Browse Source

Flesh out usage of block classes more

faster-single-tables
Andrew Cassidy 2 months ago
parent
commit
6018ab0cfb
10 changed files with 343 additions and 285 deletions
  1. +5
    -1
      .clang-format
  2. +16
    -19
      .clang-tidy
  3. +1
    -1
      .gitignore
  4. +6
    -4
      CMakeLists.txt
  5. +1
    -31
      src/blocks.cpp
  6. +80
    -132
      src/blocks.h
  7. +76
    -0
      src/color.cpp
  8. +64
    -0
      src/color.h
  9. +89
    -87
      src/rgbcx.cpp
  10. +5
    -10
      src/util.h

+ 5
- 1
.clang-format View File

@ -1,6 +1,10 @@
---
BasedOnStyle: LLVM
BasedOnStyle: google
IndentWidth: 4
ColumnLimit: 160
AllowShortBlocksOnASingleLine: Always
AllowShortFunctionsOnASingleLine: All
AlwaysBreakTemplateDeclarations: MultiLine
#AlignConsecutiveDeclarations: true
---

+ 16
- 19
.clang-tidy View File

@ -1,21 +1,18 @@
---
FormatStyle: LLVM
Checks: 'clang-diagnostic-*,clang-analyzer-*,readability-*'
FormatStyle: google
Checks: '-*,clang-diagnostic-*,llvm-*,misc-*,-misc-unused-parameters,-misc-non-private-member-variables-in-classes,readability-identifier-naming,cppcoreguidelines-narrowing-conversions'
CheckOptions:
- key: readability-identifier-naming.AbstractClassCase
value: 'CamelCase'
- key: readability-identifier-naming.ClassCase
value: 'CamelCase'
- key: readability-identifier-naming.EnumCase
value: 'CamelCase'
- key: readability-identifier-naming.StructCase
value: 'CamelCase'
- key: readability-identifier-naming.TypedefCase
value: 'CamelCase'
- key: readability-identifier-naming.MemberCase
value: 'CamelCase'
- key: readability-identifier-naming.FunctionCase
value: 'camelBack'
- key: readability-identifier-naming.ClassMethodCase
value: 'camelBack'
- { key: readability-identifier-naming.NamespaceCase, value: lower_case }
- { key: readability-identifier-naming.ClassCase, value: CamelCase }
- { key: readability-identifier-naming.StructCase, value: CamelCase }
- { key: readability-identifier-naming.TemplateParameterCase, value: CamelCase }
- { key: readability-identifier-naming.FunctionCase, value: aNy_CasE }
- { key: readability-identifier-naming.VariableCase, value: lower_case }
- { key: readability-identifier-naming.MemberCase, value: lower_case }
- { key: readability-identifier-naming.PrivateMemberPrefix, value: _ }
- { key: readability-identifier-naming.ProtectedMemberPrefix, value: _ }
- { key: readability-identifier-naming.EnumConstantCase, value: CamelCase }
- { key: readability-identifier-naming.ConstexprVariableCase, value: CamelCase }
- { key: readability-identifier-naming.GlobalConstantCase, value: CamelCase }
- { key: readability-identifier-naming.MemberConstantCase, value: CamelCase }
- { key: readability-identifier-naming.StaticConstantCase, value: CamelCase }

+ 1
- 1
.gitignore View File

@ -17,4 +17,4 @@ install_manifest.txt
compile_commands.json
CTestTestfile.cmake
_deps
cmake-build-debug
cmake-build-*

+ 6
- 4
CMakeLists.txt View File

@ -18,8 +18,7 @@ source_group(TREE ${CMAKE_CURRENT_SOURCE_DIR} FILES ${SOURCE_FILES} ${HEADER_FIL
pybind11_add_module(python_rgbcx
${SOURCE_FILES}
${HEADER_FILES}
${PYTHON_FILES}
)
${PYTHON_FILES})
add_executable(test_rgbcx
${SOURCE_FILES}
@ -27,5 +26,8 @@ add_executable(test_rgbcx
${TEST_FILES})
# Set module features, like C/C++ standards
target_compile_features(python_rgbcx PUBLIC cxx_std_17 c_std_11)
target_compile_features(test_rgbcx PUBLIC cxx_std_17 c_std_11)
target_compile_features(python_rgbcx PUBLIC cxx_std_20 c_std_11)
target_compile_features(test_rgbcx PUBLIC cxx_std_20 c_std_11)
set_property(TARGET python_rgbcx test_rgbcx PROPERTY INTERPROCEDURAL_OPTIMIZATION True) #enable FLTO if available
set_property(TARGET python_rgbcx test_rgbcx PROPERTY OSX_ARCHITECTURES_RELEASE x86_64 arm64) #Mach-O fat binary for arm and x86

