feat: Tracy time with instrumentation

barretenberg/README.md

@@ -477,4 +477,4 @@ See Tracy manual linked here <https://github.com/wolfpld/tracy> for in-depth Tra
 The basic use of Tracy is to run a benchmark with the `cmake --preset tracy` build type, create a capture file, then
 transfer it to a local machine for interactive UI introspection.
 
-All the steps to do this effectively are included in cpp/scripts/benchmark_tracy.sh
+All the steps to do this effectively are included in various scripts in cpp/scripts/.

barretenberg/cpp/CMakeLists.txt

@@ -54,21 +54,22 @@ if(CHECK_CIRCUIT_STACKTRACES)
     add_compile_options(-DCHECK_CIRCUIT_STACKTRACES)
 endif()
 
-if(ENABLE_TRACY)
+if(ENABLE_TRACY OR ENABLE_TRACY_TIME_INSTRUMENTED)
     add_compile_options(-DTRACY_ENABLE)
     SET(TRACY_LIBS Tracy::TracyClient)
 else()
     SET(TRACY_LIBS)
 endif()
 
-if(TRACY_PROFILE_MEMORY)
-  add_compile_options(-DTRACY_MEMORY)
-endif()
 
-if(TRACY_PROFILE_TIME)
-  add_compile_options(-DTRACY_TIME)
+if(ENABLE_TRACY_TIME_INSTRUMENTED)
+    add_compile_options(-DTRACY_INSTRUMENTED)
 endif()
 
+if(TRACY_PROFILE_MEMORY)
+  add_compile_options(-DTRACY_MEMORY)
+  add_compile_options(-DTRACY_INSTRUMENTED)
+endif()
 
 if(ENABLE_ASAN)
     add_compile_options(-fsanitize=address)

barretenberg/cpp/CMakePresets.json

@@ -112,8 +112,8 @@
     },
     {
       "name": "tracy-memory",
-      "displayName": "Release build with tracy, optimized for memory tracking",
-      "description": "Release build with tracy, optimized for memory tracking",
+      "displayName": "Release build with tracy with memory tracking",
+      "description": "Release build with tracy with memory tracking",
       "inherits": "clang16",
       "binaryDir": "build-tracy-memory",
       "cacheVariables": {
@@ -122,20 +122,29 @@
       }
     },
     {
-      "name": "tracy-time",
-      "displayName": "Build for tracy time profiling",
-      "description": "Build for tracy time profiling",
-      "binaryDir": "build-tracy-time",
+      "name": "tracy-time-instrumented",
+      "displayName": "Build for tracy time profiling via instrumentation",
+      "description": "Build for tracy time profiling via instrumentation",
+      "binaryDir": "build-tracy-time-instrumented",
       "inherits": "clang16",
+      "cacheVariables": {
+        "ENABLE_TRACY_TIME_INSTRUMENTED": "ON"
+      }
+    },
+    {
+      "name": "tracy-time-sampled",
+      "displayName": "Build for tracy time profiling via sampling",
+      "description": "Build for tracy time profiling via sampling",
+      "binaryDir": "build-tracy-time-sampled",
+      "inherits": "default",
       "environment": {
         "CMAKE_BUILD_TYPE": "RelWithDebInfo",
         "CFLAGS": "-g -fno-omit-frame-pointer",
         "CXXFLAGS": "-g -fno-omit-frame-pointer",
         "LDFLAGS": "-g -fno-omit-frame-pointer -rdynamic"
       },
       "cacheVariables": {
-        "ENABLE_TRACY": "ON",
-        "TRACY_PROFILE_TIME": "ON"
+        "ENABLE_TRACY": "ON"
       }
     },
     {
@@ -497,9 +506,14 @@
       "configurePreset": "tracy-memory"
     },
     {
-      "name": "tracy-time",
+      "name": "tracy-time-instrumented",
+      "inherits": "default",
+      "configurePreset": "tracy-time-instrumented"
+    },
+    {
+      "name": "tracy-time-sampled",
       "inherits": "default",
-      "configurePreset": "tracy-time"
+      "configurePreset": "tracy-time-sampled"
     },
     {
       "name": "clang16-pic",

