mirror/mpv
0
0
Fork 0
mirror of https://github.com/mpv-player/mpv.git synced 2024-09-20 12:02:23 +02:00
Code

liba52 asm optimizations ported to amd64

Browse Source 
git-svn-id: svn://svn.mplayerhq.hu/mplayer/trunk@16174 b3059339-0415-0410-9bf9-f77b7e298cf2
This commit is contained in:
aurel 2005-08-05 13:33:50 +00:00
parent 4bf7522a2e
commit 236d514567
5 changed files with 624 additions and 608 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
liba52/a52_internal.h
View File

@ -41,6 +41,20 @@
#define DELTA_BIT_NONE (2)
#define DELTA_BIT_RESERVED (3)
#ifdef ARCH_X86_64
# define REG_a "rax"
# define REG_d "rdx"
# define REG_S "rsi"
# define REG_D "rdi"
# define REG_BP "rbp"
#else
# define REG_a "eax"
# define REG_d "edx"
# define REG_S "esi"
# define REG_D "edi"
# define REG_BP "ebp"
#endif
void bit_allocate (a52_state_t * state, a52_ba_t * ba, int bndstart,
int start, int end, int fastleak, int slowleak,
uint8_t * exp, int8_t * bap);

620
liba52/downmix.c
View File

File diff suppressed because it is too large Load Diff

274
liba52/imdct.c
View File

@ -101,7 +101,7 @@ static uint8_t bit_reverse_256[] = {
0x03, 0x23, 0x13, 0x33, 0x0b, 0x2b, 0x1b, 0x3b,
0x07, 0x27, 0x17, 0x37, 0x0f, 0x2f, 0x1f, 0x3f};
#ifdef ARCH_X86
#if defined(ARCH_X86) || defined(ARCH_X86_64)
// NOTE: SSE needs 16byte alignment or it will segfault
//
static complex_t __attribute__((aligned(16))) buf[128];
@ -442,8 +442,8 @@ imdct_do_512_altivec(sample_t data[],sample_t delay[], sample_t bias)
int k;
int p,q;
int m;
int two_m;
int two_m_plus_one;
long two_m;
long two_m_plus_one;
sample_t tmp_b_i;
sample_t tmp_b_r;
@ -747,7 +747,7 @@ imdct_do_512_altivec(sample_t data[],sample_t delay[], sample_t bias)
// Stuff below this line is borrowed from libac3
#include "srfftp.h"
#ifdef ARCH_X86
#if defined(ARCH_X86) || defined(ARCH_X86_64)
#ifndef HAVE_3DNOW
#define HAVE_3DNOW 1
#endif
@ -768,9 +768,9 @@ imdct_do_512_sse(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;
int two_m_plus_one_shl3;
long two_m;
long two_m_plus_one;
long two_m_plus_one_shl3;
complex_t *buf_offset;
/* sample_t tmp_a_i;
@ -788,33 +788,33 @@ imdct_do_512_sse(sample_t data[],sample_t delay[], sample_t bias)
/* Pre IFFT complex multiply plus IFFT cmplx conjugate */
/* Bit reversed shuffling */
asm volatile(
"xorl %%esi, %%esi \n\t"
"leal "MANGLE(bit_reverse_512)", %%eax \n\t"
"movl $1008, %%edi \n\t"
"pushl %%ebp \n\t" //use ebp without telling gcc
"xor %%"REG_S", %%"REG_S" \n\t"
"lea "MANGLE(bit_reverse_512)", %%"REG_a"\n\t"
"mov $1008, %%"REG_D" \n\t"
"push %%"REG_BP" \n\t" //use ebp without telling gcc
".balign 16 \n\t"
"1: \n\t"
"movlps (%0, %%esi), %%xmm0 \n\t" // XXXI
"movhps 8(%0, %%edi), %%xmm0 \n\t" // RXXI
"movlps 8(%0, %%esi), %%xmm1 \n\t" // XXXi
"movhps (%0, %%edi), %%xmm1 \n\t" // rXXi
"movlps (%0, %%"REG_S"), %%xmm0 \n\t" // XXXI
"movhps 8(%0, %%"REG_D"), %%xmm0 \n\t" // RXXI
"movlps 8(%0, %%"REG_S"), %%xmm1 \n\t" // XXXi
"movhps (%0, %%"REG_D"), %%xmm1 \n\t" // rXXi
"shufps $0x33, %%xmm1, %%xmm0 \n\t" // irIR
"movaps "MANGLE(sseSinCos1c)"(%%esi), %%xmm2\n\t"
"movaps "MANGLE(sseSinCos1c)"(%%"REG_S"), %%xmm2\n\t"
"mulps %%xmm0, %%xmm2 \n\t"
"shufps $0xB1, %%xmm0, %%xmm0 \n\t" // riRI
"mulps "MANGLE(sseSinCos1d)"(%%esi), %%xmm0\n\t"
"mulps "MANGLE(sseSinCos1d)"(%%"REG_S"), %%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"
"movzb (%%"REG_a"), %%"REG_d" \n\t"
"movzb 1(%%"REG_a"), %%"REG_BP" \n\t"
"movlps %%xmm2, (%1, %%"REG_d", 8) \n\t"
"movhps %%xmm2, (%1, %%"REG_BP", 8) \n\t"
"add $16, %%"REG_S" \n\t"
"add $2, %%"REG_a" \n\t" // avoid complex addressing for P4 crap
"sub $16, %%"REG_D" \n\t"
"jnc 1b \n\t"
"pop %%"REG_BP" \n\t"//no we didnt touch ebp *g*
:: "r" (data), "r" (buf)
: "%"REG_S, "%"REG_D, "%"REG_a, "%"REG_d
);
@ -850,44 +850,44 @@ imdct_do_512_sse(sample_t data[],sample_t delay[], sample_t bias)
asm volatile(
"xorps %%xmm1, %%xmm1 \n\t"
"xorps %%xmm2, %%xmm2 \n\t"
"movl %0, %%esi \n\t"
"mov %0, %%"REG_S" \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]
"movlps (%%"REG_S"), %%xmm0\n\t" //buf[p]
"movlps 8(%%"REG_S"), %%xmm1\n\t" //buf[q]
"movhps (%%"REG_S"), %%xmm0\n\t" //buf[p]