+ 1
- 31
src/blocks.cpp View File

@ -18,38 +18,8 @@
*/
#include "blocks.h"
#include "color.h"
#include <algorithm>
#include <cassert>
// region Color32 implementation
Color32::Color32(uint32_t vr, uint32_t vg, uint32_t vb, uint32_t va) { set(vr, vg, vb, va); }
uint8_t Color32::operator[](uint32_t idx) const {
assert(idx < 4);
return C[idx];
}
uint8_t &Color32::operator[](uint32_t idx) {
assert(idx < 4);
return C[idx];
}
void Color32::set(uint8_t vr, uint8_t vg, uint8_t vb, uint8_t va) {
C[0] = vr;
C[1] = vg;
C[2] = vb;
C[3] = va;
}
void Color32::set(const Color32 &other) {
C[0] = other.C[0];
C[1] = other.C[1];
C[2] = other.C[2];
}
Color32 Color32::min(const Color32 &a, const Color32 &b) {
return Color32(std::min(a[0], b[0]), std::min(a[1], b[1]), std::min(a[2], b[2]), std::min(a[3], b[3]));
}
Color32 Color32::max(const Color32 &a, const Color32 &b) {
return Color32(std::max(a[0], b[0]), std::max(a[1], b[1]), std::max(a[2], b[2]), std::max(a[3], b[3]));
}
// endregion