barretenberg/cpp/scripts/benchmark_wasm.sh

@@ -1,8 +1,8 @@
 #!/usr/bin/env bash
 set -eu
 
-BENCHMARK=${1:-goblin_bench}
-COMMAND=${2:-./bin/$BENCHMARK}
+BENCHMARK=${1:-client_ivc_bench}
+COMMAND=${2:-./bin/$BENCHMARK --benchmark_filter=ClientIVCBench/Full/6}
 HARDWARE_CONCURRENCY=${HARDWARE_CONCURRENCY:-16}
 
 # Move above script dir.

barretenberg/cpp/scripts/benchmark_tracy_build_mainframe_view_local.sh → barretenberg/cpp/scripts/profile_tracy_capture_mainframe_view_local.sh

@@ -13,30 +13,31 @@ USER=${1:-$USER}
 BOX=$USER-box
 BENCHMARK=${2:-client_ivc_bench}
 COMMAND=${3:-./bin/$BENCHMARK --benchmark_filter=ClientIVCBench/Full/6"\$"}
-
+HARDWARE_CONCURRENCY=${HARDWARE_CONCURRENCY:-16}
 # Can also set PRESET=tracy-gates env variable
 PRESET=${PRESET:-tracy-memory}
 
 ssh $BOX "
 	set -eux ;
-	cd ~/aztec-packages/barretenberg/cpp/ ;
-	cmake --preset $PRESET && cmake --build --preset $PRESET --target $BENCHMARK ;
 	! [ -d ~/tracy ] && git clone https://github.com/wolfpld/tracy ~/tracy ;
 	cd ~/tracy/capture ;
         git checkout 075395620a504c0cdcaf9bab3d196db16a043de7 ;
-	sudo apt-get install -y libdbus-1-dev libdbus-glib-1-dev ;
-	mkdir -p build && cd build && cmake .. && make -j ;
+	sudo apt-get install -y libdbus-1-dev libdbus-glib-1-dev libtbb-dev libfreetype-dev ;
+	mkdir -p build && cd build && cmake -DCMAKE_MESSAGE_LOG_LEVEL=Warning .. && make -j ;
+	cd ~/aztec-packages/barretenberg/cpp/ ;
+	cmake -DCMAKE_MESSAGE_LOG_LEVEL=Warning --preset $PRESET && cmake --build --preset $PRESET --target $BENCHMARK ;
 	./tracy-capture -a 127.0.0.1 -f -o trace-$BENCHMARK & ;
 	sleep 0.1 ;
 	cd ~/aztec-packages/barretenberg/cpp/build-$PRESET ;
 	ninja $BENCHMARK ;
-	$COMMAND ;
+	HARDWARE_CONCURRENCY=$HARDWARE_CONCURRENCY $COMMAND ;
 " &
+
 wait # TODO(AD) hack - not sure why needed
 ! [ -d ~/tracy ] && git clone https://github.com/wolfpld/tracy ~/tracy
 cd ~/tracy
 git checkout 075395620a504c0cdcaf9bab3d196db16a043de7 # release 0.11.0
-cmake -B profiler/build -S profiler -DCMAKE_BUILD_TYPE=Release
+cmake -DCMAKE_MESSAGE_LOG_LEVEL=Warning -B profiler/build -S profiler -DCMAKE_BUILD_TYPE=Release
 cmake --build profiler/build --parallel
 scp $BOX:/mnt/user-data/$USER/tracy/capture/build/trace-$BENCHMARK .
 ~/tracy/profiler/build/tracy-profiler trace-$BENCHMARK

