optimize applying block changes

This commit is contained in:
p2r3
2025-08-14 07:25:38 +03:00
parent 106e3e9ba2
commit 374943669e
2 changed files with 38 additions and 9 deletions

View File

@@ -256,7 +256,6 @@ int givePlayerItem (int client_fd, uint16_t item) {
}
uint8_t getBlockChange (short x, short y, short z) {
short tmp;
for (int i = 0; i < block_changes_count; i ++) {
if (block_changes[i].block == 0xFF) continue;
if (
@@ -270,7 +269,6 @@ uint8_t getBlockChange (short x, short y, short z) {
void makeBlockChange (short x, short y, short z, uint8_t block) {
short tmp;
for (int i = 0; i < block_changes_count; i ++) {
if (
block_changes[i].x == x &&

View File

@@ -73,10 +73,7 @@ int getHeightAt (int rx, int rz, int _x, int _z, uint32_t chunk_hash) {
}
uint8_t getBlockAt (int x, int y, int z) {
uint8_t block_change = getBlockChange(x, y, z);
if (block_change != 0xFF) return block_change;
uint8_t getTerrainAt (int x, int y, int z) {
if (y > 80) return B_air;
@@ -142,21 +139,55 @@ skip_tree:
}
uint8_t getBlockAt (int x, int y, int z) {
uint8_t block_change = getBlockChange(x, y, z);
if (block_change != 0xFF) return block_change;
return getTerrainAt(x, y, z);
}
uint8_t chunk_section[4096];
void buildChunkSection (int cx, int cy, int cz) {
// Generate 4096 blocks in one buffer to reduce overhead
for (int j = 0; j < 4096; j += 8) {
// These values don't change in the lower array,
// since all of the operations are on multiples of 8
int y = j / 256 + cy;
int z = j / 16 % 16 + cz;
// The client expects "big-endian longs", which in our
// case means reversing the order in which we store/send
// each 8 block sequence.
for (int offset = 7; offset >= 0; offset--) {
int k = j + offset;
int x = k % 16 + cx;
chunk_section[j + 7 - offset] = getBlockAt(x, y, z);
chunk_section[j + 7 - offset] = getTerrainAt(x, y, z);
}
}
// Apply block changes on top of terrain
// This does mean that we're generating some terrain only to replace it,
// but it's better to apply changes in one run rather than in individual
// runs per block, as this is more expensive than terrain generation.
for (int i = 0; i < block_changes_count; i ++) {
if (block_changes[i].block == 0xFF) continue;
if ( // Check if block is within this chunk section
block_changes[i].x >= cx && block_changes[i].x < cx + 16 &&
block_changes[i].y >= cy && block_changes[i].y < cy + 16 &&
block_changes[i].z >= cz && block_changes[i].z < cz + 16
) {
int dx = block_changes[i].x - cx;
int dy = block_changes[i].y - cy;
int dz = block_changes[i].z - cz;
// Same 8-block sequence reversal as before, this time 10x dirtier
// because we're working with specific indexes.
unsigned address = (unsigned)(dx + (dz << 4) + (dy << 8));
unsigned index = (address & ~7u) | (7u - (address & 7u));
chunk_section[index] = block_changes[i].block;
}
}
}