2 * Copyright (c) 2015 Manojkumar Bhosale (Manojkumar.Bhosale@imgtec.com)
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
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.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 #ifndef AVUTIL_MIPS_GENERIC_MACROS_MSA_H
22 #define AVUTIL_MIPS_GENERIC_MACROS_MSA_H
33 #define ALLOC_ALIGNED(align) __attribute__ ((aligned((align) << 1)))
35 #define LD_V(RTYPE, psrc) *((RTYPE *)(psrc))
36 #define LD_UB(...) LD_V(v16u8, __VA_ARGS__)
37 #define LD_SB(...) LD_V(v16i8, __VA_ARGS__)
38 #define LD_UH(...) LD_V(v8u16, __VA_ARGS__)
39 #define LD_SH(...) LD_V(v8i16, __VA_ARGS__)
40 #define LD_UW(...) LD_V(v4u32, __VA_ARGS__)
41 #define LD_SW(...) LD_V(v4i32, __VA_ARGS__)
43 #define ST_V(RTYPE, in, pdst) *((RTYPE *)(pdst)) = (in)
44 #define ST_UB(...) ST_V(v16u8, __VA_ARGS__)
45 #define ST_SB(...) ST_V(v16i8, __VA_ARGS__)
46 #define ST_UH(...) ST_V(v8u16, __VA_ARGS__)
47 #define ST_SH(...) ST_V(v8i16, __VA_ARGS__)
48 #define ST_UW(...) ST_V(v4u32, __VA_ARGS__)
49 #define ST_SW(...) ST_V(v4i32, __VA_ARGS__)
51 #if (__mips_isa_rev >= 6)
54 uint16_t val_lh_m = *(uint16_t *)(psrc); \
60 uint32_t val_lw_m = *(uint32_t *)(psrc); \
67 uint64_t val_ld_m = *(uint64_t *)(psrc); \
70 #else // !(__mips == 64)
73 uint8_t *psrc_ld_m = (uint8_t *) (psrc); \
74 uint32_t val0_ld_m, val1_ld_m; \
75 uint64_t val_ld_m = 0; \
77 val0_ld_m = LW(psrc_ld_m); \
78 val1_ld_m = LW(psrc_ld_m + 4); \
80 val_ld_m = (uint64_t) (val1_ld_m); \
81 val_ld_m = (uint64_t) ((val_ld_m << 32) & 0xFFFFFFFF00000000); \
82 val_ld_m = (uint64_t) (val_ld_m | (uint64_t) val0_ld_m); \
86 #endif // (__mips == 64)
88 #define SH(val, pdst) *(uint16_t *)(pdst) = (val);
89 #define SW(val, pdst) *(uint32_t *)(pdst) = (val);
90 #define SD(val, pdst) *(uint64_t *)(pdst) = (val);
92 #else // !(__mips_isa_rev >= 6)
95 uint8_t *psrc_lh_m = (uint8_t *) (psrc); \
99 "ulh %[val_lh_m], %[psrc_lh_m] \n\t" \
101 : [val_lh_m] "=r" (val_lh_m) \
102 : [psrc_lh_m] "m" (*psrc_lh_m) \
110 uint8_t *psrc_lw_m = (uint8_t *) (psrc); \
114 "lwr %[val_lw_m], 0(%[psrc_lw_m]) \n\t" \
115 "lwl %[val_lw_m], 3(%[psrc_lw_m]) \n\t" \
117 : [val_lw_m] "=&r"(val_lw_m) \
118 : [psrc_lw_m] "r"(psrc_lw_m) \
127 uint8_t *psrc_ld_m = (uint8_t *) (psrc); \
128 uint64_t val_ld_m = 0; \
131 "ldr %[val_ld_m], 0(%[psrc_ld_m]) \n\t" \
132 "ldl %[val_ld_m], 7(%[psrc_ld_m]) \n\t" \
134 : [val_ld_m] "=&r" (val_ld_m) \
135 : [psrc_ld_m] "r" (psrc_ld_m) \
140 #else // !(__mips == 64)
143 uint8_t *psrc_ld_m = (uint8_t *) (psrc); \
144 uint32_t val0_ld_m, val1_ld_m; \
145 uint64_t val_ld_m = 0; \
147 val0_ld_m = LW(psrc_ld_m); \
148 val1_ld_m = LW(psrc_ld_m + 4); \
150 val_ld_m = (uint64_t) (val1_ld_m); \
151 val_ld_m = (uint64_t) ((val_ld_m << 32) & 0xFFFFFFFF00000000); \
