Scarfski/bimg
1
0
Fork 0
mirror of https://github.com/bkaradzic/bimg.git synced 2026-08-11 07:49:36 +00:00
Code Issues Packages Projects Releases Wiki Activity

ƒƒ‚# This is a combination of 2 commits. (#139)

Browse Source 
Updated etcpak. Added ETC2A/EAC encode and EAC/ETC2A1 decode.

.
This commit is contained in:
Branimir Karadžić
2026-06-17 21:47:27 -07:00
committed by Бранимир Караџић
parent 0d3cfc25b5
commit 41c867b0a2
19 changed files with 5058 additions and 934 deletions
Download Patch File Download Diff File Expand all files Collapse all files

719
3rdparty/etc2/ProcessRGB.cpp vendored
View File

@@ -1,719 +0,0 @@
#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 uint16 v4i[4];
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][0] = uint16( (r[2] + r[3] + 4) / 8 );
a[0][1] = uint16( (g[2] + g[3] + 4) / 8 );
a[0][2] = uint16( (b[2] + b[3] + 4) / 8 );
a[0][3] = 0;
a[1][0] = uint16( (r[0] + r[1] + 4) / 8 );
a[1][1] = uint16( (g[0] + g[1] + 4) / 8 );
a[1][2] = uint16( (b[0] + b[1] + 4) / 8 );
a[1][3] = 0;
a[2][0] = uint16( (r[1] + r[3] + 4) / 8 );
a[2][1] = uint16( (g[1] + g[3] + 4) / 8 );
a[2][2] = uint16( (b[1] + b[3] + 4) / 8 );
a[2][3] = 0;
a[3][0] = uint16( (r[0] + r[2] + 4) / 8 );
a[3][1] = uint16( (g[0] + g[2] + 4) / 8 );
a[3][2] = uint16( (b[0] + b[2] + 4) / 8 );
a[3][3] = 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]);
__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], a0);
}
for( int i=0; i<2; i++ )
{
__m128i d = _mm_loadu_si128((__m128i*)a[i*2]);
__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], 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 = (aR * 255.0f + (bR * 255.0f + dR));
float cofG = (aG * 255.0f + (bG * 255.0f + dG));
float cofB = (aB * 255.0f + (bB * 255.0f + dB));
float chfR = (aR * -425.0f + (bR * 255.0f + dR));
float chfG = (aG * -425.0f + (bG * 255.0f + dG));
float chfB = (aB * -425.0f + (bB * 255.0f + dB));
float cvfR = (aR * 255.0f + (bR * -425.0f + dR));
float cvfG = (aG * 255.0f + (bG * -425.0f + dG));
float cvfB = (aB * 255.0f + (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 );
}

9
3rdparty/etc2/ProcessRGB.hpp vendored
View File

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

109
3rdparty/etc2/Tables.cpp vendored
View File

@@ -1,109 +0,0 @@
#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

25
3rdparty/etc2/Tables.hpp vendored
View File

@@ -1,25 +0,0 @@
#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

17
3rdparty/etc2/Types.hpp vendored
View File

@@ -1,17 +0,0 @@
#ifndef __DARKRL__TYPES_HPP__
#define __DARKRL__TYPES_HPP__
#include <stdint.h>
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

