Remove old code
This commit is contained in:
parent
69e8f5ba12
commit
ce4552add5
1250
src/emu/old_emu.c
1250
src/emu/old_emu.c
File diff suppressed because it is too large
Load Diff
@ -1,586 +0,0 @@
|
||||
/* Copyright (c) 2021 Barcelona Supercomputing Center (BSC)
|
||||
* SPDX-License-Identifier: GPL-3.0-or-later */
|
||||
|
||||
#ifndef OVNI_EMU_H
|
||||
#define OVNI_EMU_H
|
||||
|
||||
#include <stdio.h>
|
||||
|
||||
#include "common.h"
|
||||
#include "heap.h"
|
||||
#include "ovni.h"
|
||||
#include "parson.h"
|
||||
#include "pcf.h"
|
||||
#include "uthash.h"
|
||||
|
||||
/* Emulated thread runtime status */
|
||||
enum ethread_state {
|
||||
TH_ST_UNKNOWN,
|
||||
TH_ST_RUNNING,
|
||||
TH_ST_PAUSED,
|
||||
TH_ST_DEAD,
|
||||
TH_ST_COOLING,
|
||||
TH_ST_WARMING,
|
||||
};
|
||||
|
||||
enum task_state {
|
||||
TASK_ST_CREATED,
|
||||
TASK_ST_RUNNING,
|
||||
TASK_ST_PAUSED,
|
||||
TASK_ST_DEAD,
|
||||
};
|
||||
|
||||
enum ovni_state {
|
||||
ST_OVNI_FLUSHING = 1,
|
||||
};
|
||||
|
||||
enum error_values {
|
||||
ST_BAD = 666,
|
||||
ST_TOO_MANY_TH = 777,
|
||||
};
|
||||
|
||||
enum nosv_ss_values {
|
||||
ST_NULL = 0,
|
||||
ST_NOSV_SCHED_HUNGRY = 6,
|
||||
ST_NOSV_SCHED_SERVING,
|
||||
ST_NOSV_SCHED_SUBMITTING,
|
||||
ST_NOSV_MEM_ALLOCATING,
|
||||
ST_NOSV_MEM_FREEING,
|
||||
ST_NOSV_TASK_RUNNING,
|
||||
ST_NOSV_API_SUBMIT,
|
||||
ST_NOSV_API_PAUSE,
|
||||
ST_NOSV_API_YIELD,
|
||||
ST_NOSV_API_WAITFOR,
|
||||
ST_NOSV_API_SCHEDPOINT,
|
||||
ST_NOSV_ATTACH,
|
||||
ST_NOSV_WORKER,
|
||||
ST_NOSV_DELEGATE,
|
||||
|
||||
EV_NOSV_SCHED_RECV,
|
||||
EV_NOSV_SCHED_SEND,
|
||||
EV_NOSV_SCHED_SELF,
|
||||
};
|
||||
|
||||
enum tampi_state {
|
||||
ST_TAMPI_SEND = 1,
|
||||
ST_TAMPI_RECV = 2,
|
||||
ST_TAMPI_ISEND = 3,
|
||||
ST_TAMPI_IRECV = 4,
|
||||
ST_TAMPI_WAIT = 5,
|
||||
ST_TAMPI_WAITALL = 6,
|
||||
};
|
||||
|
||||
enum openmp_state {
|
||||
ST_OPENMP_TASK = 1,
|
||||
ST_OPENMP_PARALLEL = 2,
|
||||
};
|
||||
|
||||
enum nodes_state {
|
||||
ST_NODES_REGISTER = 1,
|
||||
ST_NODES_UNREGISTER = 2,
|
||||
ST_NODES_IF0_WAIT = 3,
|
||||
ST_NODES_IF0_INLINE = 4,
|
||||
ST_NODES_TASKWAIT = 5,
|
||||
ST_NODES_CREATE = 6,
|
||||
ST_NODES_SUBMIT = 7,
|
||||
ST_NODES_SPAWN = 8,
|
||||
};
|
||||
|
||||
/* The values of nanos6_ss_state are synced to the previous
|
||||
* CTF implementation. */
|
||||
enum nanos6_ss_state {
|
||||
ST_NANOS6_TASK_BODY = 1,
|
||||
ST_NANOS6_TASK_CREATING,
|
||||
ST_NANOS6_TASK_SUBMIT,
|
||||
ST_NANOS6_TASK_SPAWNING,
|
||||
ST_NANOS6_TASK_FOR,
|
||||
ST_NANOS6_SCHED_ADDING,
|
||||
ST_NANOS6_SCHED_PROCESSING,
|
||||
ST_NANOS6_SCHED_SERVING,
|
||||
ST_NANOS6_DEP_REG,
|
||||
ST_NANOS6_DEP_UNREG,
|
||||
ST_NANOS6_BLK_TASKWAIT,
|
||||
ST_NANOS6_BLK_WAITFOR,
|
||||
ST_NANOS6_BLK_BLOCKING,
|
||||
ST_NANOS6_BLK_UNBLOCKING,
|
||||
ST_NANOS6_ALLOCATING,
|
||||
ST_NANOS6_FREEING,
|
||||
ST_NANOS6_HANDLING_TASK,
|
||||
ST_NANOS6_WORKER_LOOP,
|
||||
ST_NANOS6_SWITCH_TO,
|
||||
ST_NANOS6_MIGRATE,
|
||||
ST_NANOS6_SUSPEND,
|
||||
ST_NANOS6_RESUME,
|
||||
|
||||
/* Value 51 is broken in old Paraver */
|
||||
EV_NANOS6_SCHED_RECV = 60,
|
||||
EV_NANOS6_SCHED_SEND,
|
||||
EV_NANOS6_SCHED_SELF,
|
||||
EV_NANOS6_CPU_IDLE,
|
||||
EV_NANOS6_CPU_ACTIVE,
|
||||
EV_NANOS6_SIGNAL,
|
||||
};
|
||||
|
||||
enum nanos6_thread_type {
|
||||
ST_NANOS6_TH_LEADER = 1,
|
||||
ST_NANOS6_TH_MAIN = 2,
|
||||
ST_NANOS6_TH_WORKER = 3,
|
||||
ST_NANOS6_TH_EXTERNAL = 4,
|
||||
};
|
||||
|
||||
enum kernel_cs_state {
|
||||
ST_KERNEL_CSOUT = 3,
|
||||
};
|
||||
|
||||
struct ovni_ethread;
|
||||
struct ovni_eproc;
|
||||
|
||||
struct task_type {
|
||||
uint32_t id; /* Per-process task identifier */
|
||||
uint32_t gid; /* Global identifier computed from the label */
|
||||
char label[MAX_PCF_LABEL];
|
||||
UT_hash_handle hh;
|
||||
};
|
||||
|
||||
struct task {
|
||||
uint32_t id;
|
||||
struct task_type *type;
|
||||
|
||||
/* The thread that has began to execute the task. It cannot
|
||||
* change after being set, even if the task ends. */
|
||||
struct ovni_ethread *thread;
|
||||
enum task_state state;
|
