color_convert.cpp

// License: Do what you want. I don't care.

#include <algorithm>
#include <math.h>
#include <cstdio>
#include <cinttypes>
#include "lodepng.h"


// 0 = libretro GBA color, 1 = mGBA GBA color, 2 = libretro DS phat, 3 = libretro DS phat white.
#define PROFILE_IDX  (0)


typedef uint8_t  u8;
typedef uint16_t u16;
typedef uint32_t u32;
typedef uint64_t u64;

typedef int8_t  s8;
typedef int16_t s16;
typedef int32_t s32;
typedef int64_t s64;



// Compile with "g++ -std=c++17 -s -flto -O2 -fstrict-aliasing -ffunction-sections -Wall -Wextra -I./lodepng -Wl,--gc-sections ./lodepng/lodepng.cpp ./color_convert.cpp -lm -o ./color_convert"
int main(int argc, char const *argv[])
{
	if(argc != 3)
	{
		puts("Usage: color_convert input_png output_png");
		return 1;
	}

	unsigned char *buf;
	u32 width, height;
	u32 lpngErr = lodepng_decode32_file(&buf, &width, &height, argv[1]);
	if(lpngErr)
	{
		fprintf(stderr, "lodepng error: %s", lodepng_error_text(lpngErr));
		return 2;
	}

	if(width == 0 || width > 65535 || height == 0 || height > 65535)
	{
		fputs("Error: Input image too big.", stderr);
		free(buf);
		return 3;
	}

	u32 i = 0;
	do
	{
		// Normalize.
		float r = (float)buf[4 * i + 0] / 255;
		float g = (float)buf[4 * i + 1] / 255;
		float b = (float)buf[4 * i + 2] / 255;

		// Convert to linear gamma.
		static constexpr float targetGammaTable[] = {2.5f, 2.7f, 2.f, 2.f};
		r = powf(r, targetGammaTable[PROFILE_IDX]);
		g = powf(g, targetGammaTable[PROFILE_IDX]);
		b = powf(b, targetGammaTable[PROFILE_IDX]);

		// Luminance.
		static constexpr float luminanceTable[] = {0.93f, 0.99f, 1.f, 0.915f};
		r = std::clamp(r * luminanceTable[PROFILE_IDX], 0.f, 1.f);
		g = std::clamp(g * luminanceTable[PROFILE_IDX], 0.f, 1.f);
		b = std::clamp(b * luminanceTable[PROFILE_IDX], 0.f, 1.f);

		/*
		 *               Input
		 *                [r]
		 *                [g]
		 *                [b]
		 *
		 * Profile       Output
		 * [ r][gr][br]   [r]
		 * [rg][ g][bg]   [g]
		 * [rb][gb][ b]   [b]
		*/
		// Assuming alpha channel unused in original calculation.
#if (PROFILE_IDX == 0)
		// libretro GBA color (sRGB). Credits: hunterk and Pokefan531.
		// 0.800, 0.275, -0.075,
		// 0.135, 0.640,  0.225,
		// 0.195, 0.155,  0.650
		float newR = 0.800f * r + 0.275f * g + -0.075f * b;
		float newG = 0.135f * r + 0.640f * g +  0.225f * b;
		float newB = 0.195f * r + 0.155f * g +  0.650f * b;
#elif (PROFILE_IDX == 1)
		// mGBA GBA color (sRGB). Older version of the libretro shader.
		// Credits: hunterk and Pokefan531.
		// 0.84, 0.18, 0.00,
		// 0.09, 0.67, 0.26,
		// 0.15, 0.10, 0.73
		float newR = 0.84f * r + 0.18f * g + 0.00f * b;
		float newG = 0.09f * r + 0.67f * g + 0.26f * b;
		float newB = 0.15f * r + 0.10f * g + 0.73f * b;
#elif (PROFILE_IDX == 2)
		// libretro DS phat (sRGB). Credits: hunterk and Pokefan531.
		// 0.7050, 0.2350, -0.0750,
		// 0.0900, 0.5850,  0.2400,
		// 0.1075, 0.1725,  0.7200
		float newR = 0.7050f * r + 0.2350f * g + -0.0750f * b;
		float newG = 0.0900f * r + 0.5850f * g +  0.2400f * b;
		float newB = 0.1075f * r + 0.1725f * g +  0.7200f * b;
#else
		// libretro DS phat white (sRGB). Credits: hunterk and Pokefan531.
		// 0.8150, 0.2750, -0.0900,
		// 0.1000, 0.6400,  0.2600,
		// 0.1075, 0.1725,  0.7200
		float newR = 0.8150f * r + 0.2750f * g + -0.0900f * b;
		float newG = 0.1000f * r + 0.6400f * g +  0.2600f * b;
		float newB = 0.1075f * r + 0.1725f * g +  0.7200f * b;
#endif

		newR = (newR < 0.f ? 0.f : newR);
		newG = (newG < 0.f ? 0.f : newG);
		newB = (newB < 0.f ? 0.f : newB);

		// Convert to display gamma.
		static constexpr float displayGammaTable[] = {1.f / 2.f, 1.f / 2.5f + (0.5f * 0.125f), 1.f / 2.f, 1.f / 2.f};
		newR = powf(newR, displayGammaTable[PROFILE_IDX]);
		newG = powf(newG, displayGammaTable[PROFILE_IDX]);
		newB = powf(newB, displayGammaTable[PROFILE_IDX]);

		// Denormalize, clamp and convert to RGB8.
		const u32 a = buf[4 * i + 3];
		buf[3 * i + 0] = std::clamp<s32>(lroundf(newR * a), 0, 255);
		buf[3 * i + 1] = std::clamp<s32>(lroundf(newG * a), 0, 255);
		buf[3 * i + 2] = std::clamp<s32>(lroundf(newB * a), 0, 255);
	} while(++i < width * height);

	lpngErr = lodepng_encode24_file(argv[2], buf, width, height);
	if(lpngErr)
	{
		fprintf(stderr, "lodepng error: %s", lodepng_error_text(lpngErr));
		free(buf);
		return 4;
	}

	free(buf);

	return 0;
}