120
3rdparty/etcpak/Dither.cpp vendored Normal file
View File

@@ -0,0 +1,120 @@
#include <algorithm>
#include <string.h>
#include "Dither.hpp"
#include "Math.hpp"
#ifdef __SSE4_1__
# ifdef _MSC_VER
# include <intrin.h>
# include <Windows.h>
# else
# include <x86intrin.h>
# endif
#endif
#ifdef __AVX2__
void DitherAvx2( uint8_t* data, __m128i px0, __m128i px1, __m128i px2, __m128i px3 )
{
static constexpr uint8_t a31[] = { 0, 0, 0, 1, 2, 0, 4, 0, 0, 2, 0, 0, 4, 0, 3, 0 };
static constexpr uint8_t a63[] = { 0, 0, 0, 0, 1, 0, 2, 0, 0, 1, 0, 0, 2, 0, 1, 0 };
static constexpr uint8_t s31[] = { 5, 0, 4, 0, 0, 2, 0, 1, 3, 0, 4, 0, 0, 0, 0, 2 };
static constexpr uint8_t s63[] = { 2, 0, 2, 0, 0, 1, 0, 0, 1, 0, 2, 0, 0, 0, 0, 1 };
const __m256i BayerAdd0 = _mm256_setr_epi8(
a31[0], a63[0], a31[0], 0, a31[1], a63[1], a31[1], 0, a31[2], a63[2], a31[2], 0, a31[3], a63[3], a31[3], 0,
a31[4], a63[4], a31[4], 0, a31[5], a63[5], a31[5], 0, a31[6], a63[6], a31[6], 0, a31[7], a63[7], a31[7], 0
);
const __m256i BayerAdd1 = _mm256_setr_epi8(
a31[8], a63[8], a31[8], 0, a31[9], a63[9], a31[9], 0, a31[10], a63[10], a31[10], 0, a31[11], a63[11], a31[11], 0,
a31[12], a63[12], a31[12], 0, a31[13], a63[13], a31[13], 0, a31[14], a63[14], a31[14], 0, a31[15], a63[15], a31[15], 0
);
const __m256i BayerSub0 = _mm256_setr_epi8(
s31[0], s63[0], s31[0], 0, s31[1], s63[1], s31[1], 0, s31[2], s63[2], s31[2], 0, s31[3], s63[3], s31[3], 0,
s31[4], s63[4], s31[4], 0, s31[5], s63[5], s31[5], 0, s31[6], s63[6], s31[6], 0, s31[7], s63[7], s31[7], 0
);
const __m256i BayerSub1 = _mm256_setr_epi8(
s31[8], s63[8], s31[8], 0, s31[9], s63[9], s31[9], 0, s31[10], s63[10], s31[10], 0, s31[11], s63[11], s31[11], 0,
s31[12], s63[12], s31[12], 0, s31[13], s63[13], s31[13], 0, s31[14], s63[14], s31[14], 0, s31[15], s63[15], s31[15], 0
);
__m256i l0 = _mm256_inserti128_si256( _mm256_castsi128_si256( px0 ), px1, 1 );
__m256i l1 = _mm256_inserti128_si256( _mm256_castsi128_si256( px2 ), px3, 1 );
__m256i a0 = _mm256_adds_epu8( l0, BayerAdd0 );
__m256i a1 = _mm256_adds_epu8( l1, BayerAdd1 );
__m256i s0 = _mm256_subs_epu8( a0, BayerSub0 );
__m256i s1 = _mm256_subs_epu8( a1, BayerSub1 );
_mm256_storeu_si256( (__m256i*)(data ), s0 );
_mm256_storeu_si256( (__m256i*)(data+32), s1 );
}
#endif
void Dither( uint8_t* data )
{
#ifdef __AVX2__
static constexpr uint8_t a31[] = { 0, 0, 0, 1, 2, 0, 4, 0, 0, 2, 0, 0, 4, 0, 3, 0 };
static constexpr uint8_t a63[] = { 0, 0, 0, 0, 1, 0, 2, 0, 0, 1, 0, 0, 2, 0, 1, 0 };
static constexpr uint8_t s31[] = { 5, 0, 4, 0, 0, 2, 0, 1, 3, 0, 4, 0, 0, 0, 0, 2 };
static constexpr uint8_t s63[] = { 2, 0, 2, 0, 0, 1, 0, 0, 1, 0, 2, 0, 0, 0, 0, 1 };
const __m256i BayerAdd0 = _mm256_setr_epi8(
a31[0], a63[0], a31[0], 0, a31[1], a63[1], a31[1], 0, a31[2], a63[2], a31[2], 0, a31[3], a63[3], a31[3], 0,
a31[4], a63[4], a31[4], 0, a31[5], a63[5], a31[5], 0, a31[6], a63[6], a31[6], 0, a31[7], a63[7], a31[7], 0
);
const __m256i BayerAdd1 = _mm256_setr_epi8(
a31[8], a63[8], a31[8], 0, a31[9], a63[9], a31[9], 0, a31[10], a63[10], a31[10], 0, a31[11], a63[11], a31[11], 0,
a31[12], a63[12], a31[12], 0, a31[13], a63[13], a31[13], 0, a31[14], a63[14], a31[14], 0, a31[15], a63[15], a31[15], 0
);
const __m256i BayerSub0 = _mm256_setr_epi8(
s31[0], s63[0], s31[0], 0, s31[1], s63[1], s31[1], 0, s31[2], s63[2], s31[2], 0, s31[3], s63[3], s31[3], 0,
s31[4], s63[4], s31[4], 0, s31[5], s63[5], s31[5], 0, s31[6], s63[6], s31[6], 0, s31[7], s63[7], s31[7], 0
);
const __m256i BayerSub1 = _mm256_setr_epi8(
s31[8], s63[8], s31[8], 0, s31[9], s63[9], s31[9], 0, s31[10], s63[10], s31[10], 0, s31[11], s63[11], s31[11], 0,
s31[12], s63[12], s31[12], 0, s31[13], s63[13], s31[13], 0, s31[14], s63[14], s31[14], 0, s31[15], s63[15], s31[15], 0
);
__m256i px0 = _mm256_loadu_si256( (__m256i*)(data ) );
__m256i px1 = _mm256_loadu_si256( (__m256i*)(data+32) );
__m256i a0 = _mm256_adds_epu8( px0, BayerAdd0 );
__m256i a1 = _mm256_adds_epu8( px1, BayerAdd1 );
__m256i s0 = _mm256_subs_epu8( a0, BayerSub0 );
__m256i s1 = _mm256_subs_epu8( a1, BayerSub1 );
_mm256_storeu_si256( (__m256i*)(data ), s0 );
_mm256_storeu_si256( (__m256i*)(data+32), s1 );
#else
static constexpr int8_t Bayer31[16] = {
( 0-8)*2/3, ( 8-8)*2/3, ( 2-8)*2/3, (10-8)*2/3,
(12-8)*2/3, ( 4-8)*2/3, (14-8)*2/3, ( 6-8)*2/3,
( 3-8)*2/3, (11-8)*2/3, ( 1-8)*2/3, ( 9-8)*2/3,
(15-8)*2/3, ( 7-8)*2/3, (13-8)*2/3, ( 5-8)*2/3
};
static constexpr int8_t Bayer63[16] = {
( 0-8)*2/6, ( 8-8)*2/6, ( 2-8)*2/6, (10-8)*2/6,
(12-8)*2/6, ( 4-8)*2/6, (14-8)*2/6, ( 6-8)*2/6,
( 3-8)*2/6, (11-8)*2/6, ( 1-8)*2/6, ( 9-8)*2/6,
(15-8)*2/6, ( 7-8)*2/6, (13-8)*2/6, ( 5-8)*2/6
};
for( int i=0; i<16; i++ )
{
uint32_t col;
memcpy( &col, data, 4 );
uint8_t r = col & 0xFF;
uint8_t g = ( col >> 8 ) & 0xFF;
uint8_t b = ( col >> 16 ) & 0xFF;
r = clampu8( r + Bayer31[i] );
g = clampu8( g + Bayer63[i] );
b = clampu8( b + Bayer31[i] );
col = r | ( g << 8 ) | ( b << 16 );
memcpy( data, &col, 4 );
data += 4;
}
#endif
}

21
3rdparty/etcpak/Dither.hpp vendored Normal file
View File

@@ -0,0 +1,21 @@
#ifndef __DITHER_HPP__
#define __DITHER_HPP__
#include <stddef.h>
#include <stdint.h>
#ifdef __AVX2__
# ifdef _MSC_VER
# include <intrin.h>
# else
# include <x86intrin.h>
# endif
#endif
void Dither( uint8_t* data );
#ifdef __AVX2__
void DitherAvx2( uint8_t* data, __m128i px0, __m128i px1, __m128i px2, __m128i px3 );
#endif
#endif

20
3rdparty/etcpak/ForceInline.hpp vendored Normal file
View File

@@ -0,0 +1,20 @@
#ifndef __FORCEINLINE_HPP__
#define __FORCEINLINE_HPP__
#if defined(__GNUC__)
# define etcpak_force_inline __attribute__((always_inline)) inline
#elif defined(_MSC_VER)
# define etcpak_force_inline __forceinline
#else
# define etcpak_force_inline inline
#endif
#if defined(__GNUC__)
# define etcpak_no_inline __attribute__((noinline))
#elif defined(_MSC_VER)
# define etcpak_no_inline __declspec(noinline)
#else
# define etcpak_no_inline
#endif
#endif

4
3rdparty/etc2/LICENSE.txt → 3rdparty/etcpak/LICENSE.txt vendored
View File

@@ -1,4 +1,6 @@
Copyright (c) 2013, Bartosz Taudul <wolf.pld@gmail.com>
etcpak, an extremely fast ETC compression utility (https://github.com/wolfpld/etcpak)
Copyright (c) 2013-2026, Bartosz Taudul <wolf@nereid.pl>
All rights reserved.
Redistribution and use in source and binary forms, with or without

