Mercurial > mplayer.hg
annotate mp3lib/dct64_3dnow.s @ 1294:0a8e0c7ddd0c
video decoder stuff
author | arpi |
---|---|
date | Sun, 08 Jul 2001 12:29:38 +0000 |
parents | 175423b2691e |
children |
rev | line source |
---|---|
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
1 # This code was taken from http://www.mpg123.org |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
2 # See ChangeLog of mpg123-0.59s-pre.1 for detail |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
3 # Applied to mplayer by Nick Kurshev <nickols_k@mail.ru> |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
4 # Partial 3dnow! optimization by Nick Kurshev |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
5 # |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
6 # TODO: optimize scalar 3dnow! code |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
7 # Warning: Phases 7 & 8 are not tested |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
8 # |
1 | 9 |
1277 | 10 .data |
11 .align 8 | |
12 x_plus_minus_3dnow: .long 0x00000000, 0x80000000 | |
13 plus_1f: .float 1.0 | |
14 | |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
15 .text |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
16 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
17 .align 16 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
18 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
19 .globl dct64_MMX_3dnow |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
20 dct64_MMX_3dnow: |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
21 pushl %ebx |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
22 pushl %esi |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
23 pushl %edi |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
24 subl $256,%esp |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
25 movl 280(%esp),%eax |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
26 leal 128(%esp),%edx |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
27 movl 272(%esp),%esi |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
28 movl 276(%esp),%edi |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
29 movl $costab_mmx,%ebx |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
30 orl %ecx,%ecx |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
31 movl %esp,%ecx |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
32 |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
33 /* Phase 1*/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
34 movq (%eax), %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
35 movq 8(%eax), %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
36 movq %mm0, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
37 movq %mm4, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
38 movq 120(%eax), %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
39 movq 112(%eax), %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
40 /* n.b.: pswapd*/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
41 movq %mm1, %mm2 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
42 movq %mm5, %mm6 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
43 psrlq $32, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
44 psrlq $32, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
45 punpckldq %mm2, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
46 punpckldq %mm6, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
47 /**/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
48 pfadd %mm1, %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
49 pfadd %mm5, %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
50 movq %mm0, (%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
51 movq %mm4, 8(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
52 pfsub %mm1, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
53 pfsub %mm5, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
54 pfmul (%ebx), %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
55 pfmul 8(%ebx), %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
56 movd %mm3, 124(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
57 movd %mm7, 116(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
58 psrlq $32, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
59 psrlq $32, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
60 movd %mm3, 120(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
61 movd %mm7, 112(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
62 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
63 movq 16(%eax), %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
64 movq 24(%eax), %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
65 movq %mm0, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
66 movq %mm4, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
67 movq 104(%eax), %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
68 movq 96(%eax), %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
69 /* n.b.: pswapd*/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
70 movq %mm1, %mm2 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
71 movq %mm5, %mm6 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
72 psrlq $32, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
73 psrlq $32, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
74 punpckldq %mm2, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
75 punpckldq %mm6, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
76 /**/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
77 pfadd %mm1, %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
78 pfadd %mm5, %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
79 movq %mm0, 16(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
80 movq %mm4, 24(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
81 pfsub %mm1, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
82 pfsub %mm5, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
83 pfmul 16(%ebx), %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
84 pfmul 24(%ebx), %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
85 movd %mm3, 108(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
86 movd %mm7, 100(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
87 psrlq $32, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
88 psrlq $32, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
89 movd %mm3, 104(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
90 movd %mm7, 96(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
91 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
92 movq 32(%eax), %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
93 movq 40(%eax), %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
94 movq %mm0, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
95 movq %mm4, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
96 movq 88(%eax), %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
97 movq 80(%eax), %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
98 /* n.b.: pswapd*/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
99 movq %mm1, %mm2 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
100 movq %mm5, %mm6 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
101 psrlq $32, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
102 psrlq $32, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
103 punpckldq %mm2, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
