downmix.c

自己移植的linux下的流媒体播放器原代码,支持mms协议,支持ftp和http协议.

		"subps %%xmm0, %%xmm1		\n\t"
		"addps %%xmm0, %%xmm2		\n\t"
		"movaps %%xmm1, (%0, %%esi)	\n\t"
		"movaps %%xmm2, 1024(%0, %%esi)	\n\t"
		"addl $16, %%esi		\n\t"
		" jnz 1b			\n\t"
	:: "r" (samples+256), "m" (bias)
	: "%esi"
	);
}

static void mix32to2_SSE (sample_t * samples, sample_t bias)
{
	asm volatile(
	"movlps %1, %%xmm7		\n\t"
	"shufps $0x00, %%xmm7, %%xmm7	\n\t"
	"movl $-1024, %%esi		\n\t"
	".balign 16\n\t"
	"1:				\n\t"
	"movaps 1024(%0, %%esi), %%xmm0	\n\t" 
	"addps %%xmm7, %%xmm0		\n\t" // common
	"movaps %%xmm0, %%xmm1		\n\t" // common
	"addps (%0, %%esi), %%xmm0	\n\t" 
	"addps 2048(%0, %%esi), %%xmm1	\n\t" 
	"addps 3072(%0, %%esi), %%xmm0	\n\t" 
	"addps 4096(%0, %%esi), %%xmm1	\n\t" 
	"movaps %%xmm0, (%0, %%esi)	\n\t"
	"movaps %%xmm1, 1024(%0, %%esi)	\n\t"
	"addl $16, %%esi		\n\t"
	" jnz 1b			\n\t"
	:: "r" (samples+256), "m" (bias)
	: "%esi"
	);
}

static void mix32toS_SSE (sample_t * samples, sample_t bias)
{
	asm volatile(
	"movlps %1, %%xmm7		\n\t"
	"shufps $0x00, %%xmm7, %%xmm7	\n\t"
	"movl $-1024, %%esi		\n\t"
	".balign 16\n\t"
	"1:				\n\t"
	"movaps 1024(%0, %%esi), %%xmm0	\n\t" 
	"movaps 3072(%0, %%esi), %%xmm2	\n\t" 
	"addps %%xmm7, %%xmm0		\n\t" // common
	"addps 4096(%0, %%esi), %%xmm2	\n\t" // surround	
	"movaps (%0, %%esi), %%xmm1	\n\t" 
	"movaps 2048(%0, %%esi), %%xmm3	\n\t" 
	"subps %%xmm2, %%xmm1		\n\t"	
	"addps %%xmm2, %%xmm3		\n\t"	
	"addps %%xmm0, %%xmm1		\n\t"	
	"addps %%xmm0, %%xmm3		\n\t"	
	"movaps %%xmm1, (%0, %%esi)	\n\t"
	"movaps %%xmm3, 1024(%0, %%esi)	\n\t"
	"addl $16, %%esi		\n\t"
	" jnz 1b			\n\t"
	:: "r" (samples+256), "m" (bias)
	: "%esi"
	);
}

static void move2to1_SSE (sample_t * src, sample_t * dest, sample_t bias)
{
	asm volatile(
		"movlps %2, %%xmm7		\n\t"
		"shufps $0x00, %%xmm7, %%xmm7	\n\t"
		"movl $-1024, %%esi		\n\t"
		".balign 16\n\t"
		"1:				\n\t"
		"movaps (%0, %%esi), %%xmm0	\n\t"  
		"movaps 16(%0, %%esi), %%xmm1	\n\t"  
		"addps 1024(%0, %%esi), %%xmm0	\n\t"
		"addps 1040(%0, %%esi), %%xmm1	\n\t"
		"addps %%xmm7, %%xmm0		\n\t"
		"addps %%xmm7, %%xmm1		\n\t"
		"movaps %%xmm0, (%1, %%esi)	\n\t"
		"movaps %%xmm1, 16(%1, %%esi)	\n\t"
		"addl $32, %%esi		\n\t"
		" jnz 1b			\n\t"
	:: "r" (src+256), "r" (dest+256), "m" (bias)
	: "%esi"
	);
}

