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

Align output pointer so that we can use movaps instead of movups in dct64_sse;

Browse Source 
1.5% faster decode.


git-svn-id: svn://svn.mplayerhq.hu/mplayer/trunk@23484 b3059339-0415-0410-9bf9-f77b7e298cf2
This commit is contained in:
zuxy 2007-06-06 05:13:13 +00:00
parent 40c9f981a2
commit 739f79a5ff
4 changed files with 9 additions and 19 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
mp3lib/dct64_sse.c
View File

@ -5,17 +5,7 @@
* and mp3lib/dct64_MMX.c
*/
/* NOTE: The following code is suboptimal! It can be improved (at least) by
1. Replace all movups by movaps. (Can Parameter c be always aligned on
a 16-byte boundary?)
2. Rewritten using intrinsics. (GCC generally optimizes intrinsics
better. However, when __m128 locals are involved, GCC may
produce bad code that uses movaps to access a stack not aligned
on a 16-byte boundary, which leads to run-time crashes.)
*/
#include <libavutil/mem.h>
typedef float real;
@ -32,8 +22,8 @@ static const int nnnn[4] __attribute__((aligned(16))) =
void dct64_sse(short *out0,short *out1,real *c)
{
static real __attribute__ ((aligned(16))) b1[0x20];
static real __attribute__ ((aligned(16))) b2[0x20];
static DECLARE_ALIGNED(16, real, b1[0x20]);
static DECLARE_ALIGNED(16, real, b2[0x20]);
static real const one = 1.f;
{
@ -45,9 +35,9 @@ void dct64_sse(short *out0,short *out1,real *c)
asm(
"movaps %2, %%xmm3\n\t"
"shufps $27, %%xmm3, %%xmm3\n\t"
"movups %3, %%xmm1\n\t"
"movaps %3, %%xmm1\n\t"
"movaps %%xmm1, %%xmm4\n\t"
"movups %4, %%xmm2\n\t"
"movaps %4, %%xmm2\n\t"
"shufps $27, %%xmm4, %%xmm4\n\t"
"movaps %%xmm2, %%xmm0\n\t"
"shufps $27, %%xmm0, %%xmm0\n\t"

2
mp3lib/layer1.c
View File

@ -131,7 +131,7 @@ static int do_layer1(struct frame *fr,int single)
int i,stereo = fr->stereo;
unsigned int balloc[2*SBLIMIT];
unsigned int scale_index[2][SBLIMIT];
real fraction[2][SBLIMIT];
DECLARE_ALIGNED(16, real, fraction[2][SBLIMIT]);
// int single = fr->single;
// printf("do_layer1(0x%02X 0x%02X 0x%02X 0x%02X 0x%02X 0x%02X 0x%02X 0x%02X )\n",

2
mp3lib/layer2.c
View File

@ -285,7 +285,7 @@ static int do_layer2(struct frame *fr,int outmode)
int clip=0;
int i,j;
int stereo = fr->stereo;
real fraction[2][4][SBLIMIT]; /* pick_table clears unused subbands */
DECLARE_ALIGNED(16, real, fraction[2][4][SBLIMIT]); /* pick_table clears unused subbands */
unsigned int bit_alloc[64];
int scale[192];
int single = fr->single;

4
mp3lib/layer3.c
View File

@ -1260,8 +1260,8 @@ static int do_layer3(struct frame *fr,int single){
granules = (fr->lsf) ? 1 : 2;
for (gr=0;gr<granules;gr++){
static real hybridIn[2][SBLIMIT][SSLIMIT];
static real hybridOut[2][SSLIMIT][SBLIMIT];
static DECLARE_ALIGNED(16, real, hybridIn[2][SBLIMIT][SSLIMIT]);
static DECLARE_ALIGNED(16, real, hybridOut[2][SSLIMIT][SBLIMIT]);
{ struct gr_info_s *gr_info = &(sideinfo.ch[0].gr[gr]);
int part2bits;