Added ETC2 encoding.

3rdparty/etc2/LICENSE.txt

@@ -0,0 +1,24 @@
+Copyright (c) 2013, Bartosz Taudul <[email protected]>
+All rights reserved.
+
+Redistribution and use in source and binary forms, with or without
+modification, are permitted provided that the following conditions are met:
+    * Redistributions of source code must retain the above copyright
+      notice, this list of conditions and the following disclaimer.
+    * Redistributions in binary form must reproduce the above copyright
+      notice, this list of conditions and the following disclaimer in the
+      documentation and/or other materials provided with the distribution.
+    * Neither the name of the <organization> nor the
+      names of its contributors may be used to endorse or promote products
+      derived from this software without specific prior written permission.
+
+THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND
+ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED
+WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
+DISCLAIMED. IN NO EVENT SHALL <COPYRIGHT HOLDER> BE LIABLE FOR ANY
+DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
+(INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
+LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND
+ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

3rdparty/etc2/Math.hpp

@@ -0,0 +1,89 @@
+#ifndef __DARKRL__MATH_HPP__
+#define __DARKRL__MATH_HPP__
+
+#include <algorithm>
+
+#include "Types.hpp"
+
+template<typename T>
+inline T AlignPOT( T val )
+{
+    if( val == 0 ) return 1;
+    val--;
+    for( unsigned int i=1; i<sizeof( T ) * 8; i <<= 1 )
+    {
+        val |= val >> i;
+    }
+    return val + 1;
+}
+
+inline int CountSetBits( uint32 val )
+{
+    val -= ( val >> 1 ) & 0x55555555;
+    val = ( ( val >> 2 ) & 0x33333333 ) + ( val & 0x33333333 );
+    val = ( ( val >> 4 ) + val ) & 0x0f0f0f0f;
+    val += val >> 8;
+    val += val >> 16;
+    return val & 0x0000003f;
+}
+
+inline int CountLeadingZeros( uint32 val )
+{
+    val |= val >> 1;
+    val |= val >> 2;
+    val |= val >> 4;
+    val |= val >> 8;
+    val |= val >> 16;
+    return 32 - CountSetBits( val );
+}
+
+inline float sRGB2linear( float v )
+{
+    const float a = 0.055f;
+    if( v <= 0.04045f )
+    {
+        return v / 12.92f;
+    }
+    else
+    {
+        return pow( ( v + a ) / ( 1 + a ), 2.4f );
+    }
+}
+
+inline float linear2sRGB( float v )
+{
+    const float a = 0.055f;
+    if( v <= 0.0031308f )
+    {
+        return 12.92f * v;
+    }
+    else
+    {
+        return ( 1 + a ) * pow( v, 1/2.4f ) - a;
+    }
+}
+
+template<class T>
+inline T SmoothStep( T x )
+{
+    return x*x*(3-2*x);
+}
+
+inline uint8 clampu8( int32 val )
+{
+    return std::min( std::max( 0, val ), 255 );
+}
+
+template<class T>
+inline T sq( T val )
+{
+    return val * val;
+}
+
+static inline int mul8bit( int a, int b )
+{
+    int t = a*b + 128;
+    return ( t + ( t >> 8 ) ) >> 8;
+}
+
+#endif

3rdparty/etc2/ProcessCommon.hpp

@@ -0,0 +1,51 @@
+#ifndef __PROCESSCOMMON_HPP__
+#define __PROCESSCOMMON_HPP__
+
+#include <assert.h>
+#include <stddef.h>
+
+#include "Types.hpp"
+
+template<class T>
+static size_t GetLeastError( const T* err, size_t num )
+{
+    size_t idx = 0;
+    for( size_t i=1; i<num; i++ )
+    {
+        if( err[i] < err[idx] )
+        {
+            idx = i;
+        }
+    }
+    return idx;
+}
+
+static uint64 FixByteOrder( uint64 d )
+{
+    return ( ( d & 0x00000000FFFFFFFF ) ) |
+           ( ( d & 0xFF00000000000000 ) >> 24 ) |
+           ( ( d & 0x000000FF00000000 ) << 24 ) |
+           ( ( d & 0x00FF000000000000 ) >> 8 ) |
+           ( ( d & 0x0000FF0000000000 ) << 8 );
+}
+
+template<class T, class S>
+static uint64 EncodeSelectors( uint64 d, const T terr[2][8], const S tsel[16][8], const uint32* id )
+{
+    size_t tidx[2];
+    tidx[0] = GetLeastError( terr[0], 8 );
+    tidx[1] = GetLeastError( terr[1], 8 );
+
+    d |= tidx[0] << 26;
+    d |= tidx[1] << 29;
+    for( int i=0; i<16; i++ )
+    {
+        uint64 t = tsel[i][tidx[id[i]%2]];
+        d |= ( t & 0x1 ) << ( i + 32 );
+        d |= ( t & 0x2 ) << ( i + 47 );
+    }
+
+    return d;
+}
+
+#endif

