More edge case testing on the multi-threading, preppared to start moving heavy code back

This commit is contained in:
2025-10-15 21:35:45 -04:00
parent bc742b2116
commit 724b3eeba5
5 changed files with 191 additions and 80 deletions

View File

@@ -10,6 +10,7 @@ package host
//region HOST RUNTIME
load_client_api :: proc(version_id: int) -> (loaded_module: Client_API) {
profile(#procedure)
using loaded_module
// Make sure we have a dll to work with
file_io_err: OS_Error; write_time, file_io_err = file_last_write_time_by_name("sectr.dll")
@@ -17,12 +18,13 @@ load_client_api :: proc(version_id: int) -> (loaded_module: Client_API) {
panic_contextless( "Could not resolve the last write time for sectr")
}
//TODO(Ed): Lets try to minimize this...
thread_sleep( Millisecond * 50 )
thread_sleep( Millisecond * 25 )
// Get the live dll loaded up
file_copy_sync( Path_Sectr_Module, Path_Sectr_Live_Module, allocator = context.temp_allocator )
did_load: bool; lib, did_load = os_lib_load( Path_Sectr_Live_Module )
if ! did_load do panic( "Failed to load the sectr module.")
startup = cast( type_of( host_memory.client_api.startup)) os_lib_get_proc(lib, "startup")
shutdown = cast( type_of( host_memory.client_api.shutdown)) os_lib_get_proc(lib, "sectr_shutdown")
tick_lane_startup = cast( type_of( host_memory.client_api.tick_lane_startup)) os_lib_get_proc(lib, "tick_lane_startup")
job_worker_startup = cast( type_of( host_memory.client_api.job_worker_startup)) os_lib_get_proc(lib, "job_worker_startup")
hot_reload = cast( type_of( host_memory.client_api.hot_reload)) os_lib_get_proc(lib, "hot_reload")
@@ -30,6 +32,7 @@ load_client_api :: proc(version_id: int) -> (loaded_module: Client_API) {
clean_frame = cast( type_of( host_memory.client_api.clean_frame)) os_lib_get_proc(lib, "clean_frame")
jobsys_worker_tick = cast( type_of( host_memory.client_api.jobsys_worker_tick)) os_lib_get_proc(lib, "jobsys_worker_tick")
if startup == nil do panic("Failed to load sectr.startup symbol" )
if shutdown == nil do panic("Failed to load sectr.shutdown symbol" )
if tick_lane_startup == nil do panic("Failed to load sectr.tick_lane_startup symbol" )
if job_worker_startup == nil do panic("Failed to load sectr.job_worker_startup symbol" )
if hot_reload == nil do panic("Failed to load sectr.hot_reload symbol" )
@@ -69,6 +72,7 @@ main :: proc()
host_memory.spall_context = spall_context_create(Path_Sectr_Spall_Record)
grime_set_profiler_module_context(& host_memory.spall_context)
thread_memory.spall_buffer = spall_buffer_create(thread_memory.spall_buffer_backing[:], cast(u32) thread_memory.system_ctx.id)
grime_set_profiler_thread_buffer(& thread_memory.spall_buffer)
}
// Setup the logger
path_logger_finalized: string
@@ -88,8 +92,8 @@ main :: proc()
timestamp := str_pfmt_buffer( fmt_backing, "%04d-%02d-%02d_%02d-%02d-%02d", year, month, day, hour, min, sec)
host_memory.path_logger_finalized = str_pfmt_buffer( fmt_backing, "%s/sectr_%v.log", Path_Logs, timestamp)
}
logger_init( & host_memory.logger, "Sectr Host", str_pfmt_buffer( fmt_backing, "%s/sectr.log", Path_Logs ) )
context.logger = to_odin_logger( & host_memory.logger )
logger_init( & host_memory.host_logger, "Sectr Host", str_pfmt_buffer( fmt_backing, "%s/sectr.log", Path_Logs ) )
context.logger = to_odin_logger( & host_memory.host_logger )
{
// Log System Context
builder := strbuilder_from_bytes( fmt_backing )
@@ -99,7 +103,7 @@ main :: proc()
}
free_all(context.temp_allocator)
}
context.logger = to_odin_logger( & host_memory.logger )
context.logger = to_odin_logger( & host_memory.host_logger )
// Load the Enviornment API for the first-time
{
host_memory.client_api = load_client_api( 1 )
@@ -107,22 +111,23 @@ main :: proc()
}
// Client API Startup
host_memory.client_api.startup(& host_memory, & thread_memory)
// Start the tick lanes
{
profile("thread_wide_startup")
assert(thread_memory.id == .Master_Prepper)
host_memory.tick_running = true
host_memory.tick_lanes = THREAD_TICK_LANES
barrier_init(& host_memory.lane_sync, THREAD_TICK_LANES)
when THREAD_TICK_LANES > 1 {
for id in 1 ..= (THREAD_TICK_LANES - 1) {
launch_tick_lane_thread(cast(WorkerID) id)
{
profile("Tick Lanes")
host_memory.tick_running = true
host_memory.tick_lanes = THREAD_TICK_LANES
barrier_init(& host_memory.lane_sync, THREAD_TICK_LANES)
when THREAD_TICK_LANES > 1 {
for id in 1 ..= (THREAD_TICK_LANES - 1) {
launch_tick_lane_thread(cast(WorkerID) id)
}
}
}
grime_set_profiler_module_context(& host_memory.spall_context)
grime_set_profiler_thread_buffer(& thread_memory.spall_buffer)
// Job System Setup
{
profile("Job System")
host_memory.job_system.running = true
