obs-studio/libobs/media-io/format-conversion.c

/******************************************************************************
    Copyright (C) 2023 by Lain Bailey <[email protected]>

    This program is free software: you can redistribute it and/or modify
    it under the terms of the GNU General Public License as published by
    the Free Software Foundation, either version 2 of the License, or
    (at your option) any later version.

    This program is distributed in the hope that it will be useful,
    but WITHOUT ANY WARRANTY; without even the implied warranty of
    MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
    GNU General Public License for more details.

    You should have received a copy of the GNU General Public License
    along with this program.  If not, see <http://www.gnu.org/licenses/>.
******************************************************************************/

#include "format-conversion.h"

#include "../util/sse-intrin.h"

/* ...surprisingly, if I don't use a macro to force inlining, it causes the
 * CPU usage to boost by a tremendous amount in debug builds. */

#define get_m128_32_0(val) (*((uint32_t *)&val))
#define get_m128_32_1(val) (*(((uint32_t *)&val) + 1))

#define pack_shift(lum_plane, lum_pos0, lum_pos1, line1, line2, mask, sh)      \
	do {                                                                   \
		__m128i pack_val = _mm_packs_epi32(                            \
			_mm_srli_si128(_mm_and_si128(line1, mask), sh),        \
			_mm_srli_si128(_mm_and_si128(line2, mask), sh));       \
		pack_val = _mm_packus_epi16(pack_val, pack_val);               \
                                                                               \
		*(uint32_t *)(lum_plane + lum_pos0) = get_m128_32_0(pack_val); \
		*(uint32_t *)(lum_plane + lum_pos1) = get_m128_32_1(pack_val); \
	} while (false)

#define pack_val(lum_plane, lum_pos0, lum_pos1, line1, line2, mask)            \
	do {                                                                   \
		__m128i pack_val =                                             \
			_mm_packs_epi32(_mm_and_si128(line1, mask),            \
					_mm_and_si128(line2, mask));           \
		pack_val = _mm_packus_epi16(pack_val, pack_val);               \
                                                                               \
		*(uint32_t *)(lum_plane + lum_pos0) = get_m128_32_0(pack_val); \
		*(uint32_t *)(lum_plane + lum_pos1) = get_m128_32_1(pack_val); \
	} while (false)

#define pack_ch_1plane(uv_plane, chroma_pos, line1, line2, uv_mask)            \
	do {                                                                   \
		__m128i add_val =                                              \
			_mm_add_epi64(_mm_and_si128(line1, uv_mask),           \
				      _mm_and_si128(line2, uv_mask));          \
		__m128i avg_val = _mm_add_epi64(                               \
			add_val,                                               \
			_mm_shuffle_epi32(add_val, _MM_SHUFFLE(2, 3, 0, 1)));  \
		avg_val = _mm_srai_epi16(avg_val, 2);                          \
		avg_val = _mm_shuffle_epi32(avg_val, _MM_SHUFFLE(3, 1, 2, 0)); \
		avg_val = _mm_packus_epi16(avg_val, avg_val);                  \
                                                                               \
		*(uint32_t *)(uv_plane + chroma_pos) = get_m128_32_0(avg_val); \
	} while (false)

#define pack_ch_2plane(u_plane, v_plane, chroma_pos, line1, line2, uv_mask)    \
	do {                                                                   \
		uint32_t packed_vals;                                          \
                                                                               \
		__m128i add_val =                                              \
			_mm_add_epi64(_mm_and_si128(line1, uv_mask),           \
				      _mm_and_si128(line2, uv_mask));          \
		__m128i avg_val = _mm_add_epi64(                               \
			add_val,                                               \
			_mm_shuffle_epi32(add_val, _MM_SHUFFLE(2, 3, 0, 1)));  \
		avg_val = _mm_srai_epi16(avg_val, 2);                          \
		avg_val = _mm_shuffle_epi32(avg_val, _MM_SHUFFLE(3, 1, 2, 0)); \
		avg_val =                                                      \
			_mm_shufflelo_epi16(avg_val, _MM_SHUFFLE(3, 1, 2, 0)); \
		avg_val = _mm_packus_epi16(avg_val, avg_val);                  \
                                                                               \
		packed_vals = get_m128_32_0(avg_val);                          \
                                                                               \
		*(uint16_t *)(u_plane + chroma_pos) = (uint16_t)(packed_vals); \
		*(uint16_t *)(v_plane + chroma_pos) =                          \
			(uint16_t)(packed_vals >> 16);                         \
	} while (false)

static FORCE_INLINE uint32_t min_uint32(uint32_t a, uint32_t b)
{
	return a < b ? a : b;
}

void compress_uyvx_to_i420(const uint8_t *input, uint32_t in_linesize,
			   uint32_t start_y, uint32_t end_y, uint8_t *output[],
			   const uint32_t out_linesize[])
{
	uint8_t *lum_plane = output[0];
	uint8_t *u_plane = output[1];
	uint8_t *v_plane = output[2];
	uint32_t width = min_uint32(in_linesize, out_linesize[0]);
	uint32_t y;