3rdparty/etc2/ProcessRGB.cpp

@@ -0,0 +1,708 @@
+#include <array>
+#include <string.h>
+
+#include "Math.hpp"
+#include "ProcessCommon.hpp"
+#include "ProcessRGB.hpp"
+#include "Tables.hpp"
+#include "Types.hpp"
+#include "Vector.hpp"
+
+#include <bx/endian.h>
+
+#ifdef __SSE4_1__
+#  ifdef _MSC_VER
+#    include <intrin.h>
+#    include <Windows.h>
+#  else
+#    include <x86intrin.h>
+#  endif
+#endif
+
+namespace
+{
+
+typedef std::array<uint16, 4> v4i;
+
+void Average( const uint8* data, v4i* a )
+{
+#ifdef __SSE4_1__
+    __m128i d0 = _mm_loadu_si128(((__m128i*)data) + 0);
+    __m128i d1 = _mm_loadu_si128(((__m128i*)data) + 1);
+    __m128i d2 = _mm_loadu_si128(((__m128i*)data) + 2);
+    __m128i d3 = _mm_loadu_si128(((__m128i*)data) + 3);
+
+    __m128i d0l = _mm_unpacklo_epi8(d0, _mm_setzero_si128());
+    __m128i d0h = _mm_unpackhi_epi8(d0, _mm_setzero_si128());
+    __m128i d1l = _mm_unpacklo_epi8(d1, _mm_setzero_si128());
+    __m128i d1h = _mm_unpackhi_epi8(d1, _mm_setzero_si128());
+    __m128i d2l = _mm_unpacklo_epi8(d2, _mm_setzero_si128());
+    __m128i d2h = _mm_unpackhi_epi8(d2, _mm_setzero_si128());
+    __m128i d3l = _mm_unpacklo_epi8(d3, _mm_setzero_si128());
+    __m128i d3h = _mm_unpackhi_epi8(d3, _mm_setzero_si128());
+
+    __m128i sum0 = _mm_add_epi16(d0l, d1l);
+    __m128i sum1 = _mm_add_epi16(d0h, d1h);
+    __m128i sum2 = _mm_add_epi16(d2l, d3l);
+    __m128i sum3 = _mm_add_epi16(d2h, d3h);
+
+    __m128i sum0l = _mm_unpacklo_epi16(sum0, _mm_setzero_si128());
+    __m128i sum0h = _mm_unpackhi_epi16(sum0, _mm_setzero_si128());
+    __m128i sum1l = _mm_unpacklo_epi16(sum1, _mm_setzero_si128());
+    __m128i sum1h = _mm_unpackhi_epi16(sum1, _mm_setzero_si128());
+    __m128i sum2l = _mm_unpacklo_epi16(sum2, _mm_setzero_si128());
+    __m128i sum2h = _mm_unpackhi_epi16(sum2, _mm_setzero_si128());
+    __m128i sum3l = _mm_unpacklo_epi16(sum3, _mm_setzero_si128());
+    __m128i sum3h = _mm_unpackhi_epi16(sum3, _mm_setzero_si128());
+
+    __m128i b0 = _mm_add_epi32(sum0l, sum0h);
+    __m128i b1 = _mm_add_epi32(sum1l, sum1h);
+    __m128i b2 = _mm_add_epi32(sum2l, sum2h);
+    __m128i b3 = _mm_add_epi32(sum3l, sum3h);
+
+    __m128i a0 = _mm_srli_epi32(_mm_add_epi32(_mm_add_epi32(b2, b3), _mm_set1_epi32(4)), 3);
+    __m128i a1 = _mm_srli_epi32(_mm_add_epi32(_mm_add_epi32(b0, b1), _mm_set1_epi32(4)), 3);
+    __m128i a2 = _mm_srli_epi32(_mm_add_epi32(_mm_add_epi32(b1, b3), _mm_set1_epi32(4)), 3);
+    __m128i a3 = _mm_srli_epi32(_mm_add_epi32(_mm_add_epi32(b0, b2), _mm_set1_epi32(4)), 3);
+
+    _mm_storeu_si128((__m128i*)&a[0], _mm_packus_epi32(_mm_shuffle_epi32(a0, _MM_SHUFFLE(3, 0, 1, 2)), _mm_shuffle_epi32(a1, _MM_SHUFFLE(3, 0, 1, 2))));
+    _mm_storeu_si128((__m128i*)&a[2], _mm_packus_epi32(_mm_shuffle_epi32(a2, _MM_SHUFFLE(3, 0, 1, 2)), _mm_shuffle_epi32(a3, _MM_SHUFFLE(3, 0, 1, 2))));
+#else
+    uint32 r[4];
+    uint32 g[4];
+    uint32 b[4];
+
+    memset(r, 0, sizeof(r));
+    memset(g, 0, sizeof(g));
+    memset(b, 0, sizeof(b));
+
+    for( int j=0; j<4; j++ )
+    {
+        for( int i=0; i<4; i++ )
+        {
+            int index = (j & 2) + (i >> 1);
+            b[index] += *data++;
+            g[index] += *data++;
+            r[index] += *data++;
+            data++;
+        }
+    }
+
+    a[0] = v4i{ uint16( (r[2] + r[3] + 4) / 8 ), uint16( (g[2] + g[3] + 4) / 8 ), uint16( (b[2] + b[3] + 4) / 8 ), 0};
+    a[1] = v4i{ uint16( (r[0] + r[1] + 4) / 8 ), uint16( (g[0] + g[1] + 4) / 8 ), uint16( (b[0] + b[1] + 4) / 8 ), 0};
+    a[2] = v4i{ uint16( (r[1] + r[3] + 4) / 8 ), uint16( (g[1] + g[3] + 4) / 8 ), uint16( (b[1] + b[3] + 4) / 8 ), 0};
+    a[3] = v4i{ uint16( (r[0] + r[2] + 4) / 8 ), uint16( (g[0] + g[2] + 4) / 8 ), uint16( (b[0] + b[2] + 4) / 8 ), 0};
+#endif
+}
+
+void CalcErrorBlock( const uint8* data, uint err[4][4] )
+{
+#ifdef __SSE4_1__
+    __m128i d0 = _mm_loadu_si128(((__m128i*)data) + 0);
+    __m128i d1 = _mm_loadu_si128(((__m128i*)data) + 1);
+    __m128i d2 = _mm_loadu_si128(((__m128i*)data) + 2);
+    __m128i d3 = _mm_loadu_si128(((__m128i*)data) + 3);
+
+    __m128i dm0 = _mm_and_si128(d0, _mm_set1_epi32(0x00FFFFFF));
+    __m128i dm1 = _mm_and_si128(d1, _mm_set1_epi32(0x00FFFFFF));
+    __m128i dm2 = _mm_and_si128(d2, _mm_set1_epi32(0x00FFFFFF));
+    __m128i dm3 = _mm_and_si128(d3, _mm_set1_epi32(0x00FFFFFF));
+
+    __m128i d0l = _mm_unpacklo_epi8(dm0, _mm_setzero_si128());
+    __m128i d0h = _mm_unpackhi_epi8(dm0, _mm_setzero_si128());
+    __m128i d1l = _mm_unpacklo_epi8(dm1, _mm_setzero_si128());
+    __m128i d1h = _mm_unpackhi_epi8(dm1, _mm_setzero_si128());
+    __m128i d2l = _mm_unpacklo_epi8(dm2, _mm_setzero_si128());
+    __m128i d2h = _mm_unpackhi_epi8(dm2, _mm_setzero_si128());
+    __m128i d3l = _mm_unpacklo_epi8(dm3, _mm_setzero_si128());
+    __m128i d3h = _mm_unpackhi_epi8(dm3, _mm_setzero_si128());
+
+    __m128i sum0 = _mm_add_epi16(d0l, d1l);
+    __m128i sum1 = _mm_add_epi16(d0h, d1h);
