video_out_rgb.c

最新osg包

static void
__dummy_convert_yv12(uint8_t* yuv[], rgb_planar_t* rgb,
			 uint32_t pitches[], uint32_t width, uint32_t height)
{
	uint8_t* y_data = yuv[0];
	uint8_t* u_data = yuv[1];
	uint8_t* v_data = yuv[2];
	uint8_t* r_buffer = rgb->r;
	uint8_t* g_buffer = rgb->g;
	uint8_t* b_buffer = rgb->b;
	uint32_t line_size = width >> 1; /* 2 pixels at once */
	uint32_t samples   = line_size * (height >> 1);


	while(samples--)
	{
		int y, u, v;
		int m1, m2, m3, m4;
		int r, g, b;

		u  = *(u_data++) - 128;
		v  = *(v_data++) - 128;

		m1 = v * dw_factors[0];
		m2 = v * dw_factors[1];
		m3 = u * dw_factors[2];
		m4 = u * dw_factors[3];

		y = *(y_data) << 16;

		r = (int) (y + m1) >> 16;
		g = (int) (y - m2 - m3) >> 16;
		b = (int) (y + m4) >> 16;

		*(r_buffer) = range(r);
		*(g_buffer) = range(g);
		*(b_buffer) = range(b);

		y = *(y_data + 1) << 16;

		r = (int) (y + m1) >> 16;
		g = (int) (y - m2 - m3) >> 16;
		b = (int) (y + m4) >> 16;

		*(r_buffer + 1) = range(r);
		*(g_buffer + 1) = range(g);
		*(b_buffer + 1) = range(b);

		y = *(y_data + pitches[0]) << 16;

		r = (int) (y + m1) >> 16;
		g = (int) (y - m2 - m3) >> 16;
		b = (int) (y + m4) >> 16;

		*(r_buffer + width) = range(r);
		*(g_buffer + width) = range(g);
		*(b_buffer + width) = range(b);

		y = *(y_data + pitches[0] + 1) << 16;

		r = (int) (y + m1) >> 16;
		g = (int) (y - m2 - m3) >> 16;
		b = (int) (y + m4) >> 16;

		*(r_buffer + width + 1) = range(r);
		*(g_buffer + width + 1) = range(g);
		*(b_buffer + width + 1) = range(b);

		y_data   += 2;
		r_buffer += 2;
		g_buffer += 2;
		b_buffer += 2;

		if(!(--line_size))
		{
			line_size = width >> 1;
			y_data   += pitches[0];
			r_buffer += width;
			g_buffer += width;
			b_buffer += width;
		}
	}

}



static const rgbout_converter_t convert_methods[] =
{
/*        <name>   <accel>      <convert yuy2>        <convert yv12>    */
	{    NULL,        0, __dummy_convert_yuy2, __dummy_convert_yv12},
#ifdef COMPILE_ASSEMBLY
	{   "MMX",   MM_MMX,   __mmx_convert_yuy2,   __mmx_convert_yv12},
	{   "SSE",   MM_SSE,   __sse_convert_yuy2,   __sse_convert_yv12},
	{"3DNow!", MM_3DNOW, __3dnow_convert_yuy2, __3dnow_convert_yv12}
#endif
/* currently 3DNow is the best function, therefore it's preferred on AMD cpus */
};






static void
__pack_argb(rgb_planar_t* data, void* dest,
		uint32_t pixels, uint32_t accel)
{
	static const uint32_t alpha[] =
	{
		0xffffffff,
		0xffffffff
	};
	uint8_t* buffer = (uint8_t*) dest;
	uint8_t* r_data = data->r;
	uint8_t* g_data = data->g;
	uint8_t* b_data = data->b;
	uint32_t rest = pixels;

#ifdef COMPILE_ASSEMBLY

	if((accel & MM_MMXEXT) == MM_MMXEXT)
	{
		uint32_t n = pixels >> 2; /* (width * height) / 4 */
		rest = pixels & 3; /* pixels - (n * 4) */


		__asm__ __volatile__(
			"por (%0), %%mm3\n\t"
			:: "r" (alpha)
			: "memory");

		while(n--)
		{
			__asm__ __volatile__(

				"prefetchw 320(%3)\n\t"

				"movd (%0), %%mm0\n\t" /* mm0 = [0 0 0 0 r3 r2 r1 r0] */
				"movd (%1), %%mm1\n\t" /* mm1 = [0 0 0 0 g3 g2 g1 g0] */
				"movd (%2), %%mm2\n\t" /* mm2 = [0 0 0 0 b3 b2 b1 b0] */

				"punpcklbw %%mm3, %%mm0\n\t" /* mm0 = [ff r3 ff r2 ff r1 ff r0] */
				"punpcklbw %%mm1, %%mm2\n\t" /* mm2 = [g3 b3 g2 b2 g1 b1 g0 b0] */

