Backport revisions for the 2.70a releasev2.70a

d2660a0, 6e99fb0, 58c22d8, 83f2012 + ff21f6a, a7ed1db. cc6b106
7997e38, 9d4b54b, efb48fc, 3fc293c, 29f359c, 77c1d17, 92a539e,
c626462, f48828b, 6452d9f, 765d077, 74518b2, af16d46, 8da4936,
6babbf5, f0106d2, f88776b, ee72cba, 467596d, e21a7b3, eed3974,
71a2ff1, ccf9afd, 44d6c68, 30fdfc3, b69809c, b0a8e4c, bd57ec6,
3b0832d, 2a25676, 3977b76, fb25a86, 9bbb30b, 51abc2b, 0ebade5,
2c0e32f, 3deaf7d, ea01b24, c61eb64, f3db038, a6fb670, eedde31,
b66a954, 7ff123c, f5b79df, 7148c97, 54a8753, fcaa018, 4c73001,
7a21330, 07578be, e9a64e2, fd3de8b, ae792e9, b7712a7 + 3600622,
d9557d0, 6d973b8, 688257d, 4acb57a, 95ac6bc,

Also backported openmp changes to sculpt making it so number
of real CPU cores is used as a number of threads here.

1 files changed, 37 insertions, 9 deletions

diff --git a/source/blender/editors/sculpt_paint/sculpt.c b/source/blender/editors/sculpt_paint/sculpt.c
index d74e9c9e17b..35de6af2531 100644
--- a/source/blender/editors/sculpt_paint/sculpt.c
+++ b/source/blender/editors/sculpt_paint/sculpt.c
@@ -104,6 +104,19 @@
 #include <omp.h>
 #endif
 
+#if defined(__APPLE__)
+#include <sys/sysctl.h>
+
+/* Query how many cores not counting HT aka physical cores we've got. */
+static int system_physical_thread_count(void)
+{
+	int pcount;
+	size_t pcount_len = sizeof(pcount);
+	sysctlbyname("hw.physicalcpu", &pcount, &pcount_len, NULL, 0);
+	return pcount;
+}
+#endif  /* __APPLE__ */
+
 void ED_sculpt_get_average_stroke(Object *ob, float stroke[3])
 {
 	if (ob->sculpt->last_stroke_valid && ob->sculpt->average_stroke_counter > 0) {
@@ -230,7 +243,7 @@ typedef struct StrokeCache {
 	float initial_mouse[2];
 
 	/* Pre-allocated temporary storage used during smoothing */
-	int num_threads;
+	int num_threads, max_threads;
 	float (**tmpgrid_co)[3], (**tmprow_co)[3];
 	float **tmpgrid_mask, **tmprow_mask;
 
@@ -924,6 +937,7 @@ static float tex_strength(SculptSession *ss, Brush *br,
 	MTex *mtex = &br->mtex;
 	float avg = 1;
 	float rgba[4];
+	int thread_num;
 
 	if (!mtex->tex) {
 		avg = 1;
@@ -966,7 +980,12 @@ static float tex_strength(SculptSession *ss, Brush *br,
 			x += br->mtex.ofs[0];
 			y += br->mtex.ofs[1];
 
-			avg = paint_get_tex_pixel(&br->mtex, x, y, ss->tex_pool);
+#ifdef _OPENMP
+			thread_num = omp_get_thread_num();
+#else
+			thread_num = 0;
+#endif
+			avg = paint_get_tex_pixel(&br->mtex, x, y, ss->tex_pool, thread_num);
 
 			avg += br->texture_sample_bias;
 		}
@@ -3766,16 +3785,21 @@ static void sculpt_omp_start(Sculpt *sd, SculptSession *ss)
 	 * Justification: Empirically I've found that two threads per
 	 * processor gives higher throughput. */
 	if (sd->flags & SCULPT_USE_OPENMP) {
-		cache->num_threads = 2 * omp_get_num_procs();
-		omp_set_num_threads(cache->num_threads);
-	}
-	else
+#if defined(__APPLE__)
+		cache->num_threads = system_physical_thread_count();
+#else
+		cache->num_threads = omp_get_num_procs();
 #endif
-	{
-		(void)sd;
+	}
+	else {
 		cache->num_threads = 1;
 	}
-
+	cache->max_threads = omp_get_max_threads();
+	omp_set_num_threads(cache->num_threads);
+#else
+	(void)sd;
+	cache->num_threads = 1;
+#endif
 	if (ss->multires) {
 		int i, gridsize, array_mem_size;
 		BKE_pbvh_node_get_grids(ss->pbvh, NULL, NULL, NULL, NULL,
@@ -3802,6 +3826,10 @@ static void sculpt_omp_start(Sculpt *sd, SculptSession *ss)
 
 static void sculpt_omp_done(SculptSession *ss)
 {
+#ifdef _OPENMP
+	omp_set_num_threads(ss->cache->max_threads);
+#endif
+
 	if (ss->multires) {
 		int i;