S><<<NUM_BLOCKS,NUM_THREADS, 0, stream>>>(bm_sums, on_device ? final_results : d_final_results, batch_size, nof_bms, c);
+  
+  //copy final result to host
+  if (!on_device)
+    cudaMemcpyAsync(final_results, d_final_results, sizeof(P)*batch_size, cudaMemcpyDeviceToHost, stream);
+
+  //free memory
+  if (!on_device) {
+    cudaFreeAsync(d_points, stream);
+    cudaFreeAsync(d_scalars, stream);
+    cudaFreeAsync(d_final_results, stream);
+  }
+  cudaFreeAsync(buckets, stream);
+  cudaFreeAsync(bucket_indices, stream);
+  cudaFreeAsync(point_indices, stream);
+  cudaFreeAsync(sorted_bucket_indices, stream);
+  cudaFreeAsync(sorted_point_indices, stream);
+  cudaFreeAsync(single_bucket_indices, stream);
+  cudaFreeAsync(bucket_sizes, stream);
+  cudaFreeAsync(total_nof_buckets_to_compute, stream);
+  cudaFreeAsync(bucket_offsets, stream);
+  cudaFreeAsync(bm_sums, stream);
+
+  cudaStreamSynchronize(stream);
+}
+
+
+//this kernel converts affine points to projective points
+//each thread deals with a single point
+template <typename P, typename A>
+__global__ void to_proj_kernel(A* affine_points, P* proj_points, unsigned N){
+  
+  unsigned tid = (blockIdx.x * blockDim.x) + threadIdx.x;
+  if (tid < N) proj_points[tid] = P::from_affine(affine_points[tid]);
+}
+
+//the function computes msm using ssm
+template <typename S, typename P, typename A>
+void short_msm(S *h_scalars, A *h_points, unsigned size, P* h_final_result, cudaStream_t stream){ //works up to 2^8
+  S *scalars;
+  A *a_points;
+  P *p_points;
+  P *results;
+
+  cudaMallocAsync(&scalars, sizeof(S) * size, stream);
+  cudaMallocAsync(&a_points, sizeof(A) * size, stream);
+  cudaMallocAsync(&p_points, sizeof(P) * size, stream);
+  cudaMallocAsync(&results, sizeof(P) * size, stream);
+
+  //copy inputs to device
+  cudaMemcpyAsync(scalars, h_scalars, sizeof(S) * size, cudaMemcpyHostToDevice, stream);
+  cudaMemcpyAsync(a_points, h_points, sizeof(A) * size, cudaMemcpyHostToDevice, stream);
+
+  //convert to projective representation and multiply each point by its scalar using single scalar multiplication
+  unsigned NUM_THREADS = size;
+  to_proj_kernel<<<1,NUM_THREADS, 0, stream>>>(a_points, p_points, size);
+  ssm_kernel<<<1,NUM_THREADS, 0, stream>>>(scalars, p_points, results, size);
+
+  P *final_result;
+  cudaMallocAsync(&final_result, sizeof(P), stream);
+
+  //assuming msm size is a power of 2
+  //sum all the ssm results
+  NUM_THREADS = size;
+  sum_reduction_kernel<<<1,NUM_THREADS, 0, stream>>>(results, final_result);
+
+  //copy result to host
+  cudaStreamSynchronize(stream);
+  cudaMemcpyAsync(h_final_result, final_result, sizeof(P), cudaMemcpyDeviceToHost, stream);
+
+  //free memory
+  cudaFreeAsync(scalars, stream);
+  cudaFreeAsync(a_points, stream);
+  cudaFreeAsync(p_points, stream);
+  cudaFreeAsync(results, stream);
+  cudaFreeAsync(final_result, stream);
+
+}
+
+//the function computes msm on the host using the naive method
+template <typename A, typename S, typename P>
+void reference_msm(S* scalars, A* a_points, unsigned size){
+  
+  P *points = new P[size];
+  // P points[size];
+  for (unsigned i = 0; i < size ; i++)
+  {
+    points[i] = P::from_affine(a_points[i]);
+  }
+
+  P res = P::zero();
+  
+  for (unsigned i = 0; i < size; i++)
+  {
+    res = res + scalars[i]*points[i];
+  }
+
+  std::cout<<"reference results"<<std::endl;
+  std::cout<<P::to_affine(res)<<std::endl;
+  
+}
+
+unsigned get_optimal_c(const unsigned size) {
+  if (size < 17)
+    return 1;
+  // return 15;
+  return ceil(log2(size))-4;
+}
+
+//this function is used to compute msms of size larger than 256
+template <typename S, typename P, typename A>
+void large_msm(S* scalars, A* points, unsigned size, P* result, bool on_device, cudaStream_t stream){
+  unsigned c = get_optimal_c(size);
+  // unsigned c = 6;
+  // unsigned bitsize = 32;
+  unsigned bitsize = 255;
+  bucket_method_msm(bitsize, c, scalars, points, size, result, on_device, stream);
+}
+
+// this function is used to compute a batches of msms of size larger than 256
+template <typename S, typename P, typename A>
+void batched_large_msm(S* scalars, A* points, unsigned batch_size, unsigned msm_size, P* result, bool on_device, cudaStream_t stream){
+  unsigned c = get_optimal_c(msm_size);
+  // unsigned c = 6;
+  // unsigned bitsize = 32;
+  unsigned bitsize = 255;
+  batched_bucket_method_msm(bitsize, c, scalars, points, batch_size, msm_size, result, on_device, stream);
+}
+#endif

icicle-cuda/appUtils/msm/msm.cuh

@@ -3,7 +3,7 @@
 #pragma once
 
 template <typename S, typename P, typename A>
-void bucket_method_msm(unsigned bitsize, unsigned c, S *scalars, A *points, unsigned size, P* final_result, bool on_device, bool big_triangle, cudaStream_t stream);
+void bucket_method_msm(unsigned bitsize, unsigned c, S *scalars, A *points, unsigned size, P* final_result, bool on_device, cudaStream_t stream);
 
 template <typename S, typename P, typename A>
 void batched_bucket_method_msm(unsigned bitsize, unsigned c, S *scalars, A *points, unsigned batch_size, unsigned msm_size, P* final_results, bool on_device, cudaStream_t stream);
@@ -12,7 +12,7 @@ template <typename S, typename P, typename A>
 void batched_large_msm(S* scalars, A* points, unsigned batch_size, unsigned msm_size, P* result, bool on_device, cudaStream_t stream);
 
 template <typename S, typename P, typename A>
-void large_msm(S* scalars, A* points, unsigned size, P* result, bool on_device, bool big_triangle, cudaStream_t stream);
+void large_msm(S* scalars, A* points, unsigned size, P* result, bool on_device, cudaStream_t stream);
 
 template <typename S, typename P, typename A>
 void short_msm(S *h_scalars, A *h_points, unsigned size, P* h_final_result, cudaStream_t stream);

icicle-cuda/appUtils/msm/tests/msm_test.cu

@@ -5,19 +5,15 @@
 #include "../../utils/cuda_utils.cuh"
 #include "../../primitives/projective.cuh"
 #include "../../primitives/field.cuh"
-// #include "../../curves/bls12_377/curve_config.cuh"
-#include "../../curves/bn254/curve_config.cuh"
+#include "../../curves/bls12_381/curve_config.cuh"
 
-// using namespace BLS12_377;
-using namespace BN254;
+using namespace BLS12_381;
 
 class Dummy_Scalar {
   public:
     static constexpr unsigned NBITS = 32;
 
     unsigned x;
-    // unsigned p = 10;
-    unsigned p = 1<<30;
 
     friend HOST_INLINE std::ostream& operator<<(std::ostream& os, const Dummy_Scalar& scalar) {
       os << scalar.x;
@@ -29,7 +25,7 @@ class Dummy_Scalar {
     }
 
     friend HOST_DEVICE_INLINE Dummy_Scalar operator+(Dummy_Scalar p1, const Dummy_Scalar& p2) {   
-      return {(p1.x+p2.x)%p1.p};
+      return {p1.x+p2.x};
     }
 
     friend HOST_DEVICE_INLINE bool operator==(const Dummy_Scalar& p1, const Dummy_Scalar& p2) {
@@ -40,11 +36,10 @@ class Dummy_Scalar {
       return (p1.x == p2);
     }
 
-    static HOST_DEVICE_INLINE Dummy_Scalar neg(const Dummy_Scalar &scalar) { 
-      return {scalar.p-scalar.x};
-    }
+    // static HOST_DEVICE_INLINE Dummy_Scalar neg(const Dummy_Scalar &scalar) { 
+    //   return {Dummy_Scalar::neg(point.x)};
+    // }
     static HOST_INLINE Dummy_Scalar rand_host() {
-      // return {(unsigned)rand()%10};
       return {(unsigned)rand()};
     }
 };
@@ -58,10 +53,6 @@ class Dummy_Projective {
       return {0};
     }
 
-    static HOST_DEVICE_INLINE Dummy_Projective one() {
-      return {1};
-    }
-
     static HOST_DEVICE_INLINE Dummy_Projective to_affine(const Dummy_Projective &point) {
       return {point.x};
     }
@@ -70,9 +61,9 @@ class Dummy_Projective {
       return {point.x};
     }
 
-    static HOST_DEVICE_INLINE Dummy_Projective neg(const Dummy_Projective &point) { 
-      return {Dummy_Scalar::neg(point.x)};
-    }
+    // static HOST_DEVICE_INLINE Dummy_Projective neg(const Dummy_Projective &point) { 
+    //   return {Dummy_Scalar::neg(point.x)};
+    // }
 
     friend HOST_DEVICE_INLINE Dummy_Projective operator+(Dummy_Projective p1, const Dummy_Projective& p2) {   
       return {p1.x+p2.x};
@@ -128,68 +119,62 @@ typedef affine_t test_affine;
 
 int main()
 {
-  unsigned batch_size = 1;
-  unsigned msm_size = 1<<24;
+  unsigned batch_size = 4;
+  unsigned msm_size = 1<<15;
   unsigned N = batch_size*msm_size;
 
   test_scalar *scalars = new test_scalar[N];
   test_affine *points = new test_affine[N];
   
   for (unsigned i=0;i<N;i++){
-    // scalars[i] = (i%msm_size < 10)? test_scalar::rand_host() : scalars[i-10];
+    scalars[i] = (i%msm_size < 10)? test_scalar::rand_host() : scalars[i-10];
     points[i] = (i%msm_size < 10)? test_projective::to_affine(test_projective::rand_host()): points[i-10];
-    scalars[i] = test_scalar::rand_host();
+    // scalars[i] = test_scalar::rand_host();
     // points[i] = test_projective::to_affine(test_projective::rand_host());
   }
   std::cout<<"finished generating"<<std::endl;
 
   // projective_t *short_res = (projective_t*)malloc(sizeof(projective_t));
   // test_projective *large_res = (test_projective*)malloc(sizeof(test_projective));
-  test_projective large_res[batch_size*2];
-  // test_projective batched_large_res[batch_size];
+  test_projective large_res[batch_size];
+  test_projective batched_large_res[batch_size];
   // fake_point *large_res = (fake_point*)malloc(sizeof(fake_point));
   // fake_point batched_large_res[256];
 
 
   // short_msm<scalar_t, projective_t, affine_t>(scalars, points, N, short_res);
-  // for (unsigned i=0;i<batch_size;i++){
-    // large_msm<test_scalar, test_projective, test_affine>(scalars+msm_size*i, points+msm_size*i, msm_size, large_res+i, false);
+  for (unsigned i=0;i<batch_size;i++){
+    large_msm<test_scalar, test_projective, test_affine>(scalars+msm_size*i, points+msm_size*i, msm_size, large_res+i, false);
     // std::cout<<"final result large"<<std::endl;
     // std::cout<<test_projective::to_affine(*large_res)<<std::endl;
-  // }
+  }
   auto begin = std::chrono::high_resolution_clock::now();
-  // batched_large_msm<test_scalar, test_projective, test_affine>(scalars, points, batch_size, msm_size, batched_large_res, false);
-  large_msm<test_scalar, test_projective, test_affine>(scalars, points, msm_size, large_res, false, true,0);
-  // std::cout<<test_projective::to_affine(large_res[0])<<std::endl;
-  large_msm<test_scalar, test_projective, test_affine>(scalars, points, msm_size, large_res+1, false, false,0);
-  // test_reduce_triangle(scalars);
-  // test_reduce_rectangle(scalars);
-  // test_reduce_single(scalars);
+  batched_large_msm<test_scalar, test_projective, test_affine>(scalars, points, batch_size, msm_size, batched_large_res, false);
+  // large_msm<test_scalar, test_projective, test_affine>(scalars, points, msm_size, large_res, false);
   auto end = std::chrono::high_resolution_clock::now();
   auto elapsed = std::chrono::duration_cast<std::chrono::nanoseconds>(end - begin);
   printf("Time measured: %.3f seconds.\n", elapsed.count() * 1e-9);
   std::cout<<test_projective::to_affine(large_res[0])<<std::endl;
-  std::cout<<test_projective::to_affine(large_res[1])<<std::endl;
 
   // reference_msm<test_affine, test_scalar, test_projective>(scalars, points, msm_size);
 
-  // std::cout<<"final results batched large"<<std::endl;
-  // bool success = true;
-  // for (unsigned i = 0; i < batch_size; i++)
-  // {
-  //   std::cout<<test_projective::to_affine(batched_large_res[i])<<std::endl;
-  //   if (test_projective::to_affine(large_res[i])==test_projective::to_affine(batched_large_res[i])){
-  //     std::cout<<"good"<<std::endl;
-  //   }
-  //   else{
-  //     std::cout<<"miss"<<std::endl;
-  //     std::cout<<test_projective::to_affine(large_res[i])<<std::endl;
-  //     success = false;
-  //   }
-  // }
-  // if (success){
-  //   std::cout<<"success!"<<std::endl;
-  // }
+  std::cout<<"final results batched large"<<std::endl;
+  bool success = true;
+  for (unsigned i = 0; i < batch_size; i++)
+  {
+    std::cout<<test_projective::to_affine(batched_large_res[i])<<std::endl;
+    if (test_projective::to_affine(large_res[i])==test_projective::to_affine(batched_large_res[i])){
+      std::cout<<"good"<<std::endl;
+    }
+    else{
+      std::cout<<"miss"<<std::endl;
+      std::cout<<test_projective::to_affine(large_res[i])<<std::endl;
+      success = false;
+    }
+  }
+  if (success){
+    std::cout<<"success!"<<std::endl;
+  }
   
   // std::cout<<batched_large_res[0]<<std::endl;
   // std::cout<<batched_large_res[1]<<std::endl;

icicle-cuda/appUtils/poseidon/poseidon.cu

@@ -1,7 +1,7 @@
 #include "poseidon.cuh"
 
 template <typename S>
-__global__ void prepare_poseidon_states(S * states, size_t number_of_states, S domain_tag, const PoseidonConfiguration<S> config) {
+__global__ void prepare_poseidon_states(S * inp, S * states, size_t number_of_states, S domain_tag, const PoseidonConfiguration<S> config) {
     int idx = (blockIdx.x * blockDim.x) + threadIdx.x;
     int state_number = idx / config.t;
     if (state_number >= number_of_states) {
@@ -15,7 +15,7 @@ __global__ void prepare_poseidon_states(S * states, size_t number_of_states, S d
     if (element_number == 0) {
         prepared_element = domain_tag;
     } else {
-        prepared_element = states[state_number * config.t + element_number - 1];
+        prepared_element = inp[state_number * (config.t - 1) + element_number - 1];
     }
 
     // Add pre-round constant
@@ -148,20 +148,17 @@ __global__ void get_hash_results(S * states, size_t number_of_states, S * out, i
 }
 
 template <typename S>
-__host__ void Poseidon<S>::hash_blocks(const S * inp, size_t blocks, S * out, HashType hash_type, cudaStream_t stream) {
-    S * states;
+__host__ void Poseidon<S>::hash_blocks(const S * inp, size_t blocks, S * out, HashType hash_type) {
+    // Used in matrix multiplication
+
+    S * states, * inp_device;
 
     // allocate memory for {blocks} states of {t} scalars each
-    if (cudaMallocAsync(&states, blocks * this->t * sizeof(S), stream) != cudaSuccess) {
-        throw std::runtime_error("Failed memory allocation on the device");
-    }
+    cudaMalloc(&states, blocks * this->t * sizeof(S));
 
-    // This is where the input matrix of size Arity x NumberOfBlocks is
-    // padded and coppied to device in a T x NumberOfBlocks matrix
-    cudaMemcpy2DAsync(states, this->t * sizeof(S),  // Device pointer and device pitch
-                 inp, (this->t - 1) * sizeof(S),    // Host pointer and pitch
-                 (this->t - 1) * sizeof(S), blocks, // Size of the source matrix (Arity x NumberOfBlocks)
-                 cudaMemcpyHostToDevice, stream);
+    // Move input to cuda
+    cudaMalloc(&inp_device, blocks * (this->t - 1) * sizeof(S));
+    cudaMemcpy(inp_device, inp, blocks * (this->t - 1) * sizeof(S), cudaMemcpyHostToDevice);
 
     size_t rc_offset = 0;
 
@@ -194,13 +191,14 @@ __host__ void Poseidon<S>::hash_blocks(const S * inp, size_t blocks, S * out, Ha
     #endif
 
     // Domain separation and adding pre-round constants
-    prepare_poseidon_states <<< number_of_blocks, number_of_threads, 0, stream >>> (states, blocks, domain_tag, this->config);
+    prepare_poseidon_states <<< number_of_blocks, number_of_threads >>> (inp_device, states, blocks, domain_tag, this->config);
     rc_offset += this->t;
+    cudaFree(inp_device);
 
     #if !defined(__CUDA_ARCH__) && defined(DEBUG)
-    cudaStreamSynchronize(stream);
+    cudaDeviceSynchronize();
     std::cout << "Domain separation: " << rc_offset << std::endl;
-    //print_buffer_from_cuda<S>(states, blocks * this->t);
+    print_buffer_from_cuda<S>(states, blocks * this->t);
 
     auto end_time = std::chrono::high_resolution_clock::now();
     auto elapsed_time = std::chrono::duration_cast<std::chrono::milliseconds>(end_time - start_time);
@@ -209,13 +207,13 @@ __host__ void Poseidon<S>::hash_blocks(const S * inp, size_t blocks, S * out, Ha
     #endif
 
     // execute half full rounds
-    full_rounds <<< number_of_blocks, number_of_threads, sizeof(S) * hashes_per_block * this->t, stream >>> (states, blocks, rc_offset, true, this->config);
+    full_rounds <<< number_of_blocks, number_of_threads, sizeof(S) * hashes_per_block * this->t >>> (states, blocks, rc_offset, true, this->config);
     rc_offset += this->t * this->config.full_rounds_half;
 
     #if !defined(__CUDA_ARCH__) && defined(DEBUG)
-    cudaStreamSynchronize(stream);
+    cudaDeviceSynchronize();
     std::cout << "Full rounds 1. RCOFFSET: " << rc_offset << std::endl;
-    // print_buffer_from_cuda<S>(states, blocks * this->t);
+    print_buffer_from_cuda<S>(states, blocks * this->t);
 
     end_time = std::chrono::high_resolution_clock::now();
     elapsed_time = std::chrono::duration_cast<std::chrono::milliseconds>(end_time - start_time);
@@ -224,13 +222,13 @@ __host__ void Poseidon<S>::hash_blocks(const S * inp, size_t blocks, S * out, Ha
     #endif
 
     // execute partial rounds
-    partial_rounds <<< number_of_singlehash_blocks, singlehash_block_size, 0, stream >>> (states, blocks, rc_offset, this->config);
+    partial_rounds <<< number_of_singlehash_blocks, singlehash_block_size >>> (states, blocks, rc_offset, this->config);
     rc_offset += this->config.partial_rounds;
 
     #if !defined(__CUDA_ARCH__) && defined(DEBUG)
-    cudaStreamSynchronize(stream);
+    cudaDeviceSynchronize();
     std::cout << "Partial rounds. RCOFFSET: " << rc_offset << std::endl;
-    //print_buffer_from_cuda<S>(states, blocks * this->t);
+    print_buffer_from_cuda<S>(states, blocks * this->t);
 
     end_time = std::chrono::high_resolution_clock::now();
     elapsed_time = std::chrono::duration_cast<std::chrono::milliseconds>(end_time - start_time);
@@ -239,12 +237,12 @@ __host__ void Poseidon<S>::hash_blocks(const S * inp, size_t blocks, S * out, Ha
     #endif
 
     // execute half full rounds
-    full_rounds <<< number_of_blocks, number_of_threads, sizeof(S) * hashes_per_block * this->t, stream >>> (states, blocks, rc_offset, false, this->config);
+    full_rounds <<< number_of_blocks, number_of_threads, sizeof(S) * hashes_per_block * this->t >>> (states, blocks, rc_offset, false, this->config);
 
     #if !defined(__CUDA_ARCH__) && defined(DEBUG)
-    cudaStreamSynchronize(stream);
+    cudaDeviceSynchronize();
     std::cout << "Full rounds 2. RCOFFSET: " << rc_offset << std::endl;
-    //print_buffer_from_cuda<S>(states, blocks * this->t);
+    print_buffer_from_cuda<S>(states, blocks * this->t);
     end_time = std::chrono::high_resolution_clock::now();
     elapsed_time = std::chrono::duration_cast<std::chrono::milliseconds>(end_time - start_time);
     std::cout << "Elapsed time: " << elapsed_time.count() << " ms" << std::endl;
@@ -254,18 +252,18 @@ __host__ void Poseidon<S>::hash_blocks(const S * inp, size_t blocks, S * out, Ha
     // get output
     S * out_device;
     cudaMalloc(&out_device, blocks * sizeof(S));
-    get_hash_results <<< number_of_singlehash_blocks, singlehash_block_size, 0, stream >>> (states, blocks, out_device, this->config.t);
+    get_hash_results <<< number_of_singlehash_blocks, singlehash_block_size >>> (states, blocks, out_device, this->config.t);
 
     #if !defined(__CUDA_ARCH__) && defined(DEBUG)
-    cudaStreamSynchronize(stream);
+    cudaDeviceSynchronize();
     std::cout << "Get hash results" << std::endl;
     end_time = std::chrono::high_resolution_clock::now();
     elapsed_time = std::chrono::duration_cast<std::chrono::milliseconds>(end_time - start_time);
     std::cout << "Elapsed time: " << elapsed_time.count() << " ms" << std::endl;
     #endif
-    cudaMemcpyAsync(out, out_device, blocks * sizeof(S), cudaMemcpyDeviceToHost, stream);
-    cudaFreeAsync(out_device, stream);
-    cudaFreeAsync(states, stream);
+    cudaMemcpy(out, out_device, blocks * sizeof(S), cudaMemcpyDeviceToHost);
+    cudaFree(out_device);
+    cudaFree(states);
 