				"movq %%mm2, %%mm4\n\t" /* mm2 = [g3 b3 g2 b2 g1 b1 g0 b0] */
				"punpcklwd %%mm0, %%mm4\n\t" /* mm4 = [ff r1 g1 b1 ff r0 g0 b0] */
				"movntq %%mm4, (%3)\n\t"

				"punpckhwd %%mm0, %%mm2\n\t" /* mm2 = [ff r3 g3 b3 ff r2 g2 b2] */
				"movntq %%mm2, 8(%3)\n\t"

				"addl $4, %0\n\t"
				"addl $4, %1\n\t"
				"addl $4, %2\n\t"
				"addl $16, %3\n\t"

				: "=&r" (r_data), "=&r" (g_data), "=&r" (b_data), "=&r" (buffer)
				: "0" (r_data), "1" (g_data), "2" (b_data), "3" (buffer)
				: "memory");
		}

		__asm__ __volatile__("sfence; emms\n\t");

	}
	else if((accel & MM_MMX) == MM_MMX)
	{
		uint32_t n = pixels >> 2; /* (width * height) / 4 */
		rest = pixels & 3; /* pixels - (n * 4) */


		__asm__ __volatile__(
			"por (%0), %%mm3\n\t"
			:: "r" (alpha)
			: "memory");

		while(n--)
		{
			__asm__ __volatile__(

				"movd (%0), %%mm0\n\t" /* mm0 = [0 0 0 0 r3 r2 r1 r0] */
				"movd (%1), %%mm1\n\t" /* mm1 = [0 0 0 0 g3 g2 g1 g0] */
				"movd (%2), %%mm2\n\t" /* mm2 = [0 0 0 0 b3 b2 b1 b0] */

				"punpcklbw %%mm3, %%mm0\n\t" /* mm0 = [ff r3 ff r2 ff r1 ff r0] */
				"punpcklbw %%mm1, %%mm2\n\t" /* mm2 = [g3 b3 g2 b2 g1 b1 g0 b0] */

				"movq %%mm2, %%mm4\n\t" /* mm2 = [g3 b3 g2 b2 g1 b1 g0 b0] */
				"punpcklwd %%mm0, %%mm4\n\t" /* mm4 = [ff r1 g1 b1 ff r0 g0 b0] */
				"movq %%mm4, (%3)\n\t"

				"punpckhwd %%mm0, %%mm2\n\t" /* mm2 = [ff r3 g3 b3 ff r2 g2 b2] */
				"movq %%mm2, 8(%3)\n\t"

				"addl $4, %0\n\t"
				"addl $4, %1\n\t"
				"addl $4, %2\n\t"
				"addl $16, %3\n\t"

				: "=&r" (r_data), "=&r" (g_data), "=&r" (b_data), "=&r" (buffer)
				: "0" (r_data), "1" (g_data), "2" (b_data), "3" (buffer)
				: "memory");
		}

		__asm__ __volatile__("emms\n\t");

	}
#endif

	while(rest--)
	{
		*(buffer)     = *(b_data++);
		*(buffer + 1) = *(g_data++);
		*(buffer + 2) = *(r_data++);
		*(buffer + 3) = 0xff;

		buffer += 4;
	}

}


static void
__pack_argb1555(rgb_planar_t* data, void* dest,
		uint32_t pixels, uint32_t accel)
{
	static const uint16_t alpha[] =
	{
		0x8000,
		0x8000,
		0x8000,
		0x8000
	};
	uint16_t* buffer = (uint16_t*) dest;
	uint8_t* r_data = data->r;
	uint8_t* g_data = data->g;
	uint8_t* b_data = data->b;
	int32_t rest = pixels;

#ifdef COMPILE_ASSEMBLY

	if((accel & MM_MMXEXT) == MM_MMXEXT)
	{
		uint32_t n = pixels >> 2; /* (width * height) / 4 */
		rest = pixels & 3; /* pixels - (n * 4) */


		__asm__ __volatile__(
			"pxor %%mm3, %%mm3\n\t"
			"movq (%0), %%mm4\n\t"
			:: "r" (alpha)
			: "memory");

		while(n--)
		{
			__asm__ __volatile__(

				"prefetchw 320(%3)\n\t"

				"movd (%0), %%mm0\n\t" /* mm0 = [0 0 0 0 r3 r2 r1 r0] */
				"movd (%1), %%mm1\n\t" /* mm1 = [0 0 0 0 g3 g2 g1 g0] */
				"movd (%2), %%mm2\n\t" /* mm2 = [0 0 0 0 b3 b2 b1 b0] */

