Mercurial > libavcodec.hg
annotate ppc/fft_altivec.c @ 12065:1b6557975db1 libavcodec
Fix Doxygen documentation of gb parameter in ff_wma_get_large_val().
author | diego |
---|---|
date | Fri, 02 Jul 2010 11:55:42 +0000 |
parents | c7c32da068ab |
children | 8454bb880008 |
rev | line source |
---|---|
975
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
diff
changeset
|
1 /* |
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
diff
changeset
|
2 * FFT/IFFT transforms |
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
diff
changeset
|
3 * AltiVec-enabled |
12046 | 4 * Copyright (c) 2009 Loren Merritt |
975
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
diff
changeset
|
5 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3036
diff
changeset
|
6 * This file is part of FFmpeg. |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3036
diff
changeset
|
7 * |
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3036
diff
changeset
|
8 * FFmpeg is free software; you can redistribute it and/or |
975
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
diff
changeset
|
9 * modify it under the terms of the GNU Lesser General Public |
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
diff
changeset
|
10 * 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:
3036
diff
changeset
|
11 * version 2.1 of the License, or (at your option) any later version. |
975
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
diff
changeset
|
12 * |
3947
c8c591fe26f8
Change license headers to say 'FFmpeg' instead of 'this program/this library'
diego
parents:
3036
diff
changeset
|
13 * FFmpeg is distributed in the hope that it will be useful, |
975
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
diff
changeset
|
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of |
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
diff
changeset
|
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
diff
changeset
|
16 * Lesser General Public License for more details. |
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
diff
changeset
|
17 * |
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
diff
changeset
|
18 * 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:
3036
diff
changeset
|
19 * License along with FFmpeg; if not, write to the Free Software |
3036
0b546eab515d
Update licensing information: The FSF changed postal address.
diego
parents:
2979
diff
changeset
|
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA |
975
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
diff
changeset
|
21 */ |
11370 | 22 #include "libavcodec/fft.h" |
5750
09f99af1db40
Sanitize altivec code so it can be built with runtime check properly
lu_zero
parents:
5010
diff
changeset
|
23 #include "util_altivec.h" |
12046 | 24 #include "types_altivec.h" |
25 #include "regs.h" | |
11382
50415a8f1451
PPC: move prototypes to headers and make some functions static
mru
parents:
11370
diff
changeset
|
26 |
975
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
diff
changeset
|
27 /** |
1879
dd63cb7e5080
fft_*() renamed into ff_fft_*() patch by (Gildas Bazin <gbazin at altern dot org>)
michael
parents:
1352
diff
changeset
|
28 * Do a complex FFT with the parameters defined in ff_fft_init(). The |
975
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
diff
changeset
|
29 * input data must be permuted before with s->revtab table. No |
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
diff
changeset
|
30 * 1.0/sqrt(n) normalization is done. |
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
diff
changeset
|
31 * AltiVec-enabled |
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
diff
changeset
|