     #if !defined(__CUDA_ARCH__) && defined(DEBUG)
     cudaDeviceReset();

icicle-cuda/appUtils/poseidon/poseidon.cuh

@@ -0,0 +1,133 @@
+#pragma once
+#include "constants.cuh"
+
+#if !defined(__CUDA_ARCH__) && defined(DEBUG)
+#include <iostream>
+#include <iomanip>
+#include <string>
+#include <sstream>
+#include <chrono>
+
+#define ARITY 3
+
+template <typename S>
+__host__ void print_buffer_from_cuda(S * device_ptr, size_t size) {
+  S * buffer = static_cast< S * >(malloc(size * sizeof(S)));
+  cudaMemcpy(buffer, device_ptr, size * sizeof(S), cudaMemcpyDeviceToHost);
+
+  std::cout << "Start print" << std::endl;
+  for(int i = 0; i < size / ARITY; i++) {
+    std::cout << "State #" << i << std::endl;
+    for (int j = 0; j < ARITY; j++) {
+      std::cout << buffer[i * ARITY + j] << std::endl;
+    }
+    std::cout << std::endl;
+  }
+  std::cout << std::endl;
+  free(buffer);
+}
+#endif
+
+#ifdef DEBUG
+template <typename S>
+__device__ void print_scalar(S element, int data) {
+    printf("D# %d, T# %d: 0x%08x%08x%08x%08x%08x%08x%08x%08x\n",
+           data,
+           threadIdx.x,
+           element.limbs_storage.limbs[0],
+           element.limbs_storage.limbs[1],
+           element.limbs_storage.limbs[2],
+           element.limbs_storage.limbs[3],
+           element.limbs_storage.limbs[4],
+           element.limbs_storage.limbs[5],
+           element.limbs_storage.limbs[6],
+           element.limbs_storage.limbs[7]
+    );
+}
+#endif
+
+template <typename S>
+struct PoseidonConfiguration {
+    uint32_t partial_rounds, full_rounds_half, t;
+    S * round_constants, * mds_matrix, * non_sparse_matrix, *sparse_matrices;
+};
+
+template <typename S>
+class Poseidon {
+  public:
+    uint32_t t;
+    PoseidonConfiguration<S> config;
+
+    enum HashType {
+        ConstInputLen,
+        MerkleTree,
+    };
+
+    Poseidon(const uint32_t arity) {
+        t = arity + 1;
+        this->config.t = t;
+
+        // Pre-calculate domain tags
+        // Domain tags will vary for different applications of Poseidon
+        uint32_t tree_domain_tag_value = 1;
+        tree_domain_tag_value = (tree_domain_tag_value << arity) - tree_domain_tag_value;
+        tree_domain_tag = S::from(tree_domain_tag_value);
+
+        const_input_no_pad_domain_tag = S::one();
+
+        // TO-DO: implement binary shifts for scalar type
+        // const_input_no_pad_domain_tag = S::one() << 64;
+        // const_input_no_pad_domain_tag *= S::from(arity);
+
+        this->config.full_rounds_half = FULL_ROUNDS_DEFAULT;
+        this->config.partial_rounds = partial_rounds_number_from_arity(arity);
+
+        uint32_t round_constants_len = t * this->config.full_rounds_half * 2 + this->config.partial_rounds;
+        uint32_t mds_matrix_len = t * t;
+        uint32_t sparse_matrices_len = (t * 2 - 1) * this->config.partial_rounds;
+
+        // All the constants are stored in a single file
+        S * constants = load_constants<S>(arity);
+
+        S * mds_offset = constants + round_constants_len;
+        S * non_sparse_offset = mds_offset + mds_matrix_len;
+        S * sparse_matrices_offset = non_sparse_offset + mds_matrix_len;
+
+        #if !defined(__CUDA_ARCH__) && defined(DEBUG)
+        std::cout << "P: " << this->config.partial_rounds << " F: " << this->config.full_rounds_half << std::endl;
+        #endif
+
+        // Allocate the memory for constants
+        cudaMalloc(&this->config.round_constants, sizeof(S) * round_constants_len);
+        cudaMalloc(&this->config.mds_matrix, sizeof(S) * mds_matrix_len);
+        cudaMalloc(&this->config.non_sparse_matrix, sizeof(S) * mds_matrix_len);
+        cudaMalloc(&this->config.sparse_matrices, sizeof(S) * sparse_matrices_len);
+
+        // Copy the constants to device
+        cudaMemcpy(this->config.round_constants, constants,
+                sizeof(S) * round_constants_len,
+                cudaMemcpyHostToDevice);
+        cudaMemcpy(this->config.mds_matrix, mds_offset,
+                sizeof(S) * mds_matrix_len,
+                cudaMemcpyHostToDevice);
+        cudaMemcpy(this->config.non_sparse_matrix, non_sparse_offset,
+                sizeof(S) * mds_matrix_len,
+                cudaMemcpyHostToDevice);
+        cudaMemcpy(this->config.sparse_matrices, sparse_matrices_offset,
+                sizeof(S) * sparse_matrices_len,
+                cudaMemcpyHostToDevice);
+    }
+
+    ~Poseidon() {
+        cudaFree(this->config.round_constants);
+        cudaFree(this->config.mds_matrix);
+        cudaFree(this->config.non_sparse_matrix);
+        cudaFree(this->config.sparse_matrices);
+    }
+
+    // Hash multiple preimages in parallel
+    void hash_blocks(const S * inp, size_t blocks, S * out, HashType hash_type);
+
+  private:
+    S tree_domain_tag, const_input_no_pad_domain_tag;
+};

icicle-cuda/appUtils/poseidon/poseidon_test.cu

@@ -0,0 +1,48 @@
+#define DEBUG
+
+#include "../../curves/bls12_381/curve_config.cuh"
+#include "../../curves/bls12_381/poseidon.cu"
+
+#ifndef __CUDA_ARCH__
+#include <iostream>
+#include <chrono>
+#include <fstream>
+
+int main(int argc, char* argv[]) {
+  const int arity = 2;
+  const int t = arity + 1;
+
+  Poseidon<BLS12_381::scalar_t> poseidon(arity);
+
+  int number_of_blocks = 4;
+
+  BLS12_381::scalar_t input = BLS12_381::scalar_t::zero();
+  BLS12_381::scalar_t * in_ptr = static_cast< BLS12_381::scalar_t * >(malloc(number_of_blocks * arity * sizeof(BLS12_381::scalar_t)));
+  for (uint32_t i = 0; i < number_of_blocks * arity; i++) {
+    // std::cout << input << std::endl;
+    in_ptr[i] = input;
+    input = input + BLS12_381::scalar_t::one();
+  }
+  std::cout << std::endl;
+
+  BLS12_381::scalar_t * out_ptr = static_cast< BLS12_381::scalar_t * >(malloc(number_of_blocks * sizeof(BLS12_381::scalar_t)));
+
+  auto start_time = std::chrono::high_resolution_clock::now();
+
+  poseidon.hash_blocks(in_ptr, number_of_blocks, out_ptr, Poseidon<BLS12_381::scalar_t>::HashType::MerkleTree);
+
+  #ifdef DEBUG
+  for (int i = 0; i < number_of_blocks; i++) {
+    std::cout << out_ptr[i] << std::endl;
+  }
+  #endif
+
+  auto end_time = std::chrono::high_resolution_clock::now();
+  auto elapsed_time = std::chrono::duration_cast<std::chrono::milliseconds>(end_time - start_time);
+  std::cout << "Elapsed time: " << elapsed_time.count() << " ms" << std::endl;
+
+  free(in_ptr);
+  free(out_ptr);
+}
+
+#endif

icicle-cuda/curves/bls12_377/params.cuh

@@ -1,23 +1,19 @@
 #pragma once
 #include "../../utils/storage.cuh"
-
-namespace PARAMS_BLS12_377{
+namespace PARAMS{
   struct fp_config{
     static constexpr unsigned limbs_count = 8;
-    static constexpr unsigned omegas_count = 32;
-
     static constexpr storage<limbs_count> modulus = {0x00000001, 0x0a118000, 0xd0000001, 0x59aa76fe, 0x5c37b001, 0x60b44d1e, 0x9a2ca556, 0x12ab655e};
     static constexpr storage<limbs_count> modulus_2 = {0x00000002, 0x14230000, 0xa0000002, 0xb354edfd, 0xb86f6002, 0xc1689a3c, 0x34594aac, 0x2556cabd};
     static constexpr storage<limbs_count> modulus_4 = {0x00000004, 0x28460000, 0x40000004, 0x66a9dbfb, 0x70dec005, 0x82d13479, 0x68b29559, 0x4aad957a};
     static constexpr storage<2*limbs_count> modulus_wide = {0x00000001, 0x0a118000, 0xd0000001, 0x59aa76fe, 0x5c37b001, 0x60b44d1e, 0x9a2ca556, 0x12ab655e, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
-    static constexpr storage<2*limbs_count> modulus_squared = {0x00000001, 0x14230000, 0xe0000002, 0xc7dd4d2f, 0x8585d003, 0x08ee1bd4, 0xe57fc56e, 0x7e7557e3, 0x483a709d, 0x1fdebb41, 0x5678f4e6, 0x8ea77334, 0xc19c3ec5, 0xd717de29, 0xe2340781, 0x015c8d01};
-    static constexpr storage<2*limbs_count> modulus_squared_2 = {0x00000002, 0x28460000, 0xc0000004, 0x8fba9a5f, 0x0b0ba007, 0x11dc37a9, 0xcaff8adc, 0xfceaafc7, 0x9074e13a, 0x3fbd7682, 0xacf1e9cc, 0x1d4ee668, 0x83387d8b, 0xae2fbc53, 0xc4680f03, 0x02b91a03};
-    static constexpr storage<2*limbs_count> modulus_squared_4 = {0x00000004, 0x508c0000, 0x80000008, 0x1f7534bf, 0x1617400f, 0x23b86f52, 0x95ff15b8, 0xf9d55f8f, 0x20e9c275, 0x7f7aed05, 0x59e3d398, 0x3a9dccd1, 0x0670fb16, 0x5c5f78a7, 0x88d01e07, 0x05723407};
-    static constexpr unsigned modulus_bit_count = 253;
+    static constexpr storage<2*limbs_count> modulus_sqared = {0x00000001, 0x14230000, 0xe0000002, 0xc7dd4d2f, 0x8585d003, 0x08ee1bd4, 0xe57fc56e, 0x7e7557e3, 0x483a709d, 0x1fdebb41, 0x5678f4e6, 0x8ea77334, 0xc19c3ec5, 0xd717de29, 0xe2340781, 0x015c8d01};
+    static constexpr storage<2*limbs_count> modulus_sqared_2 = {0x00000002, 0x28460000, 0xc0000004, 0x8fba9a5f, 0x0b0ba007, 0x11dc37a9, 0xcaff8adc, 0xfceaafc7, 0x9074e13a, 0x3fbd7682, 0xacf1e9cc, 0x1d4ee668, 0x83387d8b, 0xae2fbc53, 0xc4680f03, 0x02b91a03};
+    static constexpr storage<2*limbs_count> modulus_sqared_4 = {0x00000004, 0x508c0000, 0x80000008, 0x1f7534bf, 0x1617400f, 0x23b86f52, 0x95ff15b8, 0xf9d55f8f, 0x20e9c275, 0x7f7aed05, 0x59e3d398, 0x3a9dccd1, 0x0670fb16, 0x5c5f78a7, 0x88d01e07, 0x05723407};
+    static constexpr unsigned modulus_bits_count = 253;
     static constexpr storage<limbs_count> m = {0x151e79ea, 0xf5204c21, 0x8d69e258, 0xfd0a180b, 0xfaa80548, 0xe4e51e49, 0xc40b2c9e, 0x36d9491e};
     static constexpr storage<limbs_count> one = {0x00000001, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
     static constexpr storage<limbs_count> zero = {0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
-
     static constexpr storage<limbs_count> omega1= {0x00000001, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
     static constexpr storage<limbs_count> omega2= {0x00000001, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
     static constexpr storage<limbs_count> omega3= {0x00000000, 0x0a118000, 0xd0000001, 0x59aa76fe, 0x5c37b001, 0x60b44d1e, 0x9a2ca556, 0x12ab655e};
@@ -50,14 +46,6 @@ namespace PARAMS_BLS12_377{
     static constexpr storage<limbs_count> omega30= {0x2be70731, 0x287abbb1, 0x7c35c5aa, 0x5cbcfd1e, 0x1671f4df, 0x7585b3fe, 0xb899c011, 0x08350ecf};
     static constexpr storage<limbs_count> omega31= {0x09f7c5e2, 0x3400c14e, 0x0a649ea1, 0xc112e60c, 0x067ce95e, 0xf7510758, 0xf9daf17c, 0x040a66a5};
     static constexpr storage<limbs_count> omega32= {0x43efecd3, 0x89d65957, 0x3bd6c318, 0x29246adc, 0xce01533c, 0xf9fb5ef6, 0x849078c3, 0x020410e4};
-
-    static constexpr storage_array<omegas_count, limbs_count> omega = {
-        omega1, omega2, omega3, omega4, omega5, omega6, omega7, omega8, 
-        omega9, omega10, omega11, omega12, omega13, omega14, omega15, omega16,
-        omega17, omega18, omega19, omega20, omega21, omega22, omega23, omega24,
-        omega25, omega26, omega27, omega28, omega29, omega30, omega31, omega32,
-    };
-
     static constexpr storage<limbs_count> omega_inv1= {0x00000001, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
     static constexpr storage<limbs_count> omega_inv2= {0x00000001, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
     static constexpr storage<limbs_count> omega_inv3= {0x00000000, 0x0a118000, 0xd0000001, 0x59aa76fe, 0x5c37b001, 0x60b44d1e, 0x9a2ca556, 0x12ab655e};
@@ -90,14 +78,6 @@ namespace PARAMS_BLS12_377{
     static constexpr storage<limbs_count> omega_inv30= {0x123a81f6, 0xc03a3272, 0x115b15e8, 0x377e6d2f, 0x2d6d7206, 0xed5575e4, 0x714004f2, 0x0b1e37e4};
     static constexpr storage<limbs_count> omega_inv31= {0xdde8ffc5, 0x62a29589, 0x618c5d62, 0xfb6716e8, 0x88d61f25, 0x787e561c, 0xd2b21c7e, 0x0e351761};
     static constexpr storage<limbs_count> omega_inv32= {0x7aca7fbe, 0xc9fea0e9, 0xb41a8854, 0x965ff314, 0x810eea7e, 0x743415d4, 0x8275bbd1, 0x0431c01b};
-    
-    static constexpr storage_array<omegas_count, limbs_count> omega_inv = {
-        omega_inv1, omega_inv2, omega_inv3, omega_inv4, omega_inv5, omega_inv6, omega_inv7, omega_inv8, 
-        omega_inv9, omega_inv10, omega_inv11, omega_inv12, omega_inv13, omega_inv14, omega_inv15, omega_inv16,
-        omega_inv17, omega_inv18, omega_inv19, omega_inv20, omega_inv21, omega_inv22, omega_inv23, omega_inv24,
-        omega_inv25, omega_inv26, omega_inv27, omega_inv28, omega_inv29, omega_inv30, omega_inv31, omega_inv32,
-    };
-
     static constexpr storage<limbs_count> inv1= {0x00000001, 0x8508c000, 0x68000000, 0xacd53b7f, 0x2e1bd800, 0x305a268f, 0x4d1652ab, 0x0955b2af};
     static constexpr storage<limbs_count> inv2= {0x00000001, 0xc78d2000, 0x1c000000, 0x033fd93f, 0xc529c401, 0xc88739d6, 0xf3a17c00, 0x0e008c06};
     static constexpr storage<limbs_count> inv3= {0x00000001, 0xe8cf5000, 0xf6000000, 0x2e75281e, 0x90b0ba01, 0x949dc37a, 0xc6e710ab, 0x1055f8b2};
@@ -130,13 +110,6 @@ namespace PARAMS_BLS12_377{
     static constexpr storage<limbs_count> inv30= {0xd7ba0001, 0xca117ffb, 0x69562405, 0xe8cbb6f9, 0xd9667b87, 0xf801b7c4, 0x4f7f0fdb, 0x12ab655e};
     static constexpr storage<limbs_count> inv31= {0xebdd0001, 0x6a117ffd, 0x1cab1203, 0xa13b16fc, 0x9acf15c4, 0x2c5b0271, 0x74d5da99, 0x12ab655e};
     static constexpr storage<limbs_count> inv32= {0xf5ee8001, 0x3a117ffe, 0x76558902, 0xfd72c6fd, 0xfb8362e2, 0xc687a7c7, 0x87813ff7, 0x12ab655e};
-
-    static constexpr storage_array<omegas_count, limbs_count> inv = {
-        inv1, inv2, inv3, inv4, inv5, inv6, inv7, inv8, 
-        inv9, inv10, inv11, inv12, inv13, inv14, inv15, inv16,
-        inv17, inv18, inv19, inv20, inv21, inv22, inv23, inv24,
-        inv25, inv26, inv27, inv28, inv29, inv30, inv31, inv32,
-    };    
   };
 
   struct fq_config{
@@ -145,10 +118,10 @@ namespace PARAMS_BLS12_377{
     static constexpr storage<limbs_count> modulus_2 = {0x00000002, 0x0a118000, 0x60000001, 0x2e16ba88, 0x74129000, 0x3de6c45f, 0x01ea271e, 0x3445b3e6, 0xd9429276, 0x8c760b80, 0x2f8a21d5, 0x035c748c};
     static constexpr storage<limbs_count> modulus_4 = {0x00000004, 0x14230000, 0xc0000002, 0x5c2d7510, 0xe8252000, 0x7bcd88be, 0x03d44e3c, 0x688b67cc, 0xb28524ec, 0x18ec1701, 0x5f1443ab, 0x06b8e918};
     static constexpr storage<2*limbs_count> modulus_wide = {0x00000001, 0x8508c000, 0x30000000, 0x170b5d44, 0xba094800, 0x1ef3622f, 0x00f5138f, 0x1a22d9f3, 0x6ca1493b, 0xc63b05c0, 0x17c510ea, 0x01ae3a46, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
-    static constexpr storage<2*limbs_count> modulus_squared = {0x00000001, 0x0a118000, 0xf0000001, 0x7338d254, 0x2e1bd800, 0x4ada268f, 0x35f1c09a, 0x6bcbfbd2, 0x58638c9d, 0x318324b9, 0x8bb70ae0, 0x460aaaaa, 0x502a4d6c, 0xc014e712, 0xb90660cd, 0x09d018af, 0x3dda4d5c, 0x1f5e7141, 0xa4aee93f, 0x4bb8b87d, 0xb361263c, 0x2256913b, 0xd0bbaffb, 0x0002d307};
-    static constexpr storage<2*limbs_count> modulus_squared_2 = {0x00000002, 0x14230000, 0xe0000002, 0xe671a4a9, 0x5c37b000, 0x95b44d1e, 0x6be38134, 0xd797f7a4, 0xb0c7193a, 0x63064972, 0x176e15c0, 0x8c155555, 0xa0549ad8, 0x8029ce24, 0x720cc19b, 0x13a0315f, 0x7bb49ab8, 0x3ebce282, 0x495dd27e, 0x977170fb, 0x66c24c78, 0x44ad2277, 0xa1775ff6, 0x0005a60f};
-    static constexpr storage<2*limbs_count> modulus_squared_4 = {0x00000004, 0x28460000, 0xc0000004, 0xcce34953, 0xb86f6001, 0x2b689a3c, 0xd7c70269, 0xaf2fef48, 0x618e3275, 0xc60c92e5, 0x2edc2b80, 0x182aaaaa, 0x40a935b1, 0x00539c49, 0xe4198337, 0x274062be, 0xf7693570, 0x7d79c504, 0x92bba4fc, 0x2ee2e1f6, 0xcd8498f1, 0x895a44ee, 0x42eebfec, 0x000b4c1f};
-    static constexpr unsigned modulus_bit_count = 377;
+    static constexpr storage<2*limbs_count> modulus_sqared = {0x00000001, 0x0a118000, 0xf0000001, 0x7338d254, 0x2e1bd800, 0x4ada268f, 0x35f1c09a, 0x6bcbfbd2, 0x58638c9d, 0x318324b9, 0x8bb70ae0, 0x460aaaaa, 0x502a4d6c, 0xc014e712, 0xb90660cd, 0x09d018af, 0x3dda4d5c, 0x1f5e7141, 0xa4aee93f, 0x4bb8b87d, 0xb361263c, 0x2256913b, 0xd0bbaffb, 0x0002d307};
+    static constexpr storage<2*limbs_count> modulus_sqared_2 = {0x00000002, 0x14230000, 0xe0000002, 0xe671a4a9, 0x5c37b000, 0x95b44d1e, 0x6be38134, 0xd797f7a4, 0xb0c7193a, 0x63064972, 0x176e15c0, 0x8c155555, 0xa0549ad8, 0x8029ce24, 0x720cc19b, 0x13a0315f, 0x7bb49ab8, 0x3ebce282, 0x495dd27e, 0x977170fb, 0x66c24c78, 0x44ad2277, 0xa1775ff6, 0x0005a60f};
+    static constexpr storage<2*limbs_count> modulus_sqared_4 = {0x00000004, 0x28460000, 0xc0000004, 0xcce34953, 0xb86f6001, 0x2b689a3c, 0xd7c70269, 0xaf2fef48, 0x618e3275, 0xc60c92e5, 0x2edc2b80, 0x182aaaaa, 0x40a935b1, 0x00539c49, 0xe4198337, 0x274062be, 0xf7693570, 0x7d79c504, 0x92bba4fc, 0x2ee2e1f6, 0xcd8498f1, 0x895a44ee, 0x42eebfec, 0x000b4c1f};
+    static constexpr unsigned modulus_bits_count = 377;
     static constexpr storage<limbs_count> m = {0x5e4daffc, 0x1f9fd58c, 0x89c42a59, 0xd0ed6877, 0xd85a6d02, 0x6af2d488, 0x6776b1a0, 0x3bbad0de, 0x582ef4f7, 0x976c3ca0, 0x0cc4060e, 0x0261508d};
     static constexpr storage<limbs_count> one = {0x00000001, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
     static constexpr storage<limbs_count> zero = {0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
@@ -157,24 +130,26 @@ namespace PARAMS_BLS12_377{
     // true if i^2 is negative
     static constexpr bool i_squared_is_negative = true;
     // G1 and G2 generators 
-    static constexpr storage<limbs_count> g1_gen_x = {0xb21be9ef, 0xeab9b16e, 0xffcd394e, 0xd5481512, 0xbd37cb5c, 0x188282c8,
+    static constexpr storage<limbs_count> generator_x = {0xb21be9ef, 0xeab9b16e, 0xffcd394e, 0xd5481512, 0xbd37cb5c, 0x188282c8,
                                                          0xaa9d41bb, 0x85951e2c, 0xbf87ff54, 0xc8fc6225, 0xfe740a67, 0x008848de};
-    static constexpr storage<limbs_count> g1_gen_y = {0x559c8ea6, 0xfd82de55, 0x34a9591a, 0xc2fe3d36, 0x4fb82305, 0x6d182ad4,
+    static constexpr storage<limbs_count> generator_y = {0x559c8ea6, 0xfd82de55, 0x34a9591a, 0xc2fe3d36, 0x4fb82305, 0x6d182ad4,
                                                          0xca3e52d9, 0xbd7fb348, 0x30afeec4, 0x1f674f5d, 0xc5102eff, 0x01914a69};
-    static constexpr storage<limbs_count> g2_gen_x_re = {0x7c005196, 0x74e3e48f, 0xbb535402, 0x71889f52, 0x57db6b9b, 0x7ea501f5, 
-                                                            0x203e5031, 0xc565f071, 0xa3841d01, 0xc89630a2, 0x71c785fe, 0x018480be};
-    static constexpr storage<limbs_count> g2_gen_x_im = {0x6ea16afe, 0xb26bfefa, 0xbff76fe6, 0x5cf89984, 0x0799c9de, 0xe7223ece, 
-                                                            0x6651cecb, 0x532777ee, 0xb1b140d5, 0x70dc5a51, 0xe7004031, 0x00ea6040};
-    static constexpr storage<limbs_count> g2_gen_y_re = {0x09fd4ddf, 0xf0940944, 0x6d8c7c2e, 0xf2cf8888, 0xf832d204, 0xe458c282, 
-                                                            0x74b49a58, 0xde03ed72, 0xcbb2efb4, 0xd960736b, 0x5d446f7b, 0x00690d66};
-    static constexpr storage<limbs_count> g2_gen_y_im = {0x85eb8f93, 0xd9a1cdd1, 0x5e52270b, 0x4279b83f, 0xcee304c2, 0x2463b01a,
-                                                            0x3d591bf1, 0x61ef11ac, 0x151a70aa, 0x9e549da3, 0xd2835518, 0x00f8169f};
+    static constexpr storage<limbs_count> generator_x_re = {0xc121bdb8, 0xd48056c8, 0xa805bbef, 0xbac0326, 0x7ae3d177, 0xb4510b64,
+                                                            0xfa403b02, 0xc6e47ad4, 0x2dc51051, 0x26080527, 0xf08f0a91, 0x24aa2b2};
+    static constexpr storage<limbs_count> generator_x_im = {0x5d042b7e, 0xe5ac7d05, 0x13945d57, 0x334cf112, 0xdc7f5049, 0xb5da61bb,
+                                                            0x9920b61a, 0x596bd0d0, 0x88274f65, 0x7dacd3a0, 0x52719f60, 0x13e02b60};
+    static constexpr storage<limbs_count> generator_y_re = {0x8b82801, 0xe1935486, 0x3baca289, 0x923ac9cc, 0x5160d12c, 0x6d429a69,
+                                                            0x8cbdd3a7, 0xadfd9baa, 0xda2e351a, 0x8cc9cdc6, 0x727d6e11, 0xce5d527};
+    static constexpr storage<limbs_count> generator_y_im = {0xf05f79be, 0xaaa9075f, 0x5cec1da1, 0x3f370d27, 0x572e99ab, 0x267492ab,
+                                                            0x85a763af, 0xcb3e287e, 0x2bc28b99, 0x32acd2b0, 0x2ea734cc, 0x606c4a0};
   };
 
   static constexpr storage<fq_config::limbs_count> weierstrass_b = {0x00000001, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000,
                                                                     0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
-  static constexpr storage<fq_config::limbs_count> weierstrass_b_g2_re = {0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000,
+
+  // TODO: correct parameters for G2 here
+  static constexpr storage<fq_config::limbs_count> weierstrass_b_g2_re = {0x00000004, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000,
+                                                                          0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
+  static constexpr storage<fq_config::limbs_count> weierstrass_b_g2_im = {0x00000004, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000,
                                                                           0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
-  static constexpr storage<fq_config::limbs_count> weierstrass_b_g2_im = {0x9999999a, 0x1c9ed999, 0x1ccccccd, 0x0dd39e5c, 0x3c6bf800, 0x129207b6,
-                                                                          0xcd5fd889, 0xdc7b4f91, 0x7460c589, 0x43bd0373, 0xdb0fd6f3, 0x010222f6};
 }

icicle-cuda/curves/bls12_381/params.cuh

@@ -2,11 +2,10 @@
 
 #include "../../utils/storage.cuh"
 
