improve gpu perf and add CI for icicle

.github/workflows/rust.yml

@@ -40,6 +40,11 @@ jobs:
         with:
           command: clippy
           args: --all
+      - name: cargo clippy icicle
+        uses: actions-rs/cargo@v1
+        with:
+          command: clippy
+          args: --all --features icicle
 
   machete:
     runs-on: ubuntu-latest
@@ -77,6 +82,23 @@ jobs:
       - name: Run jolt-core tests
         run: cargo nextest run --release -p jolt-core
 
+  test-icicle:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v3
+      - uses: actions-rust-lang/setup-rust-toolchain@v1
+      - name: Cache Jolt RISC-V Rust toolchain
+        uses: actions/cache@v4
+        with:
+          key: jolt-rust-toolchain-${{hashFiles('guest-toolchain-tag')}}
+          path: ~/.jolt
+      - name: Install Jolt RISC-V Rust toolchain
+        run: cargo run install-toolchain
+      - name: Install nextest
+        uses: taiki-e/install-action@nextest
+      - name: Run jolt-core tests
+        run: cargo nextest run --release -p jolt-core --features icicle
+
   on-chain:
     name: Onchain Verifier Tests
     runs-on: ubuntu-latest

jolt-core/benches/msm_batch.rs

@@ -17,7 +17,7 @@ use rand_core::{RngCore, SeedableRng};
 #[cfg(feature = "icicle")]
 use rayon::prelude::*;
 
-const SRS_SIZE: usize = 1 << 12;
+const SRS_SIZE: usize = 1 << 14;
 
 // Sets up the benchmark
 fn setup_bench<PCS, F, ProofTranscript>(
@@ -143,8 +143,8 @@ struct BatchConfig {
 fn main() {
     let mut criterion = Criterion::default()
         .configure_from_args()
-        .sample_size(20)
-        .warm_up_time(std::time::Duration::from_secs(5));
+        .sample_size(10)
+        .warm_up_time(std::time::Duration::from_secs(10));
     benchmark_msm_batch::<Zeromorph<Bn254, KeccakTranscript>, Fr, KeccakTranscript>(
         &mut criterion,
         "VariableBaseMSM::msm_batch(bias: Large)",

jolt-core/src/jolt/vm/mod.rs

@@ -315,6 +315,10 @@ where
         max_memory_address: usize,
         max_trace_length: usize,
     ) -> JoltPreprocessing<C, F, PCS, ProofTranscript> {
+        //TODO(sagar): This should be moved to a more appropriate place - icicle makes a network request
+        // which impacts prover time.
+        icicle::icicle_init();
+
         let bytecode_commitment_shapes = BytecodeProof::<F, PCS, ProofTranscript>::commit_shapes(
             max_bytecode_size,
             max_trace_length,
@@ -401,7 +405,6 @@ where
         println!("Trace length: {}", trace_length);
 
         JoltTraceStep::pad(&mut trace);
-        icicle::icicle_init();
 
         let mut transcript = ProofTranscript::new(b"Jolt transcript");
         Self::fiat_shamir_preamble(