				"punpcklbw %%mm3, %%mm0\n\t" /* mm0 = [0 r3 0 r2 0 r1 0 r0] */
				"punpcklbw %%mm3, %%mm1\n\t" /* mm1 = [0 g3 0 g2 0 g1 0 g0] */
				"punpcklbw %%mm3, %%mm2\n\t" /* mm2 = [0 b3 0 b2 0 b1 0 b0] */

				"psrlw $3, %%mm0\n\t" /* word = [00000000 000rrrrr] */
				"psrlw $3, %%mm1\n\t" /* word = [00000000 000ggggg] */
				"psrlw $3, %%mm2\n\t" /* word = [00000000 000bbbbb] */

				"psllw $10, %%mm0\n\t" /* word = [0rrrrr00 00000000] */
				"psllw $5,  %%mm1\n\t" /* word = [000000gg ggg00000] */
				"por %%mm2, %%mm1\n\t" /* word = [000000gg gggbbbbb] */
				"por %%mm1, %%mm0\n\t" /* word = [0rrrrrgg gggbbbbb] */
				"por %%mm4, %%mm0\n\t" /* word = [1rrrrrgg gggbbbbb] */

				"movntq %%mm0, (%3)\n\t"

				"addl $4, %0\n\t"
				"addl $4, %1\n\t"
				"addl $4, %2\n\t"
				"addl $8, %3\n\t"

				: "=&r" (r_data), "=&r" (g_data), "=&r" (b_data), "=&r" (buffer)
				: "0" (r_data), "1" (g_data), "2" (b_data), "3" (buffer)
				: "memory");

		}

		__asm__ __volatile__("sfence; emms\n\t");

	}
	else if((accel & MM_MMX) == MM_MMX)
	{
		uint32_t n = pixels >> 2; /* (width * height) / 4 */
		rest = pixels & 3; /* pixels - (n * 4) */


		__asm__ __volatile__(
			"pxor %%mm3, %%mm3\n\t"
			"movq (%0), %%mm4\n\t"
			:: "r" (alpha)
			: "memory");

		while(n--)
		{
			__asm__ __volatile__(

				"movd (%0), %%mm0\n\t" /* mm0 = [0 0 0 0 r3 r2 r1 r0] */
				"movd (%1), %%mm1\n\t" /* mm1 = [0 0 0 0 g3 g2 g1 g0] */
				"movd (%2), %%mm2\n\t" /* mm2 = [0 0 0 0 b3 b2 b1 b0] */

				"punpcklbw %%mm3, %%mm0\n\t" /* mm0 = [0 r3 0 r2 0 r1 0 r0] */
				"punpcklbw %%mm3, %%mm1\n\t" /* mm1 = [0 g3 0 g2 0 g1 0 g0] */
				"punpcklbw %%mm3, %%mm2\n\t" /* mm2 = [0 b3 0 b2 0 b1 0 b0] */

				"psrlw $3, %%mm0\n\t" /* word = [00000000 000rrrrr] */
				"psrlw $3, %%mm1\n\t" /* word = [00000000 000ggggg] */
				"psrlw $3, %%mm2\n\t" /* word = [00000000 000bbbbb] */

				"psllw $10, %%mm0\n\t" /* word = [0rrrrr00 00000000] */
				"psllw $5,  %%mm1\n\t" /* word = [000000gg ggg00000] */
				"por %%mm2, %%mm1\n\t" /* word = [000000gg gggbbbbb] */
				"por %%mm1, %%mm0\n\t" /* word = [0rrrrrgg gggbbbbb] */
				"por %%mm4, %%mm0\n\t" /* word = [arrrrrgg gggbbbbb] */

				"movq %%mm0, (%3)\n\t"

				"addl $4, %0\n\t"
				"addl $4, %1\n\t"
				"addl $4, %2\n\t"
				"addl $8, %3\n\t"

				: "=&r" (r_data), "=&r" (g_data), "=&r" (b_data), "=&r" (buffer)
				: "0" (r_data), "1" (g_data), "2" (b_data), "3" (buffer)
				: "memory");

		}

		__asm__ __volatile__("emms\n\t");

	}

#endif
	while(rest--)
	{
		uint16_t r5, g5, b5;

		r5 = *(r_data++) >> 3;
		g5 = *(g_data++) >> 3;
		b5 = *(b_data++) >> 3;

		*(buffer++) = (alpha[0] | (r5 << 10) | (g5 << 5) | b5);
	}

}


static void
__pack_rgb32(rgb_planar_t* data, void* dest,
		uint32_t pixels, uint32_t accel)
{
	uint8_t* buffer = (uint8_t*) dest;
	uint8_t* r_data = data->r;
	uint8_t* g_data = data->g;
	uint8_t* b_data = data->b;
	uint32_t rest = pixels;

#ifdef COMPILE_ASSEMBLY

	if((accel & MM_MMXEXT) == MM_MMXEXT)
	{
		uint32_t n = pixels >> 2; /* (width * height) / 4 */
		rest = pixels & 3; /* pixels - (n * 4) */