104 punpckldq %mm6, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
105 /**/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
106 pfadd %mm1, %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
107 pfadd %mm5, %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
108 movq %mm0, 32(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
109 movq %mm4, 40(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
110 pfsub %mm1, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
111 pfsub %mm5, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
112 pfmul 32(%ebx), %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
113 pfmul 40(%ebx), %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
114 movd %mm3, 92(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
115 movd %mm7, 84(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
116 psrlq $32, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
117 psrlq $32, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
118 movd %mm3, 88(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
119 movd %mm7, 80(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
120 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
121 movq 48(%eax), %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
122 movq 56(%eax), %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
123 movq %mm0, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
124 movq %mm4, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
125 movq 72(%eax), %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
126 movq 64(%eax), %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
127 /* n.b.: pswapd*/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
128 movq %mm1, %mm2 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
129 movq %mm5, %mm6 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
130 psrlq $32, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
131 psrlq $32, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
132 punpckldq %mm2, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
133 punpckldq %mm6, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
134 /**/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
135 pfadd %mm1, %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
136 pfadd %mm5, %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
137 movq %mm0, 48(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
138 movq %mm4, 56(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
139 pfsub %mm1, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
140 pfsub %mm5, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
141 pfmul 48(%ebx), %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
142 pfmul 56(%ebx), %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
143 movd %mm3, 76(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
144 movd %mm7, 68(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
145 psrlq $32, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
146 psrlq $32, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
147 movd %mm3, 72(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
148 movd %mm7, 64(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
149 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
150 /* Phase 2*/ |
1 | 151 |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
152 movq (%edx), %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
153 movq 8(%edx), %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
154 movq %mm0, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
155 movq %mm4, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
156 movq 56(%edx), %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
157 movq 48(%edx), %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
158 /* n.b.: pswapd*/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
159 movq %mm1, %mm2 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
160 movq %mm5, %mm6 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
161 psrlq $32, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
162 psrlq $32, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
163 punpckldq %mm2, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
164 punpckldq %mm6, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
165 /**/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
166 pfadd %mm1, %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
167 pfadd %mm5, %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
168 movq %mm0, (%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
169 movq %mm4, 8(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
170 pfsub %mm1, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
171 pfsub %mm5, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
172 pfmul 64(%ebx), %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
173 pfmul 72(%ebx), %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
174 movd %mm3, 60(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
175 movd %mm7, 52(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
176 psrlq $32, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
177 psrlq $32, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
178 movd %mm3, 56(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
179 movd %mm7, 48(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
180 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
181 movq 16(%edx), %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
182 movq 24(%edx), %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
183 movq %mm0, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
184 movq %mm4, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
185 movq 40(%edx), %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
186 movq 32(%edx), %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
187 /* n.b.: pswapd*/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
188 movq %mm1, %mm2 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
189 movq %mm5, %mm6 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
190 psrlq $32, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
191 psrlq $32, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
192 punpckldq %mm2, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
193 punpckldq %mm6, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
194 /**/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
195 pfadd %mm1, %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
196 pfadd %mm5, %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
197 movq %mm0, 16(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
198 movq %mm4, 24(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
199 pfsub %mm1, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
200 pfsub %mm5, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
201 pfmul 80(%ebx), %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
202 pfmul 88(%ebx), %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
203 movd %mm3, 44(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
204 movd %mm7, 36(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
205 psrlq $32, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
206 psrlq $32, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
207 movd %mm3, 40(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
208 movd %mm7, 32(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
209 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
