feat(ci): tracy gate counter preset

barretenberg/cpp/CMakePresets.json

@@ -85,14 +85,25 @@
     },
     {
       "name": "tracy",
-      "displayName": "Release build with tracy",
-      "description": "Release build with tracy",
+      "displayName": "Release build with tracy, optimized for memory tracking",
+      "description": "Release build with tracy, optimized for memory tracking",
       "inherits": "clang16",
       "binaryDir": "build-tracy",
       "cacheVariables": {
         "ENABLE_TRACY": "ON"
       }
     },
+    {
+      "name": "tracy-gates",
+      "displayName": "Release build with tracy - but hacked for gate tracking",
+      "description": "Release build with tracy - but hacker for gate tracking",
+      "inherits": "clang16",
+      "binaryDir": "build-tracy-gates",
+      "cacheVariables": {
+        "ENABLE_TRACY": "ON",
+        "CMAKE_CXX_FLAGS": "-DTRACY_HACK_GATES_AS_MEMORY"
+      }
+    },
     {
       "name": "wasm-tracy",
       "displayName": "WASM build with tracy",
@@ -435,6 +446,11 @@
       "inherits": "default",
       "configurePreset": "tracy"
     },
+    {
+      "name": "tracy-gates",
+      "inherits": "default",
+      "configurePreset": "tracy-gates"
+    },
     {
       "name": "clang16-dbg-fast",
       "inherits": "default",

barretenberg/cpp/scripts/benchmark_tracy.sh

@@ -1,31 +1,40 @@
 
 # NOTE: intended to be ran from one's external computer, connecting to Aztec mainframe
+# IF ON YOUR LOCAL COMPUTER USE NORMAL INTERACTIVE TRACY WORKFLOW
 # the benchmark runs with headless capture and then we copy the trace file and run tracy profiler
 # This is thus only really useful internally at Aztec, sorry external folks. It can be easily tweaked
-# however for any SSH setup, especially an ubuntu one, and of course if you are just tracing on the
-# same machine you can use the normal interactive tracy workflow.
+# however for any SSH setup, especially an ubuntu one.
+# on local machine run:
+# export USER=...
+# export PRESET=...tracy for memory or tracy-gates for circuit gates...
+# ssh $USER-box "cat ~/aztec-packages/barretenberg/cpp/scripts/benchmark_tracy.sh" | bash /dev/stdin $USER
 set -eux
-USER=$1
+USER=${1:-$USER}
 BOX=$USER-box
-BENCHMARK=${2:-ultra_plonk_bench}
-COMMAND=${3:-./bin/$BENCHMARK}
+BENCHMARK=${2:-client_ivc_bench}
+COMMAND=${3:-./bin/$BENCHMARK --benchmark_filter=ClientIVCBench/Full/6"\$"}
+
+# Can also set PRESET=tracy-gates env variable
+PRESET=${PRESET:-tracy}
 
 ssh $BOX "
 	set -eux ;
+	cd ~/aztec-packages/barretenberg/cpp/ ;
+	cmake --preset $PRESET && cmake --build --preset $PRESET --target $BENCHMARK ;
 	! [ -d ~/tracy ] && git clone https://github.com/wolfpld/tracy ~/tracy ;
 	cd ~/tracy/capture ;
-	sudo apt-get install libdbus-1-dev libdbus-glib-1-dev ;
+        git checkout 075395620a504c0cdcaf9bab3d196db16a043de7 ;
+	sudo apt-get install -y libdbus-1-dev libdbus-glib-1-dev ;
 	mkdir -p build && cd build && cmake .. && make -j ;
 	./tracy-capture -a 127.0.0.1 -f -o trace-$BENCHMARK & ;
 	sleep 0.1 ;
-	cd ~/aztec-packages/barretenberg/cpp/ ;
-	cmake --preset tracy && cmake --build --preset tracy --parallel $BENCHMARK ;
-	cd build-tracy ;
-	ninja $BENCHMARK ;
+	cd ~/aztec-packages/barretenberg/cpp/build-$PRESET ;
 	$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 --build profiler/build --parallel
 scp $BOX:/mnt/user-data/$USER/tracy/capture/build/trace-$BENCHMARK .

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

@@ -1,5 +1,6 @@
 #include "barretenberg/common/mem.hpp"
 
+#ifdef TRACY_ENABLE
 void* operator new(std::size_t count)
 {
     // NOLINTBEGIN(cppcoreguidelines-no-malloc)
@@ -25,3 +26,5 @@ void operator delete(void* ptr, std::size_t size) noexcept
     free(ptr);
     // NOLINTEND(cppcoreguidelines-no-malloc)
 }
+
+#endif

