server_common/recastnavigation_v1.6.0
5
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Remove redundant copying in rcBuildRegions. (#314)

Browse Source 
Remove dstReg and dstDist from rcBuildRegions entirely -- just use temporary scratch space in expandRegions.
This saves a considerable amount of time when the number of spans is large. A sample map tried has 16M spans, and only writes at most 4000 of them.
This commit is contained in:
mbabinski-at-google 2018-04-27 15:08:52 -05:00 committed by Jakob Botsch Nielsen
parent 4566d01c8f
commit 18562383f4
1 changed files with 23 additions and 33 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

56
Recast/Source/RecastRegion.cpp
View File

@ -343,12 +343,11 @@ static bool floodRegion(int x, int y, int i,
return count > 0; return count > 0;
} }
static unsigned short* expandRegions(int maxIter, unsigned short level, static void expandRegions(int maxIter, unsigned short level,
rcCompactHeightfield& chf, rcCompactHeightfield& chf,
unsigned short* srcReg, unsigned short* srcDist, unsigned short* srcReg, unsigned short* srcDist,
unsigned short* dstReg, unsigned short* dstDist, rcIntArray& stack,
rcIntArray& stack, bool fillStack)
bool fillStack)
{ {
const int w = chf.width; const int w = chf.width;
const int h = chf.height; const int h = chf.height;
@ -385,13 +384,12 @@ static unsigned short* expandRegions(int maxIter, unsigned short level,
} }
} }
rcIntArray dirtyEntries;
int iter = 0; int iter = 0;
while (stack.size() > 0) while (stack.size() > 0)
{ {
int failed = 0; int failed = 0;
dirtyEntries.resize(0);
memcpy(dstReg, srcReg, sizeof(unsigned short)*chf.spanCount);
memcpy(dstDist, srcDist, sizeof(unsigned short)*chf.spanCount);
for (int j = 0; j < stack.size(); j += 3) for (int j = 0; j < stack.size(); j += 3)
{ {
@ -427,8 +425,9 @@ static unsigned short* expandRegions(int maxIter, unsigned short level,
if (r) if (r)
{ {
stack[j+2] = -1; // mark as used stack[j+2] = -1; // mark as used
dstReg[i] = r; dirtyEntries.push(i);
dstDist[i] = d2; dirtyEntries.push(r);
dirtyEntries.push(d2);
} }
else else
{ {
@ -436,9 +435,12 @@ static unsigned short* expandRegions(int maxIter, unsigned short level,
} }
} }
// rcSwap source and dest. // Copy entries that differ between src and dst to keep them in sync.
rcSwap(srcReg, dstReg); for (int i = 0; i < dirtyEntries.size(); i+=3) {
rcSwap(srcDist, dstDist); int idx = dirtyEntries[i];
srcReg[idx] = (unsigned short)dirtyEntries[i+1];
srcDist[idx] = (unsigned short)dirtyEntries[i+2];
}
if (failed*3 == stack.size()) if (failed*3 == stack.size())
break; break;
@ -450,15 +452,13 @@ static unsigned short* expandRegions(int maxIter, unsigned short level,
break; break;
} }
} }
return srcReg;
} }
static void sortCellsByLevel(unsigned short startLevel, static void sortCellsByLevel(unsigned short startLevel,
rcCompactHeightfield& chf, rcCompactHeightfield& chf,
unsigned short* srcReg, const unsigned short* srcReg,
unsigned int nbStacks, rcIntArray* stacks, unsigned int nbStacks, rcIntArray* stacks,
unsigned short loglevelsPerStack) // the levels per stack (2 in our case) as a bit shift unsigned short loglevelsPerStack) // the levels per stack (2 in our case) as a bit shift
{ {
@ -497,7 +497,7 @@ static void sortCellsByLevel(unsigned short startLevel,
