bartvdbraak/blender
1
0
Fork 4
Code Issues 4 Pull Requests 1 Actions 1 Packages Projects Releases Wiki Activity

Sculpt: Avoid threading overhead for area/normal sampling function

Browse Source 
Remove the `any_vertex_sampled` variable, which was just used when only
sampling the area normal, and was unnecessary since we also keep track
of how many vertices passed the sample tests. Though it's a simple
change, this more than doubles the performance of the sculpt brush
benchmark file from #118145 on my hardware, from 1.76s to 0.74s for
the brush evaluation. This is likely because writing to the same
memory from many separate threads interacts very poorly with CPU
caches. If this variable weren't unnecessary, another option would
be including it in the parallel reduction.
This commit is contained in:
Hans Goudey 2024-06-18 23:19:21 -04:00
parent 903e926809
commit b9f86b831c
1 changed files with 17 additions and 45 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

62
source/blender/editors/sculpt_paint/sculpt.cc
View File

@ -1953,8 +1953,7 @@ static void calc_area_normal_and_center_node_mesh(const SculptSession &ss,
const bool use_area_nos,
const bool use_area_cos,
const PBVHNode &node,
AreaNormalCenterData &anctd,
bool &r_any_vertex_sampled)
AreaNormalCenterData &anctd)
{
const float3 &view_normal = ss.cache ? ss.cache->view_normal : ss.cursor_view_normal;
SculptBrushTest normal_test;
@ -1997,8 +1996,6 @@ static void calc_area_normal_and_center_node_mesh(const SculptSession &ss,
continue;
}
r_any_vertex_sampled = true;
const int flip_index = math::dot(view_normal, no) <= 0.0f;
if (use_area_cos && area_test_r) {
anctd.area_cos[flip_index] += area_center_calc_weighted(
@ -2018,8 +2015,7 @@ static void calc_area_normal_and_center_node_grids(const SculptSession &ss,
const bool use_area_nos,
const bool use_area_cos,
const PBVHNode &node,
AreaNormalCenterData &anctd,
bool &r_any_vertex_sampled)
AreaNormalCenterData &anctd)
{
const float3 &view_normal = ss.cache ? ss.cache->view_normal : ss.cursor_view_normal;
const CCGKey key = *BKE_pbvh_get_grid_key(*ss.pbvh);
@ -2070,8 +2066,6 @@ static void calc_area_normal_and_center_node_grids(const SculptSession &ss,
continue;
}
r_any_vertex_sampled = true;
const int flip_index = math::dot(view_normal, no) <= 0.0f;
if (use_area_cos && area_test_r) {
anctd.area_cos[flip_index] += area_center_calc_weighted(
@ -2095,8 +2089,7 @@ static void calc_area_normal_and_center_node_bmesh(const SculptSession &ss,
const bool use_area_cos,
const bool has_bm_orco,
const PBVHNode &node,
AreaNormalCenterData &anctd,
bool &r_any_vertex_sampled)
AreaNormalCenterData &anctd)
{
const float3 &view_normal = ss.cache ? ss.cache->view_normal : ss.cursor_view_normal;
SculptBrushTest normal_test;
@ -2140,8 +2133,6 @@ static void calc_area_normal_and_center_node_bmesh(const SculptSession &ss,
continue;
}
r_any_vertex_sampled = true;
float3 no;
normal_tri_v3(no, UNPACK3(co_tri));
@ -2149,7 +2140,6 @@ static void calc_area_normal_and_center_node_bmesh(const SculptSession &ss,
if (use_area_cos && area_test_r) {
anctd.area_cos[flip_index] += area_center_calc_weighted(
area_test.location, area_test.dist, area_test.radius, co);
anctd.count_co[flip_index] += 1;
}
if (use_area_nos && normal_test_r) {
@ -2184,8 +2174,6 @@ static void calc_area_normal_and_center_node_bmesh(const SculptSession &ss,
continue;
}
