// Copyright 2019-2021, Collabora, Ltd.
// SPDX-License-Identifier: BSL-1.0
/*!
 * @file
 * @brief  Code to generate disortion meshes.
 * @author Jakob Bornecrantz <jakob@collabora.com>
 * @author Moses Turner <moses@collabora.com>
 * @ingroup aux_distortion
 */

#include "util/u_misc.h"
#include "util/u_frame.h"
#include "util/u_debug.h"
#include "util/u_format.h"
#include "util/u_distortion_mesh.h"

#include "math/m_vec2.h"
#include "math/m_api.h"

#include <stdio.h>
#include <assert.h>


DEBUG_GET_ONCE_NUM_OPTION(mesh_size, "XRT_MESH_SIZE", 64)


typedef bool (*func_calc)(struct xrt_device *xdev, int view, float u, float v, struct xrt_uv_triplet *result);

static int
index_for(int row, int col, uint32_t stride, uint32_t offset)
{
	return row * stride + col + offset;
}

static void
run_func(struct xrt_device *xdev, func_calc calc, int view_count, struct xrt_hmd_parts *target, uint32_t num)
{
	assert(calc != NULL);
	assert(view_count == 2);
	assert(view_count <= 2);

	uint32_t vertex_offsets[2] = {0};
	uint32_t index_offsets[2] = {0};

	uint32_t cells_cols = num;
	uint32_t cells_rows = num;
	uint32_t vert_cols = cells_cols + 1;
	uint32_t vert_rows = cells_rows + 1;

	uint32_t vertex_count_per_view = vert_rows * vert_cols;
	uint32_t vertex_count = vertex_count_per_view * view_count;

	uint32_t uv_channels_count = 3;
	uint32_t stride_in_floats = 2 + uv_channels_count * 2;
	uint32_t float_count = vertex_count * stride_in_floats;

	float *verts = U_TYPED_ARRAY_CALLOC(float, float_count);

	// Setup the vertices for all views.
	uint32_t i = 0;
	for (int view = 0; view < view_count; view++) {
		vertex_offsets[view] = i / stride_in_floats;

		for (uint32_t r = 0; r < vert_rows; r++) {
			// This goes from 0 to 1.0 inclusive.
			float v = (float)r / (float)cells_rows;

			for (uint32_t c = 0; c < vert_cols; c++) {
				// This goes from 0 to 1.0 inclusive.
				float u = (float)c / (float)cells_cols;

				// Make the position in the range of [-1, 1]
				verts[i + 0] = u * 2.0f - 1.0f;
				verts[i + 1] = v * 2.0f - 1.0f;

				if (!calc(xdev, view, u, v, (struct xrt_uv_triplet *)&verts[i + 2])) {
					// bail on error, without updating
					// distortion.preferred
					return;
				}

				i += stride_in_floats;
			}
		}
	}

	uint32_t index_count_per_view = cells_rows * (vert_cols * 2 + 2);
	uint32_t index_count_total = index_count_per_view * view_count;
	int *indices = U_TYPED_ARRAY_CALLOC(int, index_count_total);

	// Set up indices for all views.
	i = 0;
	for (int view = 0; view < view_count; view++) {
		index_offsets[view] = i;

		uint32_t off = vertex_offsets[view];

		for (uint32_t r = 0; r < cells_rows; r++) {
			// Top vertex row for this cell row, left most vertex.
			indices[i++] = index_for(r, 0, vert_cols, off);

			for (uint32_t c = 0; c < vert_cols; c++) {
				indices[i++] = index_for(r, c, vert_cols, off);
				indices[i++] = index_for(r + 1, c, vert_cols, off);
			}

			// Bottom vertex row for this cell row, right most
			// vertex.
			indices[i++] = index_for(r + 1, vert_cols - 1, vert_cols, off);
		}
	}

	target->distortion.models |= XRT_DISTORTION_MODEL_MESHUV;
	target->distortion.mesh.vertices = verts;
	target->distortion.mesh.stride = stride_in_floats * sizeof(float);
	target->distortion.mesh.vertex_count = vertex_count;
	target->distortion.mesh.uv_channels_count = uv_channels_count;
	target->distortion.mesh.indices = indices;
	target->distortion.mesh.index_counts[0] = index_count_per_view;
	target->distortion.mesh.index_counts[1] = index_count_per_view;
	target->distortion.mesh.index_offsets[0] = index_offsets[0];
	target->distortion.mesh.index_offsets[1] = index_offsets[1];
	target->distortion.mesh.index_count_total = index_count_total;
}

bool
u_compute_distortion_vive(struct u_vive_values *values, float u, float v, struct xrt_uv_triplet *result)
{
	// Reading the whole struct like this gives the compiler more opportunity to optimize.
	const struct u_vive_values val = *values;

	const float common_factor_value = 0.5f / (1.0f + val.grow_for_undistort);
	const struct xrt_vec2 factor = {
	    common_factor_value,
	    common_factor_value * val.aspect_x_over_y,
	};