152 val_ld_m = (uint64_t) (val_ld_m | (uint64_t) val0_ld_m); \
156 #endif // (__mips == 64)
158 #define SH(val, pdst) \
160 uint8_t *pdst_sh_m = (uint8_t *) (pdst); \
161 uint16_t val_sh_m = (val); \
164 "ush %[val_sh_m], %[pdst_sh_m] \n\t" \
166 : [pdst_sh_m] "=m" (*pdst_sh_m) \
167 : [val_sh_m] "r" (val_sh_m) \
171 #define SW(val, pdst) \
173 uint8_t *pdst_sw_m = (uint8_t *) (pdst); \
174 uint32_t val_sw_m = (val); \
177 "usw %[val_sw_m], %[pdst_sw_m] \n\t" \
179 : [pdst_sw_m] "=m" (*pdst_sw_m) \
180 : [val_sw_m] "r" (val_sw_m) \
184 #define SD(val, pdst) \
186 uint8_t *pdst_sd_m = (uint8_t *) (pdst); \
187 uint32_t val0_sd_m, val1_sd_m; \
189 val0_sd_m = (uint32_t) ((val) & 0x00000000FFFFFFFF); \
190 val1_sd_m = (uint32_t) (((val) >> 32) & 0x00000000FFFFFFFF); \
192 SW(val0_sd_m, pdst_sd_m); \
193 SW(val1_sd_m, pdst_sd_m + 4); \
195 #endif // (__mips_isa_rev >= 6)
197 /* Description : Load 4 words with stride
198 Arguments : Inputs - psrc (source pointer to load from)
200 Outputs - out0, out1, out2, out3
201 Details : Loads word in 'out0' from (psrc)
202 Loads word in 'out1' from (psrc + stride)
203 Loads word in 'out2' from (psrc + 2 * stride)
204 Loads word in 'out3' from (psrc + 3 * stride)
206 #define LW4(psrc, stride, out0, out1, out2, out3) \
209 out1 = LW((psrc) + stride); \
210 out2 = LW((psrc) + 2 * stride); \
211 out3 = LW((psrc) + 3 * stride); \
214 #define LW2(psrc, stride, out0, out1) \
217 out1 = LW((psrc) + stride); \
220 /* Description : Load double words with stride
221 Arguments : Inputs - psrc (source pointer to load from)
224 Details : Loads double word in 'out0' from (psrc)
225 Loads double word in 'out1' from (psrc + stride)
227 #define LD2(psrc, stride, out0, out1) \
230 out1 = LD((psrc) + stride); \
232 #define LD4(psrc, stride, out0, out1, out2, out3) \
234 LD2((psrc), stride, out0, out1); \
235 LD2((psrc) + 2 * stride, stride, out2, out3); \
238 /* Description : Store 4 words with stride
239 Arguments : Inputs - in0, in1, in2, in3, pdst, stride
240 Details : Stores word from 'in0' to (pdst)
241 Stores word from 'in1' to (pdst + stride)
242 Stores word from 'in2' to (pdst + 2 * stride)
243 Stores word from 'in3' to (pdst + 3 * stride)
245 #define SW4(in0, in1, in2, in3, pdst, stride) \
248 SW(in1, (pdst) + stride); \
249 SW(in2, (pdst) + 2 * stride); \
250 SW(in3, (pdst) + 3 * stride); \
253 /* Description : Store 4 double words with stride
254 Arguments : Inputs - in0, in1, in2, in3, pdst, stride
255 Details : Stores double word from 'in0' to (pdst)
256 Stores double word from 'in1' to (pdst + stride)
257 Stores double word from 'in2' to (pdst + 2 * stride)
258 Stores double word from 'in3' to (pdst + 3 * stride)
260 #define SD4(in0, in1, in2, in3, pdst, stride) \
263 SD(in1, (pdst) + stride); \