+ 80
- 132
src/blocks.h View File

@ -19,172 +19,120 @@
#pragma once
#include "util.h"
#include <array>
#include <cassert>
#include <cstdint>
#include <cstdlib>
constexpr inline uint8_t DXT1SelectorBits = 2U;
#include "color.h"
#include "util.h"
#pragma pack(push, 1)
struct Color32 {
union {
struct {
uint8_t R;
uint8_t G;
uint8_t B;
uint8_t A;
};
uint8_t C[4];
};
Color32() {}
Color32(uint32_t vr, uint32_t vg, uint32_t vb, uint32_t va);
void set(uint8_t vr, uint8_t vg, uint8_t vb, uint8_t va);
void set(const Color32 &other);
uint8_t operator[](uint32_t idx) const;
uint8_t &operator[](uint32_t idx);
bool operator==(const Color32 &rhs) const {
return R == rhs.R && G == rhs.G && B == rhs.B && A == rhs.A;
class BC1Block {
public:
uint16_t GetLowColor() const { return _low_color[0] | _low_color[1] << 8U; }
uint16_t GetHighColor() const { return _high_color[0] | _high_color[1] << 8U; }
Color32 GetLowColor32() const { return Color32::unpack565(GetLowColor()); }
Color32 GetHighColor32() const { return Color32::unpack565(GetHighColor()); }
bool Is3Color() const { return GetLowColor() <= GetHighColor(); }
void SetLowColor(uint16_t c) {
_low_color[0] = c & 0xFF;
_low_color[1] = (c >> 8) & 0xFF;
}
static Color32 min(const Color32 &a, const Color32 &b);
static Color32 max(const Color32 &a, const Color32 &b);
};
struct BC1Block {
constexpr static inline size_t EndpointSize = 2;
constexpr static inline size_t SelectorSize = 4;
constexpr static inline uint8_t SelectorBits = 2;
constexpr static inline uint8_t SelectorValues = 1 << SelectorBits;
constexpr static inline uint8_t SelectorMask = SelectorValues - 1;
uint8_t LowColor[EndpointSize];
uint8_t HighColor[EndpointSize];
uint8_t Selectors[SelectorSize];
inline uint32_t get_low_color() const { return LowColor[0] | (LowColor[1] << 8U); }
inline uint32_t get_high_color() const { return HighColor[0] | (HighColor[1] << 8U); }
inline bool is_3color() const { return get_low_color() <= get_high_color(); }
inline void set_low_color(uint16_t c) {
LowColor[0] = static_cast<uint8_t>(c & 0xFF);
LowColor[1] = static_cast<uint8_t>((c >> 8) & 0xFF);
}
inline void set_high_color(uint16_t c) {
HighColor[0] = static_cast<uint8_t>(c & 0xFF);
HighColor[1] = static_cast<uint8_t>((c >> 8) & 0xFF);
void SetHighColor(uint16_t c) {
_high_color[0] = c & 0xFF;
_high_color[1] = (c >> 8) & 0xFF;
}
inline uint32_t get_selector(uint32_t x, uint32_t y) const {
uint32_t GetSelector(uint32_t x, uint32_t y) const {
assert((x < 4U) && (y < 4U));
return (Selectors[y] >> (x * SelectorBits)) & SelectorMask;
return (selectors[y] >> (x * SelectorBits)) & SelectorMask;
}
inline void set_selector(uint32_t x, uint32_t y, uint32_t val) {
void SetSelector(uint32_t x, uint32_t y, uint32_t val) {
assert((x < 4U) && (y < 4U) && (val < 4U));
Selectors[y] &= (~(SelectorMask << (x * SelectorBits)));
Selectors[y] |= (val << (x * DXT1SelectorBits));
selectors[y] &= (~(SelectorMask << (x * SelectorBits)));
selectors[y] |= (val << (x * SelectorBits));
}
static inline uint16_t pack_color(const Color32 &color, bool scaled, uint32_t bias = 127U) {
uint32_t r = color.R, g = color.G, b = color.B;
if (scaled) {
r = (r * 31U + bias) / 255U;
g = (g * 63U + bias) / 255U;
b = (b * 31U + bias) / 255U;
}
return static_cast<uint16_t>(minimum(b, 31U) | (minimum(g, 63U) << 5U) | (minimum(r, 31U) << 11U));
}
static inline uint16_t pack_unscaled_color(uint32_t r, uint32_t g, uint32_t b) { return static_cast<uint16_t>(b | (g << 5U) | (r << 11U)); }
static inline void unpack_color(uint32_t c, uint32_t &r, uint32_t &g, uint32_t &b) {
r = (c >> 11) & 31;
g = (c >> 5) & 63;
b = c & 31;
r = (r << 3) | (r >> 2);
g = (g << 2) | (g >> 4);
b = (b << 3) | (b >> 2);
}
static inline void unpack_color_unscaled(uint32_t c, uint32_t &r, uint32_t &g, uint32_t &b) {
r = (c >> 11) & 31;
g = (c >> 5) & 63;
b = c & 31;
}
};
struct BC4Block {
constexpr static inline size_t EndpointSize = 1;
constexpr static inline size_t SelectorSize = 6;
constexpr static inline uint8_t SelectorBits = 3;
constexpr static inline size_t EndpointSize = 2;
constexpr static inline size_t SelectorSize = 4;
constexpr static inline uint8_t SelectorBits = 2;
constexpr static inline uint8_t SelectorValues = 1 << SelectorBits;
constexpr static inline uint8_t SelectorMask = SelectorValues - 1;
uint8_t LowAlpha;
uint8_t HighAlpha;
uint8_t Selectors[SelectorSize];
private:
std::array<uint8_t, EndpointSize> _low_color;
std::array<uint8_t, EndpointSize> _high_color;
public:
std::array<uint8_t, 4> selectors;
};
inline uint32_t get_low_alpha() const { return LowAlpha; }
inline uint32_t get_high_alpha() const { return HighAlpha; }
inline bool is_alpha6_block() const { return get_low_alpha() <= get_high_alpha(); }
class BC4Block {
public:
inline uint32_t GetLowAlpha() const { return low_alpha; }
inline uint32_t GetHighAlpha() const { return high_alpha; }
inline bool Is6Alpha() const { return GetLowAlpha() <= GetHighAlpha(); }
inline uint64_t get_selector_bits() const {
return ((uint64_t)((uint32_t)Selectors[0] | ((uint32_t)Selectors[1] << 8U) | ((uint32_t)Selectors[2] << 16U) |
((uint32_t)Selectors[3] << 24U))) |
(((uint64_t)Selectors[4]) << 32U) | (((uint64_t)Selectors[5]) << 40U);
inline uint64_t GetSelectorBits() const {
return ((uint64_t)((uint32_t)selectors[0] | ((uint32_t)selectors[1] << 8U) | ((uint32_t)selectors[2] << 16U) | ((uint32_t)selectors[3] << 24U))) |
(((uint64_t)selectors[4]) << 32U) | (((uint64_t)selectors[5]) << 40U);
}
inline uint32_t get_selector(uint32_t x, uint32_t y, uint64_t selector_bits) const {
inline uint32_t GetSelector(uint32_t x, uint32_t y, uint64_t selector_bits) const {
assert((x < 4U) && (y < 4U));
return (selector_bits >> (((y * 4) + x) * SelectorBits)) & (SelectorMask);
}
static inline uint32_t get_block_values6(uint8_t *pDst, uint32_t l, uint32_t h) {
pDst[0] = static_cast<uint8_t>(l);
pDst[1] = static_cast<uint8_t>(h);
pDst[2] = static_cast<uint8_t>((l * 4 + h) / 5);
pDst[3] = static_cast<uint8_t>((l * 3 + h * 2) / 5);
pDst[4] = static_cast<uint8_t>((l * 2 + h * 3) / 5);
pDst[5] = static_cast<uint8_t>((l + h * 4) / 5);
pDst[6] = 0;
pDst[7] = 255;
return 6;
static inline std::array<uint8_t, 8> GetBlockValues6(uint32_t l, uint32_t h) {
return {static_cast<uint8_t>(l),
static_cast<uint8_t>(h),
static_cast<uint8_t>((l * 4 + h) / 5),
static_cast<uint8_t>((l * 3 + h * 2) / 5),
static_cast<uint8_t>((l * 2 + h * 3) / 5),
static_cast<uint8_t>((l + h * 4) / 5),
0,
255};
}
static inline uint32_t get_block_values8(uint8_t *pDst, uint32_t l, uint32_t h) {
pDst[0] = static_cast<uint8_t>(l);
pDst[1] = static_cast<uint8_t>(h);
pDst[2] = static_cast<uint8_t>((l * 6 + h) / 7);
pDst[3] = static_cast<uint8_t>((l * 5 + h * 2) / 7);
pDst[4] = static_cast<uint8_t>((l * 4 + h * 3) / 7);
pDst[5] = static_cast<uint8_t>((l * 3 + h * 4) / 7);
pDst[6] = static_cast<uint8_t>((l * 2 + h * 5) / 7);
pDst[7] = static_cast<uint8_t>((l + h * 6) / 7);
return 8;
static inline std::array<uint8_t, 8> GetBlockValues8(uint32_t l, uint32_t h) {
return {static_cast<uint8_t>(l),
static_cast<uint8_t>(h),
static_cast<uint8_t>((l * 6 + h) / 7),
static_cast<uint8_t>((l * 5 + h * 2) / 7),
static_cast<uint8_t>((l * 4 + h * 3) / 7),
static_cast<uint8_t>((l * 3 + h * 4) / 7),
static_cast<uint8_t>((l * 2 + h * 5) / 7),
static_cast<uint8_t>((l + h * 6) / 7)};
}
static inline uint32_t get_block_values(uint8_t *pDst, uint32_t l, uint32_t h) {
static inline std::array<uint8_t, 8> GetBlockValues(uint32_t l, uint32_t h) {
if (l > h)
return get_block_values8(pDst, l, h);
return GetBlockValues8(l, h);
else
return get_block_values6(pDst, l, h);
return GetBlockValues6(l, h);
}
};
struct BC3Block {
BC4Block AlphaBlock;
BC1Block ColorBlock;
constexpr static inline size_t EndpointSize = 1;
constexpr static inline size_t SelectorSize = 6;
constexpr static inline uint8_t SelectorBits = 3;
constexpr static inline uint8_t SelectorValues = 1 << SelectorBits;
constexpr static inline uint8_t SelectorMask = SelectorValues - 1;
uint8_t low_alpha;
uint8_t high_alpha;
uint8_t selectors[SelectorSize];
};
struct BC5Block {
BC4Block RBlock;
BC4Block GBlock;
class BC3Block {
public:
BC4Block alpha_block;
BC1Block color_block;
};
class BC5Block {
public:
BC4Block r_block;
BC4Block g_block;
};
#pragma pack(pop)

