Remove internal Profiler in favour of Tracy

.gitmodules

@@ -29,3 +29,6 @@
 [submodule "ext/units"]
 	path = ext/units
 	url = https://github.com/LLNL/units.git
+[submodule "ext/tracy"]
+	path = ext/tracy
+	url = https://github.com/wolfpld/tracy.git

CMakeLists.txt

@@ -21,7 +21,7 @@ set(CMAKE_POSITION_INDEPENDENT_CODE ON)
 check_ipo_supported(RESULT HAVE_LTO OUTPUT ERR_LTO)
 if(NOT DEFINED CMAKE_INTERPROCEDURAL_OPTIMIZATION)
   if(HAVE_LTO)
-    message (STATUS "LTO support found, enabling")
+    message (VERBOSE "LTO support found, enabling")
     set(CMAKE_INTERPROCEDURAL_OPTIMIZATION TRUE)
   else()
     message(STATUS "No LTO: ${ERR_LTO}")
@@ -32,6 +32,7 @@ endif()
 # redirected from the terminal (e.g. when using ninja or a pager).
 
 option(ARBDEV_COLOR "Always produce ANSI-colored output (GNU/Clang only)." OFF)
+mark_as_advanced(FORCE ARBDEV_COLOR)
 
 #----------------------------------------------------------
 # Configure-time build options for Arbor:
@@ -53,10 +54,12 @@ mark_as_advanced(ARB_USE_HWLOC)
 # Use externally built modcc?
 
 set(ARB_MODCC "" CACHE STRING "path to external modcc NMODL compiler")
+mark_as_advanced(FORCE ARB_MODCC)
 
 # Use libunwind to generate stack traces on errors?
 
 option(ARB_BACKTRACE "Enable stacktraces on assertion and exceptions (requires Boost)." OFF)
+mark_as_advanced(FORCE ARB_BACKTRACE)
 
 # Specify GPU build type
 
@@ -77,8 +80,9 @@ option(ARB_USE_BUNDLED_LIBS "Use bundled 3rd party libraries" OFF)
 # CPU architecture. Recorded in installed target, for downstream dependencies
 # to use.
 # Useful, for example, when a user wants to compile with target-specific
-# optimization flags.
+# optimization flag.spr
 set(ARB_CXX_FLAGS_TARGET "" CACHE STRING "Optional additional flags for compilation")
+mark_as_advanced(FORCE ARB_CXX_FLAGS_TARGET)
 
 #----------------------------------------------------------
 # Debug support
@@ -94,7 +98,10 @@ mark_as_advanced(ARB_CAT_VERBOSE)
 
 option(ARB_WITH_MPI "build with MPI support" OFF)
 
-option(ARB_WITH_PROFILING "use built-in profiling" OFF)
+option(ARB_WITH_PROFILING "enable Tracy profiling" OFF)
+cmake_dependent_option(ARB_WITH_STACK_PROFILING "enable stack collection in profiling" OFF "ARB_WITH_PROFILING" OFF)
+cmake_dependent_option(ARB_WITH_MEMORY_PROFILING "enable memory in profiling" OFF "ARB_WITH_PROFILING" OFF)
+mark_as_advanced(FORCE ARB_WITH_STACK_PROFILING ARB_WITH_MEMORY_PROFILING)
 
 option(ARB_WITH_ASSERTIONS "enable arb_assert() assertions in code" OFF)
 
@@ -266,7 +273,7 @@ set(ARB_USE_BUNDLED_UNITS ON CACHE STRING "Use bundled LLNL units.")
 cmake_dependent_option(ARB_USE_BUNDLED_JSON "Use bundled Niels Lohmann's json library." ON "ARB_USE_BUNDLED_LIBS" OFF)
 if(NOT ARB_USE_BUNDLED_JSON)
   find_package(nlohmann_json 3.11.2 CONFIG REQUIRED)
-  message(STATUS "Using external JSON = ${nlohmann_json_VERSION}")
+  message(VERBOSE "Using external JSON = ${nlohmann_json_VERSION}")
 endif()
 
 cmake_dependent_option(ARB_USE_BUNDLED_RANDOM123 "Use bundled Random123 lib." ON "ARB_USE_BUNDLED_LIBS" OFF)
@@ -300,14 +307,27 @@ else()
     message(FATAL, "TODO: At the time of Arbor 0.10.0 there is no Spack package")
 endif()
 
