Mercurial > libavcodec.hg
annotate ppc/float_altivec.c @ 8177:484c83d8e9a8 libavcodec
SH4: add macros for setting and restoring FPSCR.PR bit
author | mru |
---|---|
date | Thu, 20 Nov 2008 09:21:50 +0000 |
parents | 8313d5901c5c |
children | d33b47d1f4c1 |
rev | line source |
---|---|
3581 | 1 /* |
2 * Copyright (c) 2006 Luca Barbato <lu_zero@gentoo.org> | |
3 * | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3659
diff
changeset
|
4 * This file is part of FFmpeg. |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3659
diff
changeset
|
5 * |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3659
diff
changeset
|
6 * FFmpeg is free software; you can redistribute it and/or |
3581 | 7 * modify it under the terms of the GNU Lesser General Public |
8 * License as published by the Free Software Foundation; either | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3659
diff
changeset
|
9 * version 2.1 of the License, or (at your option) any later version. |
3581 | 10 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3659
diff
changeset
|
11 * FFmpeg is distributed in the hope that it will be useful, |
3581 | 12 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
14 * Lesser General Public License for more details. | |
15 * | |
16 * You should have received a copy of the GNU Lesser General Public | |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3659
diff
changeset
|
17 * License along with FFmpeg; if not, write to the Free Software |
3581 | 18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
19 */ | |
20 | |
6763 | 21 #include "libavcodec/dsputil.h" |
3581 | 22 |
23 #include "gcc_fixes.h" | |
24 | |
25 #include "dsputil_altivec.h" | |
26 | |
27 static void vector_fmul_altivec(float *dst, const float *src, int len) | |
28 { | |
29 int i; | |
30 vector float d0, d1, s, zero = (vector float)vec_splat_u32(0); | |
31 for(i=0; i<len-7; i+=8) { | |
32 d0 = vec_ld(0, dst+i); | |
33 s = vec_ld(0, src+i); | |
34 d1 = vec_ld(16, dst+i); | |
35 d0 = vec_madd(d0, s, zero); | |
36 d1 = vec_madd(d1, vec_ld(16,src+i), zero); | |
37 vec_st(d0, 0, dst+i); | |
38 vec_st(d1, 16, dst+i); | |
39 } | |
40 } | |
41 | |
42 static void vector_fmul_reverse_altivec(float *dst, const float *src0, | |
43 const float *src1, int len) | |
44 { | |
45 int i; | |
46 vector float d, s0, s1, h0, l0, | |
47 s2, s3, zero = (vector float)vec_splat_u32(0); | |
48 src1 += len-4; | |
49 for(i=0; i<len-7; i+=8) { | |
50 s1 = vec_ld(0, src1-i); // [a,b,c,d] | |
51 s0 = vec_ld(0, src0+i); | |
52 l0 = vec_mergel(s1, s1); // [c,c,d,d] | |
53 s3 = vec_ld(-16, src1-i); | |
54 h0 = vec_mergeh(s1, s1); // [a,a,b,b] | |
55 s2 = vec_ld(16, src0+i); | |
56 s1 = vec_mergeh(vec_mergel(l0,h0), // [d,b,d,b] | |
57 vec_mergeh(l0,h0)); // [c,a,c,a] | |
58 // [d,c,b,a] | |
59 l0 = vec_mergel(s3, s3); | |
60 d = vec_madd(s0, s1, zero); | |
61 h0 = vec_mergeh(s3, s3); | |
62 vec_st(d, 0, dst+i); | |
63 s3 = vec_mergeh(vec_mergel(l0,h0), | |
64 vec_mergeh(l0,h0)); | |
65 d = vec_madd(s2, s3, zero); | |
66 vec_st(d, 16, dst+i); | |
67 } | |
68 } | |
69 | |
70 static void vector_fmul_add_add_altivec(float *dst, const float *src0, | |
71 const float *src1, const float *src2, | |
72 int src3, int len, int step) | |
73 { | |
74 int i; | |
75 vector float d, s0, s1, s2, t0, t1, edges; | |
76 vector unsigned char align = vec_lvsr(0,dst), | |
77 mask = vec_lvsl(0, dst); | |
78 | |
79 #if 0 //FIXME: there is still something wrong | |
80 if (step == 2) { | |
81 int y; | |
82 vector float d0, d1, s3, t2; | |
83 vector unsigned int sel = | |
84 vec_mergeh(vec_splat_u32(-1), vec_splat_u32(0)); | |
85 t1 = vec_ld(16, dst); | |
86 for (i=0,y=0; i<len-3; i+=4,y+=8) { | |
87 | |
88 s0 = vec_ld(0,src0+i); | |
89 s1 = vec_ld(0,src1+i); | |
90 s2 = vec_ld(0,src2+i); | |
91 | |
92 // t0 = vec_ld(0, dst+y); //[x x x|a] | |
93 // t1 = vec_ld(16, dst+y); //[b c d|e] | |
94 t2 = vec_ld(31, dst+y); //[f g h|x] | |
95 | |
96 d = vec_madd(s0,s1,s2); // [A B C D] | |
97 | |
98 // [A A B B] | |
99 | |
100 // [C C D D] | |
101 | |
102 d0 = vec_perm(t0, t1, mask); // [a b c d] | |
103 | |