264 SD(in2, (pdst) + 2 * stride); \
265 SD(in3, (pdst) + 3 * stride); \
268 /* Description : Load vector elements with stride
269 Arguments : Inputs - psrc (source pointer to load from)
272 Return Type - as per RTYPE
273 Details : Loads elements in 'out0' from (psrc)
274 Loads elements in 'out1' from (psrc + stride)
276 #define LD_V2(RTYPE, psrc, stride, out0, out1) \
278 out0 = LD_V(RTYPE, (psrc)); \
279 out1 = LD_V(RTYPE, (psrc) + stride); \
281 #define LD_UB2(...) LD_V2(v16u8, __VA_ARGS__)
282 #define LD_SB2(...) LD_V2(v16i8, __VA_ARGS__)
283 #define LD_UH2(...) LD_V2(v8u16, __VA_ARGS__)
284 #define LD_SH2(...) LD_V2(v8i16, __VA_ARGS__)
285 #define LD_SW2(...) LD_V2(v4i32, __VA_ARGS__)
287 #define LD_V3(RTYPE, psrc, stride, out0, out1, out2) \
289 LD_V2(RTYPE, (psrc), stride, out0, out1); \
290 out2 = LD_V(RTYPE, (psrc) + 2 * stride); \
292 #define LD_UB3(...) LD_V3(v16u8, __VA_ARGS__)
293 #define LD_SB3(...) LD_V3(v16i8, __VA_ARGS__)
295 #define LD_V4(RTYPE, psrc, stride, out0, out1, out2, out3) \
297 LD_V2(RTYPE, (psrc), stride, out0, out1); \
298 LD_V2(RTYPE, (psrc) + 2 * stride , stride, out2, out3); \
300 #define LD_UB4(...) LD_V4(v16u8, __VA_ARGS__)
301 #define LD_SB4(...) LD_V4(v16i8, __VA_ARGS__)
302 #define LD_UH4(...) LD_V4(v8u16, __VA_ARGS__)
303 #define LD_SH4(...) LD_V4(v8i16, __VA_ARGS__)
304 #define LD_SW4(...) LD_V4(v4i32, __VA_ARGS__)
306 #define LD_V5(RTYPE, psrc, stride, out0, out1, out2, out3, out4) \
308 LD_V4(RTYPE, (psrc), stride, out0, out1, out2, out3); \
309 out4 = LD_V(RTYPE, (psrc) + 4 * stride); \
311 #define LD_UB5(...) LD_V5(v16u8, __VA_ARGS__)
312 #define LD_SB5(...) LD_V5(v16i8, __VA_ARGS__)
314 #define LD_V6(RTYPE, psrc, stride, out0, out1, out2, out3, out4, out5) \
316 LD_V4(RTYPE, (psrc), stride, out0, out1, out2, out3); \
317 LD_V2(RTYPE, (psrc) + 4 * stride, stride, out4, out5); \
319 #define LD_UB6(...) LD_V6(v16u8, __VA_ARGS__)
320 #define LD_SB6(...) LD_V6(v16i8, __VA_ARGS__)
321 #define LD_UH6(...) LD_V6(v8u16, __VA_ARGS__)
322 #define LD_SH6(...) LD_V6(v8i16, __VA_ARGS__)
324 #define LD_V7(RTYPE, psrc, stride, \
325 out0, out1, out2, out3, out4, out5, out6) \
327 LD_V5(RTYPE, (psrc), stride, out0, out1, out2, out3, out4); \
328 LD_V2(RTYPE, (psrc) + 5 * stride, stride, out5, out6); \
330 #define LD_UB7(...) LD_V7(v16u8, __VA_ARGS__)
331 #define LD_SB7(...) LD_V7(v16i8, __VA_ARGS__)
333 #define LD_V8(RTYPE, psrc, stride, \
334 out0, out1, out2, out3, out4, out5, out6, out7) \
336 LD_V4(RTYPE, (psrc), stride, out0, out1, out2, out3); \
337 LD_V4(RTYPE, (psrc) + 4 * stride, stride, out4, out5, out6, out7); \
339 #define LD_UB8(...) LD_V8(v16u8, __VA_ARGS__)
340 #define LD_SB8(...) LD_V8(v16i8, __VA_ARGS__)
341 #define LD_UH8(...) LD_V8(v8u16, __VA_ARGS__)