host_memory.job_system.worker_num = THREAD_JOB_WORKERS
for & list in host_memory.job_system.job_lists {
@@ -131,6 +136,7 @@ main :: proc()
// Determine number of physical cores
barrier_init(& host_memory.job_hot_reload_sync, THREAD_JOB_WORKERS + 1)
for id in THREAD_JOB_WORKER_ID_START ..< THREAD_JOB_WORKER_ID_END {
log_print_fmt("Spawned job worker: %v", cast(WorkerID) id)
worker_thread := thread_create(host_job_worker_entrypoint, .Normal)
worker_thread.user_index = int(id)
host_memory.threads[worker_thread.user_index] = worker_thread
@@ -141,18 +147,21 @@ main :: proc()
}
host_tick_lane()
profile_begin("Host Shutdown")
if thread_memory.id == .Master_Prepper {
thread_join_multiple(.. host_memory.threads[1:THREAD_TICK_LANES + THREAD_JOB_WORKERS])
}
unload_client_api( & host_memory.client_api )
// End profiling
spall_context_destroy( & host_memory.spall_context )
log_print("Succesfuly closed")
file_close( host_memory.logger.file )
file_close( host_memory.host_logger.file )
file_rename( str_pfmt_tmp( "%s/sectr.log", Path_Logs), host_memory.path_logger_finalized )
profile_end()
// End profiling
spall_buffer_destroy(& host_memory.spall_context, & thread_memory.spall_buffer)
spall_context_destroy( & host_memory.spall_context )
}
launch_tick_lane_thread :: proc(id : WorkerID) {
assert_contextless(thread_memory.id == .Master_Prepper)
@@ -183,12 +192,13 @@ host_tick_lane :: proc()
for ; sync_load(& host_memory.tick_running, .Relaxed);
{
profile("Host Tick")
leader := barrier_wait(& host_memory.lane_sync)
running: b64 = host_memory.client_api.tick_lane(duration_seconds(delta_ns), delta_ns) == false
if thread_memory.id == .Master_Prepper {
sync_store(& host_memory.tick_running, running, .Release)
}
// host_memory.client_api.clean_frame()
host_memory.client_api.clean_frame()
delta_ns = time_tick_lap_time( & host_tick )
host_tick = time_tick_now()
@@ -202,10 +212,25 @@ host_lane_shutdown :: proc()
{
profile(#procedure)
if thread_memory.id == .Master_Prepper {
jobs_enqueued := true
if jobs_enqueued == false {
// debug_trap()
}
for ; jobs_enqueued; {
jobs_enqueued = false
jobs_enqueued |= host_memory.job_system.job_lists[.Normal].head != nil
jobs_enqueued |= host_memory.job_system.job_lists[.Low].head != nil
jobs_enqueued |= host_memory.job_system.job_lists[.High].head != nil
if jobs_enqueued == false {
// debug_trap()
}
}
sync_store(& host_memory.job_system.running, false, .Release)
}
if thread_memory.id != .Master_Prepper {
spall_buffer_destroy( & host_memory.spall_context, & thread_memory.spall_buffer )
}
leader := barrier_wait(& host_memory.lane_job_sync)
// spall_buffer_destroy( & host_memory.spall_context, & thread_memory.spall_buffer )
}
host_job_worker_entrypoint :: proc(worker_thread: ^SysThread)
@@ -218,9 +243,13 @@ host_job_worker_entrypoint :: proc(worker_thread: ^SysThread)
host_memory.client_api.tick_lane_startup(& thread_memory)
grime_set_profiler_thread_buffer(& thread_memory.spall_buffer)
}
jobs_enqueued := false
jobs_enqueued |= host_memory.job_system.job_lists[.Normal].head != nil
jobs_enqueued |= host_memory.job_system.job_lists[.Low].head != nil
jobs_enqueued |= host_memory.job_system.job_lists[.High].head != nil
delta_ns: Duration
host_tick := time_tick_now()
for ; sync_load(& host_memory.job_system.running, .Relaxed);
for ; jobs_enqueued || sync_load(& host_memory.job_system.running, .Relaxed);
{
profile("Host Job Tick")
@@ -229,7 +258,12 @@ host_job_worker_entrypoint :: proc(worker_thread: ^SysThread)
delta_ns = time_tick_lap_time( & host_tick )
host_tick = time_tick_now()
if sync_load(& host_memory.client_api_hot_reloaded, .Acquire) {
jobs_enqueued = false
jobs_enqueued |= host_memory.job_system.job_lists[.Normal].head != nil
jobs_enqueued |= host_memory.job_system.job_lists[.Low].head != nil
jobs_enqueued |= host_memory.job_system.job_lists[.High].head != nil
if jobs_enqueued == false && sync_load(& host_memory.client_api_hot_reloaded, .Acquire) {
// Signals to main hread when all jobs have drained.
leader :=barrier_wait(& host_memory.job_hot_reload_sync)
// Job threads wait here until client module is back
@@ -237,6 +271,7 @@ host_job_worker_entrypoint :: proc(worker_thread: ^SysThread)
host_memory.client_api.hot_reload(& host_memory, & thread_memory)
}
}
spall_buffer_destroy( & host_memory.spall_context, & thread_memory.spall_buffer )
// Were exiting, wait for tick lanes.
leader := barrier_wait(& host_memory.lane_job_sync)
}
@@ -278,6 +313,7 @@ sync_client_api :: proc()
}
unload_client_api :: proc( module : ^Client_API )
{
profile(#procedure)
os_lib_unload( module.lib )
file_remove( Path_Sectr_Live_Module )
module^ = {}

