media/libtremor/lib/asm_arm.h@b8a032363ba2
media/libtremor/lib/asm_arm.h
Thu, 22 Jan 2015 13:21:57 +0100
- author
- Michael Schloh von Bennewitz <michael@schloh.com>
- date
- Thu, 22 Jan 2015 13:21:57 +0100
- branch
- TOR_BUG_9701
- changeset 15
- b8a032363ba2
- permissions
- -rw-r--r--
Incorporate requested changes from Mozilla in review:
https://bugzilla.mozilla.org/show_bug.cgi?id=1123480#c6
1 /********************************************************************
2 * *
3 * THIS FILE IS PART OF THE OggVorbis 'TREMOR' CODEC SOURCE CODE. *
4 * *
5 * USE, DISTRIBUTION AND REPRODUCTION OF THIS LIBRARY SOURCE IS *
6 * GOVERNED BY A BSD-STYLE SOURCE LICENSE INCLUDED WITH THIS SOURCE *
7 * IN 'COPYING'. PLEASE READ THESE TERMS BEFORE DISTRIBUTING. *
8 * *
9 * THE OggVorbis 'TREMOR' SOURCE CODE IS (C) COPYRIGHT 1994-2002 *
10 * BY THE Xiph.Org FOUNDATION http://www.xiph.org/ *
11 * *
12 ********************************************************************
14 function: arm7 and later wide math functions
16 ********************************************************************/
18 #ifdef _ARM_ASSEM_
20 #if !defined(_V_WIDE_MATH) && !defined(_LOW_ACCURACY_)
21 #define _V_WIDE_MATH
23 static inline ogg_int32_t MULT32(ogg_int32_t x, ogg_int32_t y) {
24 int lo,hi;
25 asm volatile("smull\t%0, %1, %2, %3"
26 : "=&r"(lo),"=&r"(hi)
27 : "%r"(x),"r"(y)
28 : "cc");
29 return(hi);
30 }
32 static inline ogg_int32_t MULT31(ogg_int32_t x, ogg_int32_t y) {
33 return MULT32(x,y)<<1;
34 }
36 static inline ogg_int32_t MULT31_SHIFT15(ogg_int32_t x, ogg_int32_t y) {
37 int lo,hi;
38 asm volatile("smull %0, %1, %2, %3\n\t"
39 "movs %0, %0, lsr #15\n\t"
40 "adc %1, %0, %1, lsl #17\n\t"
41 : "=&r"(lo),"=&r"(hi)
42 : "%r"(x),"r"(y)
43 : "cc");
44 return(hi);
45 }
47 #define MB() asm volatile ("" : : : "memory")
49 static inline void XPROD32(ogg_int32_t a, ogg_int32_t b,
50 ogg_int32_t t, ogg_int32_t v,
51 ogg_int32_t *x, ogg_int32_t *y)
52 {
53 int x1, y1, l;
54 asm( "smull %0, %1, %4, %6\n\t"
55 "smlal %0, %1, %5, %7\n\t"
56 "rsb %3, %4, #0\n\t"
57 "smull %0, %2, %5, %6\n\t"
58 "smlal %0, %2, %3, %7"
59 : "=&r" (l), "=&r" (x1), "=&r" (y1), "=r" (a)
60 : "3" (a), "r" (b), "r" (t), "r" (v)
61 : "cc" );
62 *x = x1;
63 MB();
64 *y = y1;
65 }
67 static inline void XPROD31(ogg_int32_t a, ogg_int32_t b,
68 ogg_int32_t t, ogg_int32_t v,
69 ogg_int32_t *x, ogg_int32_t *y)
70 {
71 int x1, y1, l;
72 asm( "smull %0, %1, %4, %6\n\t"
73 "smlal %0, %1, %5, %7\n\t"
74 "rsb %3, %4, #0\n\t"
75 "smull %0, %2, %5, %6\n\t"
76 "smlal %0, %2, %3, %7"
77 : "=&r" (l), "=&r" (x1), "=&r" (y1), "=r" (a)
78 : "3" (a), "r" (b), "r" (t), "r" (v)
79 : "cc" );
80 *x = x1 << 1;
81 MB();
82 *y = y1 << 1;
83 }
85 static inline void XNPROD31(ogg_int32_t a, ogg_int32_t b,
86 ogg_int32_t t, ogg_int32_t v,
87 ogg_int32_t *x, ogg_int32_t *y)
88 {
89 int x1, y1, l;
90 asm( "rsb %2, %4, #0\n\t"
91 "smull %0, %1, %3, %5\n\t"
92 "smlal %0, %1, %2, %6\n\t"
93 "smull %0, %2, %4, %5\n\t"
94 "smlal %0, %2, %3, %6"
95 : "=&r" (l), "=&r" (x1), "=&r" (y1)
96 : "r" (a), "r" (b), "r" (t), "r" (v)
97 : "cc" );
98 *x = x1 << 1;
99 MB();
100 *y = y1 << 1;
101 }
103 #endif
105 #ifndef _V_CLIP_MATH
106 #define _V_CLIP_MATH
108 static inline ogg_int32_t CLIP_TO_15(ogg_int32_t x) {
109 int tmp;
110 asm volatile("subs %1, %0, #32768\n\t"
111 "movpl %0, #0x7f00\n\t"
112 "orrpl %0, %0, #0xff\n"
113 "adds %1, %0, #32768\n\t"
114 "movmi %0, #0x8000"
115 : "+r"(x),"=r"(tmp)
116 :
117 : "cc");
118 return(x);
119 }
121 #endif
123 #ifndef _V_LSP_MATH_ASM
124 #define _V_LSP_MATH_ASM
126 static inline void lsp_loop_asm(ogg_uint32_t *qip,ogg_uint32_t *pip,
127 ogg_int32_t *qexpp,
128 ogg_int32_t *ilsp,ogg_int32_t wi,
129 ogg_int32_t m){
131 ogg_uint32_t qi=*qip,pi=*pip;
132 ogg_int32_t qexp=*qexpp;
134 asm("mov r0,%3;"
135 "movs r1,%5,asr#1;"
136 "add r0,r0,r1,lsl#3;"
137 "beq 2f;\n"
138 "1:"
140 "ldmdb r0!,{r1,r3};"
141 "subs r1,r1,%4;" //ilsp[j]-wi
142 "rsbmi r1,r1,#0;" //labs(ilsp[j]-wi)
143 "umull %0,r2,r1,%0;" //qi*=labs(ilsp[j]-wi)
145 "subs r1,r3,%4;" //ilsp[j+1]-wi
146 "rsbmi r1,r1,#0;" //labs(ilsp[j+1]-wi)
147 "umull %1,r3,r1,%1;" //pi*=labs(ilsp[j+1]-wi)
149 "cmn r2,r3;" // shift down 16?
