Merge branch 'master' into ad/test-transform-circleci

barretenberg/cpp/CMakePresets.json

@@ -274,16 +274,6 @@
         "MULTITHREADING": "ON"
       }
     },
-    {
-      "name": "wasm-bench",
-      "displayName": "WASM benchmarking.",
-      "description": "WASM benchmarking.",
-      "inherits": "wasm-threads",
-      "binaryDir": "build-wasm-bench",
-      "environment": {
-        "CXXFLAGS": "-DWASMTIME_ENV_HACK"
-      }
-    },
     {
       "name": "xray",
       "displayName": "Build with multi-threaded XRay Profiling",
@@ -444,12 +434,6 @@
       "jobs": 0,
       "targets": ["barretenberg.wasm"]
     },
-    {
-      "name": "wasm-bench",
-      "configurePreset": "wasm-bench",
-      "inheritConfigureEnvironment": true,
-      "jobs": 0
-    },
     {
       "name": "xray",
       "configurePreset": "xray",

barretenberg/cpp/scripts/_benchmark_remote_lock.sh

@@ -23,5 +23,6 @@ echo "Benchmarking lock created at ~/BENCHMARK_IN_PROGRESS."
 
 # Trap to ensure cleanup runs on ANY exit, including from a signal
 trap cleanup EXIT
+trap cleanup INT # handle ctrl-c
 
 # don't exit, the caller script will run

barretenberg/cpp/scripts/benchmark_remote.sh

@@ -11,6 +11,7 @@ BENCHMARK=${1:-goblin_bench}
 COMMAND=${2:-./$BENCHMARK}
 PRESET=${3:-clang16}
 BUILD_DIR=${4:-build}
+HARDWARE_CONCURRENCY=${HARDWARE_CONCURRENCY:-16}
 
 # Move above script dir.
 cd $(dirname $0)/..
@@ -24,4 +25,4 @@ source scripts/_benchmark_remote_lock.sh
 cd $BUILD_DIR
 scp $BB_SSH_KEY ./bin/$BENCHMARK $BB_SSH_INSTANCE:$BB_SSH_CPP_PATH/build
 ssh $BB_SSH_KEY $BB_SSH_INSTANCE \
-  "cd $BB_SSH_CPP_PATH/build ; $COMMAND"
+  "cd $BB_SSH_CPP_PATH/build ; HARDWARE_CONCURRENCY=$HARDWARE_CONCURRENCY $COMMAND"

barretenberg/cpp/scripts/benchmark_wasm.sh

@@ -3,15 +3,16 @@ set -eu
 
 BENCHMARK=${1:-goblin_bench}
 COMMAND=${2:-./bin/$BENCHMARK}
+HARDWARE_CONCURRENCY=${HARDWARE_CONCURRENCY:-16}
 
 # Move above script dir.
 cd $(dirname $0)/..
 
 # Configure and build.
-cmake --preset wasm-bench
-cmake --build --preset wasm-bench --target $BENCHMARK 
+cmake --preset wasm-threads
+cmake --build --preset wasm-threads --target $BENCHMARK 
 
-cd build-wasm-bench
+cd build-wasm-threads
 # Consistency with _wasm.sh targets / shorter $COMMAND.
 cp ./bin/$BENCHMARK .
-wasmtime run -Wthreads=y -Sthreads=y $COMMAND
+wasmtime run --env HARDWARE_CONCURRENCY=$HARDWARE_CONCURRENCY -Wthreads=y -Sthreads=y --dir=.. $COMMAND

barretenberg/cpp/scripts/benchmark_wasm_remote.sh

@@ -9,17 +9,22 @@ set -eu
 
 BENCHMARK=${1:-goblin_bench}
 COMMAND=${2:-./$BENCHMARK}
+HARDWARE_CONCURRENCY=${HARDWARE_CONCURRENCY:-16}
 
 # Move above script dir.
 cd $(dirname $0)/..
 
 # Configure and build.
-cmake --preset wasm-bench
-cmake --build --preset wasm-bench --target $BENCHMARK
+cmake --preset wasm-threads
+cmake --build --preset wasm-threads --target $BENCHMARK
 