r_any_vertex_sampled = true;
const int flip_index = math::dot(view_normal, no) <= 0.0f;
if (use_area_cos && area_test_r) {
anctd.area_cos[flip_index] += area_center_calc_weighted(
@ -2228,8 +2216,6 @@ void calc_area_center(const Brush &brush,
const bool has_bm_orco = ss.bm && dyntopo::stroke_is_dyntopo(ss, brush);
int n;
bool any_vertex_sampled = false;
const PBVHType type = BKE_pbvh_type(*ss.pbvh);
const AreaNormalCenterData anctd = threading::parallel_reduce(
nodes.index_range(),
@ -2253,17 +2239,15 @@ void calc_area_center(const Brush &brush,
false,
true,
*nodes[i],
anctd,
any_vertex_sampled);
anctd);
break;
}
case PBVH_GRIDS:
calc_area_normal_and_center_node_grids(
ss, brush, false, true, *nodes[i], anctd, any_vertex_sampled);
calc_area_normal_and_center_node_grids(ss, brush, false, true, *nodes[i], anctd);
break;
case PBVH_BMESH:
calc_area_normal_and_center_node_bmesh(
ss, brush, false, true, has_bm_orco, *nodes[i], anctd, any_vertex_sampled);
ss, brush, false, true, has_bm_orco, *nodes[i], anctd);
break;
}
}
@ -2297,8 +2281,6 @@ std::optional<float3> calc_area_normal(const Brush &brush, Object &ob, Span<PBVH
SculptSession &ss = *ob.sculpt;
const bool has_bm_orco = ss.bm && dyntopo::stroke_is_dyntopo(ss, brush);
bool any_vertex_sampled = false;
const PBVHType type = BKE_pbvh_type(*ss.pbvh);
const AreaNormalCenterData anctd = threading::parallel_reduce(
nodes.index_range(),
@ -2322,17 +2304,15 @@ std::optional<float3> calc_area_normal(const Brush &brush, Object &ob, Span<PBVH
true,
false,
*nodes[i],
anctd,
any_vertex_sampled);
anctd);
break;
}
case PBVH_GRIDS:
calc_area_normal_and_center_node_grids(
ss, brush, true, false, *nodes[i], anctd, any_vertex_sampled);
calc_area_normal_and_center_node_grids(ss, brush, true, false, *nodes[i], anctd);
break;
case PBVH_BMESH:
calc_area_normal_and_center_node_bmesh(
ss, brush, true, false, has_bm_orco, *nodes[i], anctd, any_vertex_sampled);
ss, brush, true, false, has_bm_orco, *nodes[i], anctd);
break;
}
}
@ -2340,15 +2320,11 @@ std::optional<float3> calc_area_normal(const Brush &brush, Object &ob, Span<PBVH
},
calc_area_normal_and_center_reduce);
if (!any_vertex_sampled) {
return std::nullopt;
}
/* For area normal. */
float3 result;
for (int i = 0; i < anctd.area_nos.size(); i++) {
if (normalize_v3_v3(result, anctd.area_nos[i]) != 0.0f) {
return result;
for (const int i : {0, 1}) {
if (anctd.count_no[i] != 0) {
if (!math::is_zero(anctd.area_nos[i])) {
return math::normalize(anctd.area_nos[i]);
}
}
}
return std::nullopt;
@ -2364,8 +2340,6 @@ void calc_area_normal_and_center(const Brush &brush,
const bool has_bm_orco = ss.bm && dyntopo::stroke_is_dyntopo(ss, brush);
int n;
bool any_vertex_sampled = false;
const PBVHType type = BKE_pbvh_type(*ss.pbvh);
const AreaNormalCenterData anctd = threading::parallel_reduce(
nodes.index_range(),
@ -2389,17 +2363,15 @@ void calc_area_normal_and_center(const Brush &brush,
true,
true,
*nodes[i],
anctd,
any_vertex_sampled);
anctd);
break;
}
case PBVH_GRIDS:
calc_area_normal_and_center_node_grids(
ss, brush, true, true, *nodes[i], anctd, any_vertex_sampled);
calc_area_normal_and_center_node_grids(ss, brush, true, true, *nodes[i], anctd);
break;
case PBVH_BMESH:
calc_area_normal_and_center_node_bmesh(
ss, brush, true, true, has_bm_orco, *nodes[i], anctd, any_vertex_sampled);
ss, brush, true, true, has_bm_orco, *nodes[i], anctd);
break;
}
}