	// Results r/g/b.
	struct xrt_vec2 tc[3] = {{0, 0}, {0, 0}, {0, 0}};

	// Dear compiler, please vectorize.
	for (int i = 0; i < 3; i++) {
		struct xrt_vec2 texCoord = {
		    2.f * u - 1.f,
		    2.f * v - 1.f,
		};

		texCoord.y /= val.aspect_x_over_y;
		texCoord.x -= val.center[i].x;
		texCoord.y -= val.center[i].y;

		float r2 = m_vec2_dot(texCoord, texCoord);
		float k1 = val.coefficients[i][0];
		float k2 = val.coefficients[i][1];
		float k3 = val.coefficients[i][2];
		float k4 = val.coefficients[i][3];

		/*
		 *                     1.0
		 * d = -------------------------------------- + k4
		 *      1.0 + r^2 * k1 + r^4 * k2 + r^6 * k3
		 *
		 * The variable k4 is the scaled part of DISTORT_DPOLY3_SCALED.
		 *
		 * Optimization to reduce the number of multiplications.
		 *    1.0 + r^2 * k1 + r^4 * k2 + r^6 * k3
		 *    1.0 + r^2 * ((k1 + r^2 * k2) + r^2 * k3)
		 */

		float top = 1.f;
		float bottom = 1.f + r2 * (k1 + r2 * (k2 + r2 * k3));
		float d = (top / bottom) + k4;

		struct xrt_vec2 offset = {0.5f, 0.5f};

		tc[i].x = offset.x + (texCoord.x * d + val.center[i].x) * factor.x;
		tc[i].y = offset.y + (texCoord.y * d + val.center[i].y) * factor.y;
	}

	result->r = tc[0];
	result->g = tc[1];
	result->b = tc[2];

	return true;
}


#define mul m_vec2_mul
#define mul_scalar m_vec2_mul_scalar
#define add m_vec2_add
#define sub m_vec2_sub
#define div m_vec2_div
#define div_scalar m_vec2_div_scalar
#define len m_vec2_len
#define len_sqrd m_vec2_len_sqrd

bool
u_compute_distortion_panotools(struct u_panotools_values *values, float u, float v, struct xrt_uv_triplet *result)
{
	const struct u_panotools_values val = *values;

	struct xrt_vec2 r = {u, v};
	r = mul(r, val.viewport_size);
	r = sub(r, val.lens_center);
	r = div_scalar(r, val.scale);

	float r_mag = len(r);
	r_mag = val.distortion_k[0] +                                // r^1
	        val.distortion_k[1] * r_mag +                        // r^2
	        val.distortion_k[2] * r_mag * r_mag +                // r^3
	        val.distortion_k[3] * r_mag * r_mag * r_mag +        // r^4
	        val.distortion_k[4] * r_mag * r_mag * r_mag * r_mag; // r^5

	struct xrt_vec2 r_dist = mul_scalar(r, r_mag);
	r_dist = mul_scalar(r_dist, val.scale);

	struct xrt_vec2 r_uv = mul_scalar(r_dist, val.aberration_k[0]);
	r_uv = add(r_uv, val.lens_center);
	r_uv = div(r_uv, val.viewport_size);

	struct xrt_vec2 g_uv = mul_scalar(r_dist, val.aberration_k[1]);
	g_uv = add(g_uv, val.lens_center);
	g_uv = div(g_uv, val.viewport_size);

	struct xrt_vec2 b_uv = mul_scalar(r_dist, val.aberration_k[2]);
	b_uv = add(b_uv, val.lens_center);
	b_uv = div(b_uv, val.viewport_size);

	result->r = r_uv;
	result->g = g_uv;
	result->b = b_uv;
	return true;
}

bool
u_compute_distortion_cardboard(struct u_cardboard_distortion_values *values,
                               float u,
                               float v,
                               struct xrt_uv_triplet *result)
{
	struct xrt_vec2 uv = {u, v};
	uv = sub(mul(uv, values->screen.size), values->screen.offset);

