// SPDX-License-Identifier: CC0-1.0
//
// SPDX-FileContributor: Antonio Niño Díaz, 2022

#include <stdint.h>

#define GBA_SCREEN_W            240
#define GBA_SCREEN_H            160

#define REG_DISPCNT             *((volatile uint16_t *)0x04000000)

#define REG_KEYINPUT		*((volatile uint16_t *)0x04000130)
#define DEFAULT_REG_KEYINPUT	*((volatile uint16_t *)0x04000130)
#define KEY_DOWN_NOW(key)  (~(REG_KEYINPUT) & key)

#define KEY_A			1
#define KEY_B			2
#define KEY_SELECT		3
#define KEY_START		4
#define KEY_DPAD_RIGHT		5
#define KEY_DPAD_LEFT		6
#define KEY_DPAD_UP		7
#define KEY_DPAD_DOWN		8
#define KEY_TRIGGER_LEFT	9
#define KEY_TRIGGER_RIGHT	10

#define DISPCNT_BG_MODE_MASK    (0x7)
#define DISPCNT_BG_MODE(n)      ((n) & DISPCNT_BG_MODE_MASK) // 0 to 5

#define DISPCNT_BG2_ENABLE      (1 << 10)

#define MEM_VRAM_MODE3_FB       ((uint16_t *)0x06000000)

#define FIXED_POINT int32_t
#define fp 12
#define SHIFT_THRESHOLD 0.05
#define SHIFT_THRESHOLD_FP ((1 << fp) * SHIFT_THRESHOLD)

#define FLOAT2FIXED(value)     (int)((value) * (1 << fp))
#define FIXED2FLOAT(value)     ((value) / (float)(1 << fp))

static inline uint16_t
RGB15(uint16_t r, uint16_t g, uint16_t b)
{
	return (r & 0x1F) | ((g & 0x1F) << 5) | ((b & 0x1F) << 10);
}

///////////////////////////////////////////////////////////

#include <stdio.h>
#include <math.h>
#include <unistd.h>
#include <string.h>
#include <fcntl.h>
#include <float.h>

#define VWIDTH 50
#define VHEIGHT 50
#define CUBE_WIDTH 10
#define CUBE_WIDTH_FP ((1 << fp) * CUBE_WIDTH)

enum faces {
	FACE_FRONT = 0,
	FACE_LEFT,
	FACE_RIGHT,
	FACE_BOTTOM,
	FACE_TOP,
	FACE_BACK,
	NUM_FACES,
};

#define STEP 5
#define STEP_FP ((1 << fp) * STEP)

#define ACTION_STEP 0.1
#define ACTION_STEP_FP ((1 << fp) * ACTION_STEP)

#define PITCH_STEP 0.05
#define ROLL_STEP 0.05
#define YAW_STEP 0.05

volatile FIXED_POINT  K1 = 60;
volatile FIXED_POINT K2 = (2 * CUBE_WIDTH) + 20;


#define MULT_FP(a,b) ((a * b) >> fp)

#define SQ(n) (n * n)
#define SQ_FP(n) (MULT_FP(n, n))

#define COORD2INDEX(x, y) (y * VWIDTH + x)
#define COUPLE2INDEX(x) (COORD2INDEX(x[0], x[1]))

#define GET_ROTATE_X_Q(a) ({ float _a = (FIXED2FLOAT(a)) ; \
		struct Quaternions q = {}; q.w = FLOAT2FIXED(cos(_a * .5)); \
		q.x = FLOAT2FIXED(sin(_a * .5)); q; })
#define GET_ROTATE_Y_Q(a) ({ float _a = (FIXED2FLOAT(a)) ; \
		struct Quaternions q = {}; q.w = FLOAT2FIXED(cos(_a * .5)); \
		q.y = FLOAT2FIXED(sin(_a * .5)); q; })
#define GET_ROTATE_Z_Q(a) ({ float _a = (FIXED2FLOAT(a)) ; \
		struct Quaternions q = {}; q.w = FLOAT2FIXED(cos(_a * .5)); \
		q.z = FLOAT2FIXED(sin(_a * .5)); q; })

