imdct.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.

		"shufps $0x33, %%xmm1, %%xmm0		\n\t" // irIR
		"movaps "MANGLE(sseSinCos1c)"(%%esi), %%xmm2\n\t"
		"mulps %%xmm0, %%xmm2			\n\t"
		"shufps $0xB1, %%xmm0, %%xmm0		\n\t" // riRI
		"mulps "MANGLE(sseSinCos1d)"(%%esi), %%xmm0\n\t"
		"subps %%xmm0, %%xmm2			\n\t"
		"movzbl (%%eax), %%edx			\n\t"
		"movzbl 1(%%eax), %%ebp			\n\t"
		"movlps %%xmm2, (%1, %%edx,8)		\n\t"
		"movhps %%xmm2, (%1, %%ebp,8)		\n\t"
		"addl $16, %%esi			\n\t"
		"addl $2, %%eax				\n\t" // avoid complex addressing for P4 crap
		"subl $16, %%edi			\n\t"
		" jnc 1b				\n\t"
		"popl %%ebp				\n\t"//no we didnt touch ebp *g*
		:: "b" (data), "c" (buf)
		: "%esi", "%edi", "%eax", "%edx"
	);


    /* FFT Merge */
/* unoptimized variant
    for (m=1; m < 7; m++) {
	if(m)
	    two_m = (1 << m);
	else
	    two_m = 1;

	two_m_plus_one = (1 << (m+1));

	for(i = 0; i < 128; i += two_m_plus_one) {
	    for(k = 0; k < two_m; k++) {
		p = k + i;
		q = p + two_m;
		tmp_a_r = buf[p].real;
		tmp_a_i = buf[p].imag;
		tmp_b_r = buf[q].real * w[m][k].real - buf[q].imag * w[m][k].imag;
		tmp_b_i = buf[q].imag * w[m][k].real + buf[q].real * w[m][k].imag;
		buf[p].real = tmp_a_r + tmp_b_r;
		buf[p].imag =  tmp_a_i + tmp_b_i;
		buf[q].real = tmp_a_r - tmp_b_r;
		buf[q].imag =  tmp_a_i - tmp_b_i;
	    }
	}
    }
*/
    
    /* 1. iteration */
	// Note w[0][0]={1,0}
	asm volatile(
		"xorps %%xmm1, %%xmm1	\n\t"
		"xorps %%xmm2, %%xmm2	\n\t"
		"movl %0, %%esi		\n\t"
		".balign 16				\n\t"
		"1:			\n\t"
		"movlps (%%esi), %%xmm0	\n\t" //buf[p]
		"movlps 8(%%esi), %%xmm1\n\t" //buf[q]
		"movhps (%%esi), %%xmm0	\n\t" //buf[p]
		"movhps 8(%%esi), %%xmm2\n\t" //buf[q]
		"addps %%xmm1, %%xmm0	\n\t"
		"subps %%xmm2, %%xmm0	\n\t"
		"movaps %%xmm0, (%%esi)	\n\t"
		"addl $16, %%esi	\n\t"
		"cmpl %1, %%esi		\n\t"
		" jb 1b			\n\t"
		:: "g" (buf), "r" (buf + 128)
		: "%esi"
	);
        
    /* 2. iteration */
	// Note w[1]={{1,0}, {0,-1}}
	asm volatile(
		"movaps "MANGLE(ps111_1)", %%xmm7\n\t" // 1,1,1,-1
		"movl %0, %%esi			\n\t"
		".balign 16				\n\t"
		"1:				\n\t"
		"movaps 16(%%esi), %%xmm2	\n\t" //r2,i2,r3,i3
		"shufps $0xB4, %%xmm2, %%xmm2	\n\t" //r2,i2,i3,r3
		"mulps %%xmm7, %%xmm2		\n\t" //r2,i2,i3,-r3
		"movaps (%%esi), %%xmm0		\n\t" //r0,i0,r1,i1
		"movaps (%%esi), %%xmm1		\n\t" //r0,i0,r1,i1
		"addps %%xmm2, %%xmm0		\n\t"
		"subps %%xmm2, %%xmm1		\n\t"
		"movaps %%xmm0, (%%esi)		\n\t"
		"movaps %%xmm1, 16(%%esi)	\n\t"
		"addl $32, %%esi	\n\t"
		"cmpl %1, %%esi		\n\t"
		" jb 1b			\n\t"
		:: "g" (buf), "r" (buf + 128)
		: "%esi"
	);