210 /* Phase 3*/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
211 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
212 movq 64(%edx), %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
213 movq 72(%edx), %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
214 movq %mm0, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
215 movq %mm4, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
216 movq 120(%edx), %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
217 movq 112(%edx), %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
218 /* n.b.: pswapd*/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
219 movq %mm1, %mm2 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
220 movq %mm5, %mm6 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
221 psrlq $32, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
222 psrlq $32, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
223 punpckldq %mm2, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
224 punpckldq %mm6, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
225 /**/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
226 pfadd %mm1, %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
227 pfadd %mm5, %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
228 movq %mm0, 64(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
229 movq %mm4, 72(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
230 pfsubr %mm1, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
231 pfsubr %mm5, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
232 pfmul 64(%ebx), %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
233 pfmul 72(%ebx), %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
234 movd %mm3, 124(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
235 movd %mm7, 116(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
236 psrlq $32, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
237 psrlq $32, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
238 movd %mm3, 120(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
239 movd %mm7, 112(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
240 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
241 movq 80(%edx), %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
242 movq 88(%edx), %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
243 movq %mm0, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
244 movq %mm4, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
245 movq 104(%edx), %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
246 movq 96(%edx), %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
247 /* n.b.: pswapd*/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
248 movq %mm1, %mm2 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
249 movq %mm5, %mm6 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
250 psrlq $32, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
251 psrlq $32, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
252 punpckldq %mm2, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
253 punpckldq %mm6, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
254 /**/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
255 pfadd %mm1, %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
256 pfadd %mm5, %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
257 movq %mm0, 80(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
258 movq %mm4, 88(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
259 pfsubr %mm1, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
260 pfsubr %mm5, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
261 pfmul 80(%ebx), %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
262 pfmul 88(%ebx), %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
263 movd %mm3, 108(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
264 movd %mm7, 100(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
265 psrlq $32, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
266 psrlq $32, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
267 movd %mm3, 104(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
268 movd %mm7, 96(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
269 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
270 /* Phase 4*/ |
1 | 271 |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
272 movq (%ecx), %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
273 movq 8(%ecx), %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
274 movq %mm0, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
275 movq %mm4, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
276 movq 24(%ecx), %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
277 movq 16(%ecx), %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
278 /* n.b.: pswapd*/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
279 movq %mm1, %mm2 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
280 movq %mm5, %mm6 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
281 psrlq $32, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
282 psrlq $32, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
283 punpckldq %mm2, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
284 punpckldq %mm6, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
285 /**/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
286 pfadd %mm1, %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
287 pfadd %mm5, %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
288 movq %mm0, (%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
289 movq %mm4, 8(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
290 pfsub %mm1, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
291 pfsub %mm5, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
292 pfmul 96(%ebx), %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
293 pfmul 104(%ebx), %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
294 movd %mm3, 28(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
295 movd %mm7, 20(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
296 psrlq $32, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
297 psrlq $32, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
298 movd %mm3, 24(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
299 movd %mm7, 16(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
300 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
301 movq 32(%ecx), %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
302 movq 40(%ecx), %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
303 movq %mm0, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
304 movq %mm4, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
305 movq 56(%ecx), %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
306 movq 48(%ecx), %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
307 /* n.b.: pswapd*/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
308 movq %mm1, %mm2 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
309 movq %mm5, %mm6 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
310 psrlq $32, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
311 psrlq $32, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
312 punpckldq %mm2, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
313 punpckldq %mm6, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
314 /**/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
315 pfadd %mm1, %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