barretenberg/cpp/scripts/profile_tracy_local.sh

@@ -0,0 +1,36 @@
+# Collect a profile completely locally, i.e., without using any remote machine for building or capturing.
+
+set -eux
+USER=${1:-$USER}
+BOX=$USER-box
+BENCHMARK=${2:-client_ivc_bench}
+COMMAND=${3:-./bin/$BENCHMARK --benchmark_filter=ClientIVCBench/Full/6}
+HARDWARE_CONCURRENCY=${HARDWARE_CONCURRENCY:-16}
+PRESET=${PRESET:-tracy-time-sampled}
+
+! [ -d ~/tracy ] && git clone https://github.com/wolfpld/tracy ~/tracy
+cd ~/tracy
+git checkout 075395620a504c0cdcaf9bab3d196db16a043de7 # release 0.11.0
+cmake -B profiler/build -S profiler -DCMAKE_BUILD_TYPE=Release
+cmake --build profiler/build --parallel
+
+cd ~/aztec-packages/barretenberg/cpp/
+cmake --preset $PRESET -DCMAKE_MESSAGE_LOG_LEVEL=Warning && cmake --build --preset $PRESET --target $BENCHMARK
+
+! [ -d ~/tracy ] && git clone https://github.com/wolfpld/tracy ~/tracy
+cd ~/tracy/capture
+git checkout 075395620a504c0cdcaf9bab3d196db16a043de7
+mkdir -p build && cd build && cmake .. -DCMAKE_MESSAGE_LOG_LEVEL=Warning && make -j
+
+./tracy-capture -a 127.0.0.1 -f -o ../trace-$BENCHMARK &
+sleep 0.1
+cd ~/aztec-packages/barretenberg/cpp/build-$PRESET/
+
+# Run the COMMAND with sudo if PRESET is 'tracy-time-sampled'
+if [ "$PRESET" = "tracy-time-sampled" ]; then
+    sudo HARDWARE_CONCURRENCY=$HARDWARE_CONCURRENCY $COMMAND
+else
+    HARDWARE_CONCURRENCY=$HARDWARE_CONCURRENCY $COMMAND
+fi
+
+~/tracy/profiler/build/tracy-profiler ~/tracy/capture/trace-$BENCHMARK

barretenberg/cpp/scripts/profile_wasm_samply.sh

@@ -0,0 +1,19 @@
+# This is to be run locally not in a container, so the user must handle samply installation.
+#!/usr/bin/env bash
+set -eu
+
+BENCHMARK=${1:-client_ivc_bench}
+COMMAND=${2:-./bin/$BENCHMARK --benchmark_filter=ClientIVCBench/Full/6}
+HARDWARE_CONCURRENCY=${HARDWARE_CONCURRENCY:-16}
+
+# Move above script dir.
+cd $(dirname $0)/..
+
+# Configure and build.
+cmake --preset wasm-threads -DCMAKE_MESSAGE_LOG_LEVEL=Warning
+cmake --build --preset wasm-threads --target $BENCHMARK
+
+cd build-wasm-threads
+# Consistency with _wasm.sh targets / shorter $COMMAND.
+cp ./bin/$BENCHMARK .
+samply record wasmtime run --profile=perfmap --env HARDWARE_CONCURRENCY=$HARDWARE_CONCURRENCY -Wthreads=y -Sthreads=y --dir=.. $COMMAND

barretenberg/cpp/src/barretenberg/benchmark/ultra_bench/mock_circuits.hpp

@@ -54,9 +54,8 @@ Prover get_prover(void (*test_circuit_function)(typename Prover::Flavor::Circuit
         Composer composer;
         return composer.create_prover(builder);
     } else {
-#ifdef TRACY_MEMORY
-        ZoneScopedN("creating prover");
-#endif
+        PROFILE_THIS_NAME("creating prover");
+
         return Prover(builder);
     }
 };

barretenberg/cpp/src/barretenberg/client_ivc/test_bench_shared.hpp

@@ -44,7 +44,7 @@ void perform_ivc_accumulation_rounds(size_t NUM_CIRCUITS,
     for (size_t circuit_idx = 0; circuit_idx < NUM_CIRCUITS; ++circuit_idx) {
         MegaCircuitBuilder circuit;
         {
-            BB_OP_COUNT_TIME_NAME("construct_circuits");
+            PROFILE_THIS_NAME("construct_circuits");
             circuit = circuit_producer.create_next_circuit(ivc);
         }
 