static void zero_MMX(sample_t * samples)
{
	asm volatile(
		"movl $-1024, %%esi		\n\t"
		"pxor %%mm0, %%mm0		\n\t"
		".balign 16\n\t"
		"1:				\n\t"
		"movq %%mm0, (%0, %%esi)	\n\t"
		"movq %%mm0, 8(%0, %%esi)	\n\t"
		"movq %%mm0, 16(%0, %%esi)	\n\t"
		"movq %%mm0, 24(%0, %%esi)	\n\t"
		"addl $32, %%esi		\n\t"
		" jnz 1b			\n\t"
		"emms"
	:: "r" (samples+256)
	: "%esi"
	);
}

/*
 I hope dest and src will be at least 8 byte aligned and size
 will devide on 8 without remain
 Note: untested and unused.
*/
static void copy_MMX(void *dest,const void *src,unsigned size)
{
  unsigned i;
  size /= 64;
	for(i=0;i<size;i++)
	{
	    __asm __volatile(
		"movq	%0,   %%mm0\n\t"
		"movq	8%0,  %%mm1\n\t"
		"movq	16%0, %%mm2\n\t"
		"movq	24%0, %%mm3\n\t"
		"movq	32%0, %%mm4\n\t"
		"movq	40%0, %%mm5\n\t"
		"movq	48%0, %%mm6\n\t"
		"movq	56%0, %%mm7\n\t"
		"movq	%%mm0, %1\n\t"
		"movq	%%mm1, 8%1\n\t"
		"movq	%%mm2, 16%1\n\t"
		"movq	%%mm3, 24%1\n\t"
		"movq	%%mm4, 32%1\n\t"
		"movq	%%mm5, 40%1\n\t"
		"movq	%%mm6, 48%1\n\t"
		"movq	%%mm7, 56%1\n\t"
		:
		:"m"(src),"m"(dest));
	}
}

static void downmix_SSE (sample_t * samples, int acmod, int output, sample_t bias,
	      sample_t clev, sample_t slev)
{
    switch (CONVERT (acmod, output & A52_CHANNEL_MASK)) {

    case CONVERT (A52_CHANNEL, A52_CHANNEL2):
	memcpy (samples, samples + 256, 256 * sizeof (sample_t));
	break;

    case CONVERT (A52_CHANNEL, A52_MONO):
    case CONVERT (A52_STEREO, A52_MONO):
    mix_2to1_SSE:
	mix2to1_SSE (samples, samples + 256, bias);
	break;

    case CONVERT (A52_2F1R, A52_MONO):
	if (slev == 0)
	    goto mix_2to1_SSE;
    case CONVERT (A52_3F, A52_MONO):
    mix_3to1_SSE:
	mix3to1_SSE (samples, bias);
	break;

    case CONVERT (A52_3F1R, A52_MONO):
	if (slev == 0)
	    goto mix_3to1_SSE;
    case CONVERT (A52_2F2R, A52_MONO):
	if (slev == 0)
	    goto mix_2to1_SSE;
	mix4to1_SSE (samples, bias);
	break;

    case CONVERT (A52_3F2R, A52_MONO):
	if (slev == 0)
	    goto mix_3to1_SSE;
	mix5to1_SSE (samples, bias);
	break;

    case CONVERT (A52_MONO, A52_DOLBY):
	memcpy (samples + 256, samples, 256 * sizeof (sample_t));
	break;

    case CONVERT (A52_3F, A52_STEREO):
    case CONVERT (A52_3F, A52_DOLBY):
    mix_3to2_SSE:
	mix3to2_SSE (samples, bias);
	break;