316 pfadd %mm5, %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
317 movq %mm0, 32(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
318 movq %mm4, 40(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
319 pfsubr %mm1, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
320 pfsubr %mm5, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
321 pfmul 96(%ebx), %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
322 pfmul 104(%ebx), %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
323 movd %mm3, 60(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
324 movd %mm7, 52(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
325 psrlq $32, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
326 psrlq $32, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
327 movd %mm3, 56(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
328 movd %mm7, 48(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
329 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
330 movq 64(%ecx), %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
331 movq 72(%ecx), %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
332 movq %mm0, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
333 movq %mm4, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
334 movq 88(%ecx), %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
335 movq 80(%ecx), %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
336 /* n.b.: pswapd*/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
337 movq %mm1, %mm2 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
338 movq %mm5, %mm6 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
339 psrlq $32, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
340 psrlq $32, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
341 punpckldq %mm2, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
342 punpckldq %mm6, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
343 /**/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
344 pfadd %mm1, %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
345 pfadd %mm5, %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
346 movq %mm0, 64(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
347 movq %mm4, 72(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
348 pfsub %mm1, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
349 pfsub %mm5, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
350 pfmul 96(%ebx), %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
351 pfmul 104(%ebx), %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
352 movd %mm3, 92(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
353 movd %mm7, 84(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
354 psrlq $32, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
355 psrlq $32, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
356 movd %mm3, 88(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
357 movd %mm7, 80(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
358 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
359 movq 96(%ecx), %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
360 movq 104(%ecx), %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
361 movq %mm0, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
362 movq %mm4, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
363 movq 120(%ecx), %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
364 movq 112(%ecx), %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
365 /* n.b.: pswapd*/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
366 movq %mm1, %mm2 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
367 movq %mm5, %mm6 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
368 psrlq $32, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
369 psrlq $32, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
370 punpckldq %mm2, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
371 punpckldq %mm6, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
372 /**/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
373 pfadd %mm1, %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
374 pfadd %mm5, %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
375 movq %mm0, 96(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
376 movq %mm4, 104(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
377 pfsubr %mm1, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
378 pfsubr %mm5, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
379 pfmul 96(%ebx), %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
380 pfmul 104(%ebx), %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
381 movd %mm3, 124(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
382 movd %mm7, 116(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
383 psrlq $32, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
384 psrlq $32, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
385 movd %mm3, 120(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
386 movd %mm7, 112(%edx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
387 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
388 /* Phase 5 */ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
389 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
390 movq (%edx), %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
391 movq 16(%edx), %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
392 movq %mm0, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
393 movq %mm4, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
394 movq 8(%edx), %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
395 movq 24(%edx), %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
396 /* n.b.: pswapd*/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
397 movq %mm1, %mm2 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
398 movq %mm5, %mm6 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
399 psrlq $32, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
400 psrlq $32, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
401 punpckldq %mm2, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
402 punpckldq %mm6, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
403 /**/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
404 pfadd %mm1, %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
405 pfadd %mm5, %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
406 movq %mm0, (%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
407 movq %mm4, 16(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
408 pfsub %mm1, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
409 pfsubr %mm5, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
410 pfmul 112(%ebx), %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
411 pfmul 112(%ebx), %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
412 movd %mm3, 12(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
413 movd %mm7, 28(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
414 psrlq $32, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
415 psrlq $32, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
416 movd %mm3, 8(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
417 movd %mm7, 24(%ecx) |
1 | 418 |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
419 movq 32(%edx), %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
420 movq 48(%edx), %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
421 movq %mm0, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
