drewcassidy/nvidia-texture-tools
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Fix errors in new cluster fit compressor.

Browse Source
This commit is contained in:
castano
2012-01-02 08:49:13 +00:00
parent 23bfc1b514
commit f2d90ee844
16 changed files with 677 additions and 632 deletions
Download Patch File Download Diff File Expand all files Collapse all files

58
src/nvtt/ClusterFit.cpp
View File

@ -49,7 +49,7 @@ void ClusterFit::setColourSet(const ColorSet * set)
#endif
// cache some values
m_count = set->count;
m_count = set->colorCount;
Vector3 values[16];
for (uint i = 0; i < m_count; i++)
@ -148,7 +148,7 @@ bool ClusterFit::compress3( Vector3 * start, Vector3 * end )
SimdVector besterror = SimdVector( FLT_MAX );
SimdVector x0 = zero;
int b0 = 0, b1 = 0;
// check all possible clusters for this total order
@ -191,22 +191,22 @@ bool ClusterFit::compress3( Vector3 * start, Vector3 * end )
SimdVector e3 = negativeMultiplySubtract( b, betax_sum, e2 );
SimdVector e4 = multiplyAdd( two, e3, e1 );
// apply the metric to the error term
SimdVector e5 = e4 * m_metricSqr;
SimdVector error = e5.splatX() + e5.splatY() + e5.splatZ();
// apply the metric to the error term
SimdVector e5 = e4 * m_metricSqr;
SimdVector error = e5.splatX() + e5.splatY() + e5.splatZ();
// keep the solution if it wins
if( compareAnyLessThan( error, besterror ) )
{
besterror = error;
beststart = a;
bestend = b;
b0 = c0;
b1 = c1;
}
// keep the solution if it wins
if( compareAnyLessThan( error, besterror ) )
{
besterror = error;
beststart = a;
bestend = b;
b0 = c0;
b1 = c1;
}
x1 += m_weighted[c0+c1];
}
x1 += m_weighted[c0+c1];
}
x0 += m_weighted[c0];
}
@ -218,8 +218,8 @@ bool ClusterFit::compress3( Vector3 * start, Vector3 * end )
*start = beststart.toVector3();
*end = bestend.toVector3();
// save the error
m_besterror = besterror;
// save the error
m_besterror = besterror;
return true;
}
@ -308,10 +308,10 @@ bool ClusterFit::compress4( Vector3 * start, Vector3 * end )
}
x2 += m_weighted[c0+c1+c2];
}
}
x1 += m_weighted[c0+c1];
}
x1 += m_weighted[c0+c1];
}
x0 += m_weighted[c0];
}
@ -321,9 +321,9 @@ bool ClusterFit::compress4( Vector3 * start, Vector3 * end )
{
*start = beststart.toVector3();
*end = bestend.toVector3();
// save the error
m_besterror = besterror;
// save the error
m_besterror = besterror;
return true;
}
@ -404,12 +404,12 @@ bool ClusterFit::compress3(Vector3 * start, Vector3 * end)
// save the block if necessary
if( besterror < m_besterror )
{
*start = beststart;
*end = bestend;
// save the error
m_besterror = besterror;
// save the error
m_besterror = besterror;
return true;
}
@ -420,8 +420,8 @@ bool ClusterFit::compress3(Vector3 * start, Vector3 * end)
bool ClusterFit::compress4(Vector3 * start, Vector3 * end)
{
const uint count = m_count;
Vector3 const grid( 31.0f, 63.0f, 31.0f );
Vector3 const gridrcp( 1.0f/31.0f, 1.0f/63.0f, 1.0f/31.0f );
const Vector3 grid( 31.0f, 63.0f, 31.0f );
const Vector3 gridrcp( 1.0f/31.0f, 1.0f/63.0f, 1.0f/31.0f );
// declare variables
Vector3 beststart( 0.0f );

17
src/nvtt/QuickCompressDXT.cpp
View File

@ -179,8 +179,13 @@ inline static uint computeIndices4(const ColorSet & set, Vector3::Arg maxColor,
palette[3] = lerp(palette[0], palette[1], 2.0f / 3.0f);
uint indices = 0;
for(int i = 0; i < 16; i++)
for(int i = 0; i < 16; i++)
{
if (!set.isValidIndex(i)) {
// Skip masked pixels and out of bounds.
continue;
}
Vector3 color = set.color(i).xyz();
float d0 = colorDistance(palette[0], color);
@ -237,16 +242,20 @@ inline static uint computeIndices3(const ColorSet & set, Vector3::Arg maxColor,
uint indices = 0;
for(int i = 0; i < 16; i++)
{
if (!set.isValidIndex(i)) {
// Skip masked pixels and out of bounds.
indices |= 3 << (2 * i);
continue;
}
Vector3 color = set.color(i).xyz();
float alpha = set.color(i).w;
float d0 = colorDistance(palette[0], color);
float d1 = colorDistance(palette[1], color);
float d2 = colorDistance(palette[2], color);
uint index;
if (alpha == 0) index = 3;
else if (d0 < d1 && d0 < d2) index = 0;
if (d0 < d1 && d0 < d2) index = 0;
else if (d1 < d2) index = 1;
else index = 2;

34
src/nvtt/tools/imgdiff.cpp
View File

@ -72,11 +72,11 @@ struct Error
mse = 0.0f;
}
void addSample(float e)
void addSample(double e)
{
samples++;
mabse += fabsf(e);
maxabse = nv::max(maxabse, fabsf(e));
mabse += fabs(e);
maxabse = nv::max(maxabse, fabs(e));
mse += e * e;
}
@ -84,8 +84,8 @@ struct Error
{
mabse /= samples;
mse /= samples;
rmse = sqrtf(mse);
psnr = (rmse == 0) ? 999.0f : 20.0f * log10(255.0f / rmse);
rmse = sqrt(mse);
psnr = (rmse == 0) ? 999.0 : 20.0 * log10(255.0 / rmse);
}
void print()
@ -97,11 +97,11 @@ struct Error
}
int samples;
float mabse;
float maxabse;
float mse;
float rmse;
float psnr;
double mabse;
double maxabse;
double mse;
double rmse;
double psnr;
};
struct NormalError
@ -230,10 +230,10 @@ int main(int argc, char *argv[])
const nv::Color32 c0(image0.pixel(e, i));
const nv::Color32 c1(image1.pixel(e, i));
float r = float(c0.r - c1.r);
float g = float(c0.g - c1.g);
float b = float(c0.b - c1.b);
float a = float(c0.a - c1.a);
double r = float(c0.r - c1.r);
double g = float(c0.g - c1.g);
double b = float(c0.b - c1.b);
double a = float(c0.a - c1.a);
error_r.addSample(r);
error_g.addSample(g);
@ -247,9 +247,9 @@ int main(int argc, char *argv[])
if (compareAlpha)
{
error_total.addSample(r * c0.a / 255.0f);
error_total.addSample(g * c0.a / 255.0f);
error_total.addSample(b * c0.a / 255.0f);
error_total.addSample(r * c0.a / 255.0);
error_total.addSample(g * c0.a / 255.0);
error_total.addSample(b * c0.a / 255.0);
}
else
{