From b42035d8a4f81ab7b66095cefecbbb23e6dedd47 Mon Sep 17 00:00:00 2001
From: ludamad <adam@aztecprotocol.com>
Date: Mon, 16 Oct 2023 14:35:20 +0000
Subject: [PATCH 1/8] feat: gperftools

---
 barretenberg/cpp/.gitignore                   |  4 +-
 barretenberg/cpp/CMakePresets.json            | 52 ++++++++++++-------
 .../cpp/scripts/collect_heap_information.sh   | 41 +++++++++++++++
 3 files changed, 76 insertions(+), 21 deletions(-)
 create mode 100755 barretenberg/cpp/scripts/collect_heap_information.sh

diff --git a/barretenberg/cpp/.gitignore b/barretenberg/cpp/.gitignore
index 9a5226b72fa..811aef3620e 100644
--- a/barretenberg/cpp/.gitignore
+++ b/barretenberg/cpp/.gitignore
@@ -8,4 +8,6 @@ CMakeUserPresets.json
 .vscode/settings.json
 # to be unignored when we agree on clang-tidy rules
 .clangd
-acir_tests
\ No newline at end of file
+acir_tests
+# we may download go in scripts/collect_heap_information.sh
+go*.tar.gz
diff --git a/barretenberg/cpp/CMakePresets.json b/barretenberg/cpp/CMakePresets.json
index 6bafa8fed7a..bda74911d93 100644
--- a/barretenberg/cpp/CMakePresets.json
+++ b/barretenberg/cpp/CMakePresets.json
@@ -135,6 +135,18 @@
         "CMAKE_BUILD_TYPE": "Debug"
       }
     },
+    {
+      "name": "gperftools",
+      "displayName": "Debugging build with gperftools on Clang-16",
+      "description": "Build with gperf",
+      "inherits": "clang16-dbg",
+      "binaryDir": "build-gperftools",
+      "cacheVariables": {
+        "CMAKE_BUILD_TYPE": "RelWithDebInfo",
+        "CMAKE_EXE_LINKER_FLAGS": "-ltcmalloc",
+        "CXXFLAGS": "-fno-omit-frame-pointer"
+      }
+    },
     {
       "name": "wasm",
       "displayName": "Build for WASM",
@@ -183,22 +195,6 @@
         "MULTITHREADING": "ON"
       }
     },
-    {
-      "name": "xray-1thread",
-      "displayName": "Build with single-threaded XRay Profiling",
-      "description": "Build with Clang and enable single-threaded LLVM XRay for profiling",
-      "generator": "Unix Makefiles",
-      "inherits": "clang16",
-      "environment": {
-        "CFLAGS": "-fxray-instrument -fxray-instruction-threshold=10",
-        "CXXFLAGS": "-fxray-instrument -fxray-instruction-threshold=10",
-        "LDFLAGS": "-fxray-instrument -fxray-instruction-threshold=10"
-      },
-      "cacheVariables": {
-        "MULTITHREADING": "OFF"
-      },
-      "binaryDir": "build-xray-1thread"
-    },
     {
       "name": "xray",
       "displayName": "Build with multi-threaded XRay Profiling",
@@ -211,6 +207,17 @@
         "LDFLAGS": "-fxray-instrument -fxray-instruction-threshold=10"
       },
       "binaryDir": "build-xray"
+    },
+    {
+      "name": "xray-1thread",
+      "displayName": "Build with single-threaded XRay Profiling",
+      "description": "Build with Clang and enable single-threaded LLVM XRay for profiling",
+      "generator": "Unix Makefiles",
+      "inherits": "xray",
+      "cacheVariables": {
+        "MULTITHREADING": "OFF"
+      },
+      "binaryDir": "build-xray-1thread"
     }
   ],
   "buildPresets": [
@@ -265,6 +272,11 @@
       "inherits": "clang16",
       "configurePreset": "fuzzing"
     },
+    {
+      "name": "gperftools",
+      "inherits": "clang16",
+      "configurePreset": "gperftools"
+    },
     {
       "name": "smt-verification",
       "inherits": "clang16",
@@ -304,13 +316,13 @@
       "targets": ["barretenberg.wasm"]
     },
     {
-      "name": "xray-1thread",
-      "configurePreset": "xray-1thread",
+      "name": "xray",
+      "configurePreset": "xray",
       "inherits": "default"
     },
     {
-      "name": "xray",
-      "configurePreset": "xray",
+      "name": "xray-1thread",
+      "configurePreset": "xray-1thread",
       "inherits": "default"
     }
   ],
diff --git a/barretenberg/cpp/scripts/collect_heap_information.sh b/barretenberg/cpp/scripts/collect_heap_information.sh
new file mode 100755
index 00000000000..88e799529ec
--- /dev/null
+++ b/barretenberg/cpp/scripts/collect_heap_information.sh
@@ -0,0 +1,41 @@
+#!/bin/bash
+set -eu
+
+PRESET=gperftools
+ONLY_PROCESS=${1:-}
+
+# Move above script dir.
+cd $(dirname $0)/..
+
+# Configure and build with heap profiling preset.
+cmake --preset $PRESET
+cmake --build --preset $PRESET
+
+cd build-$PRESET
+
+if [ -z "$ONLY_PROCESS" ]; then
+  # Clear old heap profile data.
+  rm -f honk_bench_main_simple.heap.*
+
+  # Run application with heap profiling.
+HEAPPROFILE=./honk_bench_main_simple ./bin/honk_bench_main_simple
+fi
+
+# Download and install Go
+if [ ! -d ~/go ]; then
+  ARCHIVE=go1.21.3.linux-amd64.tar.gz
+  echo "Downloading and installing Go..."
+  curl -O https://golang.org/dl/$ARCHIVE
+  tar -C ~/ -xvf $ARCHIVE
+  rm $ARCHIVE
+  export PATH=$PATH:~/go/bin
+fi
+
+# Install pprof
+if [ ! -f ~/go/bin/pprof ]; then
+    echo "Installing pprof..."
+    ~/go/bin/go install github.com/google/pprof@latest
+fi
+
+# Process the heap profile with pprof
+~/go/bin/pprof --text ./bin/honk_bench_main_simple honk_bench_main_simple.0001.heap
\ No newline at end of file