    case CONVERT (A52_2F1R, A52_STEREO):
	if (slev == 0)
	    break;
	mix21to2_SSE (samples, samples + 256, bias);
	break;

    case CONVERT (A52_2F1R, A52_DOLBY):
	mix21toS_SSE (samples, bias);
	break;

    case CONVERT (A52_3F1R, A52_STEREO):
	if (slev == 0)
	    goto mix_3to2_SSE;
	mix31to2_SSE (samples, bias);
	break;

    case CONVERT (A52_3F1R, A52_DOLBY):
	mix31toS_SSE (samples, bias);
	break;

    case CONVERT (A52_2F2R, A52_STEREO):
	if (slev == 0)
	    break;
	mix2to1_SSE (samples, samples + 512, bias);
	mix2to1_SSE (samples + 256, samples + 768, bias);
	break;

    case CONVERT (A52_2F2R, A52_DOLBY):
	mix22toS_SSE (samples, bias);
	break;

    case CONVERT (A52_3F2R, A52_STEREO):
	if (slev == 0)
	    goto mix_3to2_SSE;
	mix32to2_SSE (samples, bias);
	break;

    case CONVERT (A52_3F2R, A52_DOLBY):
	mix32toS_SSE (samples, bias);
	break;

    case CONVERT (A52_3F1R, A52_3F):
	if (slev == 0)
	    break;
	mix21to2_SSE (samples, samples + 512, bias);
	break;

    case CONVERT (A52_3F2R, A52_3F):
	if (slev == 0)
	    break;
	mix2to1_SSE (samples, samples + 768, bias);
	mix2to1_SSE (samples + 512, samples + 1024, bias);
	break;

    case CONVERT (A52_3F1R, A52_2F1R):
	mix3to2_SSE (samples, bias);
	memcpy (samples + 512, samples + 768, 256 * sizeof (sample_t));
	break;

    case CONVERT (A52_2F2R, A52_2F1R):
	mix2to1_SSE (samples + 512, samples + 768, bias);
	break;

    case CONVERT (A52_3F2R, A52_2F1R):
	mix3to2_SSE (samples, bias); //FIXME possible bug? (output doesnt seem to be used)
	move2to1_SSE (samples + 768, samples + 512, bias);
	break;

    case CONVERT (A52_3F2R, A52_3F1R):
	mix2to1_SSE (samples + 768, samples + 1024, bias);
	break;

    case CONVERT (A52_2F1R, A52_2F2R):
	memcpy (samples + 768, samples + 512, 256 * sizeof (sample_t));
	break;

    case CONVERT (A52_3F1R, A52_2F2R):
	mix3to2_SSE (samples, bias);
	memcpy (samples + 512, samples + 768, 256 * sizeof (sample_t));
	break;

    case CONVERT (A52_3F2R, A52_2F2R):
	mix3to2_SSE (samples, bias);
	memcpy (samples + 512, samples + 768, 256 * sizeof (sample_t));
	memcpy (samples + 768, samples + 1024, 256 * sizeof (sample_t));
	break;

    case CONVERT (A52_3F1R, A52_3F2R):
	memcpy (samples + 1024, samples + 768, 256 * sizeof (sample_t));
	break;
    }
}

static void upmix_MMX (sample_t * samples, int acmod, int output)
{
    switch (CONVERT (acmod, output & A52_CHANNEL_MASK)) {

    case CONVERT (A52_CHANNEL, A52_CHANNEL2):
	memcpy (samples + 256, samples, 256 * sizeof (sample_t));
	break;

    case CONVERT (A52_3F2R, A52_MONO):
	zero_MMX (samples + 1024);
    case CONVERT (A52_3F1R, A52_MONO):
    case CONVERT (A52_2F2R, A52_MONO):
	zero_MMX (samples + 768);
    case CONVERT (A52_3F, A52_MONO):
    case CONVERT (A52_2F1R, A52_MONO):
	zero_MMX (samples + 512);
    case CONVERT (A52_CHANNEL, A52_MONO):
    case CONVERT (A52_STEREO, A52_MONO):
	zero_MMX (samples + 256);
	break;

