SDL_mixer_MMX.c
1
2
3
4
5
6
7
8
9
10
11
12
13
/*
MMX assembler version of SDL_MixAudio for signed little endian 16 bit samples and signed 8 bit samples
Copyright 2002 Stephane Marchesin (stephane.marchesin@wanadoo.fr)
This code is licensed under the LGPL (see COPYING for details)
Assumes buffer size in bytes is a multiple of 16
Assumes SDL_MIX_MAXVOLUME = 128
*/
/***********************************************
* Mixing for 16 bit signed buffers
***********************************************/
14
15
16
17
18
19
20
21
#if defined(i386) && defined(__GNUC__) && defined(USE_ASMBLIT)
void SDL_MixAudio_MMX_S16(char* dst,char* src,unsigned int size,int volume)
{
__asm__ __volatile__ (
" movl %3,%%eax\n" // eax = volume
22
" movl %2,%%edx\n" // edx = size
24
" shrl $4,%%edx\n" // process 16 bytes per iteration = 8 samples
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
" jz .endS16\n"
" pxor %%mm0,%%mm0\n"
" movd %%eax,%%mm0\n"
" movq %%mm0,%%mm1\n"
" psllq $16,%%mm0\n"
" por %%mm1,%%mm0\n"
" psllq $16,%%mm0\n"
" por %%mm1,%%mm0\n"
" psllq $16,%%mm0\n"
" por %%mm1,%%mm0\n" // mm0 = vol|vol|vol|vol
".align 16\n"
" .mixloopS16:\n"
42
" movq (%1),%%mm1\n" // mm1 = a|b|c|d
43
44
45
" movq %%mm1,%%mm2\n" // mm2 = a|b|c|d
46
" movq 8(%1),%%mm4\n" // mm4 = e|f|g|h
47
48
// pré charger le buffer dst dans mm7
49
" movq (%0),%%mm7\n" // mm7 = dst[0]"
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
// multiplier par le volume
" pmullw %%mm0,%%mm1\n" // mm1 = l(a*v)|l(b*v)|l(c*v)|l(d*v)
" pmulhw %%mm0,%%mm2\n" // mm2 = h(a*v)|h(b*v)|h(c*v)|h(d*v)
" movq %%mm4,%%mm5\n" // mm5 = e|f|g|h
" pmullw %%mm0,%%mm4\n" // mm4 = l(e*v)|l(f*v)|l(g*v)|l(h*v)
" pmulhw %%mm0,%%mm5\n" // mm5 = h(e*v)|h(f*v)|h(g*v)|h(h*v)
" movq %%mm1,%%mm3\n" // mm3 = l(a*v)|l(b*v)|l(c*v)|l(d*v)
" punpckhwd %%mm2,%%mm1\n" // mm1 = a*v|b*v
" movq %%mm4,%%mm6\n" // mm6 = l(e*v)|l(f*v)|l(g*v)|l(h*v)
" punpcklwd %%mm2,%%mm3\n" // mm3 = c*v|d*v
" punpckhwd %%mm5,%%mm4\n" // mm4 = e*f|f*v
" punpcklwd %%mm5,%%mm6\n" // mm6 = g*v|h*v
// pré charger le buffer dst dans mm5
72
" movq 8(%0),%%mm5\n" // mm5 = dst[1]
73
74
75
// diviser par 128
" psrad $7,%%mm1\n" // mm1 = a*v/128|b*v/128 , 128 = SDL_MIX_MAXVOLUME
76
" add $16,%1\n"
77
78
79
80
81
82
83
84
85
86
87
88
89
" psrad $7,%%mm3\n" // mm3 = c*v/128|d*v/128
" psrad $7,%%mm4\n" // mm4 = e*v/128|f*v/128
// mm1 = le sample avec le volume modifié
" packssdw %%mm1,%%mm3\n" // mm3 = s(a*v|b*v|c*v|d*v)
" psrad $7,%%mm6\n" // mm6= g*v/128|h*v/128
" paddsw %%mm7,%%mm3\n" // mm3 = adjust_volume(src)+dst
// mm4 = le sample avec le volume modifié
" packssdw %%mm4,%%mm6\n" // mm6 = s(e*v|f*v|g*v|h*v)
90
" movq %%mm3,(%0)\n"
91
92
93
" paddsw %%mm5,%%mm6\n" // mm6 = adjust_volume(src)+dst
94
" movq %%mm6,8(%0)\n"
96
" add $16,%0\n"
98
" dec %%edx\n"
99
100
101
102
103
104
105
" jnz .mixloopS16\n"
" emms\n"
".endS16:\n"
:
106
: "r" (dst), "r"(src),"m"(size),
107
"m"(volume)
108
: "eax","edx","memory"
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
);
}
////////////////////////////////////////////////
// Mixing for 8 bit signed buffers
////////////////////////////////////////////////
void SDL_MixAudio_MMX_S8(char* dst,char* src,unsigned int size,int volume)
{
__asm__ __volatile__ (
" movl %3,%%eax\n" // eax = volume
124
" movd %%edx,%%mm0\n"
125
126
127
128
129
130
131
132
" movq %%mm0,%%mm1\n"
" psllq $16,%%mm0\n"
" por %%mm1,%%mm0\n"
" psllq $16,%%mm0\n"
" por %%mm1,%%mm0\n"
" psllq $16,%%mm0\n"
" por %%mm1,%%mm0\n"
133
134
" movl %2,%%edx\n" // edx = size
" shr $3,%%edx\n" // process 8 bytes per iteration = 8 samples
136
" cmp $0,%%edx\n"
137
138
139
140
141
142
" je .endS8\n"
".align 16\n"
" .mixloopS8:\n"
" pxor %%mm2,%%mm2\n" // mm2 = 0
143
" movq (%1),%%mm1\n" // mm1 = a|b|c|d|e|f|g|h
144
145
146
147
148
149
150
151
152
" movq %%mm1,%%mm3\n" // mm3 = a|b|c|d|e|f|g|h
// on va faire le "sign extension" en faisant un cmp avec 0 qui retourne 1 si <0, 0 si >0
" pcmpgtb %%mm1,%%mm2\n" // mm2 = 11111111|00000000|00000000....
" punpckhbw %%mm2,%%mm1\n" // mm1 = 0|a|0|b|0|c|0|d
" punpcklbw %%mm2,%%mm3\n" // mm3 = 0|e|0|f|0|g|0|h
153
" movq (%0),%%mm2\n" // mm2 = destination
154
155
" pmullw %%mm0,%%mm1\n" // mm1 = v*a|v*b|v*c|v*d
156
" add $8,%1\n"
157
158
159
160
161
162
163
164
165
166
" pmullw %%mm0,%%mm3\n" // mm3 = v*e|v*f|v*g|v*h
" psraw $7,%%mm1\n" // mm1 = v*a/128|v*b/128|v*c/128|v*d/128
" psraw $7,%%mm3\n" // mm3 = v*e/128|v*f/128|v*g/128|v*h/128
" packsswb %%mm1,%%mm3\n" // mm1 = v*a/128|v*b/128|v*c/128|v*d/128|v*e/128|v*f/128|v*g/128|v*h/128
" paddsb %%mm2,%%mm3\n" // add to destination buffer
167
168
" movq %%mm3,(%0)\n" // store back to ram
" add $8,%0\n"
170
" dec %%edx\n"
171
172
173
174
175
176
" jnz .mixloopS8\n"
".endS8:\n"
" emms\n"
:
177
: "r" (dst), "r"(src),"m"(size),
178
"m"(volume)
179
: "eax","edx","memory"
180
181
182
);
}
#endif