//TODO Idle animations
#define IS_IDLE (Idle.x || Idle.y || Idle.z)
#define RESET_IDLE {Idle.x = 0; Idle.y = 0; Idle.z = 0;}

struct {
	char x;
	char y;
	char z;
} Idle;


struct Quaternions {
	FIXED_POINT w;
	FIXED_POINT x;
	FIXED_POINT y;
	FIXED_POINT z;
} Target, Current;

FIXED_POINT		interpolationStep = 0;
FIXED_POINT		zBuffer[VHEIGHT * VWIDTH];
char			output[VHEIGHT * VWIDTH];

static volatile char shouldBreak = 1;
static volatile char currentlyMoving = 0;
static volatile char currentCountR = 0;
static volatile char frontFacingFace = FACE_FRONT;

void
normalize(struct Quaternions *q)
{
	float n = sqrt(FIXED2FLOAT(SQ_FP(q->w) + SQ_FP(q->x) +
				SQ_FP(q->y) + SQ_FP(q->z)));
	if (n == 0)
		return;
	q->w = FLOAT2FIXED(FIXED2FLOAT(q->w) / n);
	q->x = FLOAT2FIXED(FIXED2FLOAT(q->x) / n);
	q->y = FLOAT2FIXED(FIXED2FLOAT(q->y) / n);
	q->z = FLOAT2FIXED(FIXED2FLOAT(q->z) / n);
}

struct Quaternions
mult(struct Quaternions q, FIXED_POINT x, FIXED_POINT y, FIXED_POINT z)
{
	//p = q * p * qbar
	struct Quaternions res;

	res.w = 0;
	res.x = MULT_FP(x, (SQ_FP(q.w) + SQ_FP(q.x) - SQ_FP(q.y) - SQ_FP(q.z)))
		+ (MULT_FP(y, (MULT_FP(q.x, q.y) - MULT_FP(q.w, q.z))) * 2)
		+ (MULT_FP(z, (MULT_FP(q.x, q.z) + MULT_FP(q.w, q.y))) * 2);

	res.y = (MULT_FP(x, (MULT_FP(q.x, q.y) + MULT_FP(q.w,q.z))) * 2)
		+ (MULT_FP(y, (SQ_FP(q.w) - SQ_FP(q.x) + SQ_FP(q.y) - SQ_FP(q.z))))
		+ (MULT_FP(z, (MULT_FP(q.y, q.z) - MULT_FP(q.w, q.x))) << 2);

	res.z = (MULT_FP(x, (MULT_FP(q.x, q.z) - MULT_FP(q.w, q.y)))* 2)
		+ (MULT_FP(y, (MULT_FP(q.y, q.z) + MULT_FP(q.w, q.x))) * 2)
		+ MULT_FP(z, (SQ_FP(q.w) - SQ_FP(q.x) - SQ_FP(q.y) + SQ_FP(q.z)));

	return res;
}

struct Quaternions
multQ(struct Quaternions p, struct Quaternions q)
{
	if (p.x <= SHIFT_THRESHOLD_FP && p.x >= -SHIFT_THRESHOLD_FP
	    && p.y <= SHIFT_THRESHOLD_FP && p.y >= -SHIFT_THRESHOLD_FP
	    && p.z <= SHIFT_THRESHOLD_FP && p.z >= -SHIFT_THRESHOLD_FP)
		return q;

	if (q.x <= SHIFT_THRESHOLD_FP && q.x >= -SHIFT_THRESHOLD_FP
	    && q.y <= SHIFT_THRESHOLD_FP && q.y >= -SHIFT_THRESHOLD_FP
	    && q.z <= SHIFT_THRESHOLD_FP && q.z >= -SHIFT_THRESHOLD_FP)
		return p;

	struct Quaternions res = {
		.w = MULT_FP(p.w, q.w) - MULT_FP(p.x, q.x) -
			MULT_FP(p.y, q.y) - MULT_FP(p.z, q.z),
		.x = MULT_FP(p.w, q.x) + MULT_FP(p.x, q.w) +
			MULT_FP(p.y, q.z) - MULT_FP(p.z, q.y),
		.y = MULT_FP(p.w, q.y) - MULT_FP(p.x, q.z) +
			MULT_FP(p.y, q.w) + MULT_FP(p.z, q.x),
		.z = MULT_FP(p.w, q.z) + MULT_FP(p.x, q.y) -
			MULT_FP(p.y, q.x) + MULT_FP(p.z, q.w),
	};