422 movq %mm4, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
423 movq 40(%edx), %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
424 movq 56(%edx), %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
425 /* n.b.: pswapd*/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
426 movq %mm1, %mm2 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
427 movq %mm5, %mm6 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
428 psrlq $32, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
429 psrlq $32, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
430 punpckldq %mm2, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
431 punpckldq %mm6, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
432 /**/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
433 pfadd %mm1, %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
434 pfadd %mm5, %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
435 movq %mm0, 32(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
436 movq %mm4, 48(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
437 pfsub %mm1, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
438 pfsubr %mm5, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
439 pfmul 112(%ebx), %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
440 pfmul 112(%ebx), %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
441 movd %mm3, 44(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
442 movd %mm7, 60(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
443 psrlq $32, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
444 psrlq $32, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
445 movd %mm3, 40(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
446 movd %mm7, 56(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
447 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
448 movq 64(%edx), %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
449 movq 80(%edx), %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
450 movq %mm0, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
451 movq %mm4, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
452 movq 72(%edx), %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
453 movq 88(%edx), %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
454 /* n.b.: pswapd*/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
455 movq %mm1, %mm2 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
456 movq %mm5, %mm6 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
457 psrlq $32, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
458 psrlq $32, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
459 punpckldq %mm2, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
460 punpckldq %mm6, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
461 /**/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
462 pfadd %mm1, %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
463 pfadd %mm5, %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
464 movq %mm0, 64(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
465 movq %mm4, 80(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
466 pfsub %mm1, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
467 pfsubr %mm5, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
468 pfmul 112(%ebx), %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
469 pfmul 112(%ebx), %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
470 movd %mm3, 76(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
471 movd %mm7, 92(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
472 psrlq $32, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
473 psrlq $32, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
474 movd %mm3, 72(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
475 movd %mm7, 88(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
476 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
477 movq 96(%edx), %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
478 movq 112(%edx), %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
479 movq %mm0, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
480 movq %mm4, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
481 movq 104(%edx), %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
482 movq 120(%edx), %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
483 /* n.b.: pswapd*/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
484 movq %mm1, %mm2 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
485 movq %mm5, %mm6 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
486 psrlq $32, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
487 psrlq $32, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
488 punpckldq %mm2, %mm1 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
489 punpckldq %mm6, %mm5 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
490 /**/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
491 pfadd %mm1, %mm0 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
492 pfadd %mm5, %mm4 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
493 movq %mm0, 96(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
494 movq %mm4, 112(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
495 pfsub %mm1, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
496 pfsubr %mm5, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
497 pfmul 112(%ebx), %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
498 pfmul 112(%ebx), %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
499 movd %mm3, 108(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
500 movd %mm7, 124(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
501 psrlq $32, %mm3 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
502 psrlq $32, %mm7 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
503 movd %mm3, 104(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
504 movd %mm7, 120(%ecx) |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
505 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
506 /* Phase 6. This is the end of easy road. */ |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
507 /* Code below is coded in scalar mode. Should be optimized */ |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
508 |
1277 | 509 movd plus_1f, %mm6 |
510 punpckldq 120(%ebx), %mm6 /* mm6 = 1.0 | 120(%ebx)*/ | |
511 movq x_plus_minus_3dnow, %mm7 /* mm7 = +1 | -1 */ | |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
512 |
1277 | 513 movq 32(%ecx), %mm0 |
514 movq 64(%ecx), %mm2 | |
515 movq %mm0, %mm1 | |
516 movq %mm2, %mm3 | |
517 pxor %mm7, %mm1 | |
518 pxor %mm7, %mm3 | |
519 pfacc %mm1, %mm0 | |
520 pfacc %mm3, %mm2 | |
521 pfmul %mm6, %mm0 | |
522 pfmul %mm6, %mm2 | |
523 movq %mm0, 32(%edx) | |
524 movq %mm2, 64(%edx) | |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
525 |
1277 | 526 movd 44(%ecx), %mm0 |
527 movd 40(%ecx), %mm2 | |
528 movd 120(%ebx), %mm3 | |
529 punpckldq 76(%ecx), %mm0 | |
530 punpckldq 72(%ecx), %mm2 | |
531 punpckldq %mm3, %mm3 | |
532 movq %mm0, %mm4 | |
533 movq %mm2, %mm5 | |
534 pfsub %mm2, %mm0 | |
535 pfmul %mm3, %mm0 | |
536 movq %mm0, %mm1 | |
537 pfadd %mm5, %mm0 | |
538 pfadd %mm4, %mm0 | |
539 movq %mm0, %mm2 | |
540 punpckldq %mm1, %mm0 | |
541 punpckhdq %mm1, %mm2 | |
542 movq %mm0, 40(%edx) | |
543 movq %mm2, 72(%edx) | |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
544 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
