Welcome to mirror list, hosted at ThFree Co, Russian Federation.

git.blender.org/blender.git - Unnamed repository; edit this file 'description' to name the repository.
summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorBrecht Van Lommel <brechtvanlommel@pandora.be>2013-07-08 21:56:51 +0400
committerBrecht Van Lommel <brechtvanlommel@pandora.be>2013-07-08 21:56:51 +0400
commit3ce280e825cdd7bcbdd58cb10537a2aa11119ee3 (patch)
tree54328823ffa9f00932a98d21e2aa89c37584e530 /source/blender/windowmanager/intern/wm_jobs.c
parent861f9e10f7ea9d1d38130ecbc983fe8f8dac93d0 (diff)
Fix #35960, #36044: blender internal viewport rendering crash while editing data.
Now the viewport rendering thread will lock the main thread while it is exporting objects to render data. This is not ideal if you have big scenes that might block the UI, but Cycles does the same, and it's fairly quick because the same evaluated mesh can be used as for viewport drawing. It's the only way to get things stable until the thread safe dependency graph is here. This adds a mechanism to the job system for jobs to lock the main thread, using a new 'ticket mutex lock' which is a mutex lock that gives priority to the first thread that tries to lock the mutex. Still to solve: undo/redo crashes.
Diffstat (limited to 'source/blender/windowmanager/intern/wm_jobs.c')
-rw-r--r--source/blender/windowmanager/intern/wm_jobs.c62
1 files changed, 55 insertions, 7 deletions
diff --git a/source/blender/windowmanager/intern/wm_jobs.c b/source/blender/windowmanager/intern/wm_jobs.c
index 03af5e9e8a6..c6e067dc2f9 100644
--- a/source/blender/windowmanager/intern/wm_jobs.c
+++ b/source/blender/windowmanager/intern/wm_jobs.c
@@ -128,8 +128,43 @@ struct wmJob {
ListBase threads;
double start_time;
+
+ /* ticket mutex for main thread locking while some job accesses
+ * data that the main thread might modify at the same time */
+ TicketMutex *main_thread_mutex;
+ bool main_thread_mutex_ending;
};
+/* Main thread locking */
+
+void WM_job_main_thread_lock_acquire(wmJob *wm_job)
+{
+ BLI_ticket_mutex_lock(wm_job->main_thread_mutex);
+
+ /* if BLI_end_threads is being called to stop the job before it's finished,
+ * we no longer need to lock to get access to the main thread as it's
+ * waiting and can't respond */
+ if (wm_job->main_thread_mutex_ending)
+ BLI_ticket_mutex_unlock(wm_job->main_thread_mutex);
+}
+
+void WM_job_main_thread_lock_release(wmJob *wm_job)
+{
+ if (!wm_job->main_thread_mutex_ending)
+ BLI_ticket_mutex_unlock(wm_job->main_thread_mutex);
+}
+
+static void wm_job_main_thread_yield(wmJob *wm_job, bool ending)
+{
+ if (ending)
+ wm_job->main_thread_mutex_ending = true;
+
+ /* unlock and lock the ticket mutex. because it's a fair mutex any job that
+ * is waiting to acquire the lock will get it first, before we can lock */
+ BLI_ticket_mutex_unlock(wm_job->main_thread_mutex);
+ BLI_ticket_mutex_lock(wm_job->main_thread_mutex);
+}
+
/* finds:
* if type, compare for it, otherwise any matching job
*/
@@ -162,13 +197,16 @@ wmJob *WM_jobs_get(wmWindowManager *wm, wmWindow *win, void *owner, const char *
if (wm_job == NULL) {
wm_job = MEM_callocN(sizeof(wmJob), "new job");
-
+
BLI_addtail(&wm->jobs, wm_job);
wm_job->win = win;
wm_job->owner = owner;
wm_job->flag = flag;
wm_job->job_type = job_type;
BLI_strncpy(wm_job->name, name, sizeof(wm_job->name));
+
+ wm_job->main_thread_mutex = BLI_ticket_mutex_alloc();
+ BLI_ticket_mutex_lock(wm_job->main_thread_mutex);
}
/* else: a running job, be careful */
@@ -369,12 +407,21 @@ void WM_jobs_start(wmWindowManager *wm, wmJob *wm_job)
}
}
+static void wm_job_free(wmWindowManager *wm, wmJob *wm_job)
+{
+ BLI_remlink(&wm->jobs, wm_job);
+ BLI_ticket_mutex_unlock(wm_job->main_thread_mutex);
+ BLI_ticket_mutex_free(wm_job->main_thread_mutex);
+ MEM_freeN(wm_job);
+}
+
/* stop job, end thread, free data completely */
static void wm_jobs_kill_job(wmWindowManager *wm, wmJob *wm_job)
{
if (wm_job->running) {
/* signal job to end */
wm_job->stop = TRUE;
+ wm_job_main_thread_yield(wm_job, true);
BLI_end_threads(&wm_job->threads);
if (wm_job->endjob)
@@ -389,9 +436,7 @@ static void wm_jobs_kill_job(wmWindowManager *wm, wmJob *wm_job)
wm_job->run_free(wm_job->run_customdata);
/* remove wm_job */
- BLI_remlink(&wm->jobs, wm_job);
- MEM_freeN(wm_job);
-
+ wm_job_free(wm, wm_job);
}
/* wait until every job ended */
@@ -483,7 +528,6 @@ void wm_jobs_timer(const bContext *C, wmWindowManager *wm, wmTimer *wt)
float total_progress = 0.f;
float jobs_progress = 0;
-
for (wm_job = wm->jobs.first; wm_job; wm_job = wm_jobnext) {
wm_jobnext = wm_job->next;
@@ -491,6 +535,9 @@ void wm_jobs_timer(const bContext *C, wmWindowManager *wm, wmTimer *wt)
/* running threads */
if (wm_job->threads.first) {
+
+ /* let threads get temporary lock over main thread if needed */
+ wm_job_main_thread_yield(wm_job, false);
/* always call note and update when ready */
if (wm_job->do_update || wm_job->ready) {
@@ -522,7 +569,9 @@ void wm_jobs_timer(const bContext *C, wmWindowManager *wm, wmTimer *wt)
}
wm_job->running = FALSE;
+ wm_job_main_thread_yield(wm_job, true);
BLI_end_threads(&wm_job->threads);
+ wm_job->main_thread_mutex_ending = false;
if (wm_job->endnote)
WM_event_add_notifier(C, wm_job->endnote, NULL);
@@ -539,8 +588,7 @@ void wm_jobs_timer(const bContext *C, wmWindowManager *wm, wmTimer *wt)
wm_job->wt = NULL;
/* remove wm_job */
- BLI_remlink(&wm->jobs, wm_job);
- MEM_freeN(wm_job);
+ wm_job_free(wm, wm_job);
}
}
else if (wm_job->flag & WM_JOB_PROGRESS) {