Smoke: Patch #20955 / fix by Damien Plisson (damien78)

* Smoke got it's own thread loop now and can be esc'ed now
* Apple OpenMP bug workaround until they fix the GCC (couldn't test this one)
* removing some "static" keyword to prevent crashing on frame display
* should also fix problems with wavelet.noise crashing (Bug #20710)

Fluid:
* sleep delay has been reduced to be more responsive
* Fluid progress is displayed in percent using the mouse cursor (like smoke)

6 files changed, 153 insertions, 31 deletions

diff --git a/intern/smoke/intern/WAVELET_NOISE.h b/intern/smoke/intern/WAVELET_NOISE.h
index 4c2e7514af7..7056b1db5be 100644
--- a/intern/smoke/intern/WAVELET_NOISE.h
+++ b/intern/smoke/intern/WAVELET_NOISE.h
@@ -42,6 +42,10 @@
 
 #include <MERSENNETWISTER.h>
 
+// Tile file header, update revision upon any change done to the noise generator
+static const char tilefile_headerstring[] = "Noise Tile File rev. ";
+static const char tilefile_revision[] =  "001";
+
 #define NOISE_TILE_SIZE 128
 static const int noiseTileSize = NOISE_TILE_SIZE;
 
@@ -94,7 +98,7 @@ static void downsampleNeumann(const float *from, float *to, int n, int stride)
 {
   // if these values are not local incorrect results are generated
   float downCoeffs[32] = { DOWNCOEFFS };
-  static const float *const aCoCenter= &downCoeffs[16];
+  const float *const aCoCenter= &downCoeffs[16];
 	for (int i = 0; i < n / 2; i++) {
 		to[i * stride] = 0;
 		for (int k = 2 * i - 16; k < 2 * i + 16; k++) { 
@@ -212,6 +216,12 @@ static void upsampleZNeumann(float* to, const float* from, int sx, int sy, int s
 static bool loadTile(float* const noiseTileData, std::string filename)
 {
 	FILE* file;
+	char headerbuffer[64];
+	size_t headerlen;
+	size_t bread;
+	int endiantest = 1;
+	char endianness;
+	
 	file = fopen(filename.c_str(), "rb");
 
 	if (file == NULL) {
@@ -219,11 +229,28 @@ static bool loadTile(float* const noiseTileData, std::string filename)
 		return false;
 	}
 
+	//Check header
+	headerlen = strlen(tilefile_headerstring) + strlen(tilefile_revision) + 2;
+	bread = fread((void*)headerbuffer, 1, headerlen, file);
+	if (*((unsigned char*)&endiantest) == 1)
+		endianness = 'L';
+	else
+		endianness = 'B';
+	if ((bread != headerlen)
+		|| (strncmp(headerbuffer, tilefile_headerstring, strlen(tilefile_headerstring)))
+		|| (strncmp(headerbuffer+ strlen(tilefile_headerstring), tilefile_revision, strlen(tilefile_revision)))
+		|| (headerbuffer[headerlen-2] != endianness)
+		|| (headerbuffer[headerlen-1] != (char)((char)sizeof(long)+'0')))
+	{
+		printf("loadTile : Noise tile '%s' was generated on an incompatible platform.\n",filename.c_str());
+		return false;
+	}
+	
 	// dimensions
 	size_t gridSize = noiseTileSize * noiseTileSize * noiseTileSize;
 
 	// noiseTileData memory is managed by caller
-	size_t bread = fread((void*)noiseTileData, sizeof(float), gridSize, file);
+	bread = fread((void*)noiseTileData, sizeof(float), gridSize, file);
 	fclose(file);
 	printf("Noise tile file '%s' loaded.\n", filename.c_str());
 
@@ -241,12 +268,27 @@ static void saveTile(float* const noiseTileData, std::string filename)
 {
 	FILE* file;
 	file = fopen(filename.c_str(), "wb");
+	int endiantest=1;
+	char longsize;
 
 	if (file == NULL) {
 		printf("saveTile: Noise tile '%s' could not be saved.\n", filename.c_str());
 		return;
 	} 
 