342 #define LD_SH8(...) LD_V8(v8i16, __VA_ARGS__)
343 #define LD_SW8(...) LD_V8(v4i32, __VA_ARGS__)
345 #define LD_V16(RTYPE, psrc, stride, \
346 out0, out1, out2, out3, out4, out5, out6, out7, \
347 out8, out9, out10, out11, out12, out13, out14, out15) \
349 LD_V8(RTYPE, (psrc), stride, \
350 out0, out1, out2, out3, out4, out5, out6, out7); \
351 LD_V8(RTYPE, (psrc) + 8 * stride, stride, \
352 out8, out9, out10, out11, out12, out13, out14, out15); \
354 #define LD_SH16(...) LD_V16(v8i16, __VA_ARGS__)
356 /* Description : Store vectors with stride
357 Arguments : Inputs - in0, in1, stride
358 Outputs - pdst (destination pointer to store to)
359 Details : Stores elements from 'in0' to (pdst)
360 Stores elements from 'in1' to (pdst + stride)
362 #define ST_V2(RTYPE, in0, in1, pdst, stride) \
364 ST_V(RTYPE, in0, (pdst)); \
365 ST_V(RTYPE, in1, (pdst) + stride); \
367 #define ST_UB2(...) ST_V2(v16u8, __VA_ARGS__)
368 #define ST_SB2(...) ST_V2(v16i8, __VA_ARGS__)
369 #define ST_UH2(...) ST_V2(v8u16, __VA_ARGS__)
370 #define ST_SH2(...) ST_V2(v8i16, __VA_ARGS__)
371 #define ST_SW2(...) ST_V2(v4i32, __VA_ARGS__)
373 #define ST_V4(RTYPE, in0, in1, in2, in3, pdst, stride) \
375 ST_V2(RTYPE, in0, in1, (pdst), stride); \
376 ST_V2(RTYPE, in2, in3, (pdst) + 2 * stride, stride); \
378 #define ST_UB4(...) ST_V4(v16u8, __VA_ARGS__)
379 #define ST_SB4(...) ST_V4(v16i8, __VA_ARGS__)
380 #define ST_SH4(...) ST_V4(v8i16, __VA_ARGS__)
381 #define ST_SW4(...) ST_V4(v4i32, __VA_ARGS__)
383 #define ST_V6(RTYPE, in0, in1, in2, in3, in4, in5, pdst, stride) \
385 ST_V4(RTYPE, in0, in1, in2, in3, (pdst), stride); \
386 ST_V2(RTYPE, in4, in5, (pdst) + 4 * stride, stride); \
388 #define ST_SH6(...) ST_V6(v8i16, __VA_ARGS__)
390 #define ST_V8(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, pdst, stride) \
392 ST_V4(RTYPE, in0, in1, in2, in3, (pdst), stride); \
393 ST_V4(RTYPE, in4, in5, in6, in7, (pdst) + 4 * stride, stride); \
395 #define ST_UB8(...) ST_V8(v16u8, __VA_ARGS__)
396 #define ST_SH8(...) ST_V8(v8i16, __VA_ARGS__)
397 #define ST_SW8(...) ST_V8(v4i32, __VA_ARGS__)
399 /* Description : Store half word elements of vector with stride
400 * Arguments : Inputs - in source vector
401 * - pdst (destination pointer to store to)
403 * Details : Stores half word 'idx0' from 'in' to (pdst)
404 * Stores half word 'idx1' from 'in' to (pdst + stride)
405 * Similar for other elements
407 #define ST_H1(in, idx, pdst) \
410 out0_m = __msa_copy_u_h((v8i16) in, idx); \
411 SH(out0_m, (pdst)); \
413 #define ST_H2(in, idx0, idx1, pdst, stride) \
415 uint16_t out0_m, out1_m; \
416 out0_m = __msa_copy_u_h((v8i16) in, idx0); \
417 out1_m = __msa_copy_u_h((v8i16) in, idx1); \
418 SH(out0_m, (pdst)); \
419 SH(out1_m, (pdst) + stride); \
421 #define ST_H4(in, idx0, idx1, idx2, idx3, pdst, stride) \