 source scripts/_benchmark_remote_lock.sh
 
-cd build-wasm-bench
-scp $BB_SSH_KEY ./bin/$BENCHMARK $BB_SSH_INSTANCE:$BB_SSH_CPP_PATH/build-wasm-bench
+cd build-wasm-threads
+# ensure folder structure
+ssh $BB_SSH_KEY $BB_SSH_INSTANCE "mkdir -p $BB_SSH_CPP_PATH/build-wasm-threads"
+# copy build wasm threads
+scp $BB_SSH_KEY ./bin/$BENCHMARK $BB_SSH_INSTANCE:$BB_SSH_CPP_PATH/build-wasm-threads
+# run wasm benchmarking
 ssh $BB_SSH_KEY $BB_SSH_INSTANCE \
-  "cd $BB_SSH_CPP_PATH/build-wasm-bench ; /home/ubuntu/.wasmtime/bin/wasmtime run -Wthreads=y -Sthreads=y $COMMAND"
+  "cd $BB_SSH_CPP_PATH/build-wasm-threads ; /home/ubuntu/.wasmtime/bin/wasmtime run --env HARDWARE_CONCURRENCY=$HARDWARE_CONCURRENCY -Wthreads=y -Sthreads=y --dir=.. $COMMAND"

barretenberg/cpp/src/barretenberg/commitment_schemes/ipa/ipa.hpp

@@ -235,6 +235,9 @@ template <typename Curve> class IPA {
         // Compute G_zero
         // First construct s_vec
         std::vector<Fr> s_vec(poly_degree);
+        // TODO(https://github.com/AztecProtocol/barretenberg/issues/857): This code is not efficient as its O(nlogn).
+        // This can be optimized to be linear by computing a tree of products. Its very readable, so we're
+        // leaving it unoptimized for now.
         run_loop_in_parallel_if_effective(
             poly_degree,
             [&s_vec, &round_challenges_inv, log_poly_degree](size_t start, size_t end) {

barretenberg/cpp/src/barretenberg/env/hardware_concurrency.cpp

@@ -1,6 +1,7 @@
 #include "hardware_concurrency.hpp"
 #include <barretenberg/common/throw_or_abort.hpp>
 #include <cstdlib>
+#include <iostream>
 #include <stdexcept>
 #include <string>
 #include <thread>
@@ -12,14 +13,9 @@ uint32_t env_hardware_concurrency()
 #ifndef __wasm__
     try {
 #endif
-#ifdef WASMTIME_ENV_HACK
-        // TODO(https://github.com/AztecProtocol/barretenberg/issues/837): Undo this hack, rely on WASI.
-        return 16;
-#else
-    static auto val = std::getenv("HARDWARE_CONCURRENCY");
-    static const uint32_t cores = val ? (uint32_t)std::stoul(val) : std::thread::hardware_concurrency();
-    return cores;
-#endif
+        static auto val = std::getenv("HARDWARE_CONCURRENCY");
+        static const uint32_t cores = val ? (uint32_t)std::stoul(val) : std::thread::hardware_concurrency();
+        return cores;
 #ifndef __wasm__
     } catch (std::exception const&) {
         throw std::runtime_error("HARDWARE_CONCURRENCY invalid.");

barretenberg/cpp/src/barretenberg/polynomials/pow.bench.cpp

@@ -0,0 +1,27 @@
+#include "barretenberg/polynomials/pow.hpp"
+#include "barretenberg/ecc/curves/bn254/fr.hpp"
+#include <benchmark/benchmark.h>
+
+using namespace benchmark;
+using namespace bb;
+
+namespace {
+
+void compute_pow_poly(benchmark::State& state)
+{
+    // just set up huge vector
+    std::vector<bb::fr> betas{ 1,  2,  3,  4,  5,  6,  7,  8,  9,  10, 11, 12, 13, 14,
+                               15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28 };
+
+    for (auto _ : state) {
+        int64_t num_betas = state.range(0);
+        std::vector<bb::fr> cur_betas(betas.begin(), betas.begin() + num_betas);
+        PowPolynomial pow{ cur_betas };
+        pow.compute_values();
+    }
+}
+
+BENCHMARK(compute_pow_poly)->Unit(benchmark::kMillisecond)->Arg(20);
+
+} // namespace
+BENCHMARK_MAIN();

barretenberg/cpp/src/barretenberg/polynomials/pow.hpp

@@ -1,4 +1,6 @@
 #pragma once
+#include "barretenberg/common/compiler_hints.hpp"
+#include "barretenberg/common/op_count.hpp"
 #include "barretenberg/common/thread.hpp"
 