150 "beq 0f;"
151 "add %2,%2,#16;"
152 "mov %0,%0,lsr #16;"
153 "orr %0,%0,r2,lsl #16;"
154 "mov %1,%1,lsr #16;"
155 "orr %1,%1,r3,lsl #16;"
156 "0:"
157 "cmp r0,%3;\n"
158 "bhi 1b;\n"
160 "2:"
161 // odd filter assymetry
162 "ands r0,%5,#1;\n"
163 "beq 3f;\n"
164 "add r0,%3,%5,lsl#2;\n"
166 "ldr r1,[r0,#-4];\n"
167 "mov r0,#0x4000;\n"
169 "subs r1,r1,%4;\n" //ilsp[j]-wi
170 "rsbmi r1,r1,#0;\n" //labs(ilsp[j]-wi)
171 "umull %0,r2,r1,%0;\n" //qi*=labs(ilsp[j]-wi)
172 "umull %1,r3,r0,%1;\n" //pi*=labs(ilsp[j+1]-wi)
174 "cmn r2,r3;\n" // shift down 16?
175 "beq 3f;\n"
176 "add %2,%2,#16;\n"
177 "mov %0,%0,lsr #16;\n"
178 "orr %0,%0,r2,lsl #16;\n"
179 "mov %1,%1,lsr #16;\n"
180 "orr %1,%1,r3,lsl #16;\n"
182 //qi=(pi>>shift)*labs(ilsp[j]-wi);
183 //pi=(qi>>shift)*labs(ilsp[j+1]-wi);
184 //qexp+=shift;
186 //}
188 /* normalize to max 16 sig figs */
189 "3:"
190 "mov r2,#0;"
191 "orr r1,%0,%1;"
192 "tst r1,#0xff000000;"
193 "addne r2,r2,#8;"
194 "movne r1,r1,lsr #8;"
195 "tst r1,#0x00f00000;"
196 "addne r2,r2,#4;"
197 "movne r1,r1,lsr #4;"
198 "tst r1,#0x000c0000;"
199 "addne r2,r2,#2;"
200 "movne r1,r1,lsr #2;"
201 "tst r1,#0x00020000;"
202 "addne r2,r2,#1;"
203 "movne r1,r1,lsr #1;"
204 "tst r1,#0x00010000;"
205 "addne r2,r2,#1;"
206 "mov %0,%0,lsr r2;"
207 "mov %1,%1,lsr r2;"
208 "add %2,%2,r2;"
210 : "+r"(qi),"+r"(pi),"+r"(qexp)
211 : "r"(ilsp),"r"(wi),"r"(m)
212 : "r0","r1","r2","r3","cc");
214 *qip=qi;
215 *pip=pi;
216 *qexpp=qexp;
217 }
219 static inline void lsp_norm_asm(ogg_uint32_t *qip,ogg_int32_t *qexpp){
221 ogg_uint32_t qi=*qip;
222 ogg_int32_t qexp=*qexpp;
224 asm("tst %0,#0x0000ff00;"
225 "moveq %0,%0,lsl #8;"
226 "subeq %1,%1,#8;"
227 "tst %0,#0x0000f000;"
228 "moveq %0,%0,lsl #4;"
229 "subeq %1,%1,#4;"
230 "tst %0,#0x0000c000;"
231 "moveq %0,%0,lsl #2;"
232 "subeq %1,%1,#2;"
233 "tst %0,#0x00008000;"
234 "moveq %0,%0,lsl #1;"
235 "subeq %1,%1,#1;"
236 : "+r"(qi),"+r"(qexp)
237 :
238 : "cc");
239 *qip=qi;
240 *qexpp=qexp;
241 }
243 #endif
244 #endif
