mirror of
https://github.com/godotengine/godot.git
synced 2025-10-19 16:03:29 +00:00
Import mesh colors in 8BPP.
* Colors were imported as 16BPP (half float) * Far most common use cases only require 8BPP * If you need higher data precision, use a custom array, which are supported now. **WARNING**: 3D Scenes imported in 4.0 no longer compatible with this new format. You need to re-import them (erase them from .godot/import)
This commit is contained in:
parent
3530cb639a
commit
9ad0c6cde7
4 changed files with 33 additions and 31 deletions
|
@ -438,13 +438,14 @@ Error RenderingServer::_surface_set_data(Array p_arrays, uint32_t p_format, uint
|
|||
ERR_FAIL_COND_V(array.size() != p_vertex_array_len, ERR_INVALID_PARAMETER);
|
||||
|
||||
const Color *src = array.ptr();
|
||||
uint16_t color16[4];
|
||||
for (int i = 0; i < p_vertex_array_len; i++) {
|
||||
color16[0] = Math::make_half_float(src[i].r);
|
||||
color16[1] = Math::make_half_float(src[i].g);
|
||||
color16[2] = Math::make_half_float(src[i].b);
|
||||
color16[3] = Math::make_half_float(src[i].a);
|
||||
memcpy(&aw[p_offsets[ai] + i * p_attrib_stride], color16, 8);
|
||||
uint8_t color8[4] = {
|
||||
uint8_t(CLAMP(src[i].r * 255.0, 0.0, 255.0)),
|
||||
uint8_t(CLAMP(src[i].g * 255.0, 0.0, 255.0)),
|
||||
uint8_t(CLAMP(src[i].b * 255.0, 0.0, 255.0)),
|
||||
uint8_t(CLAMP(src[i].a * 255.0, 0.0, 255.0))
|
||||
};
|
||||
memcpy(&aw[p_offsets[ai] + i * p_attrib_stride], color8, 4);
|
||||
}
|
||||
} break;
|
||||
case RS::ARRAY_TEX_UV: {
|
||||
|
@ -761,7 +762,7 @@ void RenderingServer::mesh_surface_make_offsets_from_format(uint32_t p_format, i
|
|||
elem_size = 4;
|
||||
} break;
|
||||
case RS::ARRAY_COLOR: {
|
||||
elem_size = 8;
|
||||
elem_size = 4;
|
||||
} break;
|
||||
case RS::ARRAY_TEX_UV: {
|
||||
elem_size = 8;
|
||||
|
@ -1123,8 +1124,9 @@ Array RenderingServer::_get_array_from_surface(uint32_t p_format, Vector<uint8_t
|
|||
Color *w = arr.ptrw();
|
||||
|
||||
for (int32_t j = 0; j < p_vertex_len; j++) {
|
||||
const uint16_t *v = (const uint16_t *)&ar[j * attrib_elem_size + offsets[i]];
|
||||
w[j] = Color(Math::half_to_float(v[0]), Math::half_to_float(v[1]), Math::half_to_float(v[2]), Math::half_to_float(v[3]));
|
||||
const uint8_t *v = (const uint8_t *)&ar[j * attrib_elem_size + offsets[i]];
|
||||
|
||||
w[j] = Color(v[0] / 255.0, v[1] / 255.0, v[2] / 255.0, v[3] / 255.0);
|
||||
}
|
||||
|
||||
ret[i] = arr;
|
||||
|
|
Loading…
Add table
Add a link
Reference in a new issue