+ 76
- 0
src/color.cpp View File

@ -0,0 +1,76 @@
/* Python-rgbcx Texture Compression Library
Copyright (C) 2021 Andrew Cassidy <drewcassidy@me.com>
Partially derived from rgbcx.h written by Richard Geldreich <richgel99@gmail.com>
and licenced under the public domain
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU Affero General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU Affero General Public License for more details.
You should have received a copy of the GNU Affero General Public License
along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#include "color.h"
#include <algorithm>
#include <cassert>
// region Color32 implementation
Color32::Color32() { set(0, 0, 0, 0xFF); }
Color32::Color32(uint8_t R, uint8_t G, uint8_t B, uint8_t A) { set(R, G, B, A); }
uint16_t Color32::pack565Unscaled(uint16_t R, uint16_t G, uint16_t B) { return B | (G << 5) | (R << 11); }
uint16_t Color32::pack565(uint16_t R, uint16_t G, uint16_t B) { return pack565Unscaled(scale8To5(R), scale8To6(G), scale8To5(B)); }
Color32 Color32::unpack565(uint16_t Packed) {
uint8_t R = scale5To8((Packed >> 11) & 0x1F);
uint8_t G = scale6To8((Packed >> 5) & 0x3F);
uint8_t B = scale5To8(Packed & 0x1F);
return Color32(R, G, B);
}
uint8_t Color32::operator[](uint32_t Index) const {
assert(Index < 4);
return C[Index];
}
uint8_t &Color32::operator[](uint32_t Index) {
assert(Index < 4);
return C[Index];
}
void Color32::set(uint8_t R, uint8_t G, uint8_t B, uint8_t A) {
this->R = R;
this->G = G;
this->B = B;
this->A = A;
}
void Color32::set(const Color32 &Other) {
this->R = Other.R;
this->G = Other.G;
this->B = Other.B;
this->A = Other.A;
}
Color32 Color32::min(const Color32 &a, const Color32 &b) {
return Color32(std::min(a[0], b[0]), std::min(a[1], b[1]), std::min(a[2], b[2]), std::min(a[3], b[3]));
}
Color32 Color32::max(const Color32 &a, const Color32 &b) {
return Color32(std::max(a[0], b[0]), std::max(a[1], b[1]), std::max(a[2], b[2]), std::max(a[3], b[3]));
}
uint16_t Color32::pack565() { return pack565(R, G, B); }
uint16_t Color32::pack565Unscaled() { return pack565Unscaled(R, G, B); }
// endregion

+ 64
- 0
src/color.h View File

@ -0,0 +1,64 @@
/* Python-rgbcx Texture Compression Library
Copyright (C) 2021 Andrew Cassidy <drewcassidy@me.com>
Partially derived from rgbcx.h written by Richard Geldreich <richgel99@gmail.com>
and licenced under the public domain
This program is free software: you can redistribute it and/or modify
it under the terms of the GNU Affero General Public License as published by
the Free Software Foundation, either version 3 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU Affero General Public License for more details.
You should have received a copy of the GNU Affero General Public License
along with this program. If not, see <http://www.gnu.org/licenses/>.
*/
#pragma once
#include "util.h"
#include <cassert>
#include <cstdint>
#include <cstdlib>
#pragma pack(push, 1)
class Color32 {
public:
union {
struct {
uint8_t R;
uint8_t G;
uint8_t B;
uint8_t A;
};
uint8_t C[4];
};
Color32();
Color32(uint8_t R, uint8_t G, uint8_t B, uint8_t A = 0xFF);
static uint16_t pack565Unscaled(uint16_t R, uint16_t G, uint16_t B);
static uint16_t pack565(uint16_t R, uint16_t G, uint16_t B);
static Color32 unpack565(uint16_t Packed);
bool operator==(const Color32 &Rhs) const { return R == Rhs.R && G == Rhs.G && B == Rhs.B && A == Rhs.A; }
uint8_t operator[](uint32_t Index) const;
uint8_t &operator[](uint32_t Index);
uint16_t pack565();
uint16_t pack565Unscaled();
static Color32 min(const Color32 &A, const Color32 &B);
static Color32 max(const Color32 &A, const Color32 &B);
void set(uint8_t R, uint8_t G, uint8_t B, uint8_t A);
void set(const Color32 &Other);
};
#pragma pack(pop)