+    __m128i sum2 = _mm_add_epi16(d2l, d3l);
+    __m128i sum3 = _mm_add_epi16(d2h, d3h);
+
+    __m128i sum0l = _mm_unpacklo_epi16(sum0, _mm_setzero_si128());
+    __m128i sum0h = _mm_unpackhi_epi16(sum0, _mm_setzero_si128());
+    __m128i sum1l = _mm_unpacklo_epi16(sum1, _mm_setzero_si128());
+    __m128i sum1h = _mm_unpackhi_epi16(sum1, _mm_setzero_si128());
+    __m128i sum2l = _mm_unpacklo_epi16(sum2, _mm_setzero_si128());
+    __m128i sum2h = _mm_unpackhi_epi16(sum2, _mm_setzero_si128());
+    __m128i sum3l = _mm_unpacklo_epi16(sum3, _mm_setzero_si128());
+    __m128i sum3h = _mm_unpackhi_epi16(sum3, _mm_setzero_si128());
+
+    __m128i b0 = _mm_add_epi32(sum0l, sum0h);
+    __m128i b1 = _mm_add_epi32(sum1l, sum1h);
+    __m128i b2 = _mm_add_epi32(sum2l, sum2h);
+    __m128i b3 = _mm_add_epi32(sum3l, sum3h);
+
+    __m128i a0 = _mm_add_epi32(b2, b3);
+    __m128i a1 = _mm_add_epi32(b0, b1);
+    __m128i a2 = _mm_add_epi32(b1, b3);
+    __m128i a3 = _mm_add_epi32(b0, b2);
+
+    _mm_storeu_si128((__m128i*)&err[0], a0);
+    _mm_storeu_si128((__m128i*)&err[1], a1);
+    _mm_storeu_si128((__m128i*)&err[2], a2);
+    _mm_storeu_si128((__m128i*)&err[3], a3);
+#else
+    uint terr[4][4];
+
+    memset(terr, 0, 16 * sizeof(uint));
+
+    for( int j=0; j<4; j++ )
+    {
+        for( int i=0; i<4; i++ )
+        {
+            int index = (j & 2) + (i >> 1);
+            uint d = *data++;
+            terr[index][0] += d;
+            d = *data++;
+            terr[index][1] += d;
+            d = *data++;
+            terr[index][2] += d;
+            data++;
+        }
+    }
+
+    for( int i=0; i<3; i++ )
+    {
+        err[0][i] = terr[2][i] + terr[3][i];
+        err[1][i] = terr[0][i] + terr[1][i];
+        err[2][i] = terr[1][i] + terr[3][i];
+        err[3][i] = terr[0][i] + terr[2][i];
+    }
+    for( int i=0; i<4; i++ )
+    {
+        err[i][3] = 0;
+    }
+#endif
+}
+
+uint CalcError( const uint block[4], const v4i& average )
+{
+    uint err = 0x3FFFFFFF; // Big value to prevent negative values, but small enough to prevent overflow
+    err -= block[0] * 2 * average[2];
+    err -= block[1] * 2 * average[1];
+    err -= block[2] * 2 * average[0];
+    err += 8 * ( sq( average[0] ) + sq( average[1] ) + sq( average[2] ) );
+    return err;
+}
+
+void ProcessAverages( v4i* a )
+{
+#ifdef __SSE4_1__
+    for( int i=0; i<2; i++ )
+    {
+        __m128i d = _mm_loadu_si128((__m128i*)a[i*2].data());
+
+        __m128i t = _mm_add_epi16(_mm_mullo_epi16(d, _mm_set1_epi16(31)), _mm_set1_epi16(128));
+
+        __m128i c = _mm_srli_epi16(_mm_add_epi16(t, _mm_srli_epi16(t, 8)), 8);
+
+        __m128i c1 = _mm_shuffle_epi32(c, _MM_SHUFFLE(3, 2, 3, 2));
+        __m128i diff = _mm_sub_epi16(c, c1);
+        diff = _mm_max_epi16(diff, _mm_set1_epi16(-4));
+        diff = _mm_min_epi16(diff, _mm_set1_epi16(3));
+
+        __m128i co = _mm_add_epi16(c1, diff);
+
+        c = _mm_blend_epi16(co, c, 0xF0);
+
+        __m128i a0 = _mm_or_si128(_mm_slli_epi16(c, 3), _mm_srli_epi16(c, 2));
+
+        _mm_storeu_si128((__m128i*)a[4+i*2].data(), a0);
+    }
+
+    for( int i=0; i<2; i++ )
+    {
+        __m128i d = _mm_loadu_si128((__m128i*)a[i*2].data());
+
+        __m128i t0 = _mm_add_epi16(_mm_mullo_epi16(d, _mm_set1_epi16(15)), _mm_set1_epi16(128));
+        __m128i t1 = _mm_srli_epi16(_mm_add_epi16(t0, _mm_srli_epi16(t0, 8)), 8);
+
+        __m128i t2 = _mm_or_si128(t1, _mm_slli_epi16(t1, 4));
+
+        _mm_storeu_si128((__m128i*)a[i*2].data(), t2);
+    }
+#else
+    for( int i=0; i<2; i++ )
+    {
+        for( int j=0; j<3; j++ )
+        {
+            int32 c1 = mul8bit( a[i*2+1][j], 31 );
+            int32 c2 = mul8bit( a[i*2][j], 31 );
+
+            int32 diff = c2 - c1;
+            if( diff > 3 ) diff = 3;
+            else if( diff < -4 ) diff = -4;
+
+            int32 co = c1 + diff;
+
+            a[5+i*2][j] = ( c1 << 3 ) | ( c1 >> 2 );
+            a[4+i*2][j] = ( co << 3 ) | ( co >> 2 );
+        }
+    }
+
+    for( int i=0; i<4; i++ )
+    {
+        a[i][0] = g_avg2[mul8bit( a[i][0], 15 )];
+        a[i][1] = g_avg2[mul8bit( a[i][1], 15 )];
+        a[i][2] = g_avg2[mul8bit( a[i][2], 15 )];
+    }
+#endif
+}
+
+void EncodeAverages( uint64& _d, const v4i* a, size_t idx )
+{
+    uint64 d = _d;
+    d |= ( idx << 24 );
+    size_t base = idx << 1;
+
+    if( ( idx & 0x2 ) == 0 )
+    {
+        for( int i=0; i<3; i++ )
+        {
+            d |= uint64( a[base+0][i] >> 4 ) << ( i*8 );
+            d |= uint64( a[base+1][i] >> 4 ) << ( i*8 + 4 );
+        }
+    }
+    else
+    {
+        for( int i=0; i<3; i++ )
+        {
+            d |= uint64( a[base+1][i] & 0xF8 ) << ( i*8 );
+            int32 c = ( ( a[base+0][i] & 0xF8 ) - ( a[base+1][i] & 0xF8 ) ) >> 3;
+            c &= ~0xFFFFFFF8;
+            d |= ((uint64)c) << ( i*8 );
+        }
+    }
+    _d = d;
+}
+
+uint64 CheckSolid( const uint8* src )
+{
+#ifdef __SSE4_1__
+    __m128i d0 = _mm_loadu_si128(((__m128i*)src) + 0);