"movhps 8(%%"REG_S"), %%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"
"movaps %%xmm0, (%%"REG_S")\n\t"
"add $16, %%"REG_S" \n\t"
"cmp %1, %%"REG_S" \n\t"
" jb 1b \n\t"
:: "g" (buf), "r" (buf + 128)
: "%esi"
: "%"REG_S
);
/* 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"
"mov %0, %%"REG_S" \n\t"
".balign 16 \n\t"
"1: \n\t"
"movaps 16(%%esi), %%xmm2 \n\t" //r2,i2,r3,i3
"movaps 16(%%"REG_S"), %%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
"movaps (%%"REG_S"), %%xmm0 \n\t" //r0,i0,r1,i1
"movaps (%%"REG_S"), %%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"
"movaps %%xmm0, (%%"REG_S") \n\t"
"movaps %%xmm1, 16(%%"REG_S") \n\t"
"add $32, %%"REG_S" \n\t"
"cmp %1, %%"REG_S" \n\t"
" jb 1b \n\t"
:: "g" (buf), "r" (buf + 128)
: "%esi"
: "%"REG_S
);
/* 3. iteration */
@ -902,11 +902,11 @@ imdct_do_512_sse(sample_t data[],sample_t delay[], sample_t bias)
"movaps 16+"MANGLE(sseW2)", %%xmm7\n\t"
"xorps %%xmm5, %%xmm5 \n\t"
"xorps %%xmm2, %%xmm2 \n\t"
"movl %0, %%esi \n\t"
"mov %0, %%"REG_S" \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 32(%%"REG_S"), %%xmm2 \n\t" //r4,i4,r5,i5
"movaps 48(%%"REG_S"), %%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"
@ -915,8 +915,8 @@ imdct_do_512_sse(sample_t data[],sample_t delay[], sample_t bias)
"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
"movaps (%%"REG_S"), %%xmm0 \n\t" //r0,i0,r1,i1
"movaps 16(%%"REG_S"), %%xmm1 \n\t" //r2,i2,r3,i3
"addps %%xmm4, %%xmm2 \n\t"
"addps %%xmm5, %%xmm3 \n\t"
"movaps %%xmm2, %%xmm4 \n\t"
@ -925,15 +925,15 @@ imdct_do_512_sse(sample_t data[],sample_t delay[], sample_t bias)
"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"
"movaps %%xmm2, (%%"REG_S") \n\t"
"movaps %%xmm3, 16(%%"REG_S") \n\t"
"movaps %%xmm0, 32(%%"REG_S") \n\t"
"movaps %%xmm1, 48(%%"REG_S") \n\t"
"add $64, %%"REG_S" \n\t"
"cmp %1, %%"REG_S" \n\t"
" jb 1b \n\t"
:: "g" (buf), "r" (buf + 128)
: "%esi"
: "%"REG_S
);
/* 4-7. iterations */
@ -943,52 +943,52 @@ imdct_do_512_sse(sample_t data[],sample_t delay[], sample_t bias)
two_m_plus_one_shl3 = (two_m_plus_one<<3);
buf_offset = buf+128;
asm volatile(
"movl %0, %%esi \n\t"
"mov %0, %%"REG_S" \n\t"
".balign 16 \n\t"
"1: \n\t"
"xorl %%edi, %%edi \n\t" // k
"leal (%%esi, %3), %%edx \n\t"
"xor %%"REG_D", %%"REG_D" \n\t" // k
"lea (%%"REG_S", %3), %%"REG_d" \n\t"
"2: \n\t"
"movaps (%%edx, %%edi), %%xmm1 \n\t"
"movaps (%4, %%edi, 2), %%xmm2 \n\t"
"movaps (%%"REG_d", %%"REG_D"), %%xmm1 \n\t"
"movaps (%4, %%"REG_D", 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"
"mulps 16(%4, %%"REG_D", 2), %%xmm1 \n\t"
"movaps (%%"REG_S", %%"REG_D"), %%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"
"movaps %%xmm1, (%%"REG_S", %%"REG_D") \n\t"
"movaps %%xmm0, (%%"REG_d", %%"REG_D") \n\t"
"add $16, %%"REG_D" \n\t"
"cmp %3, %%"REG_D" \n\t" //FIXME (opt) count against 0
"jb 2b \n\t"
"add %2, %%"REG_S" \n\t"
"cmp %1, %%"REG_S" \n\t"
" jb 1b \n\t"
:: "g" (buf), "m" (buf_offset), "m" (two_m_plus_one_shl3), "r" (two_m<<3),
"r" (sseW[m])
: "%esi", "%edi", "%edx"
: "%"REG_S, "%"REG_D, "%"REG_d
);
}
/* Post IFFT complex multiply plus IFFT complex conjugate*/
asm volatile(
"movl $-1024, %%esi \n\t"
"mov $-1024, %%"REG_S" \n\t"
".balign 16 \n\t"
"1: \n\t"
"movaps (%0, %%esi), %%xmm0 \n\t"
"movaps (%0, %%esi), %%xmm1 \n\t"
"movaps (%0, %%"REG_S"), %%xmm0 \n\t"
"movaps (%0, %%"REG_S"), %%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"
"mulps 1024+"MANGLE(sseSinCos1c)"(%%"REG_S"), %%xmm1\n\t"
"mulps 1024+"MANGLE(sseSinCos1d)"(%%"REG_S"), %%xmm0\n\t"
"addps %%xmm1, %%xmm0 \n\t"
"movaps %%xmm0, (%0, %%esi) \n\t"
"addl $16, %%esi \n\t"
"movaps %%xmm0, (%0, %%"REG_S") \n\t"
"add $16, %%"REG_S" \n\t"
" jnz 1b \n\t"
:: "r" (buf+128)
: "%esi"
: "%"REG_S
);
@ -998,54 +998,54 @@ imdct_do_512_sse(sample_t data[],sample_t delay[], sample_t bias)
/* Window and convert to real valued signal */
asm volatile(
"xorl %%edi, %%edi \n\t" // 0
"xorl %%esi, %%esi \n\t" // 0
"xor %%"REG_D", %%"REG_D" \n\t" // 0
"xor %%"REG_S", %%"REG_S" \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 ?
"movlps (%0, %%"REG_S"), %%xmm0 \n\t" // ? ? A ?
"movlps 8(%0, %%"REG_S"), %%xmm1 \n\t" // ? ? C ?
"movhps -16(%0, %%"REG_D"), %%xmm1 \n\t" // ? D C ?