-namespace PARAMS_BLS12_381{
+namespace PARAMS{
   struct fp_config {
     // field structure size = 8 * 32 bit
     static constexpr unsigned limbs_count = 8;
-    static constexpr unsigned omegas_count = 32;
     // modulus = 52435875175126190479447740508185965837690552500527637822603658699938581184513
     static constexpr storage<limbs_count> modulus = {0x00000001, 0xffffffff, 0xfffe5bfe, 0x53bda402, 0x09a1d805, 0x3339d808, 0x299d7d48, 0x73eda753};
     // modulus*2 = 104871750350252380958895481016371931675381105001055275645207317399877162369026
@@ -16,13 +15,13 @@ namespace PARAMS_BLS12_381{
     static constexpr storage<2 * limbs_count> modulus_wide = {0x00000001, 0xffffffff, 0xfffe5bfe, 0x53bda402, 0x09a1d805, 0x3339d808, 0x299d7d48, 0x73eda753,
                                                               0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
     // modulus^2
-    static constexpr storage<2*limbs_count> modulus_squared = {0x00000001, 0xfffffffe, 0xfffcb7fe, 0xa77e9007, 0x1cdbb005, 0x698ae002, 0x5433f7b8, 0x48aa415e, 
+    static constexpr storage<2*limbs_count> modulus_sqared = {0x00000001, 0xfffffffe, 0xfffcb7fe, 0xa77e9007, 0x1cdbb005, 0x698ae002, 0x5433f7b8, 0x48aa415e, 
                                                               0x4aa9c661, 0xc2611f6f, 0x59934a1d, 0x0e9593f9, 0xef2cc20f, 0x520c13db, 0xf4bc2778, 0x347f60f3};
     // 2*modulus^2
-    static constexpr storage<2*limbs_count> modulus_squared_2 = {0x00000002, 0xfffffffc, 0xfff96ffd, 0x4efd200f, 0x39b7600b, 0xd315c004, 0xa867ef70, 0x915482bc, 
+    static constexpr storage<2*limbs_count> modulus_sqared_2 = {0x00000002, 0xfffffffc, 0xfff96ffd, 0x4efd200f, 0x39b7600b, 0xd315c004, 0xa867ef70, 0x915482bc, 
                                                                 0x95538cc2, 0x84c23ede, 0xb326943b, 0x1d2b27f2, 0xde59841e, 0xa41827b7, 0xe9784ef0, 0x68fec1e7};
-    static constexpr unsigned modulus_bit_count = 255;
-    // m = floor(2^(2*modulus_bit_count) / modulus)
+    static constexpr unsigned modulus_bits_count = 255;
+    // m = floor(2^(2*modulus_bits_count) / modulus)
     static constexpr storage<limbs_count> m = {0x830358e4, 0x509cde80, 0x2f92eb5c, 0xd9410fad, 0xc1f823b4, 0xe2d772d, 0x7fb78ddf, 0x8d54253b};
   
     static constexpr storage<limbs_count> one = {0x00000001, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
@@ -62,13 +61,6 @@ namespace PARAMS_BLS12_381{
     static constexpr storage<limbs_count> omega30=  {0x14ebb608, 0x8409a9ea, 0x66bac611, 0xfad0084e, 0x811c1dfb, 0x04287254, 0x23b30c29, 0x086d072b};
     static constexpr storage<limbs_count> omega31=  {0x67e4756a, 0xb427c9b3, 0x02ebc38d, 0xc7537fb9, 0xcd6a205f, 0x51de21be, 0x7923597d, 0x6064ab72};
     static constexpr storage<limbs_count> omega32=  {0x0b912f1f, 0x1b788f50, 0x70b3e094, 0xc4024ff2, 0xd168d6c0, 0x0fd56dc8, 0x5b416b6f, 0x0212d79e};
-
-    static constexpr storage_array<omegas_count, limbs_count> omega = {
-        omega1, omega2, omega3, omega4, omega5, omega6, omega7, omega8, 
-        omega9, omega10, omega11, omega12, omega13, omega14, omega15, omega16,
-        omega17, omega18, omega19, omega20, omega21, omega22, omega23, omega24,
-        omega25, omega26, omega27, omega28, omega29, omega30, omega31, omega32,
-    };
   
     // static constexpr storage<limbs_count> omega_inv[32]={ {0x00000000, 0xffffffff, 0xfffe5bfe, 0x53bda402, 0x09a1d805, 0x3339d808, 0x299d7d48, 0x73eda753}, {0x00000001, 0xfffeffff, 0x89fb5bfe, 0x67baa400, 0x939ed334, 0xa5e80b39, 0x299d7d47, 0x73eda753}, {0xae99502e, 0x6037fe81, 0x94b04fd8, 0x8e749036, 0xca86bf65, 0xbabc5aff, 0x5ce11044, 0x1333b22e}, {0x7dc08d74, 0x7f847ee4, 0x04eeaf5a, 0xbd433896, 0x1832fc60, 0xd66c91d6, 0x607e449b, 0x551115b4}, {0x4e7773cb, 0xee5bcecc, 0xf6dab086, 0x45593d6f, 0x4016e2bd, 0xa3a95d2d, 0xaf96816f, 0x047cb16c}, {0x982b68c5, 0xb891fa3f, 0x1d426b52, 0xa41e8501, 0x882952d6, 0x566009b5, 0x7b3c79d6, 0x199cdaee}, {0xcf28601b, 0x571ba2fc, 0xac74db12, 0x166fb582, 0x3501370b, 0x51420be4, 0x52f970ba, 0x1996fa8d}, {0x6a2f777a, 0xe9561c17, 0x2393991b, 0xc03cae03, 0x5a5bfd4f, 0x91b00023, 0x272e58ee, 0x6d64ed25}, {0xf02a116e, 0xfb350dbe, 0xb4543a3e, 0x1c510ebf, 0x37ad4eca, 0xf675522e, 0x80f82b2d, 0x1907a56e}, {0x4eb71aa6, 0xb0ad8003, 0xaa67e0be, 0x50a32c41, 0x19141f44, 0x105f0672, 0xa3dad316, 0x2bcd9508}, {0x0f6fb2ac, 0x3dc9e560, 0x9aa58ff5, 0x3cc5bb32, 0x36f376e1, 0xdeae67bc, 0x65ba213e, 0x394fda0d}, {0x60b82267, 0x09f239f7, 0x8b24f123, 0x14180e0e, 0x45625d95, 0xad5a5340, 0x6d174692, 0x58c3ba63}, {0x348b416f, 0x0acf21c2, 0xbc086439, 0x798b6bf6, 0xb1ca111d, 0x222d411f, 0x30ba1e0f, 0x044107b7}, {0x014abe84, 0xa3b861b8, 0x427ed008, 0x37c017e4, 0xae0ff4f5, 0xae51f613, 0xcb1218d3, 0x1a2d00e1}, {0x4de7eb2b, 0x48aaa3bf, 0x6772057d, 0x4a58d54d, 0x7093b551, 0xce25f16c, 0xd206337c, 0x242150ac}, {0x9ed57ae5, 0xdf3ec9ae, 0x7166577f, 0xea7df73a, 0x022fbbe4, 0x6ca8d281, 0x151e3f6b, 0x5850c003}, {0x645e1cfa, 0x903a0a0c, 0x34788c37, 0xfbac54cb, 0x8cf73d78, 0xdc127d11, 0x975d3c82, 0x6d0b5c7c}, {0x14b1ba04, 0xb49d6b05, 0xf00b84f2, 0x56e466b4, 0x0b904f22, 0x30c390cf, 0x3ee254cc, 0x3e11cfb7}, {0xbe8201ab, 0x84dfa547, 0x530715d2, 0x3887ce8b, 0x3eed4ed7, 0xa4c719c6, 0x8f8007b4, 0x18c44950}, {0x7d813cd1, 0xdaf0346d, 0xf755beb1, 0xeccf6f9a, 0xe08143e3, 0x167fce38, 0x6f5d6dfa, 0x545ad9b2}, {0x577605de, 0x973f5466, 0x974f953c, 0x0ce8986e, 0x074382f9, 0x8941cf4b, 0x6fa2672c, 0x156cd7f6}, {0x33b66141, 0x24315404, 0x1992f584, 0x5d1375ab, 0x8b20ca1a, 0xf193ffa6, 0x2701a503, 0x47880cd5}, {0xe9f7b9af, 0xf7b6847d, 0x62c83ce2, 0x9a339673, 0x6e5e6f79, 0xfabf4537, 0x35af33a3, 0x0975acd9}, {0x0eddd248, 0x4fb4204a, 0xc9e509b3, 0x8c98706a, 0x2bb27eb1, 0xd0be8987, 0xc831438b, 0x6ec5f960}, {0x20238f62, 0xa13c95b7, 0x83b476b9, 0x130aa097, 0x14860881, 0x758a04e0, 0x97066493, 0x58e2f8d6}, {0xe8bff41e, 0x65b09c73, 0x37f1c6a3, 0x8b3280e8, 0x2846fb21, 0xe17b82ce, 0xb1ae27df, 0x476534bf}, {0xd5fdb757, 0x8480c0e7, 0x365bf9fd, 0x3644eea0, 0xb776be86, 0x4ca116ca, 0x8b58390c, 0x17b6395f}, {0x252eb0db, 0x2c811e9a, 0x7479e161, 0x1b7d960d, 0xb0a89a26, 0xb3afc7c1, 0x32b5e793, 0x6a2f9533}, {0x08b8a7ad, 0xe877b2c4, 0x341652b4, 0x68b0e8f0, 0xe8b6a2d9, 0x2d44da3b, 0xfd09be59, 0x092778ff}, {0x7988f244, 0x84a1aa6f, 0x24faf63f, 0xa164b3d9, 0xc1bbb915, 0x7aae9724, 0xf386c0d2, 0x24e5d287}, {0x41a1b30c, 0xa70a7efd, 0x39f0e511, 0xc49c55a5, 0x033bb323, 0xab307a8f, 0x17acbd7f, 0x0158abd6}, {0x0f642025, 0x2c228b30, 0x01bd882b, 0xb0878e8d, 0xd7377fea, 0xd862b255, 0xf0490536, 0x18ac3666}};
     // Quick fix for linking issue
@@ -105,13 +97,7 @@ namespace PARAMS_BLS12_381{
     static constexpr storage<limbs_count> omega_inv31=  {0x41a1b30c, 0xa70a7efd, 0x39f0e511, 0xc49c55a5, 0x033bb323, 0xab307a8f, 0x17acbd7f, 0x0158abd6};
     static constexpr storage<limbs_count> omega_inv32=  {0x0f642025, 0x2c228b30, 0x01bd882b, 0xb0878e8d, 0xd7377fea, 0xd862b255, 0xf0490536, 0x18ac3666};
     
-    static constexpr storage_array<omegas_count, limbs_count> omega_inv = {
-        omega_inv1, omega_inv2, omega_inv3, omega_inv4, omega_inv5, omega_inv6, omega_inv7, omega_inv8, 
-        omega_inv9, omega_inv10, omega_inv11, omega_inv12, omega_inv13, omega_inv14, omega_inv15, omega_inv16,
-        omega_inv17, omega_inv18, omega_inv19, omega_inv20, omega_inv21, omega_inv22, omega_inv23, omega_inv24,
-        omega_inv25, omega_inv26, omega_inv27, omega_inv28, omega_inv29, omega_inv30, omega_inv31, omega_inv32,
-    };
-    
+    // static constexpr storage<limbs_count> inv[32]={ {0x80000001, 0x7fffffff, 0x7fff2dff, 0xa9ded201, 0x04d0ec02, 0x199cec04, 0x94cebea4, 0x39f6d3a9}, {0x40000001, 0x3fffffff, 0x3ffec4ff, 0xfece3b02, 0x07396203, 0x266b6206, 0x5f361df6, 0x56f23d7e}, {0x20000001, 0x1fffffff, 0x9ffe907f, 0xa945ef82, 0x086d9d04, 0x2cd29d07, 0xc469cd9f, 0x656ff268}, {0x10000001, 0x0fffffff, 0xcffe763f, 0xfe81c9c2, 0x8907ba84, 0xb0063a87, 0xf703a573, 0x6caeccdd}, {0x08000001, 0x07ffffff, 0xe7fe691f, 0x291fb6e2, 0xc954c945, 0xf1a00947, 0x9050915d, 0x704e3a18}, {0x04000001, 0x03ffffff, 0xf3fe628f, 0x3e6ead72, 0xe97b50a5, 0x126cf0a7, 0xdcf70753, 0x721df0b5}, {0x02000001, 0x01ffffff, 0xf9fe5f47, 0x491628ba, 0xf98e9455, 0xa2d36457, 0x834a424d, 0x7305cc04}, {0x01000001, 0x00ffffff, 0xfcfe5da3, 0x4e69e65e, 0x0198362d, 0xeb069e30, 0xd673dfca, 0x7379b9ab}, {0x00800001, 0x007fffff, 0xfe7e5cd1, 0x5113c530, 0x059d0719, 0x8f203b1c, 0x8008ae89, 0x73b3b07f}, {0x00400001, 0x003fffff, 0xff3e5c68, 0x5268b499, 0x079f6f8f, 0xe12d0992, 0x54d315e8, 0x73d0abe9}, {0x00200001, 0x801fffff, 0x7f9e5c33, 0x53132c4e, 0x08a0a3ca, 0x8a3370cd, 0x3f384998, 0x73df299e}, {0x00100001, 0x400fffff, 0xbfce5c19, 0xd3686828, 0x89213de7, 0x5eb6a46a, 0xb46ae370, 0x73e66878}, {0x00080001, 0x2007ffff, 0xdfe65c0c, 0x93930615, 0x49618af6, 0x48f83e39, 0xef04305c, 0x73ea07e5}, {0x00040001, 0x9003ffff, 0x6ff25c05, 0xf3a8550c, 0xa981b17d, 0x3e190b20, 0x8c50d6d2, 0x73ebd79c}, {0x00020001, 0x4801ffff, 0xb7f85c02, 0xa3b2fc87, 0x5991c4c1, 0x38a97194, 0xdaf72a0d, 0x73ecbf77}, {0x00010001, 0xa400ffff, 0x5bfb5c00, 0x7bb85045, 0x3199ce63, 0xb5f1a4ce, 0x824a53aa, 0x73ed3365}, {0x00008001, 0xd2007fff, 0x2dfcdbff, 0x67bafa24, 0x1d9dd334, 0x7495be6b, 0x55f3e879, 0x73ed6d5c}, {0x00004001, 0x69003fff, 0x96fd9bff, 0xddbc4f13, 0x939fd59c, 0xd3e7cb39, 0xbfc8b2e0, 0x73ed8a57}, {0x00002001, 0x34801fff, 0x4b7dfbff, 0x18bcf98b, 0xcea0d6d1, 0x8390d1a0, 0x74b31814, 0x73ed98d5}, {0x00001001, 0x1a400fff, 0x25be2bff, 0x363d4ec7, 0x6c21576b, 0x5b6554d4, 0x4f284aae, 0x73eda014}, {0x00000801, 0x0d2007ff, 0x12de43ff, 0x44fd7965, 0x3ae197b8, 0x474f966e, 0xbc62e3fb, 0x73eda3b3}, {0x00000401, 0x069003ff, 0x096e4fff, 0xcc5d8eb4, 0x2241b7de, 0xbd44b73b, 0x730030a1, 0x73eda583}, {0x00000201, 0x034801ff, 0x84b655ff, 0x100d995b, 0x95f1c7f2, 0xf83f47a1, 0x4e4ed6f4, 0x73eda66b}, {0x00000101, 0x01a400ff, 0x425a58ff, 0xb1e59eaf, 0xcfc9cffb, 0x95bc8fd4, 0x3bf62a1e, 0x73eda6df}, {0x00000081, 0x00d2007f, 0x212c5a7f, 0x82d1a159, 0x6cb5d400, 0x647b33ee, 0x32c9d3b3, 0x73eda719}, {0x00000041, 0x0069003f, 0x10955b3f, 0xeb47a2ae, 0x3b2bd602, 0xcbda85fb, 0x2e33a87d, 0x73eda736}, {0x00000021, 0x0034801f, 0x8849db9f, 0x1f82a358, 0xa266d704, 0xff8a2f01, 0xabe892e2, 0x73eda744}, {0x00000011, 0x001a400f, 0xc4241bcf, 0xb9a023ad, 0xd6045784, 0x99620384, 0xeac30815, 0x73eda74b}, {0x00000009, 0x000d2007, 0x62113be7, 0x06aee3d8, 0x6fd317c5, 0xe64dedc6, 0x8a3042ae, 0x73eda74f}, {0x00000005, 0x00069003, 0xb107cbf3, 0x2d3643ed, 0x3cba77e5, 0x8cc3e2e7, 0x59e6dffb, 0x73eda751}, {0x00000003, 0x00034801, 0x588313f9, 0x4079f3f8, 0xa32e27f5, 0xdffedd77, 0x41c22ea1, 0x73eda752}, {0x00000002, 0x0001a400, 0xac40b7fc, 0x4a1bcbfd, 0xd667fffd, 0x099c5abf, 0xb5afd5f5, 0x73eda752}};
     // Quick fix for linking issue
     static constexpr storage<limbs_count> inv1=   {0x80000001, 0x7fffffff, 0x7fff2dff, 0xa9ded201, 0x04d0ec02, 0x199cec04, 0x94cebea4, 0x39f6d3a9};
     static constexpr storage<limbs_count> inv2=   {0x40000001, 0x3fffffff, 0x3ffec4ff, 0xfece3b02, 0x07396203, 0x266b6206, 0x5f361df6, 0x56f23d7e};
@@ -145,14 +131,7 @@ namespace PARAMS_BLS12_381{
     static constexpr storage<limbs_count> inv30=  {0x00000005, 0x00069003, 0xb107cbf3, 0x2d3643ed, 0x3cba77e5, 0x8cc3e2e7, 0x59e6dffb, 0x73eda751};
     static constexpr storage<limbs_count> inv31=  {0x00000003, 0x00034801, 0x588313f9, 0x4079f3f8, 0xa32e27f5, 0xdffedd77, 0x41c22ea1, 0x73eda752};
     static constexpr storage<limbs_count> inv32=  {0x00000002, 0x0001a400, 0xac40b7fc, 0x4a1bcbfd, 0xd667fffd, 0x099c5abf, 0xb5afd5f5, 0x73eda752};
-
-    static constexpr storage_array<omegas_count, limbs_count> inv = {
-        inv1, inv2, inv3, inv4, inv5, inv6, inv7, inv8, 
-        inv9, inv10, inv11, inv12, inv13, inv14, inv15, inv16,
-        inv17, inv18, inv19, inv20, inv21, inv22, inv23, inv24,
-        inv25, inv26, inv27, inv28, inv29, inv30, inv31, inv32,
-    }; 
-};
+  };
   
   struct fq_config {
     // field structure size = 12 * 32 bit
@@ -169,19 +148,19 @@ namespace PARAMS_BLS12_381{
                                                             0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
   
     // modulus^2
-    static constexpr storage<2*limbs_count> modulus_squared = {0x1c718e39, 0x26aa0000, 0x76382eab, 0x7ced6b1d, 0x62113cfd, 0x162c3383, 0x3e71b743, 0x66bf91ed, 
+    static constexpr storage<2*limbs_count> modulus_sqared = {0x1c718e39, 0x26aa0000, 0x76382eab, 0x7ced6b1d, 0x62113cfd, 0x162c3383, 0x3e71b743, 0x66bf91ed, 
                                                               0x7091a049, 0x292e85a8, 0x86185c7b, 0x1d68619c, 0x0978ef01, 0xf5314933, 0x16ddca6e, 0x50a62cfd, 
                                                               0x349e8bd0, 0x66e59e49, 0x0e7046b4, 0xe2dc90e5, 0xa22f25e9, 0x4bd278ea, 0xb8c35fc7, 0x02a437a4};
     // 2*modulus^2
-    static constexpr storage<2*limbs_count> modulus_squared_2 = {0x38e31c72, 0x4d540000, 0xec705d56, 0xf9dad63a, 0xc42279fa, 0x2c586706, 0x7ce36e86, 0xcd7f23da, 
+    static constexpr storage<2*limbs_count> modulus_sqared_2 = {0x38e31c72, 0x4d540000, 0xec705d56, 0xf9dad63a, 0xc42279fa, 0x2c586706, 0x7ce36e86, 0xcd7f23da, 
                                                                 0xe1234092, 0x525d0b50, 0x0c30b8f6, 0x3ad0c339, 0x12f1de02, 0xea629266, 0x2dbb94dd, 0xa14c59fa, 
                                                                 0x693d17a0, 0xcdcb3c92, 0x1ce08d68, 0xc5b921ca, 0x445e4bd3, 0x97a4f1d5, 0x7186bf8e, 0x05486f49};
     // 4*modulus^2
-    static constexpr storage<2*limbs_count> modulus_squared_4 = {0x71c638e4, 0x9aa80000, 0xd8e0baac, 0xf3b5ac75, 0x8844f3f5, 0x58b0ce0d, 0xf9c6dd0c, 0x9afe47b4, 
+    static constexpr storage<2*limbs_count> modulus_sqared_4 = {0x71c638e4, 0x9aa80000, 0xd8e0baac, 0xf3b5ac75, 0x8844f3f5, 0x58b0ce0d, 0xf9c6dd0c, 0x9afe47b4, 
                                                                 0xc2468125, 0xa4ba16a1, 0x186171ec, 0x75a18672, 0x25e3bc04, 0xd4c524cc, 0x5b7729bb, 0x4298b3f4, 
                                                                 0xd27a2f41, 0x9b967924, 0x39c11ad1, 0x8b724394, 0x88bc97a7, 0x2f49e3aa, 0xe30d7f1d, 0x0a90de92};
-    static constexpr unsigned modulus_bit_count = 381;
-    // m = floor(2^(2*modulus_bit_count) / modulus)
+    static constexpr unsigned modulus_bits_count = 381;
+    // m = floor(2^(2*modulus_bits_count) / modulus)
     static constexpr storage<limbs_count> m = {0xd59646e8, 0xec4f881f, 0x8163c701, 0x4e65c59e, 0x80a19de7, 0x2f7d1dc7, 0x7fda82a5, 0xa46e09d0, 0x331e9ae8, 0x38a0406c, 0xcf327917, 0x2760d74b};
     static constexpr storage<limbs_count> one = {0x00000001, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
     static constexpr storage<limbs_count> zero = {0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
@@ -190,22 +169,23 @@ namespace PARAMS_BLS12_381{
     // true if i^2 is negative
     static constexpr bool i_squared_is_negative = true;
     // G1 and G2 generators 
-    static constexpr storage<limbs_count> g1_gen_x = {0xdb22c6bb, 0xfb3af00a, 0xf97a1aef, 0x6c55e83f, 0x171bac58, 0xa14e3a3f,
+    static constexpr storage<limbs_count> generator_x = {0xdb22c6bb, 0xfb3af00a, 0xf97a1aef, 0x6c55e83f, 0x171bac58, 0xa14e3a3f,
                                                          0x9774b905, 0xc3688c4f, 0x4fa9ac0f, 0x2695638c, 0x3197d794, 0x17f1d3a7};
-    static constexpr storage<limbs_count> g1_gen_y = {0x46c5e7e1, 0x0caa2329, 0xa2888ae4, 0xd03cc744, 0x2c04b3ed, 0x00db18cb,
+    static constexpr storage<limbs_count> generator_y = {0x46c5e7e1, 0x0caa2329, 0xa2888ae4, 0xd03cc744, 0x2c04b3ed, 0x00db18cb,
                                                          0xd5d00af6, 0xfcf5e095, 0x741d8ae4, 0xa09e30ed, 0xe3aaa0f1, 0x08b3f481};
-    static constexpr storage<limbs_count> g2_gen_x_re = {0xc121bdb8, 0xd48056c8, 0xa805bbef, 0x0bac0326, 0x7ae3d177, 0xb4510b64,
-                                                            0xfa403b02, 0xc6e47ad4, 0x2dc51051, 0x26080527, 0xf08f0a91, 0x024aa2b2};
-    static constexpr storage<limbs_count> g2_gen_x_im = {0x5d042b7e, 0xe5ac7d05, 0x13945d57, 0x334cf112, 0xdc7f5049, 0xb5da61bb,
+    static constexpr storage<limbs_count> generator_x_re = {0xc121bdb8, 0xd48056c8, 0xa805bbef, 0xbac0326, 0x7ae3d177, 0xb4510b64,
+                                                            0xfa403b02, 0xc6e47ad4, 0x2dc51051, 0x26080527, 0xf08f0a91, 0x24aa2b2};
+    static constexpr storage<limbs_count> generator_x_im = {0x5d042b7e, 0xe5ac7d05, 0x13945d57, 0x334cf112, 0xdc7f5049, 0xb5da61bb,
                                                             0x9920b61a, 0x596bd0d0, 0x88274f65, 0x7dacd3a0, 0x52719f60, 0x13e02b60};
-    static constexpr storage<limbs_count> g2_gen_y_re = {0x08b82801, 0xe1935486, 0x3baca289, 0x923ac9cc, 0x5160d12c, 0x6d429a69,
-                                                            0x8cbdd3a7, 0xadfd9baa, 0xda2e351a, 0x8cc9cdc6, 0x727d6e11, 0x0ce5d527};
-    static constexpr storage<limbs_count> g2_gen_y_im = {0xf05f79be, 0xaaa9075f, 0x5cec1da1, 0x3f370d27, 0x572e99ab, 0x267492ab,
-                                                            0x85a763af, 0xcb3e287e, 0x2bc28b99, 0x32acd2b0, 0x2ea734cc, 0x0606c4a0};
+    static constexpr storage<limbs_count> generator_y_re = {0x8b82801, 0xe1935486, 0x3baca289, 0x923ac9cc, 0x5160d12c, 0x6d429a69,
+                                                            0x8cbdd3a7, 0xadfd9baa, 0xda2e351a, 0x8cc9cdc6, 0x727d6e11, 0xce5d527};
+    static constexpr storage<limbs_count> generator_y_im = {0xf05f79be, 0xaaa9075f, 0x5cec1da1, 0x3f370d27, 0x572e99ab, 0x267492ab,
+                                                            0x85a763af, 0xcb3e287e, 0x2bc28b99, 0x32acd2b0, 0x2ea734cc, 0x606c4a0};
   };
   