+    __m128i d1 = _mm_loadu_si128(((__m128i*)src) + 1);
+    __m128i d2 = _mm_loadu_si128(((__m128i*)src) + 2);
+    __m128i d3 = _mm_loadu_si128(((__m128i*)src) + 3);
+
+    __m128i c = _mm_shuffle_epi32(d0, _MM_SHUFFLE(0, 0, 0, 0));
+
+    __m128i c0 = _mm_cmpeq_epi8(d0, c);
+    __m128i c1 = _mm_cmpeq_epi8(d1, c);
+    __m128i c2 = _mm_cmpeq_epi8(d2, c);
+    __m128i c3 = _mm_cmpeq_epi8(d3, c);
+
+    __m128i m0 = _mm_and_si128(c0, c1);
+    __m128i m1 = _mm_and_si128(c2, c3);
+    __m128i m = _mm_and_si128(m0, m1);
+
+    if (!_mm_testc_si128(m, _mm_set1_epi32(-1)))
+    {
+        return 0;
+    }
+#else
+    const uint8* ptr = src + 4;
+    for( int i=1; i<16; i++ )
+    {
+        if( memcmp( src, ptr, 4 ) != 0 )
+        {
+            return 0;
+        }
+        ptr += 4;
+    }
+#endif
+    return 0x02000000 |
+        ( uint( src[0] & 0xF8 ) << 16 ) |
+        ( uint( src[1] & 0xF8 ) << 8 ) |
+        ( uint( src[2] & 0xF8 ) );
+}
+
+void PrepareAverages( v4i a[8], const uint8* src, uint err[4] )
+{
+    Average( src, a );
+    ProcessAverages( a );
+
+    uint errblock[4][4];
+    CalcErrorBlock( src, errblock );
+
+    for( int i=0; i<4; i++ )
+    {
+        err[i/2] += CalcError( errblock[i], a[i] );
+        err[2+i/2] += CalcError( errblock[i], a[i+4] );
+    }
+}
+
+void FindBestFit( uint64 terr[2][8], uint16 tsel[16][8], v4i a[8], const uint32* id, const uint8* data )
+{
+    for( size_t i=0; i<16; i++ )
+    {
+        uint16* sel = tsel[i];
+        uint bid = id[i];
+        uint64* ter = terr[bid%2];
+
+        uint8 b = *data++;
+        uint8 g = *data++;
+        uint8 r = *data++;
+        data++;
+
+        int dr = a[bid][0] - r;
+        int dg = a[bid][1] - g;
+        int db = a[bid][2] - b;
+
+#ifdef __SSE4_1__
+        // Reference implementation
+
+        __m128i pix = _mm_set1_epi32(dr * 77 + dg * 151 + db * 28);
+        // Taking the absolute value is way faster. The values are only used to sort, so the result will be the same.
+        __m128i error0 = _mm_abs_epi32(_mm_add_epi32(pix, g_table256_SIMD[0]));
+        __m128i error1 = _mm_abs_epi32(_mm_add_epi32(pix, g_table256_SIMD[1]));
+        __m128i error2 = _mm_abs_epi32(_mm_sub_epi32(pix, g_table256_SIMD[0]));
+        __m128i error3 = _mm_abs_epi32(_mm_sub_epi32(pix, g_table256_SIMD[1]));
+
+        __m128i index0 = _mm_and_si128(_mm_cmplt_epi32(error1, error0), _mm_set1_epi32(1));
+        __m128i minError0 = _mm_min_epi32(error0, error1);
+
+        __m128i index1 = _mm_sub_epi32(_mm_set1_epi32(2), _mm_cmplt_epi32(error3, error2));
+        __m128i minError1 = _mm_min_epi32(error2, error3);
+
+        __m128i minIndex0 = _mm_blendv_epi8(index0, index1, _mm_cmplt_epi32(minError1, minError0));
+        __m128i minError = _mm_min_epi32(minError0, minError1);
+
+        // Squaring the minimum error to produce correct values when adding
+        __m128i minErrorLow = _mm_shuffle_epi32(minError, _MM_SHUFFLE(1, 1, 0, 0));
+        __m128i squareErrorLow = _mm_mul_epi32(minErrorLow, minErrorLow);
+        squareErrorLow = _mm_add_epi64(squareErrorLow, _mm_loadu_si128(((__m128i*)ter) + 0));
+        _mm_storeu_si128(((__m128i*)ter) + 0, squareErrorLow);
+        __m128i minErrorHigh = _mm_shuffle_epi32(minError, _MM_SHUFFLE(3, 3, 2, 2));
+        __m128i squareErrorHigh = _mm_mul_epi32(minErrorHigh, minErrorHigh);
+        squareErrorHigh = _mm_add_epi64(squareErrorHigh, _mm_loadu_si128(((__m128i*)ter) + 1));
+        _mm_storeu_si128(((__m128i*)ter) + 1, squareErrorHigh);
+
+        // Taking the absolute value is way faster. The values are only used to sort, so the result will be the same.
+        error0 = _mm_abs_epi32(_mm_add_epi32(pix, g_table256_SIMD[2]));
+        error1 = _mm_abs_epi32(_mm_add_epi32(pix, g_table256_SIMD[3]));
+        error2 = _mm_abs_epi32(_mm_sub_epi32(pix, g_table256_SIMD[2]));
+        error3 = _mm_abs_epi32(_mm_sub_epi32(pix, g_table256_SIMD[3]));
+
+        index0 = _mm_and_si128(_mm_cmplt_epi32(error1, error0), _mm_set1_epi32(1));
+        minError0 = _mm_min_epi32(error0, error1);
+
+        index1 = _mm_sub_epi32(_mm_set1_epi32(2), _mm_cmplt_epi32(error3, error2));
+        minError1 = _mm_min_epi32(error2, error3);
+
+        __m128i minIndex1 = _mm_blendv_epi8(index0, index1, _mm_cmplt_epi32(minError1, minError0));
+        minError = _mm_min_epi32(minError0, minError1);
+
+        // Squaring the minimum error to produce correct values when adding
+        minErrorLow = _mm_shuffle_epi32(minError, _MM_SHUFFLE(1, 1, 0, 0));
+        squareErrorLow = _mm_mul_epi32(minErrorLow, minErrorLow);
+        squareErrorLow = _mm_add_epi64(squareErrorLow, _mm_loadu_si128(((__m128i*)ter) + 2));
+        _mm_storeu_si128(((__m128i*)ter) + 2, squareErrorLow);
+        minErrorHigh = _mm_shuffle_epi32(minError, _MM_SHUFFLE(3, 3, 2, 2));
+        squareErrorHigh = _mm_mul_epi32(minErrorHigh, minErrorHigh);
