void FastClusterFit::Compress4( void* block )
{
// declare variables
Vec3 beststart( 0.0f );
Vec3 bestend( 0.0f );
float besterror = FLT_MAX;
Vec3 x0(0.0f);
Vec3 x1;
Vec3 x2;
int b0 = 0, b1 = 0, b2 = 0;
int i = 0;
// check all possible clusters for this total order
for( int c0 = 0; c0 <= 16; c0++)
{
x1 = Vec3(0.0f);
for( int c1 = 0; c1 <= 16-c0; c1++)
{
x2 = Vec3(0.0f);
for( int c2 = 0; c2 <= 16-c0-c1; c2++)
{
float const alpha2_sum = s_fourElement[i].alpha2_sum;
float const beta2_sum = s_fourElement[i].beta2_sum;
float const alphabeta_sum = s_fourElement[i].alphabeta_sum;
float const factor = s_fourElement[i].factor;
i++;
Vec3 const alphax_sum = x0 + x1 * (2.0f / 3.0f) + x2 * (1.0f / 3.0f);
Vec3 const betax_sum = m_xsum - alphax_sum;
Vec3 a = ( alphax_sum*beta2_sum - betax_sum*alphabeta_sum )*factor;
Vec3 b = ( betax_sum*alpha2_sum - alphax_sum*alphabeta_sum )*factor;
// clamp the output to [0, 1]
Vec3 const one( 1.0f );
Vec3 const zero( 0.0f );
a = Min( one, Max( zero, a ) );
b = Min( one, Max( zero, b ) );
// clamp to the grid
Vec3 const grid( 31.0f, 63.0f, 31.0f );
Vec3 const gridrcp( 0.03227752766457f, 0.01583151765563f, 0.03227752766457f );
Vec3 const half( 0.5f );
a = Floor( grid*a + half )*gridrcp;
b = Floor( grid*b + half )*gridrcp;
// compute the error
Vec3 e1 = a*a*alpha2_sum + b*b*beta2_sum + 2.0f*( a*b*alphabeta_sum - a*alphax_sum - b*betax_sum );
// apply the metric to the error term
float error = Dot( e1, m_metricSqr );
// keep the solution if it wins
if( error < besterror )
{
besterror = error;
beststart = a;
bestend = b;
b0 = c0;
b1 = c1;
b2 = c2;
}
x2 += m_unweighted[c0+c1+c2];
}
x1 += m_unweighted[c0+c1];
}
x0 += m_unweighted[c0];
}
// save the block if necessary
if( besterror < m_besterror )
{
// compute indices from cluster sizes.
/*uint bestindices = 0;
{
int i = b0;
for(; i < b0+b1; i++) {
bestindices = 2 << (2 * m_order[i]);
}
for(; i < b0+b1+b2; i++) {
bestindices = 3 << (2 * m_order[i]);
}
for(; i < 16; i++) {
bestindices = 1 << (2 * m_order[i]);
}
}*/
u8 bestindices[16];
{
int i = 0;
for(; i < b0; i++) {
bestindices[i] = 0;
}
for(; i < b0+b1; i++) {
bestindices[i] = 2;
}
for(; i < b0+b1+b2; i++) {
bestindices[i] = 3;
}
for(; i < 16; i++) {
bestindices[i] = 1;
}
}
// remap the indices
u8 ordered[16];
for( int i = 0; i < 16; ++i )
ordered[m_order[i]] = bestindices[i];
// save the block
WriteColourBlock4( beststart, bestend, ordered, block );
// save the error
m_besterror = besterror;
}
}
void WeightedClusterFit::Compress4( void* block )
{
int const count = m_colours->GetCount();
Vec3 const one( 1.0f );
Vec3 const zero( 0.0f );
Vec3 const half( 0.5f );
Vec3 const grid( 31.0f, 63.0f, 31.0f );
Vec3 const gridrcp( 1.0f/31.0f, 1.0f/63.0f, 1.0f/31.0f );
// declare variables
Vec3 beststart( 0.0f );
Vec3 bestend( 0.0f );
float besterror = FLT_MAX;
Vec3 x0(0.0f);
float w0 = 0.0f;
int b0 = 0, b1 = 0, b2 = 0;
// check all possible clusters for this total order
for( int c0 = 0; c0 < count; c0++)
{
Vec3 x1(0.0f);
float w1 = 0.0f;
for( int c1 = 0; c1 < count-c0; c1++)
{
Vec3 x2(0.0f);
float w2 = 0.0f;
for( int c2 = 0; c2 < count-c0-c1; c2++)
{
float w3 = m_wsum - w0 - w1 - w2;
float const alpha2_sum = w0 + w1 * (4.0f/9.0f) + w2 * (1.0f/9.0f);
float const beta2_sum = w3 + w2 * (4.0f/9.0f) + w1 * (1.0f/9.0f);
float const alphabeta_sum = (w1 + w2) * (2.0f/9.0f);
float const factor = 1.0f / (alpha2_sum * beta2_sum - alphabeta_sum * alphabeta_sum);
Vec3 const alphax_sum = x0 + x1 * (2.0f / 3.0f) + x2 * (1.0f / 3.0f);
Vec3 const betax_sum = m_xsum - alphax_sum;
Vec3 a = ( alphax_sum*beta2_sum - betax_sum*alphabeta_sum )*factor;
Vec3 b = ( betax_sum*alpha2_sum - alphax_sum*alphabeta_sum )*factor;
// clamp to the grid
a = Min( one, Max( zero, a ) );
b = Min( one, Max( zero, b ) );
a = Floor( grid*a + half )*gridrcp;
b = Floor( grid*b + half )*gridrcp;
// compute the error
Vec3 e1 = a*a*alpha2_sum + b*b*beta2_sum + 2.0f*( a*b*alphabeta_sum - a*alphax_sum - b*betax_sum );
// apply the metric to the error term
float error = Dot( e1, m_metricSqr );
// keep the solution if it wins
if( error < besterror )
{
besterror = error;
beststart = a;
bestend = b;
b0 = c0;
b1 = c1;
b2 = c2;
}
x2 += m_weighted[c0+c1+c2];
w2 += m_weights[c0+c1+c2];
}
x1 += m_weighted[c0+c1];
w1 += m_weights[c0+c1];
}
x0 += m_weighted[c0];
w0 += m_weights[c0];
}
// save the block if necessary
if( besterror < m_besterror )
{
// compute indices from cluster sizes.
u8 bestindices[16];
{
int i = 0;
for(; i < b0; i++) {
bestindices[i] = 0;
}
for(; i < b0+b1; i++) {
bestindices[i] = 2;
}
for(; i < b0+b1+b2; i++) {
bestindices[i] = 3;
}
for(; i < count; i++) {
bestindices[i] = 1;
}
}
// remap the indices
u8 ordered[16];
for( int i = 0; i < count; ++i )
ordered[m_order[i]] = bestindices[i];
m_colours->RemapIndices( ordered, bestindices );
// save the block
WriteColourBlock4( beststart, bestend, bestindices, block );
// save the error
m_besterror = besterror;
}
}
