bsnes/nall/image/core.hpp

173 lines
4.3 KiB
C++

#pragma once
namespace nall {
image::image(const image& source) {
operator=(source);
}
image::image(image&& source) {
operator=(forward<image>(source));
}
image::image(bool endian, unsigned depth, uint64_t alphaMask, uint64_t redMask, uint64_t greenMask, uint64_t blueMask) {
_endian = endian;
_depth = depth;
_alpha = {alphaMask, bitDepth(alphaMask), bitShift(alphaMask)};
_red = {redMask, bitDepth(redMask), bitShift(redMask )};
_green = {greenMask, bitDepth(greenMask), bitShift(greenMask)};
_blue = {blueMask, bitDepth(blueMask), bitShift(blueMask )};
}
image::image(const string& filename) {
load(filename);
}
image::image(const void* data_, uint size) {
auto data = (const uint8_t*)data_;
if(size < 4);
else if(data[0] == 'B' && data[1] == 'M') loadBMP(data, size);
else if(data[1] == 'P' && data[2] == 'N' && data[3] == 'G') loadPNG(data, size);
}
image::image(const vector<uint8_t>& buffer) : image(buffer.data(), buffer.size()) {
}
template<typename Type, uint Size> image::image(Type (&Name)[Size]) : image(Name, Size) {
}
image::image() {
}
image::~image() {
free();
}
auto image::operator=(const image& source) -> image& {
if(this == &source) return *this;
free();
_width = source._width;
_height = source._height;
_endian = source._endian;
_depth = source._depth;
_alpha = source._alpha;
_red = source._red;
_green = source._green;
_blue = source._blue;
_data = allocate(_width, _height, stride());
memory::copy(_data, source._data, source.size());
return *this;
}
auto image::operator=(image&& source) -> image& {
if(this == &source) return *this;
free();
_width = source._width;
_height = source._height;
_endian = source._endian;
_depth = source._depth;
_alpha = source._alpha;
_red = source._red;
_green = source._green;
_blue = source._blue;
_data = source._data;
source._data = nullptr;
return *this;
}
image::operator bool() const {
return _data && _width && _height;
}
auto image::operator==(const image& source) const -> bool {
if(_width != source._width) return false;
if(_height != source._height) return false;
if(_endian != source._endian) return false;
if(_depth != source._depth) return false;
if(_alpha != source._alpha) return false;
if(_red != source._red) return false;
if(_green != source._green) return false;
if(_blue != source._blue) return false;
return memory::compare(_data, source._data, size()) == 0;
}
auto image::operator!=(const image& source) const -> bool {
return !operator==(source);
}
auto image::read(const uint8_t* data) const -> uint64_t {
uint64_t result = 0;
if(_endian == 0) {
for(signed n = stride() - 1; n >= 0; n--) result = (result << 8) | data[n];
} else {
for(signed n = 0; n < stride(); n++) result = (result << 8) | data[n];
}
return result;
}
auto image::write(uint8_t* data, uint64_t value) const -> void {
if(_endian == 0) {
for(signed n = 0; n < stride(); n++) {
data[n] = value;
value >>= 8;
}
} else {
for(signed n = stride() - 1; n >= 0; n--) {
data[n] = value;
value >>= 8;
}
}
}
auto image::free() -> void {
if(_data) delete[] _data;
_data = nullptr;
}
auto image::load(const string& filename) -> bool {
if(loadBMP(filename) == true) return true;
if(loadPNG(filename) == true) return true;
return false;
}
auto image::allocate(unsigned width, unsigned height) -> void {
if(_data && _width == width && _height == height) return;
free();
_width = width;
_height = height;
_data = allocate(_width, _height, stride());
}
auto image::allocate(unsigned width, unsigned height, unsigned stride) -> uint8_t* {
//allocate 1x1 larger than requested; so that linear interpolation does not require bounds-checking
unsigned size = width * height * stride;
unsigned padding = width * stride + stride;
auto data = new uint8_t[size + padding];
memory::fill(data + size, padding);
return data;
}
//assumes image and data are in the same format; pitch is adapted to image
auto image::allocate(const void* data, uint pitch, uint width, uint height) -> void {
allocate(width, height);
for(uint y : range(height)) {
auto input = (const uint8_t*)data + y * pitch;
auto output = (uint8_t*)_data + y * this->pitch();
memory::copy(output, input, width * stride());
}
}
}