423 uint16_t out0_m, out1_m, out2_m, out3_m; \
424 out0_m = __msa_copy_u_h((v8i16) in, idx0); \
425 out1_m = __msa_copy_u_h((v8i16) in, idx1); \
426 out2_m = __msa_copy_u_h((v8i16) in, idx2); \
427 out3_m = __msa_copy_u_h((v8i16) in, idx3); \
428 SH(out0_m, (pdst)); \
429 SH(out1_m, (pdst) + stride); \
430 SH(out2_m, (pdst) + 2 * stride); \
431 SH(out3_m, (pdst) + 3 * stride); \
433 #define ST_H8(in, idx0, idx1, idx2, idx3, idx4, idx5, \
434 idx6, idx7, pdst, stride) \
436 ST_H4(in, idx0, idx1, idx2, idx3, pdst, stride) \
437 ST_H4(in, idx4, idx5, idx6, idx7, (pdst) + 4*stride, stride) \
440 /* Description : Store word elements of vector with stride
441 * Arguments : Inputs - in source vector
442 * - pdst (destination pointer to store to)
444 * Details : Stores word 'idx0' from 'in' to (pdst)
445 * Stores word 'idx1' from 'in' to (pdst + stride)
446 * Similar for other elements
448 #define ST_W1(in, idx, pdst) \
451 out0_m = __msa_copy_u_w((v4i32) in, idx); \
452 SW(out0_m, (pdst)); \
454 #define ST_W2(in, idx0, idx1, pdst, stride) \
456 uint32_t out0_m, out1_m; \
457 out0_m = __msa_copy_u_w((v4i32) in, idx0); \
458 out1_m = __msa_copy_u_w((v4i32) in, idx1); \
459 SW(out0_m, (pdst)); \
460 SW(out1_m, (pdst) + stride); \
462 #define ST_W4(in, idx0, idx1, idx2, idx3, pdst, stride) \
464 uint32_t out0_m, out1_m, out2_m, out3_m; \
465 out0_m = __msa_copy_u_w((v4i32) in, idx0); \
466 out1_m = __msa_copy_u_w((v4i32) in, idx1); \
467 out2_m = __msa_copy_u_w((v4i32) in, idx2); \
468 out3_m = __msa_copy_u_w((v4i32) in, idx3); \
469 SW(out0_m, (pdst)); \
470 SW(out1_m, (pdst) + stride); \
471 SW(out2_m, (pdst) + 2*stride); \
472 SW(out3_m, (pdst) + 3*stride); \
474 #define ST_W8(in0, in1, idx0, idx1, idx2, idx3, \
475 idx4, idx5, idx6, idx7, pdst, stride) \
477 ST_W4(in0, idx0, idx1, idx2, idx3, pdst, stride) \
478 ST_W4(in1, idx4, idx5, idx6, idx7, pdst + 4*stride, stride) \
481 /* Description : Store double word elements of vector with stride
482 * Arguments : Inputs - in source vector
483 * - pdst (destination pointer to store to)
485 * Details : Stores double word 'idx0' from 'in' to (pdst)
486 * Stores double word 'idx1' from 'in' to (pdst + stride)
487 * Similar for other elements
489 #define ST_D1(in, idx, pdst) \
492 out0_m = __msa_copy_u_d((v2i64) in, idx); \
493 SD(out0_m, (pdst)); \
495 #define ST_D2(in, idx0, idx1, pdst, stride) \
497 uint64_t out0_m, out1_m; \
498 out0_m = __msa_copy_u_d((v2i64) in, idx0); \
499 out1_m = __msa_copy_u_d((v2i64) in, idx1); \
500 SD(out0_m, (pdst)); \
501 SD(out1_m, (pdst) + stride); \
503 #define ST_D4(in0, in1, idx0, idx1, idx2, idx3, pdst, stride) \
505 uint64_t out0_m, out1_m, out2_m, out3_m; \
506 out0_m = __msa_copy_u_d((v2i64) in0, idx0); \