+        squareErrorHigh = _mm_add_epi64(squareErrorHigh, _mm_loadu_si128(((__m128i*)ter) + 3));
+        _mm_storeu_si128(((__m128i*)ter) + 3, squareErrorHigh);
+        __m128i minIndex = _mm_packs_epi32(minIndex0, minIndex1);
+        _mm_storeu_si128((__m128i*)sel, minIndex);
+#else
+        int pix = dr * 77 + dg * 151 + db * 28;
+
+        for( int t=0; t<8; t++ )
+        {
+            const int64* tab = g_table256[t];
+            uint idx = 0;
+            uint64 err = sq( tab[0] + pix );
+            for( int j=1; j<4; j++ )
+            {
+                uint64 local = sq( tab[j] + pix );
+                if( local < err )
+                {
+                    err = local;
+                    idx = j;
+                }
+            }
+            *sel++ = idx;
+            *ter++ += err;
+        }
+#endif
+    }
+}
+
+#ifdef __SSE4_1__
+// Non-reference implementation, but faster. Produces same results as the AVX2 version
+void FindBestFit( uint32 terr[2][8], uint16 tsel[16][8], v4i a[8], const uint32* id, const uint8* data )
+{
+    for( size_t i=0; i<16; i++ )
+    {
+        uint16* sel = tsel[i];
+        uint bid = id[i];
+        uint32* ter = terr[bid%2];
+
+        uint8 b = *data++;
+        uint8 g = *data++;
+        uint8 r = *data++;
+        data++;
+
+        int dr = a[bid][0] - r;
+        int dg = a[bid][1] - g;
+        int db = a[bid][2] - b;
+
+        // The scaling values are divided by two and rounded, to allow the differences to be in the range of signed int16
+        // This produces slightly different results, but is significant faster
+        __m128i pixel = _mm_set1_epi16(dr * 38 + dg * 76 + db * 14);
+        __m128i pix = _mm_abs_epi16(pixel);
+
+        // Taking the absolute value is way faster. The values are only used to sort, so the result will be the same.
+        // Since the selector table is symmetrical, we need to calculate the difference only for half of the entries.
+        __m128i error0 = _mm_abs_epi16(_mm_sub_epi16(pix, g_table128_SIMD[0]));
+        __m128i error1 = _mm_abs_epi16(_mm_sub_epi16(pix, g_table128_SIMD[1]));
+
+        __m128i index = _mm_and_si128(_mm_cmplt_epi16(error1, error0), _mm_set1_epi16(1));
+        __m128i minError = _mm_min_epi16(error0, error1);
+
+        // Exploiting symmetry of the selector table and use the sign bit
+        // This produces slightly different results, but is needed to produce same results as AVX2 implementation
+        __m128i indexBit = _mm_andnot_si128(_mm_srli_epi16(pixel, 15), _mm_set1_epi8(-1));
+        __m128i minIndex = _mm_or_si128(index, _mm_add_epi16(indexBit, indexBit));
+
+        // Squaring the minimum error to produce correct values when adding
+        __m128i squareErrorLo = _mm_mullo_epi16(minError, minError);
+        __m128i squareErrorHi = _mm_mulhi_epi16(minError, minError);
+
+        __m128i squareErrorLow = _mm_unpacklo_epi16(squareErrorLo, squareErrorHi);
+        __m128i squareErrorHigh = _mm_unpackhi_epi16(squareErrorLo, squareErrorHi);
+
+        squareErrorLow = _mm_add_epi32(squareErrorLow, _mm_loadu_si128(((__m128i*)ter) + 0));
+        _mm_storeu_si128(((__m128i*)ter) + 0, squareErrorLow);
+        squareErrorHigh = _mm_add_epi32(squareErrorHigh, _mm_loadu_si128(((__m128i*)ter) + 1));
+        _mm_storeu_si128(((__m128i*)ter) + 1, squareErrorHigh);
+
+        _mm_storeu_si128((__m128i*)sel, minIndex);
+    }
+}
+#endif
+
+uint8_t convert6(float f)
+{
+    int i = (std::min(std::max(static_cast<int>(f), 0), 1023) - 15) >> 1;
+    return (i + 11 - ((i + 11) >> 7) - ((i + 4) >> 7)) >> 3;
+}
+
+uint8_t convert7(float f)
+{
+    int i = (std::min(std::max(static_cast<int>(f), 0), 1023) - 15) >> 1;
+    return (i + 9 - ((i + 9) >> 8) - ((i + 6) >> 8)) >> 2;
+}
+
+std::pair<uint64, uint64> Planar(const uint8* src)
+{
+    int32 r = 0;
+    int32 g = 0;
+    int32 b = 0;
+
+    for (int i = 0; i < 16; ++i)
+    {
+        b += src[i * 4 + 0];
+        g += src[i * 4 + 1];
+        r += src[i * 4 + 2];
+    }
+
+    int32 difRyz = 0;
+    int32 difGyz = 0;
+    int32 difByz = 0;
+    int32 difRxz = 0;
+    int32 difGxz = 0;
+    int32 difBxz = 0;
+
+    const int32 scaling[] = { -255, -85, 85, 255 };
+
+    for (int i = 0; i < 16; ++i)
+    {
+        int32 difB = (static_cast<int>(src[i * 4 + 0]) << 4) - b;
+        int32 difG = (static_cast<int>(src[i * 4 + 1]) << 4) - g;
+        int32 difR = (static_cast<int>(src[i * 4 + 2]) << 4) - r;
+
+        difRyz += difR * scaling[i % 4];
+        difGyz += difG * scaling[i % 4];
+        difByz += difB * scaling[i % 4];
+
+        difRxz += difR * scaling[i / 4];
+        difGxz += difG * scaling[i / 4];
+        difBxz += difB * scaling[i / 4];
+    }
+
+    const float scale = -4.0f / ((255 * 255 * 8.0f + 85 * 85 * 8.0f) * 16.0f);
+
+    float aR = difRxz * scale;
+    float aG = difGxz * scale;
+    float aB = difBxz * scale;
+
+    float bR = difRyz * scale;
+    float bG = difGyz * scale;
+    float bB = difByz * scale;
