OSX vs OpenMP : implement workaround to fix crashes when using mop from a background thread

Fix# 20043 & 20392

The issue is that OSX lib does not implement TLS (Thread Local Storage), so  libgomp uses pthread functions to read/write thread specific vars.
But this implementation is currently (gcc 4.2) buggy : the write function is called only at lib start (in main thread), and the var is undefined for background thread.

The workaround is to perform this gomp_tls_key var write at beginning of background threads that use openMP. (Currently: render & fluidsim)

5 files changed, 50 insertions, 1 deletions

diff --git a/source/blender/editors/CMakeLists.txt b/source/blender/editors/CMakeLists.txt
index 04c310ba7df..c695ea02508 100644
--- a/source/blender/editors/CMakeLists.txt
+++ b/source/blender/editors/CMakeLists.txt
@@ -71,6 +71,10 @@ IF(WITH_FFMPEG)
 	ADD_DEFINITIONS(-DWITH_FFMPEG)
 ENDIF(WITH_FFMPEG)
 
+IF(WITH_OPENMP)
+		ADD_DEFINITIONS(-DPARALLEL=1)
+ENDIF(WITH_OPENMP)
+
 IF(NOT WITH_ELBEEM)
 	ADD_DEFINITIONS(-DDISABLE_ELBEEM)
 ENDIF(NOT WITH_ELBEEM)
diff --git a/source/blender/editors/physics/SConscript b/source/blender/editors/physics/SConscript
index 85a40c91b3a..fcf94eae994 100644
--- a/source/blender/editors/physics/SConscript
+++ b/source/blender/editors/physics/SConscript
@@ -8,6 +8,8 @@ incs += ' ../../windowmanager #/intern/guardedalloc #/extern/glew/include'
 incs += ' ../../gpu'
 incs += ' ../../makesrna ../../render/extern/include  #/intern/elbeem/extern'
 
+defs = ''
+
 if env['OURPLATFORM'] == 'linux2':
     cflags='-pthread'
     incs += ' ../../../extern/binreloc/include'
@@ -15,4 +17,9 @@ if env['OURPLATFORM'] == 'linux2':
 if env['OURPLATFORM'] in ('win32-vc', 'win32-mingw', 'linuxcross', 'win64-vc'):
     incs += ' ' + env['BF_PTHREADS_INC']
 
-env.BlenderLib ( 'bf_editors_physics', sources, Split(incs), [], libtype=['core'], priority=[45] )
+
+if env['OURPLATFORM'] == 'darwin':
+    if env['WITH_BF_OPENMP']:
+       defs += ' PARALLEL=1'
+
+env.BlenderLib ( 'bf_editors_physics', sources, Split(incs), Split(defs), libtype=['core'], priority=[45] )
diff --git a/source/blender/editors/physics/physics_fluid.c b/source/blender/editors/physics/physics_fluid.c
index 7893b314e3e..eec971e777e 100644
--- a/source/blender/editors/physics/physics_fluid.c
+++ b/source/blender/editors/physics/physics_fluid.c
@@ -98,6 +98,14 @@
 /* enable/disable overall compilation */
 #ifndef DISABLE_ELBEEM
 
+#if defined(__APPLE__) && (PARALLEL == 1) && (__GNUC__ == 4) && (__GNUC_MINOR__ == 2)
+/* ************** libgomp (Apple gcc 4.2.1) TLS bug workaround *************** */
+#include <pthread.h>
+extern pthread_key_t gomp_tls_key;
+static void *thread_tls_data;
+#endif
+
+
 /* XXX */
 /* from header info.c */
 static int start_progress_bar(void) {return 0;};
@@ -320,6 +328,11 @@ static void *fluidsimSimulateThread(void *unused) { // *ptr) {
 	//char* fnameCfgPath = (char*)(ptr);
 	int ret=0;
 	