	return res;
}

uint16_t
chooseColor(char c)
{
	switch (c) {
	case FACE_FRONT:
		return RGB15(31, 0, 0);
	case FACE_BACK:
		return RGB15(31, 15, 31);
	case FACE_BOTTOM:
		return RGB15(31, 0, 31);
	case FACE_LEFT:
		return RGB15(0, 0, 31);
	case FACE_RIGHT:
		return RGB15(0, 31, 31);
	case FACE_TOP:
		return RGB15(0, 31, 0);
	default:
		// BG
		return RGB15(31, 31, 31);
	}
}

char
chooseMainFace()
{
	int total = 0;
	int faces[NUM_FACES] = {0};

	for (int k = 0; k < VWIDTH * VHEIGHT; ++k)
		if (output[k] >= 0 && output[k] < NUM_FACES) {
			faces[output[k]]++;
			++total;
		}

	int max = 0, idx = 0;
	for (int k = 0; k < NUM_FACES; ++k)
		if (faces[k] > max) {
			max = faces[k];
			idx = k;
		}

	frontFacingFace = max > total * 0.9 ? idx : -1;
	return frontFacingFace;
}

char
isInQuad(char curr[2], char top[2], char left[2],
		char right[2], char bot[2])
{
	char *points[4] = {top, left, bot, right};

	char pos = 0, neg = 0;
	char x = curr[0];
	char y = curr[1];;
	int d;

	for (char i = 0; i < 4; ++i) {
		if (points[i][0] == curr[0] && points[i][1] == curr[1])
		    return 1;

		//Form a segment between the i'th point
		char x1 = points[i][0];
		char y1 = points[i][1];

		//And the i+1'th, or if i is the last, with the first point
		char i2 = (i + 1) % 4;

		char x2 = points[i2][0];
		char y2 = points[i2][1];


		//Compute the cross product
		d = (x - x1) * (y2 - y1) - (y - y1) * (x2 - x1);

		if (d > 0) ++pos;
		if (d < 0) ++neg;

		//If the sign changes, then point is outside
		if (pos > 0 && neg > 0)
		    return 0;
	}

	return 1;
}

void
fill_quads(char current_face, char top[2], char left[2],
		char right[2], char bot[2])
{
	if (current_face != 0) return;
	output[COUPLE2INDEX(top)] = RGB15(0, 0, 15);
	output[COUPLE2INDEX(left)] = RGB15(0, 0, 15);
	output[COUPLE2INDEX(right)] = RGB15(0, 0, 15);

	for (int y = top[1] ; y < bot[1] ; ++y) {
		for (int x = left[0] ; x < right[0] ; ++x) {
			char curr[2] = {x, y};
			if (isInQuad(curr, top, left, right, bot))
				//zbuffer issue
				{}
				//output[COORD2INDEX(x, y)] = current_face;
		}
	}
}

void
detect_and_fill_quads()
{
	for (int current_face = 0 ; current_face < NUM_FACES ; ++current_face) {
		char last_top [2] = {VWIDTH, VHEIGHT};
		char last_left[2] = {VWIDTH, 0};
		char last_right [2] = {0, 0};
		char last_bot[2] = {0, 0};
		char top [2] = {VWIDTH, VHEIGHT};
		char left[2] = {VWIDTH, 0};
		char right [2] = {0, 0};
		char bot[2] = {0, 0};
		for (char y = 0; y < VHEIGHT; ++y) {
			for (char x = 0; x < VWIDTH; ++x) {
				if (output[COORD2INDEX(x, y)] != current_face)
					continue;
				if (x <= left[0]) {
					left[0] = x;
					left[1] = y;
				}
				if (y <= top[1]) {
					top[0] = x;
					top[1] = y;
				}
				if (x >= right[0]) {
					right[0] = x;
					right[1] = y;
				}
				if (y >= bot[1]) {
					bot[0] = x;
					bot[1] = y;
				}
			}
		}
		fill_quads(current_face, top, left, right, bot);
	}