    case CONVERT (A52_3F2R, A52_STEREO):
    case CONVERT (A52_3F2R, A52_DOLBY):
	zero_MMX (samples + 1024);
    case CONVERT (A52_3F1R, A52_STEREO):
    case CONVERT (A52_3F1R, A52_DOLBY):
	zero_MMX (samples + 768);
    case CONVERT (A52_3F, A52_STEREO):
    case CONVERT (A52_3F, A52_DOLBY):
    mix_3to2_MMX:
	memcpy (samples + 512, samples + 256, 256 * sizeof (sample_t));
	zero_MMX (samples + 256);
	break;

    case CONVERT (A52_2F2R, A52_STEREO):
    case CONVERT (A52_2F2R, A52_DOLBY):
	zero_MMX (samples + 768);
    case CONVERT (A52_2F1R, A52_STEREO):
    case CONVERT (A52_2F1R, A52_DOLBY):
	zero_MMX (samples + 512);
	break;

    case CONVERT (A52_3F2R, A52_3F):
	zero_MMX (samples + 1024);
    case CONVERT (A52_3F1R, A52_3F):
    case CONVERT (A52_2F2R, A52_2F1R):
	zero_MMX (samples + 768);
	break;

    case CONVERT (A52_3F2R, A52_3F1R):
	zero_MMX (samples + 1024);
	break;

    case CONVERT (A52_3F2R, A52_2F1R):
	zero_MMX (samples + 1024);
    case CONVERT (A52_3F1R, A52_2F1R):
    mix_31to21_MMX:
	memcpy (samples + 768, samples + 512, 256 * sizeof (sample_t));
	goto mix_3to2_MMX;

    case CONVERT (A52_3F2R, A52_2F2R):
	memcpy (samples + 1024, samples + 768, 256 * sizeof (sample_t));
	goto mix_31to21_MMX;
    }
}

static void mix2to1_3dnow (sample_t * dest, sample_t * src, sample_t bias)
{
	asm volatile(
	"movd  %2, %%mm7	\n\t"
	"punpckldq %2, %%mm7	\n\t"
	"movl  $-1024, %%esi	\n\t"
	".balign 16\n\t"
	"1:			\n\t"
	"movq  (%0, %%esi), %%mm0	\n\t" 
	"movq  8(%0, %%esi), %%mm1	\n\t"
	"movq  16(%0, %%esi), %%mm2	\n\t" 
	"movq  24(%0, %%esi), %%mm3	\n\t"
	"pfadd (%1, %%esi), %%mm0	\n\t" 
	"pfadd 8(%1, %%esi), %%mm1	\n\t"
	"pfadd 16(%1, %%esi), %%mm2	\n\t" 
	"pfadd 24(%1, %%esi), %%mm3	\n\t"
	"pfadd %%mm7, %%mm0		\n\t"
	"pfadd %%mm7, %%mm1		\n\t"
	"pfadd %%mm7, %%mm2		\n\t"
	"pfadd %%mm7, %%mm3		\n\t"
	"movq  %%mm0, (%1, %%esi)	\n\t"
	"movq  %%mm1, 8(%1, %%esi)	\n\t"
	"movq  %%mm2, 16(%1, %%esi)	\n\t"
	"movq  %%mm3, 24(%1, %%esi)	\n\t"
	"addl $32, %%esi		\n\t"
	" jnz 1b			\n\t"
	:: "r" (src+256), "r" (dest+256), "m" (bias)
	: "%esi"
	);
}