+	//Write file header
+	fwrite(tilefile_headerstring, strlen(tilefile_headerstring), 1, file);
+	fwrite(tilefile_revision, strlen(tilefile_revision), 1, file);
+	//Endianness
+	if (*((unsigned char*)&endiantest) == 1)
+		fwrite("L", 1, 1, file); //Little endian
+	else
+		fwrite("B",1,1,file); //Big endian
+	//32/64bit
+	longsize = (char)sizeof(long)+'0';
+	fwrite(&longsize, 1, 1, file);
+	
+	
 	fwrite((void*)noiseTileData, sizeof(float), noiseTileSize * noiseTileSize * noiseTileSize, file);
 	fclose(file);
 
diff --git a/source/blender/blenkernel/CMakeLists.txt b/source/blender/blenkernel/CMakeLists.txt
index b1cbd16512e..68bc7c5bf15 100644
--- a/source/blender/blenkernel/CMakeLists.txt
+++ b/source/blender/blenkernel/CMakeLists.txt
@@ -73,6 +73,10 @@ ELSE(WITH_PYTHON)
 	ADD_DEFINITIONS(-DDISABLE_PYTHON)
 ENDIF(WITH_PYTHON)
 
+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/blenkernel/SConscript b/source/blender/blenkernel/SConscript
index 508ed58165f..37a63be6389 100644
--- a/source/blender/blenkernel/SConscript
+++ b/source/blender/blenkernel/SConscript
@@ -55,6 +55,10 @@ if env['WITH_BF_QUICKTIME']:
 if env['WITH_BF_BULLET']:
 	defs.append('USE_BULLET')
 
+if env['OURPLATFORM'] == 'darwin':
+    if env['WITH_BF_OPENMP']:
+       defs.append('PARALLEL=1')
+
 if env['BF_NO_ELBEEM']:
 	defs.append('DISABLE_ELBEEM')
 
diff --git a/source/blender/blenkernel/intern/Makefile b/source/blender/blenkernel/intern/Makefile
index 95c82d31890..70e1a785787 100644
--- a/source/blender/blenkernel/intern/Makefile
+++ b/source/blender/blenkernel/intern/Makefile
@@ -129,3 +129,8 @@ ifeq ($(WITH_QUICKTIME), true)
 	CPPFLAGS += -DWITH_QUICKTIME
 endif
 
+ifeq ($(OS), darwin)
+    ifeq ($(WITH_BF_OPENMP), true)
+      CPPFLAGS += -DPARALLEL=1
+    endif
+endif
diff --git a/source/blender/blenkernel/intern/pointcache.c b/source/blender/blenkernel/intern/pointcache.c
index 9bdc700c313..71cea63b0bc 100644
--- a/source/blender/blenkernel/intern/pointcache.c
+++ b/source/blender/blenkernel/intern/pointcache.c
@@ -42,7 +42,13 @@
 #include "DNA_smoke_types.h"
 
 #include "BLI_blenlib.h"
+#include "BLI_threads.h"
 
+#include "PIL_time.h"
+
+#include "WM_api.h"
+
+#include "BKE_blender.h"
 #include "BKE_cloth.h"
 #include "BKE_depsgraph.h"
 #include "BKE_global.h"
@@ -58,8 +64,6 @@
 #include "BKE_utildefines.h"
 #include "BIK_api.h"
 
-#include "BLI_blenlib.h"
-
 /* both in intern */
 #include "smoke_API.h"
 
@@ -84,6 +88,13 @@
   #include "BLI_winstuff.h"
 #endif
 
+#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
+
 #define PTCACHE_DATA_FROM(data, type, from)		if(data[type]) { memcpy(data[type], from, ptcache_data_size[type]); }
 #define PTCACHE_DATA_TO(data, type, index, to)	if(data[type]) { memcpy(to, (char*)data[type] + (index ? index * ptcache_data_size[type] : 0), ptcache_data_size[type]); }
 
@@ -2247,6 +2258,32 @@ void BKE_ptcache_quick_cache_all(Scene *scene)
 		BKE_ptcache_make_cache(&baker);
 }
 
