rslibcamlite 0.1.7

#include "yuv.hpp"

#include <algorithm>
#include <cassert>

// Borrowed from rpicamapps

std::vector<uint8_t> Yuv420ToRgb(const uint8_t *src, StreamInfo &src_info, StreamInfo &dst_info)
{
	std::vector<uint8_t> output(dst_info.height * dst_info.stride);
	Yuv420ToRgb(output.data(), src, src_info, dst_info);
	return output;
}

void Yuv420ToRgb(uint8_t *dst, const uint8_t *src, StreamInfo &src_info, StreamInfo &dst_info)
{
	assert(src_info.width >= dst_info.width && src_info.height >= dst_info.height);
	int off_x = ((src_info.width - dst_info.width) / 2) & ~1, off_y = ((src_info.height - dst_info.height) / 2) & ~1;
	int src_Y_size = src_info.height * src_info.stride, src_U_size = (src_info.height / 2) * (src_info.stride / 2);

	// We're going to process 4x2 pixel blocks, as far as alignment allows.
	unsigned int dst_h_aligned = dst_info.height & ~1, dst_w_aligned = dst_info.width & ~3;

	unsigned int y = 0;
	for (; y < dst_h_aligned; y += 2)
	{
		const uint8_t *src_Y0 = src + (y + off_y) * src_info.stride + off_x;
		const uint8_t *src_U = src + src_Y_size + ((y + off_y) / 2) * (src_info.stride / 2) + off_x / 2;
		const uint8_t *src_V = src_U + src_U_size;
		const uint8_t *src_Y1 = src_Y0 + src_info.stride;
		uint8_t *dst0 = &dst[y * dst_info.stride];
		uint8_t *dst1 = dst0 + dst_info.stride;

		unsigned int x = 0;
		for (; x < dst_w_aligned; x += 4)
		{
			int Y0 = *(src_Y0++);
			int U0 = *(src_U++);
			int V0 = *(src_V++);
			int Y1 = *(src_Y0++);
			int Y2 = *(src_Y0++);
			int U2 = *(src_U++);
			int V2 = *(src_V++);
			int Y3 = *(src_Y0++);
			int Y4 = *(src_Y1++);
			int Y5 = *(src_Y1++);
			int Y6 = *(src_Y1++);
			int Y7 = *(src_Y1++);

			U0 -= 128;
			V0 -= 128;
			U2 -= 128;
			V2 -= 128;
			int U1 = U0;
			int V1 = V0;
			int U4 = U0;
			int V4 = V0;
			int U5 = U0;
			int V5 = V0;
			int U3 = U2;
			int V3 = V2;
			int U6 = U2;
			int V6 = V2;
			int U7 = U2;
			int V7 = V2;

			int R0 = Y0 + 1.402 * V0;
			int G0 = Y0 - 0.345 * U0 - 0.714 * V0;
			int B0 = Y0 + 1.771 * U0;
			int R1 = Y1 + 1.402 * V1;
			int G1 = Y1 - 0.345 * U1 - 0.714 * V1;
			int B1 = Y1 + 1.771 * U1;
			int R2 = Y2 + 1.402 * V2;
			int G2 = Y2 - 0.345 * U2 - 0.714 * V2;
			int B2 = Y2 + 1.771 * U2;
			int R3 = Y3 + 1.402 * V3;
			int G3 = Y3 - 0.345 * U3 - 0.714 * V3;
			int B3 = Y3 + 1.771 * U3;
			int R4 = Y4 + 1.402 * V4;
			int G4 = Y4 - 0.345 * U4 - 0.714 * V4;
			int B4 = Y4 + 1.771 * U4;
			int R5 = Y5 + 1.402 * V5;
			int G5 = Y5 - 0.345 * U5 - 0.714 * V5;
			int B5 = Y5 + 1.771 * U5;
			int R6 = Y6 + 1.402 * V6;
			int G6 = Y6 - 0.345 * U6 - 0.714 * V6;
			int B6 = Y6 + 1.771 * U6;
			int R7 = Y7 + 1.402 * V7;
			int G7 = Y7 - 0.345 * U7 - 0.714 * V7;
			int B7 = Y7 + 1.771 * U7;