||||
UT_hash_handle hh;
|
||||
|
||||
/* List handle for nested task support */
|
||||
struct task *next;
|
||||
struct task *prev;
|
||||
};
|
||||
|
||||
struct task_info {
|
||||
/* Both hash maps of all known tasks and types */
|
||||
struct task_type *types;
|
||||
struct task *tasks;
|
||||
};
|
||||
|
||||
struct task_stack {
|
||||
union {
|
||||
struct task *top; /* Synctactic sugar */
|
||||
struct task *tasks;
|
||||
};
|
||||
struct ovni_ethread *thread;
|
||||
};
|
||||
|
||||
#define MAX_CHAN_STACK 512
|
||||
|
||||
enum chan_track {
|
||||
/* The channel is manually controlled. */
|
||||
CHAN_TRACK_NONE = 0,
|
||||
|
||||
/* Enables the channel when the thread is running only. */
|
||||
CHAN_TRACK_TH_RUNNING,
|
||||
|
||||
/* The thread active tracking mode a enables the channel when
|
||||
* the thread is running, cooling or warming. Otherwise the
|
||||
* channel is disabled. */
|
||||
CHAN_TRACK_TH_ACTIVE,
|
||||
};
|
||||
|
||||
enum chan {
|
||||
CHAN_OVNI_PID,
|
||||
CHAN_OVNI_TID,
|
||||
CHAN_OVNI_NRTHREADS,
|
||||
CHAN_OVNI_STATE,
|
||||
CHAN_OVNI_APPID,
|
||||
CHAN_OVNI_CPU,
|
||||
CHAN_OVNI_FLUSH,
|
||||
|
||||
CHAN_NOSV_TASKID,
|
||||
CHAN_NOSV_TYPE,
|
||||
CHAN_NOSV_APPID,
|
||||
CHAN_NOSV_SUBSYSTEM,
|
||||
CHAN_NOSV_RANK,
|
||||
|
||||
CHAN_TAMPI_MODE,
|
||||
CHAN_OPENMP_MODE,
|
||||
CHAN_NODES_SUBSYSTEM,
|
||||
|
||||
CHAN_NANOS6_TASKID,
|
||||
CHAN_NANOS6_TYPE,
|
||||
CHAN_NANOS6_SUBSYSTEM,
|
||||
CHAN_NANOS6_RANK,
|
||||
CHAN_NANOS6_THREAD,
|
||||
|
||||
CHAN_KERNEL_CS,
|
||||
|
||||
CHAN_MAX
|
||||
};
|
||||
|
||||
enum chan_type {
|
||||
CHAN_TH = 0,
|
||||
CHAN_CPU = 1,
|
||||
CHAN_MAXTYPE = 2,
|
||||
};
|
||||
|
||||
enum chan_dirty {
|
||||
CHAN_CLEAN = 0,
|
||||
|
||||
/* The channel is dirty because it has been enabled or disabled */
|
||||
CHAN_DIRTY_ACTIVE = 1,
|
||||
|
||||
/* The channel is dirty because it changed the state */
|
||||
CHAN_DIRTY_VALUE = 2,
|
||||
};
|
||||
|
||||
/* clang-format off */
|
||||
static const int chan_to_prvtype[CHAN_MAX] = {
|
||||
[CHAN_OVNI_PID] = 1,
|
||||
[CHAN_OVNI_TID] = 2,
|
||||
[CHAN_OVNI_NRTHREADS] = 3,
|
||||
[CHAN_OVNI_STATE] = 4,
|
||||
[CHAN_OVNI_APPID] = 5, /* Not used */
|
||||
[CHAN_OVNI_CPU] = 6,
|
||||
[CHAN_OVNI_FLUSH] = 7,
|
||||
[CHAN_NOSV_TASKID] = 10,
|
||||
[CHAN_NOSV_TYPE] = 11,
|
||||
[CHAN_NOSV_APPID] = 12,
|
||||
[CHAN_NOSV_SUBSYSTEM] = 13,
|
||||
[CHAN_NOSV_RANK] = 14,
|
||||
[CHAN_TAMPI_MODE] = 20,
|
||||
[CHAN_OPENMP_MODE] = 25,
|
||||
[CHAN_NODES_SUBSYSTEM] = 30,
|
||||
[CHAN_NANOS6_TASKID] = 35,
|
||||
[CHAN_NANOS6_TYPE] = 36,
|
||||
[CHAN_NANOS6_SUBSYSTEM] = 37,
|
||||
[CHAN_NANOS6_RANK] = 38,
|
||||
[CHAN_NANOS6_THREAD] = 39,
|
||||
[CHAN_KERNEL_CS] = 45,
|
||||
};
|
||||
/* clang-format on */
|
||||
|
||||
struct ovni_chan {
|
||||
/* Channel id */
|
||||
enum chan id;
|
||||
|
||||
/* Number of states in the stack */
|
||||
int n;
|
||||
|
||||
/* Stack of states */
|
||||
int stack[MAX_CHAN_STACK];
|
||||
|
||||
/* 1 if enabled, 0 if not. */
|
||||
int enabled;
|
||||
|
||||
/* What state should be shown in errors */
|
||||
int badst;
|
||||
|
||||
/* Last state emitted (-1 otherwise) */
|
||||
int lastst;
|
||||
|
||||
/* Punctual event: -1 if not used */
|
||||
int ev;
|
||||
|
||||
/* Emit events of this type */
|
||||
int type;
|
||||
|
||||
/* A pointer to a clock to sample the time */
|
||||
int64_t *clock;
|
||||
|
||||
/* The time of the last state or event */
|
||||
int64_t t;
|
||||
|
||||
/* Paraver row */
|
||||
int row;
|
||||
|
||||
/* Type of dirty */
|
||||
enum chan_dirty dirty;
|
||||
|
||||
/* Where should the events be written to? */
|
||||
FILE *prv;
|
||||
|
||||
/* What should cause the channel to become disabled? */
|
||||
enum chan_track track;
|
||||
|
||||
/* The thread associated with the channel if any */
|
||||
struct ovni_ethread *thread;
|
||||
|
||||
/* The CPU associated with the channel if any */
|
||||
struct ovni_cpu *cpu;
|
||||
|
||||
struct ovni_chan **update_list;
|
||||
|
||||
/* Used when the channel is a list */
|
||||
struct ovni_chan *prev;