104 d0 = vec_sel(vec_mergeh(d, d), d0, sel); // [A b B d] | |
105 | |
106 edges = vec_perm(t1, t0, mask); | |
107 | |
108 t0 = vec_perm(edges, d0, align); // [x x x|A] | |
109 | |
110 t1 = vec_perm(d0, edges, align); // [b B d|e] | |
111 | |
112 vec_stl(t0, 0, dst+y); | |
113 | |
114 d1 = vec_perm(t1, t2, mask); // [e f g h] | |
115 | |
116 d1 = vec_sel(vec_mergel(d, d), d1, sel); // [C f D h] | |
117 | |
118 edges = vec_perm(t2, t1, mask); | |
119 | |
120 t1 = vec_perm(edges, d1, align); // [b B d|C] | |
121 | |
122 t2 = vec_perm(d1, edges, align); // [f D h|x] | |
123 | |
124 vec_stl(t1, 16, dst+y); | |
125 | |
126 t0 = t1; | |
127 | |
128 vec_stl(t2, 31, dst+y); | |
129 | |
130 t1 = t2; | |
131 } | |
132 } else | |
133 #endif | |
134 if (step == 1 && src3 == 0) | |
135 for (i=0; i<len-3; i+=4) { | |
4387 | 136 t0 = vec_ld(0, dst+i); |
3581 | 137 t1 = vec_ld(15, dst+i); |
138 s0 = vec_ld(0, src0+i); | |
139 s1 = vec_ld(0, src1+i); | |
140 s2 = vec_ld(0, src2+i); | |
141 edges = vec_perm(t1 ,t0, mask); | |
142 d = vec_madd(s0,s1,s2); | |
3659
dd55fb216497
Proper fix for the corner case that would have been corrected before, praise&blame to me and exg in equal shares
lu_zero
parents:
3657
diff
changeset
|
143 t1 = vec_perm(d, edges, align); |
dd55fb216497
Proper fix for the corner case that would have been corrected before, praise&blame to me and exg in equal shares
lu_zero
parents:
3657
diff
changeset
|
144 t0 = vec_perm(edges, d, align); |
dd55fb216497
Proper fix for the corner case that would have been corrected before, praise&blame to me and exg in equal shares
lu_zero
parents:
3657
diff
changeset
|
145 vec_st(t1, 15, dst+i); |
3581 | 146 vec_st(t0, 0, dst+i); |
147 } | |
148 else | |
149 ff_vector_fmul_add_add_c(dst, src0, src1, src2, src3, len, step); | |
150 } | |
151 | |
7674 | 152 |
153 static vector signed short | |
154 float_to_int16_one_altivec(const float *src) | |
155 { | |
156 vector float s0 = vec_ld(0, src); | |
157 vector float s1 = vec_ld(16, src); | |
158 vector signed int t0 = vec_cts(s0, 0); | |
159 vector signed int t1 = vec_cts(s1, 0); | |
160 return vec_packs(t0,t1); | |
161 } | |
162 | |
7691
8313d5901c5c
Missing static in float_to_int16_altivec declaration
lu_zero
parents:
7676
diff
changeset
|
163 static void float_to_int16_altivec(int16_t *dst, const float *src, int len) |
3581 | 164 { |
165 int i; | |
166 vector signed short d0, d1, d; | |
167 vector unsigned char align; | |
168 if(((long)dst)&15) //FIXME | |
169 for(i=0; i<len-7; i+=8) { | |
170 d0 = vec_ld(0, dst+i); | |
7674 | 171 d = float_to_int16_one_altivec(src+i); |
3581 | 172 d1 = vec_ld(15, dst+i); |
173 d1 = vec_perm(d1, d0, vec_lvsl(0,dst+i)); | |
174 align = vec_lvsr(0, dst+i); | |
3657
ff6720290478
Fix float_to_int16, unaligned case, broken by the previous commit
lu_zero
parents:
3583
diff
changeset
|
175 d0 = vec_perm(d1, d, align); |
ff6720290478
Fix float_to_int16, unaligned case, broken by the previous commit
lu_zero
parents:
3583
diff
changeset
|
176 d1 = vec_perm(d, d1, align); |
3581 | 177 vec_st(d0, 0, dst+i); |
178 vec_st(d1,15, dst+i); | |
179 } | |
180 else | |
181 for(i=0; i<len-7; i+=8) { | |
7674 | 182 d = float_to_int16_one_altivec(src+i); |
3581 | 183 vec_st(d, 0, dst+i); |
184 } | |
185 } | |
186 | |
7675
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
187 static void |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
188 float_to_int16_interleave_altivec(int16_t *dst, const float **src, |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
189 long len, int channels) |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
190 { |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
191 int i; |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
192 vector signed short d0, d1, d2, c0, c1, t0, t1; |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
193 vector unsigned char align; |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
194 if(channels == 1) |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
195 float_to_int16_altivec(dst, src[0], len); |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
