Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

Performance improvements for GLES2 CPU blendshapes #51363

Merged
merged 1 commit into from
Aug 16, 2021
Merged

Performance improvements for GLES2 CPU blendshapes #51363

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
12 changes: 12 additions & 0 deletions drivers/dummy/rasterizer_dummy.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -146,6 +146,7 @@ class RasterizerStorageDummy : public RasterizerStorage {
Vector<DummySurface> surfaces;
int blend_shape_count;
VS::BlendShapeMode blend_shape_mode;
PoolRealArray blend_shape_values;
};

mutable RID_Owner<DummyTexture> texture_owner;
Expand Down Expand Up @@ -336,6 +337,17 @@ class RasterizerStorageDummy : public RasterizerStorage {
return m->blend_shape_mode;
}

void mesh_set_blend_shape_values(RID p_mesh, PoolVector<float> p_values) {
DummyMesh *m = mesh_owner.getornull(p_mesh);
ERR_FAIL_COND(!m);
m->blend_shape_values = p_values;
}
PoolVector<float> mesh_get_blend_shape_values(RID p_mesh) const {
DummyMesh *m = mesh_owner.getornull(p_mesh);
ERR_FAIL_COND_V(!m, PoolRealArray());
return m->blend_shape_values;
}

void mesh_surface_update_region(RID p_mesh, int p_surface, int p_offset, const PoolVector<uint8_t> &p_data) {}

void mesh_surface_set_material(RID p_mesh, int p_surface, RID p_material) {}
Expand Down
250 changes: 4 additions & 246 deletions drivers/gles2/rasterizer_scene_gles2.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1385,239 +1385,6 @@ bool RasterizerSceneGLES2::_setup_material(RasterizerStorageGLES2::Material *p_m
return shader_rebind;
}

void static _calculate_blend_shape_buffer(RasterizerSceneGLES2::RenderList::Element *p_element, PoolVector<float> &transform_buffer) {
RasterizerStorageGLES2::Surface *s = static_cast<RasterizerStorageGLES2::Surface *>(p_element->geometry);
if (!s->blend_shape_data.empty()) {
if (transform_buffer.size() < s->array_byte_size) {
transform_buffer.resize(s->array_byte_size);
}
for (int i = 0; i < VS::ARRAY_MAX - 1; i++) {
if (s->attribs[i].enabled) {
const float *p_weights = p_element->instance->blend_values.ptr();

PoolVector<float>::Write write = transform_buffer.write();
PoolVector<uint8_t>::Read read = s->data.read();
float attrib_array[4] = { 0.0 };

// Read all attributes
for (int j = 0; j < s->array_len; j++) {
size_t offset = s->attribs[i].offset + (j * s->attribs[i].stride);
float base_weight = 1.0;

if (s->mesh->blend_shape_mode == VS::BLEND_SHAPE_MODE_NORMALIZED) {
for (int ti = 0; ti < s->blend_shape_data.size(); ti++) {
base_weight -= p_weights[ti];
}
}