|
||||
struct ovni_chan *next;
|
||||
};
|
||||
|
||||
#define MAX_BURSTS 100
|
||||
|
||||
/* State of each emulated thread */
|
||||
struct ovni_ethread {
|
||||
/* Emulated thread tid */
|
||||
pid_t tid;
|
||||
|
||||
int index;
|
||||
int gindex;
|
||||
|
||||
/* The process associated with this thread */
|
||||
struct ovni_eproc *proc;
|
||||
|
||||
enum ethread_state state;
|
||||
int is_running;
|
||||
int is_active;
|
||||
|
||||
/* Thread stream */
|
||||
struct ovni_stream *stream;
|
||||
|
||||
/* Current cpu */
|
||||
struct ovni_cpu *cpu;
|
||||
|
||||
/* FIXME: Use a table with registrable pointers to custom data
|
||||
* structures */
|
||||
|
||||
/* Task stacks, top ones are the tasks currently runnable. */
|
||||
struct task_stack nosv_task_stack;
|
||||
struct task_stack nanos6_task_stack;
|
||||
|
||||
/* Channels are used to output the emulator state in PRV */
|
||||
struct ovni_chan chan[CHAN_MAX];
|
||||
|
||||
/* Burst times */
|
||||
int nbursts;
|
||||
int64_t burst_time[MAX_BURSTS];
|
||||
|
||||
/* These pointers keep a linked list of threads in each CPU */
|
||||
struct ovni_ethread *prev;
|
||||
struct ovni_ethread *next;
|
||||
|
||||
/* Trace file path */
|
||||
char tracefile[PATH_MAX];
|
||||
};
|
||||
|
||||
/* State of each emulated process */
|
||||
struct ovni_eproc {
|
||||
int pid;
|
||||
int index;
|
||||
int gindex;
|
||||
int appid;
|
||||
int rank;
|
||||
|
||||
/* The loom of the current process */
|
||||
struct ovni_loom *loom;
|
||||
|
||||
/* Path of the process tracedir */
|
||||
char dir[PATH_MAX];
|
||||
|
||||
/* Threads */
|
||||
size_t nthreads;
|
||||
struct ovni_ethread *thread;
|
||||
|
||||
JSON_Value *meta;
|
||||
|
||||
/* ------ Subsystem specific data --------*/
|
||||
/* TODO: Use dynamic allocation */
|
||||
|
||||
struct task_info nosv_task_info;
|
||||
struct task_info nanos6_task_info;
|
||||
};
|
||||
|
||||
|
||||
/* ------------------ emulation ---------------- */
|
||||
|
||||
enum ovni_cpu_type {
|
||||
CPU_REAL,
|
||||
CPU_VIRTUAL,
|
||||
};
|
||||
|
||||
enum ovni_cpu_state {
|
||||
CPU_ST_UNKNOWN,
|
||||
CPU_ST_READY,
|
||||
};
|
||||
|
||||
#define MAX_CPU_NAME 32
|
||||
|
||||
struct ovni_cpu {
|
||||
/* Logical index: 0 to ncpus - 1 */
|
||||
int i;
|
||||
|
||||
/* Physical id: as reported by lscpu(1) */
|
||||
int phyid;
|
||||
|
||||
/* Global index for all CPUs */
|
||||
int gindex;
|
||||
|
||||
enum ovni_cpu_state state;
|
||||
|
||||
/* The loom of the CPU */
|
||||
struct ovni_loom *loom;
|
||||
|
||||
/* CPU channels */
|
||||
struct ovni_chan chan[CHAN_MAX];
|
||||
|
||||
/* The threads assigned to this CPU */
|
||||
size_t nthreads;
|
||||
struct ovni_ethread *thread;
|
||||
|
||||
/* Running threads */
|
||||
size_t nrunning_threads;
|
||||
struct ovni_ethread *th_running;
|
||||
|
||||
/* Active threads (not paused or dead) */
|
||||
size_t nactive_threads;
|
||||
struct ovni_ethread *th_active;
|
||||
|
||||
/* Cpu name as shown in paraver row */
|
||||
char name[MAX_CPU_NAME];
|
||||
|
||||
/* Is this a virtual CPU? */
|
||||
int virtual;
|
||||
};
|
||||
|
||||
/* ----------------------- trace ------------------------ */
|
||||
|
||||
/* State of each loom on post-process */
|
||||
struct ovni_loom {
|
||||
size_t nprocs;
|
||||
char hostname[OVNI_MAX_HOSTNAME];
|
||||
char dname[PATH_MAX]; /* Loom directory name */
|
||||
char path[PATH_MAX]; /* Relative to cwd */
|
||||
|
||||
size_t max_ncpus;
|
||||
size_t max_phyid;
|
||||
size_t ncpus;
|
||||
size_t offset_ncpus;
|
||||
struct ovni_cpu *cpu;
|
||||
int rank_enabled;
|
||||
|
||||
int64_t clock_offset;
|
||||
|
||||
/* Virtual CPU */
|
||||
struct ovni_cpu vcpu;
|
||||
|
||||
struct ovni_eproc *proc;
|
||||
};
|
||||
|
||||
#define MAX_VIRTUAL_EVENTS 16
|
||||
|
||||
struct ovni_trace {
|
||||
size_t nlooms;
|
||||
struct ovni_loom *loom;
|
||||
|
||||
size_t nstreams;
|
||||
struct ovni_stream *stream;
|
||||
};
|
||||
|
||||
struct ovni_stream {
|
||||
uint8_t *buf;
|
||||
size_t size;
|
||||
size_t offset;
|
||||
|
||||
int tid;
|
||||