   static constexpr storage<fq_config::limbs_count> weierstrass_b = {0x00000004, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000,
                                                                     0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
+
   static constexpr storage<fq_config::limbs_count> weierstrass_b_g2_re = {0x00000004, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000,
                                                                           0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
   static constexpr storage<fq_config::limbs_count> weierstrass_b_g2_im = {0x00000004, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000,

icicle-cuda/curves/bn254/params.cuh

@@ -1,39 +1,10 @@
 #pragma once
 #include "../../utils/storage.cuh"
-
-namespace PARAMS_BN254{
+namespace PARAMS{
   struct fp_config{
     static constexpr unsigned limbs_count = 8;
-    static constexpr unsigned omegas_count = 32;
-    
-    static constexpr storage<limbs_count> modulus = {0xf0000001, 0x43e1f593, 0x79b97091, 0x2833e848, 0x8181585d, 0xb85045b6, 0xe131a029, 0x30644e72};
-    static constexpr storage<limbs_count> modulus_2 = {0xe0000002, 0x87c3eb27, 0xf372e122, 0x5067d090, 0x0302b0ba, 0x70a08b6d, 0xc2634053, 0x60c89ce5};
-    static constexpr storage<limbs_count> modulus_4 = {0xc0000004, 0x0f87d64f, 0xe6e5c245, 0xa0cfa121, 0x06056174, 0xe14116da, 0x84c680a6, 0xc19139cb};
-    static constexpr storage<2*limbs_count> modulus_wide = {0xf0000001, 0x43e1f593, 0x79b97091, 0x2833e848, 0x8181585d, 0xb85045b6, 0xe131a029, 0x30644e72, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
-    static constexpr storage<2*limbs_count> modulus_squared = {0xe0000001, 0x08c3eb27, 0xdcb34000, 0xc7f26223, 0x68c9bb7f, 0xffe9a62c, 0xe821ddb0, 0xa6ce1975, 0x47b62fe7, 0x2c77527b, 0xd379d3df, 0x85f73bb0, 0x0348d21c, 0x599a6f7c, 0x763cbf9c, 0x0925c4b8};
-    static constexpr storage<2*limbs_count> modulus_squared_2 = {0xc0000002, 0x1187d64f, 0xb9668000, 0x8fe4c447, 0xd19376ff, 0xffd34c58, 0xd043bb61, 0x4d9c32eb, 0x8f6c5fcf, 0x58eea4f6, 0xa6f3a7be, 0x0bee7761, 0x0691a439, 0xb334def8, 0xec797f38, 0x124b8970};
-    static constexpr storage<2*limbs_count> modulus_squared_4 = {0x80000004, 0x230fac9f, 0x72cd0000, 0x1fc9888f, 0xa326edff, 0xffa698b1, 0xa08776c3, 0x9b3865d7, 0x1ed8bf9e, 0xb1dd49ed, 0x4de74f7c, 0x17dceec3, 0x0d234872, 0x6669bdf0, 0xd8f2fe71, 0x249712e1};
-    static constexpr unsigned modulus_bit_count = 254;
-    static constexpr storage<limbs_count> m = {0xbe1de925, 0x620703a6, 0x09e880ae, 0x71448520, 0x68073014, 0xab074a58, 0x623a04a7, 0x54a47462};
-    static constexpr storage<limbs_count> one = {0x00000001, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
-    static constexpr storage<limbs_count> zero = {0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
 
-    static constexpr storage<limbs_count> omega1= {0xf0000000, 0x43e1f593, 0x79b97091, 0x2833e848, 0x8181585d, 0xb85045b6, 0xe131a029, 0x30644e72};
-    static constexpr storage<limbs_count> omega2= {0x8f703636, 0x23120470, 0xfd736bec, 0x5cea24f6, 0x3fd84104, 0x048b6e19, 0xe131a029, 0x30644e72};
-    static constexpr storage<limbs_count> omega3= {0xc1bd5e80, 0x948dad4a, 0xf8170a0a, 0x52627366, 0x96afef36, 0xec9b9e2f, 0xc8c14f22, 0x2b337de1};
-    static constexpr storage<limbs_count> omega4= {0xe306460b, 0xb11509c6, 0x174efb98, 0x996dfbe1, 0x94dd508c, 0x1c6e4f45, 0x16cbbf4e, 0x21082ca2};
-    static constexpr storage<limbs_count> omega5= {0x3bb512d0, 0x3eed4c53, 0x838eeb1d, 0x9c18d51b, 0x47c0b2a9, 0x9678200d, 0x306b93d2, 0x09c532c6};
-    static constexpr storage<limbs_count> omega6= {0x118f023a, 0xdb94fb05, 0x26e324be, 0x46a6cb24, 0x49bdadf2, 0xc24cdb76, 0x5b080fca, 0x1418144d};
-    static constexpr storage<limbs_count> omega7= {0xba9d1811, 0x9d0e470c, 0xb6f24c79, 0x1dcb5564, 0xe85943e0, 0xdf5ce19c, 0xad310991, 0x16e73dfd};
-    static constexpr storage<limbs_count> omega8= {0x74a57a76, 0xc8936191, 0x6750f230, 0x61794254, 0x9f36ffb0, 0xf086204a, 0xa6148404, 0x07b0c561};
-    static constexpr storage<limbs_count> omega9= {0x470157ce, 0x893a7fa1, 0xfc782d75, 0xe8302a41, 0xdd9b0675, 0xffc02c0e, 0xf6e72f5b, 0x0f1ded1e};
-    static constexpr storage<limbs_count> omega10= {0xbc2e5912, 0x11f995e1, 0xa8d2d7ab, 0x39ba79c0, 0xb08771e3, 0xebbebc2b, 0x7017a420, 0x06fd19c1};
-    static constexpr storage<limbs_count> omega11= {0x769a2ee2, 0xd00a58f9, 0x7494f0ca, 0xb8c12c17, 0xa5355d71, 0xb4027fd7, 0x99c5042b, 0x027a3584};
-    static constexpr storage<limbs_count> omega12= {0x0042d43a, 0x1c477572, 0x6f039bb9, 0x76f169c7, 0xfd5a90a9, 0x01ddd073, 0xde2fd10f, 0x0931d596};
-    static constexpr storage<limbs_count> omega13= {0x9bbdd310, 0x4aa49b8d, 0x8e3a2d76, 0xd31bf3e2, 0x78b2667b, 0x001deac8, 0xb869ae62, 0x006fab49};
-    static constexpr storage<limbs_count> omega14= {0x617c6e85, 0xadaa01c2, 0x7420aae6, 0xb4a93ee1, 0x0ddca8a8, 0x1f4e51b8, 0xcdd9e481, 0x2d965651};
-    static constexpr storage<limbs_count> omega15= {0x4e26ecfb, 0xa93458fd, 0x4115a009, 0x022a2a2d, 0x69ec2bd0, 0x017171fa, 0x5941dc91, 0x2d1ba66f};
-    static constexpr storage<limbs_count> omega16= {0xdaac43b7, 0xd1628ba2, 0xe4347e7d, 0x16c8601d, 0xe081dcff, 0x649abebd, 0x5981ed45, 0x00eeb2cb};
+    ////
     static constexpr storage<limbs_count> omega17= {0xec8af73d, 0x8d24de3c, 0xcf722b45, 0x50f778d4, 0x15bc7dd7, 0xf4506bc3, 0xf94a16e1, 0x0e43ba91};
     static constexpr storage<limbs_count> omega18= {0xd4405b8f, 0x0baa7b44, 0xee0f1394, 0xf8f3c7fe, 0xef0dfe6d, 0x46b153c0, 0x2dde6b95, 0x0ea2bcd9};
     static constexpr storage<limbs_count> omega19= {0x3d1fa34e, 0x5f4dc975, 0x15af81db, 0xc28e54ee, 0x04947d99, 0x83d9a55f, 0x54a2b488, 0x08ec7ccf};
@@ -50,30 +21,7 @@ namespace PARAMS_BN254{
     static constexpr storage<limbs_count> omega30= {0x2be70731, 0x287abbb1, 0x7c35c5aa, 0x5cbcfd1e, 0x1671f4df, 0x7585b3fe, 0xb899c011, 0x08350ecf};
     static constexpr storage<limbs_count> omega31= {0x09f7c5e2, 0x3400c14e, 0x0a649ea1, 0xc112e60c, 0x067ce95e, 0xf7510758, 0xf9daf17c, 0x040a66a5};
     static constexpr storage<limbs_count> omega32= {0x43efecd3, 0x89d65957, 0x3bd6c318, 0x29246adc, 0xce01533c, 0xf9fb5ef6, 0x849078c3, 0x020410e4};
-
-    static constexpr storage_array<omegas_count, limbs_count> omega = {
-        omega1, omega2, omega3, omega4, omega5, omega6, omega7, omega8, 
-        omega9, omega10, omega11, omega12, omega13, omega14, omega15, omega16,
-        omega17, omega18, omega19, omega20, omega21, omega22, omega23, omega24,
-        omega25, omega26, omega27, omega28, omega29, omega30, omega31, omega32,
-    };
-
-    static constexpr storage<limbs_count> omega_inv1= {0xf0000000, 0x43e1f593, 0x79b97091, 0x2833e848, 0x8181585d, 0xb85045b6, 0xe131a029, 0x30644e72};
-    static constexpr storage<limbs_count> omega_inv2= {0x608fc9cb, 0x20cff123, 0x7c4604a5, 0xcb49c351, 0x41a91758, 0xb3c4d79d, 0x00000000, 0x00000000};
-    static constexpr storage<limbs_count> omega_inv3= {0x07b95a9b, 0x8b11d9ab, 0x41671f56, 0x20710ead, 0x30f81dee, 0xfb3acaee, 0x9778465c, 0x130b1711};
-    static constexpr storage<limbs_count> omega_inv4= {0x373428de, 0xb85a71e6, 0xaeb0337e, 0x74954d30, 0x303402b7, 0x2bfc85eb, 0x409556c0, 0x02e40daf};
-    static constexpr storage<limbs_count> omega_inv5= {0xf210979d, 0x8c99980c, 0x34905b4d, 0xef8f3113, 0xdf25d8e7, 0x0aeaf3e7, 0x03bfbd79, 0x27247136};
-    static constexpr storage<limbs_count> omega_inv6= {0x763d698f, 0x78ce6a0b, 0x1d3213ee, 0xd80396ec, 0x67a8a676, 0x035cdc75, 0xb2a13d3a, 0x26177cf2};
-    static constexpr storage<limbs_count> omega_inv7= {0xc64427d7, 0xdddf985f, 0xa49e95bd, 0xaa4f964a, 0x5def8b04, 0x427c045f, 0x7969b732, 0x1641c053};
-    static constexpr storage<limbs_count> omega_inv8= {0x0329f5d6, 0x692c553d, 0x8712848a, 0xa54cf8c6, 0x38e2b5e6, 0x64751ad9, 0x7422fad3, 0x204bd327};
-    static constexpr storage<limbs_count> omega_inv9= {0xaf6b3e4e, 0x52f26c0f, 0xf0bcc0c8, 0x4c277a07, 0xe4fcfcab, 0x546875d5, 0xaa9995b3, 0x09d8f821};
-    static constexpr storage<limbs_count> omega_inv10= {0xb2e5cc71, 0xcaa2e1e9, 0x6e43404e, 0xed42b68e, 0x7a2c7f0a, 0x6ed80915, 0xde3c86d6, 0x1c4042c7};
-    static constexpr storage<limbs_count> omega_inv11= {0x579d71ae, 0x20a3a65d, 0x0adc4420, 0xfd7efed8, 0xfddabf54, 0x3bb6dcd7, 0xbc73d07b, 0x0fa9bb21};
-    static constexpr storage<limbs_count> omega_inv12= {0xc79e0e57, 0xb6f70f8d, 0xa04e05ac, 0x269d3fde, 0x2ba088d9, 0xcf2e371c, 0x11b88d9c, 0x1af864d2};
-    static constexpr storage<limbs_count> omega_inv13= {0xabd95dc9, 0x3b0b205a, 0x978188ca, 0xc8df74fa, 0x6a1cb6c8, 0x08e124db, 0xbfac6104, 0x1670ed58};
-    static constexpr storage<limbs_count> omega_inv14= {0x641c8410, 0xf8eee934, 0x677771c0, 0xf40976b0, 0x558e6e8c, 0x11680d42, 0x06e7e9e9, 0x281c036f};
-    static constexpr storage<limbs_count> omega_inv15= {0xb2dbc0b4, 0xc92a742f, 0x4d384e68, 0xc3f02842, 0x2fa43d0d, 0x22701b6f, 0xe4590b37, 0x05d33766};
-    static constexpr storage<limbs_count> omega_inv16= {0x02d842d4, 0x922d5ac8, 0xc830e4c6, 0x91126414, 0x082f37e0, 0xe92338c0, 0x7fe704e8, 0x0b5d56b7};
+    ////
     static constexpr storage<limbs_count> omega_inv17= {0xec8af73d, 0x8d24de3c, 0xcf722b45, 0x50f778d4, 0x15bc7dd7, 0xf4506bc3, 0xf94a16e1, 0x0e43ba91};
     static constexpr storage<limbs_count> omega_inv18= {0xd4405b8f, 0x0baa7b44, 0xee0f1394, 0xf8f3c7fe, 0xef0dfe6d, 0x46b153c0, 0x2dde6b95, 0x0ea2bcd9};
     static constexpr storage<limbs_count> omega_inv19= {0x3d1fa34e, 0x5f4dc975, 0x15af81db, 0xc28e54ee, 0x04947d99, 0x83d9a55f, 0x54a2b488, 0x08ec7ccf};
@@ -90,30 +38,8 @@ namespace PARAMS_BN254{
     static constexpr storage<limbs_count> omega_inv30= {0x2be70731, 0x287abbb1, 0x7c35c5aa, 0x5cbcfd1e, 0x1671f4df, 0x7585b3fe, 0xb899c011, 0x08350ecf};
     static constexpr storage<limbs_count> omega_inv31= {0x09f7c5e2, 0x3400c14e, 0x0a649ea1, 0xc112e60c, 0x067ce95e, 0xf7510758, 0xf9daf17c, 0x040a66a5};
     static constexpr storage<limbs_count> omega_inv32= {0x43efecd3, 0x89d65957, 0x3bd6c318, 0x29246adc, 0xce01533c, 0xf9fb5ef6, 0x849078c3, 0x020410e4};
-
-    static constexpr storage_array<omegas_count, limbs_count> omega_inv = {
-        omega_inv1, omega_inv2, omega_inv3, omega_inv4, omega_inv5, omega_inv6, omega_inv7, omega_inv8, 
-        omega_inv9, omega_inv10, omega_inv11, omega_inv12, omega_inv13, omega_inv14, omega_inv15, omega_inv16,
-        omega_inv17, omega_inv18, omega_inv19, omega_inv20, omega_inv21, omega_inv22, omega_inv23, omega_inv24,
-        omega_inv25, omega_inv26, omega_inv27, omega_inv28, omega_inv29, omega_inv30, omega_inv31, omega_inv32,
-    };
-    
-    static constexpr storage<limbs_count> inv1= {0xf8000001, 0xa1f0fac9, 0x3cdcb848, 0x9419f424, 0x40c0ac2e, 0xdc2822db, 0x7098d014, 0x18322739};
-    static constexpr storage<limbs_count> inv2= {0xf4000001, 0xf2e9782e, 0x5b4b146c, 0xde26ee36, 0xe1210245, 0x4a3c3448, 0x28e5381f, 0x244b3ad6};
-    static constexpr storage<limbs_count> inv3= {0x72000001, 0x1b65b6e1, 0x6a82427f, 0x832d6b3f, 0xb1512d51, 0x81463cff, 0x850b6c24, 0x2a57c4a4};
-    static constexpr storage<limbs_count> inv4= {0xb1000001, 0x2fa3d63a, 0xf21dd988, 0x55b0a9c3, 0x196942d7, 0x1ccb415b, 0xb31e8627, 0x2d5e098b};
-    static constexpr storage<limbs_count> inv5= {0x50800001, 0xb9c2e5e7, 0x35eba50c, 0x3ef24906, 0xcd754d9a, 0x6a8dc388, 0x4a281328, 0x2ee12bff};
-    static constexpr storage<limbs_count> inv6= {0xa0400001, 0xfed26dbd, 0x57d28ace, 0xb39318a7, 0xa77b52fb, 0x116f049f, 0x15acd9a9, 0x2fa2bd39};
-    static constexpr storage<limbs_count> inv7= {0xc8200001, 0x215a31a8, 0xe8c5fdb0, 0x6de38077, 0x147e55ac, 0x64dfa52b, 0xfb6f3ce9, 0x300385d5};
-    static constexpr storage<limbs_count> inv8= {0x5c100001, 0xb29e139e, 0x313fb720, 0xcb0bb460, 0xcaffd704, 0x8e97f570, 0x6e506e89, 0x3033ea24};
-    static constexpr storage<limbs_count> inv9= {0x26080001, 0xfb400499, 0x557c93d8, 0xf99fce54, 0xa64097b0, 0xa3741d93, 0xa7c10759, 0x304c1c4b};
-    static constexpr storage<limbs_count> inv10= {0x8b040001, 0x1f90fd16, 0x679b0235, 0x10e9db4e, 0x13e0f807, 0xade231a5, 0x447953c1, 0x3058355f};
-    static constexpr storage<limbs_count> inv11= {0x3d820001, 0x31b97955, 0x70aa3963, 0x1c8ee1cb, 0xcab12832, 0xb3193bad, 0x12d579f5, 0x305e41e9};
-    static constexpr storage<limbs_count> inv12= {0x96c10001, 0x3acdb774, 0xf531d4fa, 0xa2616509, 0x26194047, 0xb5b4c0b2, 0xfa038d0f, 0x3061482d};
-    static constexpr storage<limbs_count> inv13= {0x43608001, 0xbf57d684, 0x3775a2c5, 0x654aa6a9, 0x53cd4c52, 0xb7028334, 0x6d9a969c, 0x3062cb50};
-    static constexpr storage<limbs_count> inv14= {0x19b04001, 0x819ce60c, 0xd89789ab, 0xc6bf4778, 0x6aa75257, 0x37a96475, 0xa7661b63, 0x30638ce1};
-    static constexpr storage<limbs_count> inv15= {0x04d82001, 0x62bf6dd0, 0xa9287d1e, 0x777997e0, 0xf614555a, 0x77fcd515, 0x444bddc6, 0x3063edaa};
-    static constexpr storage<limbs_count> inv16= {0xfa6c1001, 0xd350b1b1, 0x9170f6d7, 0xcfd6c014, 0x3bcad6db, 0x18268d66, 0x92bebef8, 0x30641e0e};
+  ////
+  ////
     static constexpr storage<limbs_count> inv17= {0xec8af73d, 0x8d24de3c, 0xcf722b45, 0x50f778d4, 0x15bc7dd7, 0xf4506bc3, 0xf94a16e1, 0x0e43ba91};
     static constexpr storage<limbs_count> inv18= {0xd4405b8f, 0x0baa7b44, 0xee0f1394, 0xf8f3c7fe, 0xef0dfe6d, 0x46b153c0, 0x2dde6b95, 0x0ea2bcd9};
     static constexpr storage<limbs_count> inv19= {0x3d1fa34e, 0x5f4dc975, 0x15af81db, 0xc28e54ee, 0x04947d99, 0x83d9a55f, 0x54a2b488, 0x08ec7ccf};
@@ -131,12 +57,68 @@ namespace PARAMS_BN254{
     static constexpr storage<limbs_count> inv31= {0x09f7c5e2, 0x3400c14e, 0x0a649ea1, 0xc112e60c, 0x067ce95e, 0xf7510758, 0xf9daf17c, 0x040a66a5};
     static constexpr storage<limbs_count> inv32= {0x43efecd3, 0x89d65957, 0x3bd6c318, 0x29246adc, 0xce01533c, 0xf9fb5ef6, 0x849078c3, 0x020410e4};
 