28
3rdparty/etc2/Math.hpp → 3rdparty/etcpak/Math.hpp vendored
View File

@@ -2,12 +2,13 @@
#define __DARKRL__MATH_HPP__
#include <algorithm>
#include <math.h>
#include <cmath>
#include <stdint.h>
#include "Types.hpp"
#include "ForceInline.hpp"
template<typename T>
inline T AlignPOT( T val )
static etcpak_force_inline T AlignPOT( T val )
{
if( val == 0 ) return 1;
val--;
@@ -18,7 +19,7 @@ inline T AlignPOT( T val )
return val + 1;
}
inline int CountSetBits( uint32 val )
static etcpak_force_inline int CountSetBits( uint32_t val )
{
val -= ( val >> 1 ) & 0x55555555;
val = ( ( val >> 2 ) & 0x33333333 ) + ( val & 0x33333333 );
@@ -28,7 +29,7 @@ inline int CountSetBits( uint32 val )
return val & 0x0000003f;
}
inline int CountLeadingZeros( uint32 val )
static etcpak_force_inline int CountLeadingZeros( uint32_t val )
{
val |= val >> 1;
val |= val >> 2;
@@ -38,7 +39,7 @@ inline int CountLeadingZeros( uint32 val )
return 32 - CountSetBits( val );
}
inline float sRGB2linear( float v )
static etcpak_force_inline float sRGB2linear( float v )
{
const float a = 0.055f;
if( v <= 0.04045f )
@@ -47,11 +48,11 @@ inline float sRGB2linear( float v )
}
else
{
return powf( ( v + a ) / ( 1 + a ), 2.4f );
return pow( ( v + a ) / ( 1 + a ), 2.4f );
}
}
inline float linear2sRGB( float v )
static etcpak_force_inline float linear2sRGB( float v )
{
const float a = 0.055f;
if( v <= 0.0031308f )
@@ -65,23 +66,24 @@ inline float linear2sRGB( float v )
}
template<class T>
inline T SmoothStep( T x )
static etcpak_force_inline T SmoothStep( T x )
{
return x*x*(3-2*x);
}
inline uint8 clampu8( int32 val )
static etcpak_force_inline uint8_t clampu8( int32_t val )
{
return std::min( std::max( 0, val ), 255 );
if( ( val & ~0xFF ) == 0 ) return val;
return ( ( ~val ) >> 31 ) & 0xFF;
}
template<class T>
inline T sq( T val )
static etcpak_force_inline T sq( T val )
{
return val * val;
}
static inline int mul8bit( int a, int b )
static etcpak_force_inline int mul8bit( int a, int b )
{
int t = a*b + 128;
return ( t + ( t >> 8 ) ) >> 8;

9
3rdparty/etc2/ProcessCommon.hpp → 3rdparty/etcpak/ProcessCommon.hpp vendored
View File

@@ -3,8 +3,7 @@
#include <assert.h>
#include <stddef.h>
#include "Types.hpp"
#include <stdint.h>
template<class T>
static size_t GetLeastError( const T* err, size_t num )
@@ -20,7 +19,7 @@ static size_t GetLeastError( const T* err, size_t num )
return idx;
}
static uint64 FixByteOrder( uint64 d )
static uint64_t FixByteOrder( uint64_t d )
{
return ( ( d & 0x00000000FFFFFFFF ) ) |
( ( d & 0xFF00000000000000 ) >> 24 ) |
@@ -30,7 +29,7 @@ static uint64 FixByteOrder( uint64 d )
}
template<class T, class S>
static uint64 EncodeSelectors( uint64 d, const T terr[2][8], const S tsel[16][8], const uint32* id )
static uint64_t EncodeSelectors( uint64_t d, const T terr[2][8], const S tsel[16][8], const uint32_t* id )
{
size_t tidx[2];
tidx[0] = GetLeastError( terr[0], 8 );
@@ -40,7 +39,7 @@ static uint64 EncodeSelectors( uint64 d, const T terr[2][8], const S tsel[16][8]
d |= tidx[1] << 29;
for( int i=0; i<16; i++ )
{
uint64 t = tsel[i][tidx[id[i]%2]];
uint64_t t = tsel[i][tidx[id[i]%2]];
d |= ( t & 0x1 ) << ( i + 32 );
d |= ( t & 0x2 ) << ( i + 47 );
}

4210
3rdparty/etcpak/ProcessRGB.cpp vendored Normal file
View File

File diff suppressed because it is too large Load Diff

14
3rdparty/etcpak/ProcessRGB.hpp vendored Normal file
View File

@@ -0,0 +1,14 @@
#ifndef __PROCESSRGB_HPP__
#define __PROCESSRGB_HPP__
#include <stdint.h>
void CompressEtc1Rgb( const uint32_t* src, uint64_t* dst, uint32_t blocks, size_t width );
void CompressEtc1RgbDither( const uint32_t* src, uint64_t* dst, uint32_t blocks, size_t width );
void CompressEtc2Rgb( const uint32_t* src, uint64_t* dst, uint32_t blocks, size_t width, bool useHeuristics );
void CompressEtc2Rgba( const uint32_t* src, uint64_t* dst, uint32_t blocks, size_t width, bool useHeuristics );
void CompressEacR( const uint32_t* src, uint64_t* dst, uint32_t blocks, size_t width );
void CompressEacRg( const uint32_t* src, uint64_t* dst, uint32_t blocks, size_t width );
#endif