    /* 3. iteration */
/*
 Note sseW2+0={1,1,sqrt(2),sqrt(2))
 Note sseW2+16={0,0,sqrt(2),-sqrt(2))
 Note sseW2+32={0,0,-sqrt(2),-sqrt(2))
 Note sseW2+48={1,-1,sqrt(2),-sqrt(2))
*/
	asm volatile(
		"movaps 48+"MANGLE(sseW2)", %%xmm6\n\t" 
		"movaps 16+"MANGLE(sseW2)", %%xmm7\n\t" 
		"xorps %%xmm5, %%xmm5		\n\t"
		"xorps %%xmm2, %%xmm2		\n\t"
		"movl %0, %%esi			\n\t"
		".balign 16			\n\t"
		"1:				\n\t"
		"movaps 32(%%esi), %%xmm2	\n\t" //r4,i4,r5,i5
		"movaps 48(%%esi), %%xmm3	\n\t" //r6,i6,r7,i7
		"movaps "MANGLE(sseW2)", %%xmm4	\n\t" //r4,i4,r5,i5
		"movaps 32+"MANGLE(sseW2)", %%xmm5\n\t" //r6,i6,r7,i7
		"mulps %%xmm2, %%xmm4		\n\t"
		"mulps %%xmm3, %%xmm5		\n\t"
		"shufps $0xB1, %%xmm2, %%xmm2	\n\t" //i4,r4,i5,r5
		"shufps $0xB1, %%xmm3, %%xmm3	\n\t" //i6,r6,i7,r7
		"mulps %%xmm6, %%xmm3		\n\t"
		"mulps %%xmm7, %%xmm2		\n\t"
		"movaps (%%esi), %%xmm0		\n\t" //r0,i0,r1,i1
		"movaps 16(%%esi), %%xmm1	\n\t" //r2,i2,r3,i3
		"addps %%xmm4, %%xmm2		\n\t"
		"addps %%xmm5, %%xmm3		\n\t"
		"movaps %%xmm2, %%xmm4		\n\t"
		"movaps %%xmm3, %%xmm5		\n\t"
		"addps %%xmm0, %%xmm2		\n\t"
		"addps %%xmm1, %%xmm3		\n\t"
		"subps %%xmm4, %%xmm0		\n\t"
		"subps %%xmm5, %%xmm1		\n\t"
		"movaps %%xmm2, (%%esi)		\n\t" 
		"movaps %%xmm3, 16(%%esi)	\n\t" 
		"movaps %%xmm0, 32(%%esi)	\n\t" 
		"movaps %%xmm1, 48(%%esi)	\n\t" 
		"addl $64, %%esi	\n\t"
		"cmpl %1, %%esi		\n\t"
		" jb 1b			\n\t"
		:: "g" (buf), "r" (buf + 128)
		: "%esi"
	);