196 else |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
197 if (channels == 2) { |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
198 if(((long)dst)&15) |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
199 for(i=0; i<len-7; i+=8) { |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
200 d0 = vec_ld(0, dst + i); |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
201 t0 = float_to_int16_one_altivec(src[0] + i); |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
202 d1 = vec_ld(31, dst + i); |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
203 t1 = float_to_int16_one_altivec(src[1] + i); |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
204 c0 = vec_mergeh(t0, t1); |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
205 c1 = vec_mergel(t0, t1); |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
206 d2 = vec_perm(d1, d0, vec_lvsl(0, dst + i)); |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
207 align = vec_lvsr(0, dst + i); |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
208 d0 = vec_perm(d2, c0, align); |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
209 d1 = vec_perm(c0, c1, align); |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
210 vec_st(d0, 0, dst + i); |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
211 d0 = vec_perm(c1, d2, align); |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
212 vec_st(d1, 15, dst + i); |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
213 vec_st(d0, 31, dst + i); |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
214 dst+=8; |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
215 } |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
216 else |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
217 for(i=0; i<len-7; i+=8) { |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
218 t0 = float_to_int16_one_altivec(src[0] + i); |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
219 t1 = float_to_int16_one_altivec(src[1] + i); |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
220 d0 = vec_mergeh(t0, t1); |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
221 d1 = vec_mergel(t0, t1); |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
222 vec_st(d0, 0, dst + i); |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
223 vec_st(d1, 16, dst + i); |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
224 dst+=8; |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
225 } |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
226 } else { |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
227 DECLARE_ALIGNED(16, int16_t, tmp[len]); |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
228 int c, j; |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
229 for (c = 0; c < channels; c++) { |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
230 float_to_int16_altivec(tmp, src[c], len); |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
231 for (i = 0, j = c; i < len; i++, j+=channels) { |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
232 dst[j] = tmp[i]; |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
233 } |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
234 } |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
235 } |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
236 } |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
237 |
3581 | 238 void float_init_altivec(DSPContext* c, AVCodecContext *avctx) |
239 { | |
240 c->vector_fmul = vector_fmul_altivec; | |
241 c->vector_fmul_reverse = vector_fmul_reverse_altivec; | |
242 c->vector_fmul_add_add = vector_fmul_add_add_altivec; | |
7675
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
243 if(!(avctx->flags & CODEC_FLAG_BITEXACT)) { |
3581 | 244 c->float_to_int16 = float_to_int16_altivec; |
7675
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
245 c->float_to_int16_interleave = float_to_int16_interleave_altivec; |
ad4bf45b9b63
Introduce float_to_int16_interleave_altivec, tested with vorbis
lu_zero
parents:
7674
diff
changeset
|
246 } |
3581 | 247 } |