223
3rdparty/etcpak/Tables.cpp vendored Normal file
View File

@@ -0,0 +1,223 @@
#include "Tables.hpp"
const int32_t 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_t 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_t 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_t g_avg2[16] = {
0x00,
0x11,
0x22,
0x33,
0x44,
0x55,
0x66,
0x77,
0x88,
0x99,
0xAA,
0xBB,
0xCC,
0xDD,
0xEE,
0xFF
};
const uint32_t 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
};
const int32_t g_alpha[16][8] = {
{ -3, -6, -9, -15, 2, 5, 8, 14 },
{ -3, -7, -10, -13, 2, 6, 9, 12 },
{ -2, -5, -8, -13, 1, 4, 7, 12 },
{ -2, -4, -6, -13, 1, 3, 5, 12 },
{ -3, -6, -8, -12, 2, 5, 7, 11 },
{ -3, -7, -9, -11, 2, 6, 8, 10 },
{ -4, -7, -8, -11, 3, 6, 7, 10 },
{ -3, -5, -8, -11, 2, 4, 7, 10 },
{ -2, -6, -8, -10, 1, 5, 7, 9 },
{ -2, -5, -8, -10, 1, 4, 7, 9 },
{ -2, -4, -8, -10, 1, 3, 7, 9 },
{ -2, -5, -7, -10, 1, 4, 6, 9 },
{ -3, -4, -7, -10, 2, 3, 6, 9 },
{ -1, -2, -3, -10, 0, 1, 2, 9 },
{ -4, -6, -8, -9, 3, 5, 7, 8 },
{ -3, -5, -7, -9, 2, 4, 6, 8 }
};
const int32_t g_alpha11Mul[16] = { 1, 8, 16, 24, 32, 40, 48, 56, 64, 72, 80, 88, 96, 104, 112, 120 };
const int32_t g_alphaRange[16] = {
0x100FF / ( 1 + g_alpha[0][7] - g_alpha[0][3] ),
0x100FF / ( 1 + g_alpha[1][7] - g_alpha[1][3] ),
0x100FF / ( 1 + g_alpha[2][7] - g_alpha[2][3] ),
0x100FF / ( 1 + g_alpha[3][7] - g_alpha[3][3] ),
0x100FF / ( 1 + g_alpha[4][7] - g_alpha[4][3] ),
0x100FF / ( 1 + g_alpha[5][7] - g_alpha[5][3] ),
0x100FF / ( 1 + g_alpha[6][7] - g_alpha[6][3] ),
0x100FF / ( 1 + g_alpha[7][7] - g_alpha[7][3] ),
0x100FF / ( 1 + g_alpha[8][7] - g_alpha[8][3] ),
0x100FF / ( 1 + g_alpha[9][7] - g_alpha[9][3] ),
0x100FF / ( 1 + g_alpha[10][7] - g_alpha[10][3] ),
0x100FF / ( 1 + g_alpha[11][7] - g_alpha[11][3] ),
0x100FF / ( 1 + g_alpha[12][7] - g_alpha[12][3] ),
0x100FF / ( 1 + g_alpha[13][7] - g_alpha[13][3] ),
0x100FF / ( 1 + g_alpha[14][7] - g_alpha[14][3] ),
0x100FF / ( 1 + g_alpha[15][7] - g_alpha[15][3] ),
};
#ifdef __SSE4_1__
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)
};
const __m128i g_alpha_SIMD[16] = {
_mm_setr_epi16( g_alpha[ 0][0], g_alpha[ 0][1], g_alpha[ 0][2], g_alpha[ 0][3], g_alpha[ 0][4], g_alpha[ 0][5], g_alpha[ 0][6], g_alpha[ 0][7] ),
_mm_setr_epi16( g_alpha[ 1][0], g_alpha[ 1][1], g_alpha[ 1][2], g_alpha[ 1][3], g_alpha[ 1][4], g_alpha[ 1][5], g_alpha[ 1][6], g_alpha[ 1][7] ),
_mm_setr_epi16( g_alpha[ 2][0], g_alpha[ 2][1], g_alpha[ 2][2], g_alpha[ 2][3], g_alpha[ 2][4], g_alpha[ 2][5], g_alpha[ 2][6], g_alpha[ 2][7] ),
_mm_setr_epi16( g_alpha[ 3][0], g_alpha[ 3][1], g_alpha[ 3][2], g_alpha[ 3][3], g_alpha[ 3][4], g_alpha[ 3][5], g_alpha[ 3][6], g_alpha[ 3][7] ),
_mm_setr_epi16( g_alpha[ 4][0], g_alpha[ 4][1], g_alpha[ 4][2], g_alpha[ 4][3], g_alpha[ 4][4], g_alpha[ 4][5], g_alpha[ 4][6], g_alpha[ 4][7] ),
_mm_setr_epi16( g_alpha[ 5][0], g_alpha[ 5][1], g_alpha[ 5][2], g_alpha[ 5][3], g_alpha[ 5][4], g_alpha[ 5][5], g_alpha[ 5][6], g_alpha[ 5][7] ),
_mm_setr_epi16( g_alpha[ 6][0], g_alpha[ 6][1], g_alpha[ 6][2], g_alpha[ 6][3], g_alpha[ 6][4], g_alpha[ 6][5], g_alpha[ 6][6], g_alpha[ 6][7] ),
_mm_setr_epi16( g_alpha[ 7][0], g_alpha[ 7][1], g_alpha[ 7][2], g_alpha[ 7][3], g_alpha[ 7][4], g_alpha[ 7][5], g_alpha[ 7][6], g_alpha[ 7][7] ),
_mm_setr_epi16( g_alpha[ 8][0], g_alpha[ 8][1], g_alpha[ 8][2], g_alpha[ 8][3], g_alpha[ 8][4], g_alpha[ 8][5], g_alpha[ 8][6], g_alpha[ 8][7] ),
_mm_setr_epi16( g_alpha[ 9][0], g_alpha[ 9][1], g_alpha[ 9][2], g_alpha[ 9][3], g_alpha[ 9][4], g_alpha[ 9][5], g_alpha[ 9][6], g_alpha[ 9][7] ),
_mm_setr_epi16( g_alpha[10][0], g_alpha[10][1], g_alpha[10][2], g_alpha[10][3], g_alpha[10][4], g_alpha[10][5], g_alpha[10][6], g_alpha[10][7] ),
_mm_setr_epi16( g_alpha[11][0], g_alpha[11][1], g_alpha[11][2], g_alpha[11][3], g_alpha[11][4], g_alpha[11][5], g_alpha[11][6], g_alpha[11][7] ),
_mm_setr_epi16( g_alpha[12][0], g_alpha[12][1], g_alpha[12][2], g_alpha[12][3], g_alpha[12][4], g_alpha[12][5], g_alpha[12][6], g_alpha[12][7] ),
_mm_setr_epi16( g_alpha[13][0], g_alpha[13][1], g_alpha[13][2], g_alpha[13][3], g_alpha[13][4], g_alpha[13][5], g_alpha[13][6], g_alpha[13][7] ),
_mm_setr_epi16( g_alpha[14][0], g_alpha[14][1], g_alpha[14][2], g_alpha[14][3], g_alpha[14][4], g_alpha[14][5], g_alpha[14][6], g_alpha[14][7] ),
_mm_setr_epi16( g_alpha[15][0], g_alpha[15][1], g_alpha[15][2], g_alpha[15][3], g_alpha[15][4], g_alpha[15][5], g_alpha[15][6], g_alpha[15][7] ),
};
const __m128i g_alphaRange_SIMD = _mm_setr_epi16(
g_alphaRange[0],
g_alphaRange[1],
g_alphaRange[4],
g_alphaRange[5],
g_alphaRange[8],
g_alphaRange[14],
0,
0 );
#endif
#ifdef __AVX2__
const __m256i g_alpha_AVX[8] = {
_mm256_setr_epi16( g_alpha[ 0][0], g_alpha[ 1][0], g_alpha[ 2][0], g_alpha[ 3][0], g_alpha[ 4][0], g_alpha[ 5][0], g_alpha[ 6][0], g_alpha[ 7][0], g_alpha[ 8][0], g_alpha[ 9][0], g_alpha[10][0], g_alpha[11][0], g_alpha[12][0], g_alpha[13][0], g_alpha[14][0], g_alpha[15][0] ),
_mm256_setr_epi16( g_alpha[ 0][1], g_alpha[ 1][1], g_alpha[ 2][1], g_alpha[ 3][1], g_alpha[ 4][1], g_alpha[ 5][1], g_alpha[ 6][1], g_alpha[ 7][1], g_alpha[ 8][1], g_alpha[ 9][1], g_alpha[10][1], g_alpha[11][1], g_alpha[12][1], g_alpha[13][1], g_alpha[14][1], g_alpha[15][1] ),
_mm256_setr_epi16( g_alpha[ 0][2], g_alpha[ 1][2], g_alpha[ 2][2], g_alpha[ 3][2], g_alpha[ 4][2], g_alpha[ 5][2], g_alpha[ 6][2], g_alpha[ 7][2], g_alpha[ 8][2], g_alpha[ 9][2], g_alpha[10][2], g_alpha[11][2], g_alpha[12][2], g_alpha[13][2], g_alpha[14][2], g_alpha[15][2] ),
_mm256_setr_epi16( g_alpha[ 0][3], g_alpha[ 1][3], g_alpha[ 2][3], g_alpha[ 3][3], g_alpha[ 4][3], g_alpha[ 5][3], g_alpha[ 6][3], g_alpha[ 7][3], g_alpha[ 8][3], g_alpha[ 9][3], g_alpha[10][3], g_alpha[11][3], g_alpha[12][3], g_alpha[13][3], g_alpha[14][3], g_alpha[15][3] ),
_mm256_setr_epi16( g_alpha[ 0][4], g_alpha[ 1][4], g_alpha[ 2][4], g_alpha[ 3][4], g_alpha[ 4][4], g_alpha[ 5][4], g_alpha[ 6][4], g_alpha[ 7][4], g_alpha[ 8][4], g_alpha[ 9][4], g_alpha[10][4], g_alpha[11][4], g_alpha[12][4], g_alpha[13][4], g_alpha[14][4], g_alpha[15][4] ),
_mm256_setr_epi16( g_alpha[ 0][5], g_alpha[ 1][5], g_alpha[ 2][5], g_alpha[ 3][5], g_alpha[ 4][5], g_alpha[ 5][5], g_alpha[ 6][5], g_alpha[ 7][5], g_alpha[ 8][5], g_alpha[ 9][5], g_alpha[10][5], g_alpha[11][5], g_alpha[12][5], g_alpha[13][5], g_alpha[14][5], g_alpha[15][5] ),
_mm256_setr_epi16( g_alpha[ 0][6], g_alpha[ 1][6], g_alpha[ 2][6], g_alpha[ 3][6], g_alpha[ 4][6], g_alpha[ 5][6], g_alpha[ 6][6], g_alpha[ 7][6], g_alpha[ 8][6], g_alpha[ 9][6], g_alpha[10][6], g_alpha[11][6], g_alpha[12][6], g_alpha[13][6], g_alpha[14][6], g_alpha[15][6] ),
_mm256_setr_epi16( g_alpha[ 0][7], g_alpha[ 1][7], g_alpha[ 2][7], g_alpha[ 3][7], g_alpha[ 4][7], g_alpha[ 5][7], g_alpha[ 6][7], g_alpha[ 7][7], g_alpha[ 8][7], g_alpha[ 9][7], g_alpha[10][7], g_alpha[11][7], g_alpha[12][7], g_alpha[13][7], g_alpha[14][7], g_alpha[15][7] ),
};
const __m256i g_alphaRange_AVX = _mm256_setr_epi16(
g_alphaRange[ 0], g_alphaRange[ 1], g_alphaRange[ 2], g_alphaRange[ 3], g_alphaRange[ 4], g_alphaRange[ 5], g_alphaRange[ 6], g_alphaRange[ 7],
g_alphaRange[ 8], g_alphaRange[ 9], g_alphaRange[10], g_alphaRange[11], g_alphaRange[12], g_alphaRange[13], g_alphaRange[14], g_alphaRange[15]
);
#endif
#ifdef __ARM_NEON
const int16x8_t g_table128_NEON[2] =
{
{ 2*128, 5*128, 9*128, 13*128, 18*128, 24*128, 33*128, 47*128 },
{ 8*128, 17*128, 29*128, 42*128, 60*128, 80*128, 106*128, 183*128 }
};
const int32x4_t g_table256_NEON[4] =
{
{ 2*256, 5*256, 9*256, 13*256 },
{ 8*256, 17*256, 29*256, 42*256 },
{ 18*256, 24*256, 33*256, 47*256 },
{ 60*256, 80*256, 106*256, 183*256 }
};
const int16x8_t g_alpha_NEON[16] =
{
{ -3, -6, -9, -15, 2, 5, 8, 14 },
{ -3, -7, -10, -13, 2, 6, 9, 12 },
{ -2, -5, -8, -13, 1, 4, 7, 12 },
{ -2, -4, -6, -13, 1, 3, 5, 12 },
{ -3, -6, -8, -12, 2, 5, 7, 11 },
{ -3, -7, -9, -11, 2, 6, 8, 10 },
{ -4, -7, -8, -11, 3, 6, 7, 10 },
{ -3, -5, -8, -11, 2, 4, 7, 10 },
{ -2, -6, -8, -10, 1, 5, 7, 9 },
{ -2, -5, -8, -10, 1, 4, 7, 9 },
{ -2, -4, -8, -10, 1, 3, 7, 9 },
{ -2, -5, -7, -10, 1, 4, 6, 9 },
{ -3, -4, -7, -10, 2, 3, 6, 9 },
{ -1, -2, -3, -10, 0, 1, 2, 9 },
{ -4, -6, -8, -9, 3, 5, 7, 8 },
{ -3, -5, -7, -9, 2, 4, 6, 8 }
};
const int16x8_t g_alphaRange_NEON =
{
(int16_t)g_alphaRange[0],
(int16_t)g_alphaRange[1],
(int16_t)g_alphaRange[4],
(int16_t)g_alphaRange[5],
(int16_t)g_alphaRange[8],
(int16_t)g_alphaRange[14],
0,
0
};
#endif