-    static constexpr storage_array<omegas_count, limbs_count> inv = {
-        inv1, inv2, inv3, inv4, inv5, inv6, inv7, inv8, 
-        inv9, inv10, inv11, inv12, inv13, inv14, inv15, inv16,
-        inv17, inv18, inv19, inv20, inv21, inv22, inv23, inv24,
-        inv25, inv26, inv27, inv28, inv29, inv30, inv31, inv32,
-    }; 
+    ////
+
+
+    static constexpr storage<limbs_count> modulus = {0xf0000001, 0x43e1f593, 0x79b97091, 0x2833e848, 0x8181585d, 0xb85045b6, 0xe131a029, 0x30644e72};
+    static constexpr storage<limbs_count> modulus_2 = {0xe0000002, 0x87c3eb27, 0xf372e122, 0x5067d090, 0x0302b0ba, 0x70a08b6d, 0xc2634053, 0x60c89ce5};
+    static constexpr storage<limbs_count> modulus_4 = {0xc0000004, 0x0f87d64f, 0xe6e5c245, 0xa0cfa121, 0x06056174, 0xe14116da, 0x84c680a6, 0xc19139cb};
+    static constexpr storage<2*limbs_count> modulus_wide = {0xf0000001, 0x43e1f593, 0x79b97091, 0x2833e848, 0x8181585d, 0xb85045b6, 0xe131a029, 0x30644e72, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
+    static constexpr storage<2*limbs_count> modulus_sqared = {0xe0000001, 0x08c3eb27, 0xdcb34000, 0xc7f26223, 0x68c9bb7f, 0xffe9a62c, 0xe821ddb0, 0xa6ce1975, 0x47b62fe7, 0x2c77527b, 0xd379d3df, 0x85f73bb0, 0x0348d21c, 0x599a6f7c, 0x763cbf9c, 0x0925c4b8};
+    static constexpr storage<2*limbs_count> modulus_sqared_2 = {0xc0000002, 0x1187d64f, 0xb9668000, 0x8fe4c447, 0xd19376ff, 0xffd34c58, 0xd043bb61, 0x4d9c32eb, 0x8f6c5fcf, 0x58eea4f6, 0xa6f3a7be, 0x0bee7761, 0x0691a439, 0xb334def8, 0xec797f38, 0x124b8970};
+    static constexpr storage<2*limbs_count> modulus_sqared_4 = {0x80000004, 0x230fac9f, 0x72cd0000, 0x1fc9888f, 0xa326edff, 0xffa698b1, 0xa08776c3, 0x9b3865d7, 0x1ed8bf9e, 0xb1dd49ed, 0x4de74f7c, 0x17dceec3, 0x0d234872, 0x6669bdf0, 0xd8f2fe71, 0x249712e1};
+    static constexpr unsigned modulus_bits_count = 254;
+    static constexpr storage<limbs_count> m = {0xbe1de925, 0x620703a6, 0x09e880ae, 0x71448520, 0x68073014, 0xab074a58, 0x623a04a7, 0x54a47462};
+    static constexpr storage<limbs_count> one = {0x00000001, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
+    static constexpr storage<limbs_count> zero = {0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
+    static constexpr storage<limbs_count> omega1= {0xf0000000, 0x43e1f593, 0x79b97091, 0x2833e848, 0x8181585d, 0xb85045b6, 0xe131a029, 0x30644e72};
+    static constexpr storage<limbs_count> omega2= {0x8f703636, 0x23120470, 0xfd736bec, 0x5cea24f6, 0x3fd84104, 0x048b6e19, 0xe131a029, 0x30644e72};
+    static constexpr storage<limbs_count> omega3= {0xc1bd5e80, 0x948dad4a, 0xf8170a0a, 0x52627366, 0x96afef36, 0xec9b9e2f, 0xc8c14f22, 0x2b337de1};
+    static constexpr storage<limbs_count> omega4= {0xe306460b, 0xb11509c6, 0x174efb98, 0x996dfbe1, 0x94dd508c, 0x1c6e4f45, 0x16cbbf4e, 0x21082ca2};
+    static constexpr storage<limbs_count> omega5= {0x3bb512d0, 0x3eed4c53, 0x838eeb1d, 0x9c18d51b, 0x47c0b2a9, 0x9678200d, 0x306b93d2, 0x09c532c6};
+    static constexpr storage<limbs_count> omega6= {0x118f023a, 0xdb94fb05, 0x26e324be, 0x46a6cb24, 0x49bdadf2, 0xc24cdb76, 0x5b080fca, 0x1418144d};
+    static constexpr storage<limbs_count> omega7= {0xba9d1811, 0x9d0e470c, 0xb6f24c79, 0x1dcb5564, 0xe85943e0, 0xdf5ce19c, 0xad310991, 0x16e73dfd};
+    static constexpr storage<limbs_count> omega8= {0x74a57a76, 0xc8936191, 0x6750f230, 0x61794254, 0x9f36ffb0, 0xf086204a, 0xa6148404, 0x07b0c561};
+    static constexpr storage<limbs_count> omega9= {0x470157ce, 0x893a7fa1, 0xfc782d75, 0xe8302a41, 0xdd9b0675, 0xffc02c0e, 0xf6e72f5b, 0x0f1ded1e};
+    static constexpr storage<limbs_count> omega10= {0xbc2e5912, 0x11f995e1, 0xa8d2d7ab, 0x39ba79c0, 0xb08771e3, 0xebbebc2b, 0x7017a420, 0x06fd19c1};
+    static constexpr storage<limbs_count> omega11= {0x769a2ee2, 0xd00a58f9, 0x7494f0ca, 0xb8c12c17, 0xa5355d71, 0xb4027fd7, 0x99c5042b, 0x027a3584};
+    static constexpr storage<limbs_count> omega12= {0x0042d43a, 0x1c477572, 0x6f039bb9, 0x76f169c7, 0xfd5a90a9, 0x01ddd073, 0xde2fd10f, 0x0931d596};
+    static constexpr storage<limbs_count> omega13= {0x9bbdd310, 0x4aa49b8d, 0x8e3a2d76, 0xd31bf3e2, 0x78b2667b, 0x001deac8, 0xb869ae62, 0x006fab49};
+    static constexpr storage<limbs_count> omega14= {0x617c6e85, 0xadaa01c2, 0x7420aae6, 0xb4a93ee1, 0x0ddca8a8, 0x1f4e51b8, 0xcdd9e481, 0x2d965651};
+    static constexpr storage<limbs_count> omega15= {0x4e26ecfb, 0xa93458fd, 0x4115a009, 0x022a2a2d, 0x69ec2bd0, 0x017171fa, 0x5941dc91, 0x2d1ba66f};
+    static constexpr storage<limbs_count> omega16= {0xdaac43b7, 0xd1628ba2, 0xe4347e7d, 0x16c8601d, 0xe081dcff, 0x649abebd, 0x5981ed45, 0x00eeb2cb};
+    static constexpr storage<limbs_count> omega_inv1= {0xf0000000, 0x43e1f593, 0x79b97091, 0x2833e848, 0x8181585d, 0xb85045b6, 0xe131a029, 0x30644e72};
+    static constexpr storage<limbs_count> omega_inv2= {0x608fc9cb, 0x20cff123, 0x7c4604a5, 0xcb49c351, 0x41a91758, 0xb3c4d79d, 0x00000000, 0x00000000};
+    static constexpr storage<limbs_count> omega_inv3= {0x07b95a9b, 0x8b11d9ab, 0x41671f56, 0x20710ead, 0x30f81dee, 0xfb3acaee, 0x9778465c, 0x130b1711};
+    static constexpr storage<limbs_count> omega_inv4= {0x373428de, 0xb85a71e6, 0xaeb0337e, 0x74954d30, 0x303402b7, 0x2bfc85eb, 0x409556c0, 0x02e40daf};
+    static constexpr storage<limbs_count> omega_inv5= {0xf210979d, 0x8c99980c, 0x34905b4d, 0xef8f3113, 0xdf25d8e7, 0x0aeaf3e7, 0x03bfbd79, 0x27247136};
+    static constexpr storage<limbs_count> omega_inv6= {0x763d698f, 0x78ce6a0b, 0x1d3213ee, 0xd80396ec, 0x67a8a676, 0x035cdc75, 0xb2a13d3a, 0x26177cf2};
+    static constexpr storage<limbs_count> omega_inv7= {0xc64427d7, 0xdddf985f, 0xa49e95bd, 0xaa4f964a, 0x5def8b04, 0x427c045f, 0x7969b732, 0x1641c053};
+    static constexpr storage<limbs_count> omega_inv8= {0x0329f5d6, 0x692c553d, 0x8712848a, 0xa54cf8c6, 0x38e2b5e6, 0x64751ad9, 0x7422fad3, 0x204bd327};
+    static constexpr storage<limbs_count> omega_inv9= {0xaf6b3e4e, 0x52f26c0f, 0xf0bcc0c8, 0x4c277a07, 0xe4fcfcab, 0x546875d5, 0xaa9995b3, 0x09d8f821};
+    static constexpr storage<limbs_count> omega_inv10= {0xb2e5cc71, 0xcaa2e1e9, 0x6e43404e, 0xed42b68e, 0x7a2c7f0a, 0x6ed80915, 0xde3c86d6, 0x1c4042c7};
+    static constexpr storage<limbs_count> omega_inv11= {0x579d71ae, 0x20a3a65d, 0x0adc4420, 0xfd7efed8, 0xfddabf54, 0x3bb6dcd7, 0xbc73d07b, 0x0fa9bb21};
+    static constexpr storage<limbs_count> omega_inv12= {0xc79e0e57, 0xb6f70f8d, 0xa04e05ac, 0x269d3fde, 0x2ba088d9, 0xcf2e371c, 0x11b88d9c, 0x1af864d2};
+    static constexpr storage<limbs_count> omega_inv13= {0xabd95dc9, 0x3b0b205a, 0x978188ca, 0xc8df74fa, 0x6a1cb6c8, 0x08e124db, 0xbfac6104, 0x1670ed58};
+    static constexpr storage<limbs_count> omega_inv14= {0x641c8410, 0xf8eee934, 0x677771c0, 0xf40976b0, 0x558e6e8c, 0x11680d42, 0x06e7e9e9, 0x281c036f};
+    static constexpr storage<limbs_count> omega_inv15= {0xb2dbc0b4, 0xc92a742f, 0x4d384e68, 0xc3f02842, 0x2fa43d0d, 0x22701b6f, 0xe4590b37, 0x05d33766};
+    static constexpr storage<limbs_count> omega_inv16= {0x02d842d4, 0x922d5ac8, 0xc830e4c6, 0x91126414, 0x082f37e0, 0xe92338c0, 0x7fe704e8, 0x0b5d56b7};
+    static constexpr storage<limbs_count> inv1= {0xf8000001, 0xa1f0fac9, 0x3cdcb848, 0x9419f424, 0x40c0ac2e, 0xdc2822db, 0x7098d014, 0x18322739};
+    static constexpr storage<limbs_count> inv2= {0xf4000001, 0xf2e9782e, 0x5b4b146c, 0xde26ee36, 0xe1210245, 0x4a3c3448, 0x28e5381f, 0x244b3ad6};
+    static constexpr storage<limbs_count> inv3= {0x72000001, 0x1b65b6e1, 0x6a82427f, 0x832d6b3f, 0xb1512d51, 0x81463cff, 0x850b6c24, 0x2a57c4a4};
+    static constexpr storage<limbs_count> inv4= {0xb1000001, 0x2fa3d63a, 0xf21dd988, 0x55b0a9c3, 0x196942d7, 0x1ccb415b, 0xb31e8627, 0x2d5e098b};
+    static constexpr storage<limbs_count> inv5= {0x50800001, 0xb9c2e5e7, 0x35eba50c, 0x3ef24906, 0xcd754d9a, 0x6a8dc388, 0x4a281328, 0x2ee12bff};
+    static constexpr storage<limbs_count> inv6= {0xa0400001, 0xfed26dbd, 0x57d28ace, 0xb39318a7, 0xa77b52fb, 0x116f049f, 0x15acd9a9, 0x2fa2bd39};
+    static constexpr storage<limbs_count> inv7= {0xc8200001, 0x215a31a8, 0xe8c5fdb0, 0x6de38077, 0x147e55ac, 0x64dfa52b, 0xfb6f3ce9, 0x300385d5};
+    static constexpr storage<limbs_count> inv8= {0x5c100001, 0xb29e139e, 0x313fb720, 0xcb0bb460, 0xcaffd704, 0x8e97f570, 0x6e506e89, 0x3033ea24};
+    static constexpr storage<limbs_count> inv9= {0x26080001, 0xfb400499, 0x557c93d8, 0xf99fce54, 0xa64097b0, 0xa3741d93, 0xa7c10759, 0x304c1c4b};
+    static constexpr storage<limbs_count> inv10= {0x8b040001, 0x1f90fd16, 0x679b0235, 0x10e9db4e, 0x13e0f807, 0xade231a5, 0x447953c1, 0x3058355f};
+    static constexpr storage<limbs_count> inv11= {0x3d820001, 0x31b97955, 0x70aa3963, 0x1c8ee1cb, 0xcab12832, 0xb3193bad, 0x12d579f5, 0x305e41e9};
+    static constexpr storage<limbs_count> inv12= {0x96c10001, 0x3acdb774, 0xf531d4fa, 0xa2616509, 0x26194047, 0xb5b4c0b2, 0xfa038d0f, 0x3061482d};
+    static constexpr storage<limbs_count> inv13= {0x43608001, 0xbf57d684, 0x3775a2c5, 0x654aa6a9, 0x53cd4c52, 0xb7028334, 0x6d9a969c, 0x3062cb50};
+    static constexpr storage<limbs_count> inv14= {0x19b04001, 0x819ce60c, 0xd89789ab, 0xc6bf4778, 0x6aa75257, 0x37a96475, 0xa7661b63, 0x30638ce1};
+    static constexpr storage<limbs_count> inv15= {0x04d82001, 0x62bf6dd0, 0xa9287d1e, 0x777997e0, 0xf614555a, 0x77fcd515, 0x444bddc6, 0x3063edaa};
+    static constexpr storage<limbs_count> inv16= {0xfa6c1001, 0xd350b1b1, 0x9170f6d7, 0xcfd6c014, 0x3bcad6db, 0x18268d66, 0x92bebef8, 0x30641e0e};
   };
 
   struct fq_config{
@@ -145,10 +127,10 @@ namespace PARAMS_BN254{
     static constexpr storage<limbs_count> modulus_2 = {0xb0f9fa8e, 0x7841182d, 0xd0e3951a, 0x2f02d522, 0x0302b0bb, 0x70a08b6d, 0xc2634053, 0x60c89ce5};
     static constexpr storage<limbs_count> modulus_4 = {0x61f3f51c, 0xf082305b, 0xa1c72a34, 0x5e05aa45, 0x06056176, 0xe14116da, 0x84c680a6, 0xc19139cb};
     static constexpr storage<2*limbs_count> modulus_wide = {0xd87cfd47, 0x3c208c16, 0x6871ca8d, 0x97816a91, 0x8181585d, 0xb85045b6, 0xe131a029, 0x30644e72, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
-    static constexpr storage<2*limbs_count> modulus_squared = {0x275d69b1, 0x3b5458a2, 0x09eac101, 0xa602072d, 0x6d96cadc, 0x4a50189c, 0x7a1242c8, 0x04689e95, 0x34c6b38d, 0x26edfa5c, 0x16375606, 0xb00b8551, 0x0348d21c, 0x599a6f7c, 0x763cbf9c, 0x0925c4b8};
-    static constexpr storage<2*limbs_count> modulus_squared_2 = {0x4ebad362, 0x76a8b144, 0x13d58202, 0x4c040e5a, 0xdb2d95b9, 0x94a03138, 0xf4248590, 0x08d13d2a, 0x698d671a, 0x4ddbf4b8, 0x2c6eac0c, 0x60170aa2, 0x0691a439, 0xb334def8, 0xec797f38, 0x124b8970};
-    static constexpr storage<2*limbs_count> modulus_squared_4 = {0x9d75a6c4, 0xed516288, 0x27ab0404, 0x98081cb4, 0xb65b2b72, 0x29406271, 0xe8490b21, 0x11a27a55, 0xd31ace34, 0x9bb7e970, 0x58dd5818, 0xc02e1544, 0x0d234872, 0x6669bdf0, 0xd8f2fe71, 0x249712e1};
-    static constexpr unsigned modulus_bit_count = 254;
+    static constexpr storage<2*limbs_count> modulus_sqared = {0x275d69b1, 0x3b5458a2, 0x09eac101, 0xa602072d, 0x6d96cadc, 0x4a50189c, 0x7a1242c8, 0x04689e95, 0x34c6b38d, 0x26edfa5c, 0x16375606, 0xb00b8551, 0x0348d21c, 0x599a6f7c, 0x763cbf9c, 0x0925c4b8};
+    static constexpr storage<2*limbs_count> modulus_sqared_2 = {0x4ebad362, 0x76a8b144, 0x13d58202, 0x4c040e5a, 0xdb2d95b9, 0x94a03138, 0xf4248590, 0x08d13d2a, 0x698d671a, 0x4ddbf4b8, 0x2c6eac0c, 0x60170aa2, 0x0691a439, 0xb334def8, 0xec797f38, 0x124b8970};
+    static constexpr storage<2*limbs_count> modulus_sqared_4 = {0x9d75a6c4, 0xed516288, 0x27ab0404, 0x98081cb4, 0xb65b2b72, 0x29406271, 0xe8490b21, 0x11a27a55, 0xd31ace34, 0x9bb7e970, 0x58dd5818, 0xc02e1544, 0x0d234872, 0x6669bdf0, 0xd8f2fe71, 0x249712e1};
+    static constexpr unsigned modulus_bits_count = 254;
     static constexpr storage<limbs_count> m = {0x19bf90e5, 0x6f3aed8a, 0x67cd4c08, 0xae965e17, 0x68073013, 0xab074a58, 0x623a04a7, 0x54a47462};
     static constexpr storage<limbs_count> one = {0x00000001, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
     static constexpr storage<limbs_count> zero = {0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
@@ -157,15 +139,17 @@ namespace PARAMS_BN254{
     // true if i^2 is negative
     static constexpr bool i_squared_is_negative = true;
     // G1 and G2 generators 
-    static constexpr storage<limbs_count> g1_gen_x = {0x00000001, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
-    static constexpr storage<limbs_count> g1_gen_y = {0x00000002, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
-    static constexpr storage<limbs_count> g2_gen_x_re = {0xd992f6ed, 0x46debd5c, 0xf75edadd, 0x674322d4, 0x5e5c4479, 0x426a0066, 0x121f1e76, 0x1800deef};
-    static constexpr storage<limbs_count> g2_gen_x_im = {0xaef312c2, 0x97e485b7, 0x35a9e712, 0xf1aa4933, 0x31fb5d25, 0x7260bfb7, 0x920d483a, 0x198e9393};
-    static constexpr storage<limbs_count> g2_gen_y_re = {0x66fa7daa, 0x4ce6cc01, 0x0c43d37b, 0xe3d1e769, 0x8dcb408f, 0x4aab7180, 0xdb8c6deb, 0x12c85ea5};
-    static constexpr storage<limbs_count> g2_gen_y_im = {0xd122975b, 0x55acdadc, 0x70b38ef3, 0xbc4b3133, 0x690c3395, 0xec9e99ad, 0x585ff075, 0x090689d0};
+    static constexpr storage<limbs_count> generator_x = {0x00000001, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
+    static constexpr storage<limbs_count> generator_y = {0x00000002, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
+    static constexpr storage<limbs_count> generator_x_re = {0x00000001, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
+    static constexpr storage<limbs_count> generator_x_im = {0x00000001, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
+    static constexpr storage<limbs_count> generator_y_re = {0x00000001, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
+    static constexpr storage<limbs_count> generator_y_im = {0x00000001, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
   };
 
   static constexpr storage<fq_config::limbs_count> weierstrass_b = {0x00000003, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
-  static constexpr storage<fq_config::limbs_count> weierstrass_b_g2_re = {0x24a138e5, 0x3267e6dc, 0x59dbefa3, 0xb5b4c5e5, 0x1be06ac3, 0x81be1899, 0xceb8aaae, 0x2b149d40};
-  static constexpr storage<fq_config::limbs_count> weierstrass_b_g2_im = {0x85c315d2, 0xe4a2bd06, 0xe52d1852, 0xa74fa084, 0xeed8fdf4, 0xcd2cafad, 0x3af0fed4, 0x009713b0};
+
+  // TODO: correct parameters for G2 here
+  static constexpr storage<fq_config::limbs_count> weierstrass_b_g2_re = {0x00000001, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
+  static constexpr storage<fq_config::limbs_count> weierstrass_b_g2_im = {0x00000001, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000, 0x00000000};
 }

icicle-cuda/curves/curve_config.cuh

@@ -0,0 +1,28 @@
+#pragma once
+
+#include "../primitives/field.cuh"
+#include "../primitives/projective.cuh"
+
+#if defined(FEATURE_BLS12_381)
+#include "bls12_381/params.cuh"
+#elif defined(FEATURE_BLS12_377)
+#include "bls12_377/params.cuh"
+#elif defined(FEATURE_BN254)
+#include "bn254/params.cuh"
+#else
+# error "no FEATURE"
+#endif
+
+typedef Field<PARAMS::fp_config> scalar_field_t;
+typedef scalar_field_t scalar_t;
+typedef Field<PARAMS::fq_config> point_field_t;
+static constexpr point_field_t b = point_field_t{ PARAMS::weierstrass_b };
+typedef Projective<point_field_t, scalar_field_t, b> projective_t;
+typedef Affine<point_field_t> affine_t;
+#if defined(G2_DEFINED)
+typedef ExtensionField<PARAMS::fq_config> g2_point_field_t;
+static constexpr g2_point_field_t b_g2 = g2_point_field_t{ point_field_t{ PARAMS::weierstrass_b_g2_re },
+                                                            point_field_t{ PARAMS::weierstrass_b_g2_im }};
+typedef Projective<g2_point_field_t, scalar_field_t, b_g2> g2_projective_t;
+typedef Affine<g2_point_field_t> g2_affine_t;
+#endif

icicle-cuda/curves/index.cu

@@ -1,5 +1,5 @@
-
 #include "projective.cu"
 #include "lde.cu"
 #include "msm.cu"
 #include "ve_mod_mult.cu"
+#include "poseidon.cu"

icicle-cuda/curves/lde.cu

@@ -1,20 +1,20 @@
-#ifndef _BN254_LDE
-#define _BN254_LDE
+#ifndef _LDE
+#define _LDE
 #include <cuda.h>
-#include "../../appUtils/ntt/lde.cu"
-#include "../../appUtils/ntt/ntt.cuh"
-#include "../../appUtils/vector_manipulation/ve_mod_mult.cuh"
+#include "../appUtils/ntt/lde.cu"
+#include "../appUtils/ntt/ntt.cuh"
+#include "../appUtils/vector_manipulation/ve_mod_mult.cuh"
 #include "curve_config.cuh"
 
-extern "C" BN254::scalar_t* build_domain_cuda_bn254(uint32_t domain_size, uint32_t logn, bool inverse, size_t device_id = 0, cudaStream_t stream = 0)
+extern "C" scalar_t* build_domain_cuda(uint32_t domain_size, uint32_t logn, bool inverse, size_t device_id = 0, cudaStream_t stream = 0)
 {
     try
     {
         cudaStreamCreate(&stream);
         if (inverse) {
-            return fill_twiddle_factors_array(domain_size, BN254::scalar_t::omega_inv(logn), stream);
+            return fill_twiddle_factors_array(domain_size, scalar_t::omega_inv(logn), stream);
         } else {
-            return fill_twiddle_factors_array(domain_size, BN254::scalar_t::omega(logn), stream);
+            return fill_twiddle_factors_array(domain_size, scalar_t::omega(logn), stream);
         }
     }
     catch (const std::runtime_error &ex)
@@ -24,12 +24,12 @@ extern "C" BN254::scalar_t* build_domain_cuda_bn254(uint32_t domain_size, uint32
     }
 }
 
-extern "C" int ntt_cuda_bn254(BN254::scalar_t *arr, uint32_t n, bool inverse, size_t device_id = 0, cudaStream_t stream = 0)
+extern "C" int ntt_cuda(scalar_t *arr, uint32_t n, bool inverse, size_t device_id = 0, cudaStream_t stream = 0)
 {
     try
     {
         cudaStreamCreate(&stream);
-        return ntt_end2end_template<BN254::scalar_t,BN254::scalar_t>(arr, n, inverse, stream); // TODO: pass device_id
+        return ntt_end2end_template<scalar_t,scalar_t>(arr, n, inverse, stream); // TODO: pass device_id
     }
     catch (const std::runtime_error &ex)
     {
@@ -39,12 +39,12 @@ extern "C" int ntt_cuda_bn254(BN254::scalar_t *arr, uint32_t n, bool inverse, si
     }
 }
 
-extern "C" int ecntt_cuda_bn254(BN254::projective_t *arr, uint32_t n, bool inverse, size_t device_id = 0, cudaStream_t stream = 0)
+extern "C" int ecntt_cuda(projective_t *arr, uint32_t n, bool inverse, size_t device_id = 0, cudaStream_t stream = 0)
 {
     try
     {
         cudaStreamCreate(&stream);
-        return ntt_end2end_template<BN254::projective_t,BN254::scalar_t>(arr, n, inverse, stream); // TODO: pass device_id
+        return ntt_end2end_template<projective_t,scalar_t>(arr, n, inverse, stream); // TODO: pass device_id
     }
     catch (const std::runtime_error &ex)
     {
@@ -53,12 +53,12 @@ extern "C" int ecntt_cuda_bn254(BN254::projective_t *arr, uint32_t n, bool inver
     }
 }
 
-extern "C" int ntt_batch_cuda_bn254(BN254::scalar_t *arr, uint32_t arr_size, uint32_t batch_size, bool inverse, size_t device_id = 0, cudaStream_t stream = 0)
+extern "C" int ntt_batch_cuda(scalar_t *arr, uint32_t arr_size, uint32_t batch_size, bool inverse, size_t device_id = 0, cudaStream_t stream = 0)
 {
     try
     {
         cudaStreamCreate(&stream);
-        return ntt_end2end_batch_template<BN254::scalar_t,BN254::scalar_t>(arr, arr_size, batch_size, inverse, stream); // TODO: pass device_id
+        return ntt_end2end_batch_template<scalar_t,scalar_t>(arr, arr_size, batch_size, inverse, stream); // TODO: pass device_id
     }
     catch (const std::runtime_error &ex)
     {
@@ -67,12 +67,12 @@ extern "C" int ntt_batch_cuda_bn254(BN254::scalar_t *arr, uint32_t arr_size, uin
     }
 }
 
