diff --git a/evm/src/cpu/kernel/aggregator.rs b/evm/src/cpu/kernel/aggregator.rs
index 114a3d83..81be7f96 100644
--- a/evm/src/cpu/kernel/aggregator.rs
+++ b/evm/src/cpu/kernel/aggregator.rs
@@ -90,5 +90,6 @@ mod tests {
         // Make sure we can parse and assemble the entire kernel.
         let kernel = combined_kernel();
         debug!("Total kernel size: {} bytes", kernel.code.len());
+        dbg!("Total kernel size: {} bytes", kernel.code.len());
     }
 }
diff --git a/evm/src/cpu/kernel/asm/moddiv.asm b/evm/src/cpu/kernel/asm/moddiv.asm
index 2b76d054..780473b9 100644
--- a/evm/src/cpu/kernel/asm/moddiv.asm
+++ b/evm/src/cpu/kernel/asm/moddiv.asm
@@ -30,16 +30,10 @@
     // stack: x
     PROVER_INPUT(ff::bn254_base::inverse)
     // stack: x^-1, x
-    %stack (inv, x) -> (inv, x, @BN_BASE, inv, x)
-    // stack: x^-1, x, N, x^-1, x
+    %stack (inv, x) -> (inv, x, @BN_BASE, inv)
+    // stack: x^-1, x, N, x^-1
     MULMOD
-    // stack: x^-1 * x, x^-1, x
-    PUSH 1
-    // stack: 1, x^-1 * x, x^-1, x
-    %assert_eq
-    // stack: x^-1, x
-    SWAP1
-    // stack: x, x^-1
-    POP
+    // stack: x^-1 * x, x^-1
+    %assert_eq_const(1)
     // stack: x^-1
 %endmacro
diff --git a/evm/src/cpu/kernel/asm/secp256k1/inverse_scalar.asm b/evm/src/cpu/kernel/asm/secp256k1/inverse_scalar.asm
index 7b859b4f..6e1563e2 100644
--- a/evm/src/cpu/kernel/asm/secp256k1/inverse_scalar.asm
+++ b/evm/src/cpu/kernel/asm/secp256k1/inverse_scalar.asm
@@ -22,16 +22,10 @@
     // stack: x
     PROVER_INPUT(ff::secp256k1_scalar::inverse)
     // stack: x^-1, x
-    %stack (inv, x) -> (inv, x, @SECP_SCALAR, inv, x)
-    // stack: x^-1, x, N, x^-1, x
+    %stack (inv, x) -> (inv, x, @SECP_SCALAR, inv)
+    // stack: x^-1, x, N, x^-1
     MULMOD
-    // stack: x^-1 * x, x^-1, x
-    PUSH 1
-    // stack: 1, x^-1 * x, x^-1, x
-    %assert_eq
-    // stack: x^-1, x
-    SWAP1
-    // stack: x, x^-1
-    POP
+    // stack: x^-1 * x, x^-1
+    %assert_eq_const(1)
     // stack: x^-1
 %endmacro
diff --git a/evm/src/cpu/kernel/asm/secp256k1/lift_x.asm b/evm/src/cpu/kernel/asm/secp256k1/lift_x.asm
index 4bef700e..dc765518 100644
--- a/evm/src/cpu/kernel/asm/secp256k1/lift_x.asm
+++ b/evm/src/cpu/kernel/asm/secp256k1/lift_x.asm
@@ -10,7 +10,7 @@
     // stack: x^3+7, x, parity
     DUP1
     // stack: x^3+7, x^3+7, parity
-    %sqrt_secp_base
+    %sqrt_secp_base_unsafe
     // stack: y, x^3+7, x, parity
     SWAP1
     // stack: x^3+7, y, parity
@@ -61,7 +61,8 @@
 %endmacro
 
 // Non-deterministically provide the square root modulo N.
-%macro sqrt_secp_base
+// Note: The square root is not checked and the macro doesn't not panic if `x` is not a square.
+%macro sqrt_secp_base_unsafe
     // stack: x
     PROVER_INPUT(ff::secp256k1_base::sqrt)
     // stack: √x, x
diff --git a/evm/src/cpu/kernel/asm/secp256k1/moddiv.asm b/evm/src/cpu/kernel/asm/secp256k1/moddiv.asm
index fd077b11..d878dc14 100644
--- a/evm/src/cpu/kernel/asm/secp256k1/moddiv.asm
+++ b/evm/src/cpu/kernel/asm/secp256k1/moddiv.asm
@@ -30,16 +30,10 @@
     // stack: x
     PROVER_INPUT(ff::secp256k1_base::inverse)
     // stack: x^-1, x
-    %stack (inv, x) -> (inv, x, @SECP_BASE, inv, x)
-    // stack: x^-1, x, N, x^-1, x
+    %stack (inv, x) -> (inv, x, @SECP_BASE, inv)
+    // stack: x^-1, x, N, x^-1
     MULMOD
-    // stack: x^-1 * x, x^-1, x
-    PUSH 1
-    // stack: 1, x^-1 * x, x^-1, x
-    %assert_eq
-    // stack: x^-1, x
-    SWAP1
-    // stack: x, x^-1
-    POP
+    // stack: x^-1 * x, x^-1
+    %assert_eq_const(1)
     // stack: x^-1
 %endmacro
diff --git a/evm/src/cpu/kernel/interpreter.rs b/evm/src/cpu/kernel/interpreter.rs
index 0634b864..61272c24 100644
--- a/evm/src/cpu/kernel/interpreter.rs
+++ b/evm/src/cpu/kernel/interpreter.rs
@@ -336,7 +336,7 @@ impl<'a> Interpreter<'a> {
             .prover_inputs_map
             .get(&(self.offset - 1))
             .ok_or_else(|| anyhow!("Offset not in prover inputs."))?;
-        let output = prover_input_fn.run(self.stack.clone());
+        let output = prover_input_fn.run(&self.stack);
         self.stack.push(output);
         self.prover_inputs.push(output);
         Ok(())
diff --git a/evm/src/cpu/kernel/mod.rs b/evm/src/cpu/kernel/mod.rs
index d146884a..5b9b1b4a 100644
--- a/evm/src/cpu/kernel/mod.rs
+++ b/evm/src/cpu/kernel/mod.rs
@@ -1,14 +1,15 @@
 pub mod aggregator;
 pub mod assembler;
 mod ast;
-pub mod interpreter;
 pub(crate) mod keccak_util;
 mod opcodes;
 mod parser;
-mod prover_input;
+pub mod prover_input;
 mod stack_manipulation;
 mod txn_fields;
 
+#[cfg(test)]
+mod interpreter;
 #[cfg(test)]
 mod tests;
 
diff --git a/evm/src/cpu/kernel/prover_input.rs b/evm/src/cpu/kernel/prover_input.rs
index b27eb561..38e1914e 100644
--- a/evm/src/cpu/kernel/prover_input.rs
+++ b/evm/src/cpu/kernel/prover_input.rs
@@ -20,7 +20,7 @@ impl From<Vec<String>> for ProverInputFn {
 
 impl ProverInputFn {
     /// Run the function on the stack.
-    pub(crate) fn run(&self, stack: Vec<U256>) -> U256 {
+    pub fn run(&self, stack: &[U256]) -> U256 {
         match self.0[0].as_str() {
             "ff" => self.run_ff(stack),
             "mpt" => todo!(),
@@ -29,10 +29,10 @@ impl ProverInputFn {
     }
 
     // Finite field operations.
-    fn run_ff(&self, mut stack: Vec<U256>) -> U256 {
+    fn run_ff(&self, stack: &[U256]) -> U256 {
         let field = Field::from_str(self.0[1].as_str()).unwrap();
         let op = FieldOp::from_str(self.0[2].as_str()).unwrap();
-        let x = stack.pop().expect("Empty stack");
+        let x = *stack.last().expect("Empty stack");
         field.op(op, x)
     }
 
@@ -130,7 +130,7 @@ fn modexp(x: U256, e: U256, n: U256) -> U256 {
     let mut product = U256::one();
 
     for j in 0..256 {
-        if !(e >> j & U256::one()).is_zero() {
+        if e.bit(j) {
             product = U256::try_from(product.full_mul(current) % n).unwrap();
         }
         current = U256::try_from(current.full_mul(current) % n).unwrap();