feat: designing dedicated chips for memcpy

Cargo.toml

@@ -61,6 +61,9 @@ members = [
     "extensions/ecc/tests",
     "extensions/pairing/circuit",
     "extensions/pairing/guest",
+    "extensions/memcpy/circuit",
+    "extensions/memcpy/transpiler",
+    "extensions/memcpy/tests",
     "guest-libs/ff_derive/",
     "guest-libs/k256/",
     "guest-libs/p256/",
@@ -171,6 +174,8 @@ openvm-ecc-sw-macros = { path = "extensions/ecc/sw-macros", default-features = f
 openvm-pairing-circuit = { path = "extensions/pairing/circuit", default-features = false }
 openvm-pairing-transpiler = { path = "extensions/pairing/transpiler", default-features = false }
 openvm-pairing-guest = { path = "extensions/pairing/guest", default-features = false }
+openvm-memcpy-circuit = { path = "extensions/memcpy/circuit", default-features = false }
+openvm-memcpy-transpiler = { path = "extensions/memcpy/transpiler", default-features = false }
 openvm-verify-stark = { path = "guest-libs/verify_stark", default-features = false }
 
 # Benchmarking

benchmarks/execute/Cargo.toml

@@ -24,6 +24,8 @@ openvm-pairing-guest.workspace = true
 openvm-pairing-transpiler.workspace = true
 openvm-keccak256-circuit.workspace = true
 openvm-keccak256-transpiler.workspace = true
+openvm-memcpy-circuit.workspace = true
+openvm-memcpy-transpiler.workspace = true
 openvm-rv32im-circuit.workspace = true
 openvm-rv32im-transpiler.workspace = true
 openvm-sha256-circuit.workspace = true

benchmarks/execute/benches/execute.rs

@@ -32,6 +32,8 @@ use openvm_ecc_circuit::{EccCpuProverExt, WeierstrassExtension, WeierstrassExten
 use openvm_ecc_transpiler::EccTranspilerExtension;
 use openvm_keccak256_circuit::{Keccak256, Keccak256CpuProverExt, Keccak256Executor};
 use openvm_keccak256_transpiler::Keccak256TranspilerExtension;
+use openvm_memcpy_circuit::{Memcpy, MemcpyCpuProverExt, MemcpyExecutor};
+use openvm_memcpy_transpiler::MemcpyTranspilerExtension;
 use openvm_native_circuit::{NativeCpuBuilder, NATIVE_MAX_TRACE_HEIGHTS};
 use openvm_native_recursion::hints::Hintable;
 use openvm_pairing_circuit::{
@@ -107,6 +109,8 @@ pub struct ExecuteConfig {
     #[extension]
     pub keccak: Keccak256,
     #[extension]
+    pub memcpy: Memcpy,
+    #[extension]
     pub sha256: Sha256,
     #[extension]
     pub modular: ModularExtension,
@@ -128,6 +132,7 @@ impl Default for ExecuteConfig {
             io: Rv32Io,
             bigint: Int256::default(),
             keccak: Keccak256,
+            memcpy: Memcpy,
             sha256: Sha256,
             modular: ModularExtension::new(vec![
                 bn_config.modulus.clone(),
@@ -188,6 +193,11 @@ where
             &config.keccak,
             inventory,
         )?;
+        VmProverExtension::<E, _, _>::extend_prover(
+            &MemcpyCpuProverExt,
+            &config.memcpy,
+            inventory,
+        )?;
         VmProverExtension::<E, _, _>::extend_prover(&Sha2CpuProverExt, &config.sha256, inventory)?;
         VmProverExtension::<E, _, _>::extend_prover(
             &AlgebraCpuProverExt,
@@ -216,6 +226,7 @@ fn create_default_transpiler() -> Transpiler<BabyBear> {
         .with_extension(Rv32MTranspilerExtension)
         .with_extension(Int256TranspilerExtension)
         .with_extension(Keccak256TranspilerExtension)
+        .with_extension(MemcpyTranspilerExtension)
         .with_extension(Sha256TranspilerExtension)
         .with_extension(ModularTranspilerExtension)
         .with_extension(Fp2TranspilerExtension)

benchmarks/guest/fibonacci/openvm.toml

@@ -1,3 +1,4 @@
 [app_vm_config.rv32i]
 [app_vm_config.rv32m]
 [app_vm_config.io]
+[app_vm_config.memcpy]

benchmarks/guest/fibonacci/src/main.rs

@@ -1,14 +1,44 @@
-use openvm::io::{read, reveal_u32};
+use core::ptr;
 