+ 89
- 87
src/rgbcx.cpp View File

@ -1,16 +1,16 @@
// rgbcx.h v1.12
// High-performance scalar BC1-5 encoders. Public Domain or MIT license (you choose - see below), written by Richard Geldreich 2020 <richgel99@gmail.com>.
#include "rgbcx.h"
#include "blocks.h"
#include "color.h"
#include "tables.h"
#include "util.h"
#include <algorithm>
#include <cassert>
#include <climits>
#include <cstdlib>
#include <cstring>
#include <cmath>
#include "util.h"
#include "tables.h"
#include "blocks.h"
#include "rgbcx.h"
namespace rgbcx {
@ -159,7 +159,7 @@ static inline int interp_half_5_6_amd(int c0, int c1) {
}
static inline int interp_5(int v0, int v1, int c0, int c1, bc1_approx_mode mode) {
// assert(scale_5_to_8(v0) == c0 && scale_5_to_8(v1) == c1);
// assert(scale_5_to_8(v0) == c0 && scale5To8(v1) == c1);
switch (mode) {
case bc1_approx_mode::cBC1NVidia:
return interp_5_nv(v0, v1);
@ -176,7 +176,7 @@ static inline int interp_5(int v0, int v1, int c0, int c1, bc1_approx_mode mode)
static inline int interp_6(int v0, int v1, int c0, int c1, bc1_approx_mode mode) {
(void)v0;
(void)v1;
// assert(scale_6_to_8(v0) == c0 && scale_6_to_8(v1) == c1);
// assert(scale_6_to_8(v0) == c0 && scale6To8(v1) == c1);
switch (mode) {
case bc1_approx_mode::cBC1NVidia:
return interp_6_nv(c0, c1);
@ -191,7 +191,7 @@ static inline int interp_6(int v0, int v1, int c0, int c1, bc1_approx_mode mode)
}
static inline int interp_half_5(int v0, int v1, int c0, int c1, bc1_approx_mode mode) {
assert(scale_5_to_8(v0) == c0 && scale_5_to_8(v1) == c1);
assert(scale5To8(v0) == c0 && scale5To8(v1) == c1);
switch (mode) {
case bc1_approx_mode::cBC1NVidia:
return interp_half_5_nv(v0, v1);
@ -207,7 +207,7 @@ static inline int interp_half_5(int v0, int v1, int c0, int c1, bc1_approx_mode
static inline int interp_half_6(int v0, int v1, int c0, int c1, bc1_approx_mode mode) {
(void)v0;
(void)v1;
assert(scale_6_to_8(v0) == c0 && scale_6_to_8(v1) == c1);
assert(scale6To8(v0) == c0 && scale6To8(v1) == c1);
switch (mode) {
case bc1_approx_mode::cBC1NVidia:
return interp_half_6_nv(c0, c1);
@ -434,12 +434,12 @@ void encode_bc1_solid_block(void *pDst, uint32_t fr, uint32_t fg, uint32_t fb, b
}
}
pDst_block->set_low_color(static_cast<uint16_t>(max16));
pDst_block->set_high_color(static_cast<uint16_t>(min16));
pDst_block->Selectors[0] = static_cast<uint8_t>(mask);
pDst_block->Selectors[1] = static_cast<uint8_t>(mask);
pDst_block->Selectors[2] = static_cast<uint8_t>(mask);
pDst_block->Selectors[3] = static_cast<uint8_t>(mask);
pDst_block->SetLowColor(static_cast<uint16_t>(max16));
pDst_block->SetHighColor(static_cast<uint16_t>(min16));
pDst_block->selectors[0] = static_cast<uint8_t>(mask);
pDst_block->selectors[1] = static_cast<uint8_t>(mask);
pDst_block->selectors[2] = static_cast<uint8_t>(mask);
pDst_block->selectors[3] = static_cast<uint8_t>(mask);
}
static const float g_midpoint5[32] = {.015686f, .047059f, .078431f, .111765f, .145098f, .176471f, .207843f, .241176f, .274510f, .305882f, .337255f,
@ -944,8 +944,8 @@ static inline void precise_round_565_noscale(vec3F xl, vec3F xh, int &trial_lr,
}
static inline void bc1_encode4(BC1Block *pDst_block, int lr, int lg, int lb, int hr, int hg, int hb, const uint8_t sels[16]) {
uint32_t lc16 = BC1Block::pack_unscaled_color(lr, lg, lb);
uint32_t hc16 = BC1Block::pack_unscaled_color(hr, hg, hb);
uint16_t lc16 = Color32::pack565Unscaled(lr, lg, lb);
uint16_t hc16 = Color32::pack565Unscaled(hr, hg, hb);
// Always forbid 3 color blocks
if (lc16 == hc16) {
@ -964,13 +964,13 @@ static inline void bc1_encode4(BC1Block *pDst_block, int lr, int lg, int lb, int
}
assert(lc16 > hc16);
pDst_block->set_low_color(static_cast<uint16_t>(lc16));
pDst_block->set_high_color(static_cast<uint16_t>(hc16));
pDst_block->SetLowColor(static_cast<uint16_t>(lc16));
pDst_block->SetHighColor(static_cast<uint16_t>(hc16));
pDst_block->Selectors[0] = mask;
pDst_block->Selectors[1] = mask;
pDst_block->Selectors[2] = mask;
pDst_block->Selectors[3] = mask;
pDst_block->selectors[0] = mask;
pDst_block->selectors[1] = mask;
pDst_block->selectors[2] = mask;
pDst_block->selectors[3] = mask;
} else {
uint8_t invert_mask = 0;
if (lc16 < hc16) {
@ -979,24 +979,25 @@ static inline void bc1_encode4(BC1Block *pDst_block, int lr, int lg, int lb, int
}
assert(lc16 > hc16);
pDst_block->set_low_color((uint16_t)lc16);
pDst_block->set_high_color((uint16_t)hc16);
pDst_block->SetLowColor((uint16_t)lc16);
pDst_block->SetHighColor((uint16_t)hc16);
uint32_t packed_sels = 0;
static const uint8_t s_sel_trans[4] = {0, 2, 3, 1};
for (uint32_t i = 0; i < 16; i++)