-
+# Reduce CMAKE Log clutter
+set(_saved_CMAKE_MESSAGE_LOG_LEVEL ${CMAKE_MESSAGE_LOG_LEVEL})
+set(CMAKE_MESSAGE_LOG_LEVEL WARNING)
 add_subdirectory(ext)
+set(CMAKE_MESSAGE_LOG_LEVEL ${_saved_CMAKE_MESSAGE_LOG_LEVEL})
+
 install(TARGETS ext-hwloc EXPORT arbor-targets)
 install(TARGETS ext-random123 EXPORT arbor-targets)
 target_link_libraries(arbor-public-deps INTERFACE ext-units)
 install(TARGETS ext-units EXPORT arbor-targets)
 install(TARGETS units compile_flags_target EXPORT arbor-targets)
 
+# Hide dependency options
+mark_as_advanced(FORCE JSON_CI JSON_Diagnostics JSON_DisableEnumSerialization JSON_GlobalUDLs JSON_ImplicitConversions JSON_Install JSON_LegacyDiscardedValueComparison JSON_MultipleHeaders JSON_SystemInclude)
+mark_as_advanced(FORCE TRACY_ENABLE TRACY_ON_DEMAND TRACY_CALLSTACK TRACY_NO_CALLSTACK TRACY_NO_CALLSTACK_INLINES TRACY_ONLY_LOCALHOST TRACY_NO_BROADCAST TRACY_ONLY_IPV4 TRACY_NO_CODE_TRANSFER TRACY_NO_CONTEXT_SWITCH TRACY_NO_EXIT TRACY_NO_SAMPLING TRACY_NO_VERIFY TRACY_NO_VSYNC_CAPTURE TRACY_NO_FRAME_IMAGE TRACY_NO_SYSTEM_TRACING TRACY_PATCHABLE_NOPSLEDS TRACY_DELAYED_INIT TRACY_MANUAL_LIFETIME TRACY_FIBERS TRACY_NO_CRASH_HANDLER TRACY_TIMER_FALLBACK TRACY_CALLSTACK TRACY_DELAYED_INIT TRACY_ENABLE TRACY_FIBERS TRACY_MANUAL_LIFETIME TRACY_NO_BROADCAST TRACY_NO_CALLSTACK TRACY_NO_CALLSTACK_INLINES TRACY_NO_CODE_TRANSFER TRACY_NO_CONTEXT_SWITCH TRACY_NO_CRASH_HANDLER TRACY_NO_EXIT TRACY_NO_FRAME_IMAGE TRACY_NO_SAMPLING TRACY_NO_SYSTEM_TRACING TRACY_NO_VERIFY TRACY_NO_VSYNC_CAPTURE TRACY_ONLY_IPV4 TRACY_ONLY_LOCALHOST TRACY_ON_DEMAND TRACY_PATCHABLE_NOPSLEDS TRACY_STATIC TRACY_TIMER_FALLBACK)
+mark_as_advanced(FORCE ARB_USE_BUNDLED_FMT ARB_USE_BUNDLED_GTEST ARB_USE_BUNDLED_JSON ARB_USE_BUNDLED_PUGIXML ARB_USE_BUNDLED_PYBIND11 ARB_USE_BUNDLED_RANDOM123 ARB_USE_BUNDLED_UNITS)
+
+# hide OSX advance config
+mark_as_advanced(FORCE CMAKE_OSX_ARCHITECTURES CMAKE_OSX_DEPLOYMENT_TARGET CMAKE_OSX_SYSROOT)
+
+
 # Keep track of packages we need to add to the generated CMake config
 # file for arbor.
 
