Mercurial > audlegacy-plugins
annotate src/madplug/SFMT-alti.c @ 1146:183d03932c9e trunk
[svn] - curl, stdio: convert to plugin2 architecture
author | nenolod |
---|---|
date | Mon, 28 May 2007 03:12:40 -0700 |
parents | 88bf800eba26 |
children |
rev | line source |
---|---|
922
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
1 /** |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
2 * This function represents the recursion formula in AltiVec and BIG ENDIAN. |
926
fa7f7cd029af
[svn] Spell internal & pseudorandom in a consistent way. Add missing altivec.h include in SFMT-alti.h; more work to be done as it still does not build.
chainsaw
parents:
922
diff
changeset
|
3 * @param a a 128-bit part of the internal state array |
fa7f7cd029af
[svn] Spell internal & pseudorandom in a consistent way. Add missing altivec.h include in SFMT-alti.h; more work to be done as it still does not build.
chainsaw
parents:
922
diff
changeset
|
4 * @param b a 128-bit part of the internal state array |
fa7f7cd029af
[svn] Spell internal & pseudorandom in a consistent way. Add missing altivec.h include in SFMT-alti.h; more work to be done as it still does not build.
chainsaw
parents:
922
diff
changeset
|
5 * @param c a 128-bit part of the internal state array |
fa7f7cd029af
[svn] Spell internal & pseudorandom in a consistent way. Add missing altivec.h include in SFMT-alti.h; more work to be done as it still does not build.
chainsaw
parents:
922
diff
changeset
|
6 * @param d a 128-bit part of the internal state array |
922
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
7 * @return output |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
8 */ |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
9 inline static vector unsigned int vec_recursion(vector unsigned int a, |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
10 vector unsigned int b, |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
11 vector unsigned int c, |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
12 vector unsigned int d) { |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
13 |
928
88bf800eba26
[svn] - use C-style vectors instead of apple's stuff
nenolod
parents:
926
diff
changeset
|
14 const vector unsigned int sl1 = (vector unsigned int){SL1, SL1, SL1, SL1}; |
88bf800eba26
[svn] - use C-style vectors instead of apple's stuff
nenolod
parents:
926
diff
changeset
|
15 const vector unsigned int sr1 = (vector unsigned int){SR1, SR1, SR1, SR1}; |
922
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
16 #ifdef ONLY64 |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
17 const vector unsigned int mask = (vector unsigned int) |
928
88bf800eba26
[svn] - use C-style vectors instead of apple's stuff
nenolod
parents:
926
diff
changeset
|
18 {MSK2, MSK1, MSK4, MSK3}; |
922
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
19 const vector unsigned char perm_sl = ALTI_SL2_PERM64; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
20 const vector unsigned char perm_sr = ALTI_SR2_PERM64; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
21 #else |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
22 const vector unsigned int mask = (vector unsigned int) |
928
88bf800eba26
[svn] - use C-style vectors instead of apple's stuff
nenolod
parents:
926
diff
changeset
|
23 {MSK1, MSK2, MSK3, MSK4}; |
922
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
24 const vector unsigned char perm_sl = ALTI_SL2_PERM; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
25 const vector unsigned char perm_sr = ALTI_SR2_PERM; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
26 #endif |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
27 vector unsigned int v, w, x, y, z; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
28 x = vec_perm(a, (vector unsigned int)perm_sl, perm_sl); |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
29 v = a; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
30 y = vec_sr(b, sr1); |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
31 z = vec_perm(c, (vector unsigned int)perm_sr, perm_sr); |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
32 w = vec_sl(d, sl1); |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
33 z = vec_xor(z, w); |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
34 y = vec_and(y, mask); |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
35 v = vec_xor(v, x); |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
36 z = vec_xor(z, y); |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
37 z = vec_xor(z, v); |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
38 return z; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
39 } |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
40 |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
41 /** |
926
fa7f7cd029af
[svn] Spell internal & pseudorandom in a consistent way. Add missing altivec.h include in SFMT-alti.h; more work to be done as it still does not build.