+/* Simulation thread, no need for interlocks as data written in both threads
+ are only unitary integers (I/O assumed to be atomic for them) */
+typedef struct {
+	int break_operation;
+	int thread_ended;
+	int endframe;
+	int step;
+	int *cfra_ptr;
+	Scene *scene;
+} ptcache_make_cache_data;
+
+static void *ptcache_make_cache_thread(void *ptr) {
+	ptcache_make_cache_data *data = (ptcache_make_cache_data*)ptr;
+
+#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
+	
+	for(; (*data->cfra_ptr <= data->endframe) && !data->break_operation; *data->cfra_ptr+=data->step)
+		scene_update_for_newframe(data->scene, data->scene->lay);
+
+	data->thread_ended = TRUE;
+	return NULL;
+}
+
 /* if bake is not given run simulations to current frame */
 void BKE_ptcache_make_cache(PTCacheBaker* baker)
 {
@@ -2258,10 +2295,16 @@ void BKE_ptcache_make_cache(PTCacheBaker* baker)
 	float frameleno = scene->r.framelen;
 	int cfrao = CFRA;
 	int startframe = MAXFRAME;
-	int endframe = baker->anim_init ? scene->r.sfra : CFRA;
 	int bake = baker->bake;
 	int render = baker->render;
-	int step = baker->quick_step;
+	ListBase threads;
+	ptcache_make_cache_data thread_data;
+	int progress, old_progress;
+	
+	thread_data.endframe = baker->anim_init ? scene->r.sfra : CFRA;
+	thread_data.step = baker->quick_step;
+	thread_data.cfra_ptr = &CFRA;
+	thread_data.scene = baker->scene;
 
 	G.afbreek = 0;
 
@@ -2299,11 +2342,11 @@ void BKE_ptcache_make_cache(PTCacheBaker* baker)
 			startframe = MAX2(cache->last_exact, cache->startframe);
 
 			if(bake) {
-				endframe = cache->endframe;
+				thread_data.endframe = cache->endframe;
 				cache->flag |= PTCACHE_BAKING;
 			}
 			else {
-				endframe = MIN2(endframe, cache->endframe);
+				thread_data.endframe = MIN2(thread_data.endframe, cache->endframe);
 			}
 
 			cache->flag &= ~PTCACHE_BAKED;
@@ -2335,7 +2378,7 @@ void BKE_ptcache_make_cache(PTCacheBaker* baker)
 					cache->flag |= PTCACHE_BAKING;
 
 					if(bake)
-						endframe = MAX2(endframe, cache->endframe);
+						thread_data.endframe = MAX2(thread_data.endframe, cache->endframe);
 				}
 
 				cache->flag &= ~PTCACHE_BAKED;
@@ -2345,30 +2388,45 @@ void BKE_ptcache_make_cache(PTCacheBaker* baker)
 		BLI_freelistN(&pidlist);
 	}
 
-	CFRA= startframe;
+	CFRA = startframe;
 	scene->r.framelen = 1.0;
-
-	for(; CFRA <= endframe; CFRA+=step) {
-		int prog;
+	thread_data.break_operation = FALSE;
+	thread_data.thread_ended = FALSE;
+	old_progress = -1;
+	
+#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, ptcache_make_cache_thread, 1);
+	BLI_insert_thread(&threads, (void*)&thread_data);
+	
+	while (thread_data.thread_ended == FALSE) {
 
 		if(bake)
-			prog = (int)(100.0f * (float)(CFRA - startframe)/(float)(endframe-startframe));
+			progress = (int)(100.0f * (float)(CFRA - startframe)/(float)(thread_data.endframe-startframe));
 		else
-			prog = CFRA;
+			progress = CFRA;
 
 		/* NOTE: baking should not redraw whole ui as this slows things down */
-		if(baker->progressbar)
-			baker->progressbar(baker->progresscontext, prog);
+		if ((baker->progressbar) && (progress != old_progress)) {
+			baker->progressbar(baker->progresscontext, progress);
+			old_progress = progress;
+		}
 		
-		scene_update_for_newframe(scene, scene->lay);
+		/* Delay to lessen CPU load from UI thread */
+		PIL_sleep_ms(200);
 
