29 #define vs16(v) ((vector signed short) (v))
30 #define vs32(v) ((vector signed int) (v))
31 #define vu8(v) ((vector unsigned char) (v))
32 #define vu16(v) ((vector unsigned short) (v))
33 #define vu32(v) ((vector unsigned int) (v))
35 #define C1 0.98078525066375732421875000
36 #define C2 0.92387950420379638671875000
37 #define C3 0.83146959543228149414062500
38 #define C4 0.70710676908493041992187500
39 #define C5 0.55557024478912353515625000
40 #define C6 0.38268342614173889160156250
41 #define C7 0.19509032368659973144531250
42 #define SQRT_2 1.41421353816986083984375000
46 #define W2 (SQRT_2 * C6)
47 #define W3 (SQRT_2 * C3)
48 #define W4 (SQRT_2 * (-C1 + C3 + C5 - C7))
49 #define W5 (SQRT_2 * (C1 + C3 - C5 + C7))
50 #define W6 (SQRT_2 * (C1 + C3 + C5 - C7))
51 #define W7 (SQRT_2 * (C1 + C3 - C5 - C7))
52 #define W8 (SQRT_2 * (C7 - C3))
53 #define W9 (SQRT_2 * (-C1 - C3))
54 #define WA (SQRT_2 * (-C3 - C5))
55 #define WB (SQRT_2 * (C5 - C3))
63 #define LD_W0 vec_splat(cnsts0, 0)
64 #define LD_W1 vec_splat(cnsts0, 1)
65 #define LD_W2 vec_splat(cnsts0, 2)
66 #define LD_W3 vec_splat(cnsts0, 3)
67 #define LD_W4 vec_splat(cnsts1, 0)
68 #define LD_W5 vec_splat(cnsts1, 1)
69 #define LD_W6 vec_splat(cnsts1, 2)
70 #define LD_W7 vec_splat(cnsts1, 3)
71 #define LD_W8 vec_splat(cnsts2, 0)
72 #define LD_W9 vec_splat(cnsts2, 1)
73 #define LD_WA vec_splat(cnsts2, 2)
74 #define LD_WB vec_splat(cnsts2, 3)
76 #define FDCTROW(b0, b1, b2, b3, b4, b5, b6, b7) \
77 x0 = vec_add(b0, b7); \
78 x7 = vec_sub(b0, b7); \
79 x1 = vec_add(b1, b6); \
80 x6 = vec_sub(b1, b6); \
81 x2 = vec_add(b2, b5); \
82 x5 = vec_sub(b2, b5); \
83 x3 = vec_add(b3, b4); \
84 x4 = vec_sub(b3, b4); \
86 b7 = vec_add(x0, x3); \
87 b1 = vec_add(x1, x2); \
88 b0 = vec_add(b7, b1); \
89 b4 = vec_sub(b7, b1); \
91 b2 = vec_sub(x0, x3); \
92 b6 = vec_sub(x1, x2); \
93 b5 = vec_add(b6, b2); \
95 b5 = vec_madd(cnst, b5, mzero); \
97 b2 = vec_madd(cnst, b2, b5); \
99 b6 = vec_madd(cnst, b6, b5); \
101 x0 = vec_add(x4, x7); \
102 x1 = vec_add(x5, x6); \
103 x2 = vec_add(x4, x6); \
104 x3 = vec_add(x5, x7); \
105 x8 = vec_add(x2, x3); \
107 x8 = vec_madd(cnst, x8, mzero); \
110 x0 = vec_madd(cnst, x0, mzero); \
112 x1 = vec_madd(cnst, x1, mzero); \
114 x2 = vec_madd(cnst, x2, x8); \
116 x3 = vec_madd(cnst, x3, x8); \
119 b7 = vec_madd(cnst, x4, x0); \
121 b5 = vec_madd(cnst, x5, x1); \
123 b3 = vec_madd(cnst, x6, x1); \
125 b1 = vec_madd(cnst, x7, x0); \
127 b7 = vec_add(b7, x2); \
