griefith/test2
1
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

Cycles: Change OptiX curve vertex data generation to use more compact representation

Browse Source 
OptiX has accepted Catmull-Rom curve data natively since OptiX 7.4, but due to the previous conversion to B-Spline code, the format that data is fed to OptiX wasn't optimal.

Each curve segment was put in the vertex buffer as four independent control points, even though continuous segments actually share control points between each other. This patch compacts that so shared control points only occur once in the vertex buffer.
This compact form uses less memory and also allows OptiX to easily identify segments that belong together into a curve (those where the step between indices is one).

Pull Request: https://projects.blender.org/blender/blender/pulls/125899
This commit is contained in:
Patrick Mours
2024-08-15 15:00:56 +02:00
parent d8913812ef
commit 013a2ce765
1 changed files with 96 additions and 51 deletions
Download Patch File Download Diff File Expand all files Collapse all files

147
intern/cycles/device/optix/device_impl.cpp
View File

@@ -1185,13 +1185,17 @@ void OptiXDevice::build_bvh(BVH *bvh, Progress &progress, bool refit)
device_vector<int> index_data(this, "optix temp index data", MEM_READ_ONLY);
device_vector<float4> vertex_data(this, "optix temp vertex data", MEM_READ_ONLY);
/* Four control points for each curve segment. */
const size_t num_vertices = num_segments * 4;
size_t num_vertices = num_segments * 4;
if (hair->curve_shape == CURVE_THICK) {
# if OPTIX_ABI_VERSION >= 55
num_vertices = hair->num_keys() + 2 * hair->num_curves();
# endif
index_data.alloc(num_segments);
vertex_data.alloc(num_vertices * num_motion_steps);
}
else
else {
aabb_data.alloc(num_segments * num_motion_steps);
}
/* Get AABBs for each motion step. */
for (size_t step = 0; step < num_motion_steps; ++step) {
@@ -1204,59 +1208,100 @@ void OptiXDevice::build_bvh(BVH *bvh, Progress &progress, bool refit)
keys = motion_keys->data_float3() + attr_offset * hair->get_curve_keys().size();
}
for (size_t j = 0, i = 0; j < hair->num_curves(); ++j) {
const Hair::Curve curve = hair->get_curve(j);
const array<float> &curve_radius = hair->get_curve_radius();
for (int segment = 0; segment < curve.num_segments(); ++segment, ++i) {
if (hair->curve_shape == CURVE_THICK) {
int k0 = curve.first_key + segment;
int k1 = k0 + 1;
int ka = max(k0 - 1, curve.first_key);
int kb = min(k1 + 1, curve.first_key + curve.num_keys - 1);
index_data[i] = i * 4;
float4 *const v = vertex_data.data() + step * num_vertices + index_data[i];
# if OPTIX_ABI_VERSION >= 55
v[0] = make_float4(keys[ka].x, keys[ka].y, keys[ka].z, curve_radius[ka]);
v[1] = make_float4(keys[k0].x, keys[k0].y, keys[k0].z, curve_radius[k0]);
v[2] = make_float4(keys[k1].x, keys[k1].y, keys[k1].z, curve_radius[k1]);
v[3] = make_float4(keys[kb].x, keys[kb].y, keys[kb].z, curve_radius[kb]);
# else
const float4 px = make_float4(keys[ka].x, keys[k0].x, keys[k1].x, keys[kb].x);
const float4 py = make_float4(keys[ka].y, keys[k0].y, keys[k1].y, keys[kb].y);
const float4 pz = make_float4(keys[ka].z, keys[k0].z, keys[k1].z, keys[kb].z);
const float4 pw = make_float4(
curve_radius[ka], curve_radius[k0], curve_radius[k1], curve_radius[kb]);
if (hair->curve_shape == CURVE_THICK) {
for (size_t curve_index = 0, segment_index = 0, vertex_index = step * num_vertices;
curve_index < hair->num_curves();
++curve_index)
{
const Hair::Curve curve = hair->get_curve(curve_index);
const array<float> &curve_radius = hair->get_curve_radius();
/* Convert Catmull-Rom data to B-spline. */
static const float4 cr2bsp0 = make_float4(+7, -4, +5, -2) / 6.f;
static const float4 cr2bsp1 = make_float4(-2, 11, -4, +1) / 6.f;