static void appendStacks(rcIntArray& srcStack, rcIntArray& dstStack, static void appendStacks(rcIntArray& srcStack, rcIntArray& dstStack,
unsigned short* srcReg) const unsigned short* srcReg)
{ {
for (int j=0; j<srcStack.size(); j+=3) for (int j=0; j<srcStack.size(); j+=3)
{ {
@ -671,7 +671,7 @@ static bool isRegionConnectedToBorder(const rcRegion& reg)
return false; return false;
} }
static bool isSolidEdge(rcCompactHeightfield& chf, unsigned short* srcReg, static bool isSolidEdge(rcCompactHeightfield& chf, const unsigned short* srcReg,
int x, int y, int i, int dir) int x, int y, int i, int dir)
{ {
const rcCompactSpan& s = chf.spans[i]; const rcCompactSpan& s = chf.spans[i];
@ -690,7 +690,7 @@ static bool isSolidEdge(rcCompactHeightfield& chf, unsigned short* srcReg,
static void walkContour(int x, int y, int i, int dir, static void walkContour(int x, int y, int i, int dir,
rcCompactHeightfield& chf, rcCompactHeightfield& chf,
unsigned short* srcReg, const unsigned short* srcReg,
rcIntArray& cont) rcIntArray& cont)
{ {
int startDir = dir; int startDir = dir;
@ -1535,7 +1535,7 @@ bool rcBuildRegions(rcContext* ctx, rcCompactHeightfield& chf,
const int w = chf.width; const int w = chf.width;
const int h = chf.height; const int h = chf.height;
rcScopedDelete<unsigned short> buf((unsigned short*)rcAlloc(sizeof(unsigned short)*chf.spanCount*4, RC_ALLOC_TEMP)); rcScopedDelete<unsigned short> buf((unsigned short*)rcAlloc(sizeof(unsigned short)*chf.spanCount*2, RC_ALLOC_TEMP));
if (!buf) if (!buf)
{ {
ctx->log(RC_LOG_ERROR, "rcBuildRegions: Out of memory 'tmp' (%d).", chf.spanCount*4); ctx->log(RC_LOG_ERROR, "rcBuildRegions: Out of memory 'tmp' (%d).", chf.spanCount*4);
@ -1555,8 +1555,6 @@ bool rcBuildRegions(rcContext* ctx, rcCompactHeightfield& chf,
unsigned short* srcReg = buf; unsigned short* srcReg = buf;
unsigned short* srcDist = buf+chf.spanCount; unsigned short* srcDist = buf+chf.spanCount;
unsigned short* dstReg = buf+chf.spanCount*2;
unsigned short* dstDist = buf+chf.spanCount*3;
memset(srcReg, 0, sizeof(unsigned short)*chf.spanCount); memset(srcReg, 0, sizeof(unsigned short)*chf.spanCount);
memset(srcDist, 0, sizeof(unsigned short)*chf.spanCount); memset(srcDist, 0, sizeof(unsigned short)*chf.spanCount);
@ -1604,11 +1602,7 @@ bool rcBuildRegions(rcContext* ctx, rcCompactHeightfield& chf,
rcScopedTimer timerExpand(ctx, RC_TIMER_BUILD_REGIONS_EXPAND); rcScopedTimer timerExpand(ctx, RC_TIMER_BUILD_REGIONS_EXPAND);
// Expand current regions until no empty connected cells found. // Expand current regions until no empty connected cells found.
if (expandRegions(expandIters, level, chf, srcReg, srcDist, dstReg, dstDist, lvlStacks[sId], false) != srcReg) expandRegions(expandIters, level, chf, srcReg, srcDist, lvlStacks[sId], false);
{
rcSwap(srcReg, dstReg);
rcSwap(srcDist, dstDist);
}
} }
{ {
@ -1638,11 +1632,7 @@ bool rcBuildRegions(rcContext* ctx, rcCompactHeightfield& chf,
} }
// Expand current regions until no empty connected cells found. // Expand current regions until no empty connected cells found.
if (expandRegions(expandIters*8, 0, chf, srcReg, srcDist, dstReg, dstDist, stack, true) != srcReg) expandRegions(expandIters*8, 0, chf, srcReg, srcDist, stack, true);
{
rcSwap(srcReg, dstReg);
rcSwap(srcDist, dstDist);
}
ctx->stopTimer(RC_TIMER_BUILD_REGIONS_WATERSHED); ctx->stopTimer(RC_TIMER_BUILD_REGIONS_WATERSHED);