}

void
printAscii()
{
	// TODO scale up
	MEM_VRAM_MODE3_FB[120 + 80 * GBA_SCREEN_W] = RGB15(currentCountR, 31 - currentCountR, 0);
	MEM_VRAM_MODE3_FB[136 + 80 * GBA_SCREEN_W] = RGB15(currentCountR, 31 - currentCountR, 0);
	MEM_VRAM_MODE3_FB[120 + 96 * GBA_SCREEN_W] = RGB15(currentCountR, 31 - currentCountR, 0);
	currentCountR = currentCountR == 31 ? 0 : 31;

	detect_and_fill_quads();

	for (int i = 0; i < VHEIGHT; ++i) {
		for (int j = 0; j < VWIDTH; ++j) {
			char prevc = 0;
			char *c = output + (i * VWIDTH + j);
			MEM_VRAM_MODE3_FB[(i + 50) * GBA_SCREEN_W + j + 50] = chooseColor(*c);
		}
	}
}

void
rotateCube(FIXED_POINT cubeX, FIXED_POINT cubeY, FIXED_POINT cubeZ, char ch)
{
	struct Quaternions q = mult(Current, cubeX, cubeY, cubeZ);

	int x = q.x >> fp;
	int y = q.y >> fp;

	// not fixed point yet!!
	float invZ = (1 << fp) / (float)(q.z + K2 * (1 << fp));

	int screenX = (int)(VWIDTH * 0.5) + (int)((x) * K1) * invZ;
	int screenY = (int)(VHEIGHT * 0.5) + (int)((y) * K1) * invZ;
	//TODO luminescence

	if (screenX > VWIDTH || screenX < 0) return;

	int idx = screenY * VWIDTH + screenX;
	if (idx >= 0 && idx < VWIDTH * VHEIGHT) {
		invZ = FLOAT2FIXED(invZ);
		if (zBuffer[idx] < invZ) {
			zBuffer[idx] = invZ;
			output[idx] = ch;
		}
	}
}

struct Quaternions
interpolate(struct Quaternions qa, struct Quaternions qb)
{
	frontFacingFace = -1;
	struct Quaternions res;
	float cosHalfTheta =
		FIXED2FLOAT(MULT_FP(qa.w, qb.w) +
		MULT_FP(qa.x, qb.x) +
		MULT_FP(qa.y, qb.y) +
		MULT_FP(qa.z, qb.z));
	//if qa = qb or qa = -qb then theta = 0 and we can return qa
	if (cosHalfTheta >= 1.0 || cosHalfTheta <= -1.0) {
		res.w = qa.w;
		res.x = qa.x;
		res.y = qa.y;
		res.z = qa.z;
		goto exit;
	}
	if (cosHalfTheta < 0) {
		qb.w = -qb.w;
		qb.x = -qb.x;
		qb.y = -qb.y;
		qb.z = qb.z;
		cosHalfTheta = -cosHalfTheta;
	}

	float		halfTheta = acos(cosHalfTheta);
	float		sinHalfTheta = sqrt(1.0 - cosHalfTheta * cosHalfTheta);
	//if theta = 180 degrees then result is not fully defined
	// we could rotate around any axis normal to qa or qb
	if (sinHalfTheta < 0.001 && sinHalfTheta > -0.001) {
		res.w = ((qa.w >> 1) + (qb.w >> 1));
		res.x = ((qa.x >> 1) + (qb.x >> 1));
		res.y = ((qa.y >> 1) + (qb.y >> 1));
		res.z = ((qa.z >> 1) + (qb.z >> 1));
		goto exit;
	}


	FIXED_POINT ratioA = FLOAT2FIXED(sin((1 - FIXED2FLOAT(interpolationStep)) * halfTheta) / sinHalfTheta);
	FIXED_POINT ratioB = FLOAT2FIXED(sin(FIXED2FLOAT(interpolationStep) * halfTheta) / sinHalfTheta);

	res.w = (MULT_FP(qa.w, ratioA) + MULT_FP(qb.w, ratioB));
	res.x = (MULT_FP(qa.x, ratioA) + MULT_FP(qb.x, ratioB));
	res.y = (MULT_FP(qa.y, ratioA) + MULT_FP(qb.y, ratioB));
	res.z = (MULT_FP(qa.z, ratioA) + MULT_FP(qb.z, ratioB));