View File

@@ -16,11 +16,12 @@ ModuleAPI :: struct {
write_time: FileTime,
lib_version : int,
startup: type_of( startup ),
startup: type_of( startup),
shutdown: type_of( sectr_shutdown),
tick_lane_startup: type_of( tick_lane_startup),
job_worker_startup: type_of( job_worker_startup),
hot_reload: type_of( hot_reload ),
tick_lane: type_of( tick_lane ),
hot_reload: type_of( hot_reload),
tick_lane: type_of( tick_lane),
clean_frame: type_of( clean_frame),
jobsys_worker_tick: type_of( jobsys_worker_tick)
}
@@ -35,13 +36,18 @@ then prepare for multi-threaded "laned" tick: thread_wide_startup.
startup :: proc(host_mem: ^ProcessMemory, thread_mem: ^ThreadMemory)
{
// Rad Debugger driving me crazy..
// NOTE(Ed): This is problably not necessary, they're just loops for my sanity.
// NOTE(Ed): This is not necessary, they're just loops for my sanity.
for ; memory == nil; { memory = host_mem }
for ; thread == nil; { thread = thread_mem }
grime_set_profiler_module_context(& memory.spall_context)
grime_set_profiler_thread_buffer(& thread.spall_buffer)
profile(#procedure)
startup_tick := tick_now()
logger_init(& memory.client_memory.logger, "Sectr", memory.host_logger.file_path, memory.host_logger.file)
context.logger = to_odin_logger(& memory.client_memory.logger)
using memory.client_memory
// Configuration Load
@@ -78,7 +84,34 @@ startup :: proc(host_mem: ^ProcessMemory, thread_mem: ^ThreadMemory)
Desired_OS_Scheduler_MS :: 1
sleep_is_granular = set__scheduler_granularity( Desired_OS_Scheduler_MS )
// TODO(Ed): String Cache (Not backed by slab!)
// TODO(Ed): Setup input system
// TODO(Ed): Setup sokol_app
// TODO(Ed): Setup sokol_gfx
// TODO(Ed): Setup sokol_gp
// TODO(Ed): Use job system to load fonts!!!
// TODO(Ed): Setup screen ui state
// TODO(Ed): Setup proper workspace scaffold
startup_ms := duration_ms( tick_lap_time( & startup_tick))
log_print_fmt("Startup time: %v ms", startup_ms)
}
// For some reason odin's symbols conflict with native foreign symbols...
@export
sectr_shutdown :: proc()
{
context.logger = to_odin_logger(& memory.client_memory.logger)
// TODO(Ed): Shut down font system
// TODO(Ed): Shutdown sokol gp, gfx, and app.
log_print("Client module shutdown complete")
}
/*
@@ -105,12 +138,21 @@ hot_reload :: proc(host_mem: ^ProcessMemory, thread_mem: ^ThreadMemory)
profile(#procedure)
// Do hot-reload stuff...
{
// Test dispatching 64 jobs during hot_reload loop (when the above store is uncommented)
for job_id := 1; job_id < 64; job_id += 1 {
memory.job_info_reload[job_id].id = job_id
memory.job_reload[job_id] = make_job_raw(& memory.job_group_reload, & memory.job_info_reload[job_id], test_job, {}, "Job Test (Hot-Reload)")
job_dispatch_single(& memory.job_reload[job_id], .Normal)
context.logger = to_odin_logger(& memory.client_memory.logger)
// TODO(Ed): Setup context alloators
// TODO(Ed): Patch Sokol contextes
// We hopefully don't have to patch third-party allocators anymore per-hot-reload.
{
}
// TODO(Ed): Reload the font system
log_print("Module reloaded")
}
// Critical reference synchronization
{
@@ -158,37 +200,68 @@ Host handles the loop.
tick_lane :: proc(host_delta_time_ms: f64, host_delta_ns: Duration) -> (should_close: b64 = false)
{
profile(#procedure)
@thread_local dummy: int = 0
dummy += 1
EXIT_TIME :: 1
// profile_begin("sokol_app: pre_client_tick")
// should_close |= cast(b64) sokol_app.pre_client_frame()
@static timer: f64
if thread.id == .Master_Prepper {
timer += host_delta_time_ms
// sync_store(& should_close, timer > EXIT_TIME, .Release)
}
// profile_end()
profile_begin("Client Tick")
if thread.id == .Master_Prepper && timer > EXIT_TIME {
// Test dispatching 64 jobs during the last iteration before exiting.
for job_id := 1; job_id < 64; job_id += 1 {
memory.job_info_exit[job_id].id = job_id
memory.job_exit[job_id] = make_job_raw(& memory.job_group_exit, & memory.job_info_exit[job_id], test_job, {}, "Job Test (Exit)")
job_dispatch_single(& memory.job_exit[job_id], .Normal)