32 * This code assumes that the 'z' pointer is 16 bytes-aligned |
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
diff
changeset
|
33 * It also assumes all FFTComplex are 8 bytes-aligned pair of float |
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
diff
changeset
|
34 */ |
12046 | 35 |
36 // Pointers to functions. Not using function pointer syntax, because | |
37 // that involves an extra level of indirection on some PPC ABIs. | |
38 extern void *ff_fft_dispatch_altivec[2][15]; | |
39 | |
12049 | 40 #if HAVE_GNU_AS |
12046 | 41 // Convert from simd order to C order. |
42 static void swizzle(vec_f *z, int n) | |
43 { | |
44 int i; | |
45 n >>= 1; | |
46 for (i = 0; i < n; i += 2) { | |
47 vec_f re = z[i]; | |
48 vec_f im = z[i+1]; | |
49 z[i] = vec_mergeh(re, im); | |
50 z[i+1] = vec_mergel(re, im); | |
51 } | |
52 } | |
53 | |
12049 | 54 static av_always_inline void fft_dispatch(FFTContext *s, FFTComplex *z, int do_swizzle) |
975
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
diff
changeset
|
55 { |
12046 | 56 register vec_f v14 __asm__("v14") = {0,0,0,0}; |
57 register vec_f v15 __asm__("v15") = *(const vec_f*)ff_cos_16; | |
58 register vec_f v16 __asm__("v16") = {0, 0.38268343, M_SQRT1_2, 0.92387953}; | |
59 register vec_f v17 __asm__("v17") = {-M_SQRT1_2, M_SQRT1_2, M_SQRT1_2,-M_SQRT1_2}; | |
60 register vec_f v18 __asm__("v18") = { M_SQRT1_2, M_SQRT1_2, M_SQRT1_2, M_SQRT1_2}; | |
61 register vec_u8 v19 __asm__("v19") = vcprm(s0,3,2,1); | |
62 register vec_u8 v20 __asm__("v20") = vcprm(0,1,s2,s1); | |
63 register vec_u8 v21 __asm__("v21") = vcprm(2,3,s0,s3); | |
64 register vec_u8 v22 __asm__("v22") = vcprm(2,s3,3,s2); | |
65 register vec_u8 v23 __asm__("v23") = vcprm(0,1,s0,s1); | |
66 register vec_u8 v24 __asm__("v24") = vcprm(2,3,s2,s3); | |
67 register vec_u8 v25 __asm__("v25") = vcprm(2,3,0,1); | |
68 register vec_u8 v26 __asm__("v26") = vcprm(1,2,s3,s0); | |
69 register vec_u8 v27 __asm__("v27") = vcprm(0,3,s2,s1); | |
70 register vec_u8 v28 __asm__("v28") = vcprm(0,2,s1,s3); | |
71 register vec_u8 v29 __asm__("v29") = vcprm(1,3,s0,s2); | |
72 register FFTSample *const*cos_tabs __asm__("r12") = ff_cos_tabs; | |
73 register FFTComplex *zarg __asm__("r3") = z; | |
74 __asm__( | |
75 "mtctr %0 \n" | |
76 "li "r(9)", 16 \n" | |
77 "subi "r(1)","r(1) ",%1 \n" | |
78 "bctrl \n" | |
79 "addi "r(1)","r(1) ",%1 \n" | |
12049 | 80 ::"r"(ff_fft_dispatch_altivec[do_swizzle][s->nbits-2]), "i"(12*sizeof(void*)), |
12046 | 81 "r"(zarg), "r"(cos_tabs), |
82 "v"(v14),"v"(v15),"v"(v16),"v"(v17),"v"(v18),"v"(v19),"v"(v20),"v"(v21), | |
83 "v"(v22),"v"(v23),"v"(v24),"v"(v25),"v"(v26),"v"(v27),"v"(v28),"v"(v29) | |
84 : "lr","ctr","r0","r4","r5","r6","r7","r8","r9","r10","r11", | |
85 "v0","v1","v2","v3","v4","v5","v6","v7","v8","v9","v10","v11","v12","v13" | |
86 ); | |
12049 | 87 if (do_swizzle && s->nbits <= 4) |
12046 | 88 swizzle((vec_f*)z, 1<<s->nbits); |
975
e05d525505c5
fft altivec by Romain Dolbeau - simplified patch, test it on PPC with fft-test and wma decoding
bellard
parents:
diff
changeset
|
89 } |
10175
5cf49858179a
Move per-arch fft init bits into the corresponding subdirs
mru
parents:
9364
diff
changeset
|
90 |
12049 | 91 static void ff_fft_calc_altivec(FFTContext *s, FFTComplex *z) |
92 { | |
93 fft_dispatch(s, z, 1); | |
94 } | |
95 | |
96 static void ff_imdct_half_altivec(FFTContext *s, FFTSample *output, const FFTSample *input) | |
97 { | |
98 int j, k; | |
99 int n = 1 << s->mdct_bits; | |
100 int n4 = n >> 2; | |
101 int n8 = n >> 3; | |
102 int n32 = n >> 5; | |
103 const uint16_t *revtabj = s->revtab; | |
104 const uint16_t *revtabk = s->revtab+n4; | |
105 const vec_f *tcos = (const vec_f*)(s->tcos+n8); | |