From 5224ebffd39328b9e1a8ab669a80b191fa1a60a0 Mon Sep 17 00:00:00 2001
From: ludamad <adam@aztecprotocol.com>
Date: Mon, 16 Oct 2023 15:04:38 +0000
Subject: [PATCH 2/8] feat: revert bbmalloc using slabs

---
 barretenberg/cpp/src/barretenberg/common/mem.cpp | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/barretenberg/cpp/src/barretenberg/common/mem.cpp b/barretenberg/cpp/src/barretenberg/common/mem.cpp
index d119eaf24a4..979c10d11e9 100644
--- a/barretenberg/cpp/src/barretenberg/common/mem.cpp
+++ b/barretenberg/cpp/src/barretenberg/common/mem.cpp
@@ -4,10 +4,10 @@
 
 WASM_EXPORT void* bbmalloc(size_t size)
 {
-    return barretenberg::get_mem_slab_raw(size);
+    return aligned_alloc(32, size);
 }
 
 WASM_EXPORT void bbfree(void* ptr)
 {
-    barretenberg::free_mem_slab_raw(ptr);
+    aligned_free(ptr);
 }

From 128d3e6735824f6bf60d76fb1a5e7707e222cde4 Mon Sep 17 00:00:00 2001
From: ludamad <adam@aztecprotocol.com>
Date: Mon, 16 Oct 2023 19:24:05 +0000
Subject: [PATCH 3/8] REVERTME: temporary allocation changes

---
 barretenberg/cpp/scripts/collect_heap_information.sh      | 5 +++--
 .../cpp/src/barretenberg/common/slab_allocator.cpp        | 8 ++++----
 .../cpp/src/barretenberg/common/slab_allocator.hpp        | 8 +++++++-
 .../cpp/src/barretenberg/polynomials/polynomial.cpp       | 3 ++-
 4 files changed, 16 insertions(+), 8 deletions(-)

diff --git a/barretenberg/cpp/scripts/collect_heap_information.sh b/barretenberg/cpp/scripts/collect_heap_information.sh
index 88e799529ec..9a7407b2fe7 100755
--- a/barretenberg/cpp/scripts/collect_heap_information.sh
+++ b/barretenberg/cpp/scripts/collect_heap_information.sh
@@ -8,6 +8,7 @@ ONLY_PROCESS=${1:-}
 cd $(dirname $0)/..
 