545 movd 48(%ecx), %mm3 |
1282 | 546 movd 60(%ecx), %mm2 |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
547 pfsub 52(%ecx), %mm3 |
1282 | 548 pfsub 56(%ecx), %mm2 |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
549 pfmul 120(%ebx), %mm3 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
550 pfmul 120(%ebx), %mm2 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
551 movq %mm2, %mm1 |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
552 |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
553 pfadd 56(%ecx), %mm1 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
554 pfadd 60(%ecx), %mm1 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
555 movq %mm1, %mm0 |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
556 |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
557 pfadd 48(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
558 pfadd 52(%ecx), %mm0 |
1277 | 559 pfadd %mm3, %mm1 |
560 punpckldq %mm2, %mm1 | |
561 pfadd %mm3, %mm2 | |
562 punpckldq %mm2, %mm0 | |
563 movq %mm1, 56(%edx) | |
564 movq %mm0, 48(%edx) | |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
565 |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
566 /*---*/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
567 |
1277 | 568 movd 92(%ecx), %mm1 |
569 pfsub 88(%ecx), %mm1 | |
570 pfmul 120(%ebx), %mm1 | |
571 movd %mm1, 92(%edx) | |
572 pfadd 92(%ecx), %mm1 | |
573 pfadd 88(%ecx), %mm1 | |
574 movq %mm1, %mm0 | |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
575 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
576 pfadd 80(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
577 pfadd 84(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
578 movd %mm0, 80(%edx) |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
579 |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
580 movd 80(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
581 pfsub 84(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
582 pfmul 120(%ebx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
583 pfadd %mm0, %mm1 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
584 pfadd 92(%edx), %mm0 |
1282 | 585 punpckldq %mm1, %mm0 |
586 movq %mm0, 84(%edx) | |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
587 |
1277 | 588 movq 96(%ecx), %mm0 |
589 movq %mm0, %mm1 | |
590 pxor %mm7, %mm1 | |
591 pfacc %mm1, %mm0 | |
592 pfmul %mm6, %mm0 | |
593 movq %mm0, 96(%edx) | |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
594 |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
595 movd 108(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
596 pfsub 104(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
597 pfmul 120(%ebx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
598 movd %mm0, 108(%edx) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
599 pfadd 104(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
600 pfadd 108(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
601 movd %mm0, 104(%edx) |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
602 |
1277 | 603 movd 124(%ecx), %mm1 |
604 pfsub 120(%ecx), %mm1 | |
605 pfmul 120(%ebx), %mm1 | |
606 movd %mm1, 124(%edx) | |
607 pfadd 120(%ecx), %mm1 | |
608 pfadd 124(%ecx), %mm1 | |
609 movq %mm1, %mm0 | |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
610 |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
611 pfadd 112(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
612 pfadd 116(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
613 movd %mm0, 112(%edx) |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
614 |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
615 movd 112(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
616 pfsub 116(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
617 pfmul 120(%ebx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
618 pfadd %mm0,%mm1 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
619 pfadd 124(%edx), %mm0 |
1282 | 620 punpckldq %mm1, %mm0 |
621 movq %mm0, 116(%edx) | |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
622 |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
623 jnz .L01 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
624 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
625 /* Phase 7*/ |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
626 /* Code below is coded in scalar mode. Should be optimized */ |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
627 |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
628 movd (%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
629 pfadd 4(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
630 movd %mm0, 1024(%esi) |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
631 |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
632 movd (%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
633 pfsub 4(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
634 pfmul 120(%ebx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
635 movd %mm0, (%esi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
636 movd %mm0, (%edi) |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
637 |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
638 movd 12(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
639 pfsub 8(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
640 pfmul 120(%ebx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
641 movd %mm0, 512(%edi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
642 pfadd 12(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
643 pfadd 8(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
644 movd %mm0, 512(%esi) |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
645 |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
646 movd 16(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
647 pfsub 20(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
648 pfmul 120(%ebx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
649 movq %mm0, %mm3 |
1 | 650 |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
651 movd 28(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
652 pfsub 24(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
653 pfmul 120(%ebx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
654 movd %mm0, 768(%edi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
655 movq %mm0, %mm2 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
656 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
657 pfadd 24(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
658 pfadd 28(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
659 movq %mm0, %mm1 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
660 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
661 pfadd 16(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
662 pfadd 20(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
663 movd %mm0, 768(%esi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
664 pfadd %mm3, %mm1 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
665 movd %mm1, 256(%esi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
666 pfadd %mm3, %mm2 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
667 movd %mm2, 256(%edi) |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
668 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
669 /* Phase 8*/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
670 |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