packed_sels |= ((uint32_t)s_sel_trans[sels[i]] << (i * 2));
pDst_block->Selectors[0] = (uint8_t)packed_sels ^ invert_mask;
pDst_block->Selectors[1] = (uint8_t)(packed_sels >> 8) ^ invert_mask;
pDst_block->Selectors[2] = (uint8_t)(packed_sels >> 16) ^ invert_mask;
pDst_block->Selectors[3] = (uint8_t)(packed_sels >> 24) ^ invert_mask;
// todo: make this less silly to prevent packing and unpacking
pDst_block->selectors[0] = (uint8_t)packed_sels ^ invert_mask;
pDst_block->selectors[1] = (uint8_t)(packed_sels >> 8) ^ invert_mask;
pDst_block->selectors[2] = (uint8_t)(packed_sels >> 16) ^ invert_mask;
pDst_block->selectors[3] = (uint8_t)(packed_sels >> 24) ^ invert_mask;
}
}
static inline void bc1_encode3(BC1Block *pDst_block, int lr, int lg, int lb, int hr, int hg, int hb, const uint8_t sels[16]) {
uint32_t lc16 = BC1Block::pack_unscaled_color(lr, lg, lb);
uint32_t hc16 = BC1Block::pack_unscaled_color(hr, hg, hb);
uint16_t lc16 = Color32::pack565Unscaled(lr, lg, lb);
uint16_t hc16 = Color32::pack565Unscaled(hr, hg, hb);
bool invert_flag = false;
if (lc16 > hc16) {
@ -1006,8 +1007,8 @@ static inline void bc1_encode3(BC1Block *pDst_block, int lr, int lg, int lb, int
assert(lc16 <= hc16);
pDst_block->set_low_color((uint16_t)lc16);
pDst_block->set_high_color((uint16_t)hc16);
pDst_block->SetLowColor((uint16_t)lc16);
pDst_block->SetHighColor((uint16_t)hc16);
uint32_t packed_sels = 0;
@ -1021,10 +1022,11 @@ static inline void bc1_encode3(BC1Block *pDst_block, int lr, int lg, int lb, int
packed_sels |= ((uint32_t)sels[i] << (i * 2));
}
pDst_block->Selectors[0] = (uint8_t)packed_sels;
pDst_block->Selectors[1] = (uint8_t)(packed_sels >> 8);
pDst_block->Selectors[2] = (uint8_t)(packed_sels >> 16);
pDst_block->Selectors[3] = (uint8_t)(packed_sels >> 24);
// todo: make this less silly to prevent packing and unpacking
pDst_block->selectors[0] = (uint8_t)packed_sels;
pDst_block->selectors[1] = (uint8_t)(packed_sels >> 8);
pDst_block->selectors[2] = (uint8_t)(packed_sels >> 16);
pDst_block->selectors[3] = (uint8_t)(packed_sels >> 24);
}
struct bc1_encode_results {
@ -1138,13 +1140,13 @@ static bool try_3color_block_useblack(const Color32 *pSrc_pixels, uint32_t flags
}
}
int lr = scale_8_to_5(pSrc_pixels[low_c].R);
int lg = scale_8_to_6(pSrc_pixels[low_c].G);
int lb = scale_8_to_5(pSrc_pixels[low_c].B);
int lr = scale8To5(pSrc_pixels[low_c].R);
int lg = scale8To6(pSrc_pixels[low_c].G);
int lb = scale8To5(pSrc_pixels[low_c].B);
int hr = scale_8_to_5(pSrc_pixels[high_c].R);
int hg = scale_8_to_6(pSrc_pixels[high_c].G);
int hb = scale_8_to_5(pSrc_pixels[high_c].B);
int hr = scale8To5(pSrc_pixels[high_c].R);
int hg = scale8To6(pSrc_pixels[high_c].G);
int hb = scale8To5(pSrc_pixels[high_c].B);
uint8_t trial_sels[16];
uint32_t trial_err = bc1_find_sels3_fullerr(true, pSrc_pixels, lr, lg, lb, hr, hg, hb, trial_sels, UINT32_MAX);
@ -1491,14 +1493,14 @@ static inline void encode_bc1_pick_initial(const Color32 *pSrc_pixels, uint32_t
// Grayscale blocks are a common enough case to specialize.
if ((max_r - min_r) < 2) {
lr = lb = hr = hb = scale_8_to_5(fr);
lg = hg = scale_8_to_6(fr);
lr = lb = hr = hb = scale8To5(fr);
lg = hg = scale8To6(fr);
} else {
lr = lb = scale_8_to_5(min_r);
lg = scale_8_to_6(min_r);
lr = lb = scale8To5(min_r);
lg = scale8To6(min_r);
hr = hb = scale_8_to_5(max_r);
hg = scale_8_to_6(max_r);
hr = hb = scale8To5(max_r);
hg = scale8To6(max_r);
}
} else if (flags & cEncodeBC1Use2DLS) {
// 2D Least Squares approach from Humus's example, with added inset and optimal rounding.
@ -1721,13 +1723,13 @@ static inline void encode_bc1_pick_initial(const Color32 *pSrc_pixels, uint32_t
if (icov_yz < 0)
std::swap(y0, y1);
lr = scale_8_to_5(x0);
lg = scale_8_to_6(y0);
lb = scale_8_to_5(min_b);
lr = scale8To5(x0);
lg = scale8To6(y0);
lb = scale8To5(min_b);
hr = scale_8_to_5(x1);
hg = scale_8_to_6(y1);
hb = scale_8_to_5(max_b);
hr = scale8To5(x1);
hg = scale8To6(y1);
hb = scale8To5(max_b);
} else {
// Select 2 colors along the principle axis. (There must be a faster/simpler way.)
uint32_t low_c = 0, high_c = 0;
@ -1806,13 +1808,13 @@ static inline void encode_bc1_pick_initial(const Color32 *pSrc_pixels, uint32_t
low_c = low_dot & 15;
high_c = high_dot & 15;
lr = scale_8_to_5(pSrc_pixels[low_c].R);
lg = scale_8_to_6(pSrc_pixels[low_c].G);
lb = scale_8_to_5(pSrc_pixels[low_c].B);
lr = scale8To5(pSrc_pixels[low_c].R);
lg = scale8To6(pSrc_pixels[low_c].G);
lb = scale8To5(pSrc_pixels[low_c].B);
hr = scale_8_to_5(pSrc_pixels[high_c].R);
hg = scale_8_to_6(pSrc_pixels[high_c].G);
hb = scale_8_to_5(pSrc_pixels[high_c].B);
hr = scale8To5(pSrc_pixels[high_c].R);