-pub fn main() {
-    let n: u64 = read();
-    let mut a: u64 = 0;
-    let mut b: u64 = 1;
-    for _ in 0..n {
-        let c: u64 = a.wrapping_add(b);
-        a = b;
-        b = c;
+openvm::entry!(main);
+
+/// Moves all the elements of `src` into `dst`, leaving `src` empty.
+#[no_mangle]
+pub fn append<T>(dst: &mut [T], src: &mut [T], shift: usize) {
+    let src_len = src.len();
+    let dst_len = dst.len();
+
+    unsafe {
+        // The call to add is always safe because `Vec` will never
+        // allocate more than `isize::MAX` bytes.
+        let dst_ptr = dst.as_mut_ptr().wrapping_add(shift);
+        let src_ptr = src.as_ptr();
+        println!("dst_ptr: {:?}", dst_ptr);
+        println!("src_ptr: {:?}", src_ptr);
+        println!("src_len: {:?}", src_len);
+
+        // The two regions cannot overlap because mutable references do
+        // not alias, and two different vectors cannot own the same
+        // memory.
+        ptr::copy_nonoverlapping(src_ptr, dst_ptr, src_len);
     }
-    reveal_u32(a as u32, 0);
-    reveal_u32((a >> 32) as u32, 1);
 }
+
+pub fn main() {
+    let mut a: [u8; 1000] = [1; 1000];
+    let mut b: [u8; 500] = [2; 500];
+
+    let shift: usize = 0;
+    append(&mut a, &mut b, shift);
+
+    for i in 0..1000 {
+        if i < shift || i >= shift + b.len() {
+            assert_eq!(a[i], 1);
+        } else {
+            assert_eq!(a[i], 2);
+        }
+    }
+
+    println!("a: {:?}", a);
+    println!("b: {:?}", b);
+}

crates/circuits/primitives/src/assert_less_than/mod.rs

@@ -3,7 +3,7 @@ use openvm_circuit_primitives_derive::AlignedBorrow;
 use openvm_stark_backend::{
     interaction::InteractionBuilder,
     p3_air::AirBuilder,
-    p3_field::{Field, FieldAlgebra},
+    p3_field::{Field, FieldAlgebra, PrimeField32},
 };
 
 use crate::{
@@ -58,6 +58,14 @@ pub struct LessThanAuxCols<T, const AUX_LEN: usize> {
     pub lower_decomp: [T; AUX_LEN],
 }
 
+impl<F: PrimeField32, const AUX_LEN: usize> Default for LessThanAuxCols<F, AUX_LEN> {
+    fn default() -> Self {
+        Self {
+            lower_decomp: [F::ZERO; AUX_LEN],
+        }
+    }
+}
+
 /// This is intended for use as a **SubAir**, not as a standalone Air.
 ///
 /// This SubAir constrains that `x < y` when `count != 0`, assuming

crates/sdk/Cargo.toml

@@ -18,6 +18,8 @@ openvm-ecc-circuit = { workspace = true }
 openvm-ecc-transpiler = { workspace = true }
 openvm-keccak256-circuit = { workspace = true }
 openvm-keccak256-transpiler = { workspace = true }
+openvm-memcpy-circuit = { workspace = true }
+openvm-memcpy-transpiler = { workspace = true }
 openvm-sha256-circuit = { workspace = true }
 openvm-sha256-transpiler = { workspace = true }
 openvm-pairing-circuit = { workspace = true }