barretenberg/cpp/src/barretenberg/commitment_schemes/commitment_key.hpp

@@ -85,7 +85,7 @@ template <class Curve> class CommitmentKey {
      */
     Commitment commit(PolynomialSpan<const Fr> polynomial)
     {
-        BB_OP_COUNT_TIME();
+        PROFILE_THIS();
         // We must have a power-of-2 SRS points *after* subtracting by start_index.
         const size_t consumed_srs = numeric::round_up_power_2(polynomial.size()) + polynomial.start_index;
         auto srs = srs::get_crs_factory<Curve>()->get_prover_crs(consumed_srs);
@@ -120,7 +120,7 @@ template <class Curve> class CommitmentKey {
      */
     Commitment commit_sparse(PolynomialSpan<const Fr> polynomial)
     {
-        BB_OP_COUNT_TIME();
+        PROFILE_THIS();
         const size_t poly_size = polynomial.size();
         ASSERT(polynomial.end_index() <= srs->get_monomial_size());
 

barretenberg/cpp/src/barretenberg/common/mem.cpp

@@ -1,6 +1,6 @@
 #include "barretenberg/common/mem.hpp"
 
-#ifdef TRACY_ENABLE
+#ifdef TRACY_MEMORY
 void* operator new(std::size_t count)
 {
     // NOLINTBEGIN(cppcoreguidelines-no-malloc)

barretenberg/cpp/src/barretenberg/common/op_count.hpp

@@ -3,6 +3,18 @@
 
 #include <memory>
 #include <tracy/Tracy.hpp>
+
+#ifdef BB_USE_OP_COUNT_TIME_ONLY
+#define PROFILE_THIS() BB_OP_COUNT_TIME_NAME(__func__)
+#define PROFILE_THIS_NAME(name) BB_OP_COUNT_TIME_NAME(name)
+#elif defined TRACY_INSTRUMENTED
+#define PROFILE_THIS() ZoneScopedN(__func__)
+#define PROFILE_THIS_NAME(name) ZoneScopedN(name)
+#else
+#define PROFILE_THIS() (void)0
+#define PROFILE_THIS_NAME(name) (void)0
+#endif
+
 #ifndef BB_USE_OP_COUNT
 // require a semicolon to appease formatters
 // NOLINTNEXTLINE(cppcoreguidelines-macro-usage)
@@ -12,18 +24,11 @@
 // NOLINTNEXTLINE(cppcoreguidelines-macro-usage)
 #define BB_OP_COUNT_CYCLES_NAME(name) (void)0
 // NOLINTNEXTLINE(cppcoreguidelines-macro-usage)
-#define BB_OP_COUNT_CYCLES() (void)0
-#ifndef TRACY_TIME
-// NOLINTNEXTLINE(cppcoreguidelines-macro-usage)
 #define BB_OP_COUNT_TIME_NAME(name) (void)0
 // NOLINTNEXTLINE(cppcoreguidelines-macro-usage)
-#define BB_OP_COUNT_TIME() (void)0
-#else
-// NOLINTNEXTLINE(cppcoreguidelines-macro-usage)
-#define BB_OP_COUNT_TIME_NAME(name) ZoneScopedN(name)
+#define BB_OP_COUNT_CYCLES() (void)0
 // NOLINTNEXTLINE(cppcoreguidelines-macro-usage)
-#define BB_OP_COUNT_TIME() BB_OP_COUNT_TIME_NAME(__func__)
-#endif
+#define BB_OP_COUNT_TIME() (void)0
 #else
 /**
  * Provides an abstraction that counts operations based on function names.