128 b5 = vec_add(b5, x3); \
129 b3 = vec_add(b3, x2); \
130 b1 = vec_add(b1, x3) \
133 #define FDCTCOL(b0, b1, b2, b3, b4, b5, b6, b7) \
134 x0 = vec_add(b0, b7); \
135 x7 = vec_sub(b0, b7); \
136 x1 = vec_add(b1, b6); \
137 x6 = vec_sub(b1, b6); \
138 x2 = vec_add(b2, b5); \
139 x5 = vec_sub(b2, b5); \
140 x3 = vec_add(b3, b4); \
141 x4 = vec_sub(b3, b4); \
143 b7 = vec_add(x0, x3); \
144 b1 = vec_add(x1, x2); \
145 b0 = vec_add(b7, b1); \
146 b4 = vec_sub(b7, b1); \
148 b2 = vec_sub(x0, x3); \
149 b6 = vec_sub(x1, x2); \
150 b5 = vec_add(b6, b2); \
152 b5 = vec_madd(cnst, b5, mzero); \
154 b2 = vec_madd(cnst, b2, b5); \
156 b6 = vec_madd(cnst, b6, b5); \
158 x0 = vec_add(x4, x7); \
159 x1 = vec_add(x5, x6); \
160 x2 = vec_add(x4, x6); \
161 x3 = vec_add(x5, x7); \
162 x8 = vec_add(x2, x3); \
164 x8 = vec_madd(cnst, x8, mzero); \
167 x0 = vec_madd(cnst, x0, mzero); \
169 x1 = vec_madd(cnst, x1, mzero); \
171 x2 = vec_madd(cnst, x2, x8); \
173 x3 = vec_madd(cnst, x3, x8); \
176 b7 = vec_madd(cnst, x4, x0); \
178 b5 = vec_madd(cnst, x5, x1); \
180 b3 = vec_madd(cnst, x6, x1); \
182 b1 = vec_madd(cnst, x7, x0); \
184 b7 = vec_add(b7, x2); \
185 b5 = vec_add(b5, x3); \
186 b3 = vec_add(b3, x2); \
187 b1 = vec_add(b1, x3) \
193 vector
signed short *bp;
195 vector
float b00, b10, b20, b30, b40, b50, b60, b70;
196 vector
float b01, b11, b21, b31, b41, b51, b61, b71;
197 vector
float mzero, cnst, cnsts0, cnsts1, cnsts2;
198 vector
float x0, x1, x2, x3, x4, x5, x6, x7, x8;
202 mzero = ((vector float) vec_splat_u32(-1));
203 mzero = ((vector float) vec_sl(
vu32(mzero),
vu32(mzero)));
204 cnsts0 = vec_ld(0, cp);
206 cnsts1 = vec_ld(0, cp);
208 cnsts2 = vec_ld(0, cp);
212 #define MERGE_S16(hl, a, b) vec_merge ## hl(vs16(a), vs16(b))
214 bp = (vector
signed short *) block;
215 b00 = ((vector float) vec_ld(0, bp));
216 b40 = ((vector float) vec_ld(16 * 4, bp));
217 b01 = ((vector float)
MERGE_S16(h, b00, b40));
218 b11 = ((vector float)
MERGE_S16(l, b00, b40));
220 b10 = ((vector float) vec_ld(0, bp));
221 b50 = ((vector float) vec_ld(16 * 4, bp));
222 b21 = ((vector float)
MERGE_S16(h, b10, b50));
223 b31 = ((vector float)
MERGE_S16(l, b10, b50));
225 b20 = ((vector float) vec_ld(0, bp));
226 b60 = ((vector float) vec_ld(16 * 4, bp));
227 b41 = ((vector float)
MERGE_S16(h, b20, b60));
228 b51 = ((vector float)
MERGE_S16(l, b20, b60));
230 b30 = ((vector float) vec_ld(0, bp));
231 b70 = ((vector float) vec_ld(16 * 4, bp));
232 b61 = ((vector float)
