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;
36 TinyProfiler (
const char* funcname,
bool start_)
noexcept;
51 std::map<std::
string,
MemStat>& memstats) noexcept;
55 static
void Finalize (
bool bFlushing = false);
61 std::map<std::
string,
MemStat>& memstats) noexcept;
65 static
void StartRegion (std::
string regname) noexcept;
66 static
void StopRegion (const std::
string& regname) noexcept;
88 Stats ()
noexcept =
default;
99 Long nmin{std::numeric_limits<Long>::max()};
100 Long navg{0L}, nmax{0L};
101 double dtinmin{std::numeric_limits<double>::max()};
102 double dtinavg{0.0}, dtinmax{0.0};
103 double dtexmin{std::numeric_limits<double>::max()};
104 double dtexavg{0.0}, dtexmax{0.0};
107 static bool compex (
const ProcStats& lhs,
const ProcStats& rhs) {
108 return lhs.dtexmax > rhs.dtexmax;
110 static bool compin (
const ProcStats& lhs,
const ProcStats& rhs) {
111 return lhs.dtinmax > rhs.dtinmax;
120 Long avgmem_min = std::numeric_limits<Long>::max();
123 Long maxmem_min = std::numeric_limits<Long>::max();
127 static bool compmem (
const MemProcStats& lhs,
const MemProcStats& rhs) {
128 return lhs.maxmem_max > rhs.maxmem_max;
133 bool in_parallel_region =
false;
134 int global_depth = -1;
135 std::vector<Stats*> stats;
137 const char* prev_name =
nullptr;
139 static std::deque<const TinyProfiler*> mem_stack;
142 struct aligned_deque {
144 alignas(64) std::deque<const TinyProfiler*> deque;
147 static std::vector<aligned_deque> mem_stack_thread_private;
149 static std::vector<std::map<std::string, MemStat>*> all_memstats;
150 static std::vector<std::string> all_memnames;
152 static std::vector<std::string> regionstack;
153 static std::vector<std::pair<std::string,bool> > regionstartstack;
154 static std::deque<std::tuple<double,double,std::string*> > ttstack;
155 static std::map<std::string,std::map<std::string, Stats> > statsmap;
156 static double t_init;
157 static double t_memory_init;
158 static bool device_synchronize_around_region;
159 static int n_print_tabs;
161 static double print_threshold;
163 static bool memprof_enabled;
164 static std::string output_file;
166 static std::string
const& get_output_file ();
167 static void PrintStats (std::map<std::string,Stats>& regstats,
double dt_max,
169 static void PrintMemStats (std::map<std::string, MemStat>& memstats,
170 std::string
const& memname,
double dt_max,
171 double t_final, std::ostream* os,
bool only_local);
Definition AMReX_TinyProfiler.H:175
~TinyProfileRegion()
Definition AMReX_TinyProfiler.cpp:1028
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:601
static void Initialize()
Definition AMReX_TinyProfiler.cpp:359
static bool RegisterArena(const std::string &memory_name, std::map< std::string, MemStat > &memstats) noexcept
Definition AMReX_TinyProfiler.cpp:526
static void PrintMemoryUsage(std::ostream *os, bool only_local) noexcept
Definition AMReX_TinyProfiler.cpp:1054
void start()
Definition AMReX_TinyProfiler.cpp:135
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:315
static void PrintCallStack(std::ostream &os)
Definition AMReX_TinyProfiler.cpp:1035
void stop()
Definition AMReX_TinyProfiler.cpp:202
static void MemoryInitialize()
Definition AMReX_TinyProfiler.cpp:389
void memory_start() const noexcept
Definition AMReX_TinyProfiler.cpp:276
void memory_stop() const noexcept
Definition AMReX_TinyProfiler.cpp:293
static void StartRegion(std::string regname) noexcept
Definition AMReX_TinyProfiler.cpp:985
~TinyProfiler()
Definition AMReX_TinyProfiler.cpp:129
static void StopRegion(const std::string ®name) noexcept
Definition AMReX_TinyProfiler.cpp:998
static void memory_free(std::size_t nbytes, MemStat *stat) noexcept
Definition AMReX_TinyProfiler.cpp:344
static void MemoryFinalize(bool bFlushing=false)
Definition AMReX_TinyProfiler.cpp:497
TinyProfiler(TinyProfiler &&)=delete
static void DeregisterArena(std::map< std::string, MemStat > &memstats) noexcept
Definition AMReX_TinyProfiler.cpp:537
static const char * CurrentName() noexcept
Without the tiny profiler, there is no active profiler.
Definition AMReX_BLProfiler.H:604
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