"movhps -8(%0, %%"REG_D"), %%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"
"mulps "MANGLE(sseWindow)"(%%"REG_S"), %%xmm0\n\t"
"addps (%2, %%"REG_S"), %%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"
"movaps %%xmm0, (%1, %%"REG_S") \n\t"
"add $16, %%"REG_S" \n\t"
"sub $16, %%"REG_D" \n\t"
"cmp $512, %%"REG_S" \n\t"
" jb 1b \n\t"
:: "r" (buf+64), "r" (data_ptr), "r" (delay_ptr), "m" (bias)
: "%esi", "%edi"
: "%"REG_S, "%"REG_D
);
data_ptr+=128;
delay_ptr+=128;
// window_ptr+=128;
asm volatile(
"movl $1024, %%edi \n\t" // 512
"xorl %%esi, %%esi \n\t" // 0
"mov $1024, %%"REG_D" \n\t" // 512
"xor %%"REG_S", %%"REG_S" \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
"movlps (%0, %%"REG_S"), %%xmm0 \n\t" // ? ? ? A
"movlps 8(%0, %%"REG_S"), %%xmm1 \n\t" // ? ? ? C
"movhps -16(%0, %%"REG_D"), %%xmm1 \n\t" // D ? ? C
"movhps -8(%0, %%"REG_D"), %%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"
"mulps 512+"MANGLE(sseWindow)"(%%"REG_S"), %%xmm0\n\t"
"addps (%2, %%"REG_S"), %%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"
"movaps %%xmm0, (%1, %%"REG_S") \n\t"
"add $16, %%"REG_S" \n\t"
"sub $16, %%"REG_D" \n\t"
"cmp $512, %%"REG_S" \n\t"
" jb 1b \n\t"
:: "r" (buf), "r" (data_ptr), "r" (delay_ptr), "m" (bias)
: "%esi", "%edi"
: "%"REG_S, "%"REG_D
);
data_ptr+=128;
// window_ptr+=128;
@ -1054,48 +1054,48 @@ imdct_do_512_sse(sample_t data[],sample_t delay[], sample_t bias)
delay_ptr = delay;
asm volatile(
"xorl %%edi, %%edi \n\t" // 0
"xorl %%esi, %%esi \n\t" // 0
"xor %%"REG_D", %%"REG_D" \n\t" // 0
"xor %%"REG_S", %%"REG_S" \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
"movlps (%0, %%"REG_S"), %%xmm0 \n\t" // ? ? ? A
"movlps 8(%0, %%"REG_S"), %%xmm1 \n\t" // ? ? ? C
"movhps -16(%0, %%"REG_D"), %%xmm1 \n\t" // D ? ? C
"movhps -8(%0, %%"REG_D"), %%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"
"mulps 1024+"MANGLE(sseWindow)"(%%"REG_S"), %%xmm0\n\t"
"movaps %%xmm0, (%1, %%"REG_S") \n\t"
"add $16, %%"REG_S" \n\t"
"sub $16, %%"REG_D" \n\t"
"cmp $512, %%"REG_S" \n\t"
" jb 1b \n\t"
:: "r" (buf+64), "r" (delay_ptr)
: "%esi", "%edi"
: "%"REG_S, "%"REG_D
);
delay_ptr+=128;
// window_ptr-=128;
asm volatile(
"movl $1024, %%edi \n\t" // 1024
"xorl %%esi, %%esi \n\t" // 0
"mov $1024, %%"REG_D" \n\t" // 1024
"xor %%"REG_S", %%"REG_S" \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 ?
"movlps (%0, %%"REG_S"), %%xmm0 \n\t" // ? ? A ?
"movlps 8(%0, %%"REG_S"), %%xmm1 \n\t" // ? ? C ?
"movhps -16(%0, %%"REG_D"), %%xmm1 \n\t" // ? D C ?
"movhps -8(%0, %%"REG_D"), %%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"
"mulps 1536+"MANGLE(sseWindow)"(%%"REG_S"), %%xmm0\n\t"
"movaps %%xmm0, (%1, %%"REG_S") \n\t"
"add $16, %%"REG_S" \n\t"
"sub $16, %%"REG_D" \n\t"
"cmp $512, %%"REG_S" \n\t"
" jb 1b \n\t"
:: "r" (buf), "r" (delay_ptr)
: "%esi", "%edi"
: "%"REG_S, "%"REG_D
);
}
#endif //arch_x86
#endif // ARCH_X86 || ARCH_X86_64
void
imdct_do_256(sample_t data[],sample_t delay[],sample_t bias)
@ -1242,7 +1242,7 @@ void imdct_init (uint32_t mm_accel)
xcos1[i] = -cos ((M_PI / 2048) * (8 * i + 1));
xsin1[i] = -sin ((M_PI / 2048) * (8 * i + 1));
}
#ifdef ARCH_X86
#if defined(ARCH_X86) || defined(ARCH_X86_64)
for (i = 0; i < 128; i++) {
sseSinCos1c[2*i+0]= xcos1[i];
sseSinCos1c[2*i+1]= -xcos1[i];
@ -1264,7 +1264,7 @@ void imdct_init (uint32_t mm_accel)
w[i][k].imag = sin (-M_PI * k / j);
}
}
#ifdef ARCH_X86
#if defined(ARCH_X86) || defined(ARCH_X86_64)
for (i = 1; i < 7; i++) {
j = 1 << i;
for (k = 0; k < j; k+=2) {
@ -1307,10 +1307,10 @@ void imdct_init (uint32_t mm_accel)
sseWindow[384 + 2*i+0]= imdct_window[126 - 2*i+1];
sseWindow[384 + 2*i+1]= -imdct_window[126 - 2*i+0];
}
#endif // arch_x86
#endif // ARCH_X86 || ARCH_X86_64
imdct_512 = imdct_do_512;
#ifdef ARCH_X86
#if defined(ARCH_X86) || defined(ARCH_X86_64)
if(mm_accel & MM_ACCEL_X86_SSE)
{
fprintf (stderr, "Using SSE optimized IMDCT transform\n");