+
+    float dR = r * (4.0f / 16.0f);
+    float dG = g * (4.0f / 16.0f);
+    float dB = b * (4.0f / 16.0f);
+
+    // calculating the three colors RGBO, RGBH, and RGBV.  RGB = df - af * x - bf * y;
+    float cofR = std::fma(aR,  255.0f, std::fma(bR,  255.0f, dR));
+    float cofG = std::fma(aG,  255.0f, std::fma(bG,  255.0f, dG));
+    float cofB = std::fma(aB,  255.0f, std::fma(bB,  255.0f, dB));
+    float chfR = std::fma(aR, -425.0f, std::fma(bR,  255.0f, dR));
+    float chfG = std::fma(aG, -425.0f, std::fma(bG,  255.0f, dG));
+    float chfB = std::fma(aB, -425.0f, std::fma(bB,  255.0f, dB));
+    float cvfR = std::fma(aR,  255.0f, std::fma(bR, -425.0f, dR));
+    float cvfG = std::fma(aG,  255.0f, std::fma(bG, -425.0f, dG));
+    float cvfB = std::fma(aB,  255.0f, std::fma(bB, -425.0f, dB));
+
+    // convert to r6g7b6
+    int32 coR = convert6(cofR);
+    int32 coG = convert7(cofG);
+    int32 coB = convert6(cofB);
+    int32 chR = convert6(chfR);
+    int32 chG = convert7(chfG);
+    int32 chB = convert6(chfB);
+    int32 cvR = convert6(cvfR);
+    int32 cvG = convert7(cvfG);
+    int32 cvB = convert6(cvfB);
+
+    // Error calculation
+    int32 ro0 = coR;
+    int32 go0 = coG;
+    int32 bo0 = coB;
+    int32 ro1 = (ro0 >> 4) | (ro0 << 2);
+    int32 go1 = (go0 >> 6) | (go0 << 1);
+    int32 bo1 = (bo0 >> 4) | (bo0 << 2);
+    int32 ro2 = (ro1 << 2) + 2;
+    int32 go2 = (go1 << 2) + 2;
+    int32 bo2 = (bo1 << 2) + 2;
+
+    int32 rh0 = chR;
+    int32 gh0 = chG;
+    int32 bh0 = chB;
+    int32 rh1 = (rh0 >> 4) | (rh0 << 2);
+    int32 gh1 = (gh0 >> 6) | (gh0 << 1);
+    int32 bh1 = (bh0 >> 4) | (bh0 << 2);
+
+    int32 rh2 = rh1 - ro1;
+    int32 gh2 = gh1 - go1;
+    int32 bh2 = bh1 - bo1;
+
+    int32 rv0 = cvR;
+    int32 gv0 = cvG;
+    int32 bv0 = cvB;
+    int32 rv1 = (rv0 >> 4) | (rv0 << 2);
+    int32 gv1 = (gv0 >> 6) | (gv0 << 1);
+    int32 bv1 = (bv0 >> 4) | (bv0 << 2);
+
+    int32 rv2 = rv1 - ro1;
+    int32 gv2 = gv1 - go1;
+    int32 bv2 = bv1 - bo1;
+
+    uint64 error = 0;
+
+    for (int i = 0; i < 16; ++i)
+    {
+        int32 cR = clampu8((rh2 * (i / 4) + rv2 * (i % 4) + ro2) >> 2);
+        int32 cG = clampu8((gh2 * (i / 4) + gv2 * (i % 4) + go2) >> 2);
+        int32 cB = clampu8((bh2 * (i / 4) + bv2 * (i % 4) + bo2) >> 2);
+
+        int32 difB = static_cast<int>(src[i * 4 + 0]) - cB;
+        int32 difG = static_cast<int>(src[i * 4 + 1]) - cG;
+        int32 difR = static_cast<int>(src[i * 4 + 2]) - cR;
+
+        int32 dif = difR * 38 + difG * 76 + difB * 14;
+
+        error += dif * dif;
+    }
+
+    /**/
+    uint32 rgbv = cvB | (cvG << 6) | (cvR << 13);
+    uint32 rgbh = chB | (chG << 6) | (chR << 13);
+    uint32 hi = rgbv | ((rgbh & 0x1FFF) << 19);
+    uint32 lo = (chR & 0x1) | 0x2 | ((chR << 1) & 0x7C);
+    lo |= ((coB & 0x07) <<  7) | ((coB & 0x18) <<  8) | ((coB & 0x20) << 11);
+    lo |= ((coG & 0x3F) << 17) | ((coG & 0x40) << 18);
+    lo |= coR << 25;
+
+    const int32 idx = (coR & 0x20) | ((coG & 0x20) >> 1) | ((coB & 0x1E) >> 1);
+
+    lo |= g_flags[idx];
+
+    uint64 result = static_cast<uint32>(bx::endianSwap(lo));
+    result |= static_cast<uint64>(static_cast<uint32>(bx::endianSwap(hi))) << 32;
+
+    return std::make_pair(result, error);
+}
+
+template<class T, class S>
+uint64 EncodeSelectors( uint64 d, const T terr[2][8], const S tsel[16][8], const uint32* id, const uint64 value, const uint64 error)
+{
+    size_t tidx[2];
+    tidx[0] = GetLeastError( terr[0], 8 );
+    tidx[1] = GetLeastError( terr[1], 8 );
+
+    if ((terr[0][tidx[0]] + terr[1][tidx[1]]) >= error)
+    {
+        return value;
+    }
+
+    d |= tidx[0] << 26;
+    d |= tidx[1] << 29;
+    for( int i=0; i<16; i++ )
+    {
+        uint64 t = tsel[i][tidx[id[i]%2]];
+        d |= ( t & 0x1 ) << ( i + 32 );
+        d |= ( t & 0x2 ) << ( i + 47 );
+    }
+
+    return FixByteOrder(d);
+}
+}
+
+uint64 ProcessRGB( const uint8* src )
+{
+    uint64 d = CheckSolid( src );
+    if( d != 0 ) return d;
+
+    v4i a[8];
+    uint err[4] = {};
+    PrepareAverages( a, src, err );
+    size_t idx = GetLeastError( err, 4 );
+    EncodeAverages( d, a, idx );
+
+#if defined __SSE4_1__ && !defined REFERENCE_IMPLEMENTATION
+    uint32 terr[2][8] = {};
+#else
+    uint64 terr[2][8] = {};
+#endif
+    uint16 tsel[16][8];
+    const uint32* id = g_id[idx];
+    FindBestFit( terr, tsel, a, id, src );
+
+    return FixByteOrder( EncodeSelectors( d, terr, tsel, id ) );
+}
+
+uint64 ProcessRGB_ETC2( const uint8* src )
+{
+    std::pair<uint64, uint64> result = Planar( src );
+
+    uint64 d = 0;
+
+    v4i a[8];
+    uint err[4] = {};
+    PrepareAverages( a, src, err );
+    size_t idx = GetLeastError( err, 4 );
+    EncodeAverages( d, a, idx );
+
+    uint64 terr[2][8] = {};
+    uint16 tsel[16][8];
+    const uint32* id = g_id[idx];
+    FindBestFit( terr, tsel, a, id, src );
+
+    return EncodeSelectors( d, terr, tsel, id, result.first, result.second );
+}