    /* 4-7. iterations */
    for (m=3; m < 7; m++) {
	two_m = (1 << m);
	two_m_plus_one = two_m<<1;
	asm volatile(
		"movl %0, %%esi				\n\t"
		".balign 16				\n\t"
		"1:					\n\t"
		"xorl %%edi, %%edi			\n\t" // k
		"leal (%%esi, %3), %%edx		\n\t"
		"2:					\n\t"
		"movaps (%%edx, %%edi), %%xmm1		\n\t"
		"movaps (%4, %%edi, 2), %%xmm2		\n\t"
		"mulps %%xmm1, %%xmm2			\n\t"
		"shufps $0xB1, %%xmm1, %%xmm1		\n\t"
		"mulps 16(%4, %%edi, 2), %%xmm1		\n\t"
		"movaps (%%esi, %%edi), %%xmm0		\n\t"
		"addps %%xmm2, %%xmm1			\n\t"
		"movaps %%xmm1, %%xmm2			\n\t"
		"addps %%xmm0, %%xmm1			\n\t"
		"subps %%xmm2, %%xmm0			\n\t"
		"movaps %%xmm1, (%%esi, %%edi)		\n\t"
		"movaps %%xmm0, (%%edx, %%edi)		\n\t"
		"addl $16, %%edi			\n\t"
		"cmpl %3, %%edi				\n\t" //FIXME (opt) count against 0 
		" jb 2b					\n\t"
		"addl %2, %%esi				\n\t"
		"cmpl %1, %%esi				\n\t"
		" jb 1b					\n\t"
		:: "g" (buf), "m" (buf+128), "m" (two_m_plus_one<<3), "r" (two_m<<3),
		   "r" (sseW[m])
		: "%esi", "%edi", "%edx"
	);
    }

    /* Post IFFT complex multiply  plus IFFT complex conjugate*/
	asm volatile(
		"movl $-1024, %%esi			\n\t"
		".balign 16				\n\t"
		"1:					\n\t"
		"movaps (%0, %%esi), %%xmm0		\n\t"
		"movaps (%0, %%esi), %%xmm1		\n\t"
		"shufps $0xB1, %%xmm0, %%xmm0		\n\t"
		"mulps 1024+"MANGLE(sseSinCos1c)"(%%esi), %%xmm1\n\t"
		"mulps 1024+"MANGLE(sseSinCos1d)"(%%esi), %%xmm0\n\t"
		"addps %%xmm1, %%xmm0			\n\t"
		"movaps %%xmm0, (%0, %%esi)		\n\t"
		"addl $16, %%esi			\n\t"
		" jnz 1b				\n\t"
		:: "r" (buf+128)
		: "%esi"
	);   

	
    data_ptr = data;
    delay_ptr = delay;
    window_ptr = imdct_window;

    /* Window and convert to real valued signal */
	asm volatile(
		"xorl %%edi, %%edi			\n\t"  // 0
		"xorl %%esi, %%esi			\n\t"  // 0
		"movss %3, %%xmm2			\n\t"  // bias
		"shufps $0x00, %%xmm2, %%xmm2		\n\t"  // bias, bias, ...
		".balign 16				\n\t"
		"1:					\n\t"
		"movlps (%0, %%esi), %%xmm0		\n\t" // ? ? A ?
		"movlps 8(%0, %%esi), %%xmm1		\n\t" // ? ? C ?
		"movhps -16(%0, %%edi), %%xmm1		\n\t" // ? D C ?
		"movhps -8(%0, %%edi), %%xmm0		\n\t" // ? B A ?
		"shufps $0x99, %%xmm1, %%xmm0		\n\t" // D C B A
		"mulps "MANGLE(sseWindow)"(%%esi), %%xmm0\n\t"
		"addps (%2, %%esi), %%xmm0		\n\t"
		"addps %%xmm2, %%xmm0			\n\t"
		"movaps %%xmm0, (%1, %%esi)		\n\t"
		"addl $16, %%esi			\n\t"
		"subl $16, %%edi			\n\t"
		"cmpl $512, %%esi			\n\t" 
		" jb 1b					\n\t"
		:: "r" (buf+64), "r" (data_ptr), "r" (delay_ptr), "m" (bias)
		: "%esi", "%edi"
	);
	data_ptr+=128;
	delay_ptr+=128;
//	window_ptr+=128;
	