		__asm__ __volatile__(
			"pxor %%mm3, %%mm3\n\t"
			::: "memory");

		while(n--)
		{
			__asm__ __volatile__(

				"prefetchw 320(%0)\n\t"

				"movd (%0), %%mm0\n\t" /* mm0 = [0 0 0 0 r3 r2 r1 r0] */
				"movd (%1), %%mm1\n\t" /* mm1 = [0 0 0 0 g3 g2 g1 g0] */
				"movd (%2), %%mm2\n\t" /* mm2 = [0 0 0 0 b3 b2 b1 b0] */

				"punpcklbw %%mm3, %%mm0\n\t" /* mm0 = [0 r3 0 r2 0 r1 0 r0] */
				"punpcklbw %%mm1, %%mm2\n\t" /* mm2 = [g3 b3 g2 b2 g1 b1 g0 b0] */

				"movq %%mm2, %%mm4\n\t" /* mm2 = [g3 b3 g2 b2 g1 b1 g0 b0] */
				"punpcklwd %%mm0, %%mm4\n\t" /* mm4 = [0 r1 g1 b1 0 r0 g0 b0] */
				"movntq %%mm4, (%3)\n\t"

				"punpckhwd %%mm0, %%mm2\n\t" /* mm2 = [0 r3 g3 b3 0 r2 g2 b2] */
				"movntq %%mm2, 8(%3)\n\t"

				"addl  $4, %0\n\t"
				"addl  $4, %1\n\t"
				"addl  $4, %2\n\t"
				"addl $16, %3\n\t"

				: "=&r" (r_data), "=&r" (g_data), "=&r" (b_data), "=&r" (buffer)
				: "0" (r_data), "1" (g_data), "2" (b_data), "3" (buffer)
				: "memory");
		}

		__asm__ __volatile__("sfence; emms\n\t");

	}
	else if((accel & MM_MMX) == MM_MMX)
	{
		uint32_t n = pixels >> 2; /* (width * height) / 4 */
		rest = pixels & 3; /* pixels - (n * 4) */


		__asm__ __volatile__(
			"pxor %%mm3, %%mm3\n\t"
			::: "memory");

		while(n--)
		{
			__asm__ __volatile__(

				"movd (%0), %%mm0\n\t" /* mm0 = [0 0 0 0 r3 r2 r1 r0] */
				"movd (%1), %%mm1\n\t" /* mm1 = [0 0 0 0 g3 g2 g1 g0] */
				"movd (%2), %%mm2\n\t" /* mm2 = [0 0 0 0 b3 b2 b1 b0] */

				"punpcklbw %%mm3, %%mm0\n\t" /* mm0 = [0 r3 0 r2 0 r1 0 r0] */
				"punpcklbw %%mm1, %%mm2\n\t" /* mm2 = [g3 b3 g2 b2 g1 b1 g0 b0] */

				"movq %%mm2, %%mm4\n\t" /* mm2 = [g3 b3 g2 b2 g1 b1 g0 b0] */
				"punpcklwd %%mm0, %%mm4\n\t" /* mm4 = [0 r1 g1 b1 0 r0 g0 b0] */
				"movq %%mm4, (%3)\n\t"

				"punpckhwd %%mm0, %%mm2\n\t" /* mm2 = [0 r3 g3 b3 0 r2 g2 b2] */
				"movq %%mm2, 8(%3)\n\t"

				"addl $4, %0\n\t"
				"addl $4, %1\n\t"
				"addl $4, %2\n\t"
				"addl $16, %3\n\t"

				: "=&r" (r_data), "=&r" (g_data), "=&r" (b_data), "=&r" (buffer)
				: "0" (r_data), "1" (g_data), "2" (b_data), "3" (buffer)
				: "memory");
		}

		__asm__ __volatile__("emms\n\t");
	}
#endif

	while(rest--)
	{
		*(buffer)     = *(b_data++);
		*(buffer + 1) = *(g_data++);
		*(buffer + 2) = *(r_data++);

		buffer += 4;
	}

}


static void
__pack_rgb24(rgb_planar_t* data, void* dest,
		uint32_t pixels, uint32_t accel)
{
	uint8_t* buffer = (uint8_t*) dest;
	uint8_t* r_data = data->r;
	uint8_t* g_data = data->g;
	uint8_t* b_data = data->b;
	uint32_t rest = pixels;


/* MMXEXT doesn't speed up here */
#ifdef COMPILE_ASSEMBLY

	if((accel & MM_MMX) == MM_MMX)
	{
		uint32_t n = pixels >> 2; /* (width * height) / 4 */
		rest = pixels & 3; /* pixels - (n * 4) */


		__asm__ __volatile__(
			"pxor %%mm3, %%mm3\n\t"