 # Configure and build with heap profiling preset.
+
 cmake --preset $PRESET
 cmake --build --preset $PRESET
 
@@ -15,10 +16,10 @@ cd build-$PRESET
 
 if [ -z "$ONLY_PROCESS" ]; then
   # Clear old heap profile data.
-  rm -f honk_bench_main_simple.heap.*
+  rm -f honk_bench_main_simple.heap*
 
   # Run application with heap profiling.
-HEAPPROFILE=./honk_bench_main_simple ./bin/honk_bench_main_simple
+  HEAPPROFILE=./honk_bench_main_simple ./bin/honk_bench_main_simple
 fi
 
 # Download and install Go
diff --git a/barretenberg/cpp/src/barretenberg/common/slab_allocator.cpp b/barretenberg/cpp/src/barretenberg/common/slab_allocator.cpp
index 7b6d2020caf..bc0f9b7dd68 100644
--- a/barretenberg/cpp/src/barretenberg/common/slab_allocator.cpp
+++ b/barretenberg/cpp/src/barretenberg/common/slab_allocator.cpp
@@ -211,10 +211,10 @@ void init_slab_allocator(size_t circuit_subgroup_size)
 //     return 0;
 // })();
 
-std::shared_ptr<void> get_mem_slab(size_t size)
-{
-    return allocator.get(size);
-}
+// std::shared_ptr<void> get_mem_slab(size_t size)
+// {
+//     return allocator.get(size);
+// }
 
 void* get_mem_slab_raw(size_t size)
 {
diff --git a/barretenberg/cpp/src/barretenberg/common/slab_allocator.hpp b/barretenberg/cpp/src/barretenberg/common/slab_allocator.hpp
index 050360d4db5..f15e6385274 100644
--- a/barretenberg/cpp/src/barretenberg/common/slab_allocator.hpp
+++ b/barretenberg/cpp/src/barretenberg/common/slab_allocator.hpp
@@ -31,7 +31,13 @@ void init_slab_allocator(size_t circuit_subgroup_size);
  * Returns a slab from the preallocated pool of slabs, or fallback to a new heap allocation (32 byte aligned).
  * Ref counted result so no need to manually free.
  */
-std::shared_ptr<void> get_mem_slab(size_t size);
+// std::shared_ptr<void> get_mem_slab(size_t size);
+
+inline std::shared_ptr<void> get_mem_slab(size_t size)
+{
+    // NOLINTNEXTLINE(cppcoreguidelines-no-malloc)
+    return { malloc(size), free };
+}
 
 /**
  * Sometimes you want a raw pointer to a slab so you can manage when it's released manually (e.g. c_binds, containers).
diff --git a/barretenberg/cpp/src/barretenberg/polynomials/polynomial.cpp b/barretenberg/cpp/src/barretenberg/polynomials/polynomial.cpp
index e3f94e9c872..8509c717eb1 100644
--- a/barretenberg/cpp/src/barretenberg/polynomials/polynomial.cpp
+++ b/barretenberg/cpp/src/barretenberg/polynomials/polynomial.cpp
@@ -461,7 +461,8 @@ template <typename Fr> Polynomial<Fr> Polynomial<Fr>::partial_evaluate_mle(std::
     return result;
 }
 
-template <typename Fr> typename Polynomial<Fr>::pointer Polynomial<Fr>::allocate_aligned_memory(const size_t size) const
+template <typename Fr>
+BBERG_INLINE typename Polynomial<Fr>::pointer Polynomial<Fr>::allocate_aligned_memory(const size_t size) const
 {
     return std::static_pointer_cast<Fr[]>(get_mem_slab(size));
 }

From 1c07648f6da80dec9ab5c5aa2140b66b024ffd08 Mon Sep 17 00:00:00 2001
From: ludamad <adam@aztecprotocol.com>
Date: Tue, 17 Oct 2023 22:22:53 +0000
Subject: [PATCH 4/8] Memory tests