507 out1_m = __msa_copy_u_d((v2i64) in0, idx1); \
508 out2_m = __msa_copy_u_d((v2i64) in1, idx2); \
509 out3_m = __msa_copy_u_d((v2i64) in1, idx3); \
510 SD(out0_m, (pdst)); \
511 SD(out1_m, (pdst) + stride); \
512 SD(out2_m, (pdst) + 2 * stride); \
513 SD(out3_m, (pdst) + 3 * stride); \
515 #define ST_D8(in0, in1, in2, in3, idx0, idx1, idx2, idx3, \
516 idx4, idx5, idx6, idx7, pdst, stride) \
518 ST_D4(in0, in1, idx0, idx1, idx2, idx3, pdst, stride) \
519 ST_D4(in2, in3, idx4, idx5, idx6, idx7, pdst + 4 * stride, stride) \
522 /* Description : Store as 12x8 byte block to destination memory from
524 Arguments : Inputs - in0, in1, in2, in3, in4, in5, in6, in7, pdst, stride
525 Details : Index 0 double word element from input vector 'in0' is copied
526 and stored to destination memory at (pblk_12x8_m) followed by
527 index 2 word element from same input vector 'in0' at
529 Similar to remaining lines
531 #define ST12x8_UB(in0, in1, in2, in3, in4, in5, in6, in7, pdst, stride) \
533 uint64_t out0_m, out1_m, out2_m, out3_m; \
534 uint64_t out4_m, out5_m, out6_m, out7_m; \
535 uint32_t out8_m, out9_m, out10_m, out11_m; \
536 uint32_t out12_m, out13_m, out14_m, out15_m; \
537 uint8_t *pblk_12x8_m = (uint8_t *) (pdst); \
539 out0_m = __msa_copy_u_d((v2i64) in0, 0); \
540 out1_m = __msa_copy_u_d((v2i64) in1, 0); \
541 out2_m = __msa_copy_u_d((v2i64) in2, 0); \
542 out3_m = __msa_copy_u_d((v2i64) in3, 0); \
543 out4_m = __msa_copy_u_d((v2i64) in4, 0); \
544 out5_m = __msa_copy_u_d((v2i64) in5, 0); \
545 out6_m = __msa_copy_u_d((v2i64) in6, 0); \
546 out7_m = __msa_copy_u_d((v2i64) in7, 0); \
548 out8_m = __msa_copy_u_w((v4i32) in0, 2); \
549 out9_m = __msa_copy_u_w((v4i32) in1, 2); \
550 out10_m = __msa_copy_u_w((v4i32) in2, 2); \
551 out11_m = __msa_copy_u_w((v4i32) in3, 2); \
552 out12_m = __msa_copy_u_w((v4i32) in4, 2); \
553 out13_m = __msa_copy_u_w((v4i32) in5, 2); \
554 out14_m = __msa_copy_u_w((v4i32) in6, 2); \
555 out15_m = __msa_copy_u_w((v4i32) in7, 2); \
557 SD(out0_m, pblk_12x8_m); \
558 SW(out8_m, pblk_12x8_m + 8); \
559 pblk_12x8_m += stride; \
560 SD(out1_m, pblk_12x8_m); \
561 SW(out9_m, pblk_12x8_m + 8); \
562 pblk_12x8_m += stride; \
563 SD(out2_m, pblk_12x8_m); \
564 SW(out10_m, pblk_12x8_m + 8); \
565 pblk_12x8_m += stride; \
566 SD(out3_m, pblk_12x8_m); \
567 SW(out11_m, pblk_12x8_m + 8); \
568 pblk_12x8_m += stride; \
569 SD(out4_m, pblk_12x8_m); \
570 SW(out12_m, pblk_12x8_m + 8); \
571 pblk_12x8_m += stride; \
572 SD(out5_m, pblk_12x8_m); \
573 SW(out13_m, pblk_12x8_m + 8); \
574 pblk_12x8_m += stride; \
575 SD(out6_m, pblk_12x8_m); \
576 SW(out14_m, pblk_12x8_m + 8); \
577 pblk_12x8_m += stride; \
578 SD(out7_m, pblk_12x8_m); \
579 SW(out15_m, pblk_12x8_m + 8); \
582 /* Description : average with rounding (in0 + in1 + 1) / 2.