671 movq 32(%edx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
672 movq 48(%edx), %mm1 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
673 pfadd 48(%edx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
674 pfadd 40(%edx), %mm1 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
675 movd %mm0, 896(%esi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
676 movd %mm1, 640(%esi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
677 psrlq $32, %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
678 psrlq $32, %mm1 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
679 movd %mm0, 128(%edi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
680 movd %mm1, 384(%edi) |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
681 |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
682 movd 40(%edx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
683 pfadd 56(%edx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
684 movd %mm0, 384(%esi) |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
685 |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
686 movd 56(%edx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
687 pfadd 36(%edx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
688 movd %mm0, 128(%esi) |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
689 |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
690 movd 60(%edx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
691 movd %mm0, 896(%edi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
692 pfadd 44(%edx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
693 movd %mm0, 640(%edi) |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
694 |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
695 movq 96(%edx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
696 movq 112(%edx), %mm2 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
697 movq 104(%edx), %mm4 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
698 pfadd 112(%edx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
699 pfadd 104(%edx), %mm2 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
700 pfadd 120(%edx), %mm4 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
701 movq %mm0, %mm1 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
702 movq %mm2, %mm3 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
703 movq %mm4, %mm5 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
704 pfadd 64(%edx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
705 pfadd 80(%edx), %mm2 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
706 pfadd 72(%edx), %mm4 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
707 movd %mm0, 960(%esi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
708 movd %mm2, 704(%esi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
709 movd %mm4, 448(%esi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
710 psrlq $32, %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
711 psrlq $32, %mm2 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
712 psrlq $32, %mm4 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
713 movd %mm0, 64(%edi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
714 movd %mm2, 320(%edi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
715 movd %mm4, 576(%edi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
716 pfadd 80(%edx), %mm1 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
717 pfadd 72(%edx), %mm3 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
718 pfadd 88(%edx), %mm5 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
719 movd %mm1, 832(%esi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
720 movd %mm3, 576(%esi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
721 movd %mm5, 320(%esi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
722 psrlq $32, %mm1 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
723 psrlq $32, %mm3 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
724 psrlq $32, %mm5 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
725 movd %mm1, 192(%edi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
726 movd %mm3, 448(%edi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
727 movd %mm5, 704(%edi) |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
728 |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
729 movd 120(%edx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
730 pfadd 100(%edx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
731 movq %mm0, %mm1 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
732 pfadd 88(%edx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
733 movd %mm0, 192(%esi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
734 pfadd 68(%edx), %mm1 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
735 movd %mm1, 64(%esi) |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
736 |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
737 movd 124(%edx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
738 movd %mm0, 960(%edi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
739 pfadd 92(%edx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
740 movd %mm0, 832(%edi) |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
741 |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
742 jmp .L_bye |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
743 .L01: |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
744 /* Phase 9*/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
745 |
1277 | 746 movq (%ecx), %mm0 |
747 movq %mm0, %mm1 | |
748 pxor %mm7, %mm1 | |
749 pfacc %mm1, %mm0 | |
750 pfmul %mm6, %mm0 | |
751 pf2id %mm0, %mm0 | |
752 movd %mm0, %eax | |
753 movw %ax, 512(%esi) | |
754 psrlq $32, %mm0 | |
755 movd %mm0, %eax | |
756 movw %ax, (%esi) | |
1 | 757 |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
758 movd 12(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
759 pfsub 8(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
760 pfmul 120(%ebx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
761 pf2id %mm0, %mm7 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
762 movd %mm7, %eax |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
763 movw %ax, 256(%edi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
764 pfadd 12(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
765 pfadd 8(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
766 pf2id %mm0, %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
767 movd %mm0, %eax |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
768 movw %ax, 256(%esi) |
1 | 769 |
1277 | 770 movd 16(%ecx), %mm3 |
771 pfsub 20(%ecx), %mm3 | |
772 pfmul 120(%ebx), %mm3 | |
773 movq %mm3, %mm2 | |
1 | 774 |
1277 | 775 movd 28(%ecx), %mm2 |
776 pfsub 24(%ecx), %mm2 | |
777 pfmul 120(%ebx), %mm2 | |
778 movq %mm2, %mm1 | |
779 | |
780 pf2id %mm2, %mm7 | |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
781 movd %mm7, %eax |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
782 movw %ax, 384(%edi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
783 |
1277 | 784 pfadd 24(%ecx), %mm1 |
785 pfadd 28(%ecx), %mm1 | |
786 movq %mm1, %mm0 | |
787 | |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
788 pfadd 16(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
789 pfadd 20(%ecx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
790 pf2id %mm0, %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
791 movd %mm0, %eax |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