-extern "C" int ecntt_batch_cuda_bn254(BN254::projective_t *arr, uint32_t arr_size, uint32_t batch_size, bool inverse, size_t device_id = 0, cudaStream_t stream = 0)
+extern "C" int ecntt_batch_cuda(projective_t *arr, uint32_t arr_size, uint32_t batch_size, bool inverse, size_t device_id = 0, cudaStream_t stream = 0)
 {
     try
     {
         cudaStreamCreate(&stream);
-        return ntt_end2end_batch_template<BN254::projective_t,BN254::scalar_t>(arr, arr_size, batch_size, inverse, stream); // TODO: pass device_id
+        return ntt_end2end_batch_template<projective_t,scalar_t>(arr, arr_size, batch_size, inverse, stream); // TODO: pass device_id
     }
     catch (const std::runtime_error &ex)
     {
@@ -81,7 +81,7 @@ extern "C" int ecntt_batch_cuda_bn254(BN254::projective_t *arr, uint32_t arr_siz
     }
 }
 
-extern "C" int interpolate_scalars_cuda_bn254(BN254::scalar_t* d_out, BN254::scalar_t *d_evaluations, BN254::scalar_t *d_domain, unsigned n, unsigned device_id = 0, cudaStream_t stream = 0)
+extern "C" int interpolate_scalars_cuda(scalar_t* d_out, scalar_t *d_evaluations, scalar_t *d_domain, unsigned n, unsigned device_id = 0, cudaStream_t stream = 0)
 {
     try
     {
@@ -94,7 +94,7 @@ extern "C" int interpolate_scalars_cuda_bn254(BN254::scalar_t* d_out, BN254::sca
     }
 }
 
-extern "C" int interpolate_scalars_batch_cuda_bn254(BN254::scalar_t* d_out, BN254::scalar_t* d_evaluations, BN254::scalar_t* d_domain, unsigned n,
+extern "C" int interpolate_scalars_batch_cuda(scalar_t* d_out, scalar_t* d_evaluations, scalar_t* d_domain, unsigned n,
                                               unsigned batch_size, size_t device_id = 0, cudaStream_t stream = 0)
 {
     try
@@ -109,7 +109,7 @@ extern "C" int interpolate_scalars_batch_cuda_bn254(BN254::scalar_t* d_out, BN25
     }
 }
 
-extern "C" int interpolate_points_cuda_bn254(BN254::projective_t* d_out, BN254::projective_t *d_evaluations, BN254::scalar_t *d_domain, unsigned n, size_t device_id = 0, cudaStream_t stream = 0)
+extern "C" int interpolate_points_cuda(projective_t* d_out, projective_t *d_evaluations, scalar_t *d_domain, unsigned n, size_t device_id = 0, cudaStream_t stream = 0)
 {
     try
     {
@@ -122,7 +122,7 @@ extern "C" int interpolate_points_cuda_bn254(BN254::projective_t* d_out, BN254::
     }
 }
 
-extern "C" int interpolate_points_batch_cuda_bn254(BN254::projective_t* d_out, BN254::projective_t* d_evaluations, BN254::scalar_t* d_domain,
+extern "C" int interpolate_points_batch_cuda(projective_t* d_out, projective_t* d_evaluations, scalar_t* d_domain,
                                              unsigned n, unsigned batch_size, size_t device_id = 0, cudaStream_t stream = 0)
 {
     try
@@ -137,12 +137,12 @@ extern "C" int interpolate_points_batch_cuda_bn254(BN254::projective_t* d_out, B
     }
 }
 
-extern "C" int evaluate_scalars_cuda_bn254(BN254::scalar_t* d_out, BN254::scalar_t *d_coefficients, BN254::scalar_t *d_domain, 
+extern "C" int evaluate_scalars_cuda(scalar_t* d_out, scalar_t *d_coefficients, scalar_t *d_domain, 
                                      unsigned domain_size, unsigned n, unsigned device_id = 0, cudaStream_t stream = 0)
 {
     try
     {
-        BN254::scalar_t* _null = nullptr;
+        scalar_t* _null = nullptr;
         cudaStreamCreate(&stream);
         return evaluate(d_out, d_coefficients, d_domain, domain_size, n, false, _null, stream);
     }
@@ -153,12 +153,12 @@ extern "C" int evaluate_scalars_cuda_bn254(BN254::scalar_t* d_out, BN254::scalar
     }
 }
 
-extern "C" int evaluate_scalars_batch_cuda_bn254(BN254::scalar_t* d_out, BN254::scalar_t* d_coefficients, BN254::scalar_t* d_domain, unsigned domain_size,
+extern "C" int evaluate_scalars_batch_cuda(scalar_t* d_out, scalar_t* d_coefficients, scalar_t* d_domain, unsigned domain_size,
                                            unsigned n, unsigned batch_size, size_t device_id = 0, cudaStream_t stream = 0)
 {
     try
     {
-        BN254::scalar_t* _null = nullptr;
+        scalar_t* _null = nullptr;
         cudaStreamCreate(&stream);
         return evaluate_batch(d_out, d_coefficients, d_domain, domain_size, n, batch_size, false, _null, stream);
     }
@@ -169,12 +169,12 @@ extern "C" int evaluate_scalars_batch_cuda_bn254(BN254::scalar_t* d_out, BN254::
     }
 }
 
-extern "C" int evaluate_points_cuda_bn254(BN254::projective_t* d_out, BN254::projective_t *d_coefficients, BN254::scalar_t *d_domain, 
+extern "C" int evaluate_points_cuda(projective_t* d_out, projective_t *d_coefficients, scalar_t *d_domain, 
                                     unsigned domain_size, unsigned n, size_t device_id = 0, cudaStream_t stream = 0)
 {
     try
     {
-        BN254::scalar_t* _null = nullptr;
+        scalar_t* _null = nullptr;
         cudaStreamCreate(&stream);
         return evaluate(d_out, d_coefficients, d_domain, domain_size, n, false, _null, stream);
     }
@@ -185,12 +185,12 @@ extern "C" int evaluate_points_cuda_bn254(BN254::projective_t* d_out, BN254::pro
     }
 }
 
-extern "C" int evaluate_points_batch_cuda_bn254(BN254::projective_t* d_out, BN254::projective_t* d_coefficients, BN254::scalar_t* d_domain, unsigned domain_size,
+extern "C" int evaluate_points_batch_cuda(projective_t* d_out, projective_t* d_coefficients, scalar_t* d_domain, unsigned domain_size,
                                           unsigned n, unsigned batch_size, size_t device_id = 0, cudaStream_t stream = 0)
 {
     try
     {
-        BN254::scalar_t* _null = nullptr;
+        scalar_t* _null = nullptr;
         cudaStreamCreate(&stream);
         return evaluate_batch(d_out, d_coefficients, d_domain, domain_size, n, batch_size, false, _null, stream);
     }
@@ -201,8 +201,8 @@ extern "C" int evaluate_points_batch_cuda_bn254(BN254::projective_t* d_out, BN25
     }
 }
 
-extern "C" int evaluate_scalars_on_coset_cuda_bn254(BN254::scalar_t* d_out, BN254::scalar_t *d_coefficients, BN254::scalar_t *d_domain, unsigned domain_size,
-                                              unsigned n, BN254::scalar_t *coset_powers, unsigned device_id = 0, cudaStream_t stream = 0)
+extern "C" int evaluate_scalars_on_coset_cuda(scalar_t* d_out, scalar_t *d_coefficients, scalar_t *d_domain, unsigned domain_size,
+                                              unsigned n, scalar_t *coset_powers, unsigned device_id = 0, cudaStream_t stream = 0)
 {
     try
     {
@@ -216,8 +216,8 @@ extern "C" int evaluate_scalars_on_coset_cuda_bn254(BN254::scalar_t* d_out, BN25
     }
 }
 
-extern "C" int evaluate_scalars_on_coset_batch_cuda_bn254(BN254::scalar_t* d_out, BN254::scalar_t* d_coefficients, BN254::scalar_t* d_domain, unsigned domain_size, 
-                                                    unsigned n, unsigned batch_size, BN254::scalar_t *coset_powers, size_t device_id = 0, cudaStream_t stream = 0)
+extern "C" int evaluate_scalars_on_coset_batch_cuda(scalar_t* d_out, scalar_t* d_coefficients, scalar_t* d_domain, unsigned domain_size, 
+                                                    unsigned n, unsigned batch_size, scalar_t *coset_powers, size_t device_id = 0, cudaStream_t stream = 0)
 {
     try
     {
@@ -231,8 +231,8 @@ extern "C" int evaluate_scalars_on_coset_batch_cuda_bn254(BN254::scalar_t* d_out
     }
 }
 
-extern "C" int evaluate_points_on_coset_cuda_bn254(BN254::projective_t* d_out, BN254::projective_t *d_coefficients, BN254::scalar_t *d_domain, unsigned domain_size,
-                                             unsigned n, BN254::scalar_t *coset_powers, size_t device_id = 0, cudaStream_t stream = 0)
+extern "C" int evaluate_points_on_coset_cuda(projective_t* d_out, projective_t *d_coefficients, scalar_t *d_domain, unsigned domain_size,
+                                             unsigned n, scalar_t *coset_powers, size_t device_id = 0, cudaStream_t stream = 0)
 {
     try
     {
@@ -246,8 +246,8 @@ extern "C" int evaluate_points_on_coset_cuda_bn254(BN254::projective_t* d_out, B
     }
 }
 
-extern "C" int evaluate_points_on_coset_batch_cuda_bn254(BN254::projective_t* d_out, BN254::projective_t* d_coefficients, BN254::scalar_t* d_domain, unsigned domain_size, 
-                                                   unsigned n, unsigned batch_size, BN254::scalar_t *coset_powers, size_t device_id = 0, cudaStream_t stream = 0)
+extern "C" int evaluate_points_on_coset_batch_cuda(projective_t* d_out, projective_t* d_coefficients, scalar_t* d_domain, unsigned domain_size, 
+                                                   unsigned n, unsigned batch_size, scalar_t *coset_powers, size_t device_id = 0, cudaStream_t stream = 0)
 {
     try
     {
@@ -261,7 +261,7 @@ extern "C" int evaluate_points_on_coset_batch_cuda_bn254(BN254::projective_t* d_
     }
 }
 
-extern "C" int reverse_order_scalars_cuda_bn254(BN254::scalar_t* arr, int n, size_t device_id = 0, cudaStream_t stream = 0)
+extern "C" int reverse_order_scalars_cuda(scalar_t* arr, int n, size_t device_id = 0, cudaStream_t stream = 0)
 {
     try
     {
@@ -277,7 +277,7 @@ extern "C" int reverse_order_scalars_cuda_bn254(BN254::scalar_t* arr, int n, siz
     }
 }
 
-extern "C" int reverse_order_scalars_batch_cuda_bn254(BN254::scalar_t* arr, int n, int batch_size, size_t device_id = 0, cudaStream_t stream = 0)
+extern "C" int reverse_order_scalars_batch_cuda(scalar_t* arr, int n, int batch_size, size_t device_id = 0, cudaStream_t stream = 0)
 {
     try
     {
@@ -293,7 +293,7 @@ extern "C" int reverse_order_scalars_batch_cuda_bn254(BN254::scalar_t* arr, int
     }
 }
 
-extern "C" int reverse_order_points_cuda_bn254(BN254::projective_t* arr, int n, size_t device_id = 0, cudaStream_t stream = 0)
+extern "C" int reverse_order_points_cuda(projective_t* arr, int n, size_t device_id = 0, cudaStream_t stream = 0)
 {
     try
     {
@@ -309,7 +309,7 @@ extern "C" int reverse_order_points_cuda_bn254(BN254::projective_t* arr, int n,
     }
 }
 
-extern "C" int reverse_order_points_batch_cuda_bn254(BN254::projective_t* arr, int n, int batch_size, size_t device_id = 0, cudaStream_t stream = 0)
+extern "C" int reverse_order_points_batch_cuda(projective_t* arr, int n, int batch_size, size_t device_id = 0, cudaStream_t stream = 0)
 {
     try
     {

icicle-cuda/curves/msm.cu

@@ -1,18 +1,18 @@
-#ifndef _BN254_MSM
-#define _BN254_MSM
-#include "../../appUtils/msm/msm.cu"
+#ifndef _MSM
+#define _MSM
+#include "../appUtils/msm/msm.cu"
 #include <stdexcept>
 #include <cuda.h>
 #include "curve_config.cuh"
 
 
 extern "C"
-int msm_cuda_bn254(BN254::projective_t *out, BN254::affine_t points[],
-              BN254::scalar_t scalars[], size_t count, size_t device_id = 0, cudaStream_t stream = 0)
+int msm_cuda(projective_t *out, affine_t points[],
+              scalar_t scalars[], size_t count, size_t device_id = 0, cudaStream_t stream = 0)
 {
     try
     {
-        large_msm<BN254::scalar_t, BN254::projective_t, BN254::affine_t>(scalars, points, count, out, false, false, stream);
+        large_msm<scalar_t, projective_t, affine_t>(scalars, points, count, out, false, stream);
         return CUDA_SUCCESS;
     }
     catch (const std::runtime_error &ex)
@@ -22,13 +22,13 @@ int msm_cuda_bn254(BN254::projective_t *out, BN254::affine_t points[],
     }
 }
 
-extern "C" int msm_batch_cuda_bn254(BN254::projective_t* out, BN254::affine_t points[],
-                              BN254::scalar_t scalars[], size_t batch_size, size_t msm_size, size_t device_id = 0, cudaStream_t stream = 0)
+extern "C" int msm_batch_cuda(projective_t* out, affine_t points[],
+                              scalar_t scalars[], size_t batch_size, size_t msm_size, size_t device_id = 0, cudaStream_t stream = 0)
 {
   try
   {
     cudaStreamCreate(&stream);
-    batched_large_msm<BN254::scalar_t, BN254::projective_t, BN254::affine_t>(scalars, points, batch_size, msm_size, out, false, stream);
+    batched_large_msm<scalar_t, projective_t, affine_t>(scalars, points, batch_size, msm_size, out, false, stream);
     cudaStreamSynchronize(stream);
     return CUDA_SUCCESS;
   }
@@ -48,11 +48,11 @@ extern "C" int msm_batch_cuda_bn254(BN254::projective_t* out, BN254::affine_t po
  * @param count Length of `d_scalars` and `d_points` arrays (they should have equal length).
  */
  extern "C"
- int commit_cuda_bn254(BN254::projective_t* d_out, BN254::scalar_t* d_scalars, BN254::affine_t* d_points, size_t count, size_t device_id = 0, cudaStream_t stream = 0)
+ int commit_cuda(projective_t* d_out, scalar_t* d_scalars, affine_t* d_points, size_t count, size_t device_id = 0, cudaStream_t stream = 0)
  {
      try
      {
-         large_msm(d_scalars, d_points, count, d_out, true, false, stream);
+         large_msm(d_scalars, d_points, count, d_out, true, stream);
          cudaStreamSynchronize(stream);
          return 0;
      }
@@ -73,7 +73,7 @@ extern "C" int msm_batch_cuda_bn254(BN254::projective_t* out, BN254::affine_t po
   * @param batch_size Size of the batch.
   */
  extern "C"
- int commit_batch_cuda_bn254(BN254::projective_t* d_out, BN254::scalar_t* d_scalars, BN254::affine_t* d_points, size_t count, size_t batch_size, size_t device_id = 0, cudaStream_t stream = 0)
+ int commit_batch_cuda(projective_t* d_out, scalar_t* d_scalars, affine_t* d_points, size_t count, size_t batch_size, size_t device_id = 0, cudaStream_t stream = 0)
  {
      try
      {

icicle-cuda/curves/poseidon.cu

@@ -0,0 +1,26 @@
+#ifndef _POSEIDON
+#define _POSEIDON
+#include <cuda.h>
+#include <stdexcept>
+#include "../appUtils/poseidon/poseidon.cu"
+#include "curve_config.cuh"
+
+template class Poseidon<scalar_t>;
+
+extern "C" int poseidon_multi_cuda(scalar_t input[], scalar_t* out,
+                                             size_t number_of_blocks, int arity, size_t device_id = 0)
+{
+  try
+  {
+    Poseidon<scalar_t> poseidon(arity);
+    poseidon.hash_blocks(input, number_of_blocks, out, Poseidon<scalar_t>::HashType::MerkleTree);
+
+    return CUDA_SUCCESS;
+  }
+  catch (const std::runtime_error &ex)
+  {
+    printf("error %s", ex.what());
+    return -1;
+  }
+}
+#endif

icicle-cuda/curves/projective.cu

@@ -0,0 +1,19 @@
+#include <cuda.h>
+#include "curve_config.cuh"
+#include "../primitives/projective.cuh"
+
+extern "C" bool eq(projective_t *point1, projective_t *point2)
+{
+  return (*point1 == *point2) && 
+  !((point1->x == point_field_t::zero()) && (point1->y == point_field_t::zero()) && (point1->z == point_field_t::zero())) && 
+  !((point2->x == point_field_t::zero()) && (point2->y == point_field_t::zero()) && (point2->z == point_field_t::zero()));
+}
+
+#if defined(G2_DEFINED)
+extern "C" bool eq_g2(g2_projective_t *point1, g2_projective_t *point2)
+{
+  return (*point1 == *point2) && 
+  !((point1->x == g2_point_field_t::zero()) && (point1->y == g2_point_field_t::zero()) && (point1->z == g2_point_field_t::zero())) && 
+  !((point2->x == g2_point_field_t::zero()) && (point2->y == g2_point_field_t::zero()) && (point2->z == g2_point_field_t::zero()));
+}
+#endif

icicle-cuda/curves/ve_mod_mult.cu

@@ -1,16 +1,16 @@
-#ifndef _BN254_VEC_MULT
-#define _BN254_VEC_MULT
+#ifndef _VEC_MULT
+#define _VEC_MULT
 #include <stdio.h>
 #include <iostream>
-#include "../../primitives/field.cuh"
-#include "../../utils/storage.cuh"
-#include "../../primitives/projective.cuh"
+#include "../primitives/field.cuh"
+#include "../utils/storage.cuh"
+#include "../primitives/projective.cuh"
 #include "curve_config.cuh"
-#include "../../appUtils/vector_manipulation/ve_mod_mult.cuh"
+#include "../appUtils/vector_manipulation/ve_mod_mult.cuh"
 
 
-extern "C" int32_t vec_mod_mult_point_bn254(BN254::projective_t *inout,
-                                      BN254::scalar_t *scalar_vec,
+extern "C" int32_t vec_mod_mult_point(projective_t *inout,
+                                      scalar_t *scalar_vec,
                                       size_t n_elments,
                                       size_t device_id,
                                       cudaStream_t stream = 0)
@@ -20,7 +20,7 @@ extern "C" int32_t vec_mod_mult_point_bn254(BN254::projective_t *inout,
   try
   {
     // TODO: device_id
-    vector_mod_mult<BN254::projective_t, BN254::scalar_t>(scalar_vec, inout, inout, n_elments, stream);
+    vector_mod_mult<projective_t, scalar_t>(scalar_vec, inout, inout, n_elments, stream);
     return CUDA_SUCCESS;
   }
   catch (const std::runtime_error &ex)
@@ -30,8 +30,8 @@ extern "C" int32_t vec_mod_mult_point_bn254(BN254::projective_t *inout,
   }
 }
 
-extern "C" int32_t vec_mod_mult_scalar_bn254(BN254::scalar_t *inout,
-                                       BN254::scalar_t *scalar_vec,
+extern "C" int32_t vec_mod_mult_scalar(scalar_t *inout,
+                                       scalar_t *scalar_vec,
                                        size_t n_elments,
                                        size_t device_id,
                                        cudaStream_t stream = 0)
@@ -41,7 +41,7 @@ extern "C" int32_t vec_mod_mult_scalar_bn254(BN254::scalar_t *inout,
   try
   {
     // TODO: device_id
-    vector_mod_mult<BN254::scalar_t, BN254::scalar_t>(scalar_vec, inout, inout, n_elments, stream);
+    vector_mod_mult<scalar_t, scalar_t>(scalar_vec, inout, inout, n_elments, stream);
     return CUDA_SUCCESS;
   }
   catch (const std::runtime_error &ex)
@@ -51,9 +51,9 @@ extern "C" int32_t vec_mod_mult_scalar_bn254(BN254::scalar_t *inout,
   }
 }
 
-extern "C" int32_t matrix_vec_mod_mult_bn254(BN254::scalar_t *matrix_flattened,
-                                       BN254::scalar_t *input,
-                                       BN254::scalar_t *output,
+extern "C" int32_t matrix_vec_mod_mult(scalar_t *matrix_flattened,
+                                       scalar_t *input,
+                                       scalar_t *output,
                                        size_t n_elments,
                                        size_t device_id,
                                        cudaStream_t stream = 0)
@@ -63,7 +63,7 @@ extern "C" int32_t matrix_vec_mod_mult_bn254(BN254::scalar_t *matrix_flattened,
   try
   {
     // TODO: device_id
-    matrix_mod_mult<BN254::scalar_t>(matrix_flattened, input, output, n_elments, stream);
+    matrix_mod_mult<scalar_t>(matrix_flattened, input, output, n_elments, stream);
     return CUDA_SUCCESS;
   }
   catch (const std::runtime_error &ex)

icicle-cuda/primitives/extension_field.cuh

@@ -48,11 +48,11 @@ template <typename CONFIG> class ExtensionField {
     }
 
     static constexpr HOST_DEVICE_INLINE ExtensionField generator_x() {
-      return ExtensionField { FF { CONFIG::g2_gen_x_re }, FF { CONFIG::g2_gen_x_im } };
+      return ExtensionField { FF { CONFIG::generator_x_re }, FF { CONFIG::generator_x_im } };
     }
 
     static constexpr HOST_DEVICE_INLINE ExtensionField generator_y() {
-      return ExtensionField { FF { CONFIG::g2_gen_y_re }, FF { CONFIG::g2_gen_y_im } };
+      return ExtensionField { FF { CONFIG::generator_y_re }, FF { CONFIG::generator_y_im } };
     }
 
 

icicle-cuda/primitives/field.cuh

@@ -16,7 +16,7 @@
 template <class CONFIG> class Field {
   public:
     static constexpr unsigned TLC = CONFIG::limbs_count;
-    static constexpr unsigned NBITS = CONFIG::modulus_bit_count;
+    static constexpr unsigned NBITS = CONFIG::modulus_bits_count;
 
     static constexpr HOST_DEVICE_INLINE Field zero() {
       return Field { CONFIG::zero };
@@ -36,50 +36,231 @@ template <class CONFIG> class Field {
     }
 
     static constexpr HOST_DEVICE_INLINE Field generator_x() {
-      return Field { CONFIG::g1_gen_x };
+      return Field { CONFIG::generator_x };
     }
 
     static constexpr HOST_DEVICE_INLINE Field generator_y() {
-      return Field { CONFIG::g1_gen_y };
+      return Field { CONFIG::generator_y };
     }
 
-    static HOST_INLINE Field omega(uint32_t logn) {
-        if (logn == 0) {
-            return Field { CONFIG::one };
-        }
-
-        if (logn > CONFIG::omegas_count) {
-            throw std::invalid_argument( "Field: Invalid omega index" );
-        }
-
-        storage_array<CONFIG::omegas_count, TLC> const omega = CONFIG::omega;
-        return Field { omega.storages[logn-1] };
-    }      
-
-    static HOST_INLINE Field omega_inv(uint32_t logn) {
-        if (logn == 0) {
-            return Field { CONFIG::one };
-        }
-
-        if (logn > CONFIG::omegas_count) {
-            throw std::invalid_argument( "Field: Invalid omega_inv index" );
-        }
-
-        storage_array<CONFIG::omegas_count, TLC> const omega_inv = CONFIG::omega_inv;
-        return Field { omega_inv.storages[logn-1] };
+    static constexpr HOST_INLINE Field omega(uint32_t log_size) {
+      // Quick fix to linking issue, permanent fix will follow
+      switch (log_size) {
+        case 0:
+          return Field { CONFIG::one };
+        case 1:
+          return Field { CONFIG::omega1 };
+        case 2:
+          return Field { CONFIG::omega2 };
+        case 3:
+          return Field { CONFIG::omega3 };
+        case 4:
+          return Field { CONFIG::omega4 };
+        case 5:
+          return Field { CONFIG::omega5 };
+        case 6:
+          return Field { CONFIG::omega6 };
+        case 7:
+          return Field { CONFIG::omega7 };
+        case 8:
+          return Field { CONFIG::omega8 };
+        case 9:
+          return Field { CONFIG::omega9 };
+        case 10:
+          return Field { CONFIG::omega10 };
+        case 11:
+          return Field { CONFIG::omega11 };
+        case 12:
+          return Field { CONFIG::omega12 };
+        case 13:
+          return Field { CONFIG::omega13 };
+        case 14:
+          return Field { CONFIG::omega14 };
+        case 15:
+          return Field { CONFIG::omega15 };
+        case 16:
+          return Field { CONFIG::omega16 };
+        case 17:
+          return Field { CONFIG::omega17 };
+        case 18:
+          return Field { CONFIG::omega18 };
+        case 19:
+          return Field { CONFIG::omega19 };
+        case 20:
+          return Field { CONFIG::omega20 };
+        case 21:
+          return Field { CONFIG::omega21 };
+        case 22:
+          return Field { CONFIG::omega22 };
+        case 23:
+          return Field { CONFIG::omega23 };
+        case 24:
+          return Field { CONFIG::omega24 };
+        case 25:
+          return Field { CONFIG::omega25 };
+        case 26:
+          return Field { CONFIG::omega26 };
+        case 27:
+          return Field { CONFIG::omega27 };
+        case 28:
+          return Field { CONFIG::omega28 };
+        case 29:
+          return Field { CONFIG::omega29 };
+        case 30:
+          return Field { CONFIG::omega30 };
+        case 31:
+          return Field { CONFIG::omega31 };
+        case 32:
+          return Field { CONFIG::omega32 };        
+      }
+      return Field { CONFIG::one };
+      // return Field { CONFIG::omega[log_size-1] };
     }
 