@ -1329,7 +1329,7 @@ void imdct_init (uint32_t mm_accel)
imdct_512 = imdct_do_512_3dnow;
}
else
#endif // arch_x86
#endif // ARCH_X86 || ARCH_X86_64
#ifdef HAVE_ALTIVEC
if (mm_accel & MM_ACCEL_PPC_ALTIVEC)
{

4
liba52/resample.c
View File

@ -15,7 +15,7 @@ int (* a52_resample) (float * _f, int16_t * s16)=NULL;
#include "resample_c.c"
#ifdef ARCH_X86
#if defined(ARCH_X86) || defined(ARCH_X86_64)
#include "resample_mmx.c"
#endif
@ -26,7 +26,7 @@ int (* a52_resample) (float * _f, int16_t * s16)=NULL;
void* a52_resample_init(uint32_t mm_accel,int flags,int chans){
void* tmp;
#ifdef ARCH_X86
#if defined(ARCH_X86) || defined(ARCH_X86_64)
if(mm_accel&MM_ACCEL_X86_MMX){
tmp=a52_resample_MMX(flags,chans);
if(tmp){

320
liba52/resample_mmx.c
View File

@ -7,6 +7,9 @@
and it would mean (C / MMX2 / MMX / 3DNOW) versions
*/
#include "a52_internal.h"
static uint64_t attribute_used __attribute__((aligned(8))) magicF2W= 0x43c0000043c00000LL;
static uint64_t attribute_used __attribute__((aligned(8))) wm1010= 0xFFFF0000FFFF0000LL;
static uint64_t attribute_used __attribute__((aligned(8))) wm0101= 0x0000FFFF0000FFFFLL;
@ -15,36 +18,36 @@ static uint64_t attribute_used __attribute__((aligned(8))) wm1100= 0xFFFFFFFF000
static int a52_resample_MONO_to_5_MMX(float * _f, int16_t * s16){
int32_t * f = (int32_t *) _f;
asm volatile(
"movl $-512, %%esi \n\t"
"mov $-512, %%"REG_S" \n\t"
"movq "MANGLE(magicF2W)", %%mm7 \n\t"
"movq "MANGLE(wm1100)", %%mm3 \n\t"
"movq "MANGLE(wm0101)", %%mm4 \n\t"
"movq "MANGLE(wm1010)", %%mm5 \n\t"
"pxor %%mm6, %%mm6 \n\t"
"1: \n\t"
"movq (%1, %%esi, 2), %%mm0 \n\t"
"movq 8(%1, %%esi, 2), %%mm1 \n\t"
"leal (%%esi, %%esi, 4), %%edi \n\t"
"movq (%1, %%"REG_S", 2), %%mm0 \n\t"
"movq 8(%1, %%"REG_S", 2), %%mm1\n\t"
"lea (%%"REG_S", %%"REG_S", 4), %%"REG_D"\n\t"
"psubd %%mm7, %%mm0 \n\t"
"psubd %%mm7, %%mm1 \n\t"
"packssdw %%mm1, %%mm0 \n\t"
"movq %%mm0, %%mm1 \n\t"
"pand %%mm4, %%mm0 \n\t"
"pand %%mm5, %%mm1 \n\t"
"movq %%mm6, (%0, %%edi) \n\t" // 0 0 0 0
"movd %%mm0, 8(%0, %%edi) \n\t" // A 0
"movq %%mm6, (%0, %%"REG_D") \n\t" // 0 0 0 0
"movd %%mm0, 8(%0, %%"REG_D") \n\t" // A 0
"pand %%mm3, %%mm0 \n\t"
"movd %%mm6, 12(%0, %%edi) \n\t" // 0 0
"movd %%mm1, 16(%0, %%edi) \n\t" // 0 B
"movd %%mm6, 12(%0, %%"REG_D") \n\t" // 0 0
"movd %%mm1, 16(%0, %%"REG_D") \n\t" // 0 B
"pand %%mm3, %%mm1 \n\t"
"movd %%mm6, 20(%0, %%edi) \n\t" // 0 0
"movq %%mm0, 24(%0, %%edi) \n\t" // 0 0 C 0
"movq %%mm1, 32(%0, %%edi) \n\t" // 0 0 0 B
"addl $8, %%esi \n\t"
"movd %%mm6, 20(%0, %%"REG_D") \n\t" // 0 0
"movq %%mm0, 24(%0, %%"REG_D") \n\t" // 0 0 C 0
"movq %%mm1, 32(%0, %%"REG_D") \n\t" // 0 0 0 B
"add $8, %%"REG_S" \n\t"
" jnz 1b \n\t"
"emms \n\t"
:: "r" (s16+1280), "r" (f+256)
:"%esi", "%edi", "memory"
:"%"REG_S, "%"REG_D, "memory"
);
return 5*256;
}
@ -54,29 +57,29 @@ static int a52_resample_STEREO_to_2_MMX(float * _f, int16_t * s16){
/* benchmark scores are 0.3% better with SSE but we would need to set bias=0 and premultiply it
#ifdef HAVE_SSE
asm volatile(
"movl $-1024, %%esi \n\t"
"mov $-1024, %%"REG_S" \n\t"
"1: \n\t"
"cvtps2pi (%1, %%esi), %%mm0 \n\t"
"cvtps2pi 1024(%1, %%esi), %%mm2\n\t"
"cvtps2pi (%1, %%"REG_S"), %%mm0\n\t"
"cvtps2pi 1024(%1, %%"REG_S"), %%mm2\n\t"
"movq %%mm0, %%mm1 \n\t"
"punpcklwd %%mm2, %%mm0 \n\t"
"punpckhwd %%mm2, %%mm1 \n\t"
"movq %%mm0, (%0, %%esi) \n\t"
"movq %%mm1, 8(%0, %%esi) \n\t"
"addl $16, %%esi \n\t"
"movq %%mm0, (%0, %%"REG_S") \n\t"
"movq %%mm1, 8(%0, %%"REG_S") \n\t"
"add $16, %%"REG_S" \n\t"
" jnz 1b \n\t"
"emms \n\t"
:: "r" (s16+512), "r" (f+256)
:"%esi", "memory"
:"%"REG_S, "memory"
);*/
asm volatile(
"movl $-1024, %%esi \n\t"
"mov $-1024, %%"REG_S" \n\t"
"movq "MANGLE(magicF2W)", %%mm7 \n\t"
"1: \n\t"
"movq (%1, %%esi), %%mm0 \n\t"
"movq 8(%1, %%esi), %%mm1 \n\t"
"movq 1024(%1, %%esi), %%mm2 \n\t"
"movq 1032(%1, %%esi), %%mm3 \n\t"
"movq (%1, %%"REG_S"), %%mm0 \n\t"
