1#ifndef AMREX_TINY_PROFILER_H_
2#define AMREX_TINY_PROFILER_H_
3#include <AMReX_Config.H>
34 TinyProfiler (std::string funcname,
bool start_)
noexcept;
35 TinyProfiler (std::string funcname,
bool start_,
bool gpu_sync_)
noexcept;
37 TinyProfiler (
const char* funcname,
bool start_)
noexcept;
38 TinyProfiler (
const char* funcname,
bool start_,
bool gpu_sync_)
noexcept;
53 std::map<std::
string,
MemStat>& memstats) noexcept;
57 static
void Finalize (
bool bFlushing = false);
63 std::map<std::
string,
MemStat>& memstats) noexcept;
67 static
void StartRegion (std::
string regname) noexcept;
68 static
void StartRegion (std::
string regname,
bool gpu_sync) noexcept;
69 static
void StopRegion (const std::
string& regname) noexcept;
91 Stats ()
noexcept =
default;
101 bool gpu_sync =
false;
103 bool operator< (TimerKey
const& rhs)
const noexcept
105 return std::tie(name, gpu_sync) < std::tie(rhs.name, rhs.gpu_sync);
109 using RegionStats = std::map<TimerKey, Stats>;
110 using StatsMap = std::map<std::string, RegionStats>;
116 Long nmin{std::numeric_limits<Long>::max()};
117 Long navg{0L}, nmax{0L};
118 double dtinmin{std::numeric_limits<double>::max()};
119 double dtinavg{0.0}, dtinmax{0.0};
120 double dtexmin{std::numeric_limits<double>::max()};
121 double dtexavg{0.0}, dtexmax{0.0};
124 static bool compex (
const ProcStats& lhs,
const ProcStats& rhs) {
125 return lhs.dtexmax > rhs.dtexmax;
127 static bool compin (
const ProcStats& lhs,
const ProcStats& rhs) {
128 return lhs.dtinmax > rhs.dtinmax;
137 Long avgmem_min = std::numeric_limits<Long>::max();
140 Long maxmem_min = std::numeric_limits<Long>::max();
144 static bool compmem (
const MemProcStats& lhs,
const MemProcStats& rhs) {
145 return lhs.maxmem_max > rhs.maxmem_max;
151 [[maybe_unused]]
bool gpu_sync =
false;
152 bool in_parallel_region =
false;
153 int global_depth = -1;
154 std::vector<Stats*> stats;
155 std::vector<Stats*> selective_stats;
157 const char* prev_name =
nullptr;
159 static std::deque<const TinyProfiler*> mem_stack;
162 struct aligned_deque {
164 alignas(64) std::deque<const TinyProfiler*> deque;
167 static std::vector<aligned_deque> mem_stack_thread_private;
169 static std::vector<std::map<std::string, MemStat>*> all_memstats;
170 static std::vector<std::string> all_memnames;
172 static std::vector<std::string> regionstack;
173 static std::vector<std::string> selective_regionstack;
174 static std::vector<std::tuple<std::string,bool,bool> > regionstartstack;
176 static std::deque<std::tuple<double,double,std::string*,double> > ttstack;
177 static StatsMap statsmap;
178 static StatsMap selective_statsmap;
179 static bool selective_reporting;
180 static double t_init;
181 static double t_memory_init;
182 static bool device_synchronize_around_region;
183 static int n_print_tabs;
185 static double print_threshold;
187 static bool memprof_enabled;
188 static std::string output_file;
190 static std::string
const& get_output_file ();
191 static void PrintStats (RegionStats& regstats,
double dt_max,
193 static void PrintMemStats (std::map<std::string, MemStat>& memstats,
194 std::string
const& memname,
double dt_max,
195 double t_final, std::ostream* os,
bool only_local);
Definition AMReX_TinyProfiler.H:199
~TinyProfileRegion()
Definition AMReX_TinyProfiler.cpp:1138
TinyProfileRegion & operator=(TinyProfileRegion const &)=delete
TinyProfileRegion(TinyProfileRegion const &)=delete
TinyProfileRegion(TinyProfileRegion &&)=delete
A simple profiler that returns basic performance information (e.g. min, max, and average running time...
Definition AMReX_BLProfiler.H:636
static void Initialize()
Definition AMReX_TinyProfiler.cpp:413
static bool RegisterArena(const std::string &memory_name, std::map< std::string, MemStat > &memstats) noexcept
Definition AMReX_TinyProfiler.cpp:593
static void PrintMemoryUsage(std::ostream *os, bool only_local) noexcept
Definition AMReX_TinyProfiler.cpp:1164
void start()
Definition AMReX_TinyProfiler.cpp:149
TinyProfiler(TinyProfiler const &)=delete
TinyProfiler & operator=(TinyProfiler const &)=delete
static MemStat * memory_alloc(std::size_t nbytes, std::map< std::string, MemStat > &memstats) noexcept
Definition AMReX_TinyProfiler.cpp:369
static void PrintCallStack(std::ostream &os)
Definition AMReX_TinyProfiler.cpp:1145
void stop()
Definition AMReX_TinyProfiler.cpp:240
static void MemoryInitialize()
Definition AMReX_TinyProfiler.cpp:447
void memory_start() const noexcept
Definition AMReX_TinyProfiler.cpp:330
void memory_stop() const noexcept
Definition AMReX_TinyProfiler.cpp:347
static void StartRegion(std::string regname) noexcept
Definition AMReX_TinyProfiler.cpp:1059
~TinyProfiler()
Definition AMReX_TinyProfiler.cpp:143
static void StopRegion(const std::string ®name) noexcept
Definition AMReX_TinyProfiler.cpp:1086
static void memory_free(std::size_t nbytes, MemStat *stat) noexcept
Definition AMReX_TinyProfiler.cpp:398
static void MemoryFinalize(bool bFlushing=false)
Definition AMReX_TinyProfiler.cpp:564
TinyProfiler(TinyProfiler &&)=delete
static void DeregisterArena(std::map< std::string, MemStat > &memstats) noexcept
Definition AMReX_TinyProfiler.cpp:604
static const char * CurrentName() noexcept
Without the tiny profiler, there is no active profiler.
Definition AMReX_BLProfiler.H:639
amrex_long Long
Definition AMReX_INT.H:30
Definition AMReX_Amr.cpp:50
void Finalize()
Convenience overload that finalizes the instance at the top of the stack.
Definition AMReX.cpp:835
Definition AMReX_Arena.H:21
Long currentmem
amount of currently used memory in bytes
Definition AMReX_TinyProfiler.H:24
Long maxmem
running maximum of currentmem
Definition AMReX_TinyProfiler.H:26
Long nalloc
number of allocations
Definition AMReX_TinyProfiler.H:22
Long nfree
number of frees
Definition AMReX_TinyProfiler.H:23
double avgmem
memory used (bytes) times time in use (seconds)
Definition AMReX_TinyProfiler.H:25