barretenberg/cpp/src/barretenberg/common/slab_allocator.cpp

@@ -1,7 +1,8 @@
 #include "slab_allocator.hpp"
-#include <barretenberg/common/assert.hpp>
-#include <barretenberg/common/log.hpp>
-#include <barretenberg/common/mem.hpp>
+#include "barretenberg/common/assert.hpp"
+#include "barretenberg/common/log.hpp"
+#include "barretenberg/common/mem.hpp"
+#include "barretenberg/common/op_count.hpp"
 #include <cstddef>
 #include <numeric>
 #include <unordered_map>
@@ -211,6 +212,8 @@ void init_slab_allocator(size_t circuit_subgroup_size)
 
 std::shared_ptr<void> get_mem_slab(size_t size)
 {
+    PROFILE_THIS();
+
     return allocator.get(size);
 }
 

barretenberg/cpp/src/barretenberg/ecc/groups/element_impl.hpp

@@ -792,7 +792,7 @@ template <class Fq, class Fr, class T>
 std::vector<affine_element<Fq, Fr, T>> element<Fq, Fr, T>::batch_mul_with_endomorphism(
     const std::span<const affine_element<Fq, Fr, T>>& points, const Fr& scalar) noexcept
 {
-    BB_OP_COUNT_TIME();
+    PROFILE_THIS();
     typedef affine_element<Fq, Fr, T> affine_element;
     const size_t num_points = points.size();
 

barretenberg/cpp/src/barretenberg/ecc/scalar_multiplication/runtime_states.cpp

@@ -40,6 +40,8 @@ pippenger_runtime_state<Curve>::pippenger_runtime_state(const size_t num_initial
     , bucket_empty_status(reinterpret_cast<bool*>(aligned_alloc(64, num_threads * num_buckets * sizeof(bool))))
     , round_counts(reinterpret_cast<uint64_t*>(aligned_alloc(32, MAX_NUM_ROUNDS * sizeof(uint64_t))))
 {
+    PROFILE_THIS();
+
     using Fq = typename Curve::BaseField;
     using AffineElement = typename Curve::AffineElement;
 
@@ -51,6 +53,7 @@ pippenger_runtime_state<Curve>::pippenger_runtime_state(const size_t num_initial
 
     const size_t points_per_thread = static_cast<size_t>(num_points) / num_threads;
     parallel_for(num_threads, [&](size_t i) {
+        PROFILE_THIS_NAME("memset in Pippenger runtime state creation");
         const size_t thread_offset = i * points_per_thread;
         memset(reinterpret_cast<void*>(point_pairs_1 + thread_offset + (i * 16)),
                0,
@@ -96,6 +99,8 @@ pippenger_runtime_state<Curve>::pippenger_runtime_state(pippenger_runtime_state&
     , round_counts(other.round_counts)
 
 {
+    PROFILE_THIS();
+
     other.point_schedule = nullptr;
     other.skew_table = nullptr;
     other.point_pairs_1 = nullptr;
@@ -111,6 +116,8 @@ template <typename Curve>
 pippenger_runtime_state<Curve>& pippenger_runtime_state<Curve>::operator=(
     pippenger_runtime_state<Curve>&& other) noexcept
 {
+    PROFILE_THIS();
+
     if (skew_table != nullptr) {
         aligned_free(skew_table);
     }
@@ -164,6 +171,8 @@ template <typename Curve>
 affine_product_runtime_state<Curve> pippenger_runtime_state<Curve>::get_affine_product_runtime_state(
     const size_t num_threads, const size_t thread_index)
 {
+    PROFILE_THIS();
+
     const auto points_per_thread = static_cast<size_t>(num_points / num_threads);
     const auto num_buckets =
         static_cast<size_t>(1U << scalar_multiplication::get_optimal_bucket_width(static_cast<size_t>(num_points) / 2));
@@ -181,6 +190,8 @@ affine_product_runtime_state<Curve> pippenger_runtime_state<Curve>::get_affine_p
 
 template <typename Curve> pippenger_runtime_state<Curve>::~pippenger_runtime_state() noexcept
 {
+    PROFILE_THIS();
+
     if (skew_table != nullptr) {
         aligned_free(skew_table);
     }