---
 barretenberg/cpp/scripts/collect_heap_information.sh |  6 +++++-
 .../benchmark/honk_bench/main.simple.cpp             |  4 ++--
 .../cpp/src/barretenberg/common/slab_allocator.cpp   | 12 ++++++------
 .../cpp/src/barretenberg/common/slab_allocator.hpp   |  6 +++++-
 4 files changed, 18 insertions(+), 10 deletions(-)

diff --git a/barretenberg/cpp/scripts/collect_heap_information.sh b/barretenberg/cpp/scripts/collect_heap_information.sh
index 9a7407b2fe7..92fdbc26298 100755
--- a/barretenberg/cpp/scripts/collect_heap_information.sh
+++ b/barretenberg/cpp/scripts/collect_heap_information.sh
@@ -38,5 +38,9 @@ if [ ! -f ~/go/bin/pprof ]; then
     ~/go/bin/go install github.com/google/pprof@latest
 fi
 
+# Collect the heap files
+files=(./honk_bench_main_simple.*.heap)
+# Find the middle index based on the count
+middle_index=$(( (${#files[@]} + 1) / 2 - 1))
 # Process the heap profile with pprof
-~/go/bin/pprof --text ./bin/honk_bench_main_simple honk_bench_main_simple.0001.heap
\ No newline at end of file
+~/go/bin/pprof --text ./bin/honk_bench_main_simple ${files[$middle_index]}
\ No newline at end of file
diff --git a/barretenberg/cpp/src/barretenberg/benchmark/honk_bench/main.simple.cpp b/barretenberg/cpp/src/barretenberg/benchmark/honk_bench/main.simple.cpp
index f33faf554d1..588e2727991 100644
--- a/barretenberg/cpp/src/barretenberg/benchmark/honk_bench/main.simple.cpp
+++ b/barretenberg/cpp/src/barretenberg/benchmark/honk_bench/main.simple.cpp
@@ -45,12 +45,12 @@ void construct_proof_ultra() noexcept
     barretenberg::srs::init_crs_factory("../srs_db/ignition");
     // Constuct circuit and prover; don't include this part in measurement
     auto builder = typename UltraHonk::CircuitBuilder();
-    generate_sha256_test_circuit<UltraBuilder>(builder, 1);
+    generate_sha256_test_circuit<UltraBuilder>(builder, 100);
 
     auto composer = UltraHonk();
     auto instance = composer.create_instance(builder);
     auto ext_prover = composer.create_prover(instance);
-    for (size_t i = 0; i < 10; i++) {
+    for (size_t i = 0; i < 1; i++) {
         auto proof = ext_prover.construct_proof();
     }
 }
diff --git a/barretenberg/cpp/src/barretenberg/common/slab_allocator.cpp b/barretenberg/cpp/src/barretenberg/common/slab_allocator.cpp
index bc0f9b7dd68..c123294d091 100644
--- a/barretenberg/cpp/src/barretenberg/common/slab_allocator.cpp
+++ b/barretenberg/cpp/src/barretenberg/common/slab_allocator.cpp
@@ -216,12 +216,12 @@ void init_slab_allocator(size_t circuit_subgroup_size)
 //     return allocator.get(size);
 // }
 
-void* get_mem_slab_raw(size_t size)
-{
-    auto slab = get_mem_slab(size);
-    manual_slabs[slab.get()] = slab;
-    return slab.get();
-}
+// void* get_mem_slab_raw(size_t size)
+// {
+//     auto slab = get_mem_slab(size);
+//     manual_slabs[slab.get()] = slab;
+//     return slab.get();
+// }
 
 void free_mem_slab_raw(void* p)
 {
diff --git a/barretenberg/cpp/src/barretenberg/common/slab_allocator.hpp b/barretenberg/cpp/src/barretenberg/common/slab_allocator.hpp
index f15e6385274..11c88961a91 100644
--- a/barretenberg/cpp/src/barretenberg/common/slab_allocator.hpp
+++ b/barretenberg/cpp/src/barretenberg/common/slab_allocator.hpp
@@ -43,7 +43,11 @@ inline std::shared_ptr<void> get_mem_slab(size_t size)
  * Sometimes you want a raw pointer to a slab so you can manage when it's released manually (e.g. c_binds, containers).
  * This still gets a slab with a shared_ptr, but holds the shared_ptr internally until free_mem_slab_raw is called.
  */
-void* get_mem_slab_raw(size_t size);
+inline void* get_mem_slab_raw(size_t size)
+{
+    // NOLINTNEXTLINE(cppcoreguidelines-no-malloc)
+    return malloc(size);
+}
 
 void free_mem_slab_raw(void*);
 

