From 5e65a939d6fc9bb929c3faab99cd0022efeec878 Mon Sep 17 00:00:00 2001
From: Ben Cumming <bcumming@cscs.ch>
Date: Thu, 7 Jun 2018 12:56:59 +0200
Subject: [PATCH] profile multicore mechanism state and current calls
individually (#492)
The built in profiler generates timings for state and current for individual multicore mechanisms.
Modcc generates and PE(advance_integrate_{state,current}_X) profiler calls (along with corresponding PL() for calls to multicore mechanism nrn_state and nrn_current API calls.
No timings are made for the gpu back end, which is not properly supported by the current profiling tools.
---
modcc/printer/cprinter.cpp | 14 ++++++++++++++
src/fvm_lowered_cell_impl.hpp | 6 ++----
2 files changed, 16 insertions(+), 4 deletions(-)
diff --git a/modcc/printer/cprinter.cpp b/modcc/printer/cprinter.cpp
index c856f4b6..e7b478b6 100644
--- a/modcc/printer/cprinter.cpp
+++ b/modcc/printer/cprinter.cpp
@@ -14,6 +14,13 @@ using io::indent;
using io::popindent;
using io::quote;
+constexpr bool with_profiling() {
+#ifdef ARB_HAVE_PROFILING
+ return true;
+#else
+ return false;
+#endif
+}
void emit_procedure_proto(std::ostream&, ProcedureExpression*, const std::string& qualified = "");
void emit_simd_procedure_proto(std::ostream&, ProcedureExpression*, const std::string& qualified = "");
@@ -106,6 +113,9 @@ std::string emit_cpp_source(const Module& module_, const std::string& ns, simd_s
"#include <" << arb_header_prefix() << "backends/multicore/mechanism.hpp>\n"
"#include <" << arb_header_prefix() << "math.hpp>\n";
+ with_profiling() &&
+ out << "#include <" << arb_header_prefix() << "profiling/profiler.hpp>\n";
+
if (with_simd) {
out << "#include <" << arb_header_prefix() << "simd/simd.hpp>\n";
}
@@ -306,11 +316,15 @@ std::string emit_cpp_source(const Module& module_, const std::string& ns, simd_s
out << popindent << "}\n\n";
out << "void " << class_name << "::nrn_state() {\n" << indent;
+ with_profiling() && out << "PE(advance_integrate_state_" << name << ");\n";
emit_body(state_api);
+ with_profiling() && out << "PL();\n";
out << popindent << "}\n\n";
out << "void " << class_name << "::nrn_current() {\n" << indent;
+ with_profiling() && out << "PE(advance_integrate_current_" << name << ");\n";
emit_body(current_api);
+ with_profiling() && out << "PL();\n";
out << popindent << "}\n\n";
out << "void " << class_name << "::write_ions() {\n" << indent;
diff --git a/src/fvm_lowered_cell_impl.hpp b/src/fvm_lowered_cell_impl.hpp
index 765dfb2a..c793659b 100644
--- a/src/fvm_lowered_cell_impl.hpp
+++ b/src/fvm_lowered_cell_impl.hpp
@@ -170,13 +170,13 @@ fvm_integration_result fvm_lowered_cell_impl<Backend>::integrate(
state_->deliverable_events.mark_until_after(state_->time);
PL();
- PE(advance_integrate_current);
+ PE(advance_integrate_current_zero);
state_->zero_currents();
+ PL();
for (auto& m: mechanisms_) {
m->deliver_events();
m->nrn_current();
}
- PL();
PE(advance_integrate_events);
state_->deliverable_events.drop_marked_events();
@@ -208,11 +208,9 @@ fvm_integration_result fvm_lowered_cell_impl<Backend>::integrate(
// Integrate mechanism state.
- PE(advance_integrate_state);
for (auto& m: mechanisms_) {
m->nrn_state();
}
- PL();
// Update ion concentrations.
--
GitLab