struct ovni_loom *loom;
|
||||
struct ovni_eproc *proc;
|
||||
struct ovni_ethread *thread;
|
||||
int loaded;
|
||||
int active;
|
||||
|
||||
double progress;
|
||||
|
||||
struct ovni_ev *cur_ev;
|
||||
int64_t lastclock;
|
||||
int64_t clock_offset;
|
||||
|
||||
heap_node_t hh;
|
||||
};
|
||||
|
||||
struct ovni_emu {
|
||||
struct ovni_trace trace;
|
||||
|
||||
struct ovni_stream *cur_stream;
|
||||
struct ovni_ev *cur_ev;
|
||||
|
||||
struct ovni_loom *cur_loom;
|
||||
struct ovni_eproc *cur_proc;
|
||||
struct ovni_ethread *cur_thread;
|
||||
|
||||
/* Indexed by gindex */
|
||||
struct ovni_ethread **global_thread;
|
||||
struct ovni_cpu **global_cpu;
|
||||
|
||||
/* Global processed size and offset of all streams */
|
||||
size_t global_size;
|
||||
size_t global_offset;
|
||||
double start_emulation_time;
|
||||
|
||||
int64_t firstclock;
|
||||
int64_t lastclock;
|
||||
int64_t delta_time;
|
||||
|
||||
/* Counters for statistics */
|
||||
int64_t nev_processed;
|
||||
|
||||
/* Be strict */
|
||||
int enable_linter;
|
||||
|
||||
FILE *prv_thread;
|
||||
FILE *prv_cpu;
|
||||
|
||||
struct pcf_file pcf[CHAN_MAXTYPE];
|
||||
|
||||
char *clock_offset_file;
|
||||
char *tracedir;
|
||||
|
||||
/* Total counters */
|
||||
size_t total_nthreads;
|
||||
size_t total_nprocs;
|
||||
size_t total_ncpus;
|
||||
|
||||
uint32_t nosv_type_counter;
|
||||
|
||||
/* Keep a list of dirty channels for the CPUs and threads */
|
||||
struct ovni_chan *cpu_chan;
|
||||
struct ovni_chan *th_chan;
|
||||
|
||||
heap_head_t sorted_stream;
|
||||
};
|
||||
|
||||
/* Emulator function declaration */
|
||||
|
||||
void edie(struct ovni_emu *emu, const char *fmt, ...);
|
||||
void eerr(struct ovni_emu *emu, const char *fmt, ...);
|
||||
|
||||
void hook_init_ovni(struct ovni_emu *emu);
|
||||
void hook_pre_ovni(struct ovni_emu *emu);
|
||||
|
||||
void hook_init_nosv(struct ovni_emu *emu);
|
||||
void hook_pre_nosv(struct ovni_emu *emu);
|
||||
void hook_end_nosv(struct ovni_emu *emu);
|
||||
|
||||
void hook_init_tampi(struct ovni_emu *emu);
|
||||
void hook_pre_tampi(struct ovni_emu *emu);
|
||||
|
||||
void hook_init_openmp(struct ovni_emu *emu);
|
||||
void hook_pre_openmp(struct ovni_emu *emu);
|
||||
|
||||
void hook_init_nodes(struct ovni_emu *emu);
|
||||
void hook_pre_nodes(struct ovni_emu *emu);
|
||||
|
||||
void hook_init_kernel(struct ovni_emu *emu);
|
||||
void hook_pre_kernel(struct ovni_emu *emu);
|
||||
|
||||
void hook_init_nanos6(struct ovni_emu *emu);
|
||||
void hook_pre_nanos6(struct ovni_emu *emu);
|
||||
void hook_end_nanos6(struct ovni_emu *emu);
|
||||
|
||||
struct ovni_cpu *emu_get_cpu(struct ovni_loom *loom, int cpuid);
|
||||
|
||||
struct ovni_ethread *emu_get_thread(struct ovni_eproc *proc, int tid);
|
||||
|
||||
void emu_cpu_update_chan(struct ovni_cpu *cpu, struct ovni_chan *cpu_chan);
|
||||
|
||||
void emu_init(struct ovni_emu *emu, int argc, char *argv[]);
|
||||
void emu_run(struct ovni_emu *emu);
|
||||
void emu_post(struct ovni_emu *emu);
|
||||
void emu_destroy(struct ovni_emu *emu);
|
||||
|
||||
#endif /* OVNI_EMU_H */
|
||||
@ -1,686 +0,0 @@
|
||||
/* Copyright (c) 2021-2022 Barcelona Supercomputing Center (BSC)
|
||||
* SPDX-License-Identifier: GPL-3.0-or-later */
|
||||
|
||||
#include "trace.h"
|
||||
|
||||
#define _GNU_SOURCE
|
||||
#include <unistd.h>
|
||||
|
||||
#include <dirent.h>
|
||||
#include <errno.h>
|
||||
#include <fcntl.h>
|
||||
#include <linux/limits.h>
|
||||
#include <stdatomic.h>
|
||||
#include <stdint.h>
|
||||
#include <stdio.h>
|
||||
#include <stdlib.h>
|
||||
#include <string.h>
|
||||
#include <sys/mman.h>
|
||||
#include <sys/stat.h>
|
||||
#include <time.h>
|
||||
#include <unistd.h>
|
||||
|
||||
static int
|
||||
find_dir_prefix_str(const char *dirname, const char *prefix, const char **str)
|
||||
{
|
||||
const char *p = dirname;
|
||||
|
||||
/* Check the prefix */
|
||||
if (strncmp(p, prefix, strlen(prefix)) != 0)
|
||||
return -1;
|
||||
|
||||
p += strlen(prefix);
|
||||
|
||||
/* Find the dot */
|
||||
if (*p != '.')