chainsaw
parents:
922
diff
changeset
|
42 * This function fills the internal state array with pseudorandom |
922
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
43 * integers. |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
44 */ |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
45 inline static void gen_rand_all(void) { |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
46 int i; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
47 vector unsigned int r, r1, r2; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
48 |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
49 r1 = sfmt[N - 2].s; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
50 r2 = sfmt[N - 1].s; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
51 for (i = 0; i < N - POS1; i++) { |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
52 r = vec_recursion(sfmt[i].s, sfmt[i + POS1].s, r1, r2); |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
53 sfmt[i].s = r; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
54 r1 = r2; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
55 r2 = r; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
56 } |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
57 for (; i < N; i++) { |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
58 r = vec_recursion(sfmt[i].s, sfmt[i + POS1 - N].s, r1, r2); |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
59 sfmt[i].s = r; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
60 r1 = r2; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
61 r2 = r; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
62 } |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
63 } |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
64 |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
65 /** |
926
fa7f7cd029af
[svn] Spell internal & pseudorandom in a consistent way. Add missing altivec.h include in SFMT-alti.h; more work to be done as it still does not build.
chainsaw
parents:
922
diff
changeset
|
66 * This function fills the user-specified array with pseudorandom |
922
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
67 * integers. |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
68 * |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
69 * @param array an 128-bit array to be filled by pseudorandom numbers. |
926
fa7f7cd029af
[svn] Spell internal & pseudorandom in a consistent way. Add missing altivec.h include in SFMT-alti.h; more work to be done as it still does not build.
chainsaw
parents:
922
diff
changeset
|
70 * @param size number of 128-bit pseudorandom numbers to be generated. |
922
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
71 */ |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
72 inline static void gen_rand_array(w128_t array[], int size) { |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
73 int i, j; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
74 vector unsigned int r, r1, r2; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
75 |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
76 r1 = sfmt[N - 2].s; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
77 r2 = sfmt[N - 1].s; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
78 for (i = 0; i < N - POS1; i++) { |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
79 r = vec_recursion(sfmt[i].s, sfmt[i + POS1].s, r1, r2); |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
80 array[i].s = r; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
81 r1 = r2; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
82 r2 = r; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
83 } |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
84 for (; i < N; i++) { |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
85 r = vec_recursion(sfmt[i].s, array[i + POS1 - N].s, r1, r2); |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
86 array[i].s = r; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
87 r1 = r2; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
88 r2 = r; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
89 } |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
90 /* main loop */ |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
91 for (; i < size - N; i++) { |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
92 r = vec_recursion(array[i - N].s, array[i + POS1 - N].s, r1, r2); |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
93 array[i].s = r; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
94 r1 = r2; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
95 r2 = r; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
96 } |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
97 for (j = 0; j < 2 * N - size; j++) { |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
98 sfmt[j].s = array[j + size - N].s; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
99 } |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
100 for (; i < size; i++) { |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
101 r = vec_recursion(array[i - N].s, array[i + POS1 - N].s, r1, r2); |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
102 array[i].s = r; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
103 sfmt[j++].s = r; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
104 r1 = r2; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
105 r2 = r; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
106 } |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
107 } |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
108 |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
109 #ifndef ONLY64 |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
110 /** |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
111 * This function swaps high and low 32-bit of 64-bit integers in user |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
112 * specified array. |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
113 * |
926
fa7f7cd029af
[svn] Spell internal & pseudorandom in a consistent way. Add missing altivec.h include in SFMT-alti.h; more work to be done as it still does not build.
chainsaw
parents:
922
diff
changeset
|
114 * @param array an 128-bit array to be swapped. |
922
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
115 * @param size size of 128-bit array. |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
116 */ |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
117 inline static void swap(w128_t array[], int size) { |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
118 int i; |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
119 const vector unsigned char perm = (vector unsigned char) |
928
88bf800eba26
[svn] - use C-style vectors instead of apple's stuff
nenolod
parents:
926
diff
changeset
|
120 {4, 5, 6, 7, 0, 1, 2, 3, 12, 13, 14, 15, 8, 9, 10, 11}; |
922
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
121 |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
122 for (i = 0; i < size; i++) { |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
123 array[i].s = vec_perm(array[i].s, (vector unsigned int)perm, perm); |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
124 } |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
125 } |
7e14701aef54
[svn] - replace random number generator in dithering code with SIMD-oriented Fast Mersenne Twister (SFMT). it reduces CPU load on SSE2 or AltiVec capable platform.
yaz
parents:
diff
changeset
|
126 #endif |