792 movw %ax, 384(%esi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
793 pfadd %mm3, %mm1 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
794 pf2id %mm1, %mm1 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
795 movd %mm1, %eax |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
796 movw %ax, 128(%esi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
797 pfadd %mm3, %mm2 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
798 pf2id %mm2, %mm2 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
799 movd %mm2, %eax |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
800 movw %ax, 128(%edi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
801 |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
802 /* Phase 10*/ |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
803 |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
804 movq 32(%edx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
805 movq 48(%edx), %mm1 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
806 pfadd 48(%edx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
807 pfadd 40(%edx), %mm1 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
808 pf2id %mm0, %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
809 pf2id %mm1, %mm1 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
810 movd %mm0, %eax |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
811 movd %mm1, %ecx |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
812 movw %ax, 448(%esi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
813 movw %cx, 320(%esi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
814 psrlq $32, %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
815 psrlq $32, %mm1 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
816 movd %mm0, %eax |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
817 movd %mm1, %ecx |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
818 movw %ax, 64(%edi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
819 movw %cx, 192(%edi) |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
820 |
1277 | 821 movd 40(%edx), %mm3 |
822 movd 56(%edx), %mm4 | |
823 movd 60(%edx), %mm0 | |
824 movd 44(%edx), %mm2 | |
825 movd 120(%edx), %mm5 | |
826 punpckldq %mm4, %mm3 | |
827 punpckldq 124(%edx), %mm0 | |
828 pfadd 100(%edx), %mm5 | |
829 punpckldq 36(%edx), %mm4 | |
830 punpckldq 92(%edx), %mm2 | |
831 movq %mm5, %mm6 | |
832 pfadd %mm4, %mm3 | |
833 pf2id %mm0, %mm1 | |
834 pf2id %mm3, %mm3 | |
835 pfadd 88(%edx), %mm5 | |
836 movd %mm1, %eax | |
837 movd %mm3, %ecx | |
838 movw %ax, 448(%edi) | |
839 movw %cx, 192(%esi) | |
840 pf2id %mm5, %mm5 | |
841 psrlq $32, %mm1 | |
842 psrlq $32, %mm3 | |
843 movd %mm5, %ebx | |
844 movd %mm1, %eax | |
845 movd %mm3, %ecx | |
846 movw %bx, 96(%esi) | |
847 movw %ax, 480(%edi) | |
848 movw %cx, 64(%esi) | |
849 pfadd %mm2, %mm0 | |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
850 pf2id %mm0, %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
851 movd %mm0, %eax |
1277 | 852 pfadd 68(%edx), %mm6 |
853 movw %ax, 320(%edi) | |
854 psrlq $32, %mm0 | |
855 pf2id %mm6, %mm6 | |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
856 movd %mm0, %eax |
1277 | 857 movd %mm6, %ebx |
858 movw %ax, 416(%edi) | |
859 movw %bx, 32(%esi) | |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
860 |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
861 movq 96(%edx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
862 movq 112(%edx), %mm2 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
863 movq 104(%edx), %mm4 |
1277 | 864 pfadd %mm2, %mm0 |
865 pfadd %mm4, %mm2 | |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
866 pfadd 120(%edx), %mm4 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
867 movq %mm0, %mm1 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
868 movq %mm2, %mm3 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
869 movq %mm4, %mm5 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
870 pfadd 64(%edx), %mm0 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
871 pfadd 80(%edx), %mm2 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
872 pfadd 72(%edx), %mm4 |
1277 | 873 pf2id %mm0, %mm0 |
874 pf2id %mm2, %mm2 | |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
875 pf2id %mm4, %mm4 |
1277 | 876 movd %mm0, %eax |
877 movd %mm2, %ecx | |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
878 movd %mm4, %ebx |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
879 movw %ax, 480(%esi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
880 movw %cx, 352(%esi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
881 movw %bx, 224(%esi) |
1277 | 882 psrlq $32, %mm0 |
883 psrlq $32, %mm2 | |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
884 psrlq $32, %mm4 |
1277 | 885 movd %mm0, %eax |
886 movd %mm2, %ecx | |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
887 movd %mm4, %ebx |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
888 movw %ax, 32(%edi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
889 movw %cx, 160(%edi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
890 movw %bx, 288(%edi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
891 pfadd 80(%edx), %mm1 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
892 pfadd 72(%edx), %mm3 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
893 pfadd 88(%edx), %mm5 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
894 pf2id %mm1, %mm1 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
895 pf2id %mm3, %mm3 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
896 pf2id %mm5, %mm5 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
897 movd %mm1, %eax |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
898 movd %mm3, %ecx |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
899 movd %mm5, %ebx |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
900 movw %ax, 416(%esi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
901 movw %cx, 288(%esi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
902 movw %bx, 160(%esi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
903 psrlq $32, %mm1 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
904 psrlq $32, %mm3 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
905 psrlq $32, %mm5 |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
906 movd %mm1, %eax |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
907 movd %mm3, %ecx |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
908 movd %mm5, %ebx |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
909 movw %ax, 96(%edi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
910 movw %cx, 224(%edi) |
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
911 movw %bx, 352(%edi) |
1 | 912 |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
913 movsw |
1 | 914 |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
915 .L_bye: |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
916 addl $256,%esp |
1271
2864e32cd267
Finished 3dnow optimization (in scalar mode) and minor improvements
nick
parents:
1245
diff
changeset
|
917 femms |
1245
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
918 popl %edi |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
919 popl %esi |
03b7e2955a20
Added newest MMX-optimized decore which speedups decoding at least on 13% for any cpu.
nick
parents:
1
diff
changeset
|
920 popl %ebx |
1282 | 921 ret $12 |