|
||||
return -1;
|
||||
|
||||
p++;
|
||||
|
||||
if (str)
|
||||
*str = p;
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
static int
|
||||
find_dir_prefix_int(const char *dirname, const char *prefix, int *num)
|
||||
{
|
||||
const char *p = NULL;
|
||||
|
||||
if (find_dir_prefix_str(dirname, prefix, &p) != 0)
|
||||
return -1;
|
||||
|
||||
/* Convert the suffix string to a number */
|
||||
*num = atoi(p);
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
static size_t
|
||||
count_dir_prefix(DIR *dir, const char *prefix)
|
||||
{
|
||||
struct dirent *dirent;
|
||||
size_t n = 0;
|
||||
|
||||
while ((dirent = readdir(dir)) != NULL) {
|
||||
if (find_dir_prefix_str(dirent->d_name, prefix, NULL) != 0)
|
||||
continue;
|
||||
|
||||
n++;
|
||||
}
|
||||
|
||||
return n;
|
||||
}
|
||||
|
||||
static int
|
||||
load_thread(struct ovni_ethread *thread, struct ovni_eproc *proc, int index, int tid, char *filepath)
|
||||
{
|
||||
static int total_threads = 0;
|
||||
|
||||
thread->tid = tid;
|
||||
thread->index = index;
|
||||
thread->gindex = total_threads++;
|
||||
thread->state = TH_ST_UNKNOWN;
|
||||
thread->proc = proc;
|
||||
|
||||
if (strlen(filepath) >= PATH_MAX) {
|
||||
err("filepath too large: %s\n", filepath);
|
||||
return -1;
|
||||
}
|
||||
|
||||
strcpy(thread->tracefile, filepath);
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
static void
|
||||
load_proc_metadata(struct ovni_eproc *proc, int *rank_enabled)
|
||||
{
|
||||
JSON_Object *meta = json_value_get_object(proc->meta);
|
||||
if (meta == NULL)
|
||||
die("load_proc_metadata: json_value_get_object() failed\n");
|
||||
|
||||
JSON_Value *appid_val = json_object_get_value(meta, "app_id");
|
||||
if (appid_val == NULL)
|
||||
die("process %d is missing app_id in metadata\n", proc->pid);
|
||||
|
||||
proc->appid = (int) json_number(appid_val);
|
||||
|
||||
JSON_Value *rank_val = json_object_get_value(meta, "rank");
|
||||
|
||||
if (rank_val != NULL) {
|
||||
proc->rank = (int) json_number(rank_val);
|
||||
*rank_enabled = 1;
|
||||
} else {
|
||||
proc->rank = -1;
|
||||
}
|
||||
}
|
||||
|
||||
static void
|
||||
check_metadata_version(struct ovni_eproc *proc)
|
||||
{
|
||||
JSON_Object *meta = json_value_get_object(proc->meta);
|
||||
if (meta == NULL)
|
||||
die("check_metadata_version: json_value_get_object() failed\n");
|
||||
|
||||
JSON_Value *version_val = json_object_get_value(meta, "version");
|
||||
if (version_val == NULL) {
|
||||
die("process %d is missing attribute \"version\" in metadata\n",
|
||||
proc->pid);
|
||||
}
|
||||
|
||||
int version = (int) json_number(version_val);
|
||||
|
||||
if (version != OVNI_METADATA_VERSION) {
|
||||
die("pid %d: metadata version mismatch %d (expected %d)\n",
|
||||
proc->pid, version,
|
||||
OVNI_METADATA_VERSION);
|
||||
}
|
||||
|
||||
JSON_Value *mversion_val = json_object_get_value(meta, "model_version");
|
||||
if (mversion_val == NULL) {
|
||||
die("process %d is missing attribute \"model_version\" in metadata\n",
|
||||
proc->pid);
|
||||
}
|
||||
|
||||
const char *mversion = json_string(mversion_val);
|
||||
|
||||
if (strcmp(mversion, OVNI_MODEL_VERSION) != 0) {
|
||||
die("pid %d: metadata model version mismatch '%s' (expected '%s')\n",
|
||||
proc->pid, mversion,
|
||||
OVNI_MODEL_VERSION);
|
||||
}
|
||||
}
|
||||
|
||||
static int
|
||||
compare_int(const void *a, const void *b)
|
||||
{
|
||||
int aa = *(const int *) a;
|
||||
int bb = *(const int *) b;
|
||||
|
||||
if (aa < bb)
|
||||
return -1;
|
||||
else if (aa > bb)
|
||||
return +1;
|
||||
else
|
||||
return 0;
|
||||
}
|
||||
|
||||
static void
|
||||
check_metadata_version(struct ovni_eproc *proc)
|
||||
{
|
||||
JSON_Object *meta = json_value_get_object(proc->meta);
|
||||
if (meta == NULL)
|
||||
die("check_metadata_version: json_value_get_object() failed\n");
|
||||
|
||||
JSON_Value *version_val = json_object_get_value(meta, "version");
|
||||
if (version_val == NULL) {
|
||||
die("process %d is missing attribute \"version\" in metadata\n",
|
||||
proc->pid);
|
||||
}
|
||||
|
||||
int version = (int) json_number(version_val);
|
||||
|
||||
if (version != OVNI_METADATA_VERSION) {
|
||||
die("pid %d: metadata version mismatch %d (expected %d)\n",
|
||||
proc->pid, version,
|
||||
OVNI_METADATA_VERSION);
|
||||
}
|
||||
|
||||
JSON_Value *mversion_val = json_object_get_value(meta, "model_version");
|
||||
if (mversion_val == NULL) {
|
||||
die("process %d is missing attribute \"model_version\" in metadata\n",
|
||||
proc->pid);
|
||||
}
|
||||
|
||||
const char *mversion = json_string(mversion_val);