3rdparty/etc2/ProcessRGB.hpp

@@ -0,0 +1,9 @@
+#ifndef __PROCESSRGB_HPP__
+#define __PROCESSRGB_HPP__
+
+#include "Types.hpp"
+
+uint64 ProcessRGB( const uint8* src );
+uint64 ProcessRGB_ETC2( const uint8* src );
+
+#endif

3rdparty/etc2/Tables.cpp

@@ -0,0 +1,109 @@
+#include "Tables.hpp"
+
+const int32 g_table[8][4] = {
+    {  2,  8,   -2,   -8 },
+    {  5, 17,   -5,  -17 },
+    {  9, 29,   -9,  -29 },
+    { 13, 42,  -13,  -42 },
+    { 18, 60,  -18,  -60 },
+    { 24, 80,  -24,  -80 },
+    { 33, 106, -33, -106 },
+    { 47, 183, -47, -183 }
+};
+
+const int64 g_table256[8][4] = {
+    {  2*256,  8*256,   -2*256,   -8*256 },
+    {  5*256, 17*256,   -5*256,  -17*256 },
+    {  9*256, 29*256,   -9*256,  -29*256 },
+    { 13*256, 42*256,  -13*256,  -42*256 },
+    { 18*256, 60*256,  -18*256,  -60*256 },
+    { 24*256, 80*256,  -24*256,  -80*256 },
+    { 33*256, 106*256, -33*256, -106*256 },
+    { 47*256, 183*256, -47*256, -183*256 }
+};
+
+const uint32 g_id[4][16] = {
+    { 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0 },
+    { 3, 3, 2, 2, 3, 3, 2, 2, 3, 3, 2, 2, 3, 3, 2, 2 },
+    { 5, 5, 5, 5, 5, 5, 5, 5, 4, 4, 4, 4, 4, 4, 4, 4 },
+    { 7, 7, 6, 6, 7, 7, 6, 6, 7, 7, 6, 6, 7, 7, 6, 6 }
+};
+
+const uint32 g_avg2[16] = {
+    0x00,
+    0x11,
+    0x22,
+    0x33,
+    0x44,
+    0x55,
+    0x66,
+    0x77,
+    0x88,
+    0x99,
+    0xAA,
+    0xBB,
+    0xCC,
+    0xDD,
+    0xEE,
+    0xFF
+};
+
+const uint32 g_flags[64] = {
+    0x80800402, 0x80800402, 0x80800402, 0x80800402,
+    0x80800402, 0x80800402, 0x80800402, 0x8080E002,
+    0x80800402, 0x80800402, 0x8080E002, 0x8080E002,
+    0x80800402, 0x8080E002, 0x8080E002, 0x8080E002,
+    0x80000402, 0x80000402, 0x80000402, 0x80000402,
+    0x80000402, 0x80000402, 0x80000402, 0x8000E002,
+    0x80000402, 0x80000402, 0x8000E002, 0x8000E002,
+    0x80000402, 0x8000E002, 0x8000E002, 0x8000E002,
+    0x00800402, 0x00800402, 0x00800402, 0x00800402,
+    0x00800402, 0x00800402, 0x00800402, 0x0080E002,
+    0x00800402, 0x00800402, 0x0080E002, 0x0080E002,
+    0x00800402, 0x0080E002, 0x0080E002, 0x0080E002,
+    0x00000402, 0x00000402, 0x00000402, 0x00000402,
+    0x00000402, 0x00000402, 0x00000402, 0x0000E002,
+    0x00000402, 0x00000402, 0x0000E002, 0x0000E002,
+    0x00000402, 0x0000E002, 0x0000E002, 0x0000E002
+};
+
+#ifdef __SSE4_1__
+const uint8 g_flags_AVX2[64] =
+{
+    0x63, 0x63, 0x63, 0x63,
+    0x63, 0x63, 0x63, 0x7D,
+    0x63, 0x63, 0x7D, 0x7D,
+    0x63, 0x7D, 0x7D, 0x7D,
+    0x43, 0x43, 0x43, 0x43,
+    0x43, 0x43, 0x43, 0x5D,
+    0x43, 0x43, 0x5D, 0x5D,
+    0x43, 0x5D, 0x5D, 0x5D,
+    0x23, 0x23, 0x23, 0x23,
+    0x23, 0x23, 0x23, 0x3D,
+    0x23, 0x23, 0x3D, 0x3D,
+    0x23, 0x3D, 0x3D, 0x3D,
+    0x03, 0x03, 0x03, 0x03,
+    0x03, 0x03, 0x03, 0x1D,
+    0x03, 0x03, 0x1D, 0x1D,
+    0x03, 0x1D, 0x1D, 0x1D,
+};
+
+const __m128i g_table_SIMD[2] =
+{
+    _mm_setr_epi16(   2,   5,   9,  13,  18,  24,  33,  47),
+    _mm_setr_epi16(   8,  17,  29,  42,  60,  80, 106, 183)
+};
+const __m128i g_table128_SIMD[2] =
+{
+    _mm_setr_epi16(   2*128,   5*128,   9*128,  13*128,  18*128,  24*128,  33*128,  47*128),
+    _mm_setr_epi16(   8*128,  17*128,  29*128,  42*128,  60*128,  80*128, 106*128, 183*128)
+};
+const __m128i g_table256_SIMD[4] =
+{
+    _mm_setr_epi32(  2*256,   5*256,   9*256,  13*256),
+    _mm_setr_epi32(  8*256,  17*256,  29*256,  42*256),
+    _mm_setr_epi32( 18*256,  24*256,  33*256,  47*256),
+    _mm_setr_epi32( 60*256,  80*256, 106*256, 183*256)
+};
+#endif
+

3rdparty/etc2/Tables.hpp

@@ -0,0 +1,25 @@
+#ifndef __TABLES_HPP__
+#define __TABLES_HPP__
+
+#include "Types.hpp"
+#ifdef __SSE4_1__
+#include <smmintrin.h>
+#endif
+
+extern const int32 g_table[8][4];
+extern const int64 g_table256[8][4];
+
+extern const uint32 g_id[4][16];
+
+extern const uint32 g_avg2[16];
+
+extern const uint32 g_flags[64];
+
+#ifdef __SSE4_1__
+extern const uint8 g_flags_AVX2[64];
+extern const __m128i g_table_SIMD[2];
+extern const __m128i g_table128_SIMD[2];
+extern const __m128i g_table256_SIMD[4];
+#endif
+
+#endif

3rdparty/etc2/Types.hpp

@@ -0,0 +1,17 @@
+#ifndef __DARKRL__TYPES_HPP__
+#define __DARKRL__TYPES_HPP__
+
+#include <cstdint>
+
+typedef int8_t      int8;
+typedef uint8_t     uint8;
+typedef int16_t     int16;
+typedef uint16_t    uint16;
+typedef int32_t     int32;
+typedef uint32_t    uint32;
+typedef int64_t     int64;
+typedef uint64_t    uint64;
+
+typedef unsigned int uint;
+
+#endif