	float sqrd = len_sqrd(uv);
	float r = 1.0f;
	float fact = 1.0f;
	r *= sqrd;
	fact += values->distortion_k[0] * r;
	r *= sqrd;
	fact += values->distortion_k[1] * r;
	r *= sqrd;
	fact += values->distortion_k[2] * r;
	r *= sqrd;
	fact += values->distortion_k[3] * r;
	r *= sqrd;
	fact += values->distortion_k[4] * r;

	uv = mul_scalar(uv, fact);

	uv = div(add(uv, values->texture.offset), values->texture.size);

	result->r.x = uv.x;
	result->r.y = uv.y;
	result->g.x = uv.x;
	result->g.y = uv.y;
	result->b.x = uv.x;
	result->b.y = uv.y;
	return true;
}

/*
 *
 * North Star "2D Polynomial" distortion
 * Sometimes known as "v2", filename is often NorthStarCalibration.json
 *
 */

static float
u_ns_polyval2d(float X, float Y, float C[16])
{
	float X2 = X * X;
	float X3 = X2 * X;
	float Y2 = Y * Y;
	float Y3 = Y2 * Y;
	return (((C[0]) + (C[1] * Y) + (C[2] * Y2) + (C[3] * Y3)) +
	        ((C[4] * X) + (C[5] * X * Y) + (C[6] * X * Y2) + (C[7] * X * Y3)) +
	        ((C[8] * X2) + (C[9] * X2 * Y) + (C[10] * X2 * Y2) + (C[11] * X2 * Y3)) +
	        ((C[12] * X3) + (C[13] * X3 * Y) + (C[14] * X3 * Y2) + (C[15] * X3 * Y3)));
}


bool
u_compute_distortion_ns_p2d(struct u_ns_p2d_values *values, int view, float u, float v, struct xrt_uv_triplet *result)
{
	// I think that OpenCV and Monado have different definitions of v coordinates, but not sure. if not,
	// unexplainable
	v = 1.0f - v;

	float x_ray = u_ns_polyval2d(u, v, view ? values->x_coefficients_left : values->x_coefficients_right);
	float y_ray = u_ns_polyval2d(u, v, view ? values->y_coefficients_left : values->y_coefficients_right);

	struct xrt_fov fov = values->fov[view];

	float left_ray_bound = tanf(fov.angle_left);
	float right_ray_bound = tanf(fov.angle_right);
	float up_ray_bound = tanf(fov.angle_up);
	float down_ray_bound = tanf(fov.angle_down);

	float u_eye = (float)math_map_ranges(x_ray, left_ray_bound, right_ray_bound, 0, 1);

	float v_eye = (float)math_map_ranges(y_ray, down_ray_bound, up_ray_bound, 0, 1);

	// boilerplate, put the UV coordinates in all the RGB slots
	result->r.x = u_eye;
	result->r.y = v_eye;
	result->g.x = u_eye;
	result->g.y = v_eye;
	result->b.x = u_eye;
	result->b.y = v_eye;

	return true;
}


/*
 *
 * Moses Turner's mesh-grid-based North Star distortion correction.
 * This is a relatively ad-hoc thing I wrote; if this ends up going unused feel free to remove it.
 *
 */

bool
u_compute_distortion_ns_meshgrid(
    struct u_ns_meshgrid_values *values, int view, float u, float v, struct xrt_uv_triplet *result)
{
	int u_edge_num = (values->num_grid_points_u - 1);
	int v_edge_num = (values->num_grid_points_v - 1);

	int u_index_int = floorf(u * u_edge_num);
	int v_index_int = floorf(v * v_edge_num);
	float u_index_frac = (u * u_edge_num) - u_index_int;
	float v_index_frac = (v * v_edge_num) - v_index_int;