// Set the base
switch (i) {
case VS::ARRAY_VERTEX: {
if (s->format & VS::ARRAY_COMPRESS_VERTEX) {
const uint16_t *v = (const uint16_t *)(read.ptr() + offset);
attrib_array[0] = Math::halfptr_to_float(&v[0]) * base_weight;
attrib_array[1] = Math::halfptr_to_float(&v[1]) * base_weight;
attrib_array[2] = Math::halfptr_to_float(&v[2]) * base_weight;
} else {
const float *v = (const float *)(read.ptr() + offset);
attrib_array[0] = v[0] * base_weight;
attrib_array[1] = v[1] * base_weight;
attrib_array[2] = v[2] * base_weight;
}
} break;
case VS::ARRAY_NORMAL: {
if (s->format & VS::ARRAY_COMPRESS_NORMAL) {
const int8_t *v = (const int8_t *)(read.ptr() + offset);
attrib_array[0] = (v[0] / 127.0) * base_weight;
attrib_array[1] = (v[1] / 127.0) * base_weight;
attrib_array[2] = (v[2] / 127.0) * base_weight;
} else {
const float *v = (const float *)(read.ptr() + offset);
attrib_array[0] = v[0] * base_weight;
attrib_array[1] = v[1] * base_weight;
attrib_array[2] = v[2] * base_weight;
}
} break;
case VS::ARRAY_TANGENT: {
if (s->format & VS::ARRAY_COMPRESS_TANGENT) {
const int8_t *v = (const int8_t *)(read.ptr() + offset);
attrib_array[0] = (v[0] / 127.0) * base_weight;
attrib_array[1] = (v[1] / 127.0) * base_weight;
attrib_array[2] = (v[2] / 127.0) * base_weight;
attrib_array[3] = (v[3] / 127.0) * base_weight;
} else {
const float *v = (const float *)(read.ptr() + offset);
attrib_array[0] = v[0] * base_weight;
attrib_array[1] = v[1] * base_weight;
attrib_array[2] = v[2] * base_weight;
attrib_array[3] = v[3] * base_weight;
}
} break;
case VS::ARRAY_COLOR: {
if (s->format & VS::ARRAY_COMPRESS_COLOR) {
const uint8_t *v = (const uint8_t *)(read.ptr() + offset);
attrib_array[0] = (v[0] / 255.0) * base_weight;
attrib_array[1] = (v[1] / 255.0) * base_weight;
attrib_array[2] = (v[2] / 255.0) * base_weight;
attrib_array[3] = (v[3] / 255.0) * base_weight;
} else {
const float *v = (const float *)(read.ptr() + offset);
attrib_array[0] = v[0] * base_weight;
attrib_array[1] = v[1] * base_weight;
attrib_array[2] = v[2] * base_weight;
attrib_array[3] = v[3] * base_weight;
}
} break;
case VS::ARRAY_TEX_UV: {
if (s->format & VS::ARRAY_COMPRESS_TEX_UV) {
const uint16_t *v = (const uint16_t *)(read.ptr() + offset);
attrib_array[0] = Math::halfptr_to_float(&v[0]) * base_weight;
attrib_array[1] = Math::halfptr_to_float(&v[1]) * base_weight;
} else {
const float *v = (const float *)(read.ptr() + offset);
attrib_array[0] = v[0] * base_weight;
attrib_array[1] = v[1] * base_weight;
}
} break;
case VS::ARRAY_TEX_UV2: {
if (s->format & VS::ARRAY_COMPRESS_TEX_UV2) {
const uint16_t *v = (const uint16_t *)(read.ptr() + offset);
attrib_array[0] = Math::halfptr_to_float(&v[0]) * base_weight;
attrib_array[1] = Math::halfptr_to_float(&v[1]) * base_weight;
} else {
const float *v = (const float *)(read.ptr() + offset);
attrib_array[0] = v[0] * base_weight;
attrib_array[1] = v[1] * base_weight;
}
} break;
case VS::ARRAY_WEIGHTS: {
if (s->format & VS::ARRAY_COMPRESS_WEIGHTS) {
const uint16_t *v = (const uint16_t *)(read.ptr() + offset);
attrib_array[0] = (v[0] / 65535.0) * base_weight;
attrib_array[1] = (v[1] / 65535.0) * base_weight;
attrib_array[2] = (v[2] / 65535.0) * base_weight;
attrib_array[3] = (v[3] / 65535.0) * base_weight;
} else {
const float *v = (const float *)(read.ptr() + offset);
attrib_array[0] = v[0] * base_weight;
attrib_array[1] = v[1] * base_weight;
attrib_array[2] = v[2] * base_weight;
attrib_array[3] = v[3] * base_weight;
}
} break;
}

// Add all blend shapes
for (int ti = 0; ti < s->blend_shape_data.size(); ti++) {
PoolVector<uint8_t>::Read blend = s->blend_shape_data[ti].read();
float weight = p_weights[ti];
if (Math::is_zero_approx(weight)) {
continue;
}