3rdparty/etc2/Vector.hpp

@@ -0,0 +1,222 @@
+#ifndef __DARKRL__VECTOR_HPP__
+#define __DARKRL__VECTOR_HPP__
+
+#include <assert.h>
+#include <algorithm>
+#include <math.h>
+
+#include "Math.hpp"
+#include "Types.hpp"
+
+template<class T>
+struct Vector2
+{
+    Vector2() : x( 0 ), y( 0 ) {}
+    Vector2( T v ) : x( v ), y( v ) {}
+    Vector2( T _x, T _y ) : x( _x ), y( _y ) {}
+
+    bool operator==( const Vector2<T>& rhs ) const { return x == rhs.x && y == rhs.y; }
+    bool operator!=( const Vector2<T>& rhs ) const { return !( *this == rhs ); }
+
+    Vector2<T>& operator+=( const Vector2<T>& rhs )
+    {
+        x += rhs.x;
+        y += rhs.y;
+        return *this;
+    }
+    Vector2<T>& operator-=( const Vector2<T>& rhs )
+    {
+        x -= rhs.x;
+        y -= rhs.y;
+        return *this;
+    }
+    Vector2<T>& operator*=( const Vector2<T>& rhs )
+    {
+        x *= rhs.x;
+        y *= rhs.y;
+        return *this;
+    }
+
+    T x, y;
+};
+
+template<class T>
+Vector2<T> operator+( const Vector2<T>& lhs, const Vector2<T>& rhs )
+{
+    return Vector2<T>( lhs.x + rhs.x, lhs.y + rhs.y );
+}
+
+template<class T>
+Vector2<T> operator-( const Vector2<T>& lhs, const Vector2<T>& rhs )
+{
+    return Vector2<T>( lhs.x - rhs.x, lhs.y - rhs.y );
+}
+
+template<class T>
+Vector2<T> operator*( const Vector2<T>& lhs, const float& rhs )
+{
+    return Vector2<T>( lhs.x * rhs, lhs.y * rhs );
+}
+
+template<class T>
+Vector2<T> operator/( const Vector2<T>& lhs, const T& rhs )
+{
+    return Vector2<T>( lhs.x / rhs, lhs.y / rhs );
+}
+
+
+typedef Vector2<int32> v2i;
+typedef Vector2<float> v2f;
+
+
+template<class T>
+struct Vector3
+{
+    Vector3() : x( 0 ), y( 0 ), z( 0 ) {}
+    Vector3( T v ) : x( v ), y( v ), z( v ) {}
+    Vector3( T _x, T _y, T _z ) : x( _x ), y( _y ), z( _z ) {}
+    template<class Y>
+    Vector3( const Vector3<Y>& v ) : x( T( v.x ) ), y( T( v.y ) ), z( T( v.z ) ) {}
+
+    T Luminance() const { return T( x * 0.3f + y * 0.59f + z * 0.11f ); }
+    void Clamp()
+    {
+        x = std::min( T(1), std::max( T(0), x ) );
+        y = std::min( T(1), std::max( T(0), y ) );
+        z = std::min( T(1), std::max( T(0), z ) );
+    }
+
+    bool operator==( const Vector3<T>& rhs ) const { return x == rhs.x && y == rhs.y && z == rhs.z; }
+    bool operator!=( const Vector2<T>& rhs ) const { return !( *this == rhs ); }
+
+    T& operator[]( uint idx ) { assert( idx < 3 ); return ((T*)this)[idx]; }
+    const T& operator[]( uint idx ) const { assert( idx < 3 ); return ((T*)this)[idx]; }
+
+    Vector3<T> operator+=( const Vector3<T>& rhs )
+    {
+        x += rhs.x;
+        y += rhs.y;
+        z += rhs.z;
+        return *this;
+    }
+
+    Vector3<T> operator*=( const Vector3<T>& rhs )
+    {
+        x *= rhs.x;
+        y *= rhs.y;
+        z *= rhs.z;
+        return *this;
+    }
+
+    Vector3<T> operator*=( const float& rhs )
+    {
+        x *= rhs;
+        y *= rhs;
+        z *= rhs;
+        return *this;
+    }
+
+    T x, y, z;
+    T padding;
+};
+
+template<class T>
+Vector3<T> operator+( const Vector3<T>& lhs, const Vector3<T>& rhs )
+{
+    return Vector3<T>( lhs.x + rhs.x, lhs.y + rhs.y, lhs.z + rhs.z );
+}
+
+template<class T>
+Vector3<T> operator-( const Vector3<T>& lhs, const Vector3<T>& rhs )
+{
+    return Vector3<T>( lhs.x - rhs.x, lhs.y - rhs.y, lhs.z - rhs.z );
+}
+
+template<class T>
+Vector3<T> operator*( const Vector3<T>& lhs, const Vector3<T>& rhs )
+{
+    return Vector3<T>( lhs.x * rhs.x, lhs.y * rhs.y, lhs.z * rhs.z );
+}
+
+template<class T>
+Vector3<T> operator*( const Vector3<T>& lhs, const float& rhs )
+{
+    return Vector3<T>( T( lhs.x * rhs ), T( lhs.y * rhs ), T( lhs.z * rhs ) );
+}
+
+template<class T>
+Vector3<T> operator/( const Vector3<T>& lhs, const T& rhs )
+{
+    return Vector3<T>( lhs.x / rhs, lhs.y / rhs, lhs.z / rhs );
+}
+
+template<class T>
+bool operator<( const Vector3<T>& lhs, const Vector3<T>& rhs )
+{
+    return lhs.Luminance() < rhs.Luminance();
+}
+
+typedef Vector3<int32> v3i;
+typedef Vector3<float> v3f;
+typedef Vector3<uint8> v3b;
+
+
+static inline v3b v3f_to_v3b( const v3f& v )
+{
+    return v3b( uint8( std::min( 1.f, v.x ) * 255 ), uint8( std::min( 1.f, v.y ) * 255 ), uint8( std::min( 1.f, v.z ) * 255 ) );
+}
+
+template<class T>
+Vector3<T> Mix( const Vector3<T>& v1, const Vector3<T>& v2, float amount )
+{
+    return v1 + ( v2 - v1 ) * amount;
+}
+
+template<>
+inline v3b Mix( const v3b& v1, const v3b& v2, float amount )
+{
+    return v3b( v3f( v1 ) + ( v3f( v2 ) - v3f( v1 ) ) * amount );
+}
+
+template<class T>
+Vector3<T> Desaturate( const Vector3<T>& v )
+{
+    T l = v.Luminance();
+    return Vector3<T>( l, l, l );
+}
+
+template<class T>
+Vector3<T> Desaturate( const Vector3<T>& v, float mul )
+{
+    T l = T( v.Luminance() * mul );
+    return Vector3<T>( l, l, l );
+}
+
+template<class T>
+Vector3<T> pow( const Vector3<T>& base, float exponent )
+{
+    return Vector3<T>(
+        pow( base.x, exponent ),
+        pow( base.y, exponent ),
+        pow( base.z, exponent ) );
+}
+
+template<class T>
+Vector3<T> sRGB2linear( const Vector3<T>& v )
+{
+    return Vector3<T>(
+        sRGB2linear( v.x ),
+        sRGB2linear( v.y ),
+        sRGB2linear( v.z ) );
+}
+
+template<class T>
+Vector3<T> linear2sRGB( const Vector3<T>& v )
+{
+    return Vector3<T>(
+        linear2sRGB( v.x ),
+        linear2sRGB( v.y ),
+        linear2sRGB( v.z ) );
+}
+
+#endif

scripts/texturec.lua

@@ -21,6 +21,8 @@ project "texturec"
 		path.join(BGFX_DIR, "3rdparty/libsquish/**.h"),
 		path.join(BGFX_DIR, "3rdparty/etc1/**.cpp"),
 		path.join(BGFX_DIR, "3rdparty/etc1/**.h"),
+		path.join(BGFX_DIR, "3rdparty/etc2/**.cpp"),
+		path.join(BGFX_DIR, "3rdparty/etc2/**.hpp"),
 		path.join(BGFX_DIR, "3rdparty/nvtt/**.cpp"),
 		path.join(BGFX_DIR, "3rdparty/nvtt/**.h"),
 		path.join(BGFX_DIR, "3rdparty/pvrtc/**.cpp"),

tools/texturec/texturec.cpp

@@ -13,6 +13,7 @@
 #include "image.h"
 #include <libsquish/squish.h>
 #include <etc1/etc1.h>
+#include <etc2/ProcessRGB.hpp>
 #include <nvtt/nvtt.h>
 #include <pvrtc/PvrTcEncoder.h>
 #include <tinyexr/tinyexr.h>
@@ -86,6 +87,23 @@ namespace bgfx
 			etc1_encode_image( (const uint8_t*)_src, _width, _height, 4, _width*4, (uint8_t*)_dst);
 			return true;
 
+		case TextureFormat::ETC2:
+			{
+				const uint32_t pitch  = _width*4;
+				const uint32_t width  = _width/4;
+				const uint32_t height = _height/4;
+				const uint8_t* src = (const uint8_t*)_src;
+				uint64_t* dst = (uint64_t*)_dst;
+				for (uint32_t yy = 0; yy < height; ++yy)
+				{
+					for (uint32_t xx = 0; xx < width; ++xx)
+					{
+						*dst++ = ProcessRGB_ETC2(&src[(yy*pitch+xx)*4]);
+					}
+				}
+			}
+			return true;
+
 		case TextureFormat::PTC14:
 			{
 				using namespace Javelin;