exit:
	interpolationStep += ACTION_STEP_FP;
	return res;
}

void
handleAngle(char input)
{
	// TODO
	if (currentlyMoving == 0) {
		currentlyMoving = input;
		switch (input) {
		case 'w':
		case 'W':
			Target = multQ(GET_ROTATE_X_Q(FLOAT2FIXED(M_PI_2)), Current);
			break;
		case 'a':
		case 'A':
			Target = multQ(GET_ROTATE_Y_Q(-FLOAT2FIXED(M_PI_2)), Current);
			break;
		case 's':
		case 'S':
			Target = multQ(GET_ROTATE_X_Q(-FLOAT2FIXED(M_PI_2)), Current);
			break;
		case 'd':
		case 'D':
			Target = multQ(GET_ROTATE_Y_Q(FLOAT2FIXED(M_PI_2)), Current);
			break;
		case 'q':
		case 'Q':
			Target = multQ(GET_ROTATE_Z_Q(-FLOAT2FIXED(M_PI_2)), Current);
			break;
		case 'e':
		case 'E':
			Target = multQ(GET_ROTATE_Z_Q(FLOAT2FIXED(M_PI_2)), Current);
			break;
		default:
			currentlyMoving = 0;
			//TODO idle movement
		}
		normalize(&Target);
	} else {
		if (interpolationStep < (1 << fp) - ACTION_STEP_FP * 2) {
			Current = interpolate(Current, Target);
			normalize(&Current);
		}
		else {
			Current = Target;
			interpolationStep = 0;
			currentlyMoving = 0;
		}
	}
}

char
getInput()
{
	// TODO
	char c = 'd';
	handleAngle(c);
	return c;
}


int
main()
{
	REG_DISPCNT = DISPCNT_BG_MODE(3) | DISPCNT_BG2_ENABLE;

	Current = GET_ROTATE_Z_Q(0);

	while (1) {
		memset(output, NUM_FACES, VWIDTH * VHEIGHT);
		memset(zBuffer, 0xffffffff, VWIDTH * VHEIGHT * sizeof(FIXED_POINT));

		for (FIXED_POINT cubeX = -CUBE_WIDTH_FP + STEP_FP ;
				cubeX <= CUBE_WIDTH_FP - STEP_FP; cubeX += STEP_FP) {
			for (FIXED_POINT cubeY = -CUBE_WIDTH_FP + STEP_FP;
					cubeY <= CUBE_WIDTH_FP - STEP_FP; cubeY += STEP_FP) {
				switch (FACE_FRONT) {
				case FACE_FRONT:
					rotateCube(cubeX, cubeY, -CUBE_WIDTH_FP, FACE_FRONT);
					break;
				case FACE_LEFT:
					rotateCube(-CUBE_WIDTH_FP, cubeX, cubeY, FACE_LEFT);
					break;
				case FACE_RIGHT:
					rotateCube(CUBE_WIDTH_FP, cubeX, cubeY, FACE_RIGHT);
					break;
				case FACE_BOTTOM:
					rotateCube(cubeX, -CUBE_WIDTH_FP, cubeY, FACE_BOTTOM);
					break;
				case FACE_TOP:
					rotateCube(cubeX, CUBE_WIDTH_FP, cubeY, FACE_TOP);
					break;
				case FACE_BACK:
					rotateCube(cubeX, cubeY, CUBE_WIDTH_FP, FACE_BACK);
					break;
				default: // idk render all
					rotateCube(cubeX, cubeY, -CUBE_WIDTH_FP, FACE_FRONT);
					rotateCube(-CUBE_WIDTH_FP, cubeX, cubeY, FACE_LEFT);
					rotateCube(CUBE_WIDTH_FP, cubeX, cubeY, FACE_RIGHT);
					rotateCube(cubeX, -CUBE_WIDTH_FP, cubeY, FACE_TOP);
					rotateCube(cubeX, CUBE_WIDTH_FP, cubeY, FACE_BOTTOM);
					rotateCube(cubeX, cubeY, CUBE_WIDTH_FP, FACE_BACK);
				}
			}
		}

		printAscii();
		getInput();
	}
}