50
3rdparty/etcpak/Tables.hpp vendored Normal file
View File

@@ -0,0 +1,50 @@
#ifndef __TABLES_HPP__
#define __TABLES_HPP__
#include <stdint.h>
#ifdef __AVX2__
# include <immintrin.h>
#endif
#ifdef __SSE4_1__
# include <smmintrin.h>
#endif
#ifdef __ARM_NEON
# include <arm_neon.h>
#endif
extern const int32_t g_table[8][4];
extern const int64_t g_table256[8][4];
extern const uint32_t g_id[4][16];
extern const uint32_t g_avg2[16];
extern const uint32_t g_flags[64];
extern const int32_t g_alpha[16][8];
extern const int32_t g_alpha11Mul[16];
extern const int32_t g_alphaRange[16];
#ifdef __SSE4_1__
extern const __m128i g_table_SIMD[2];
extern const __m128i g_table128_SIMD[2];
extern const __m128i g_table256_SIMD[4];
extern const __m128i g_alpha_SIMD[16];
extern const __m128i g_alphaRange_SIMD;
#endif
#ifdef __AVX2__
extern const __m256i g_alpha_AVX[8];
extern const __m256i g_alphaRange_AVX;
#endif
#ifdef __ARM_NEON
extern const int16x8_t g_table128_NEON[2];
extern const int32x4_t g_table256_NEON[4];
extern const int16x8_t g_alpha_NEON[16];
extern const int16x8_t g_alphaRange_NEON;
#endif
#endif