@@ -339,10 +359,10 @@ mark_as_advanced(ARB_SVE_WIDTH)
 if (ARB_SVE_WIDTH STREQUAL "auto")
     get_sve_length(ARB_HAS_SVE ARB_SVE_BITS)
     if (ARB_HAS_SVE)
-        message(STATUS "SVE detected with vector size = ${ARB_SVE_BITS} bits")
+        message(VERBOSE "SVE detected with vector size = ${ARB_SVE_BITS} bits")
         set(ARB_CXX_SVE_FLAGS " -msve-vector-bits=${ARB_SVE_BITS}")
     else()
-        message(STATUS "NO SVE detected")
+        message(VERBOSE "NO SVE detected")
         set(ARB_CXX_SVE_FLAGS "")
     endif()
 else()
@@ -369,7 +389,24 @@ target_compile_options(arborio-private-deps INTERFACE ${ARB_CXX_FLAGS_TARGET_FUL
 
 if(ARB_WITH_PROFILING)
     target_compile_definitions(arbor-config-defs INTERFACE ARB_HAVE_PROFILING)
+    option(TRACY_ENABLE "" ON)
+    option(TRACY_ON_DEMAND "" ON)
+
+    set(_saved_CMAKE_MESSAGE_LOG_LEVEL ${CMAKE_MESSAGE_LOG_LEVEL})
+    set(CMAKE_MESSAGE_LOG_LEVEL WARNING)
+    add_subdirectory (ext/tracy) # target: TracyClient or alias Tracy :: TracyClient
+    set(CMAKE_MESSAGE_LOG_LEVEL ${_saved_CMAKE_MESSAGE_LOG_LEVEL})
+
+    target_link_libraries (arbor-private-deps INTERFACE Tracy::TracyClient)
+    list(APPEND arbor_export_dependencies "Tracy::TracyClient")
+    if(ARB_WITH_STACK_PROFILING)
+        target_compile_definitions(arbor-config-defs INTERFACE ARB_HAVE_STACK_PROFILING)
+    endif()
+    if(ARB_WITH_MEMORY_PROFILING)
+        target_compile_definitions(arbor-config-defs INTERFACE ARB_HAVE_MEMORY_PROFILING)
+    endif()
 endif()
+
 if(ARB_WITH_ASSERTIONS)
     target_compile_definitions(arbor-config-defs INTERFACE ARB_HAVE_ASSERTIONS)
 endif()
@@ -401,7 +438,7 @@ endif()
 
 if(${Python3_FOUND})
     set(PYTHON_EXECUTABLE "${Python3_EXECUTABLE}")
-    message(STATUS "PYTHON_EXECUTABLE: ${PYTHON_EXECUTABLE}")
+    message(VERBOSE "PYTHON_EXECUTABLE: ${PYTHON_EXECUTABLE}")
 endif()
 
 

arbor/CMakeLists.txt

@@ -27,6 +27,7 @@ set(arbor_sources
     label_resolution.cpp
     lif_cell_group.cpp
     cable_cell_group.cpp
+    mechanism.cpp
     mechcat.cpp
     mechinfo.cpp
     memory/gpu_wrappers.cpp
@@ -50,7 +51,6 @@ set(arbor_sources
     profile/memory_meter.cpp
     profile/meter_manager.cpp
     profile/power_meter.cpp
-    profile/profiler.cpp
     schedule.cpp
     spike_event_io.cpp
     spike_source_cell_group.cpp
@@ -122,7 +122,10 @@ install(TARGETS arbor-private-headers EXPORT arbor-targets)
 # variable, build_all_mods target. Note: CMake source file properties are
 # directory-local.
 