From 488fc0e7e05f3b470db241a7e86ae08e04e65e77 Mon Sep 17 00:00:00 2001
From: ludamad <adam@aztecprotocol.com>
Date: Fri, 27 Oct 2023 17:24:27 +0000
Subject: [PATCH 5/8] feat: update

---
 barretenberg/cpp/CMakePresets.json | 12 ------------
 1 file changed, 12 deletions(-)

diff --git a/barretenberg/cpp/CMakePresets.json b/barretenberg/cpp/CMakePresets.json
index 0c051c88253..85193274ed8 100644
--- a/barretenberg/cpp/CMakePresets.json
+++ b/barretenberg/cpp/CMakePresets.json
@@ -221,12 +221,6 @@
       "binaryDir": "build-xray"
     },
     {
-<<<<<<< HEAD
-      "name": "xray-1thread",
-      "displayName": "Build with single-threaded XRay Profiling",
-      "description": "Build with Clang and enable single-threaded LLVM XRay for profiling",
-      "generator": "Unix Makefiles",
-=======
       "name": "xray-verbose",
       "displayName": "Build with detailed XRay Profiling",
       "description": "Build with Clang and enable detailed LLVM XRay for profiling",
@@ -242,7 +236,6 @@
       "name": "xray-1thread",
       "displayName": "Build with single-threaded XRay Profiling",
       "description": "Build with Clang and enable single-threaded LLVM XRay for profiling",
->>>>>>> origin/master
       "inherits": "xray",
       "cacheVariables": {
         "MULTITHREADING": "OFF"
@@ -351,10 +344,6 @@
       "targets": ["barretenberg.wasm"]
     },
     {
-<<<<<<< HEAD
-      "name": "xray",
-      "configurePreset": "xray",
-=======
       "name": "xray-verbose",
       "configurePreset": "xray-verbose",
       "inherits": "default"
@@ -362,7 +351,6 @@
     {
       "name": "xray-1thread",
       "configurePreset": "xray-1thread",
->>>>>>> origin/master
       "inherits": "default"
     },
     {

From 1faa038c204965fc5a56a76eb4971664faabb694 Mon Sep 17 00:00:00 2001
From: ludamad <adam@aztecprotocol.com>
Date: Fri, 27 Oct 2023 17:34:43 +0000
Subject: [PATCH 6/8] fixes post merge

---
 barretenberg/cpp/CMakePresets.json | 7 +------
 1 file changed, 1 insertion(+), 6 deletions(-)

diff --git a/barretenberg/cpp/CMakePresets.json b/barretenberg/cpp/CMakePresets.json
index 85193274ed8..0174d8b873b 100644
--- a/barretenberg/cpp/CMakePresets.json
+++ b/barretenberg/cpp/CMakePresets.json
@@ -151,7 +151,7 @@
       "name": "gperftools",
       "displayName": "Debugging build with gperftools on Clang-16",
       "description": "Build with gperf",
-      "inherits": "clang16-dbg",
+      "inherits": "clang16",
       "binaryDir": "build-gperftools",
       "cacheVariables": {
         "CMAKE_BUILD_TYPE": "RelWithDebInfo",
@@ -348,11 +348,6 @@
       "configurePreset": "xray-verbose",
       "inherits": "default"
     },
-    {
-      "name": "xray-1thread",
-      "configurePreset": "xray-1thread",
-      "inherits": "default"
-    },
     {
       "name": "xray-1thread",
       "configurePreset": "xray-1thread",

From cae72061fe97002cfcb9be1c2214da5dc5155f0b Mon Sep 17 00:00:00 2001
From: ludamad <adam@aztecprotocol.com>
Date: Fri, 27 Oct 2023 18:33:30 +0000
Subject: [PATCH 7/8] Revert code. Fix script