|
||||
|
||||
if (strcmp(mversion, OVNI_MODEL_VERSION) != 0) {
|
||||
die("pid %d: metadata model version mismatch '%s' (expected '%s')\n",
|
||||
proc->pid, mversion,
|
||||
OVNI_MODEL_VERSION);
|
||||
}
|
||||
}
|
||||
|
||||
static int
|
||||
load_proc(struct ovni_eproc *proc, struct ovni_loom *loom, int index, int pid, char *procdir)
|
||||
{
|
||||
static int total_procs = 0;
|
||||
|
||||
proc->pid = pid;
|
||||
proc->index = index;
|
||||
proc->gindex = total_procs++;
|
||||
proc->loom = loom;
|
||||
|
||||
char path[PATH_MAX];
|
||||
if (snprintf(path, PATH_MAX, "%s/%s", procdir, "metadata.json") >= PATH_MAX) {
|
||||
err("snprintf: path too large: %s\n", procdir);
|
||||
abort();
|
||||
}
|
||||
|
||||
proc->meta = json_parse_file_with_comments(path);
|
||||
if (proc->meta == NULL) {
|
||||
err("error loading metadata from %s\n", path);
|
||||
return -1;
|
||||
}
|
||||
|
||||
check_metadata_version(proc);
|
||||
|
||||
/* The appid is populated from the metadata */
|
||||
load_proc_metadata(proc, &loom->rank_enabled);
|
||||
|
||||
DIR *dir;
|
||||
if ((dir = opendir(procdir)) == NULL) {
|
||||
fprintf(stderr, "opendir %s failed: %s\n",
|
||||
procdir, strerror(errno));
|
||||
return -1;
|
||||
}
|
||||
|
||||
proc->nthreads = count_dir_prefix(dir, "thread");
|
||||
|
||||
if (proc->nthreads <= 0) {
|
||||
err("cannot find any thread for process %d\n",
|
||||
proc->pid);
|
||||
return -1;
|
||||
}
|
||||
|
||||
proc->thread = calloc(proc->nthreads, sizeof(struct ovni_ethread));
|
||||
|
||||
if (proc->thread == NULL) {
|
||||
perror("calloc failed");
|
||||
return -1;
|
||||
}
|
||||
|
||||
int *tids;
|
||||
|
||||
if ((tids = calloc(proc->nthreads, sizeof(int))) == NULL) {
|
||||
perror("calloc failed\n");
|
||||
return -1;
|
||||
}
|
||||
|
||||
rewinddir(dir);
|
||||
|
||||
for (size_t i = 0; i < proc->nthreads;) {
|
||||
struct dirent *dirent = readdir(dir);
|
||||
|
||||
if (dirent == NULL) {
|
||||
err("inconsistent: readdir returned NULL\n");
|
||||
return -1;
|
||||
}
|
||||
|
||||
if (find_dir_prefix_int(dirent->d_name, "thread", &tids[i]) != 0)
|
||||
continue;
|
||||
|
||||
i++;
|
||||
}
|
||||
|
||||
closedir(dir);
|
||||
|
||||
/* Sort threads by ascending TID */
|
||||
qsort(tids, proc->nthreads, sizeof(int), compare_int);
|
||||
|
||||
for (size_t i = 0; i < proc->nthreads; i++) {
|
||||
int tid = tids[i];
|
||||
|
||||
if (snprintf(path, PATH_MAX, "%s/thread.%d", procdir, tid) >= PATH_MAX) {
|
||||
err("snprintf: path too large: %s\n", procdir);
|
||||
abort();
|
||||
}
|
||||
|
||||
struct ovni_ethread *thread = &proc->thread[i];
|
||||
|
||||
if (load_thread(thread, proc, i, tid, path) != 0)
|
||||
return -1;
|
||||
}
|
||||
|
||||
free(tids);
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
static int
|
||||
load_loom(struct ovni_loom *loom, char *loomdir)
|
||||
{
|
||||
DIR *dir = NULL;
|
||||
|
||||
if ((dir = opendir(loomdir)) == NULL) {
|
||||
fprintf(stderr, "opendir %s failed: %s\n",
|
||||
loomdir, strerror(errno));
|
||||
return -1;
|
||||
}
|
||||
|
||||
loom->rank_enabled = 0;
|
||||
loom->nprocs = count_dir_prefix(dir, "proc");
|
||||
|
||||
if (loom->nprocs <= 0) {
|
||||
err("cannot find any process directory in loom %s\n",
|
||||
loom->hostname);
|
||||
return -1;
|
||||
}
|
||||
|
||||
loom->proc = calloc(loom->nprocs, sizeof(struct ovni_eproc));
|
||||
|
||||
if (loom->proc == NULL) {
|
||||
perror("calloc failed");
|
||||
return -1;
|
||||
}
|
||||
|
||||
rewinddir(dir);
|
||||
|
||||
size_t i = 0;
|
||||
struct dirent *dirent = NULL;
|
||||
while ((dirent = readdir(dir)) != NULL) {
|
||||
int pid;
|
||||
if (find_dir_prefix_int(dirent->d_name, "proc", &pid) != 0)
|
||||
continue;
|
||||
|
||||
if (i >= loom->nprocs) {
|
||||
err("more process than expected\n");
|
||||
abort();
|
||||
}
|
||||
|
||||
struct ovni_eproc *proc = &loom->proc[i];
|
||||
|
||||
if (snprintf(proc->dir, PATH_MAX, "%s/%s", loomdir, dirent->d_name) >= PATH_MAX) {
|
||||
err("error: process dir name %s too long\n", dirent->d_name);
|
||||
return -1;
|
||||
}
|
||||
|
||||
if (load_proc(&loom->proc[i], loom, i, pid, proc->dir) != 0)
|
||||
return -1;
|
||||
|
||||
i++;
|
||||
}
|
||||
|
||||
if (i != loom->nprocs) {
|
||||
err("unexpected number of processes\n");
|
||||
abort();
|
||||
}
|
||||
|
||||
closedir(dir);
|
||||
|
||||
/* Ensure all process have the rank, if enabled in any */
|
||||
if (loom->rank_enabled) {
|
||||
for (i = 0; i < loom->nprocs; i++) {
|
||||
struct ovni_eproc *proc = &loom->proc[i];
|
||||
if (proc->rank < 0) {
|
||||
die("process %d is missing the rank\n",