+set(_saved_CMAKE_MESSAGE_LOG_LEVEL ${CMAKE_MESSAGE_LOG_LEVEL})
+set(CMAKE_MESSAGE_LOG_LEVEL WARNING)
 add_subdirectory(../mechanisms "${CMAKE_BINARY_DIR}/mechanisms")
+set(CMAKE_MESSAGE_LOG_LEVEL ${_saved_CMAKE_MESSAGE_LOG_LEVEL})
 set_source_files_properties(${arbor-builtin-mechanisms} DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR} PROPERTIES GENERATED TRUE)
 
 if(ARB_WITH_CUDA_CLANG OR ARB_WITH_HIP_CLANG)

arbor/backends/gpu/diffusion_state.hpp

@@ -15,6 +15,7 @@
 #include "diffusion.hpp"
 #include "forest.hpp"
 #include "fine.hpp"
+#include "profile/profiler_macro.hpp"
 
 namespace arb {
 namespace gpu {
@@ -402,6 +403,7 @@ struct diffusion_state {
     //   current density [A/m²]
     //   conductivity [kS/m²]
     void assemble(const value_type dt, const_view concentration, const_view voltage, const_view current, const_view conductivity, arb_value_type q) {
+        PROFILE_ZONE();
         assemble_diffusion(d.data(),
                            rhs.data(),
                            invariant_d.data(),
@@ -417,6 +419,7 @@ struct diffusion_state {
     }
 
     void solve(array& to) {
+        PROFILE_ZONE();
         solve_diffusion(rhs.data(),
                         d.data(),
                         u.data(),
@@ -438,6 +441,7 @@ struct diffusion_state {
                const_view current,
                const_view conductivity,
                arb_value_type q) {
+        PROFILE_ZONE();
         assemble(dt, concentration, voltage, current, conductivity, q);
         solve(concentration);
     }

arbor/backends/gpu/matrix_state_fine.hpp

@@ -1,7 +1,6 @@
 #pragma once
 
 #include <cstring>
-
 #include <vector>
 #include <type_traits>
 
@@ -15,6 +14,7 @@
 #include "fine.hpp"
 #include "matrix_fine.hpp"
 #include "forest.hpp"
+#include "profile/profiler_macro.hpp"
 
 namespace arb {
 namespace gpu {
@@ -409,6 +409,7 @@ struct matrix_state_fine {
     //   current density [A/m²]
     //   conductivity [kS/m²]
     void assemble(const T dt, const_view voltage, const_view current, const_view conductivity) {
+        PROFILE_ZONE();
         assemble_matrix_fine(
             d.data(),
             rhs.data(),
@@ -424,6 +425,7 @@ struct matrix_state_fine {
     }
 
     void solve(array& to) {
+        PROFILE_ZONE();
         solve_matrix_fine(rhs.data(),
                           d.data(),
                           u.data(),
@@ -450,13 +452,15 @@ struct matrix_state_fine {
 
 private:
     void flat_to_packed(const array& from, array& to ) {
+        PROFILE_ZONE();
         arb_assert(from.size()==matrix_size);
         arb_assert(to.size()==data_size);
 
         scatter(from.data(), to.data(), perm.data(), perm.size());
     }
 
     void packed_to_flat(const array& from, array& to ) {
+        PROFILE_ZONE();
         arb_assert(from.size()==data_size);
         arb_assert(to.size()==matrix_size);
 

arbor/backends/gpu/shared_state.cpp

@@ -21,6 +21,7 @@
 #include "util/meta.hpp"
 #include "util/range.hpp"
 #include "util/strprintf.hpp"
+#include "profile/profiler_macro.hpp"
 