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

@@ -7,9 +7,31 @@
 #include <memory>
 
 // This can be altered to capture stack traces, though more expensive
-// This is the only reason we wrap TracyAlloc or TracyAllocS
+// so wrap TracyAlloc or TracyAllocS. We disable these if gates are being tracked
+// Gates are hackishly tracked as if they were memory, for the sweet sweet memory
+// stack tree that doesn't seem to be available for other metric types.
+#ifndef TRACY_HACK_GATES_AS_MEMORY
 #define TRACY_ALLOC(t, size) TracyAllocS(t, size, /*stack depth*/ 10)
 #define TRACY_FREE(t) TracyFreeS(t, /*stack depth*/ 10)
+#define TRACY_GATE_ALLOC(t)
+#define TRACY_GATE_FREE(t)
+#else
+#include <mutex>
+#include <set>
+#define TRACY_ALLOC(t, size)
+#define TRACY_FREE(t)
+
+namespace bb {
+// These are hacks to make sure tracy plays along
+// If we free an ID not allocated, or allocate an index twice without a free it will complain
+// so we hack thread-safety and an incrementing global ID.
+static std::mutex GLOBAL_GATE_MUTEX;
+static size_t GLOBAL_GATE = 0;
+static std::set<size_t> FREED_GATES; // hack to prevent instrumentation failures
+} // namespace bb
+#define TRACY_GATE_ALLOC(index) TracyAllocS(reinterpret_cast<void*>(index), 1, /*stack depth*/ 50)
+#define TRACY_GATE_FREE(index) TracyFreeS(reinterpret_cast<void*>(index), /*stack depth*/ 50)
+#endif
 // #define TRACY_ALLOC(t, size) TracyAlloc(t, size)
 // #define TRACY_FREE(t) TracyFree(t)
 
@@ -108,4 +130,4 @@ inline void tracy_free(void* mem)
     TRACY_FREE(mem);
     // NOLINTNEXTLINE(cppcoreguidelines-owning-memory, cppcoreguidelines-no-malloc)
     free(mem);
-}
+}

barretenberg/cpp/src/barretenberg/plonk_honk_shared/arithmetization/arithmetization.hpp

@@ -1,4 +1,5 @@
 #pragma once
+#include "barretenberg/common/mem.hpp"
 #include "barretenberg/common/ref_array.hpp"
 #include "barretenberg/common/slab_allocator.hpp"
 #include <cstddef>
@@ -49,6 +50,18 @@ template <typename FF, size_t NUM_WIRES, size_t NUM_SELECTORS> class ExecutionTr
     // If enabled, we keep slow stack traces to be able to correlate gates with code locations where they were added
     StackTraces stack_traces;
 #endif
+#ifdef TRACY_HACK_GATES_AS_MEMORY
+    std::vector<size_t> allocated_gates;
+#endif
+    void tracy_gate()
+    {
+#ifdef TRACY_HACK_GATES_AS_MEMORY
+        std::unique_lock<std::mutex> lock(GLOBAL_GATE_MUTEX);
+        GLOBAL_GATE++;
+        TRACY_GATE_ALLOC(GLOBAL_GATE);
+        allocated_gates.push_back(GLOBAL_GATE);
+#endif
+    }
 
     Wires wires; // vectors of indices into a witness variables array
     Selectors selectors;
@@ -76,6 +89,18 @@ template <typename FF, size_t NUM_WIRES, size_t NUM_SELECTORS> class ExecutionTr
 
     uint32_t get_fixed_size() const { return fixed_size; }
     void set_fixed_size(uint32_t size_in) { fixed_size = size_in; }
+#ifdef TRACY_HACK_GATES_AS_MEMORY
+    ~ExecutionTraceBlock()
+    {
+        std::unique_lock<std::mutex> lock(GLOBAL_GATE_MUTEX);
+        for ([[maybe_unused]] size_t gate : allocated_gates) {
+            if (!FREED_GATES.contains(gate)) {
+                TRACY_GATE_FREE(gate);
+                FREED_GATES.insert(gate);
+            }
+        }
+    }
+#endif
 };
 
 } // namespace bb

barretenberg/cpp/src/barretenberg/plonk_honk_shared/arithmetization/mega_arithmetization.hpp

@@ -124,6 +124,7 @@ template <typename FF_> class MegaArith {
 #ifdef CHECK_CIRCUIT_STACKTRACES
             this->stack_traces.populate();
 #endif
+            this->tracy_gate();
             this->wires[0].emplace_back(idx_1);
             this->wires[1].emplace_back(idx_2);
             this->wires[2].emplace_back(idx_3);

barretenberg/cpp/src/barretenberg/plonk_honk_shared/arithmetization/standard_arithmetization.hpp

@@ -17,6 +17,7 @@ template <typename FF_> class StandardArith {
 #ifdef CHECK_CIRCUIT_STACKTRACES
             this->stack_traces.populate();
 #endif
+            this->tracy_gate();
             this->wires[0].emplace_back(idx_1);
             this->wires[1].emplace_back(idx_2);
             this->wires[2].emplace_back(idx_3);

barretenberg/cpp/src/barretenberg/plonk_honk_shared/arithmetization/ultra_arithmetization.hpp

@@ -63,6 +63,7 @@ template <typename FF_> class UltraArith {
 #ifdef CHECK_CIRCUIT_STACKTRACES
             this->stack_traces.populate();
 #endif
+            this->tracy_gate();
             this->wires[0].emplace_back(idx_1);
             this->wires[1].emplace_back(idx_2);
             this->wires[2].emplace_back(idx_3);