feat: Circuit simulator by codygunton · Pull Request #580 · AztecProtocol/barretenberg

codygunton · 2023-07-05T15:48:57Z

Description

Please provide a paragraph or two giving a summary of the change, including relevant motivation and context.

Checklist:

ludamad · 2023-07-05T16:25:01Z

cpp/src/barretenberg/stdlib/primitives/field/field.cpp

+        multiplicative_constant = 1;
+        witness_index = IS_CONSTANT;
+        // context -> stored_value = value.witness;
+    } else {


Hmm why is this if constexpr needed?

Normally values of field elements are stored in the variables array of a given circuit builder. The index in that array is called a "real variable index". In order to avoid redundantly storing variable values, a witness object wraps an index idx and the corresponding variable value is accessed as in variables[real_variable_index[idx]]. I'm trying out a model where we do away with the variables array altogether, but just have a witness wrap a value. Therefore to initialize a field element from a witness I need to supply the witness info not through providing a witness index, but a witness value itself. The right way to do that is through setting the additive_constant, since 'normalizing' the field element would put the same value there. That said, I'm not sure this is the best approach.

But if you're just talking about the constexpr then I think we could avoid code generation by just defining the function for the particular template parameter CircuitSimulatorBN254. We will also have a CircuitSimulatorGrumpkin though... anyway, experimenting.

codygunton · 2023-07-06T16:26:07Z

cpp/src/barretenberg/proof_system/circuit_builder/circuit_simulator.hpp

@@ -0,0 +1,113 @@
+#pragma once


TODO: share code with CircuitConstructorBase

codygunton · 2023-07-06T16:26:58Z

cpp/src/barretenberg/proof_system/flavor/flavor.hpp

 template <typename T> 
 concept IsHonkFlavor = IsAnyOf<T, honk::flavor::Standard, honk::flavor::Ultra, honk::flavor::StandardGrumpkin, honk::flavor::UltraGrumpkin>;

+// WORKTODO: move? smart way of not referring to instances?


We will probably want a CircuitSimulatorGrumpkin

codygunton · 2023-07-06T16:28:02Z

cpp/src/barretenberg/stdlib/hash/blake2s/blake2s.test.cpp

@@ -1,13 +1,14 @@
 #include "barretenberg/crypto/blake2s/blake2s.hpp"
-#include "barretenberg/proof_system/circuit_builder/ultra_circuit_builder.hpp"
+#include "barretenberg/proof_system/circuit_builder/circuit_simulator.hpp"


(blake2s tests pass)

codygunton · 2023-07-07T07:53:39Z

cpp/src/barretenberg/stdlib/primitives/address/address.hpp

 #include "barretenberg/stdlib/primitives/point/point.hpp"
 #include "barretenberg/stdlib/primitives/witness/witness.hpp"

+// TODO: This does not belong in barretenberg.


The notion of an "address" I mean... just a side note.

codygunton · 2023-07-07T08:03:51Z

cpp/src/barretenberg/stdlib/primitives/bool/bool.cpp

    const bool_t lhs = *this;
    ComposerContext* ctx = lhs.get_context() ? lhs.get_context() : rhs.get_context();

-    if (lhs.is_constant() && rhs.is_constant()) {


We want to put the composer in a failing state without changing existing code. We can't reuse the "both are witnesses" code because the assert_equal that takes in values (ant not witness indices) only exists on the simulator.

codygunton · 2023-07-07T08:04:36Z

cpp/src/barretenberg/stdlib/primitives/byte_array/byte_array.cpp


    if (byte.is_constant()) {
        field_t<Composer> low(context, low_value);
-        field_t<Composer> high(context, high_value);


CIRCUIT BUG: we were not correctly setting bits in the constant case.

codygunton · 2023-07-07T08:05:11Z

cpp/src/barretenberg/stdlib/primitives/byte_array/byte_array.test.cpp

    EXPECT_EQ(out[0], uint8_t(0));
-    EXPECT_EQ(out[1], uint8_t(7));
-    EXPECT_EQ(out[3], uint8_t(5));
+    EXPECT_EQ(out[1], uint8_t(7)); // start with 0000'0010, want 0000'0111, get 0000'0110


TODO: clean up comment

codygunton · 2023-07-07T08:06:47Z

cpp/src/barretenberg/proof_system/circuit_builder/circuit_simulator.hpp

+    uint32_t put_constant_variable([[maybe_unused]] const barretenberg::fr& variable) { return 1028; }
+    void set_public_input([[maybe_unused]] const uint32_t witness_index)
+    {
+        // WORKTODO Public input logic?