static void mix3to1_3dnow (sample_t * samples, sample_t bias)
{
	asm volatile(
	"movd  %1, %%mm7	\n\t"
	"punpckldq %1, %%mm7	\n\t"
	"movl $-1024, %%esi	\n\t"
	".balign 16\n\t"
	"1:			\n\t"
	"movq  (%0, %%esi), %%mm0	\n\t" 
	"movq  8(%0, %%esi), %%mm1	\n\t"
	"movq  1024(%0, %%esi), %%mm2	\n\t" 
	"movq  1032(%0, %%esi), %%mm3	\n\t"
	"pfadd 2048(%0, %%esi), %%mm0	\n\t" 
	"pfadd 2056(%0, %%esi), %%mm1	\n\t"
	"pfadd %%mm7, %%mm0		\n\t"
	"pfadd %%mm7, %%mm1		\n\t"
	"pfadd %%mm2, %%mm0		\n\t"
	"pfadd %%mm3, %%mm1		\n\t"
	"movq  %%mm0, (%0, %%esi)	\n\t"
	"movq  %%mm1, 8(%0, %%esi)	\n\t"
	"addl $16, %%esi		\n\t"
	" jnz 1b			\n\t"
	:: "r" (samples+256), "m" (bias)
	: "%esi"
	);
}

static void mix4to1_3dnow (sample_t * samples, sample_t bias)
{
	asm volatile(
	"movd  %1, %%mm7	\n\t"
	"punpckldq %1, %%mm7	\n\t"
	"movl $-1024, %%esi	\n\t"
	".balign 16\n\t"
	"1:			\n\t"
	"movq  (%0, %%esi), %%mm0	\n\t" 
	"movq  8(%0, %%esi), %%mm1	\n\t"
	"movq  1024(%0, %%esi), %%mm2	\n\t" 
	"movq  1032(%0, %%esi), %%mm3	\n\t"
	"pfadd 2048(%0, %%esi), %%mm0	\n\t" 
	"pfadd 2056(%0, %%esi), %%mm1	\n\t"
	"pfadd 3072(%0, %%esi), %%mm2	\n\t" 
	"pfadd 3080(%0, %%esi), %%mm3	\n\t"
	"pfadd %%mm7, %%mm0		\n\t"
	"pfadd %%mm7, %%mm1		\n\t"
	"pfadd %%mm2, %%mm0		\n\t"
	"pfadd %%mm3, %%mm1		\n\t"
	"movq  %%mm0, (%0, %%esi)	\n\t"
	"movq  %%mm1, 8(%0, %%esi)	\n\t"
	"addl $16, %%esi		\n\t"
	" jnz 1b			\n\t"
	:: "r" (samples+256), "m" (bias)
	: "%esi"
	);
}

static void mix5to1_3dnow (sample_t * samples, sample_t bias)
{
	asm volatile(
	"movd  %1, %%mm7	\n\t"
	"punpckldq %1, %%mm7	\n\t"
	"movl $-1024, %%esi	\n\t"
	".balign 16\n\t"
	"1:			\n\t"
	"movq  (%0, %%esi), %%mm0	\n\t" 
	"movq  8(%0, %%esi), %%mm1	\n\t"
	"movq  1024(%0, %%esi), %%mm2	\n\t" 
	"movq  1032(%0, %%esi), %%mm3	\n\t"
	"pfadd 2048(%0, %%esi), %%mm0	\n\t" 
	"pfadd 2056(%0, %%esi), %%mm1	\n\t"
	"pfadd 3072(%0, %%esi), %%mm2	\n\t" 
	"pfadd 3080(%0, %%esi), %%mm3	\n\t"
	"pfadd %%mm7, %%mm0		\n\t"
	"pfadd %%mm7, %%mm1		\n\t"
	"pfadd 4096(%0, %%esi), %%mm2	\n\t" 
	"pfadd 4104(%0, %%esi), %%mm3	\n\t"
	"pfadd %%mm2, %%mm0		\n\t"
	"pfadd %%mm3, %%mm1		\n\t"
	"movq  %%mm0, (%0, %%esi)	\n\t"
	"movq  %%mm1, 8(%0, %%esi)	\n\t"