 		/* NOTE: breaking baking should leave calculated frames in cache, not clear it */
-		if(baker->break_test && baker->break_test(baker->break_data))
-			break;
+		if(blender_test_break() && !thread_data.break_operation) {
+			thread_data.break_operation = TRUE;
+			if (baker->progressend)
+				baker->progressend(baker->progresscontext);
+			WM_cursor_wait(1);
+		}
 	}
 
-	if (baker->progressend)
-		baker->progressend(baker->progresscontext);
+	BLI_end_threads(&threads);
 
 	/* clear baking flag */
 	if(pid) {
@@ -2391,7 +2449,7 @@ void BKE_ptcache_make_cache(PTCacheBaker* baker)
 		
 			cache = pid->cache;
 
-			if(step > 1)
+			if(thread_data.step > 1)
 				cache->flag &= ~(PTCACHE_BAKING|PTCACHE_OUTDATED);
 			else
 				cache->flag &= ~(PTCACHE_BAKING|PTCACHE_REDO_NEEDED);
@@ -2413,6 +2471,11 @@ void BKE_ptcache_make_cache(PTCacheBaker* baker)
 	if(bake) /* already on cfra unless baking */
 		scene_update_for_newframe(scene, scene->lay);
 
+	if (thread_data.break_operation)
+		WM_cursor_wait(0);
+	else if (baker->progressend)
+		baker->progressend(baker->progresscontext);
+
 	/* TODO: call redraw all windows somehow */
 }
 /* Helpers */
diff --git a/source/blender/editors/physics/physics_fluid.c b/source/blender/editors/physics/physics_fluid.c
index eec971e777e..b1a3a3be28d 100644
--- a/source/blender/editors/physics/physics_fluid.c
+++ b/source/blender/editors/physics/physics_fluid.c
@@ -109,9 +109,9 @@ static void *thread_tls_data;
 /* XXX */
 /* from header info.c */
 static int start_progress_bar(void) {return 0;};
-static void end_progress_bar(void) {};
+static void end_progress_bar(wmWindow *win) {WM_cursor_restore(win);};
 static void waitcursor(int val) {};
-static int progress_bar(float done, char *busy_info) {return 0;}
+static int progress_bar(wmWindow *win, float done, char *busy_info) { WM_timecursor(win,done*100); return 0;}
 static int pupmenu() {return 0;}
 /* XXX */
 
@@ -1060,7 +1060,7 @@ int fluidsimBake(bContext *C, ReportList *reports, Object *ob)
 		{
 			int done = 0;
 			float noFramesf = (float)noFrames;
-			float percentdone = 0.0;
+			float percentdone = 0.0, oldpercentdone = -1.0;
 			int lastRedraw = -1;
 			
 			g_break= 0;
@@ -1075,11 +1075,15 @@ int fluidsimBake(bContext *C, ReportList *reports, Object *ob)
 				
 				// lukep we add progress bar as an interim mesure
 				percentdone = globalBakeFrame / noFramesf;
-				sprintf(busy_mess, "baking fluids %d / %d       |||", globalBakeFrame, (int) noFramesf);
-				progress_bar(percentdone, busy_mess );
+				if (percentdone != oldpercentdone) {
+					sprintf(busy_mess, "baking fluids %d / %d       |||", globalBakeFrame, (int) noFramesf);
+					percentdone = percentdone < 0.0 ? 0.0:percentdone;
+					progress_bar(CTX_wm_window(C), percentdone, busy_mess );
+					oldpercentdone = percentdone;
+				}
 				
-				// longer delay to prevent frequent redrawing
-				PIL_sleep_ms(2000);
+				//XXX no more need for longer delay to prevent frequent redrawing
+				PIL_sleep_ms(200);
 				
 				BLI_lock_thread(LOCK_CUSTOM1);
 				if(globalBakeState != 0) done = 1; // 1=ok, <0=error/abort
@@ -1121,7 +1125,7 @@ int fluidsimBake(bContext *C, ReportList *reports, Object *ob)
 #endif
 				} // redraw
 			}
-			end_progress_bar();
+			end_progress_bar(CTX_wm_window(C));
 		}
 		BLI_end_threads(&threads);
 	} // El'Beem API init, thread creation