Haven't thought through potential subtleties with set_public_input or fix_witness.

codygunton · 2023-07-07T08:07:51Z

cpp/src/barretenberg/proof_system/circuit_builder/circuit_simulator.hpp

+    //   return 0; // WORKTODO: return part of `in` for debugging purposes?
+    // }
+
+    inline uint32_t add_variable([[maybe_unused]] const barretenberg::fr index) const { return 1028; }


1028 = 0x404, recognizable when printing if something went wrong in the simulation.

codygunton · 2023-07-07T08:10:40Z

cpp/src/barretenberg/proof_system/circuit_builder/circuit_simulator.hpp

+    size_t get_num_constant_gates() { return 1028; };
+    // maybe this shouldn't be implemented?
+
+    bool create_range_constraint(FF const& elt,


My goal is to preserve the composer interface when I can, but without tracking witness indices. In this case I decided to change the interface (input is a field element elt rather than a witness index). One way to unify interfaces would be to pass around witness_t's rather than indices, since this is a container for both an index and a vale. But this would undermine the optimization of tracking witness indices.

codygunton · 2023-07-07T08:11:39Z

cpp/src/barretenberg/proof_system/circuit_builder/circuit_simulator.hpp

+        return { 1028 };
+    };
+
+    void assert_equal(FF left, FF right, std::string const& msg)


Another departure of the form "witness index becomes a value".

codygunton · 2023-07-07T08:12:29Z

cpp/src/barretenberg/proof_system/polynomial_store/polynomial_store_cache.hpp

 * to max_cache_size_ polynomials. This saves on many expensive copies of large amounts of memory to the external
 * store. Smaller polynomials get swapped out, but they're also much cheaper to read/write.
- * The default ctor sets the cache size to 70.
+ * The default ctor sets the cache size to 40.


Unrelated: I should have changed this in a PR earlier this week.

codygunton · 2023-07-07T08:13:26Z

cpp/src/barretenberg/proof_system/types/merkle_hash_type.hpp

@@ -2,5 +2,5 @@

 namespace proof_system::merkle {
 // TODO(Cody) Get rid of this?


IMO the various enum's that track types should go away.

codygunton · 2023-07-07T08:14:35Z

cpp/src/barretenberg/stdlib/primitives/bool/bool.test.cpp

    EXPECT_EQ(result, true);

-    auto gates_after = composer.get_num_gates();
-    EXPECT_EQ(gates_after - gates_before, 6UL);


These tests that function like benchmarks should break, since we don't track gates, but they aren't testing circuit logic, so I felt comfortable deleting them.

…berg into cg/simulate-spike

codygunton · 2024-05-27T13:33:07Z

Subsumed by AztecProtocol/aztec-packages#1195, which @maramihali took over the line

codygunton added 5 commits July 5, 2023 07:46

bool tests pass

ad0f6e2

Fix comment.

3c3c821

Merge branch 'master' into cg/simulate-spike

f5ad944

Stdlib bool and field build and tests pass.

6bbf664

Temp explicit inst macros; uncomment stuff.

c7c67d0

codygunton self-assigned this Jul 5, 2023

ludamad reviewed Jul 5, 2023

View reviewed changes

codygunton added 9 commits July 6, 2023 10:41

Test field add_two

7848146

safe_uint works (fix bug)

4da38c9

byte_array works (fix bug)

213ac0f

Everything compiles

be9d58c

packed_byte_array tests pass

bf57c0e

1028 == 0x404

a7c1f0b

group, schnorr signatures work

1322151

Reinstate field tests.

e347cd1

Leave note.

b903a8a

codygunton commented Jul 6, 2023

View reviewed changes

codygunton commented Jul 7, 2023

View reviewed changes

Fix GCC error.

3d70094

codygunton force-pushed the cg/simulate-spike branch from d720a1a to 3d70094 Compare July 17, 2023 14:59

codygunton added 2 commits July 17, 2023 15:30

Merge branch 'cg/simulate-spike' of github.com:AztecProtocol/barreten…

9fae028

…berg into cg/simulate-spike

Fix private function call segfault.

3b79321

codygunton changed the title ~~feat: Add circuit simulator~~ feat: Circuit simulator Jul 17, 2023

codygunton added 22 commits July 18, 2023 20:59

Partial private kernel.

29fba08

Merge branch 'master' into cg/simulate-spike

e908e94

Simulate pedersen using plookup.

3f92696

Plookup tests pass

a6e50ef

Squash; uint works.

7639b4b

More pedersen tests.

0069914

AES

af52eca

SHA and ECDSA

fbc541c

Partial keccak (bug in constant case?).

66a927e

blake3s

3c45641

bit array

0319741

bigfield

d56a5f2

biggroup

7a6cd60

array

6f0cf3c

Logic test

60bced7

Fix/hack

6a9d027

Schnorr signatures

5d8715b

Fixes for private kernel.

0ca164a

Handle PIs in simulator.

1bcd9a1

Change circuit type.

79eb049

Align hash types.

8adc31c

Pedersen hack: skip problematic test.

e8ed76b

codygunton closed this May 27, 2024

ludamad deleted the cg/simulate-spike branch February 20, 2026 00:01

		@@ -2,5 +2,5 @@

		namespace proof_system::merkle {
		// TODO(Cody) Get rid of this?

Conversation

codygunton commented Jul 5, 2023

Description

Checklist:

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

codygunton Jul 6, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

codygunton Jul 7, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

codygunton commented May 27, 2024

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

codygunton Jul 6, 2023 •

edited

Loading

codygunton Jul 7, 2023 •

edited

Loading