 #include <cstddef>
@@ -121,8 +123,9 @@ template <typename FF> struct PowPolynomial {
      * @brief Given \vec{β} = {β_0,...,β_{d-1}} compute pow_\vec{β}(i) for i=0,...,2^{d}-1
      *
      */
-    void compute_values()
+    BB_PROFILE void compute_values()
     {
+        BB_OP_COUNT_TIME();
         size_t pow_size = 1 << betas.size();
         pow_betas = std::vector<FF>(pow_size);
 
@@ -136,6 +139,11 @@ template <typename FF> struct PowPolynomial {
         size_t num_threads = std::min(desired_num_threads, max_num_threads); // fewer than max if justified
         num_threads = num_threads > 0 ? num_threads : 1;                     // ensure num threads is >= 1
         size_t iterations_per_thread = pow_size / num_threads;               // actual iterations per thread
+
+        // TODO(https://github.com/AztecProtocol/barretenberg/issues/864): This computation is asymtotically slow as it
+        // does pow_size * log(pow_size) work. However, in practice, its super efficient because its trivially
+        // parallelizable and only takes 45ms for the whole 6 iter IVC benchmark. Its also very readable, so we're
+        // leaving it unoptimized for now.
         parallel_for(num_threads, [&](size_t thread_idx) {
             size_t start = thread_idx * iterations_per_thread;
             size_t end = (thread_idx + 1) * iterations_per_thread;

barretenberg/cpp/src/barretenberg/srs/global_crs.cpp

@@ -24,23 +24,7 @@ void init_crs_factory(std::string crs_path)
     if (crs_factory != nullptr) {
         return;
     }
-#ifdef WASMTIME_ENV_HACK
-    static_cast<void>(crs_path);
-    // We only need this codepath in wasmtime because the SRS cannot be loaded in our usual ways
-    // and we don't need a real CRS for our purposes.
-    // TODO(https://github.com/AztecProtocol/barretenberg/issues/837): make this a real SRS.
-    std::cout << "WASMTIME_ENV_HACK: started generating fake bn254 curve" << std::endl;
-    std::vector<g1::affine_element> points;
-    // 2**19 points
-    points.reserve(1 << 19);
-    for (int i = 0; i < (1 << 19); i++) {
-        points.push_back(g1::affine_element::random_element());
-    }
-    init_crs_factory(points, g2::affine_element{ fq::random_element(), fq::random_element() });
-    std::cout << "WASMTIME_ENV_HACK: finished generating fake bn254 curve" << std::endl;
-#else
     crs_factory = std::make_shared<factories::FileCrsFactory<curve::BN254>>(crs_path);
-#endif
 }
 
 // Initializes the crs using the memory buffers
@@ -54,23 +38,7 @@ void init_grumpkin_crs_factory(std::string crs_path)
     if (grumpkin_crs_factory != nullptr) {
         return;
     }
-#ifdef WASMTIME_ENV_HACK
-    // We only need this codepath in wasmtime because the SRS cannot be loaded in our usual ways
-    // and we don't need a real CRS for our purposes.
-    // TODO(https://github.com/AztecProtocol/barretenberg/issues/837): make this a real SRS.
-    static_cast<void>(crs_path);
-    std::cout << "WASMTIME_ENV_HACK: started generating fake grumpkin curve" << std::endl;
-    std::vector<curve::Grumpkin::AffineElement> points;
-    // 2**18 points
-    points.reserve(1 << 18);
-    for (int i = 0; i < (1 << 18); i++) {
-        points.push_back(curve::Grumpkin::AffineElement::random_element());
-    }
-    std::cout << "WASMTIME_ENV_HACK: finished generating fake grumpkin curve" << std::endl;
-    init_grumpkin_crs_factory(points);
-#else
     grumpkin_crs_factory = std::make_shared<factories::FileCrsFactory<curve::Grumpkin>>(crs_path);
-#endif
 }
 
 std::shared_ptr<factories::CrsFactory<curve::BN254>> get_crs_factory()