14
3rdparty/etc2/Vector.hpp → 3rdparty/etcpak/Vector.hpp vendored
View File

@@ -4,9 +4,9 @@
#include <assert.h>
#include <algorithm>
#include <math.h>
#include <stdint.h>
#include "Math.hpp"
#include "Types.hpp"
template<class T>
struct Vector2
@@ -65,7 +65,7 @@ Vector2<T> operator/( const Vector2<T>& lhs, const T& rhs )
}
typedef Vector2<int32> v2i;
typedef Vector2<int32_t> v2i;
typedef Vector2<float> v2f;
@@ -89,8 +89,8 @@ struct Vector3
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]; }
T& operator[]( unsigned int idx ) { assert( idx < 3 ); return ((T*)this)[idx]; }
const T& operator[]( unsigned int idx ) const { assert( idx < 3 ); return ((T*)this)[idx]; }
Vector3<T> operator+=( const Vector3<T>& rhs )
{
@@ -156,14 +156,14 @@ bool operator<( const Vector3<T>& lhs, const Vector3<T>& rhs )
return lhs.Luminance() < rhs.Luminance();
}
typedef Vector3<int32> v3i;
typedef Vector3<int32_t> v3i;
typedef Vector3<float> v3f;
typedef Vector3<uint8> v3b;
typedef Vector3<uint8_t> 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 ) );
return v3b( uint8_t( std::min( 1.f, v.x ) * 255 ), uint8_t( std::min( 1.f, v.y ) * 255 ), uint8_t( std::min( 1.f, v.z ) * 255 ) );
}
template<class T>

4
scripts/bimg_encode.lua
View File

@@ -25,8 +25,8 @@ project "bimg_encode"
path.join(BIMG_DIR, "3rdparty/edtaa3/**.h"),
path.join(BIMG_DIR, "3rdparty/etc1/**.cpp"),
path.join(BIMG_DIR, "3rdparty/etc1/**.h"),
path.join(BIMG_DIR, "3rdparty/etc2/**.cpp"),
path.join(BIMG_DIR, "3rdparty/etc2/**.hpp"),
path.join(BIMG_DIR, "3rdparty/etcpak/**.cpp"),
path.join(BIMG_DIR, "3rdparty/etcpak/**.hpp"),
path.join(BIMG_DIR, "3rdparty/nvtt/**.cpp"),
path.join(BIMG_DIR, "3rdparty/nvtt/**.h"),
path.join(BIMG_DIR, "3rdparty/pvrtc/**.cpp"),