|
||||
proc->pid);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
static int
|
||||
compare_looms(const void *a, const void *b)
|
||||
{
|
||||
struct ovni_loom *la = (struct ovni_loom *) a;
|
||||
struct ovni_loom *lb = (struct ovni_loom *) b;
|
||||
return strcmp(la->dname, lb->dname);
|
||||
}
|
||||
|
||||
static void
|
||||
loom_to_host(const char *loom_name, char *host, int n)
|
||||
{
|
||||
int i = 0;
|
||||
for (i = 0; i < n; i++) {
|
||||
/* Copy until dot or end */
|
||||
if (loom_name[i] != '.' && loom_name[i] != '\0')
|
||||
host[i] = loom_name[i];
|
||||
else
|
||||
break;
|
||||
}
|
||||
|
||||
if (i == n)
|
||||
die("loom host name %s too long\n", loom_name);
|
||||
|
||||
host[i] = '\0';
|
||||
}
|
||||
|
||||
int
|
||||
ovni_load_trace(struct ovni_trace *trace, char *tracedir)
|
||||
{
|
||||
DIR *dir = NULL;
|
||||
|
||||
if ((dir = opendir(tracedir)) == NULL) {
|
||||
err("opendir %s failed: %s\n", tracedir, strerror(errno));
|
||||
return -1;
|
||||
}
|
||||
|
||||
trace->nlooms = count_dir_prefix(dir, "loom");
|
||||
|
||||
if (trace->nlooms == 0) {
|
||||
err("cannot find any loom in %s\n", tracedir);
|
||||
return -1;
|
||||
}
|
||||
|
||||
trace->loom = calloc(trace->nlooms, sizeof(struct ovni_loom));
|
||||
|
||||
if (trace->loom == NULL) {
|
||||
perror("calloc failed\n");
|
||||
return -1;
|
||||
}
|
||||
|
||||
rewinddir(dir);
|
||||
|
||||
size_t l = 0;
|
||||
struct dirent *dirent = NULL;
|
||||
|
||||
while ((dirent = readdir(dir)) != NULL) {
|
||||
struct ovni_loom *loom = &trace->loom[l];
|
||||
const char *loom_name;
|
||||
if (find_dir_prefix_str(dirent->d_name, "loom", &loom_name) != 0) {
|
||||
/* Ignore other files in tracedir */
|
||||
continue;
|
||||
}
|
||||
|
||||
if (l >= trace->nlooms) {
|
||||
err("extra loom detected\n");
|
||||
return -1;
|
||||
}
|
||||
|
||||
/* Copy the complete loom directory name to looms */
|
||||
if (snprintf(loom->dname, PATH_MAX, "%s", dirent->d_name) >= PATH_MAX) {
|
||||
err("error: loom name %s too long\n", dirent->d_name);
|
||||
return -1;
|
||||
}
|
||||
|
||||
l++;
|
||||
}
|
||||
|
||||
closedir(dir);
|
||||
|
||||
/* Sort the looms, so we get the hostnames in alphanumeric order */
|
||||
qsort(trace->loom, trace->nlooms, sizeof(struct ovni_loom),
|
||||
compare_looms);
|
||||
|
||||
for (size_t i = 0; i < trace->nlooms; i++) {
|
||||
struct ovni_loom *loom = &trace->loom[i];
|
||||
const char *name = NULL;
|
||||
|
||||
if (find_dir_prefix_str(loom->dname, "loom", &name) != 0) {
|
||||
err("error: mismatch for loom %s\n", loom->dname);
|
||||
return -1;
|
||||
}
|
||||
|
||||
loom_to_host(name, loom->hostname, sizeof(loom->hostname));
|
||||
|
||||
if (snprintf(loom->path, PATH_MAX, "%s/%s",
|
||||
tracedir, loom->dname)
|
||||
>= PATH_MAX) {
|
||||
err("error: loom path %s/%s too long\n",
|
||||
tracedir, loom->dname);
|
||||
return -1;
|
||||
}
|
||||
|
||||
if (load_loom(loom, loom->path) != 0)
|
||||
return -1;
|
||||
}
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
static int
|
||||
check_stream_header(struct ovni_stream *stream)
|
||||
{
|
||||
int ret = 0;
|
||||
|
||||
if (stream->size < sizeof(struct ovni_stream_header)) {
|
||||
err("stream %d: incomplete stream header\n",
|
||||
stream->tid);
|
||||
return -1;
|
||||
}
|
||||
|
||||
struct ovni_stream_header *h =
|
||||
(struct ovni_stream_header *) stream->buf;
|
||||
|
||||
if (memcmp(h->magic, OVNI_STREAM_MAGIC, 4) != 0) {
|
||||
char magic[5];
|
||||
memcpy(magic, h->magic, 4);
|
||||
magic[4] = '\0';
|
||||
err("stream %d: wrong stream magic '%s' (expected '%s')\n",
|
||||
stream->tid, magic, OVNI_STREAM_MAGIC);
|
||||
ret = -1;
|
||||
}
|
||||
|
||||
if (h->version != OVNI_STREAM_VERSION) {
|
||||
err("stream %d: stream version mismatch %u (expected %u)\n",
|
||||
stream->tid, h->version, OVNI_STREAM_VERSION);
|
||||
ret = -1;
|
||||
}
|
||||
|
||||
return ret;
|
||||
}
|
||||
|
||||
static int
|
||||
load_stream_fd(struct ovni_stream *stream, int fd)
|
||||
{
|
||||
struct stat st;
|
||||
if (fstat(fd, &st) < 0) {
|
||||
perror("fstat failed");
|
||||
return -1;
|
||||
}
|
||||
|
||||
/* Error because it doesn't have the header */
|
||||
if (st.st_size == 0) {
|
||||
err("stream %d is empty\n", stream->tid);
|
||||
return -1;
|
||||
}
|
||||
|
||||
int prot = PROT_READ | PROT_WRITE;
|
||||
stream->buf = mmap(NULL, st.st_size, prot, MAP_PRIVATE, fd, 0);
|
||||
|
||||
if (stream->buf == MAP_FAILED) {
|
||||
perror("mmap failed");
|
||||
return -1;
|
||||
}
|
||||
|
||||
stream->size = st.st_size;
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