-    static HOST_INLINE Field inv_log_size(uint32_t logn) {
-        if (logn == 0) {
-            return Field { CONFIG::one };
-        }
+    static constexpr HOST_INLINE Field omega_inv(uint32_t log_size) {
+      // Quick fix to linking issue, permanent fix will follow
+      switch (log_size) {
+        case 0:
+          return Field { CONFIG::one };
+        case 1:
+          return Field { CONFIG::omega_inv1 };
+        case 2:
+          return Field { CONFIG::omega_inv2 };
+        case 3:
+          return Field { CONFIG::omega_inv3 };
+        case 4:
+          return Field { CONFIG::omega_inv4 };
+        case 5:
+          return Field { CONFIG::omega_inv5 };
+        case 6:
+          return Field { CONFIG::omega_inv6 };
+        case 7:
+          return Field { CONFIG::omega_inv7 };
+        case 8:
+          return Field { CONFIG::omega_inv8 };
+        case 9:
+          return Field { CONFIG::omega_inv9 };
+        case 10:
+          return Field { CONFIG::omega_inv10 };
+        case 11:
+          return Field { CONFIG::omega_inv11 };
+        case 12:
+          return Field { CONFIG::omega_inv12 };
+        case 13:
+          return Field { CONFIG::omega_inv13 };
+        case 14:
+          return Field { CONFIG::omega_inv14 };
+        case 15:
+          return Field { CONFIG::omega_inv15 };
+        case 16:
+          return Field { CONFIG::omega_inv16 };
+        case 17:
+          return Field { CONFIG::omega_inv17 };
+        case 18:
+          return Field { CONFIG::omega_inv18 };
+        case 19:
+          return Field { CONFIG::omega_inv19 };
+        case 20:
+          return Field { CONFIG::omega_inv20 };
+        case 21:
+          return Field { CONFIG::omega_inv21 };
+        case 22:
+          return Field { CONFIG::omega_inv22 };
+        case 23:
+          return Field { CONFIG::omega_inv23 };
+        case 24:
+          return Field { CONFIG::omega_inv24 };
+        case 25:
+          return Field { CONFIG::omega_inv25 };
+        case 26:
+          return Field { CONFIG::omega_inv26 };
+        case 27:
+          return Field { CONFIG::omega_inv27 };
+        case 28:
+          return Field { CONFIG::omega_inv28 };
+        case 29:
+          return Field { CONFIG::omega_inv29 };
+        case 30:
+          return Field { CONFIG::omega_inv30 };
+        case 31:
+          return Field { CONFIG::omega_inv31 };
+        case 32:
+          return Field { CONFIG::omega_inv32 };        
+      }
+      return Field { CONFIG::one };
+      // return Field { CONFIG::omega_inv[log_size-1] };
+    }
 
-        if (logn > CONFIG::omegas_count) {
-            throw std::invalid_argument( "Field: Invalid inv index" );
-        }
-
-        storage_array<CONFIG::omegas_count, TLC> const inv = CONFIG::inv;
-        return Field { inv.storages[logn-1] };
+    static constexpr HOST_INLINE Field inv_log_size(uint32_t log_size) {
+      // Quick fix to linking issue, permanent fix will follow
+      switch (log_size) {
+        case 1:
+          return Field { CONFIG::inv1 };
+        case 2:
+          return Field { CONFIG::inv2 };
+        case 3:
+          return Field { CONFIG::inv3 };
+        case 4:
+          return Field { CONFIG::inv4 };
+        case 5:
+          return Field { CONFIG::inv5 };
+        case 6:
+          return Field { CONFIG::inv6 };
+        case 7:
+          return Field { CONFIG::inv7 };
+        case 8:
+          return Field { CONFIG::inv8 };
+        case 9:
+          return Field { CONFIG::inv9 };
+        case 10:
+          return Field { CONFIG::inv10 };
+        case 11:
+          return Field { CONFIG::inv11 };
+        case 12:
+          return Field { CONFIG::inv12 };
+        case 13:
+          return Field { CONFIG::inv13 };
+        case 14:
+          return Field { CONFIG::inv14 };
+        case 15:
+          return Field { CONFIG::inv15 };
+        case 16:
+          return Field { CONFIG::inv16 };
+        case 17:
+          return Field { CONFIG::inv17 };
+        case 18:
+          return Field { CONFIG::inv18 };
+        case 19:
+          return Field { CONFIG::inv19 };
+        case 20:
+          return Field { CONFIG::inv20 };
+        case 21:
+          return Field { CONFIG::inv21 };
+        case 22:
+          return Field { CONFIG::inv22 };
+        case 23:
+          return Field { CONFIG::inv23 };
+        case 24:
+          return Field { CONFIG::inv24 };
+        case 25:
+          return Field { CONFIG::inv25 };
+        case 26:
+          return Field { CONFIG::inv26 };
+        case 27:
+          return Field { CONFIG::inv27 };
+        case 28:
+          return Field { CONFIG::inv28 };
+        case 29:
+          return Field { CONFIG::inv29 };
+        case 30:
+          return Field { CONFIG::inv30 };
+        case 31:
+          return Field { CONFIG::inv31 };
+        case 32:
+          return Field { CONFIG::inv32 };        
+      }
+      return Field { CONFIG::one };
+      // return Field { CONFIG::inv[log_size-1] };
     }
 
     static constexpr HOST_DEVICE_INLINE Field modulus() {

icicle-cuda/utils/storage.cuh

@@ -7,7 +7,3 @@ template <unsigned LIMBS_COUNT> struct __align__(LIMBS_ALIGNMENT(LIMBS_COUNT)) s
   static constexpr unsigned LC = LIMBS_COUNT;
   uint32_t limbs[LIMBS_COUNT];
 };
-
-template <unsigned OMEGAS_COUNT, unsigned LIMBS_COUNT> struct __align__(LIMBS_ALIGNMENT(LIMBS_COUNT)) storage_array {
-    storage<LIMBS_COUNT> storages[OMEGAS_COUNT];
-};

icicle/appUtils/poseidon/poseidon.cuh

@@ -1,163 +0,0 @@
-#pragma once
-#include "constants.cuh"
-
-#if !defined(__CUDA_ARCH__) && defined(DEBUG)
-#include <iostream>
-#include <iomanip>
-#include <string>
-#include <sstream>
-#include <chrono>
-
-template <typename S>
-__host__ void print_buffer_from_cuda(S * device_ptr, size_t size, size_t t) {
-  S * buffer = static_cast< S * >(malloc(size * sizeof(S)));
-  cudaMemcpy(buffer, device_ptr, size * sizeof(S), cudaMemcpyDeviceToHost);
-
-  std::cout << "Start print" << std::endl;
-  for(int i = 0; i < size / t; i++) {
-    std::cout << "State #" << i << std::endl;
-    for (int j = 0; j < t; j++) {
-      std::cout << buffer[i * t + j] << std::endl;
-    }
-    std::cout << std::endl;
-  }
-  std::cout << std::endl;
-  free(buffer);
-}
-#endif
-
-#ifdef DEBUG
-template <typename S>
-__device__ void print_scalar(S element, int data) {
-    printf("D# %d, T# %d: 0x%08x%08x%08x%08x%08x%08x%08x%08x\n",
-           data,
-           threadIdx.x,
-           element.limbs_storage.limbs[0],
-           element.limbs_storage.limbs[1],
-           element.limbs_storage.limbs[2],
-           element.limbs_storage.limbs[3],
-           element.limbs_storage.limbs[4],
-           element.limbs_storage.limbs[5],
-           element.limbs_storage.limbs[6],
-           element.limbs_storage.limbs[7]
-    );
-}
-#endif
-
-template <typename S>
-struct PoseidonConfiguration {
-    uint32_t partial_rounds, full_rounds_half, t;
-    S * round_constants, * mds_matrix, * non_sparse_matrix, *sparse_matrices;
-};
-
-template <typename S>
-class Poseidon {
-  public:
-    uint32_t t;
-    PoseidonConfiguration<S> config;
-
-    enum HashType {
-        ConstInputLen,
-        MerkleTree,
-    };
-
-    Poseidon(const uint32_t arity, cudaStream_t stream) {
-        t = arity + 1;
-        this->config.t = t;
-        this->stream = stream;
-
-        // Pre-calculate domain tags
-        // Domain tags will vary for different applications of Poseidon
-        uint32_t tree_domain_tag_value = 1;
-        tree_domain_tag_value = (tree_domain_tag_value << arity) - tree_domain_tag_value;
-        tree_domain_tag = S::from(tree_domain_tag_value);
-
-        const_input_no_pad_domain_tag = S::one();
-
-        // TO-DO: implement binary shifts for scalar type
-        // const_input_no_pad_domain_tag = S::one() << 64;
-        // const_input_no_pad_domain_tag *= S::from(arity);
-
-        this->config.full_rounds_half = FULL_ROUNDS_DEFAULT;
-        this->config.partial_rounds = partial_rounds_number_from_arity(arity);
-
-        uint32_t round_constants_len = t * this->config.full_rounds_half * 2 + this->config.partial_rounds;
-        uint32_t mds_matrix_len = t * t;
-        uint32_t sparse_matrices_len = (t * 2 - 1) * this->config.partial_rounds;
-
-        // All the constants are stored in a single file
-        S * constants = load_constants<S>(arity);
-
-        S * mds_offset = constants + round_constants_len;
-        S * non_sparse_offset = mds_offset + mds_matrix_len;
-        S * sparse_matrices_offset = non_sparse_offset + mds_matrix_len;
-
-        #if !defined(__CUDA_ARCH__) && defined(DEBUG)
-        std::cout << "P: " << this->config.partial_rounds << " F: " << this->config.full_rounds_half << std::endl;
-        #endif
-
-        // Create streams for copying constants
-        cudaStream_t stream_copy_round_constants, stream_copy_mds_matrix, stream_copy_non_sparse, stream_copy_sparse_matrices;
-        cudaStreamCreate(&stream_copy_round_constants);
-        cudaStreamCreate(&stream_copy_mds_matrix);
-        cudaStreamCreate(&stream_copy_non_sparse);
-        cudaStreamCreate(&stream_copy_sparse_matrices);
-        
-        // Create events for copying constants
-        cudaEvent_t event_copied_round_constants, event_copy_mds_matrix, event_copy_non_sparse, event_copy_sparse_matrices;
-        cudaEventCreateWithFlags(&event_copied_round_constants, cudaEventDisableTiming);
-        cudaEventCreateWithFlags(&event_copy_mds_matrix, cudaEventDisableTiming);
-        cudaEventCreateWithFlags(&event_copy_non_sparse, cudaEventDisableTiming);
-        cudaEventCreateWithFlags(&event_copy_sparse_matrices, cudaEventDisableTiming);
-
-        // Malloc memory for copying constants
-        cudaMallocAsync(&this->config.round_constants, sizeof(S) * round_constants_len, stream_copy_round_constants);
-        cudaMallocAsync(&this->config.mds_matrix, sizeof(S) * mds_matrix_len, stream_copy_mds_matrix);
-        cudaMallocAsync(&this->config.non_sparse_matrix, sizeof(S) * mds_matrix_len, stream_copy_non_sparse);
-        cudaMallocAsync(&this->config.sparse_matrices, sizeof(S) * sparse_matrices_len, stream_copy_sparse_matrices);
-
-        // Copy constants
-        cudaMemcpyAsync(this->config.round_constants, constants,
-            sizeof(S) * round_constants_len,
-            cudaMemcpyHostToDevice, stream_copy_round_constants
-        );
-        cudaMemcpyAsync(this->config.mds_matrix, mds_offset,
-            sizeof(S) * mds_matrix_len,
-            cudaMemcpyHostToDevice, stream_copy_mds_matrix
-        );
-        cudaMemcpyAsync(this->config.non_sparse_matrix, non_sparse_offset,
-            sizeof(S) * mds_matrix_len,
-            cudaMemcpyHostToDevice, stream_copy_non_sparse
-        );
-        cudaMemcpyAsync(this->config.sparse_matrices, sparse_matrices_offset,
-            sizeof(S) * sparse_matrices_len,
-            cudaMemcpyHostToDevice, stream_copy_sparse_matrices
-        );
-
-        // Record finished copying event for streams
-        cudaEventRecord(event_copied_round_constants, stream_copy_round_constants);
-        cudaEventRecord(event_copy_mds_matrix, stream_copy_mds_matrix);
-        cudaEventRecord(event_copy_non_sparse, stream_copy_non_sparse);
-        cudaEventRecord(event_copy_sparse_matrices, stream_copy_sparse_matrices);
-
-        // Main stream waits for copying to finish
-        cudaStreamWaitEvent(stream, event_copied_round_constants);
-        cudaStreamWaitEvent(stream, event_copy_mds_matrix);
-        cudaStreamWaitEvent(stream, event_copy_non_sparse);
-        cudaStreamWaitEvent(stream, event_copy_sparse_matrices);
-    }
-
-    ~Poseidon() {
-        cudaFreeAsync(this->config.round_constants, this->stream);
-        cudaFreeAsync(this->config.mds_matrix, this->stream);
-        cudaFreeAsync(this->config.non_sparse_matrix, this->stream);
-        cudaFreeAsync(this->config.sparse_matrices, this->stream);
-    }
-
-    // Hash multiple preimages in parallel
-    void hash_blocks(const S * inp, size_t blocks, S * out, HashType hash_type, cudaStream_t stream);
-
-  private:
-    S tree_domain_tag, const_input_no_pad_domain_tag;
-    cudaStream_t stream;
-};

icicle/curves/bls12_377/curve_config.cuh

@@ -1,22 +0,0 @@
-#pragma once
-
-#include "../../primitives/field.cuh"
-#include "../../primitives/projective.cuh"
-
-#include "params.cuh"
-
-namespace BLS12_377 {
-    typedef Field<PARAMS_BLS12_377::fp_config> scalar_field_t;
-    typedef scalar_field_t scalar_t;
-    typedef Field<PARAMS_BLS12_377::fq_config> point_field_t;
-    static constexpr point_field_t b = point_field_t{ PARAMS_BLS12_377::weierstrass_b };
-    typedef Projective<point_field_t, scalar_field_t, b> projective_t;
-    typedef Affine<point_field_t> affine_t;
-    #if defined(G2_DEFINED)
-    typedef ExtensionField<PARAMS_BLS12_377::fq_config> g2_point_field_t;
-    static constexpr g2_point_field_t b_g2 = g2_point_field_t{ point_field_t{ PARAMS_BLS12_377::weierstrass_b_g2_re },
-                                                               point_field_t{ PARAMS_BLS12_377::weierstrass_b_g2_im }};
-    typedef Projective<g2_point_field_t, scalar_field_t, b_g2> g2_projective_t;
-    typedef Affine<g2_point_field_t> g2_affine_t;
-    #endif
-}

icicle/curves/bls12_377/lde.cu

@@ -1,327 +0,0 @@
-#ifndef _BLS12_377_LDE
-#define _BLS12_377_LDE
-#include <cuda.h>
-#include "../../appUtils/ntt/lde.cu"
-#include "../../appUtils/ntt/ntt.cuh"
-#include "../../appUtils/vector_manipulation/ve_mod_mult.cuh"
-#include "curve_config.cuh"
-
-extern "C" BLS12_377::scalar_t* build_domain_cuda_bls12_377(uint32_t domain_size, uint32_t logn, bool inverse, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        cudaStreamCreate(&stream);
-        if (inverse) {
-            return fill_twiddle_factors_array(domain_size, BLS12_377::scalar_t::omega_inv(logn), stream);
-        } else {
-            return fill_twiddle_factors_array(domain_size, BLS12_377::scalar_t::omega(logn), stream);
-        }
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return nullptr;
-    }
-}
-
-extern "C" int ntt_cuda_bls12_377(BLS12_377::scalar_t *arr, uint32_t n, bool inverse, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        cudaStreamCreate(&stream);
-        return ntt_end2end_template<BLS12_377::scalar_t,BLS12_377::scalar_t>(arr, n, inverse, stream); // TODO: pass device_id
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        
-        return -1;        
-    }
-}
-
-extern "C" int ecntt_cuda_bls12_377(BLS12_377::projective_t *arr, uint32_t n, bool inverse, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        cudaStreamCreate(&stream);
-        return ntt_end2end_template<BLS12_377::projective_t,BLS12_377::scalar_t>(arr, n, inverse, stream); // TODO: pass device_id
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int ntt_batch_cuda_bls12_377(BLS12_377::scalar_t *arr, uint32_t arr_size, uint32_t batch_size, bool inverse, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        cudaStreamCreate(&stream);
-        return ntt_end2end_batch_template<BLS12_377::scalar_t,BLS12_377::scalar_t>(arr, arr_size, batch_size, inverse, stream); // TODO: pass device_id
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int ecntt_batch_cuda_bls12_377(BLS12_377::projective_t *arr, uint32_t arr_size, uint32_t batch_size, bool inverse, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        cudaStreamCreate(&stream);
-        return ntt_end2end_batch_template<BLS12_377::projective_t,BLS12_377::scalar_t>(arr, arr_size, batch_size, inverse, stream); // TODO: pass device_id
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int interpolate_scalars_cuda_bls12_377(BLS12_377::scalar_t* d_out, BLS12_377::scalar_t *d_evaluations, BLS12_377::scalar_t *d_domain, unsigned n, unsigned device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        return interpolate(d_out, d_evaluations, d_domain, n, stream);
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int interpolate_scalars_batch_cuda_bls12_377(BLS12_377::scalar_t* d_out, BLS12_377::scalar_t* d_evaluations, BLS12_377::scalar_t* d_domain, unsigned n,
-                                              unsigned batch_size, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        cudaStreamCreate(&stream);
-        return interpolate_batch(d_out, d_evaluations, d_domain, n, batch_size, stream);
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int interpolate_points_cuda_bls12_377(BLS12_377::projective_t* d_out, BLS12_377::projective_t *d_evaluations, BLS12_377::scalar_t *d_domain, unsigned n, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        return interpolate(d_out, d_evaluations, d_domain, n, stream);
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int interpolate_points_batch_cuda_bls12_377(BLS12_377::projective_t* d_out, BLS12_377::projective_t* d_evaluations, BLS12_377::scalar_t* d_domain,
-                                             unsigned n, unsigned batch_size, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        cudaStreamCreate(&stream);
-        return interpolate_batch(d_out, d_evaluations, d_domain, n, batch_size, stream);
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int evaluate_scalars_cuda_bls12_377(BLS12_377::scalar_t* d_out, BLS12_377::scalar_t *d_coefficients, BLS12_377::scalar_t *d_domain, 
-                                     unsigned domain_size, unsigned n, unsigned device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        BLS12_377::scalar_t* _null = nullptr;
-        cudaStreamCreate(&stream);
-        return evaluate(d_out, d_coefficients, d_domain, domain_size, n, false, _null, stream);
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int evaluate_scalars_batch_cuda_bls12_377(BLS12_377::scalar_t* d_out, BLS12_377::scalar_t* d_coefficients, BLS12_377::scalar_t* d_domain, unsigned domain_size,
-                                           unsigned n, unsigned batch_size, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        BLS12_377::scalar_t* _null = nullptr;
-        cudaStreamCreate(&stream);
-        return evaluate_batch(d_out, d_coefficients, d_domain, domain_size, n, batch_size, false, _null, stream);
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int evaluate_points_cuda_bls12_377(BLS12_377::projective_t* d_out, BLS12_377::projective_t *d_coefficients, BLS12_377::scalar_t *d_domain, 
-                                    unsigned domain_size, unsigned n, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        BLS12_377::scalar_t* _null = nullptr;
-        cudaStreamCreate(&stream);
-        return evaluate(d_out, d_coefficients, d_domain, domain_size, n, false, _null, stream);
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int evaluate_points_batch_cuda_bls12_377(BLS12_377::projective_t* d_out, BLS12_377::projective_t* d_coefficients, BLS12_377::scalar_t* d_domain, unsigned domain_size,
-                                          unsigned n, unsigned batch_size, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        BLS12_377::scalar_t* _null = nullptr;
-        cudaStreamCreate(&stream);
-        return evaluate_batch(d_out, d_coefficients, d_domain, domain_size, n, batch_size, false, _null, stream);
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int evaluate_scalars_on_coset_cuda_bls12_377(BLS12_377::scalar_t* d_out, BLS12_377::scalar_t *d_coefficients, BLS12_377::scalar_t *d_domain, unsigned domain_size,
-                                              unsigned n, BLS12_377::scalar_t *coset_powers, unsigned device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        cudaStreamCreate(&stream);
-        return evaluate(d_out, d_coefficients, d_domain, domain_size, n, true, coset_powers, stream);
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int evaluate_scalars_on_coset_batch_cuda_bls12_377(BLS12_377::scalar_t* d_out, BLS12_377::scalar_t* d_coefficients, BLS12_377::scalar_t* d_domain, unsigned domain_size, 
-                                                    unsigned n, unsigned batch_size, BLS12_377::scalar_t *coset_powers, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        cudaStreamCreate(&stream);
-        return evaluate_batch(d_out, d_coefficients, d_domain, domain_size, n, batch_size, true, coset_powers, stream);
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int evaluate_points_on_coset_cuda_bls12_377(BLS12_377::projective_t* d_out, BLS12_377::projective_t *d_coefficients, BLS12_377::scalar_t *d_domain, unsigned domain_size,
-                                             unsigned n, BLS12_377::scalar_t *coset_powers, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        cudaStreamCreate(&stream);
-        return evaluate(d_out, d_coefficients, d_domain, domain_size, n, true, coset_powers, stream);
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int evaluate_points_on_coset_batch_cuda_bls12_377(BLS12_377::projective_t* d_out, BLS12_377::projective_t* d_coefficients, BLS12_377::scalar_t* d_domain, unsigned domain_size, 
-                                                   unsigned n, unsigned batch_size, BLS12_377::scalar_t *coset_powers, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        cudaStreamCreate(&stream);
-        return evaluate_batch(d_out, d_coefficients, d_domain, domain_size, n, batch_size, true, coset_powers, stream);
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int reverse_order_scalars_cuda_bls12_377(BLS12_377::scalar_t* arr, int n, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        uint32_t logn = uint32_t(log(n) / log(2));
-        cudaStreamCreate(&stream);
-        reverse_order(arr, n, logn, stream);
-        return 0;
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int reverse_order_scalars_batch_cuda_bls12_377(BLS12_377::scalar_t* arr, int n, int batch_size, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        uint32_t logn = uint32_t(log(n) / log(2));
-        cudaStreamCreate(&stream);
-        reverse_order_batch(arr, n, logn, batch_size, stream);
-        return 0;
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int reverse_order_points_cuda_bls12_377(BLS12_377::projective_t* arr, int n, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        uint32_t logn = uint32_t(log(n) / log(2));
-        cudaStreamCreate(&stream);
-        reverse_order(arr, n, logn, stream);
-        return 0;
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int reverse_order_points_batch_cuda_bls12_377(BLS12_377::projective_t* arr, int n, int batch_size, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        uint32_t logn = uint32_t(log(n) / log(2));
-        cudaStreamCreate(&stream);
-        reverse_order_batch(arr, n, logn, batch_size, stream);
-        return 0;
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-#endif