	asm volatile(
		"movl $1024, %%edi			\n\t"  // 512
		"xorl %%esi, %%esi			\n\t"  // 0
		"movss %3, %%xmm2			\n\t"  // bias
		"shufps $0x00, %%xmm2, %%xmm2		\n\t"  // bias, bias, ...
		".balign 16				\n\t"
		"1:					\n\t"
		"movlps (%0, %%esi), %%xmm0		\n\t" // ? ? ? A
		"movlps 8(%0, %%esi), %%xmm1		\n\t" // ? ? ? C
		"movhps -16(%0, %%edi), %%xmm1		\n\t" // D ? ? C
		"movhps -8(%0, %%edi), %%xmm0		\n\t" // B ? ? A
		"shufps $0xCC, %%xmm1, %%xmm0		\n\t" // D C B A
		"mulps 512+"MANGLE(sseWindow)"(%%esi), %%xmm0\n\t"
		"addps (%2, %%esi), %%xmm0		\n\t"
		"addps %%xmm2, %%xmm0			\n\t"
		"movaps %%xmm0, (%1, %%esi)		\n\t"
		"addl $16, %%esi			\n\t"
		"subl $16, %%edi			\n\t"
		"cmpl $512, %%esi			\n\t" 
		" jb 1b					\n\t"
		:: "r" (buf), "r" (data_ptr), "r" (delay_ptr), "m" (bias)
		: "%esi", "%edi"
	);
	data_ptr+=128;
//	window_ptr+=128;

    /* The trailing edge of the window goes into the delay line */
    delay_ptr = delay;

	asm volatile(
		"xorl %%edi, %%edi			\n\t"  // 0
		"xorl %%esi, %%esi			\n\t"  // 0
		".balign 16				\n\t"
		"1:					\n\t"
		"movlps (%0, %%esi), %%xmm0		\n\t" // ? ? ? A
		"movlps 8(%0, %%esi), %%xmm1		\n\t" // ? ? ? C
		"movhps -16(%0, %%edi), %%xmm1		\n\t" // D ? ? C 
		"movhps -8(%0, %%edi), %%xmm0		\n\t" // B ? ? A 
		"shufps $0xCC, %%xmm1, %%xmm0		\n\t" // D C B A
		"mulps 1024+"MANGLE(sseWindow)"(%%esi), %%xmm0\n\t"
		"movaps %%xmm0, (%1, %%esi)		\n\t"
		"addl $16, %%esi			\n\t"
		"subl $16, %%edi			\n\t"
		"cmpl $512, %%esi			\n\t" 
		" jb 1b					\n\t"
		:: "r" (buf+64), "r" (delay_ptr)
		: "%esi", "%edi"
	);
	delay_ptr+=128;
//	window_ptr-=128;
	
	asm volatile(
		"movl $1024, %%edi			\n\t"  // 1024
		"xorl %%esi, %%esi			\n\t"  // 0
		".balign 16				\n\t"
		"1:					\n\t"
		"movlps (%0, %%esi), %%xmm0		\n\t" // ? ? A ?
		"movlps 8(%0, %%esi), %%xmm1		\n\t" // ? ? C ?
		"movhps -16(%0, %%edi), %%xmm1		\n\t" // ? D C ? 
		"movhps -8(%0, %%edi), %%xmm0		\n\t" // ? B A ? 
		"shufps $0x99, %%xmm1, %%xmm0		\n\t" // D C B A
		"mulps 1536+"MANGLE(sseWindow)"(%%esi), %%xmm0\n\t"
		"movaps %%xmm0, (%1, %%esi)		\n\t"
		"addl $16, %%esi			\n\t"
		"subl $16, %%edi			\n\t"
		"cmpl $512, %%esi			\n\t" 
		" jb 1b					\n\t"
		:: "r" (buf), "r" (delay_ptr)
		: "%esi", "%edi"
	);
}
#endif //arch_x86

void
imdct_do_256(sample_t data[],sample_t delay[],sample_t bias)
{
    int i,k;
    int p,q;
    int m;
    int two_m;
    int two_m_plus_one;

    sample_t tmp_a_i;
    sample_t tmp_a_r;
    sample_t tmp_b_i;
    sample_t tmp_b_r;

    sample_t *data_ptr;
    sample_t *delay_ptr;
    sample_t *window_ptr;

    complex_t *buf_1, *buf_2;

    buf_1 = &buf[0];
    buf_2 = &buf[64];