297
src/image.cpp
View File

@@ -3094,6 +3094,148 @@ namespace bimg
}
}
static const int32_t s_etc2a1Mod[8][4] =
{
{ 0, 8, 0, -8 },
{ 0, 17, 0, -17 },
{ 0, 29, 0, -29 },
{ 0, 42, 0, -42 },
{ 0, 60, 0, -60 },
{ 0, 80, 0, -80 },
{ 0, 106, 0, -106 },
{ 0, 183, 0, -183 },
};
static void decodeBlockEtc2PunchThrough(uint8_t _dst[16*4], const uint8_t _src[8])
{
uint32_t indexMsb = (_src[4]<<8) | _src[5];
uint32_t indexLsb = (_src[6]<<8) | _src[7];
for (uint32_t ii = 0; ii < 16; ++ii)
{
const uint32_t idx = (ii&0xc) | ( (ii & 0x3)<<4);
const uint32_t pixelIndex = (indexLsb & 1) | ( (indexMsb & 1)<<1);
if (2 == pixelIndex)
{
_dst[idx + 0] = 0;
_dst[idx + 1] = 0;
_dst[idx + 2] = 0;
_dst[idx + 3] = 0;
}
indexLsb >>= 1;
indexMsb >>= 1;
}
}
static void decodeBlockEtc2Rgb8A1(uint8_t _dst[16*4], const uint8_t _src[8])
{
if (!BX_ENABLED(BIMG_CONFIG_DECODE_ETC2) )
{
return;
}
const bool flipBit = 0 != (_src[3] & 0x1);
const bool opaque = 0 != (_src[3] & 0x2);
uint8_t rgb[8];
rgb[0] = _src[0] >> 3;
rgb[1] = _src[1] >> 3;
rgb[2] = _src[2] >> 3;
int8_t diff[3];
diff[0] = int8_t( (_src[0] & 0x7)<<5)>>5;
diff[1] = int8_t( (_src[1] & 0x7)<<5)>>5;
diff[2] = int8_t( (_src[2] & 0x7)<<5)>>5;
const int8_t rr = rgb[0] + diff[0];
const int8_t gg = rgb[1] + diff[1];
const int8_t bb = rgb[2] + diff[2];
if (0 > rr
|| 31 < rr)
{
decodeBlockEtc2ModeT(_dst, _src);
if (!opaque)
{
decodeBlockEtc2PunchThrough(_dst, _src);
}
return;
}
if (0 > gg
|| 31 < gg)
{
decodeBlockEtc2ModeH(_dst, _src);
if (!opaque)
{
decodeBlockEtc2PunchThrough(_dst, _src);
}
return;
}
if (0 > bb
|| 31 < bb)
{
decodeBlockEtc2ModePlanar(_dst, _src);
return;
}
rgb[0] = bitRangeConvert(rgb[0], 5, 8);
rgb[1] = bitRangeConvert(rgb[1], 5, 8);
rgb[2] = bitRangeConvert(rgb[2], 5, 8);
rgb[4] = bitRangeConvert(rr, 5, 8);
rgb[5] = bitRangeConvert(gg, 5, 8);
rgb[6] = bitRangeConvert(bb, 5, 8);
uint32_t table[2];
table[0] = (_src[3] >> 5) & 0x7;
table[1] = (_src[3] >> 2) & 0x7;
uint32_t indexMsb = (_src[4]<<8) | _src[5];
uint32_t indexLsb = (_src[6]<<8) | _src[7];
for (uint32_t ii = 0; ii < 16; ++ii)
{
const uint32_t block = flipBit
? (ii>>1)&1
: ii>>3
;
const uint32_t color = block<<2;
const uint32_t idx = (ii&0xc) | ( (ii & 0x3)<<4);
const uint32_t pixelIndex = (indexLsb & 1) | ( (indexMsb & 1)<<1);
if (!opaque
&& 2 == pixelIndex)
{
_dst[idx + 0] = 0;
_dst[idx + 1] = 0;
_dst[idx + 2] = 0;
_dst[idx + 3] = 0;
}
else
{
const int32_t mod = opaque
? s_etc1Mod [table[block] ][pixelIndex]
: s_etc2a1Mod[table[block] ][pixelIndex]
;
_dst[idx + 0] = uint8_satadd(rgb[color+2], mod);
_dst[idx + 1] = uint8_satadd(rgb[color+1], mod);
_dst[idx + 2] = uint8_satadd(rgb[color+0], mod);
_dst[idx + 3] = 255;
}
indexLsb >>= 1;
indexMsb >>= 1;
}
}
static const int8_t s_etc2aMod[16][8] =
{
{ -3, -6, -9, -15, 2, 5, 8, 14 },
@@ -3131,7 +3273,8 @@ namespace bimg
| ((uint64_t)_src[6] << 8)
| _src[7];
for (int ii = 0; ii < 16; ii++) {
for (int ii = 0; ii < 16; ii++)
{
const uint32_t idx = (ii & 0xc) | ((ii & 0x3) << 4);
const int32_t mod = modTable[(indices >> (45 - ii * 3)) & 0x7];
@@ -3139,6 +3282,48 @@ namespace bimg
}
}
template<bool SignedT>
static void decodeBlockEac(uint8_t _dst[16*4], const uint8_t _src[8], uint32_t _byteOffset)
{
if (!BX_ENABLED(BIMG_CONFIG_DECODE_ETC2) )
{
return;
}
const int32_t base = SignedT
? int32_t(int8_t(_src[0]) )
: int32_t(_src[0])
;
const int8_t* modTable = s_etc2aMod[_src[1] & 0x0f];
const int32_t mult = (_src[1] & 0xf0) >> 4;
const uint64_t indices = ( (uint64_t)_src[2] << 40)
| ( (uint64_t)_src[3] << 32)
| ( (uint64_t)_src[4] << 24)
| ( (uint64_t)_src[5] << 16)
| ( (uint64_t)_src[6] << 8)
| _src[7]
;
for (int32_t ii = 0; ii < 16; ++ii)
{
const uint32_t idx = (ii & 0xc) | ( (ii & 0x3) << 4);
const int32_t mod = modTable[(indices >> (45 - ii*3) ) & 0x7];
if (SignedT)
{
const int32_t b = (base < -127 ? -127 : base) * 8;
const int32_t value = bx::clamp(0 != mult ? b + mod*mult*8 : b + mod, -1023, 1023);
_dst[idx + _byteOffset] = uint8_t( (value + 1023)*255/2046);
}
else
{
const int32_t b = base*8 + 4;
const int32_t value = bx::clamp(0 != mult ? b + mod*mult*8 : b + mod, 0, 2047);
_dst[idx + _byteOffset] = uint8_t(value >> 3);
}
}
}
static const uint8_t s_pvrtcFactors[16][4] =
{
@@ -6027,16 +6212,118 @@ namespace bimg
break;
case TextureFormat::ETC2A1:
BX_WARN(false, "ETC2A1 decoder is not implemented.");
imageCheckerboard(_dst, _width, _height, 16, UINT32_C(0xff000000), UINT32_C(0xffff0000) );
if (BX_ENABLED(BIMG_CONFIG_DECODE_ETC2) )
{
for (uint32_t yy = 0; yy < height; ++yy)
{
for (uint32_t xx = 0; xx < width; ++xx)
{
decodeBlockEtc2Rgb8A1(temp, src);
src += 8;
uint8_t* block = &dst[yy*_dstPitch*4 + xx*16];
bx::memCopy(&block[0*_dstPitch], &temp[ 0], 16);
bx::memCopy(&block[1*_dstPitch], &temp[16], 16);
bx::memCopy(&block[2*_dstPitch], &temp[32], 16);
bx::memCopy(&block[3*_dstPitch], &temp[48], 16);
}
}
}
else
{
BX_WARN(false, "ETC2A1 decoder is disabled (BIMG_CONFIG_DECODE_ETC2).");
imageCheckerboard(_dst, _width, _height, 16, UINT32_C(0xff000000), UINT32_C(0xffff0000) );
}
break;
case TextureFormat::EACR11:
case TextureFormat::EACR11S:
if (BX_ENABLED(BIMG_CONFIG_DECODE_ETC2) )
{
const bool sign = TextureFormat::EACR11S == _srcFormat;
for (uint32_t yy = 0; yy < height; ++yy)
{
for (uint32_t xx = 0; xx < width; ++xx)
{
for (uint32_t ii = 0; ii < 16; ++ii)
{
temp[ii*4+0] = 0;
temp[ii*4+1] = 0;
temp[ii*4+2] = 0;
temp[ii*4+3] = 255;
}
if (sign)
{
decodeBlockEac<true >(temp, src, 2);
}
else
{
decodeBlockEac<false>(temp, src, 2);
}
src += 8;
uint8_t* block = &dst[yy*_dstPitch*4 + xx*16];
bx::memCopy(&block[0*_dstPitch], &temp[ 0], 16);
bx::memCopy(&block[1*_dstPitch], &temp[16], 16);
bx::memCopy(&block[2*_dstPitch], &temp[32], 16);
bx::memCopy(&block[3*_dstPitch], &temp[48], 16);
}
}
}
else
{
BX_WARN(false, "EAC decoder is disabled (BIMG_CONFIG_DECODE_ETC2).");
imageCheckerboard(_dst, _width, _height, 16, UINT32_C(0xff000000), UINT32_C(0xff00ffff) );
}
break;
case TextureFormat::EACRG11:
case TextureFormat::EACRG11S:
BX_WARN(false, "EAC decoder is not implemented.");
imageCheckerboard(_dst, _width, _height, 16, UINT32_C(0xff000000), UINT32_C(0xff00ffff) );
if (BX_ENABLED(BIMG_CONFIG_DECODE_ETC2) )
{
const bool sign = TextureFormat::EACRG11S == _srcFormat;
for (uint32_t yy = 0; yy < height; ++yy)
{
for (uint32_t xx = 0; xx < width; ++xx)
{
for (uint32_t ii = 0; ii < 16; ++ii)
{
temp[ii*4+0] = 0;
temp[ii*4+1] = 0;
temp[ii*4+2] = 0;
temp[ii*4+3] = 255;
}
if (sign)
{
decodeBlockEac<true >(temp, src, 2);
decodeBlockEac<true >(temp, src + 8, 1);
}
else
{
decodeBlockEac<false>(temp, src, 2);
decodeBlockEac<false>(temp, src + 8, 1);
}
src += 16;
uint8_t* block = &dst[yy*_dstPitch*4 + xx*16];
bx::memCopy(&block[0*_dstPitch], &temp[ 0], 16);
bx::memCopy(&block[1*_dstPitch], &temp[16], 16);
bx::memCopy(&block[2*_dstPitch], &temp[32], 16);
bx::memCopy(&block[3*_dstPitch], &temp[48], 16);
}
}
}
else
{
BX_WARN(false, "EAC decoder is disabled (BIMG_CONFIG_DECODE_ETC2).");
imageCheckerboard(_dst, _width, _height, 16, UINT32_C(0xff000000), UINT32_C(0xff00ffff) );
}
break;
case TextureFormat::PTC12:

