Skip to content

Commit

Permalink
Browse files Browse the repository at this point in the history
Optimise VoxelManipulator::copyFrom
~3-4x faster

This indirectly optimises MapBlock:copyTo() which in turn improves performance of MeshMakeData::fill()
  • Loading branch information
Zeno- committed Dec 4, 2014
1 parent f6912f4 commit c16b1e2
Showing 1 changed file with 42 additions and 7 deletions.
49 changes: 42 additions & 7 deletions src/voxel.cpp
Expand Up @@ -224,13 +224,48 @@ void VoxelManipulator::addArea(const VoxelArea &area)
void VoxelManipulator::copyFrom(MapNode *src, const VoxelArea& src_area,
v3s16 from_pos, v3s16 to_pos, v3s16 size)
{
for(s16 z=0; z<size.Z; z++)
for(s16 y=0; y<size.Y; y++)
{
s32 i_src = src_area.index(from_pos.X, from_pos.Y+y, from_pos.Z+z);
s32 i_local = m_area.index(to_pos.X, to_pos.Y+y, to_pos.Z+z);
memcpy(&m_data[i_local], &src[i_src], size.X*sizeof(MapNode));
memset(&m_flags[i_local], 0, size.X);
/* The reason for this optimised code is that we're a member function
* and the data type/layout of m_data is know to us: it's stored as
* [z*h*w + y*h + x]. Therefore we can take the calls to m_area index
* (which performs the preceding mapping/indexing of m_data) out of the
* inner loop and calculate the next index as we're iterating to gain
* performance.
*
* src_step and dest_step is the amount required to be added to our index
* every time y increments. Because the destination area may be larger
* than the source area we need one additional variable (otherwise we could
* just continue adding dest_step as is done for the source data): dest_mod.
* dest_mod is the difference in size between a "row" in the source data
* and a "row" in the destination data (I am using the term row loosely
* and for illustrative purposes). E.g.
*
* src <-------------------->|'''''' dest mod ''''''''
* dest <--------------------------------------------->
*
* dest_mod (it's essentially a modulus) is added to the destination index
* after every full iteration of the y span.
*
* This method falls under the category "linear array and incrementing
* index".
*/

s32 src_step = src_area.getExtent().X;
s32 dest_step = m_area.getExtent().X;
s32 dest_mod = m_area.index(to_pos.X, to_pos.Y, to_pos.Z + 1)
- m_area.index(to_pos.X, to_pos.Y, to_pos.Z)
- dest_step * size.Y;

s32 i_src = src_area.index(from_pos.X, from_pos.Y, from_pos.Z);
s32 i_local = m_area.index(to_pos.X, to_pos.Y, to_pos.Z);

for (s16 z = 0; z < size.Z; z++) {
for (s16 y = 0; y < size.Y; y++) {
memcpy(&m_data[i_local], &src[i_src], size.X * sizeof(*m_data));
memset(&m_flags[i_local], 0, size.X);
i_src += src_step;
i_local += dest_step;
}
i_local += dest_mod;
}
}

Expand Down

0 comments on commit c16b1e2

Please sign in to comment.