icicle icon indicating copy to clipboard operation
icicle copied to clipboard

a GPU Library for Zero-Knowledge Acceleration

Results 54 icicle issues
Sort by recently updated
recently updated
newest added

## Description Add Cuda graph support ## Motivation GPU utilization and performance improvement

type:feature
lang:cuda/cpp
backlog

Currently using pageable memory using cudaMalloc. Switching to pinned host memory using cudaMallocHost will perform better with higher bandwidth.

backlog

Curious what the rational for the window size of c = 10-bit scalars (26 windows) is? It feels like this choice can be optimized?

area:msm
backlog

The for loop in this kernel can be eliminated with the integration of **cooperative groups**. Instead of single thread looping over all the limbs for a single scalar, multiple threads...

area:msm
lang:cuda/cpp
backlog

## Description Currently we don't have the functionality to open committed polynomials at arbitrary points. ## Motivation In Danksharding, KZG openings are performed at specific carefully chosen points, which makes...

type:feature
area:msm
backlog

## Description Please provide a clear and concise description of the feature you would like included. ## Motivation Please provide a clear and concise description of the motivation for adding...

type:feature
lang:cuda/cpp
backlog

## Description The current [msm.cu](https://github.com/ingonyama-zk/icicle/blob/main/icicle/appUtils/msm/msm.cu) implementation launches a monolithic kernel in several places like this: ```cuda unsigned NUM_THREADS = 1

type:feature
area:msm
lang:cuda/cpp
backlog

## Description Please provide a clear and concise description of the feature you would like included. ## Motivation Please provide a clear and concise description of the motivation for adding...

lang:rust
backlog

1. https://github.com/ingonyama-zk/icicle/blob/ef8beb8d0c7d481b8f3f3a47d48695889e4079b3/icicle/appUtils/ntt/ntt.cuh#L323 2. https://github.com/ingonyama-zk/icicle/blob/ef8beb8d0c7d481b8f3f3a47d48695889e4079b3/icicle/appUtils/ntt/ntt.cuh#L199 3.https://github.com/ingonyama-zk/icicle/blob/ef8beb8d0c7d481b8f3f3a47d48695889e4079b3/icicle/appUtils/ntt/ntt.cuh#L338

area:ntt
lang:cuda/cpp
backlog

## Description Aleo created their version of Marlin called [Varuna](https://drive.google.com/file/d/1W9vsn5xT1vUmJbzO8VXoNS4W1wGWLDHN/view) which is (as far as I understand) what is ran when https://github.com/AleoHQ/snarkVM/blob/testnet3/synthesizer/src/vm/execute.rs#L26 is executed. Because the time outside witness generation...

type:feature
lang:rust
backlog