583 Arguments : Inputs - in0, in1, in2, in3,
585 Return Type - as per RTYPE
586 Details : Each byte element from 'in0' vector is added with each byte
587 element from 'in1' vector. The addition of the elements plus 1
588 (for rounding) is done unsigned with full precision,
589 i.e. the result has one extra bit. Unsigned division by 2
590 (or logical shift right by one bit) is performed before writing
591 the result to vector 'out0'
592 Similar for the pair of 'in2' and 'in3'
594 #define AVER_UB2(RTYPE, in0, in1, in2, in3, out0, out1) \
596 out0 = (RTYPE) __msa_aver_u_b((v16u8) in0, (v16u8) in1); \
597 out1 = (RTYPE) __msa_aver_u_b((v16u8) in2, (v16u8) in3); \
599 #define AVER_UB2_UB(...) AVER_UB2(v16u8, __VA_ARGS__)
601 #define AVER_UB4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, \
602 out0, out1, out2, out3) \
604 AVER_UB2(RTYPE, in0, in1, in2, in3, out0, out1) \
605 AVER_UB2(RTYPE, in4, in5, in6, in7, out2, out3) \
607 #define AVER_UB4_UB(...) AVER_UB4(v16u8, __VA_ARGS__)
609 /* Description : Immediate number of columns to slide
610 Arguments : Inputs - s, d, slide_val
612 Return Type - as per RTYPE
613 Details : Byte elements from 'd' vector are slide into 's' by
614 number of elements specified by 'slide_val'
616 #define SLDI_B(RTYPE, d, s, slide_val, out) \
618 out = (RTYPE) __msa_sldi_b((v16i8) d, (v16i8) s, slide_val); \
621 #define SLDI_B2(RTYPE, d0, s0, d1, s1, slide_val, out0, out1) \
623 SLDI_B(RTYPE, d0, s0, slide_val, out0) \
624 SLDI_B(RTYPE, d1, s1, slide_val, out1) \
626 #define SLDI_B2_UB(...) SLDI_B2(v16u8, __VA_ARGS__)
627 #define SLDI_B2_SB(...) SLDI_B2(v16i8, __VA_ARGS__)
628 #define SLDI_B2_SH(...) SLDI_B2(v8i16, __VA_ARGS__)
629 #define SLDI_B2_SW(...) SLDI_B2(v4i32, __VA_ARGS__)
631 #define SLDI_B3(RTYPE, d0, s0, d1, s1, d2, s2, slide_val, \
634 SLDI_B2(RTYPE, d0, s0, d1, s1, slide_val, out0, out1) \
635 SLDI_B(RTYPE, d2, s2, slide_val, out2) \
637 #define SLDI_B3_UB(...) SLDI_B3(v16u8, __VA_ARGS__)
638 #define SLDI_B3_SB(...) SLDI_B3(v16i8, __VA_ARGS__)
639 #define SLDI_B3_UH(...) SLDI_B3(v8u16, __VA_ARGS__)
641 #define SLDI_B4(RTYPE, d0, s0, d1, s1, d2, s2, d3, s3, \
642 slide_val, out0, out1, out2, out3) \
644 SLDI_B2(RTYPE, d0, s0, d1, s1, slide_val, out0, out1) \
645 SLDI_B2(RTYPE, d2, s2, d3, s3, slide_val, out2, out3) \
647 #define SLDI_B4_UB(...) SLDI_B4(v16u8, __VA_ARGS__)
648 #define SLDI_B4_SB(...) SLDI_B4(v16i8, __VA_ARGS__)
649 #define SLDI_B4_SH(...) SLDI_B4(v8i16, __VA_ARGS__)
651 /* Description : Shuffle byte vector elements as per mask vector
652 Arguments : Inputs - in0, in1, in2, in3, mask0, mask1
654 Return Type - as per RTYPE
655 Details : Selective byte elements from in0 & in1 are copied to out0 as
656 per control vector mask0
657 Selective byte elements from in2 & in3 are copied to out1 as
658 per control vector mask1
660 #define VSHF_B2(RTYPE, in0, in1, in2, in3, mask0, mask1, out0, out1) \
662 out0 = (RTYPE) __msa_vshf_b((v16i8) mask0, (v16i8) in1, (v16i8) in0); \
663 out1 = (RTYPE) __msa_vshf_b((v16i8) mask1, (v16i8) in3, (v16i8) in2); \
665 #define VSHF_B2_UB(...) VSHF_B2(v16u8, __VA_ARGS__)
666 #define VSHF_B2_SB(...) VSHF_B2(v16i8, __VA_ARGS__)
667 #define VSHF_B2_UH(...) VSHF_B2(v8u16, __VA_ARGS__)
668 #define VSHF_B2_SH(...) VSHF_B2(v8i16, __VA_ARGS__)
670 #define VSHF_B3(RTYPE, in0, in1, in2, in3, in4, in5, mask0, mask1, mask2, \