---
 .../cpp/scripts/collect_heap_information.sh   | 11 +++++-----
 .../cpp/src/barretenberg/common/mem.cpp       |  4 ++--
 .../barretenberg/common/slab_allocator.cpp    | 22 +++++++++----------
 .../barretenberg/common/slab_allocator.hpp    | 14 ++----------
 .../barretenberg/polynomials/polynomial.cpp   |  3 +--
 5 files changed, 22 insertions(+), 32 deletions(-)

diff --git a/barretenberg/cpp/scripts/collect_heap_information.sh b/barretenberg/cpp/scripts/collect_heap_information.sh
index 92fdbc26298..74bb18b2652 100755
--- a/barretenberg/cpp/scripts/collect_heap_information.sh
+++ b/barretenberg/cpp/scripts/collect_heap_information.sh
@@ -3,6 +3,7 @@ set -eu
 
 PRESET=gperftools
 ONLY_PROCESS=${1:-}
+EXECUTABLE=${2:-ultra_honk_rounds_bench}
 
 # Move above script dir.
 cd $(dirname $0)/..
@@ -16,10 +17,10 @@ cd build-$PRESET
 
 if [ -z "$ONLY_PROCESS" ]; then
   # Clear old heap profile data.
-  rm -f honk_bench_main_simple.heap*
+  rm -f $EXECUTABLE.heap*
 
-  # Run application with heap profiling.
-  HEAPPROFILE=./honk_bench_main_simple ./bin/honk_bench_main_simple
+  # Run application with heap profiling to a file with prefix '$EXECUTABLE'.
+  HEAPPROFILE=./$EXECUTABLE ./bin/$EXECUTABLE
 fi
 
 # Download and install Go
@@ -39,8 +40,8 @@ if [ ! -f ~/go/bin/pprof ]; then
 fi
 
 # Collect the heap files
-files=(./honk_bench_main_simple.*.heap)
+files=(./$EXECUTABLE.*.heap)
 # Find the middle index based on the count
 middle_index=$(( (${#files[@]} + 1) / 2 - 1))
 # Process the heap profile with pprof
-~/go/bin/pprof --text ./bin/honk_bench_main_simple ${files[$middle_index]}
\ No newline at end of file
+~/go/bin/pprof --text ./bin/$EXECUTABLE ${files[$middle_index]}
\ No newline at end of file
diff --git a/barretenberg/cpp/src/barretenberg/common/mem.cpp b/barretenberg/cpp/src/barretenberg/common/mem.cpp
index 979c10d11e9..d119eaf24a4 100644
--- a/barretenberg/cpp/src/barretenberg/common/mem.cpp
+++ b/barretenberg/cpp/src/barretenberg/common/mem.cpp
@@ -4,10 +4,10 @@
 
 WASM_EXPORT void* bbmalloc(size_t size)
 {
-    return aligned_alloc(32, size);
+    return barretenberg::get_mem_slab_raw(size);
 }
 
 WASM_EXPORT void bbfree(void* ptr)
 {
-    aligned_free(ptr);
+    barretenberg::free_mem_slab_raw(ptr);
 }
diff --git a/barretenberg/cpp/src/barretenberg/common/slab_allocator.cpp b/barretenberg/cpp/src/barretenberg/common/slab_allocator.cpp
index c123294d091..7b6d2020caf 100644
--- a/barretenberg/cpp/src/barretenberg/common/slab_allocator.cpp
+++ b/barretenberg/cpp/src/barretenberg/common/slab_allocator.cpp
@@ -211,17 +211,17 @@ void init_slab_allocator(size_t circuit_subgroup_size)
 //     return 0;
 // })();
 