"movq 8(%1, %%"REG_S"), %%mm1 \n\t"
"movq 1024(%1, %%"REG_S"), %%mm2\n\t"
"movq 1032(%1, %%"REG_S"), %%mm3\n\t"
"psubd %%mm7, %%mm0 \n\t"
"psubd %%mm7, %%mm1 \n\t"
"psubd %%mm7, %%mm2 \n\t"
@ -86,13 +89,13 @@ static int a52_resample_STEREO_to_2_MMX(float * _f, int16_t * s16){
"movq %%mm0, %%mm1 \n\t"
"punpcklwd %%mm2, %%mm0 \n\t"
"punpckhwd %%mm2, %%mm1 \n\t"
"movq %%mm0, (%0, %%esi) \n\t"
"movq %%mm1, 8(%0, %%esi) \n\t"
"addl $16, %%esi \n\t"
"movq %%mm0, (%0, %%"REG_S") \n\t"
"movq %%mm1, 8(%0, %%"REG_S") \n\t"
"add $16, %%"REG_S" \n\t"
" jnz 1b \n\t"
"emms \n\t"
:: "r" (s16+512), "r" (f+256)
:"%esi", "memory"
:"%"REG_S, "memory"
);
return 2*256;
}
@ -100,23 +103,23 @@ static int a52_resample_STEREO_to_2_MMX(float * _f, int16_t * s16){
static int a52_resample_3F_to_5_MMX(float * _f, int16_t * s16){
int32_t * f = (int32_t *) _f;
asm volatile(
"movl $-1024, %%esi \n\t"
"mov $-1024, %%"REG_S" \n\t"
"movq "MANGLE(magicF2W)", %%mm7 \n\t"
"pxor %%mm6, %%mm6 \n\t"
"movq %%mm7, %%mm5 \n\t"
"punpckldq %%mm6, %%mm5 \n\t"
"1: \n\t"
"movd (%1, %%esi), %%mm0 \n\t"
"punpckldq 2048(%1, %%esi), %%mm0\n\t"
"movd 1024(%1, %%esi), %%mm1 \n\t"
"punpckldq 4(%1, %%esi), %%mm1 \n\t"
"movd 2052(%1, %%esi), %%mm2 \n\t"
"movd (%1, %%"REG_S"), %%mm0 \n\t"
"punpckldq 2048(%1, %%"REG_S"), %%mm0\n\t"
"movd 1024(%1, %%"REG_S"), %%mm1\n\t"
"punpckldq 4(%1, %%"REG_S"), %%mm1\n\t"
"movd 2052(%1, %%"REG_S"), %%mm2\n\t"
"movq %%mm7, %%mm3 \n\t"
"punpckldq 1028(%1, %%esi), %%mm3\n\t"
"movd 8(%1, %%esi), %%mm4 \n\t"
"punpckldq 2056(%1, %%esi), %%mm4\n\t"
"leal (%%esi, %%esi, 4), %%edi \n\t"
"sarl $1, %%edi \n\t"
"punpckldq 1028(%1, %%"REG_S"), %%mm3\n\t"
"movd 8(%1, %%"REG_S"), %%mm4 \n\t"
"punpckldq 2056(%1, %%"REG_S"), %%mm4\n\t"
"lea (%%"REG_S", %%"REG_S", 4), %%"REG_D"\n\t"
"sar $1, %%"REG_D" \n\t"
"psubd %%mm7, %%mm0 \n\t"
"psubd %%mm7, %%mm1 \n\t"
"psubd %%mm5, %%mm2 \n\t"
@ -125,29 +128,28 @@ static int a52_resample_3F_to_5_MMX(float * _f, int16_t * s16){
"packssdw %%mm6, %%mm0 \n\t"
"packssdw %%mm2, %%mm1 \n\t"
"packssdw %%mm4, %%mm3 \n\t"
"movq %%mm0, (%0, %%edi) \n\t"
"movq %%mm1, 8(%0, %%edi) \n\t"
"movq %%mm3, 16(%0, %%edi) \n\t"
"movd 1032(%1, %%esi), %%mm1 \n\t"
"punpckldq 12(%1, %%esi), %%mm1\n\t"
"movd 2060(%1, %%esi), %%mm2 \n\t"
"movq %%mm0, (%0, %%"REG_D") \n\t"
"movq %%mm1, 8(%0, %%"REG_D") \n\t"
"movq %%mm3, 16(%0, %%"REG_D") \n\t"
"movd 1032(%1, %%"REG_S"), %%mm1\n\t"
"punpckldq 12(%1, %%"REG_S"), %%mm1\n\t"
"movd 2060(%1, %%"REG_S"), %%mm2\n\t"
"movq %%mm7, %%mm3 \n\t"
"punpckldq 1036(%1, %%esi), %%mm3\n\t"
"punpckldq 1036(%1, %%"REG_S"), %%mm3\n\t"
"pxor %%mm0, %%mm0 \n\t"
"psubd %%mm7, %%mm1 \n\t"
"psubd %%mm5, %%mm2 \n\t"
"psubd %%mm7, %%mm3 \n\t"
"packssdw %%mm1, %%mm0 \n\t"
"packssdw %%mm3, %%mm2 \n\t"
"movq %%mm0, 24(%0, %%edi) \n\t"
"movq %%mm2, 32(%0, %%edi) \n\t"
"movq %%mm0, 24(%0, %%"REG_D") \n\t"
"movq %%mm2, 32(%0, %%"REG_D") \n\t"
"addl $16, %%esi \n\t"
"add $16, %%"REG_S" \n\t"
" jnz 1b \n\t"
"emms \n\t"
:: "r" (s16+1280), "r" (f+256)
:"%esi", "%edi", "memory"
:"%"REG_S, "%"REG_D, "memory"
);
return 5*256;
}
@ -155,23 +157,23 @@ static int a52_resample_3F_to_5_MMX(float * _f, int16_t * s16){
static int a52_resample_2F_2R_to_4_MMX(float * _f, int16_t * s16){
int32_t * f = (int32_t *) _f;
asm volatile(
"movl $-1024, %%esi \n\t"
"mov $-1024, %%"REG_S" \n\t"
"movq "MANGLE(magicF2W)", %%mm7 \n\t"
"1: \n\t"
"movq (%1, %%esi), %%mm0 \n\t"
"movq 8(%1, %%esi), %%mm1 \n\t"
"movq 1024(%1, %%esi), %%mm2 \n\t"
"movq 1032(%1, %%esi), %%mm3 \n\t"
"movq (%1, %%"REG_S"), %%mm0 \n\t"
"movq 8(%1, %%"REG_S"), %%mm1 \n\t"
"movq 1024(%1, %%"REG_S"), %%mm2\n\t"
"movq 1032(%1, %%"REG_S"), %%mm3\n\t"
"psubd %%mm7, %%mm0 \n\t"
"psubd %%mm7, %%mm1 \n\t"
"psubd %%mm7, %%mm2 \n\t"
"psubd %%mm7, %%mm3 \n\t"
"packssdw %%mm1, %%mm0 \n\t"
"packssdw %%mm3, %%mm2 \n\t"
"movq 2048(%1, %%esi), %%mm3 \n\t"
"movq 2056(%1, %%esi), %%mm4 \n\t"