}
}
client_tick := tick_now()
profile_begin("sokol_app: pre_client_tick")
// should_close |= cast(b64) sokol_app.pre_client_frame() // TODO(Ed): SOKOL!
profile_end()
// profile_begin("sokol_app: post_client_tick")
// profile_end()
profile_begin("Client Tick")
{
profile("Work frame")
context.logger = to_odin_logger( & memory.client_memory.logger )
// TODO(Ed): Setup frame alloator
if thread.id == .Master_Prepper
{
// config := & memory.client_memory.config
// debug := & memory.client_memory.debug
// debug.draw_ui_box_bounds_points = false
// debug.draw_ui_padding_bounds = false
// debug.draw_ui_content_bounds = false
// config.engine_refresh_hz = 165
// config.color_theme = App_Thm_Light
// config.color_theme = App_Thm_Dusk
// config.color_theme = App_Thm_Dark
// sokol_width := sokol_app.widthf()
// sokol_height := sokol_app.heightf()
// window := & get_state().app_window
// if int(window.extent.x) != int(sokol_width) || int(window.extent.y) != int(sokol_height) {
// window.resized = true
// window.extent.x = sokol_width * 0.5
// window.extent.y = sokol_height * 0.5
// log("sokol_app: Event-based frame callback triggered (detected a resize")
// }
}
// Test dispatching 64 jobs during hot_reload loop (when the above store is uncommented)
if true
{
if thread.id == .Master_Prepper {
profile("dispatching")
for job_id := 1; job_id < JOB_TEST_NUM; job_id += 1 {
memory.job_info_reload[job_id].id = job_id
memory.job_reload[job_id] = make_job_raw(& memory.job_group_reload, & memory.job_info_reload[job_id], test_job, {}, "Job Test (Hot-Reload)")
job_dispatch_single(& memory.job_reload[job_id], .Normal)
}
}
should_close = true
}
// should_close |= update( host_delta_time_ms )
// render()
}
client_tick := tick_now()
profile_end()
profile_begin("sokol_app: post_client_tick")
// sokol_app.post_client_frame() // TODO(Ed): SOKOL!
profile_end()
tick_lane_frametime(& client_tick, host_delta_time_ms, host_delta_ns)
return sync_load(& should_close, .Acquire)
@@ -197,7 +270,8 @@ tick_lane :: proc(host_delta_time_ms: f64, host_delta_ns: Duration) -> (should_c
@export
jobsys_worker_tick :: proc(host_delta_time_ms: f64, host_delta_ns: Duration)
{
// profile("Worker Tick")
profile("Worker Tick")
context.logger = to_odin_logger(& memory.client_memory.logger)
ORDERED_PRIORITIES :: [len(JobPriority)]JobPriority{.High, .Normal, .Low}
block: for priority in ORDERED_PRIORITIES
@@ -225,11 +299,9 @@ jobsys_worker_tick :: proc(host_delta_time_ms: f64, host_delta_ns: Duration)
sync_mutex_unlock(& memory.job_system.job_lists[priority].mutex)
}
}
// Updating worker frametime
// Updating worker timing
{
// TODO(Ed): Setup
// TODO(Ed): Setup timing
}
}
@@ -240,7 +312,7 @@ test_job :: proc(data: rawptr)
{
profile(#procedure)
info := cast(^TestJobInfo) data
// log_print_fmt("Test job succeeded: %v", info.id)
log_print_fmt("Test job succeeded: %v", info.id)
}
Frametime_High_Perf_Threshold_MS :: 1 / 240.0
@@ -248,14 +320,13 @@ Frametime_High_Perf_Threshold_MS :: 1 / 240.0
tick_lane_frametime :: proc(client_tick: ^Tick, host_delta_time_ms: f64, host_delta_ns: Duration, can_sleep := true)
{
profile(#procedure)
config := app_config()
frametime := & memory.client_memory.frametime
// context.allocator = frame_slab_allocator()
// context.temp_allocator = transient_allocator()
config := app_config()
if thread.id == .Master_Prepper
{
frametime.target_ms = 1.0 / f64(config.engine_refresh_hz) * S_To_MS
frametime := & memory.client_memory.frametime
frametime.target_ms = 1.0 / f64(config.engine_refresh_hz)
sub_ms_granularity_required := frametime.target_ms <= Frametime_High_Perf_Threshold_MS
frametime.delta_ns = tick_lap_time( client_tick )
@@ -309,9 +380,11 @@ tick_lane_frametime :: proc(client_tick: ^Tick, host_delta_time_ms: f64, host_de
clean_frame :: proc()
{
profile(#procedure)
context.logger = to_odin_logger(& memory.client_memory.logger)
if thread.id == .Master_Prepper
{
// mem_reset( frame_allocator() )
}
return
}