106 const vec_f *tsin = (const vec_f*)(s->tsin+n8); | |
107 const vec_f *pin = (const vec_f*)(input+n4); | |
108 vec_f *pout = (vec_f*)(output+n4); | |
109 | |
110 /* pre rotation */ | |
111 k = n32-1; | |
112 do { | |
113 vec_f cos,sin,cos0,sin0,cos1,sin1,re,im,r0,i0,r1,i1,a,b,c,d; | |
114 #define CMULA(p,o0,o1,o2,o3)\ | |
115 a = pin[ k*2+p]; /* { z[k].re, z[k].im, z[k+1].re, z[k+1].im } */\ | |
116 b = pin[-k*2-p-1]; /* { z[-k-2].re, z[-k-2].im, z[-k-1].re, z[-k-1].im } */\ | |
117 re = vec_perm(a, b, vcprm(0,2,s0,s2)); /* { z[k].re, z[k+1].re, z[-k-2].re, z[-k-1].re } */\ | |
118 im = vec_perm(a, b, vcprm(s3,s1,3,1)); /* { z[-k-1].im, z[-k-2].im, z[k+1].im, z[k].im } */\ | |
119 cos = vec_perm(cos0, cos1, vcprm(o0,o1,s##o2,s##o3)); /* { cos[k], cos[k+1], cos[-k-2], cos[-k-1] } */\ | |
120 sin = vec_perm(sin0, sin1, vcprm(o0,o1,s##o2,s##o3));\ | |
121 r##p = im*cos - re*sin;\ | |
122 i##p = re*cos + im*sin; | |
123 #define STORE2(v,dst)\ | |
124 j = dst;\ | |
125 vec_ste(v, 0, output+j*2);\ | |
126 vec_ste(v, 4, output+j*2); | |
127 #define STORE8(p)\ | |
128 a = vec_perm(r##p, i##p, vcprm(0,s0,0,s0));\ | |
129 b = vec_perm(r##p, i##p, vcprm(1,s1,1,s1));\ | |
130 c = vec_perm(r##p, i##p, vcprm(2,s2,2,s2));\ | |
131 d = vec_perm(r##p, i##p, vcprm(3,s3,3,s3));\ | |
132 STORE2(a, revtabk[ p*2-4]);\ | |
133 STORE2(b, revtabk[ p*2-3]);\ | |
134 STORE2(c, revtabj[-p*2+2]);\ | |
135 STORE2(d, revtabj[-p*2+3]); | |
136 | |
137 cos0 = tcos[k]; | |
138 sin0 = tsin[k]; | |
139 cos1 = tcos[-k-1]; | |
140 sin1 = tsin[-k-1]; | |
141 CMULA(0, 0,1,2,3); | |
142 CMULA(1, 2,3,0,1); | |
143 STORE8(0); | |
144 STORE8(1); | |
145 revtabj += 4; | |
146 revtabk -= 4; | |
147 k--; | |
148 } while(k >= 0); | |
149 | |
150 fft_dispatch(s, (FFTComplex*)output, 0); | |
151 | |
152 /* post rotation + reordering */ | |
153 j = -n32; | |
154 k = n32-1; | |
155 do { | |
156 vec_f cos,sin,re,im,a,b,c,d; | |
157 #define CMULB(d0,d1,o)\ | |
158 re = pout[o*2];\ | |
159 im = pout[o*2+1];\ | |
160 cos = tcos[o];\ | |
161 sin = tsin[o];\ | |
162 d0 = im*sin - re*cos;\ | |
163 d1 = re*sin + im*cos; | |
164 | |
165 CMULB(a,b,j); | |
166 CMULB(c,d,k); | |
167 pout[2*j] = vec_perm(a, d, vcprm(0,s3,1,s2)); | |
168 pout[2*j+1] = vec_perm(a, d, vcprm(2,s1,3,s0)); | |
169 pout[2*k] = vec_perm(c, b, vcprm(0,s3,1,s2)); | |
170 pout[2*k+1] = vec_perm(c, b, vcprm(2,s1,3,s0)); | |
171 j++; | |
172 k--; | |
173 } while(k >= 0); | |
174 } | |
175 | |
176 static void ff_imdct_calc_altivec(FFTContext *s, FFTSample *output, const FFTSample *input) | |
177 { | |
178 int k; | |
179 int n = 1 << s->mdct_bits; | |
180 int n4 = n >> 2; | |
181 int n16 = n >> 4; | |
182 vec_u32 sign = {1<<31,1<<31,1<<31,1<<31}; | |
183 vec_u32 *p0 = (vec_u32*)(output+n4); | |
184 vec_u32 *p1 = (vec_u32*)(output+n4*3); | |
185 | |
186 ff_imdct_half_altivec(s, output+n4, input); | |
187 | |
188 for (k = 0; k < n16; k++) { | |
189 vec_u32 a = p0[k] ^ sign; | |
190 vec_u32 b = p1[-k-1]; | |
191 p0[-k-1] = vec_perm(a, a, vcprm(3,2,1,0)); | |
192 p1[k] = vec_perm(b, b, vcprm(3,2,1,0)); | |
193 } | |
194 } | |
195 #endif /* HAVE_GNU_AS */ | |
196 | |
10175
5cf49858179a
Move per-arch fft init bits into the corresponding subdirs
mru
parents:
9364
diff
changeset
|
197 av_cold void ff_fft_init_altivec(FFTContext *s) |
5cf49858179a
Move per-arch fft init bits into the corresponding subdirs
mru
parents:
9364
diff
changeset
|
198 { |
12052 | 199 #if HAVE_GNU_AS |
200 s->fft_calc = ff_fft_calc_altivec; | |
201 s->imdct_calc = ff_imdct_calc_altivec; | |
202 s->imdct_half = ff_imdct_half_altivec; | |
203 #endif | |
10175
5cf49858179a
Move per-arch fft init bits into the corresponding subdirs
mru
parents:
9364
diff
changeset
|
204 } |