hg = scale8To6(pSrc_pixels[high_c].G);
hb = scale8To5(pSrc_pixels[high_c].B);
}
}
@ -2349,6 +2351,7 @@ void encode_bc4(void *pDst, const uint8_t *pPixels, uint32_t stride) {
const uint64_t f = a0 | a1 | a2 | a3;
// TODO: make this less silly by using the BC4Block class
pDst_bytes[2] = (uint8_t)f;
pDst_bytes[3] = (uint8_t)(f >> 8U);
pDst_bytes[4] = (uint8_t)(f >> 16U);
@ -2357,28 +2360,28 @@ void encode_bc4(void *pDst, const uint8_t *pPixels, uint32_t stride) {
pDst_bytes[7] = (uint8_t)(f >> 40U);
}
void encode_bc3(void *pDst, const uint8_t *pPixels, uint32_t flags, uint32_t total_orderings_to_try) {
void encode_bc3(BC3Block *pDst, const uint8_t *pPixels, uint32_t flags, uint32_t total_orderings_to_try) {
assert(g_initialized);
// 3-color blocks are not allowed with BC3 (on most GPU's).
flags &= ~(cEncodeBC1Use3ColorBlocksForBlackPixels | cEncodeBC1Use3ColorBlocks);
encode_bc4(pDst, pPixels + 3, 4);
encode_bc1(static_cast<uint8_t *>(pDst) + 8, pPixels, flags, total_orderings_to_try);
encode_bc4(&pDst->alpha_block, pPixels + 3, 4);
encode_bc1(&pDst->color_block, pPixels, flags, total_orderings_to_try);
}
void encode_bc3(uint32_t level, void *pDst, const uint8_t *pPixels) {
void encode_bc3(uint32_t level, BC3Block *pDst, const uint8_t *pPixels) {
assert(g_initialized);
encode_bc4(pDst, pPixels + 3, 4);
encode_bc1(level, static_cast<uint8_t *>(pDst) + 8, pPixels, false, false);
encode_bc4(&pDst->alpha_block, pPixels + 3, 4);
encode_bc1(level, &pDst->color_block, pPixels, false, false);
}
void encode_bc5(void *pDst, const uint8_t *pPixels, uint32_t chan0, uint32_t chan1, uint32_t stride) {
void encode_bc5(BC5Block *pDst, const uint8_t *pPixels, uint32_t chan0, uint32_t chan1, uint32_t stride) {
assert(g_initialized);
encode_bc4(pDst, pPixels + chan0, stride);
encode_bc4(static_cast<uint8_t *>(pDst) + 8, pPixels + chan1, stride);
encode_bc4(&pDst->r_block, pPixels + chan0, stride);
encode_bc4(&pDst->g_block, pPixels + chan1, stride);
}
// Returns true if the block uses 3 color punchthrough alpha mode.
@ -2390,8 +2393,8 @@ bool unpack_bc1(const void *pBlock_bits, void *pPixels, bool set_alpha, bc1_appr
const BC1Block *pBlock = static_cast<const BC1Block *>(pBlock_bits);
const uint32_t l = pBlock->get_low_color();
const uint32_t h = pBlock->get_high_color();
const uint32_t l = pBlock->GetLowColor();
const uint32_t h = pBlock->GetHighColor();
Color32 c[4];
@ -2454,17 +2457,17 @@ bool unpack_bc1(const void *pBlock_bits, void *pPixels, bool set_alpha, bc1_appr
if (set_alpha) {
for (uint32_t y = 0; y < 4; y++, pDst_pixels += 4) {
pDst_pixels[0] = c[pBlock->get_selector(0, y)];
pDst_pixels[1] = c[pBlock->get_selector(1, y)];
pDst_pixels[2] = c[pBlock->get_selector(2, y)];
pDst_pixels[3] = c[pBlock->get_selector(3, y)];
pDst_pixels[0] = c[pBlock->GetSelector(0, y)];
pDst_pixels[1] = c[pBlock->GetSelector(1, y)];
pDst_pixels[2] = c[pBlock->GetSelector(2, y)];
pDst_pixels[3] = c[pBlock->GetSelector(3, y)];
}
} else {
for (uint32_t y = 0; y < 4; y++, pDst_pixels += 4) {
pDst_pixels[0].set(c[pBlock->get_selector(0, y)]);
pDst_pixels[1].set(c[pBlock->get_selector(1, y)]);
pDst_pixels[2].set(c[pBlock->get_selector(2, y)]);
pDst_pixels[3].set(c[pBlock->get_selector(3, y)]);
pDst_pixels[0].set(c[pBlock->GetSelector(0, y)]);
pDst_pixels[1].set(c[pBlock->GetSelector(1, y)]);
pDst_pixels[2].set(c[pBlock->GetSelector(2, y)]);
pDst_pixels[3].set(c[pBlock->GetSelector(3, y)]);
}
}
@ -2476,16 +2479,15 @@ void unpack_bc4(const void *pBlock_bits, uint8_t *pPixels, uint32_t stride) {
const BC4Block *pBlock = static_cast<const BC4Block *>(pBlock_bits);
uint8_t sel_values[8];
BC4Block::get_block_values(sel_values, pBlock->get_low_alpha(), pBlock->get_high_alpha());
auto sel_values = BC4Block::GetBlockValues(pBlock->GetLowAlpha(), pBlock->GetHighAlpha());
const uint64_t selector_bits = pBlock->get_selector_bits();
const uint64_t selector_bits = pBlock->GetSelectorBits();
for (uint32_t y = 0; y < 4; y++, pPixels += (stride * 4U)) {
pPixels[0] = sel_values[pBlock->get_selector(0, y, selector_bits)];
pPixels[stride * 1] = sel_values[pBlock->get_selector(1, y, selector_bits)];
pPixels[stride * 2] = sel_values[pBlock->get_selector(2, y, selector_bits)];
pPixels[stride * 3] = sel_values[pBlock->get_selector(3, y, selector_bits)];
pPixels[0] = sel_values[pBlock->GetSelector(0, y, selector_bits)];
pPixels[stride * 1] = sel_values[pBlock->GetSelector(1, y, selector_bits)];
pPixels[stride * 2] = sel_values[pBlock->GetSelector(2, y, selector_bits)];
pPixels[stride * 3] = sel_values[pBlock->GetSelector(3, y, selector_bits)];
}
}