    /* Pre IFFT complex multiply plus IFFT cmplx conjugate */
    for(k=0; k<64; k++) { 
	/* X1[k] = X[2*k]  */
	/* X2[k] = X[2*k+1]     */

	p = 2 * (128-2*k-1);
	q = 2 * (2 * k);

	/* Z1[k] = (X1[128-2*k-1] + j * X1[2*k]) * (xcos2[k] + j * xsin2[k]); */ 
	buf_1[k].real =         data[p] * xcos2[k] - data[q] * xsin2[k];
	buf_1[k].imag = -1.0f * (data[q] * xcos2[k] + data[p] * xsin2[k]); 
	/* Z2[k] = (X2[128-2*k-1] + j * X2[2*k]) * (xcos2[k] + j * xsin2[k]); */ 
	buf_2[k].real =          data[p + 1] * xcos2[k] - data[q + 1] * xsin2[k];
	buf_2[k].imag = -1.0f * ( data[q + 1] * xcos2[k] + data[p + 1] * xsin2[k]); 
    }

    /* IFFT Bit reversed shuffling */
    for(i=0; i<64; i++) { 
	k = bit_reverse_256[i];
	if (k < i) {
	    swap_cmplx(&buf_1[i],&buf_1[k]);
	    swap_cmplx(&buf_2[i],&buf_2[k]);
	}
    }

    /* FFT Merge */
    for (m=0; m < 6; m++) {
	two_m = (1 << m);
	two_m_plus_one = (1 << (m+1));

	/* FIXME */
	if(m)
	    two_m = (1 << m);
	else
	    two_m = 1;

	for(k = 0; k < two_m; k++) {
	    for(i = 0; i < 64; i += two_m_plus_one) {
		p = k + i;
		q = p + two_m;
		/* Do block 1 */
		tmp_a_r = buf_1[p].real;
		tmp_a_i = buf_1[p].imag;
		tmp_b_r = buf_1[q].real * w[m][k].real - buf_1[q].imag * w[m][k].imag;
		tmp_b_i = buf_1[q].imag * w[m][k].real + buf_1[q].real * w[m][k].imag;
		buf_1[p].real = tmp_a_r + tmp_b_r;
		buf_1[p].imag =  tmp_a_i + tmp_b_i;
		buf_1[q].real = tmp_a_r - tmp_b_r;
		buf_1[q].imag =  tmp_a_i - tmp_b_i;

		/* Do block 2 */
		tmp_a_r = buf_2[p].real;
		tmp_a_i = buf_2[p].imag;
		tmp_b_r = buf_2[q].real * w[m][k].real - buf_2[q].imag * w[m][k].imag;
		tmp_b_i = buf_2[q].imag * w[m][k].real + buf_2[q].real * w[m][k].imag;
		buf_2[p].real = tmp_a_r + tmp_b_r;
		buf_2[p].imag =  tmp_a_i + tmp_b_i;
		buf_2[q].real = tmp_a_r - tmp_b_r;
		buf_2[q].imag =  tmp_a_i - tmp_b_i;
	    }
	}
    }

    /* Post IFFT complex multiply */
    for( i=0; i < 64; i++) {
	/* y1[n] = z1[n] * (xcos2[n] + j * xs in2[n]) ; */ 
	tmp_a_r =  buf_1[i].real;
	tmp_a_i = -buf_1[i].imag;
	buf_1[i].real =(tmp_a_r * xcos2[i])  -  (tmp_a_i  * xsin2[i]);
	buf_1[i].imag =(tmp_a_r * xsin2[i])  +  (tmp_a_i  * xcos2[i]);
	/* y2[n] = z2[n] * (xcos2[n] + j * xsin2[n]) ; */ 
	tmp_a_r =  buf_2[i].real;
	tmp_a_i = -buf_2[i].imag;
	buf_2[i].real =(tmp_a_r * xcos2[i])  -  (tmp_a_i  * xsin2[i]);
	buf_2[i].imag =(tmp_a_r * xsin2[i])  +  (tmp_a_i  * xcos2[i]);
    }
	
    data_ptr = data;
    delay_ptr = delay;