switch (i) {
case VS::ARRAY_VERTEX: {
if (s->format & VS::ARRAY_COMPRESS_VERTEX) {
const uint16_t *v = (const uint16_t *)(blend.ptr() + offset);
attrib_array[0] += Math::halfptr_to_float(&v[0]) * weight;
attrib_array[1] += Math::halfptr_to_float(&v[1]) * weight;
attrib_array[2] += Math::halfptr_to_float(&v[2]) * weight;
} else {
const float *v = (const float *)(blend.ptr() + offset);
attrib_array[0] += v[0] * weight;
attrib_array[1] += v[1] * weight;
attrib_array[2] += v[2] * weight;
}
} break;
case VS::ARRAY_NORMAL: {
if (s->format & VS::ARRAY_COMPRESS_NORMAL) {
const int8_t *v = (const int8_t *)(blend.ptr() + offset);
attrib_array[0] += (float(v[0]) / 127.0) * weight;
attrib_array[1] += (float(v[1]) / 127.0) * weight;
attrib_array[2] += (float(v[2]) / 127.0) * weight;
} else {
const float *v = (const float *)(blend.ptr() + offset);
attrib_array[0] += v[0] * weight;
attrib_array[1] += v[1] * weight;
attrib_array[2] += v[2] * weight;
}
} break;
case VS::ARRAY_TANGENT: {
if (s->format & VS::ARRAY_COMPRESS_TANGENT) {
const int8_t *v = (const int8_t *)(read.ptr() + offset);
attrib_array[0] += (float(v[0]) / 127.0) * weight;
attrib_array[1] += (float(v[1]) / 127.0) * weight;
attrib_array[2] += (float(v[2]) / 127.0) * weight;
attrib_array[3] = (float(v[3]) / 127.0);
} else {
const float *v = (const float *)(read.ptr() + offset);
attrib_array[0] += v[0] * weight;
attrib_array[1] += v[1] * weight;
attrib_array[2] += v[2] * weight;
attrib_array[3] = v[3];
}
} break;
case VS::ARRAY_COLOR: {
if (s->format & VS::ARRAY_COMPRESS_COLOR) {
const uint8_t *v = (const uint8_t *)(blend.ptr() + offset);
attrib_array[0] += (v[0] / 255.0) * weight;
attrib_array[1] += (v[1] / 255.0) * weight;
attrib_array[2] += (v[2] / 255.0) * weight;
attrib_array[3] += (v[3] / 255.0) * weight;
} else {
const float *v = (const float *)(blend.ptr() + offset);
attrib_array[0] += v[0] * weight;
attrib_array[1] += v[1] * weight;
attrib_array[2] += v[2] * weight;
attrib_array[3] += v[3] * weight;
}
} break;
case VS::ARRAY_TEX_UV: {
if (s->format & VS::ARRAY_COMPRESS_TEX_UV) {
const uint16_t *v = (const uint16_t *)(blend.ptr() + offset);
attrib_array[0] += Math::halfptr_to_float(&v[0]) * weight;
attrib_array[1] += Math::halfptr_to_float(&v[1]) * weight;
} else {
const float *v = (const float *)(blend.ptr() + offset);
attrib_array[0] += v[0] * weight;
attrib_array[1] += v[1] * weight;
}
} break;
case VS::ARRAY_TEX_UV2: {
if (s->format & VS::ARRAY_COMPRESS_TEX_UV2) {
const uint16_t *v = (const uint16_t *)(blend.ptr() + offset);
attrib_array[0] += Math::halfptr_to_float(&v[0]) * weight;
attrib_array[1] += Math::halfptr_to_float(&v[1]) * weight;
} else {
const float *v = (const float *)(blend.ptr() + offset);
attrib_array[0] += v[0] * weight;
attrib_array[1] += v[1] * weight;
}
} break;
case VS::ARRAY_WEIGHTS: {
if (s->format & VS::ARRAY_COMPRESS_WEIGHTS) {
const uint16_t *v = (const uint16_t *)(blend.ptr() + offset);
attrib_array[0] += (v[0] / 65535.0) * weight;
attrib_array[1] += (v[1] / 65535.0) * weight;
attrib_array[2] += (v[2] / 65535.0) * weight;
attrib_array[3] += (v[3] / 65535.0) * weight;
} else {
const float *v = (const float *)(blend.ptr() + offset);
attrib_array[0] += v[0] * weight;
attrib_array[1] += v[1] * weight;
attrib_array[2] += v[2] * weight;
attrib_array[3] += v[3] * weight;
}
} break;
}
}
memcpy(&write[offset], attrib_array, sizeof(float) * s->attribs[i].size);
}
}
}
}
}

void RasterizerSceneGLES2::_setup_geometry(RenderList::Element *p_element, RasterizerStorageGLES2::Skeleton *p_skeleton) {
switch (p_element->instance->base_type) {
case VS::INSTANCE_MESH: {
Expand All @@ -1627,25 +1394,16 @@ void RasterizerSceneGLES2::_setup_geometry(RenderList::Element *p_element, Raste
glBindBuffer(GL_ELEMENT_ARRAY_BUFFER, s->index_id);
}

if (!s->blend_shape_data.empty()) {
_calculate_blend_shape_buffer(p_element, storage->resources.blend_shapes_transform_cpu_buffer);
storage->_update_blend_shape_transform_buffer(storage->resources.blend_shapes_transform_cpu_buffer, s->array_byte_size);
}

for (int i = 0; i < VS::ARRAY_MAX - 1; i++) {
if (s->attribs[i].enabled) {
if (!s->blend_shape_data.empty() && (i != VS::ARRAY_BONES)) {
glBindBuffer(GL_ARRAY_BUFFER, storage->resources.blend_shape_transform_buffer);

glEnableVertexAttribArray(i);
glEnableVertexAttribArray(i);

glVertexAttribPointer(s->attribs[i].index, s->attribs[i].size, GL_FLOAT, GL_FALSE, s->attribs[i].stride * sizeof(float), CAST_INT_TO_UCHAR_PTR(s->attribs[i].offset * sizeof(float)));
if (!s->blend_shape_data.empty() && i != VS::ARRAY_BONES && s->blend_shape_buffer_size > 0) {
glBindBuffer(GL_ARRAY_BUFFER, s->blend_shape_buffer_id);
glVertexAttribPointer(s->attribs[i].index, s->attribs[i].size, GL_FLOAT, GL_FALSE, 8 * 4 * sizeof(float), CAST_INT_TO_UCHAR_PTR(i * 4 * sizeof(float)));

} else {
glBindBuffer(GL_ARRAY_BUFFER, s->vertex_id);

glEnableVertexAttribArray(i);

glVertexAttribPointer(s->attribs[i].index, s->attribs[i].size, s->attribs[i].type, s->attribs[i].normalized, s->attribs[i].stride, CAST_INT_TO_UCHAR_PTR(s->attribs[i].offset));
}
} else {
Expand Down
Loading