icicle/curves/bls12_377/msm.cu

@@ -1,93 +0,0 @@
-#ifndef _BLS12_377_MSM
-#define _BLS12_377_MSM
-#include "../../appUtils/msm/msm.cu"
-#include <stdexcept>
-#include <cuda.h>
-#include "curve_config.cuh"
-
-
-extern "C"
-int msm_cuda_bls12_377(BLS12_377::projective_t *out, BLS12_377::affine_t points[],
-              BLS12_377::scalar_t scalars[], size_t count, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        large_msm<BLS12_377::scalar_t, BLS12_377::projective_t, BLS12_377::affine_t>(scalars, points, count, out, false, false, stream);
-        return CUDA_SUCCESS;
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int msm_batch_cuda_bls12_377(BLS12_377::projective_t* out, BLS12_377::affine_t points[],
-                              BLS12_377::scalar_t scalars[], size_t batch_size, size_t msm_size, size_t device_id = 0, cudaStream_t stream = 0)
-{
-  try
-  {
-    cudaStreamCreate(&stream);
-    batched_large_msm<BLS12_377::scalar_t, BLS12_377::projective_t, BLS12_377::affine_t>(scalars, points, batch_size, msm_size, out, false, stream);
-    cudaStreamSynchronize(stream);
-
-    return CUDA_SUCCESS;
-  }
-  catch (const std::runtime_error &ex)
-  {
-    printf("error %s", ex.what());
-    return -1;
-  }
-}
-
-/**
- * Commit to a polynomial using the MSM.
- * Note: this function just calls the MSM, it doesn't convert between evaluation and coefficient form of scalars or points.
- * @param d_out Ouptut point to write the result to.
- * @param d_scalars Scalars for the MSM. Must be on device.
- * @param d_points Points for the MSM. Must be on device.
- * @param count Length of `d_scalars` and `d_points` arrays (they should have equal length).
- */
- extern "C"
- int commit_cuda_bls12_377(BLS12_377::projective_t* d_out, BLS12_377::scalar_t* d_scalars, BLS12_377::affine_t* d_points, size_t count, size_t device_id = 0, cudaStream_t stream = 0)
- {
-     try
-     {
-         large_msm<BLS12_377::scalar_t, BLS12_377::projective_t, BLS12_377::affine_t>(d_scalars, d_points, count, d_out, true, false, stream);
-         cudaStreamSynchronize(stream);
-         return 0;
-     }
-     catch (const std::runtime_error &ex)
-     {
-         printf("error %s", ex.what());
-         return -1;
-     }
- }
- 
- /**
-  * Commit to a batch of polynomials using the MSM.
-  * Note: this function just calls the MSM, it doesn't convert between evaluation and coefficient form of scalars or points.
-  * @param d_out Ouptut point to write the results to.
-  * @param d_scalars Scalars for the MSMs of all polynomials. Must be on device.
-  * @param d_points Points for the MSMs. Must be on device. It is assumed that this set of bases is used for each MSM.
-  * @param count Length of `d_points` array, `d_scalar` has length `count` * `batch_size`.
-  * @param batch_size Size of the batch.
-  */
- extern "C"
- int commit_batch_cuda_bls12_377(BLS12_377::projective_t* d_out, BLS12_377::scalar_t* d_scalars, BLS12_377::affine_t* d_points, size_t count, size_t batch_size, size_t device_id = 0, cudaStream_t stream = 0)
- {
-     try
-     {
-        cudaStreamCreate(&stream);
-         batched_large_msm(d_scalars, d_points, batch_size, count, d_out, true, stream);
-         cudaStreamSynchronize(stream);
-         return 0;
-     }
-     catch (const std::runtime_error &ex)
-     {
-         printf("error %s", ex.what());
-         return -1;
-     }
- }
-
-#endif

icicle/curves/bls12_377/projective.cu

@@ -1,22 +0,0 @@
-
-#include <cuda.h>
-
-#include "curve_config.cuh"
-
-#include "../../primitives/projective.cuh"
-
-extern "C" bool eq_bls12_377(BLS12_377::projective_t *point1, BLS12_377::projective_t *point2)
-{
-    return (*point1 == *point2) && 
-    !((point1->x == BLS12_377::point_field_t::zero()) && (point1->y == BLS12_377::point_field_t::zero()) && (point1->z == BLS12_377::point_field_t::zero())) && 
-    !((point2->x == BLS12_377::point_field_t::zero()) && (point2->y == BLS12_377::point_field_t::zero()) && (point2->z == BLS12_377::point_field_t::zero()));
-}
-
-#if defined(G2_DEFINED)
-extern "C" bool eq_g2_bls12_377(BLS12_377::g2_projective_t *point1, BLS12_377::g2_projective_t *point2)
-{
-  return (*point1 == *point2) && 
-  !((point1->x == BLS12_377::g2_point_field_t::zero()) && (point1->y == BLS12_377::g2_point_field_t::zero()) && (point1->z == BLS12_377::g2_point_field_t::zero())) && 
-  !((point2->x == BLS12_377::g2_point_field_t::zero()) && (point2->y == BLS12_377::g2_point_field_t::zero()) && (point2->z == BLS12_377::g2_point_field_t::zero()));
-}
-#endif

icicle/curves/bls12_377/ve_mod_mult.cu

@@ -1,69 +0,0 @@
-#ifndef _BLS12_377_VEC_MULT
-#define _BLS12_377_VEC_MULT
-#include <stdio.h>
-#include <iostream>
-#include "../../primitives/field.cuh"
-#include "../../utils/storage.cuh"
-#include "../../primitives/projective.cuh"
-#include "curve_config.cuh"
-#include "../../appUtils/vector_manipulation/ve_mod_mult.cuh"
-
-
-extern "C" int32_t vec_mod_mult_point_bls12_377(BLS12_377::projective_t *inout,
-                                      BLS12_377::scalar_t *scalar_vec,
-                                      size_t n_elments,
-                                      size_t device_id, 
-                                      cudaStream_t stream = 0)
-{
-  try
-  {
-    // TODO: device_id
-    vector_mod_mult<BLS12_377::projective_t, BLS12_377::scalar_t>(scalar_vec, inout, inout, n_elments, stream);
-    return CUDA_SUCCESS;
-  }
-  catch (const std::runtime_error &ex)
-  {
-    printf("error %s", ex.what()); // TODO: error code and message
-    return -1;
-  }
-}
-
-extern "C" int32_t vec_mod_mult_scalar_bls12_377(BLS12_377::scalar_t *inout,
-                                       BLS12_377::scalar_t *scalar_vec,
-                                       size_t n_elments,
-                                       size_t device_id,
-                                       cudaStream_t stream = 0)
-{
-  try
-  {
-    // TODO: device_id
-    vector_mod_mult<BLS12_377::scalar_t, BLS12_377::scalar_t>(scalar_vec, inout, inout, n_elments, stream);
-    return CUDA_SUCCESS;
-  }
-  catch (const std::runtime_error &ex)
-  {
-    printf("error %s", ex.what()); // TODO: error code and message
-    return -1;
-  }
-}
-
-extern "C" int32_t matrix_vec_mod_mult_bls12_377(BLS12_377::scalar_t *matrix_flattened,
-                                       BLS12_377::scalar_t *input,
-                                       BLS12_377::scalar_t *output,
-                                       size_t n_elments,
-                                       size_t device_id,
-                                       cudaStream_t stream = 0)
-{
-  try
-  {
-    // TODO: device_id
-    matrix_mod_mult<BLS12_377::scalar_t>(matrix_flattened, input, output, n_elments, stream);
-    return CUDA_SUCCESS;
-  }
-  catch (const std::runtime_error &ex)
-  {
-    printf("error %s", ex.what()); // TODO: error code and message
-    return -1;
-  }
-}
-#endif

icicle/curves/bls12_381/curve_config.cuh

@@ -1,22 +0,0 @@
-#pragma once
-
-#include "../../primitives/field.cuh"
-#include "../../primitives/projective.cuh"
-
-#include "params.cuh"
-
-namespace BLS12_381 {
-    typedef Field<PARAMS_BLS12_381::fp_config> scalar_field_t;
-    typedef scalar_field_t scalar_t;
-    typedef Field<PARAMS_BLS12_381::fq_config> point_field_t;
-    static constexpr point_field_t b = point_field_t{ PARAMS_BLS12_381::weierstrass_b };
-    typedef Projective<point_field_t, scalar_field_t, b> projective_t;
-    typedef Affine<point_field_t> affine_t;
-    #if defined(G2_DEFINED)
-    typedef ExtensionField<PARAMS_BLS12_381::fq_config> g2_point_field_t;
-    static constexpr g2_point_field_t b_g2 = g2_point_field_t{ point_field_t{ PARAMS_BLS12_381::weierstrass_b_g2_re },
-                                                               point_field_t{ PARAMS_BLS12_381::weierstrass_b_g2_im }};
-    typedef Projective<g2_point_field_t, scalar_field_t, b_g2> g2_projective_t;
-    typedef Affine<g2_point_field_t> g2_affine_t;
-    #endif
-}

icicle/curves/bls12_381/lde.cu

@@ -1,327 +0,0 @@
-#ifndef _BLS12_381_LDE
-#define _BLS12_381_LDE
-#include <cuda.h>
-#include "../../appUtils/ntt/lde.cu"
-#include "../../appUtils/ntt/ntt.cuh"
-#include "../../appUtils/vector_manipulation/ve_mod_mult.cuh"
-#include "curve_config.cuh"
-
-extern "C" BLS12_381::scalar_t* build_domain_cuda_bls12_381(uint32_t domain_size, uint32_t logn, bool inverse, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        cudaStreamCreate(&stream);
-        if (inverse) {
-            return fill_twiddle_factors_array(domain_size, BLS12_381::scalar_t::omega_inv(logn), stream);
-        } else {
-            return fill_twiddle_factors_array(domain_size, BLS12_381::scalar_t::omega(logn), stream);
-        }
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return nullptr;
-    }
-}
-
-extern "C" int ntt_cuda_bls12_381(BLS12_381::scalar_t *arr, uint32_t n, bool inverse, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        cudaStreamCreate(&stream);
-        return ntt_end2end_template<BLS12_381::scalar_t,BLS12_381::scalar_t>(arr, n, inverse, stream); // TODO: pass device_id
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        
-        return -1;        
-    }
-}
-
-extern "C" int ecntt_cuda_bls12_381(BLS12_381::projective_t *arr, uint32_t n, bool inverse, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        cudaStreamCreate(&stream);
-        return ntt_end2end_template<BLS12_381::projective_t,BLS12_381::scalar_t>(arr, n, inverse, stream); // TODO: pass device_id
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int ntt_batch_cuda_bls12_381(BLS12_381::scalar_t *arr, uint32_t arr_size, uint32_t batch_size, bool inverse, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        cudaStreamCreate(&stream);
-        return ntt_end2end_batch_template<BLS12_381::scalar_t,BLS12_381::scalar_t>(arr, arr_size, batch_size, inverse, stream); // TODO: pass device_id
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int ecntt_batch_cuda_bls12_381(BLS12_381::projective_t *arr, uint32_t arr_size, uint32_t batch_size, bool inverse, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        cudaStreamCreate(&stream);
-        return ntt_end2end_batch_template<BLS12_381::projective_t,BLS12_381::scalar_t>(arr, arr_size, batch_size, inverse, stream); // TODO: pass device_id
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int interpolate_scalars_cuda_bls12_381(BLS12_381::scalar_t* d_out, BLS12_381::scalar_t *d_evaluations, BLS12_381::scalar_t *d_domain, unsigned n, unsigned device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        return interpolate(d_out, d_evaluations, d_domain, n, stream);
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int interpolate_scalars_batch_cuda_bls12_381(BLS12_381::scalar_t* d_out, BLS12_381::scalar_t* d_evaluations, BLS12_381::scalar_t* d_domain, unsigned n,
-                                              unsigned batch_size, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        cudaStreamCreate(&stream);
-        return interpolate_batch(d_out, d_evaluations, d_domain, n, batch_size, stream);
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int interpolate_points_cuda_bls12_381(BLS12_381::projective_t* d_out, BLS12_381::projective_t *d_evaluations, BLS12_381::scalar_t *d_domain, unsigned n, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        return interpolate(d_out, d_evaluations, d_domain, n, stream);
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int interpolate_points_batch_cuda_bls12_381(BLS12_381::projective_t* d_out, BLS12_381::projective_t* d_evaluations, BLS12_381::scalar_t* d_domain,
-                                             unsigned n, unsigned batch_size, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        cudaStreamCreate(&stream);
-        return interpolate_batch(d_out, d_evaluations, d_domain, n, batch_size, stream);
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int evaluate_scalars_cuda_bls12_381(BLS12_381::scalar_t* d_out, BLS12_381::scalar_t *d_coefficients, BLS12_381::scalar_t *d_domain, 
-                                     unsigned domain_size, unsigned n, unsigned device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        BLS12_381::scalar_t* _null = nullptr;
-        cudaStreamCreate(&stream);
-        return evaluate(d_out, d_coefficients, d_domain, domain_size, n, false, _null, stream);
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int evaluate_scalars_batch_cuda_bls12_381(BLS12_381::scalar_t* d_out, BLS12_381::scalar_t* d_coefficients, BLS12_381::scalar_t* d_domain, unsigned domain_size,
-                                           unsigned n, unsigned batch_size, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        BLS12_381::scalar_t* _null = nullptr;
-        cudaStreamCreate(&stream);
-        return evaluate_batch(d_out, d_coefficients, d_domain, domain_size, n, batch_size, false, _null, stream);
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int evaluate_points_cuda_bls12_381(BLS12_381::projective_t* d_out, BLS12_381::projective_t *d_coefficients, BLS12_381::scalar_t *d_domain, 
-                                    unsigned domain_size, unsigned n, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        BLS12_381::scalar_t* _null = nullptr;
-        cudaStreamCreate(&stream);
-        return evaluate(d_out, d_coefficients, d_domain, domain_size, n, false, _null, stream);
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int evaluate_points_batch_cuda_bls12_381(BLS12_381::projective_t* d_out, BLS12_381::projective_t* d_coefficients, BLS12_381::scalar_t* d_domain, unsigned domain_size,
-                                          unsigned n, unsigned batch_size, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        BLS12_381::scalar_t* _null = nullptr;
-        cudaStreamCreate(&stream);
-        return evaluate_batch(d_out, d_coefficients, d_domain, domain_size, n, batch_size, false, _null, stream);
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int evaluate_scalars_on_coset_cuda_bls12_381(BLS12_381::scalar_t* d_out, BLS12_381::scalar_t *d_coefficients, BLS12_381::scalar_t *d_domain, unsigned domain_size,
-                                              unsigned n, BLS12_381::scalar_t *coset_powers, unsigned device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        cudaStreamCreate(&stream);
-        return evaluate(d_out, d_coefficients, d_domain, domain_size, n, true, coset_powers, stream);
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int evaluate_scalars_on_coset_batch_cuda_bls12_381(BLS12_381::scalar_t* d_out, BLS12_381::scalar_t* d_coefficients, BLS12_381::scalar_t* d_domain, unsigned domain_size, 
-                                                    unsigned n, unsigned batch_size, BLS12_381::scalar_t *coset_powers, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        cudaStreamCreate(&stream);
-        return evaluate_batch(d_out, d_coefficients, d_domain, domain_size, n, batch_size, true, coset_powers, stream);
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int evaluate_points_on_coset_cuda_bls12_381(BLS12_381::projective_t* d_out, BLS12_381::projective_t *d_coefficients, BLS12_381::scalar_t *d_domain, unsigned domain_size,
-                                             unsigned n, BLS12_381::scalar_t *coset_powers, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        cudaStreamCreate(&stream);
-        return evaluate(d_out, d_coefficients, d_domain, domain_size, n, true, coset_powers, stream);
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int evaluate_points_on_coset_batch_cuda_bls12_381(BLS12_381::projective_t* d_out, BLS12_381::projective_t* d_coefficients, BLS12_381::scalar_t* d_domain, unsigned domain_size, 
-                                                   unsigned n, unsigned batch_size, BLS12_381::scalar_t *coset_powers, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        cudaStreamCreate(&stream);
-        return evaluate_batch(d_out, d_coefficients, d_domain, domain_size, n, batch_size, true, coset_powers, stream);
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int reverse_order_scalars_cuda_bls12_381(BLS12_381::scalar_t* arr, int n, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        uint32_t logn = uint32_t(log(n) / log(2));
-        cudaStreamCreate(&stream);
-        reverse_order(arr, n, logn, stream);
-        return 0;
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int reverse_order_scalars_batch_cuda_bls12_381(BLS12_381::scalar_t* arr, int n, int batch_size, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        uint32_t logn = uint32_t(log(n) / log(2));
-        cudaStreamCreate(&stream);
-        reverse_order_batch(arr, n, logn, batch_size, stream);
-        return 0;
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int reverse_order_points_cuda_bls12_381(BLS12_381::projective_t* arr, int n, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        uint32_t logn = uint32_t(log(n) / log(2));
-        cudaStreamCreate(&stream);
-        reverse_order(arr, n, logn, stream);
-        return 0;
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int reverse_order_points_batch_cuda_bls12_381(BLS12_381::projective_t* arr, int n, int batch_size, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        uint32_t logn = uint32_t(log(n) / log(2));
-        cudaStreamCreate(&stream);
-        reverse_order_batch(arr, n, logn, batch_size, stream);
-        return 0;
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-#endif

icicle/curves/bls12_381/msm.cu

@@ -1,92 +0,0 @@
-#ifndef _BLS12_381_MSM
-#define _BLS12_381_MSM
-#include "../../appUtils/msm/msm.cu"
-#include <stdexcept>
-#include <cuda.h>
-#include "curve_config.cuh"
-
-
-extern "C"
-int msm_cuda_bls12_381(BLS12_381::projective_t *out, BLS12_381::affine_t points[],
-              BLS12_381::scalar_t scalars[], size_t count, size_t device_id = 0, cudaStream_t stream = 0)
-{
-    try
-    {
-        large_msm<BLS12_381::scalar_t, BLS12_381::projective_t, BLS12_381::affine_t>(scalars, points, count, out, false, false, stream);
-        return CUDA_SUCCESS;
-    }
-    catch (const std::runtime_error &ex)
-    {
-        printf("error %s", ex.what());
-        return -1;
-    }
-}
-
-extern "C" int msm_batch_cuda_bls12_381(BLS12_381::projective_t* out, BLS12_381::affine_t points[],
-                              BLS12_381::scalar_t scalars[], size_t batch_size, size_t msm_size, size_t device_id = 0, cudaStream_t stream = 0)
-{
-  try
-  {
-    cudaStreamCreate(&stream);
-    batched_large_msm<BLS12_381::scalar_t, BLS12_381::projective_t, BLS12_381::affine_t>(scalars, points, batch_size, msm_size, out, false, stream);
-    cudaStreamSynchronize(stream);
-    return CUDA_SUCCESS;
-  }
-  catch (const std::runtime_error &ex)
-  {
-    printf("error %s", ex.what());
-    return -1;
-  }
-}
-
-/**
- * Commit to a polynomial using the MSM.
- * Note: this function just calls the MSM, it doesn't convert between evaluation and coefficient form of scalars or points.
- * @param d_out Ouptut point to write the result to.
- * @param d_scalars Scalars for the MSM. Must be on device.
- * @param d_points Points for the MSM. Must be on device.
- * @param count Length of `d_scalars` and `d_points` arrays (they should have equal length).
- */
- extern "C"
- int commit_cuda_bls12_381(BLS12_381::projective_t* d_out, BLS12_381::scalar_t* d_scalars, BLS12_381::affine_t* d_points, size_t count, size_t device_id = 0, cudaStream_t stream = 0)
- {
-     try
-     {
-         large_msm(d_scalars, d_points, count, d_out, true, false, stream);
-         cudaStreamSynchronize(stream);
-         return 0;
-     }
-     catch (const std::runtime_error &ex)
-     {
-         printf("error %s", ex.what());
-         return -1;
-     }
- }
- 
- /**
-  * Commit to a batch of polynomials using the MSM.
-  * Note: this function just calls the MSM, it doesn't convert between evaluation and coefficient form of scalars or points.
-  * @param d_out Ouptut point to write the results to.
-  * @param d_scalars Scalars for the MSMs of all polynomials. Must be on device.
-  * @param d_points Points for the MSMs. Must be on device. It is assumed that this set of bases is used for each MSM.
-  * @param count Length of `d_points` array, `d_scalar` has length `count` * `batch_size`.
-  * @param batch_size Size of the batch.
-  */
- extern "C"
- int commit_batch_cuda_bls12_381(BLS12_381::projective_t* d_out, BLS12_381::scalar_t* d_scalars, BLS12_381::affine_t* d_points, size_t count, size_t batch_size, size_t device_id = 0, cudaStream_t stream = 0)
- {
-     try
-     {  
-        cudaStreamCreate(&stream);
-         batched_large_msm(d_scalars, d_points, batch_size, count, d_out, true, stream);
-         cudaStreamSynchronize(stream);
-         return 0;
-     }
-     catch (const std::runtime_error &ex)
-     {
-         printf("error %s", ex.what());
-         return -1;
-     }
- }
-
- #endif

icicle/curves/bls12_381/poseidon.cu

@@ -1,47 +0,0 @@
-#ifndef _BLS12_381_POSEIDON
-#define _BLS12_381_POSEIDON
-#include <cuda.h>
-#include <stdexcept>
-#include "../../appUtils/poseidon/poseidon.cu"
-#include "curve_config.cuh"
-
-template class Poseidon<BLS12_381::scalar_t>;
-
-extern "C" int poseidon_multi_cuda_bls12_381(BLS12_381::scalar_t input[], BLS12_381::scalar_t* out,
-                                             size_t number_of_blocks, int arity, size_t device_id = 0, cudaStream_t stream = 0)
-{
-  try
-  {
-    // TODO: once we get bindings to pass a stream, we should make {stream} a required parameter and use it instead of 
-    // creating a new stream
-    if (stream == 0) {
-        cudaStreamCreate(&stream);
-    }
-    
-    cudaEvent_t start_event, end_event;
-    cudaEventCreate(&start_event);
-    cudaEventCreate(&end_event);
-    cudaEventRecord(start_event, stream);
-    Poseidon<BLS12_381::scalar_t> poseidon(arity, stream);
-    poseidon.hash_blocks(input, number_of_blocks, out, Poseidon<BLS12_381::scalar_t>::HashType::MerkleTree, stream);
-    cudaEventRecord(end_event, stream);
-    cudaEventSynchronize(end_event);
-
-    #ifdef DEBUG
-    float elapsedTime;
-    cudaEventElapsedTime(&elapsedTime, start_event, end_event);
-    printf("Time elapsed: %f", elapsedTime);
-    #endif
-
-    cudaEventDestroy(start_event);
-    cudaEventDestroy(end_event);
-
-    return CUDA_SUCCESS;
-  }
-  catch (const std::runtime_error &ex)
-  {
-    printf("error %s", ex.what());
-    return -1;
-  }
-}
-#endif

icicle/curves/bls12_381/projective.cu

@@ -1,19 +0,0 @@
-#include <cuda.h>
-#include "curve_config.cuh"
-#include "../../primitives/projective.cuh"
-
-extern "C" bool eq_bls12_381(BLS12_381::projective_t *point1, BLS12_381::projective_t *point2)
-{
-    return (*point1 == *point2) && 
-    !((point1->x == BLS12_381::point_field_t::zero()) && (point1->y == BLS12_381::point_field_t::zero()) && (point1->z == BLS12_381::point_field_t::zero())) && 
-    !((point2->x == BLS12_381::point_field_t::zero()) && (point2->y == BLS12_381::point_field_t::zero()) && (point2->z == BLS12_381::point_field_t::zero()));
-}
-
-#if defined(G2_DEFINED)
-extern "C" bool eq_g2_bls12_381(BLS12_381::g2_projective_t *point1, BLS12_381::g2_projective_t *point2)
-{
-  return (*point1 == *point2) && 
-  !((point1->x == BLS12_381::g2_point_field_t::zero()) && (point1->y == BLS12_381::g2_point_field_t::zero()) && (point1->z == BLS12_381::g2_point_field_t::zero())) && 
-  !((point2->x == BLS12_381::g2_point_field_t::zero()) && (point2->y == BLS12_381::g2_point_field_t::zero()) && (point2->z == BLS12_381::g2_point_field_t::zero()));
-}
-#endif

icicle/curves/bls12_381/supported_operations.cu

@@ -1,5 +0,0 @@
-#include "projective.cu"
-#include "lde.cu"
-#include "msm.cu"
-#include "ve_mod_mult.cu"
-#include "poseidon.cu"