	__m128i lum_mask = _mm_set1_epi32(0x0000FF00);
	__m128i uv_mask = _mm_set1_epi16(0x00FF);

	for (y = start_y; y < end_y; y += 2) {
		uint32_t y_pos = y * in_linesize;
		uint32_t chroma_y_pos = (y >> 1) * out_linesize[1];
		uint32_t lum_y_pos = y * out_linesize[0];
		uint32_t x;

		for (x = 0; x < width; x += 4) {
			const uint8_t *img = input + y_pos + x * 4;
			uint32_t lum_pos0 = lum_y_pos + x;
			uint32_t lum_pos1 = lum_pos0 + out_linesize[0];

			__m128i line1 = _mm_load_si128((const __m128i *)img);
			__m128i line2 = _mm_load_si128(
				(const __m128i *)(img + in_linesize));

			pack_shift(lum_plane, lum_pos0, lum_pos1, line1, line2,
				   lum_mask, 1);
			pack_ch_2plane(u_plane, v_plane,
				       chroma_y_pos + (x >> 1), line1, line2,
				       uv_mask);
		}
	}
}

void compress_uyvx_to_nv12(const uint8_t *input, uint32_t in_linesize,
			   uint32_t start_y, uint32_t end_y, uint8_t *output[],
			   const uint32_t out_linesize[])
{
	uint8_t *lum_plane = output[0];
	uint8_t *chroma_plane = output[1];
	uint32_t width = min_uint32(in_linesize, out_linesize[0]);
	uint32_t y;

	__m128i lum_mask = _mm_set1_epi32(0x0000FF00);
	__m128i uv_mask = _mm_set1_epi16(0x00FF);

	for (y = start_y; y < end_y; y += 2) {
		uint32_t y_pos = y * in_linesize;
		uint32_t chroma_y_pos = (y >> 1) * out_linesize[1];
		uint32_t lum_y_pos = y * out_linesize[0];
		uint32_t x;

		for (x = 0; x < width; x += 4) {
			const uint8_t *img = input + y_pos + x * 4;
			uint32_t lum_pos0 = lum_y_pos + x;
			uint32_t lum_pos1 = lum_pos0 + out_linesize[0];

			__m128i line1 = _mm_load_si128((const __m128i *)img);
			__m128i line2 = _mm_load_si128(
				(const __m128i *)(img + in_linesize));

			pack_shift(lum_plane, lum_pos0, lum_pos1, line1, line2,
				   lum_mask, 1);
			pack_ch_1plane(chroma_plane, chroma_y_pos + x, line1,
				       line2, uv_mask);
		}
	}
}

void convert_uyvx_to_i444(const uint8_t *input, uint32_t in_linesize,
			  uint32_t start_y, uint32_t end_y, uint8_t *output[],
			  const uint32_t out_linesize[])
{
	uint8_t *lum_plane = output[0];
	uint8_t *u_plane = output[1];
	uint8_t *v_plane = output[2];
	uint32_t width = min_uint32(in_linesize, out_linesize[0]);
	uint32_t y;

	__m128i lum_mask = _mm_set1_epi32(0x0000FF00);
	__m128i u_mask = _mm_set1_epi32(0x000000FF);
	__m128i v_mask = _mm_set1_epi32(0x00FF0000);

	for (y = start_y; y < end_y; y += 2) {
		uint32_t y_pos = y * in_linesize;
		uint32_t lum_y_pos = y * out_linesize[0];
		uint32_t x;

		for (x = 0; x < width; x += 4) {
			const uint8_t *img = input + y_pos + x * 4;
			uint32_t lum_pos0 = lum_y_pos + x;
			uint32_t lum_pos1 = lum_pos0 + out_linesize[0];