View File

@@ -22,7 +22,7 @@ ProcessMemory :: struct {
host_api: Host_API, // Client -> Host Interface
// Textual Logging
logger: Logger,
host_logger: Logger,
path_logger_finalized: string,
// Profiling
@@ -45,12 +45,10 @@ ProcessMemory :: struct {
// Testing
job_group_reload: JobGroup,
job_info_reload: [64]TestJobInfo,
job_reload: [64]Job,
job_group_exit: JobGroup,
job_info_exit: [64]TestJobInfo,
job_exit: [64]Job,
job_info_reload: [JOB_TEST_NUM]TestJobInfo,
job_reload: [JOB_TEST_NUM]Job,
}
JOB_TEST_NUM :: 64
Host_API :: struct {
request_virtual_memory: #type proc(), // All dynamic allocations will utilize vmem interfaces
@@ -61,7 +59,7 @@ ThreadMemory :: struct {
using _: ThreadWorkerContext,
// Per-thread profiling
spall_buffer_backing: [SPALL_BUFFER_DEFAULT_SIZE * 4]byte,
spall_buffer_backing: [SPALL_BUFFER_DEFAULT_SIZE]byte,
spall_buffer: Spall_Buffer,
client_memory: ThreadState,

View File

@@ -57,13 +57,15 @@ import "core:time"
tick_now :: time.tick_now
import "codebase:grime"
Logger :: grime.Logger
Logger :: grime.Logger
logger_init :: grime.logger_init
to_odin_logger :: grime.to_odin_logger
set__scheduler_granularity :: grime.set__scheduler_granularity
grime_set_profiler_module_context :: grime.set_profiler_module_context
grime_set_profiler_thread_buffer :: grime.set_profiler_thread_buffer
set__scheduler_granularity :: grime.set__scheduler_granularity
Kilo :: 1024
Mega :: Kilo * 1024
Giga :: Mega * 1024

View File

@@ -67,6 +67,8 @@ State :: struct {
// Overall frametime of the tick frame (currently main thread's)
using frametime : FrameTime,
logger: Logger,
}
ThreadState :: struct {