			R0 = std::clamp(R0, 0, 255);
			G0 = std::clamp(G0, 0, 255);
			B0 = std::clamp(B0, 0, 255);
			R1 = std::clamp(R1, 0, 255);
			G1 = std::clamp(G1, 0, 255);
			B1 = std::clamp(B1, 0, 255);
			R2 = std::clamp(R2, 0, 255);
			G2 = std::clamp(G2, 0, 255);
			B2 = std::clamp(B2, 0, 255);
			R3 = std::clamp(R3, 0, 255);
			G3 = std::clamp(G3, 0, 255);
			B3 = std::clamp(B3, 0, 255);
			R4 = std::clamp(R4, 0, 255);
			G4 = std::clamp(G4, 0, 255);
			B4 = std::clamp(B4, 0, 255);
			R5 = std::clamp(R5, 0, 255);
			G5 = std::clamp(G5, 0, 255);
			B5 = std::clamp(B5, 0, 255);
			R6 = std::clamp(R6, 0, 255);
			G6 = std::clamp(G6, 0, 255);
			B6 = std::clamp(B6, 0, 255);
			R7 = std::clamp(R7, 0, 255);
			G7 = std::clamp(G7, 0, 255);
			B7 = std::clamp(B7, 0, 255);

			*(dst0++) = R0;
			*(dst0++) = G0;
			*(dst0++) = B0;
			*(dst0++) = R1;
			*(dst0++) = G1;
			*(dst0++) = B1;
			*(dst0++) = R2;
			*(dst0++) = G2;
			*(dst0++) = B2;
			*(dst0++) = R3;
			*(dst0++) = G3;
			*(dst0++) = B3;
			*(dst1++) = R4;
			*(dst1++) = G4;
			*(dst1++) = B4;
			*(dst1++) = R5;
			*(dst1++) = G5;
			*(dst1++) = B5;
			*(dst1++) = R6;
			*(dst1++) = G6;
			*(dst1++) = B6;
			*(dst1++) = R7;
			*(dst1++) = G7;
			*(dst1++) = B7;
		}
		// Straggling pixel columns - we must still do both rows.
		for (; x < dst_info.width; x++)
		{
			int Y0 = *(src_Y0++);
			int U0 = *(src_U);
			int V0 = *(src_V);
			int Y4 = *(src_Y1++);
			src_U += (x & 1);
			src_V += (x & 1);

			U0 -= 128;
			V0 -= 128;
			int U4 = U0;
			int V4 = V0;

			int R0 = Y0 + 1.402 * V0;
			int G0 = Y0 - 0.345 * U0 - 0.714 * V0;
			int B0 = Y0 + 1.771 * U0;
			int R4 = Y4 + 1.402 * V4;
			int G4 = Y4 - 0.345 * U4 - 0.714 * V4;
			int B4 = Y4 + 1.771 * U4;

			R0 = std::clamp(R0, 0, 255);
			G0 = std::clamp(G0, 0, 255);
			B0 = std::clamp(B0, 0, 255);
			R4 = std::clamp(R4, 0, 255);
			G4 = std::clamp(G4, 0, 255);
			B4 = std::clamp(B4, 0, 255);

			*(dst0++) = R0;
			*(dst0++) = G0;
			*(dst0++) = B0;
			*(dst1++) = R4;
			*(dst1++) = G4;
			*(dst1++) = B4;
		}
	}
	// Any straggling final row is done with extreme steam power.
	for (; y < dst_info.height; y++)
	{
		const uint8_t *src_Y0 = src + (y + off_y) * src_info.stride + off_x;
		const uint8_t *src_U = src + src_Y_size + ((y + off_y) / 2) * (src_info.stride / 2) + off_x / 2;
		const uint8_t *src_V = src_U + src_U_size;
		uint8_t *dst0 = &dst[y * dst_info.stride];

		unsigned int x = 0;
		for (; x < dst_info.width; x++)
		{
			int Y0 = *(src_Y0++);
			int U0 = *(src_U);
			int V0 = *(src_V);
			src_U += (x & 1);
			src_V += (x & 1);

			U0 -= 128;
			V0 -= 128;

			int R0 = Y0 + 1.402 * V0;
			int G0 = Y0 - 0.345 * U0 - 0.714 * V0;
			int B0 = Y0 + 1.771 * U0;

			R0 = std::clamp(R0, 0, 255);
			G0 = std::clamp(G0, 0, 255);
			B0 = std::clamp(B0, 0, 255);

			*(dst0++) = R0;
			*(dst0++) = G0;
			*(dst0++) = B0;
		}
	}
}