[Bf-blender-cvs] [0b157f7] master: Sculpt: OMP -> BLI_task, step III.

Bastien Montagne noreply at git.blender.org
Mon Jan 4 19:30:15 CET 2016


Commit: 0b157f7c6cf3b0dfd7b3f2d0317739dc0dc1f236
Author: Bastien Montagne
Date:   Mon Jan 4 17:16:13 2016 +0100
Branches: master
https://developer.blender.org/rB0b157f7c6cf3b0dfd7b3f2d0317739dc0dc1f236

Sculpt: OMP -> BLI_task, step III.

The remaining bits in editors' sculpt.c, which is now free from OMP.

All in all, changes in this file allow from below 2% to above %20 speedup per stroke, depending
on brush complexity, size, and definition of the mesh below it...

Note that this commit also unifies all 'threaded data' into a single struct, a few more
floats/pointers do not hurt here, and it makes code a bit simpler.

===================================================================

M	source/blender/editors/sculpt_paint/sculpt.c

===================================================================

diff --git a/source/blender/editors/sculpt_paint/sculpt.c b/source/blender/editors/sculpt_paint/sculpt.c
index 8781b46..2b708a8 100644
--- a/source/blender/editors/sculpt_paint/sculpt.c
+++ b/source/blender/editors/sculpt_paint/sculpt.c
@@ -414,57 +414,105 @@ static bool sculpt_stroke_is_dynamic_topology(
 
 /*** paint mesh ***/
 
-static void paint_mesh_restore_co(Sculpt *sd, Object *ob)
+/* Single struct used by all BLI_task threaded callbacks, let's avoid adding 10's of those... */
+typedef struct SculptThreadedTaskData {
+	Sculpt *sd;
+	Object *ob;
+	Brush *brush;
+    PBVHNode **nodes;
+	int totnode;
+
+	/* Data specific to some callbacks. */
+	/* Note: even if only one or two of those are used at a time, keeping them separated, names help figuring out
+	 *       what it is, and memory overhead is ridiculous anyway... */
+	float flippedbstrength;
+	float angle;
+	float strength;
+	bool smooth_mask;
+	bool has_bm_orco;
+
+	SculptProjectVector *spvc;
+	float *offset;
+	float *grab_delta;
+	float *cono;
+	float *area_no;
+	float *area_no_sp;
+	float *area_co;
+	float (*mat)[4];
+	float (*vertCos)[3];
+
+	/* 0=towards view, 1=flipped */
+	float (*area_cos)[3];
+	float (*area_nos)[3];
+	int *count;
+
+	ThreadMutex mutex;
+} SculptThreadedTaskData;
+
+static void paint_mesh_restore_co_task_cb(void *userdata, void *UNUSED(userdata_chunk), int n)
 {
-	SculptSession *ss = ob->sculpt;
-	const Brush *brush = BKE_paint_brush(&sd->paint);
+	SculptThreadedTaskData *data = userdata;
+	SculptSession *ss = data->ob->sculpt;
 
-	PBVHNode **nodes;
-	int n, totnode;
+	SculptUndoNode *unode;
+	SculptUndoType type = (data->brush->sculpt_tool == SCULPT_TOOL_MASK ? SCULPT_UNDO_MASK : SCULPT_UNDO_COORDS);
 
-	BKE_pbvh_search_gather(ss->pbvh, NULL, NULL, &nodes, &totnode);
+	if (ss->bm) {
+		unode = sculpt_undo_push_node(data->ob, data->nodes[n], type);
+	}
+	else {
+		unode = sculpt_undo_get_node(data->nodes[n]);
+	}
 
-	/* Disable OpenMP when dynamic-topology is enabled. Otherwise, new
-	 * entries might be inserted by sculpt_undo_push_node() into the
-	 * GHash used internally by BM_log_original_vert_co() by a
-	 * different thread. [#33787] */
-#pragma omp parallel for schedule(guided) if ((sd->flags & SCULPT_USE_OPENMP) && !ss->bm && totnode > SCULPT_OMP_LIMIT)
-	for (n = 0; n < totnode; n++) {
-		SculptUndoNode *unode;
-		SculptUndoType type = (brush->sculpt_tool == SCULPT_TOOL_MASK ?
-		                       SCULPT_UNDO_MASK : SCULPT_UNDO_COORDS);
+	if (unode) {
+		PBVHVertexIter vd;
+		SculptOrigVertData orig_data;
 
-		if (ss->bm) {
-			unode = sculpt_undo_push_node(ob, nodes[n], type);
-		}
-		else {
-			unode = sculpt_undo_get_node(nodes[n]);
-		}
-		if (unode) {
-			PBVHVertexIter vd;
-			SculptOrigVertData orig_data;
+		sculpt_orig_vert_data_unode_init(&orig_data, data->ob, unode);
 
-			sculpt_orig_vert_data_unode_init(&orig_data, ob, unode);
-		
-			BKE_pbvh_vertex_iter_begin(ss->pbvh, nodes[n], vd, PBVH_ITER_UNIQUE)
-			{
-				sculpt_orig_vert_data_update(&orig_data, &vd);
+		BKE_pbvh_vertex_iter_begin(ss->pbvh, data->nodes[n], vd, PBVH_ITER_UNIQUE)
+		{
+			sculpt_orig_vert_data_update(&orig_data, &vd);
 