+ 5
- 10
src/util.h View File

@ -23,17 +23,17 @@
static inline uint32_t iabs(int32_t i) { return (i < 0) ? static_cast<uint32_t>(-i) : static_cast<uint32_t>(i); }
static inline uint64_t iabs(int64_t i) { return (i < 0) ? static_cast<uint64_t>(-i) : static_cast<uint64_t>(i); }
static inline uint8_t scale_8_to_5(uint32_t v) {
static inline uint8_t scale8To5(uint32_t v) {
v = v * 31 + 128;
return (uint8_t)((v + (v >> 8)) >> 8);
}
static inline uint8_t scale_8_to_6(uint32_t v) {
static inline uint8_t scale8To6(uint32_t v) {
v = v * 63 + 128;
return (uint8_t)((v + (v >> 8)) >> 8);
}
static inline int scale_5_to_8(int v) { return (v << 3) | (v >> 2); }
static inline int scale_6_to_8(int v) { return (v << 2) | (v >> 4); }
static inline int scale5To8(int v) { return (v << 3) | (v >> 2); }
static inline int scale6To8(int v) { return (v << 2) | (v >> 4); }
template <typename S> inline S maximum(S a, S b) { return (a > b) ? a : b; }
template <typename S> inline S maximum(S a, S b, S c) { return maximum(maximum(a, b), c); }
@ -55,7 +55,7 @@ static inline float clampf(float value, float low, float high) {
static inline uint8_t clamp255(int32_t i) { return (uint8_t)((i & 0xFFFFFF00U) ? (~(i >> 31)) : i); }
template <typename S> inline S clamp(S value, S low, S high) { return (value < low) ? low : ((value > high) ? high : value); }
static inline int32_t clampi(int32_t value, int32_t low, int32_t high) {
static inline int32_t clampi(int32_t value, int32_t low, int32_t high) {
if (value < low)
value = low;
else if (value > high)
@ -67,8 +67,3 @@ static inline int squarei(int a) { return a * a; }
static inline int absi(int a) { return (a < 0) ? -a : a; }
template <typename F> inline F lerp(F a, F b, F s) { return a + (b - a) * s; }

Loading…
Cancel
Save