	// Imagine this like a ray coming out of your eye with x, y coordinate bearing and z coordinate -1.0f
	struct xrt_vec2 bearing = {0};

	int stride = values->num_grid_points_u;

	float eps = 0.000001;

	struct xrt_vec2 *grid = values->grid[view];
	int topleft_i = (v_index_int * stride) + u_index_int;
	int topright_i = (v_index_int * stride) + u_index_int + 1;
	int bottomleft_i = ((v_index_int + 1) * stride) + u_index_int;
	int bottomright_i = ((v_index_int + 1) * stride) + u_index_int + 1;

	if (u_index_frac > eps && v_index_frac > eps) {
		// Usual case - we're in the middle of a cell
		// {top,bottom}-{left,right} notation might be inaccurate. The code *works* right now but don't take its
		// word when reading
		struct xrt_vec2 topleft = grid[topleft_i];
		struct xrt_vec2 topright = grid[topright_i];
		struct xrt_vec2 bottomleft = grid[bottomleft_i];
		struct xrt_vec2 bottomright = grid[bottomright_i];

		struct xrt_vec2 left_point_on_line_segment = m_vec2_lerp(topleft, bottomleft, v_index_frac);
		struct xrt_vec2 right_point_on_line_segment = m_vec2_lerp(topright, bottomright, v_index_frac);

		bearing = m_vec2_lerp(left_point_on_line_segment, right_point_on_line_segment, u_index_frac);

	} else if (v_index_frac > eps) {
		// We're on a vertical edge
		struct xrt_vec2 top = values->grid[view][topleft_i];
		struct xrt_vec2 bottom = values->grid[view][bottomleft_i];
		bearing = m_vec2_lerp(top, bottom, v_index_frac);
	} else if (u_index_frac > eps) {
		// We're on a horizontal edge
		struct xrt_vec2 left = values->grid[view][topleft_i];
		struct xrt_vec2 right = values->grid[view][topright_i];
		bearing = m_vec2_lerp(left, right, u_index_frac);
	} else {
		int acc_idx = (v_index_int * stride) + u_index_int;
		bearing = values->grid[view][acc_idx];
	}

	struct xrt_fov fov = values->fov[view];

	float left_ray_bound = tan(fov.angle_left);
	float right_ray_bound = tan(fov.angle_right);
	float up_ray_bound = tan(fov.angle_up);
	float down_ray_bound = tan(fov.angle_down);

	float u_eye = math_map_ranges(bearing.x, left_ray_bound, right_ray_bound, 0, 1);
	float v_eye = math_map_ranges(bearing.y, down_ray_bound, up_ray_bound, 0, 1);

	// boilerplate, put the UV coordinates in all the RGB slots
	result->r.x = u_eye;
	result->r.y = v_eye;
	result->g.x = u_eye;
	result->g.y = v_eye;
	result->b.x = u_eye;
	result->b.y = v_eye;

	return true;
}


bool
u_compute_distortion_none(float u, float v, struct xrt_uv_triplet *result)
{
	result->r.x = u;
	result->r.y = v;
	result->g.x = u;
	result->g.y = v;
	result->b.x = u;
	result->b.y = v;
	return true;
}



/*
 *
 * No distortion.
 *
 */

bool
u_distortion_mesh_none(struct xrt_device *xdev, int view, float u, float v, struct xrt_uv_triplet *result)
{
	return u_compute_distortion_none(u, v, result);
}

void
u_distortion_mesh_fill_in_none(struct xrt_device *xdev)
{
	struct xrt_hmd_parts *target = xdev->hmd;

	// Do the generation.
	run_func(xdev, u_distortion_mesh_none, 2, target, 1);

	// Make the target mostly usable.
	target->distortion.models |= XRT_DISTORTION_MODEL_NONE;
	target->distortion.models |= XRT_DISTORTION_MODEL_MESHUV;
	target->distortion.preferred = XRT_DISTORTION_MODEL_MESHUV;
}

void
u_distortion_mesh_set_none(struct xrt_device *xdev)
{
	struct xrt_hmd_parts *target = xdev->hmd;

	// Reset to none.
	target->distortion.models = XRT_DISTORTION_MODEL_NONE;

	u_distortion_mesh_fill_in_none(xdev);

	// Make sure that the xdev implements the compute_distortion function.
	xdev->compute_distortion = u_distortion_mesh_none;

	// Make the target completely usable.
	target->distortion.models |= XRT_DISTORTION_MODEL_COMPUTE;
}


/*
 *
 *
 *
 */

void
u_distortion_mesh_fill_in_compute(struct xrt_device *xdev)
{
	func_calc calc = xdev->compute_distortion;
	if (calc == NULL) {
		u_distortion_mesh_fill_in_none(xdev);
		return;
	}

	struct xrt_hmd_parts *target = xdev->hmd;

	uint32_t num = (uint32_t)debug_get_num_option_mesh_size();
	run_func(xdev, calc, 2, target, num);
}