-				if (orig_data.unode->type == SCULPT_UNDO_COORDS) {
-					copy_v3_v3(vd.co, orig_data.co);
-					if (vd.no) copy_v3_v3_short(vd.no, orig_data.no);
-					else normal_short_to_float_v3(vd.fno, orig_data.no);
-				}
-				else if (orig_data.unode->type == SCULPT_UNDO_MASK) {
-					*vd.mask = orig_data.mask;
-				}
-				if (vd.mvert) vd.mvert->flag |= ME_VERT_PBVH_UPDATE;
+			if (orig_data.unode->type == SCULPT_UNDO_COORDS) {
+				copy_v3_v3(vd.co, orig_data.co);
+				if (vd.no)
+					copy_v3_v3_short(vd.no, orig_data.no);
+				else
+					normal_short_to_float_v3(vd.fno, orig_data.no);
+			}
+			else if (orig_data.unode->type == SCULPT_UNDO_MASK) {
+				*vd.mask = orig_data.mask;
 			}
-			BKE_pbvh_vertex_iter_end;
 
-			BKE_pbvh_node_mark_update(nodes[n]);
+			if (vd.mvert)
+				vd.mvert->flag |= ME_VERT_PBVH_UPDATE;
 		}
+		BKE_pbvh_vertex_iter_end;
+
+		BKE_pbvh_node_mark_update(data->nodes[n]);
 	}
+}
+
+static void paint_mesh_restore_co(Sculpt *sd, Object *ob)
+{
+	SculptSession *ss = ob->sculpt;
+	Brush *brush = BKE_paint_brush(&sd->paint);
+
+	PBVHNode **nodes;
+	int totnode;
+
+	BKE_pbvh_search_gather(ss->pbvh, NULL, NULL, &nodes, &totnode);
+
+	/* Disable OpenMP when dynamic-topology is enabled. Otherwise, new entries might be inserted by
+	 * sculpt_undo_push_node() into the GHash used internally by BM_log_original_vert_co() by a different thread.
+	 * See T33787. */
+	SculptThreadedTaskData data = {
+		.sd = sd, .ob = ob, .brush = brush, .nodes = nodes,
+	};
+
+	BLI_task_parallel_range_ex(0, totnode, &data, NULL, 0, paint_mesh_restore_co_task_cb,
+	                           ((sd->flags & SCULPT_USE_OPENMP) && !ss->bm && totnode > SCULPT_OMP_LIMIT), false);
 
 	if (nodes)
 		MEM_freeN(nodes);
@@ -771,25 +819,13 @@ static float calc_symmetry_feather(Sculpt *sd, StrokeCache *cache)
  * \note These are all _very_ similar, when changing one, check others.
  * \{ */
 
-typedef struct SculptCalcAreaData {
-	Sculpt *sd;
-	Object *ob;
-    PBVHNode **nodes;
-	int totnode;
-	bool has_bm_orco;
-
-	/* 0=towards view, 1=flipped */
-	float (*area_co)[3];
-	float (*area_no)[3];
-	int *count;
-
-	ThreadMutex mutex;
-} SculptCalcAreaData;
-
-static void calc_area_normal_and_centr_task_cb(void *userdata, void *UNUSED(userdata_chunk), int n)
+static void calc_area_normal_and_center_task_cb(void *userdata, void *UNUSED(userdata_chunk), int n)
 {
-	SculptCalcAreaData *data = userdata;
+	SculptThreadedTaskData *data = userdata;
 	SculptSession *ss = data->ob->sculpt;
+	float (*area_nos)[3] = data->area_nos;
+	float (*area_cos)[3] = data->area_cos;
+
 	PBVHVertexIter vd;
 	SculptBrushTest test;
 	SculptUndoNode *unode;
@@ -812,9 +848,7 @@ static void calc_area_normal_and_centr_task_cb(void *userdata, void *UNUSED(user
 		int     orco_tris_num;
 		int i;
 
-		BKE_pbvh_node_get_bm_orco_data(
-		        data->nodes[n],
-		        &orco_tris, &orco_tris_num, &orco_coords);
+		BKE_pbvh_node_get_bm_orco_data(data->nodes[n], &orco_tris, &orco_tris_num, &orco_coords);
 
 		for (i = 0; i < orco_tris_num; i++) {
 			const float *co_tri[3] = {
@@ -833,9 +867,9 @@ static void calc_area_normal_and_centr_task_cb(void *userdata, void *UNUSED(user
 				normal_tri_v3(no, UNPACK3(co_tri));
 
 				flip_index = (dot_v3v3(ss->cache->view_normal, no) <= 0.0f);
-				if (data->area_co)
+				if (area_cos)
 					add_v3_v3(private_co[flip_index], co);
-				if (data->area_no)
+				if (area_nos)
 					add_v3_v3(private_no[flip_index], no);
 				private_count[flip_index] += 1;
 			}
@@ -880,9 +914,9 @@ static void calc_area_normal_and_centr_task_cb(void *userdata, void *UNUSED(user
 				}
 
 				flip_index = (dot_v3v3(ss->cache->view_normal, no) <= 0.0f);
-				if (data->area_co)
+				if (area_cos)
 					add_v3_v3(private_co[flip_index], co);
-				if (data->area_no)
+				if (area_nos)
 					add_v3_v3(private_no[flip_index], no);
 				private_count[flip_index] += 1;
 			}
@@ -893,15 +927,15 @@ static void calc_area_normal_and_centr_task_cb(void *userdata, void *UNUSED(user
 	BLI_mutex_lock(&data->mutex);
 
