use icicle-jolt to compile cuda deps

README.md

@@ -68,6 +68,29 @@ Examples in the [`examples`](./examples/) directory can be run using e.g.
 
 ```cargo run --release -p sha2-chain```
 
+## CUDA Support
+
+JOLT supports CUDA acceleration via [icicle](https://github.com/ingonyama-zk/icicle-jolt).
+
+Dependencies:
+1. Install [CUDA Toolkit](https://developer.nvidia.com/cuda-toolkit)
+2. Install [CMake](https://cmake.org/)
+
+Now you may build Jolt with CUDA acceleration using the `--features icicle` flag.
+
+### Build
+
+```cargo build -p jolt-core --features icicle```
+
+### Bench
+
+```
+# Set the icicle backend path - this won't be needed in the future
+export ICICLE_BACKEND_INSTALL_DIR=$(pwd)/target/debug/deps/icicle/lib/backend
+cargo bench --bench msm_batch --no-fail-fast -p jolt-core --features icicle
+```
+
+Note - NVIDIA doesn't support cross compilation on MacOS. Only Windows or Linux.
 
 ## Performance profiling
 

jolt-core/Cargo.toml

@@ -119,9 +119,9 @@ name = "jolt_core"
 path = "src/lib.rs"
 
 [target.'cfg(not(target_arch = "wasm32"))'.dependencies]
-icicle-runtime = { git = "https://github.com/ingonyama-zk/icicle.git", tag = "v3.1.0", optional = true }
-icicle-core = { git = "https://github.com/ingonyama-zk/icicle.git", tag = "v3.1.0", optional = true }
-icicle-bn254 = { git = "https://github.com/ingonyama-zk/icicle.git", tag = "v3.1.0", optional = true }
+icicle-runtime = { git = "https://github.com/ingonyama-zk/icicle-jolt.git", features = ["cuda_backend"], rev = "ed93e21", optional = true }
+icicle-core = { git = "https://github.com/ingonyama-zk/icicle-jolt.git", rev = "ed93e21", optional = true }
+icicle-bn254 = { git = "https://github.com/ingonyama-zk/icicle-jolt.git", features = ["cuda_backend"], rev = "ed93e21", optional = true }
 memory-stats = "1.0.0"
 sys-info = "0.9.1"
 tokio = { version = "1.38.0", optional = true, features = ["rt-multi-thread"] }