static int
|
||||
load_stream_buf(struct ovni_stream *stream, struct ovni_ethread *thread)
|
||||
{
|
||||
int fd;
|
||||
|
||||
if ((fd = open(thread->tracefile, O_RDWR)) == -1) {
|
||||
perror("open failed");
|
||||
return -1;
|
||||
}
|
||||
|
||||
if (load_stream_fd(stream, fd) != 0)
|
||||
return -1;
|
||||
|
||||
if (check_stream_header(stream) != 0) {
|
||||
err("stream %d: bad header\n", stream->tid);
|
||||
return -1;
|
||||
}
|
||||
|
||||
stream->offset = sizeof(struct ovni_stream_header);
|
||||
|
||||
if (stream->offset == stream->size)
|
||||
stream->active = 0;
|
||||
else
|
||||
stream->active = 1;
|
||||
|
||||
/* No need to keep the fd open */
|
||||
if (close(fd)) {
|
||||
perror("close failed");
|
||||
return -1;
|
||||
}
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
/* Populates the streams in a single array */
|
||||
int
|
||||
ovni_load_streams(struct ovni_trace *trace)
|
||||
{
|
||||
trace->nstreams = 0;
|
||||
|
||||
for (size_t i = 0; i < trace->nlooms; i++) {
|
||||
struct ovni_loom *loom = &trace->loom[i];
|
||||
for (size_t j = 0; j < loom->nprocs; j++) {
|
||||
struct ovni_eproc *proc = &loom->proc[j];
|
||||
for (size_t k = 0; k < proc->nthreads; k++) {
|
||||
trace->nstreams++;
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
trace->stream = calloc(trace->nstreams, sizeof(struct ovni_stream));
|
||||
|
||||
if (trace->stream == NULL) {
|
||||
perror("calloc failed");
|
||||
return -1;
|
||||
}
|
||||
|
||||
err("loaded %ld streams\n", trace->nstreams);
|
||||
|
||||
size_t s = 0;
|
||||
for (size_t i = 0; i < trace->nlooms; i++) {
|
||||
struct ovni_loom *loom = &trace->loom[i];
|
||||
for (size_t j = 0; j < loom->nprocs; j++) {
|
||||
struct ovni_eproc *proc = &loom->proc[j];
|
||||
for (size_t k = 0; k < proc->nthreads; k++) {
|
||||
struct ovni_ethread *thread = &proc->thread[k];
|
||||
struct ovni_stream *stream = &trace->stream[s++];
|
||||
|
||||
stream->tid = thread->tid;
|
||||
stream->thread = thread;
|
||||
stream->proc = proc;
|
||||
stream->loom = loom;
|
||||
stream->lastclock = 0;
|
||||
stream->offset = 0;
|
||||
stream->cur_ev = NULL;
|
||||
|
||||
if (load_stream_buf(stream, thread) != 0) {
|
||||
err("load_stream_buf failed\n");
|
||||
return -1;
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
return 0;
|
||||
}
|
||||
|
||||
void
|
||||
ovni_free_streams(struct ovni_trace *trace)
|
||||
{
|
||||
for (size_t i = 0; i < trace->nstreams; i++) {
|
||||
struct ovni_stream *stream = &trace->stream[i];
|
||||
if (munmap(stream->buf, stream->size) != 0)
|
||||
die("munmap stream failed: %s\n", strerror(errno));
|
||||
}
|
||||
|
||||
free(trace->stream);
|
||||
}
|
||||
|
||||
void
|
||||
ovni_free_trace(struct ovni_trace *trace)
|
||||
{
|
||||
for (size_t i = 0; i < trace->nlooms; i++) {
|
||||
for (size_t j = 0; j < trace->loom[i].nprocs; j++) {
|
||||
free(trace->loom[i].proc[j].thread);
|
||||
}
|
||||
|
||||
free(trace->loom[i].proc);
|
||||
}
|
||||
|
||||
free(trace->loom);
|
||||
}
|
||||
|
||||
int
|
||||
ovni_load_next_event(struct ovni_stream *stream)
|
||||
{
|
||||
if (stream->active == 0) {
|
||||
dbg("stream is inactive, cannot load more events\n");
|
||||
return -1;
|
||||
}
|
||||
|
||||
/* Only step the offset if we have load an event */
|
||||
if (stream->cur_ev != NULL)
|
||||
stream->offset += ovni_ev_size(stream->cur_ev);
|
||||
|
||||
/* It cannot overflow, otherwise we are reading garbage */
|
||||
if (stream->offset > stream->size)
|
||||
die("ovni_load_next_event: stream offset exceeds size\n");
|
||||
|
||||
/* We have reached the end */
|
||||
if (stream->offset == stream->size) {
|
||||
stream->active = 0;
|
||||
stream->cur_ev = NULL;
|
||||
dbg("stream %d runs out of events\n", stream->tid);
|
||||
return -1;
|
||||
}
|
||||
|
||||
stream->cur_ev = (struct ovni_ev *) &stream->buf[stream->offset];
|
||||
|
||||
return 0;
|
||||
}
|
||||
@ -1,20 +0,0 @@
|
||||
/* Copyright (c) 2021 Barcelona Supercomputing Center (BSC)
|
||||
* SPDX-License-Identifier: GPL-3.0-or-later */
|
||||
|
||||
#ifndef OVNI_TRACE_H
|
||||
#define OVNI_TRACE_H
|
||||
|
||||
#include "emu.h"
|
||||
#include "ovni.h"
|
||||
|
||||
int ovni_load_next_event(struct ovni_stream *stream);
|
||||
|
||||
int ovni_load_trace(struct ovni_trace *trace, char *tracedir);
|
||||
|
||||
int ovni_load_streams(struct ovni_trace *trace);
|
||||
|
||||
void ovni_free_streams(struct ovni_trace *trace);
|
||||
|
||||
void ovni_free_trace(struct ovni_trace *trace);
|
||||
|
||||
#endif /* OVNI_TRACE_H */
|
||||
Loading…
Reference in New Issue
Block a user