 	/* for flatten center */
-	if (data->area_co) {
-		add_v3_v3(data->area_co[0], private_co[0]);
-		add_v3_v3(data->area_co[1], private_co[1]);
+	if (area_cos) {
+		add_v3_v3(area_cos[0], private_co[0]);
+		add_v3_v3(area_cos[1], private_co[1]);
 	}
 
 	/* for area normal */
-	if (data->area_no) {
-		add_v3_v3(data->area_no[0], private_no[0]);
-		add_v3_v3(data->area_no[1], private_no[1]);
+	if (area_nos) {
+		add_v3_v3(area_nos[0], private_no[0]);
+		add_v3_v3(area_nos[1], private_no[1]);
 	}
 
 	/* weights */
@@ -922,25 +956,25 @@ static void calc_area_center(
 	int n;
 
 	/* 0=towards view, 1=flipped */
-	float area_co[2][3] = {{0.0f}};
+	float area_cos[2][3] = {{0.0f}};
 
 	int count[2] = {0};
 
-	SculptCalcAreaData data = {
-		.sd = sd, .ob = ob, .nodes = nodes, .totnode = totnode, .has_bm_orco = has_bm_orco,
-		.area_co = area_co, .area_no = NULL, .count = count,
+	SculptThreadedTaskData data = {
+		.sd = sd, .ob = ob, .nodes = nodes, .totnode = totnode,
+	    .has_bm_orco = has_bm_orco, .area_cos = area_cos, .area_nos = NULL, .count = count,
 	};
 	BLI_mutex_init(&data.mutex);
 
-	BLI_task_parallel_range_ex(0, totnode, &data, NULL, 0, calc_area_normal_and_centr_task_cb,
+	BLI_task_parallel_range_ex(0, totnode, &data, NULL, 0, calc_area_normal_and_center_task_cb,
 	                           ((sd->flags & SCULPT_USE_OPENMP) && totnode > SCULPT_OMP_LIMIT), false);
 
 	BLI_mutex_end(&data.mutex);
 
 	/* for flatten center */
-	for (n = 0; n < ARRAY_SIZE(area_co); n++) {
+	for (n = 0; n < ARRAY_SIZE(area_cos); n++) {
 		if (count[n] != 0) {
-			mul_v3_v3fl(r_area_co, area_co[n], 1.0f / count[n]);
+			mul_v3_v3fl(r_area_co, area_cos[n], 1.0f / count[n]);
 			break;
 		}
 	}
@@ -961,24 +995,24 @@ static void calc_area_normal(
 	int n;
 
 	/* 0=towards view, 1=flipped */
-	float area_no[2][3] = {{0.0f}};
+	float area_nos[2][3] = {{0.0f}};
 
 	int count[2] = {0};
 
-	SculptCalcAreaData data = {
-		.sd = sd, .ob = ob, .nodes = nodes, .totnode = totnode, .has_bm_orco = has_bm_orco,
-		.area_co = NULL, .area_no = area_no, .count = count,
+	SculptThreadedTaskData data = {
+		.sd = sd, .ob = ob, .nodes = nodes, .totnode = totnode,
+	    .has_bm_orco = has_bm_orco, .area_cos = NULL, .area_nos = area_nos, .count = count,
 	};
 	BLI_mutex_init(&data.mutex);
 
-	BLI_task_parallel_range_ex(0, totnode, &data, NULL, 0, calc_area_normal_and_centr_task_cb,
+	BLI_task_parallel_range_ex(0, totnode, &data, NULL, 0, calc_area_normal_and_center_task_cb,
 	                           ((sd->flags & SCULPT_USE_OPENMP) && totnode > SCULPT_OMP_LIMIT), false);
 
 	BLI_mutex_end(&data.mutex);
 
 	/* for area normal */
-	for (n = 0; n < ARRAY_SIZE(area_no); n++) {
-		if (normalize_v3_v3(r_area_no, area_no[n]) != 0.0f) {
+	for (n = 0; n < ARRAY_SIZE(area_nos); n++) {
+		if (normalize_v3_v3(r_area_no, area_nos[n]) != 0.0f) {
 			break;
 		}
 	}
@@ -997,26 +1031,26 @@ static void calc_area_normal_and_center(
 	int n;
 
 	/* 0=towards view, 1=flipped */
-	float area_co[2][3] = {{0.0f}};
-	float area_no[2][3] = {{0.0f}};

@@ Diff output truncated at 10240 characters. @@




More information about the Bf-blender-cvs mailing list