99
src/image_encode.cpp
View File

@@ -8,7 +8,7 @@
#include <libsquish/squish.h>
#include <etc1/etc1.h>
#include <etc2/ProcessRGB.hpp>
#include <etcpak/ProcessRGB.hpp>
#include <nvtt/nvtt.h>
#include <pvrtc/PvrTcEncoder.h>
#include <edtaa3/edtaa3func.h>
@@ -54,6 +54,48 @@ namespace bimg
};
static_assert(Quality::Count == BX_COUNTOF(s_astcQuality) );
static uint32_t* etcpakAllocBgraBlocks(
bx::AllocatorI* _allocator
, const uint8_t* _src
, uint32_t _width
, uint32_t _height
, uint32_t& _outNumBlocks
, uint32_t& _outPaddedWidth
)
{
const uint32_t blockWidth = (_width + 3)/4;
const uint32_t blockHeight = (_height + 3)/4;
const uint32_t paddedWidth = blockWidth *4;
const uint32_t paddedHeight = blockHeight*4;
const uint32_t srcPitch = _width*4;
uint32_t* bgra = (uint32_t*)bx::alloc(_allocator, paddedWidth*paddedHeight*sizeof(uint32_t) );
for (uint32_t yy = 0; yy < paddedHeight; ++yy)
{
const uint32_t sy = bx::min(yy, _height-1);
const uint8_t* srcRow = &_src[sy*srcPitch];
uint32_t* dstRow = &bgra[yy*paddedWidth];
for (uint32_t xx = 0; xx < paddedWidth; ++xx)
{
const uint32_t sx = bx::min(xx, _width-1);
const uint8_t* px = &srcRow[sx*4];
dstRow[xx] = 0
| (uint32_t(px[3])<<24) // A
| (uint32_t(px[0])<<16) // R
| (uint32_t(px[1])<< 8) // G
| (uint32_t(px[2]) ) // B
;
}
}
_outNumBlocks = blockWidth*blockHeight;
_outPaddedWidth = paddedWidth;
return bgra;
}
void imageEncodeFromRgba8(bx::AllocatorI* _allocator, void* _dst, const void* _src, uint32_t _width, uint32_t _height, uint32_t _depth, TextureFormat::Enum _format, Quality::Enum _quality, bx::Error* _err)
{
const uint8_t* src = (const uint8_t*)_src;
@@ -89,9 +131,8 @@ namespace bimg
BX_ERROR_SET(_err, BIMG_ERROR, "Unable to convert between input/output formats!");
break;
case TextureFormat::ETC2A:
case TextureFormat::ETC2A1:
BX_ERROR_SET(_err, BIMG_ERROR, "Encoding to ETC2A/ETC2A1 is not supported.");
BX_ERROR_SET(_err, BIMG_ERROR, "Encoding to ETC2A1 is not supported.");
break;
case TextureFormat::ETC1:
@@ -100,25 +141,37 @@ namespace bimg
case TextureFormat::ETC2:
{
const uint32_t blockWidth = (_width +3)/4;
const uint32_t blockHeight = (_height+3)/4;
uint64_t* dstBlock = (uint64_t*)dst;
for (uint32_t yy = 0; yy < blockHeight; ++yy)
{
for (uint32_t xx = 0; xx < blockWidth; ++xx)
{
uint8_t block[4*4*4];
const uint8_t* ptr = &src[(yy*srcPitch+xx*4)*4];
uint32_t numBlocks, paddedWidth;
uint32_t* bgra = etcpakAllocBgraBlocks(_allocator, src, _width, _height, numBlocks, paddedWidth);
CompressEtc2Rgb(bgra, (uint64_t*)dst, numBlocks, paddedWidth, true);
bx::free(_allocator, bgra);
}
break;
for (uint32_t ii = 0; ii < 16; ++ii)
{ // BGRx
bx::memCopy(&block[ii*4], &ptr[(ii%4)*srcPitch + (ii&~3)], 4);
bx::swap(block[ii*4+0], block[ii*4+2]);
}
case TextureFormat::ETC2A:
{
uint32_t numBlocks, paddedWidth;
uint32_t* bgra = etcpakAllocBgraBlocks(_allocator, src, _width, _height, numBlocks, paddedWidth);
CompressEtc2Rgba(bgra, (uint64_t*)dst, numBlocks, paddedWidth, true);
bx::free(_allocator, bgra);
}
break;
*dstBlock++ = ProcessRGB_ETC2(block);
}
}
case TextureFormat::EACR11:
{
uint32_t numBlocks, paddedWidth;
uint32_t* bgra = etcpakAllocBgraBlocks(_allocator, src, _width, _height, numBlocks, paddedWidth);
CompressEacR(bgra, (uint64_t*)dst, numBlocks, paddedWidth);
bx::free(_allocator, bgra);
}
break;
case TextureFormat::EACRG11:
{
uint32_t numBlocks, paddedWidth;
uint32_t* bgra = etcpakAllocBgraBlocks(_allocator, src, _width, _height, numBlocks, paddedWidth);
CompressEacRg(bgra, (uint64_t*)dst, numBlocks, paddedWidth);
bx::free(_allocator, bgra);
}
break;
@@ -332,6 +385,9 @@ namespace bimg
case TextureFormat::BC5:
case TextureFormat::ETC1:
case TextureFormat::ETC2:
case TextureFormat::ETC2A:
case TextureFormat::EACR11:
case TextureFormat::EACRG11:
case TextureFormat::PTC14:
case TextureFormat::PTC14A:
case TextureFormat::ASTC4x4:
@@ -366,9 +422,8 @@ namespace bimg
}
break;
case bimg::TextureFormat::ETC2A:
case bimg::TextureFormat::ETC2A1:
BX_ERROR_SET(_err, BIMG_ERROR, "Encoding to ETC2A/ETC2A1 is not supported.");
BX_ERROR_SET(_err, BIMG_ERROR, "Encoding to ETC2A1 is not supported.");
break;
default: