Merge NOT and POP flags.

evm/src/cpu/columns/ops.rs

@@ -12,11 +12,10 @@ pub struct OpsColumnsView<T: Copy> {
     pub fp254_op: T,   // Combines ADD_FP254, MUL_FP254 and SUB_FP254 flags.
     pub eq_iszero: T,  // Combines EQ and ISZERO flags.
     pub logic_op: T,   // Combines AND, OR and XOR flags.
-    pub not: T,
-    pub shift: T, // Combines SHL and SHR flags.
+    pub not_pop: T,    // Combines NOT and POP flags.
+    pub shift: T,      // Combines SHL and SHR flags.
     pub keccak_general: T,
     pub prover_input: T,
-    pub pop: T,
     pub jumps: T, // Combines JUMP and JUMPI flags.
     pub pc: T,
     pub jumpdest: T,

evm/src/cpu/control_flow.rs

@@ -8,17 +8,16 @@ use crate::constraint_consumer::{ConstraintConsumer, RecursiveConstraintConsumer
 use crate::cpu::columns::{CpuColumnsView, COL_MAP};
 use crate::cpu::kernel::aggregator::KERNEL;
 
-const NATIVE_INSTRUCTIONS: [usize; 17] = [
+const NATIVE_INSTRUCTIONS: [usize; 16] = [
     COL_MAP.op.binary_op,
     COL_MAP.op.ternary_op,
     COL_MAP.op.fp254_op,
     COL_MAP.op.eq_iszero,
     COL_MAP.op.logic_op,
-    COL_MAP.op.not,
+    COL_MAP.op.not_pop,
     COL_MAP.op.shift,
     COL_MAP.op.keccak_general,
     COL_MAP.op.prover_input,
-    COL_MAP.op.pop,
     // not JUMPS (possible need to jump)
     COL_MAP.op.pc,
     COL_MAP.op.jumpdest,

evm/src/cpu/decode.rs

@@ -23,18 +23,17 @@ use crate::cpu::columns::{CpuColumnsView, COL_MAP};
 /// behavior.
 /// Note: invalid opcodes are not represented here. _Any_ opcode is permitted to decode to
 /// `is_invalid`. The kernel then verifies that the opcode was _actually_ invalid.
-const OPCODES: [(u8, usize, bool, usize); 16] = [
+const OPCODES: [(u8, usize, bool, usize); 14] = [
     // (start index of block, number of top bits to check (log2), kernel-only, flag column)
     // ADD, MUL, SUB, DIV, MOD, LT, GT and BYTE flags are handled partly manually here, and partly through the Arithmetic table CTL.
     // ADDMOD, MULMOD and SUBMOD flags are handled partly manually here, and partly through the Arithmetic table CTL.
     // FP254 operation flags are handled partly manually here, and partly through the Arithmetic table CTL.
     (0x14, 1, false, COL_MAP.op.eq_iszero),
     // AND, OR and XOR flags are handled partly manually here, and partly through the Logic table CTL.
-    (0x19, 0, false, COL_MAP.op.not),
+    // NOT and POP are handled manually here.
     // SHL and SHR flags are handled partly manually here, and partly through the Logic table CTL.
     (0x21, 0, true, COL_MAP.op.keccak_general),
     (0x49, 0, true, COL_MAP.op.prover_input),
-    (0x50, 0, false, COL_MAP.op.pop),
     (0x56, 1, false, COL_MAP.op.jumps), // 0x56-0x57
     (0x58, 0, false, COL_MAP.op.pc),
     (0x5b, 0, false, COL_MAP.op.jumpdest),
@@ -52,13 +51,14 @@ const OPCODES: [(u8, usize, bool, usize); 16] = [
 /// List of combined opcodes requiring a special handling.
 /// Each index in the list corresponds to an arbitrary combination
 /// of opcodes defined in evm/src/cpu/columns/ops.rs.
-const COMBINED_OPCODES: [usize; 6] = [
+const COMBINED_OPCODES: [usize; 7] = [
     COL_MAP.op.logic_op,
     COL_MAP.op.fp254_op,
     COL_MAP.op.binary_op,
     COL_MAP.op.ternary_op,
     COL_MAP.op.shift,
     COL_MAP.op.m_op_general,
+    COL_MAP.op.not_pop,
 ];
 
 /// Break up an opcode (which is 8 bits long) into its eight bits.
@@ -147,6 +147,14 @@ pub fn eval_packed_generic<P: PackedField>(
         * (opcode - P::Scalar::from_canonical_usize(0xfc_usize))
         * lv.op.m_op_general;
     yield_constr.constraint(m_op_constr);
+
+    // Manually check lv.op.not_pop.
+    // Both NOT and POP can be called outside of the kernel mode:
+    // there is no need to constrain them in that regard.
+    let not_pop_op = (opcode - P::Scalar::from_canonical_usize(0x19_usize))
+        * (opcode - P::Scalar::from_canonical_usize(0x50_usize))
+        * lv.op.not_pop;
+    yield_constr.constraint(not_pop_op);
 }
 
 pub fn eval_ext_circuit<F: RichField + Extendable<D>, const D: usize>(
@@ -249,4 +257,17 @@ pub fn eval_ext_circuit<F: RichField + Extendable<D>, const D: usize>(
     m_op_constr = builder.mul_extension(m_op_constr, lv.op.m_op_general);
 
     yield_constr.constraint(builder, m_op_constr);
+
+    // Manually check lv.op.not_pop.
+    // Both NOT and POP can be called outside of the kernel mode:
+    // there is no need to constrain them in that regard.
+    let not_opcode = builder.constant_extension(F::Extension::from_canonical_usize(0x19_usize));
+    let pop_opcode = builder.constant_extension(F::Extension::from_canonical_usize(0x50_usize));
+
+    let not_constr = builder.sub_extension(opcode, not_opcode);
+    let pop_constr = builder.sub_extension(opcode, pop_opcode);
+
+    let mut not_pop_constr = builder.mul_extension(not_constr, pop_constr);
+    not_pop_constr = builder.mul_extension(lv.op.not_pop, not_pop_constr);
+    yield_constr.constraint(builder, not_pop_constr);
 }

evm/src/cpu/gas.rs

@@ -24,11 +24,10 @@ const SIMPLE_OPCODES: OpsColumnsView<Option<u32>> = OpsColumnsView {
     fp254_op: KERNEL_ONLY_INSTR,
     eq_iszero: G_VERYLOW,
     logic_op: G_VERYLOW,
-    not: G_VERYLOW,
+    not_pop: None, // This is handled manually below
     shift: G_VERYLOW,
     keccak_general: KERNEL_ONLY_INSTR,
     prover_input: KERNEL_ONLY_INSTR,
-    pop: G_BASE,
     jumps: None, // Combined flag handled separately.
     pc: G_BASE,
     jumpdest: G_JUMPDEST,
@@ -105,6 +104,13 @@ fn eval_packed_accumulate<P: PackedField>(
     let ternary_op_cost = P::Scalar::from_canonical_u32(G_MID.unwrap())
         - lv.opcode_bits[1] * P::Scalar::from_canonical_u32(G_MID.unwrap());
     yield_constr.constraint_transition(lv.op.ternary_op * (gas_diff - ternary_op_cost));
+
+    // For NOT and POP.
+    // NOT is differentiated from POP by its first bit set to 1.
+    let not_pop_cost = (P::ONES - lv.opcode_bits[0])
+        * P::Scalar::from_canonical_u32(G_BASE.unwrap())
+        + lv.opcode_bits[0] * P::Scalar::from_canonical_u32(G_VERYLOW.unwrap());
+    yield_constr.constraint_transition(lv.op.not_pop * (gas_diff - not_pop_cost));
 }
 
 fn eval_packed_init<P: PackedField>(
@@ -233,6 +239,20 @@ fn eval_ext_circuit_accumulate<F: RichField + Extendable<D>, const D: usize>(
     let gas_diff = builder.sub_extension(nv_lv_diff, ternary_op_cost);
     let constr = builder.mul_extension(filter, gas_diff);
     yield_constr.constraint_transition(builder, constr);
+
+    // For NOT and POP.
+    // NOT is differentiated from POP by its first bit set to 1.
+    let filter = lv.op.not_pop;
+    let one = builder.one_extension();
+    let mut not_pop_cost =
+        builder.mul_const_extension(F::from_canonical_u32(G_VERYLOW.unwrap()), lv.opcode_bits[0]);
+    let mut pop_cost = builder.sub_extension(one, lv.opcode_bits[0]);
+    pop_cost = builder.mul_const_extension(F::from_canonical_u32(G_BASE.unwrap()), pop_cost);
+    not_pop_cost = builder.add_extension(not_pop_cost, pop_cost);
+
+    let not_pop_gas_diff = builder.sub_extension(nv_lv_diff, not_pop_cost);
+    let not_pop_constr = builder.mul_extension(filter, not_pop_gas_diff);
+    yield_constr.constraint_transition(builder, not_pop_constr);
 }
 
 fn eval_ext_circuit_init<F: RichField + Extendable<D>, const D: usize>(

evm/src/cpu/simple_logic/mod.rs

@@ -14,7 +14,7 @@ pub fn eval_packed<P: PackedField>(
     nv: &CpuColumnsView<P>,
     yield_constr: &mut ConstraintConsumer<P>,
 ) {
-    not::eval_packed(lv, yield_constr);
+    not::eval_packed(lv, nv, yield_constr);
     eq_iszero::eval_packed(lv, nv, yield_constr);
 }
 
@@ -24,6 +24,6 @@ pub fn eval_ext_circuit<F: RichField + Extendable<D>, const D: usize>(
     nv: &CpuColumnsView<ExtensionTarget<D>>,
     yield_constr: &mut RecursiveConstraintConsumer<F, D>,
 ) {
-    not::eval_ext_circuit(builder, lv, yield_constr);
+    not::eval_ext_circuit(builder, lv, nv, yield_constr);
     eq_iszero::eval_ext_circuit(builder, lv, nv, yield_constr);
 }

evm/src/cpu/simple_logic/not.rs

@@ -7,33 +7,39 @@ use plonky2::iop::ext_target::ExtensionTarget;
 use crate::constraint_consumer::{ConstraintConsumer, RecursiveConstraintConsumer};
 use crate::cpu::columns::CpuColumnsView;
 use crate::cpu::membus::NUM_GP_CHANNELS;
+use crate::cpu::stack;
 
 const LIMB_SIZE: usize = 32;
 const ALL_1_LIMB: u64 = (1 << LIMB_SIZE) - 1;
 
 pub fn eval_packed<P: PackedField>(
     lv: &CpuColumnsView<P>,
+    nv: &CpuColumnsView<P>,
     yield_constr: &mut ConstraintConsumer<P>,
 ) {
     // This is simple: just do output = 0xffffffff - input.
     let input = lv.mem_channels[0].value;
     let output = lv.mem_channels[NUM_GP_CHANNELS - 1].value;
-    let filter = lv.op.not;
+    let filter = lv.op.not_pop * lv.opcode_bits[0];
     for (input_limb, output_limb) in input.into_iter().zip(output) {
         yield_constr.constraint(
             filter * (output_limb + input_limb - P::Scalar::from_canonical_u64(ALL_1_LIMB)),
         );
     }
+
+    // Stack constraints.
+    stack::eval_packed_one(lv, nv, filter, stack::BASIC_UNARY_OP.unwrap(), yield_constr);
 }
 
 pub fn eval_ext_circuit<F: RichField + Extendable<D>, const D: usize>(
     builder: &mut plonky2::plonk::circuit_builder::CircuitBuilder<F, D>,
     lv: &CpuColumnsView<ExtensionTarget<D>>,
+    nv: &CpuColumnsView<ExtensionTarget<D>>,
     yield_constr: &mut RecursiveConstraintConsumer<F, D>,
 ) {
     let input = lv.mem_channels[0].value;
     let output = lv.mem_channels[NUM_GP_CHANNELS - 1].value;
-    let filter = lv.op.not;
+    let filter = builder.mul_extension(lv.op.not_pop, lv.opcode_bits[0]);
     for (input_limb, output_limb) in input.into_iter().zip(output) {
         let constr = builder.add_extension(output_limb, input_limb);
         let constr = builder.arithmetic_extension(
@@ -45,4 +51,14 @@ pub fn eval_ext_circuit<F: RichField + Extendable<D>, const D: usize>(
         );
         yield_constr.constraint(builder, constr);
     }
+
+    // Stack constraints.
+    stack::eval_ext_circuit_one(
+        builder,
+        lv,
+        nv,
+        filter,
+        stack::BASIC_UNARY_OP.unwrap(),
+        yield_constr,
+    );
 }