MERGE_S16(h, b30, b70));
233 b71 = ((vector float)
MERGE_S16(l, b30, b70));
235 x0 = ((vector float)
MERGE_S16(h, b01, b41));
236 x1 = ((vector float)
MERGE_S16(l, b01, b41));
237 x2 = ((vector float)
MERGE_S16(h, b11, b51));
238 x3 = ((vector float)
MERGE_S16(l, b11, b51));
239 x4 = ((vector float)
MERGE_S16(h, b21, b61));
240 x5 = ((vector float)
MERGE_S16(l, b21, b61));
241 x6 = ((vector float)
MERGE_S16(h, b31, b71));
242 x7 = ((vector float)
MERGE_S16(l, b31, b71));
244 b00 = ((vector float)
MERGE_S16(h, x0, x4));
245 b10 = ((vector float)
MERGE_S16(l, x0, x4));
246 b20 = ((vector float)
MERGE_S16(h, x1, x5));
247 b30 = ((vector float)
MERGE_S16(l, x1, x5));
248 b40 = ((vector float)
MERGE_S16(h, x2, x6));
249 b50 = ((vector float)
MERGE_S16(l, x2, x6));
250 b60 = ((vector float)
MERGE_S16(h, x3, x7));
251 b70 = ((vector float)
MERGE_S16(l, x3, x7));
261 x0 = ((vector float) vec_add(
vs16(b00),
vs16(b70)));
262 x7 = ((vector float) vec_sub(
vs16(b00),
vs16(b70)));
263 x1 = ((vector float) vec_add(
vs16(b10),
vs16(b60)));
264 x6 = ((vector float) vec_sub(
vs16(b10),
vs16(b60)));
265 x2 = ((vector float) vec_add(
vs16(b20),
vs16(b50)));
266 x5 = ((vector float) vec_sub(
vs16(b20),
vs16(b50)));
267 x3 = ((vector float) vec_add(
vs16(b30),
vs16(b40)));
268 x4 = ((vector float) vec_sub(
vs16(b30),
vs16(b40)));
270 b70 = ((vector float) vec_add(
vs16(x0),
vs16(x3)));
271 b10 = ((vector float) vec_add(
vs16(x1),
vs16(x2)));
273 b00 = ((vector float) vec_add(
vs16(b70),
vs16(b10)));
274 b40 = ((vector float) vec_sub(
vs16(b70),
vs16(b10)));
277 b ## n ## 1 = ((vector float) vec_unpackl(vs16(b ## n ## 0))); \
278 b ## n ## 0 = ((vector float) vec_unpackh(vs16(b ## n ## 0))); \
279 b ## n ## 1 = vec_ctf(vs32(b ## n ## 1), 0); \
280 b ## n ## 0 = vec_ctf(vs32(b ## n ## 0), 0)
285 b20 = ((vector float) vec_sub(
vs16(x0),
vs16(x3)));
286 b60 = ((vector float) vec_sub(
vs16(x1),
vs16(x2)));
293 x0 = vec_add(b60, b20);
294 x1 = vec_add(b61, b21);
297 x0 = vec_madd(cnst, x0, mzero);
298 x1 = vec_madd(cnst, x1, mzero);
300 b20 = vec_madd(cnst, b20, x0);
301 b21 = vec_madd(cnst, b21, x1);
303 b60 = vec_madd(cnst, b60, x0);
304 b61 = vec_madd(cnst, b61, x1);
307 b ## 0 = ((vector float) vec_unpackh(vs16(x))); \
308 b ## 1 = ((vector float) vec_unpackl(vs16(x))); \
309 b ## 0 = vec_ctf(vs32(b ## 0), 0); \
310 b ## 1 = vec_ctf(vs32(b ## 1), 0)
319 x0 = vec_add(b70, b10);
320 x1 = vec_add(b50, b30);
321 x2 = vec_add(b70, b30);
322 x3 = vec_add(b50, b10);