+#if defined(__APPLE__) && (PARALLEL == 1) && (__GNUC__ == 4) && (__GNUC_MINOR__ == 2)
+	// Workaround for Apple gcc 4.2.1 omp vs background thread bug
+	pthread_setspecific (gomp_tls_key, thread_tls_data);
+#endif
+
 	ret = elbeemSimulate();
 	BLI_lock_thread(LOCK_CUSTOM1);
 	if(globalBakeState==0) {
@@ -1036,6 +1049,11 @@ int fluidsimBake(bContext *C, ReportList *reports, Object *ob)
 		// set to neutral, -1 means user abort, -2 means init error
 		globalBakeState = 0;
 		globalBakeFrame = 0;
+		
+#if defined(__APPLE__) && (PARALLEL == 1) && (__GNUC__ == 4) && (__GNUC_MINOR__ == 2)
+		// Workaround for Apple gcc 4.2.1 omp vs background thread bug
+		thread_tls_data = pthread_getspecific(gomp_tls_key);
+#endif
 		BLI_init_threads(&threads, fluidsimSimulateThread, 1);
 		BLI_insert_thread(&threads, targetFile);
 		
diff --git a/source/blender/editors/screen/SConscript b/source/blender/editors/screen/SConscript
index afd6ce68b68..703a16f753b 100644
--- a/source/blender/editors/screen/SConscript
+++ b/source/blender/editors/screen/SConscript
@@ -22,4 +22,8 @@ if env['OURPLATFORM'] == 'linux2':
 if env['OURPLATFORM'] in ('win32-vc', 'win32-mingw', 'linuxcross', 'win64-vc'):
 	incs += ' ' + env['BF_PTHREADS_INC']
 
+if env['OURPLATFORM'] == 'darwin':
+    if env['WITH_BF_OPENMP']:
+       defs += ' PARALLEL=1'
+
 env.BlenderLib ( 'bf_editors_screen', sources, Split(incs), Split(defs), libtype=['core'], priority=[105] )
diff --git a/source/blender/editors/screen/screen_ops.c b/source/blender/editors/screen/screen_ops.c
index 2a9859330af..977d04a7f45 100644
--- a/source/blender/editors/screen/screen_ops.c
+++ b/source/blender/editors/screen/screen_ops.c
@@ -96,6 +96,12 @@
 #define KM_MODAL_STEP10		3
 #define KM_MODAL_STEP10_OFF	4
 
+#if defined(__APPLE__) && (PARALLEL == 1) && (__GNUC__ == 4) && (__GNUC_MINOR__ == 2)
+/* ************** libgomp (Apple gcc 4.2.1) TLS bug workaround *************** */
+#include <pthread.h>
+extern pthread_key_t gomp_tls_key;
+static void *thread_tls_data;
+#endif
 /* ************** Exported Poll tests ********************** */
 
 int ED_operator_regionactive(bContext *C)
@@ -3015,6 +3021,11 @@ static void render_startjob(void *rjv, short *stop, short *do_update)
 	rj->stop= stop;
 	rj->do_update= do_update;
 	
+#if defined(__APPLE__) && (PARALLEL == 1) && (__GNUC__ == 4) && (__GNUC_MINOR__ == 2)
+	// Workaround for Apple gcc 4.2.1 omp vs background thread bug
+	pthread_setspecific (gomp_tls_key, thread_tls_data);
+#endif
+	
 	if(rj->anim)
 		RE_BlenderAnim(rj->re, rj->scene, rj->scene->r.sfra, rj->scene->r.efra, rj->scene->r.frame_step);
 	else
@@ -3096,6 +3107,11 @@ static int screen_render_invoke(bContext *C, wmOperator *op, wmEvent *event)
 	WM_jobs_timer(steve, 0.2, NC_SCENE|ND_RENDER_RESULT, 0);
 	WM_jobs_callbacks(steve, render_startjob, NULL, NULL);
 	
+#if defined(__APPLE__) && (PARALLEL == 1) && (__GNUC__ == 4) && (__GNUC_MINOR__ == 2)
+	// Workaround for Apple gcc 4.2.1 omp vs background thread bug
+	thread_tls_data = pthread_getspecific(gomp_tls_key);
+#endif
+	
 	/* get a render result image, and make sure it is empty */
 	ima= BKE_image_verify_viewer(IMA_TYPE_R_RESULT, "Render Result");
 	BKE_image_signal(ima, NULL, IMA_SIGNAL_FREE);