comparison armv4l/mpegvideo_iwmmxt.c @ 2734:aeea63c97878 libavcodec

Better ARM support for mplayer/ffmpeg, ported from atty fork while playing with some new hardware, I found it's running a forked mplayer -- and it looks like they're following the GPL. The maintainer's page is here: http://atty.jp/?Zaurus/mplayer Unfortunately it's mostly in Japanese, so it's hard to figure out any details. Their code looks quite interesting (at least to those of us w/ ARM CPUs). The patches I've attached are the patches from atty.jp with a couple of modifications by myself: - ported to current CVS - reverted their change of removing SNOW support from ffmpeg - cleaned up their bswap mess - removed DOS-style linebreaks from various files patch by (Bernhard Rosenkraenzer: bero, arklinux org)
author michael
date Thu, 26 May 2005 14:32:46 +0000
parents
children 930e56f92c57
comparison
equal deleted inserted replaced
2733:32336384162e 2734:aeea63c97878
1 #include "../dsputil.h"
2 #include "../mpegvideo.h"
3 #include "../avcodec.h"
4
5 static void dct_unquantize_h263_intra_iwmmxt(MpegEncContext *s,
6 DCTELEM *block, int n, int qscale)
7 {
8 int level, qmul, qadd;
9 int nCoeffs;
10 DCTELEM *block_orig = block;
11
12 assert(s->block_last_index[n]>=0);
13
14 qmul = qscale << 1;
15
16 if (!s->h263_aic) {
17 if (n < 4)
18 level = block[0] * s->y_dc_scale;
19 else
20 level = block[0] * s->c_dc_scale;
21 qadd = (qscale - 1) | 1;
22 }else{
23 qadd = 0;
24 level = block[0];
25 }
26 if(s->ac_pred)
27 nCoeffs=63;
28 else
29 nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
30
31 __asm__ __volatile__ (
32 /* "movd %1, %%mm6 \n\t" //qmul */
33 /* "packssdw %%mm6, %%mm6 \n\t" */
34 /* "packssdw %%mm6, %%mm6 \n\t" */
35 "tbcsth wr6, %[qmul] \n\t"
36 /* "movd %2, %%mm5 \n\t" //qadd */
37 /* "packssdw %%mm5, %%mm5 \n\t" */
38 /* "packssdw %%mm5, %%mm5 \n\t" */
39 "tbcsth wr5, %[qadd] \n\t"
40 "wzero wr7 \n\t" /* "pxor %%mm7, %%mm7 \n\t" */
41 "wzero wr4 \n\t" /* "pxor %%mm4, %%mm4 \n\t" */
42 "wsubh wr7, wr5, wr7 \n\t" /* "psubw %%mm5, %%mm7 \n\t" */
43 "1: \n\t"
44 "wldrd wr2, [%[block]] \n\t" /* "movq (%0, %3), %%mm0 \n\t" */
45 "wldrd wr3, [%[block], #8] \n\t" /* "movq 8(%0, %3), %%mm1 \n\t" */
46 "wmulsl wr0, wr6, wr2 \n\t" /* "pmullw %%mm6, %%mm0 \n\t" */
47 "wmulsl wr1, wr6, wr3 \n\t" /* "pmullw %%mm6, %%mm1 \n\t" */
48 /* "movq (%0, %3), %%mm2 \n\t" */
49 /* "movq 8(%0, %3), %%mm3 \n\t" */
50 "wcmpgtsh wr2, wr4, wr2 \n\t" /* "pcmpgtw %%mm4, %%mm2 \n\t" // block[i] < 0 ? -1 : 0 */
51 "wcmpgtsh wr3, wr4, wr2 \n\t" /* "pcmpgtw %%mm4, %%mm3 \n\t" // block[i] < 0 ? -1 : 0 */
52 "wxor wr0, wr2, wr0 \n\t" /* "pxor %%mm2, %%mm0 \n\t" */
53 "wxor wr1, wr3, wr1 \n\t" /* "pxor %%mm3, %%mm1 \n\t" */
54 "waddh wr0, wr7, wr0 \n\t" /* "paddw %%mm7, %%mm0 \n\t" */
55 "waddh wr1, wr7, wr1 \n\t" /* "paddw %%mm7, %%mm1 \n\t" */
56 "wxor wr2, wr0, wr2 \n\t" /* "pxor %%mm0, %%mm2 \n\t" */
57 "wxor wr3, wr1, wr3 \n\t" /* "pxor %%mm1, %%mm3 \n\t" */
58 "wcmpeqh wr0, wr7, wr0 \n\t" /* "pcmpeqw %%mm7, %%mm0 \n\t" // block[i] == 0 ? -1 : 0 */
59 "wcmpeqh wr1, wr7, wr1 \n\t" /* "pcmpeqw %%mm7, %%mm1 \n\t" // block[i] == 0 ? -1 : 0 */
60 "wandn wr0, wr2, wr0 \n\t" /* "pandn %%mm2, %%mm0 \n\t" */
61 "wandn wr1, wr3, wr1 \n\t" /* "pandn %%mm3, %%mm1 \n\t" */
62 "wstrd wr0, [%[block]] \n\t" /* "movq %%mm0, (%0, %3) \n\t" */
63 "wstrd wr1, [%[block], #8] \n\t" /* "movq %%mm1, 8(%0, %3) \n\t" */
64 "add %[block], %[block], #16 \n\t" /* "addl $16, %3 \n\t" */
65 "subs %[i], %[i], #1 \n\t"
66 "bne 1b \n\t" /* "jng 1b \n\t" */
67 :[block]"+r"(block)
68 :[i]"r"((nCoeffs + 8) / 8), [qmul]"r"(qmul), [qadd]"r"(qadd)
69 :"memory");
70
71 block_orig[0] = level;
72 }
73
74 #if 0
75 static void dct_unquantize_h263_inter_iwmmxt(MpegEncContext *s,
76 DCTELEM *block, int n, int qscale)
77 {
78 int nCoeffs;
79
80 assert(s->block_last_index[n]>=0);
81
82 if(s->ac_pred)
83 nCoeffs=63;
84 else
85 nCoeffs= s->inter_scantable.raster_end[ s->block_last_index[n] ];
86
87 ippiQuantInvInter_Compact_H263_16s_I(block, nCoeffs+1, qscale);
88 }
89 #endif
90
91 void MPV_common_init_iwmmxt(MpegEncContext *s)
92 {
93 s->dct_unquantize_h263_intra = dct_unquantize_h263_intra_iwmmxt;
94 #if 0
95 s->dct_unquantize_h263_inter = dct_unquantize_h263_inter_iwmmxt;
96 #endif
97 }