323 x8 = vec_add(x2, x3);
325 x8 = vec_madd(cnst, x8, mzero);
328 x0 = vec_madd(cnst, x0, mzero);
330 x1 = vec_madd(cnst, x1, mzero);
332 x2 = vec_madd(cnst, x2, x8);
334 x3 = vec_madd(cnst, x3, x8);
337 b70 = vec_madd(cnst, b70, x0);
339 b50 = vec_madd(cnst, b50, x1);
341 b30 = vec_madd(cnst, b30, x1);
343 b10 = vec_madd(cnst, b10, x0);
345 b70 = vec_add(b70, x2);
346 b50 = vec_add(b50, x3);
347 b30 = vec_add(b30, x2);
348 b10 = vec_add(b10, x3);
350 x0 = vec_add(b71, b11);
351 x1 = vec_add(b51, b31);
352 x2 = vec_add(b71, b31);
353 x3 = vec_add(b51, b11);
354 x8 = vec_add(x2, x3);
356 x8 = vec_madd(cnst, x8, mzero);
359 x0 = vec_madd(cnst, x0, mzero);
361 x1 = vec_madd(cnst, x1, mzero);
363 x2 = vec_madd(cnst, x2, x8);
365 x3 = vec_madd(cnst, x3, x8);
368 b71 = vec_madd(cnst, b71, x0);
370 b51 = vec_madd(cnst, b51, x1);
372 b31 = vec_madd(cnst, b31, x1);
374 b11 = vec_madd(cnst, b11, x0);
376 b71 = vec_add(b71, x2);
377 b51 = vec_add(b51, x3);
378 b31 = vec_add(b31, x2);
379 b11 = vec_add(b11, x3);
383 x0 = vec_mergel(b00, b20);
384 x1 = vec_mergeh(b00, b20);
385 x2 = vec_mergel(b10, b30);
386 x3 = vec_mergeh(b10, b30);
388 b00 = vec_mergeh(x1, x3);
389 b10 = vec_mergel(x1, x3);
390 b20 = vec_mergeh(x0, x2);
391 b30 = vec_mergel(x0, x2);
393 x4 = vec_mergel(b41, b61);
394 x5 = vec_mergeh(b41, b61);
395 x6 = vec_mergel(b51, b71);
396 x7 = vec_mergeh(b51, b71);
398 b41 = vec_mergeh(x5, x7);
399 b51 = vec_mergel(x5, x7);
400 b61 = vec_mergeh(x4, x6);
401 b71 = vec_mergel(x4, x6);
403 x0 = vec_mergel(b01, b21);
404 x1 = vec_mergeh(b01, b21);
405 x2 = vec_mergel(b11, b31);
406 x3 = vec_mergeh(b11, b31);
408 x4 = vec_mergel(b40, b60);
409 x5 = vec_mergeh(b40, b60);
410 x6 = vec_mergel(b50, b70);
411 x7 = vec_mergeh(b50, b70);
413 b40 = vec_mergeh(x1, x3);
414 b50 = vec_mergel(x1, x3);
415 b60 = vec_mergeh(x0, x2);
416 b70 = vec_mergel(x0, x2);
418 b01 = vec_mergeh(x5, x7);
419 b11 = vec_mergel(x5, x7);
420 b21 = vec_mergeh(x4, x6);
421 b31 = vec_mergel(x4, x6);
424 FDCTCOL(b00, b10, b20, b30, b40, b50, b60, b70);
425 FDCTCOL(b01, b11, b21, b31, b41, b51, b61, b71);
429 b ## n ## 0 = vec_round(b ## n ## 0); \
430 b ## n ## 1 = vec_round(b ## n ## 1); \
431 b ## n ## 0 = ((vector float) vec_cts(b ## n ## 0, 0)); \
432 b ## n ## 1 = ((vector float) vec_cts(b ## n ## 1, 0)); \
433 b ## n ## 0 = ((vector float) vec_pack(vs32(b ## n ## 0), \
434 vs32(b ## n ## 1))); \
435 vec_st(vs16(b ## n ## 0), 0, bp)
437 bp = (vector
signed short *) block;