"movq 3072(%1, %%esi), %%mm5 \n\t"
"movq 3080(%1, %%esi), %%mm6 \n\t"
"movq 2048(%1, %%"REG_S"), %%mm3\n\t"
"movq 2056(%1, %%"REG_S"), %%mm4\n\t"
"movq 3072(%1, %%"REG_S"), %%mm5\n\t"
"movq 3080(%1, %%"REG_S"), %%mm6\n\t"
"psubd %%mm7, %%mm3 \n\t"
"psubd %%mm7, %%mm4 \n\t"
"psubd %%mm7, %%mm5 \n\t"
@ -190,15 +192,15 @@ static int a52_resample_2F_2R_to_4_MMX(float * _f, int16_t * s16){
"punpckhdq %%mm3, %%mm2 \n\t"
"punpckldq %%mm4, %%mm1 \n\t"
"punpckhdq %%mm4, %%mm5 \n\t"
"movq %%mm0, (%0, %%esi,2) \n\t"
"movq %%mm2, 8(%0, %%esi,2) \n\t"
"movq %%mm1, 16(%0, %%esi,2) \n\t"
"movq %%mm5, 24(%0, %%esi,2) \n\t"
"addl $16, %%esi \n\t"
"movq %%mm0, (%0, %%"REG_S",2) \n\t"
"movq %%mm2, 8(%0, %%"REG_S",2) \n\t"
"movq %%mm1, 16(%0, %%"REG_S",2)\n\t"
"movq %%mm5, 24(%0, %%"REG_S",2)\n\t"
"add $16, %%"REG_S" \n\t"
" jnz 1b \n\t"
"emms \n\t"
:: "r" (s16+1024), "r" (f+256)
:"%esi", "memory"
:"%"REG_S, "memory"
);
return 4*256;
}
@ -206,23 +208,23 @@ static int a52_resample_2F_2R_to_4_MMX(float * _f, int16_t * s16){
static int a52_resample_3F_2R_to_5_MMX(float * _f, int16_t * s16){
int32_t * f = (int32_t *) _f;
asm volatile(
"movl $-1024, %%esi \n\t"
"mov $-1024, %%"REG_S" \n\t"
"movq "MANGLE(magicF2W)", %%mm7 \n\t"
"1: \n\t"
"movd (%1, %%esi), %%mm0 \n\t"
"punpckldq 2048(%1, %%esi), %%mm0\n\t"
"movd 3072(%1, %%esi), %%mm1 \n\t"
"punpckldq 4096(%1, %%esi), %%mm1\n\t"
"movd 1024(%1, %%esi), %%mm2 \n\t"
"punpckldq 4(%1, %%esi), %%mm2 \n\t"
"movd 2052(%1, %%esi), %%mm3 \n\t"
"punpckldq 3076(%1, %%esi), %%mm3\n\t"
"movd 4100(%1, %%esi), %%mm4 \n\t"
"punpckldq 1028(%1, %%esi), %%mm4\n\t"
"movd 8(%1, %%esi), %%mm5 \n\t"
"punpckldq 2056(%1, %%esi), %%mm5\n\t"
"leal (%%esi, %%esi, 4), %%edi \n\t"
"sarl $1, %%edi \n\t"
"movd (%1, %%"REG_S"), %%mm0 \n\t"
"punpckldq 2048(%1, %%"REG_S"), %%mm0\n\t"
"movd 3072(%1, %%"REG_S"), %%mm1\n\t"
"punpckldq 4096(%1, %%"REG_S"), %%mm1\n\t"
"movd 1024(%1, %%"REG_S"), %%mm2\n\t"
"punpckldq 4(%1, %%"REG_S"), %%mm2\n\t"
"movd 2052(%1, %%"REG_S"), %%mm3\n\t"
"punpckldq 3076(%1, %%"REG_S"), %%mm3\n\t"
"movd 4100(%1, %%"REG_S"), %%mm4\n\t"
"punpckldq 1028(%1, %%"REG_S"), %%mm4\n\t"
"movd 8(%1, %%"REG_S"), %%mm5 \n\t"
"punpckldq 2056(%1, %%"REG_S"), %%mm5\n\t"
"lea (%%"REG_S", %%"REG_S", 4), %%"REG_D"\n\t"
"sar $1, %%"REG_D" \n\t"
"psubd %%mm7, %%mm0 \n\t"
"psubd %%mm7, %%mm1 \n\t"
"psubd %%mm7, %%mm2 \n\t"
@ -232,32 +234,32 @@ static int a52_resample_3F_2R_to_5_MMX(float * _f, int16_t * s16){
"packssdw %%mm1, %%mm0 \n\t"
"packssdw %%mm3, %%mm2 \n\t"
"packssdw %%mm5, %%mm4 \n\t"
"movq %%mm0, (%0, %%edi) \n\t"
"movq %%mm2, 8(%0, %%edi) \n\t"
"movq %%mm4, 16(%0, %%edi) \n\t"
"movq %%mm0, (%0, %%"REG_D") \n\t"
"movq %%mm2, 8(%0, %%"REG_D") \n\t"
"movq %%mm4, 16(%0, %%"REG_D") \n\t"
"movd 3080(%1, %%esi), %%mm0 \n\t"
"punpckldq 4104(%1, %%esi), %%mm0\n\t"
"movd 1032(%1, %%esi), %%mm1 \n\t"
"punpckldq 12(%1, %%esi), %%mm1\n\t"
"movd 2060(%1, %%esi), %%mm2 \n\t"
"punpckldq 3084(%1, %%esi), %%mm2\n\t"
"movd 4108(%1, %%esi), %%mm3 \n\t"
"punpckldq 1036(%1, %%esi), %%mm3\n\t"
"movd 3080(%1, %%"REG_S"), %%mm0\n\t"
"punpckldq 4104(%1, %%"REG_S"), %%mm0\n\t"
"movd 1032(%1, %%"REG_S"), %%mm1\n\t"
"punpckldq 12(%1, %%"REG_S"), %%mm1\n\t"
"movd 2060(%1, %%"REG_S"), %%mm2\n\t"
"punpckldq 3084(%1, %%"REG_S"), %%mm2\n\t"
"movd 4108(%1, %%"REG_S"), %%mm3\n\t"
"punpckldq 1036(%1, %%"REG_S"), %%mm3\n\t"
"psubd %%mm7, %%mm0 \n\t"
"psubd %%mm7, %%mm1 \n\t"
"psubd %%mm7, %%mm2 \n\t"
"psubd %%mm7, %%mm3 \n\t"
"packssdw %%mm1, %%mm0 \n\t"
"packssdw %%mm3, %%mm2 \n\t"
"movq %%mm0, 24(%0, %%edi) \n\t"
"movq %%mm2, 32(%0, %%edi) \n\t"
"movq %%mm0, 24(%0, %%"REG_D") \n\t"
"movq %%mm2, 32(%0, %%"REG_D") \n\t"
"addl $16, %%esi \n\t"
"add $16, %%"REG_S" \n\t"
" jnz 1b \n\t"
"emms \n\t"
:: "r" (s16+1280), "r" (f+256)
:"%esi", "%edi", "memory"
:"%"REG_S, "%"REG_D, "memory"
);
return 5*256;
}
@ -265,14 +267,14 @@ static int a52_resample_3F_2R_to_5_MMX(float * _f, int16_t * s16){