static const float4 cr2bsp2 = make_float4(+1, -4, 11, -2) / 6.f;
static const float4 cr2bsp3 = make_float4(-2, +5, -4, +7) / 6.f;
v[0] = make_float4(
dot(cr2bsp0, px), dot(cr2bsp0, py), dot(cr2bsp0, pz), dot(cr2bsp0, pw));
v[1] = make_float4(
dot(cr2bsp1, px), dot(cr2bsp1, py), dot(cr2bsp1, pz), dot(cr2bsp1, pw));
v[2] = make_float4(
dot(cr2bsp2, px), dot(cr2bsp2, py), dot(cr2bsp2, pz), dot(cr2bsp2, pw));
v[3] = make_float4(
dot(cr2bsp3, px), dot(cr2bsp3, py), dot(cr2bsp3, pz), dot(cr2bsp3, pw));
# endif
const int first_key_index = curve.first_key;
{
vertex_data[vertex_index++] = make_float4(keys[first_key_index].x,
keys[first_key_index].y,
keys[first_key_index].z,
curve_radius[first_key_index]);
}
else {
BoundBox bounds = BoundBox::empty;
curve.bounds_grow(segment, keys, hair->get_curve_radius().data(), bounds);
const size_t index = step * num_segments + i;
aabb_data[index].minX = bounds.min.x;
aabb_data[index].minY = bounds.min.y;
aabb_data[index].minZ = bounds.min.z;
aabb_data[index].maxX = bounds.max.x;
aabb_data[index].maxY = bounds.max.y;
aabb_data[index].maxZ = bounds.max.z;
for (int k = 0; k < curve.num_segments(); ++k) {
if (step == 0) {
index_data[segment_index++] = vertex_index - 1;
}
vertex_data[vertex_index++] = make_float4(keys[first_key_index + k].x,
keys[first_key_index + k].y,
keys[first_key_index + k].z,
curve_radius[first_key_index + k]);
}
const int last_key_index = first_key_index + curve.num_keys - 1;
{
vertex_data[vertex_index++] = make_float4(keys[last_key_index].x,
keys[last_key_index].y,
keys[last_key_index].z,
curve_radius[last_key_index]);
vertex_data[vertex_index++] = make_float4(keys[last_key_index].x,
keys[last_key_index].y,
keys[last_key_index].z,
curve_radius[last_key_index]);
}
}
}
else
# endif
{
for (size_t curve_index = 0, i = 0; curve_index < hair->num_curves(); ++curve_index) {
const Hair::Curve curve = hair->get_curve(curve_index);
for (int segment = 0; segment < curve.num_segments(); ++segment, ++i) {
# if OPTIX_ABI_VERSION < 55
if (hair->curve_shape == CURVE_THICK) {
const array<float> &curve_radius = hair->get_curve_radius();
int k0 = curve.first_key + segment;
int k1 = k0 + 1;
int ka = max(k0 - 1, curve.first_key);
int kb = min(k1 + 1, curve.first_key + curve.num_keys - 1);
index_data[i] = i * 4;
float4 *const v = vertex_data.data() + step * num_vertices + index_data[i];
const float4 px = make_float4(keys[ka].x, keys[k0].x, keys[k1].x, keys[kb].x);
const float4 py = make_float4(keys[ka].y, keys[k0].y, keys[k1].y, keys[kb].y);
const float4 pz = make_float4(keys[ka].z, keys[k0].z, keys[k1].z, keys[kb].z);
const float4 pw = make_float4(
curve_radius[ka], curve_radius[k0], curve_radius[k1], curve_radius[kb]);
/* Convert Catmull-Rom data to B-spline. */
static const float4 cr2bsp0 = make_float4(+7, -4, +5, -2) / 6.f;
static const float4 cr2bsp1 = make_float4(-2, 11, -4, +1) / 6.f;
static const float4 cr2bsp2 = make_float4(+1, -4, 11, -2) / 6.f;
static const float4 cr2bsp3 = make_float4(-2, +5, -4, +7) / 6.f;
v[0] = make_float4(
dot(cr2bsp0, px), dot(cr2bsp0, py), dot(cr2bsp0, pz), dot(cr2bsp0, pw));
v[1] = make_float4(
dot(cr2bsp1, px), dot(cr2bsp1, py), dot(cr2bsp1, pz), dot(cr2bsp1, pw));
v[2] = make_float4(
dot(cr2bsp2, px), dot(cr2bsp2, py), dot(cr2bsp2, pz), dot(cr2bsp2, pw));
v[3] = make_float4(
dot(cr2bsp3, px), dot(cr2bsp3, py), dot(cr2bsp3, pz), dot(cr2bsp3, pw));
}
else
# endif
{
BoundBox bounds = BoundBox::empty;
curve.bounds_grow(segment, keys, hair->get_curve_radius().data(), bounds);
const size_t index = step * num_segments + i;
aabb_data[index].minX = bounds.min.x;
aabb_data[index].minY = bounds.min.y;
aabb_data[index].minZ = bounds.min.z;
aabb_data[index].maxX = bounds.max.x;
aabb_data[index].maxY = bounds.max.y;
aabb_data[index].maxZ = bounds.max.z;
}
}
}
}