-// std::shared_ptr<void> get_mem_slab(size_t size)
-// {
-//     return allocator.get(size);
-// }
-
-// void* get_mem_slab_raw(size_t size)
-// {
-//     auto slab = get_mem_slab(size);
-//     manual_slabs[slab.get()] = slab;
-//     return slab.get();
-// }
+std::shared_ptr<void> get_mem_slab(size_t size)
+{
+    return allocator.get(size);
+}
+
+void* get_mem_slab_raw(size_t size)
+{
+    auto slab = get_mem_slab(size);
+    manual_slabs[slab.get()] = slab;
+    return slab.get();
+}
 
 void free_mem_slab_raw(void* p)
 {
diff --git a/barretenberg/cpp/src/barretenberg/common/slab_allocator.hpp b/barretenberg/cpp/src/barretenberg/common/slab_allocator.hpp
index 11c88961a91..050360d4db5 100644
--- a/barretenberg/cpp/src/barretenberg/common/slab_allocator.hpp
+++ b/barretenberg/cpp/src/barretenberg/common/slab_allocator.hpp
@@ -31,23 +31,13 @@ void init_slab_allocator(size_t circuit_subgroup_size);
  * Returns a slab from the preallocated pool of slabs, or fallback to a new heap allocation (32 byte aligned).
  * Ref counted result so no need to manually free.
  */
-// std::shared_ptr<void> get_mem_slab(size_t size);
-
-inline std::shared_ptr<void> get_mem_slab(size_t size)
-{
-    // NOLINTNEXTLINE(cppcoreguidelines-no-malloc)
-    return { malloc(size), free };
-}
+std::shared_ptr<void> get_mem_slab(size_t size);
 
 /**
  * Sometimes you want a raw pointer to a slab so you can manage when it's released manually (e.g. c_binds, containers).
  * This still gets a slab with a shared_ptr, but holds the shared_ptr internally until free_mem_slab_raw is called.
  */
-inline void* get_mem_slab_raw(size_t size)
-{
-    // NOLINTNEXTLINE(cppcoreguidelines-no-malloc)
-    return malloc(size);
-}
+void* get_mem_slab_raw(size_t size);
 
 void free_mem_slab_raw(void*);
 
diff --git a/barretenberg/cpp/src/barretenberg/polynomials/polynomial.cpp b/barretenberg/cpp/src/barretenberg/polynomials/polynomial.cpp
index 02ce5dcd011..727575b7d56 100644
--- a/barretenberg/cpp/src/barretenberg/polynomials/polynomial.cpp
+++ b/barretenberg/cpp/src/barretenberg/polynomials/polynomial.cpp
@@ -485,8 +485,7 @@ template <typename Fr> Polynomial<Fr> Polynomial<Fr>::partial_evaluate_mle(std::
     return result;
 }
 
-template <typename Fr>
-BBERG_INLINE typename Polynomial<Fr>::pointer Polynomial<Fr>::allocate_aligned_memory(const size_t size) const
+template <typename Fr> typename Polynomial<Fr>::pointer Polynomial<Fr>::allocate_aligned_memory(const size_t size) const
 {
     return std::static_pointer_cast<Fr[]>(get_mem_slab(size));
 }

From 77b31519036f2988b29b563f2a1c5aeb565e43af Mon Sep 17 00:00:00 2001
From: ludamad <adam.domurad@gmail.com>
Date: Thu, 2 Nov 2023 12:55:50 -0500
Subject: [PATCH 8/8] fix: collect_heap_information.sh

---
 barretenberg/cpp/scripts/collect_heap_information.sh | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/barretenberg/cpp/scripts/collect_heap_information.sh b/barretenberg/cpp/scripts/collect_heap_information.sh
index 74bb18b2652..1d25c5a791c 100755
--- a/barretenberg/cpp/scripts/collect_heap_information.sh
+++ b/barretenberg/cpp/scripts/collect_heap_information.sh
@@ -27,7 +27,7 @@ fi
 if [ ! -d ~/go ]; then
   ARCHIVE=go1.21.3.linux-amd64.tar.gz
   echo "Downloading and installing Go..."
-  curl -O https://golang.org/dl/$ARCHIVE
+  wget https://go.dev/dl/$ARCHIVE
   tar -C ~/ -xvf $ARCHIVE
   rm $ARCHIVE
   export PATH=$PATH:~/go/bin
@@ -44,4 +44,4 @@ files=(./$EXECUTABLE.*.heap)
 # Find the middle index based on the count
 middle_index=$(( (${#files[@]} + 1) / 2 - 1))
 # Process the heap profile with pprof
-~/go/bin/pprof --text ./bin/$EXECUTABLE ${files[$middle_index]}
\ No newline at end of file
+~/go/bin/pprof --text ./bin/$EXECUTABLE ${files[$middle_index]}