			__m128i line1 = _mm_load_si128((const __m128i *)img);
			__m128i line2 = _mm_load_si128(
				(const __m128i *)(img + in_linesize));

			pack_shift(lum_plane, lum_pos0, lum_pos1, line1, line2,
				   lum_mask, 1);
			pack_val(u_plane, lum_pos0, lum_pos1, line1, line2,
				 u_mask);
			pack_shift(v_plane, lum_pos0, lum_pos1, line1, line2,
				   v_mask, 2);
		}
	}
}

void decompress_420(const uint8_t *const input[], const uint32_t in_linesize[],
		    uint32_t start_y, uint32_t end_y, uint8_t *output,
		    uint32_t out_linesize)
{
	uint32_t start_y_d2 = start_y / 2;
	uint32_t width_d2 = in_linesize[0] / 2;
	uint32_t height_d2 = end_y / 2;
	uint32_t y;

	for (y = start_y_d2; y < height_d2; y++) {
		const uint8_t *chroma0 = input[1] + y * in_linesize[1];
		const uint8_t *chroma1 = input[2] + y * in_linesize[2];
		register const uint8_t *lum0, *lum1;
		register uint32_t *output0, *output1;
		uint32_t x;

		lum0 = input[0] + y * 2 * in_linesize[0];
		lum1 = lum0 + in_linesize[0];
		output0 = (uint32_t *)(output + y * 2 * out_linesize);
		output1 = (uint32_t *)((uint8_t *)output0 + out_linesize);

		for (x = 0; x < width_d2; x++) {
			uint32_t out;
			out = (*(chroma0++) << 8) | *(chroma1++);

			*(output0++) = (*(lum0++) << 16) | out;
			*(output0++) = (*(lum0++) << 16) | out;

			*(output1++) = (*(lum1++) << 16) | out;
			*(output1++) = (*(lum1++) << 16) | out;
		}
	}
}

void decompress_nv12(const uint8_t *const input[], const uint32_t in_linesize[],
		     uint32_t start_y, uint32_t end_y, uint8_t *output,
		     uint32_t out_linesize)
{
	uint32_t start_y_d2 = start_y / 2;
	uint32_t width_d2 = min_uint32(in_linesize[0], out_linesize) / 2;
	uint32_t height_d2 = end_y / 2;
	uint32_t y;

	for (y = start_y_d2; y < height_d2; y++) {
		const uint16_t *chroma;
		register const uint8_t *lum0, *lum1;
		register uint32_t *output0, *output1;
		uint32_t x;

		chroma = (const uint16_t *)(input[1] + y * in_linesize[1]);
		lum0 = input[0] + y * 2 * in_linesize[0];
		lum1 = lum0 + in_linesize[0];
		output0 = (uint32_t *)(output + y * 2 * out_linesize);
		output1 = (uint32_t *)((uint8_t *)output0 + out_linesize);

		for (x = 0; x < width_d2; x++) {
			uint32_t out = *(chroma++) << 8;

			*(output0++) = *(lum0++) | out;
			*(output0++) = *(lum0++) | out;

			*(output1++) = *(lum1++) | out;
			*(output1++) = *(lum1++) | out;
		}
	}
}

void decompress_422(const uint8_t *input, uint32_t in_linesize,
		    uint32_t start_y, uint32_t end_y, uint8_t *output,
		    uint32_t out_linesize, bool leading_lum)
{
	uint32_t width_d2 = min_uint32(in_linesize, out_linesize) / 2;
	uint32_t y;

	register const uint32_t *input32;
	register const uint32_t *input32_end;
	register uint32_t *output32;

	if (leading_lum) {
		for (y = start_y; y < end_y; y++) {
			input32 = (const uint32_t *)(input + y * in_linesize);
			input32_end = input32 + width_d2;
			output32 = (uint32_t *)(output + y * out_linesize);

			while (input32 < input32_end) {
				register uint32_t dw = *input32;

				output32[0] = dw;
				dw &= 0xFFFFFF00;
				dw |= (uint8_t)(dw >> 16);
				output32[1] = dw;

				output32 += 2;
				input32++;
			}
		}
	} else {
		for (y = start_y; y < end_y; y++) {
			input32 = (const uint32_t *)(input + y * in_linesize);
			input32_end = input32 + width_d2;
			output32 = (uint32_t *)(output + y * out_linesize);

			while (input32 < input32_end) {
				register uint32_t dw = *input32;

				output32[0] = dw;
				dw &= 0xFFFF00FF;
				dw |= (dw >> 16) & 0xFF00;
				output32[1] = dw;

				output32 += 2;
				input32++;
			}
		}
	}
}