static int a52_resample_MONO_LFE_to_6_MMX(float * _f, int16_t * s16){
int32_t * f = (int32_t *) _f;
asm volatile(
"movl $-1024, %%esi \n\t"
"mov $-1024, %%"REG_S" \n\t"
"movq "MANGLE(magicF2W)", %%mm7 \n\t"
"pxor %%mm6, %%mm6 \n\t"
"1: \n\t"
"movq 1024(%1, %%esi), %%mm0 \n\t"
"movq 1032(%1, %%esi), %%mm1 \n\t"
"movq (%1, %%esi), %%mm2 \n\t"
"movq 8(%1, %%esi), %%mm3 \n\t"
"movq 1024(%1, %%"REG_S"), %%mm0\n\t"
"movq 1032(%1, %%"REG_S"), %%mm1\n\t"
"movq (%1, %%"REG_S"), %%mm2 \n\t"
"movq 8(%1, %%"REG_S"), %%mm3 \n\t"
"psubd %%mm7, %%mm0 \n\t"
"psubd %%mm7, %%mm1 \n\t"
"psubd %%mm7, %%mm2 \n\t"
@ -282,22 +284,22 @@ static int a52_resample_MONO_LFE_to_6_MMX(float * _f, int16_t * s16){
"movq %%mm0, %%mm1 \n\t"
"punpcklwd %%mm2, %%mm0 \n\t"
"punpckhwd %%mm2, %%mm1 \n\t"
"leal (%%esi, %%esi, 2), %%edi \n\t"
"movq %%mm6, (%0, %%edi) \n\t"
"movd %%mm0, 8(%0, %%edi) \n\t"
"lea (%%"REG_S", %%"REG_S", 2), %%"REG_D"\n\t"
"movq %%mm6, (%0, %%"REG_D") \n\t"
"movd %%mm0, 8(%0, %%"REG_D") \n\t"
"punpckhdq %%mm0, %%mm0 \n\t"
"movq %%mm6, 12(%0, %%edi) \n\t"
"movd %%mm0, 20(%0, %%edi) \n\t"
"movq %%mm6, 24(%0, %%edi) \n\t"
"movd %%mm1, 32(%0, %%edi) \n\t"
"movq %%mm6, 12(%0, %%"REG_D") \n\t"
"movd %%mm0, 20(%0, %%"REG_D") \n\t"
"movq %%mm6, 24(%0, %%"REG_D") \n\t"
"movd %%mm1, 32(%0, %%"REG_D") \n\t"
"punpckhdq %%mm1, %%mm1 \n\t"
"movq %%mm6, 36(%0, %%edi) \n\t"
"movd %%mm1, 44(%0, %%edi) \n\t"
"addl $16, %%esi \n\t"
"movq %%mm6, 36(%0, %%"REG_D") \n\t"
"movd %%mm1, 44(%0, %%"REG_D") \n\t"
"add $16, %%"REG_S" \n\t"
" jnz 1b \n\t"
"emms \n\t"
:: "r" (s16+1536), "r" (f+256)
:"%esi", "%edi", "memory"
:"%"REG_S, "%"REG_D, "memory"
);
return 6*256;
}
@ -305,17 +307,17 @@ static int a52_resample_MONO_LFE_to_6_MMX(float * _f, int16_t * s16){
static int a52_resample_STEREO_LFE_to_6_MMX(float * _f, int16_t * s16){
int32_t * f = (int32_t *) _f;
asm volatile(
"movl $-1024, %%esi \n\t"
"mov $-1024, %%"REG_S" \n\t"
"movq "MANGLE(magicF2W)", %%mm7 \n\t"
"pxor %%mm6, %%mm6 \n\t"
"1: \n\t"
"movq 1024(%1, %%esi), %%mm0 \n\t"
"movq 2048(%1, %%esi), %%mm1 \n\t"
"movq (%1, %%esi), %%mm5 \n\t"
"movq 1024(%1, %%"REG_S"), %%mm0\n\t"
"movq 2048(%1, %%"REG_S"), %%mm1\n\t"
"movq (%1, %%"REG_S"), %%mm5 \n\t"
"psubd %%mm7, %%mm0 \n\t"
"psubd %%mm7, %%mm1 \n\t"
"psubd %%mm7, %%mm5 \n\t"
"leal (%%esi, %%esi, 2), %%edi \n\t"
"lea (%%"REG_S", %%"REG_S", 2), %%"REG_D"\n\t"
"pxor %%mm4, %%mm4 \n\t"
"packssdw %%mm5, %%mm0 \n\t" // FfAa
@ -327,15 +329,15 @@ static int a52_resample_STEREO_LFE_to_6_MMX(float * _f, int16_t * s16){
"punpckldq %%mm6, %%mm0 \n\t" // 00ba
"punpckhdq %%mm1, %%mm3 \n\t" // BAf0
"movq %%mm0, (%0, %%edi) \n\t" // 00ba
"movq %%mm0, (%0, %%"REG_D") \n\t" // 00ba
"punpckhdq %%mm4, %%mm0 \n\t" // F000
"movq %%mm3, 8(%0, %%edi) \n\t" // BAf0
"movq %%mm0, 16(%0, %%edi) \n\t" // F000
"addl $8, %%esi \n\t"
"movq %%mm3, 8(%0, %%"REG_D") \n\t" // BAf0
"movq %%mm0, 16(%0, %%"REG_D") \n\t" // F000
"add $8, %%"REG_S" \n\t"
" jnz 1b \n\t"
"emms \n\t"
:: "r" (s16+1536), "r" (f+256)
:"%esi", "%edi", "memory"
:"%"REG_S, "%"REG_D, "memory"
);
return 6*256;
}
@ -343,19 +345,19 @@ static int a52_resample_STEREO_LFE_to_6_MMX(float * _f, int16_t * s16){
static int a52_resample_3F_LFE_to_6_MMX(float * _f, int16_t * s16){
int32_t * f = (int32_t *) _f;
asm volatile(
"movl $-1024, %%esi \n\t"
"mov $-1024, %%"REG_S" \n\t"
"movq "MANGLE(magicF2W)", %%mm7 \n\t"
"pxor %%mm6, %%mm6 \n\t"
"1: \n\t"
"movq 1024(%1, %%esi), %%mm0 \n\t"
"movq 3072(%1, %%esi), %%mm1 \n\t"
"movq 2048(%1, %%esi), %%mm4 \n\t"
"movq (%1, %%esi), %%mm5 \n\t"
"movq 1024(%1, %%"REG_S"), %%mm0\n\t"
"movq 3072(%1, %%"REG_S"), %%mm1\n\t"
"movq 2048(%1, %%"REG_S"), %%mm4\n\t"
"movq (%1, %%"REG_S"), %%mm5 \n\t"
"psubd %%mm7, %%mm0 \n\t"
"psubd %%mm7, %%mm1 \n\t"
"psubd %%mm7, %%mm4 \n\t"
"psubd %%mm7, %%mm5 \n\t"
"leal (%%esi, %%esi, 2), %%edi \n\t"
"lea (%%"REG_S", %%"REG_S", 2), %%"REG_D"\n\t"