 #include <iostream>
 
@@ -161,6 +162,7 @@ void istim_state::reset() {
 }
 
 void istim_state::add_current(const arb_value_type time, array& current_density) {
+    PROFILE_ZONE();
     ppack_.time = time;
     ppack_.current_density = current_density.data();
     istim_add_current_impl((int)size(), ppack_);
@@ -202,6 +204,7 @@ shared_state::shared_state(task_system_handle tp,
 }
 
 void shared_state::update_prng_state(mechanism& m) {
+    PROFILE_ZONE();
     if (!m.mech_.n_random_variables) return;
     auto const mech_id = m.mechanism_id();
     auto& store = storage[mech_id];
@@ -213,6 +216,8 @@ void shared_state::instantiate(mechanism& m,
                                const mechanism_overrides& overrides,
                                const mechanism_layout& pos_data,
                                const std::vector<std::pair<std::string, std::vector<arb_value_type>>>& params) {
+    PROFILE_ZONE();
+    ANNOTATE_ZONE(m.mech_.name, strlen(m.mech_.name));
     assert(m.iface_.backend == arb_backend_kind_gpu);
     using util::make_range;
     using util::make_span;
@@ -355,6 +360,7 @@ void shared_state::instantiate(mechanism& m,
 }
 
 void shared_state::reset() {
+    PROFILE_ZONE();
     memory::copy(init_voltage, voltage);
     memory::fill(current_density, 0);
     memory::fill(conductivity, 0);
@@ -369,6 +375,7 @@ void shared_state::reset() {
 }
 
 void shared_state::zero_currents() {
+    PROFILE_ZONE();
     memory::fill(current_density, 0);
     memory::fill(conductivity, 0);
     for (auto& i: ion_data) {
@@ -382,6 +389,7 @@ std::pair<arb_value_type, arb_value_type> shared_state::voltage_bounds() const {
 }
 
 void shared_state::take_samples() {
+    PROFILE_ZONE();
     sample_events.mark();
     if (!sample_events.empty()) {
         const auto state = sample_events.marked_events();

arbor/backends/multicore/cable_solver.hpp

@@ -68,6 +68,7 @@ struct cable_solver {
     // * will likewise overwrite the first argument with the solction
     template<typename T>
     void solve(T& rhs, const value_type dt, const_view current, const_view conductivity) {
+        PROFILE_ZONE();
         value_type * const ARB_NO_ALIAS d_ = d.data();
         value_type * const ARB_NO_ALIAS r_ = rhs.data();
 

arbor/backends/multicore/diffusion_solver.hpp

@@ -6,6 +6,7 @@
 #include <memory/memory.hpp>
 
 #include "multicore_common.hpp"
+#include "profile/profiler_macro.hpp"
 
 namespace arb {
 namespace multicore {
@@ -77,6 +78,7 @@ struct diffusion_solver {
                const_view current,
                const_view conductivity,
                arb_value_type q) {
+        PROFILE_ZONE();
         auto cell_cv_part = util::partition_view(cell_cv_divs);
         index_type ncells = cell_cv_part.size();
         // loop over submatrices
@@ -101,6 +103,7 @@ struct diffusion_solver {
     // Separate solver; analoguos with cable solver
     template<typename T>
     void solve(T& rhs) {
+        PROFILE_ZONE();
         // loop over submatrices
         for (const auto& [first, last]: util::partition_view(cell_cv_divs)) {
             if (first >= last) continue; // skip cell with no CVs

arbor/backends/multicore/fvm.hpp

@@ -1,19 +1,14 @@
 #pragma once
 
 #include <string>
-#include <vector>
 
 #include <arbor/mechanism.hpp>
 
-#include "backends/event.hpp"
-#include "backends/multicore/event_stream.hpp"
 #include "backends/multicore/multicore_common.hpp"
 #include "backends/multicore/shared_state.hpp"
 #include "backends/multicore/diffusion_solver.hpp"
 #include "backends/multicore/cable_solver.hpp"
 #include "backends/multicore/threshold_watcher.hpp"
-#include "execution_context.hpp"
-#include "util/padded_alloc.hpp"
 #include "util/range.hpp"
 #include "util/rangeutil.hpp"
 

arbor/backends/multicore/multicore_common.hpp

@@ -5,7 +5,6 @@
 //
 // Defines array, iarray, and specialized multi-event stream classes.
 
-#include <utility>
 #include <vector>
 
 #include <arbor/fvm_types.hpp>