"packssdw %%mm4, %%mm0 \n\t" // EeAa
"packssdw %%mm5, %%mm1 \n\t" // FfBb
@ -366,16 +368,16 @@ static int a52_resample_3F_LFE_to_6_MMX(float * _f, int16_t * s16){
"punpckldq %%mm6, %%mm0 \n\t" // 00ba
"punpckhdq %%mm1, %%mm1 \n\t" // BABA
"movq %%mm0, (%0, %%edi) \n\t"
"movq %%mm0, (%0, %%"REG_D") \n\t"
"punpckhdq %%mm2, %%mm0 \n\t" // FE00
"punpckldq %%mm1, %%mm2 \n\t" // BAfe
"movq %%mm2, 8(%0, %%edi) \n\t"
"movq %%mm0, 16(%0, %%edi) \n\t"
"addl $8, %%esi \n\t"
"movq %%mm2, 8(%0, %%"REG_D") \n\t"
"movq %%mm0, 16(%0, %%"REG_D") \n\t"
"add $8, %%"REG_S" \n\t"
" jnz 1b \n\t"
"emms \n\t"
:: "r" (s16+1536), "r" (f+256)
:"%esi", "%edi", "memory"
:"%"REG_S, "%"REG_D, "memory"
);
return 6*256;
}
@ -383,21 +385,21 @@ static int a52_resample_3F_LFE_to_6_MMX(float * _f, int16_t * s16){
static int a52_resample_2F_2R_LFE_to_6_MMX(float * _f, int16_t * s16){
int32_t * f = (int32_t *) _f;
asm volatile(
"movl $-1024, %%esi \n\t"
"mov $-1024, %%"REG_S" \n\t"
"movq "MANGLE(magicF2W)", %%mm7 \n\t"
// "pxor %%mm6, %%mm6 \n\t"
"1: \n\t"
"movq 1024(%1, %%esi), %%mm0 \n\t"
"movq 2048(%1, %%esi), %%mm1 \n\t"
"movq 3072(%1, %%esi), %%mm2 \n\t"
"movq 4096(%1, %%esi), %%mm3 \n\t"
"movq (%1, %%esi), %%mm5 \n\t"
"movq 1024(%1, %%"REG_S"), %%mm0\n\t"
"movq 2048(%1, %%"REG_S"), %%mm1\n\t"
"movq 3072(%1, %%"REG_S"), %%mm2\n\t"
"movq 4096(%1, %%"REG_S"), %%mm3\n\t"
"movq (%1, %%"REG_S"), %%mm5 \n\t"
"psubd %%mm7, %%mm0 \n\t"
"psubd %%mm7, %%mm1 \n\t"
"psubd %%mm7, %%mm2 \n\t"
"psubd %%mm7, %%mm3 \n\t"
"psubd %%mm7, %%mm5 \n\t"
"leal (%%esi, %%esi, 2), %%edi \n\t"
"lea (%%"REG_S", %%"REG_S", 2), %%"REG_D"\n\t"
"packssdw %%mm2, %%mm0 \n\t" // CcAa
"packssdw %%mm3, %%mm1 \n\t" // DdBb
@ -414,14 +416,14 @@ static int a52_resample_2F_2R_LFE_to_6_MMX(float * _f, int16_t * s16){
"punpckldq %%mm1, %%mm4 \n\t" // BAf0
"punpckhdq %%mm3, %%mm2 \n\t" // F0DC
"movq %%mm0, (%0, %%edi) \n\t"
"movq %%mm4, 8(%0, %%edi) \n\t"
"movq %%mm2, 16(%0, %%edi) \n\t"
"addl $8, %%esi \n\t"
"movq %%mm0, (%0, %%"REG_D") \n\t"
"movq %%mm4, 8(%0, %%"REG_D") \n\t"
"movq %%mm2, 16(%0, %%"REG_D") \n\t"
"add $8, %%"REG_S" \n\t"
" jnz 1b \n\t"
"emms \n\t"
:: "r" (s16+1536), "r" (f+256)
:"%esi", "%edi", "memory"
:"%"REG_S, "%"REG_D, "memory"
);
return 6*256;
}
@ -429,23 +431,23 @@ static int a52_resample_2F_2R_LFE_to_6_MMX(float * _f, int16_t * s16){
static int a52_resample_3F_2R_LFE_to_6_MMX(float * _f, int16_t * s16){
int32_t * f = (int32_t *) _f;
asm volatile(
"movl $-1024, %%esi \n\t"
"mov $-1024, %%"REG_S" \n\t"
"movq "MANGLE(magicF2W)", %%mm7 \n\t"
// "pxor %%mm6, %%mm6 \n\t"
"1: \n\t"
"movq 1024(%1, %%esi), %%mm0 \n\t"
"movq 3072(%1, %%esi), %%mm1 \n\t"
"movq 4096(%1, %%esi), %%mm2 \n\t"
"movq 5120(%1, %%esi), %%mm3 \n\t"
"movq 2048(%1, %%esi), %%mm4 \n\t"
"movq (%1, %%esi), %%mm5 \n\t"
"movq 1024(%1, %%"REG_S"), %%mm0\n\t"
"movq 3072(%1, %%"REG_S"), %%mm1\n\t"
"movq 4096(%1, %%"REG_S"), %%mm2\n\t"
"movq 5120(%1, %%"REG_S"), %%mm3\n\t"
"movq 2048(%1, %%"REG_S"), %%mm4\n\t"
"movq (%1, %%"REG_S"), %%mm5 \n\t"
"psubd %%mm7, %%mm0 \n\t"
"psubd %%mm7, %%mm1 \n\t"
"psubd %%mm7, %%mm2 \n\t"
"psubd %%mm7, %%mm3 \n\t"
"psubd %%mm7, %%mm4 \n\t"
"psubd %%mm7, %%mm5 \n\t"
"leal (%%esi, %%esi, 2), %%edi \n\t"
"lea (%%"REG_S", %%"REG_S", 2), %%"REG_D"\n\t"
"packssdw %%mm2, %%mm0 \n\t" // CcAa
"packssdw %%mm3, %%mm1 \n\t" // DdBb
@ -462,14 +464,14 @@ static int a52_resample_3F_2R_LFE_to_6_MMX(float * _f, int16_t * s16){
"punpckldq %%mm1, %%mm4 \n\t" // BAfe
"punpckhdq %%mm3, %%mm2 \n\t" // FEDC
"movq %%mm0, (%0, %%edi) \n\t"
"movq %%mm4, 8(%0, %%edi) \n\t"
"movq %%mm2, 16(%0, %%edi) \n\t"
"addl $8, %%esi \n\t"
"movq %%mm0, (%0, %%"REG_D") \n\t"
"movq %%mm4, 8(%0, %%"REG_D") \n\t"
"movq %%mm2, 16(%0, %%"REG_D") \n\t"
"add $8, %%"REG_S" \n\t"
" jnz 1b \n\t"
"emms \n\t"
:: "r" (s16+1536), "r" (f+256)
:"%esi", "%edi", "memory"
:"%"REG_S, "%"REG_D, "memory"
);
return 6*256;
}