673 VSHF_B2(RTYPE, in0, in1, in2, in3, mask0, mask1, out0, out1); \
674 out2 = (RTYPE) __msa_vshf_b((v16i8) mask2, (v16i8) in5, (v16i8) in4); \
676 #define VSHF_B3_SB(...) VSHF_B3(v16i8, __VA_ARGS__)
678 #define VSHF_B4(RTYPE, in0, in1, mask0, mask1, mask2, mask3, \
679 out0, out1, out2, out3) \
681 VSHF_B2(RTYPE, in0, in1, in0, in1, mask0, mask1, out0, out1); \
682 VSHF_B2(RTYPE, in0, in1, in0, in1, mask2, mask3, out2, out3); \
684 #define VSHF_B4_SB(...) VSHF_B4(v16i8, __VA_ARGS__)
685 #define VSHF_B4_SH(...) VSHF_B4(v8i16, __VA_ARGS__)
687 /* Description : Shuffle halfword vector elements as per mask vector
688 Arguments : Inputs - in0, in1, in2, in3, mask0, mask1
690 Return Type - as per RTYPE
691 Details : Selective halfword elements from in0 & in1 are copied to out0
692 as per control vector mask0
693 Selective halfword elements from in2 & in3 are copied to out1
694 as per control vector mask1
696 #define VSHF_H2(RTYPE, in0, in1, in2, in3, mask0, mask1, out0, out1) \
698 out0 = (RTYPE) __msa_vshf_h((v8i16) mask0, (v8i16) in1, (v8i16) in0); \
699 out1 = (RTYPE) __msa_vshf_h((v8i16) mask1, (v8i16) in3, (v8i16) in2); \
701 #define VSHF_H2_SH(...) VSHF_H2(v8i16, __VA_ARGS__)
703 #define VSHF_H3(RTYPE, in0, in1, in2, in3, in4, in5, mask0, mask1, mask2, \
706 VSHF_H2(RTYPE, in0, in1, in2, in3, mask0, mask1, out0, out1); \
707 out2 = (RTYPE) __msa_vshf_h((v8i16) mask2, (v8i16) in5, (v8i16) in4); \
709 #define VSHF_H3_SH(...) VSHF_H3(v8i16, __VA_ARGS__)
711 /* Description : Shuffle byte vector elements as per mask vector
712 Arguments : Inputs - in0, in1, in2, in3, mask0, mask1
714 Return Type - as per RTYPE
715 Details : Selective byte elements from in0 & in1 are copied to out0 as
716 per control vector mask0
717 Selective byte elements from in2 & in3 are copied to out1 as
718 per control vector mask1
720 #define VSHF_W2(RTYPE, in0, in1, in2, in3, mask0, mask1, out0, out1) \
722 out0 = (RTYPE) __msa_vshf_w((v4i32) mask0, (v4i32) in1, (v4i32) in0); \
723 out1 = (RTYPE) __msa_vshf_w((v4i32) mask1, (v4i32) in3, (v4i32) in2); \
725 #define VSHF_W2_SB(...) VSHF_W2(v16i8, __VA_ARGS__)
727 /* Description : Dot product of byte vector elements
728 Arguments : Inputs - mult0, mult1
731 Return Type - as per RTYPE
732 Details : Unsigned byte elements from mult0 are multiplied with
733 unsigned byte elements from cnst0 producing a result
734 twice the size of input i.e. unsigned halfword.
735 Then this multiplication results of adjacent odd-even elements
736 are added together and stored to the out vector
737 (2 unsigned halfword results)
739 #define DOTP_UB2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1) \
741 out0 = (RTYPE) __msa_dotp_u_h((v16u8) mult0, (v16u8) cnst0); \
742 out1 = (RTYPE) __msa_dotp_u_h((v16u8) mult1, (v16u8) cnst1); \
744 #define DOTP_UB2_UH(...) DOTP_UB2(v8u16, __VA_ARGS__)
746 #define DOTP_UB4(RTYPE, mult0, mult1, mult2, mult3, \
747 cnst0, cnst1, cnst2, cnst3, \
748 out0, out1, out2, out3) \
750 DOTP_UB2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1); \
751 DOTP_UB2(RTYPE, mult2, mult3, cnst2, cnst3, out2, out3); \
753 #define DOTP_UB4_UH(...) DOTP_UB4(v8u16, __VA_ARGS__)
755 /* Description : Dot product of byte vector elements
756 Arguments : Inputs - mult0, mult1
759 Return Type - as per RTYPE
760 Details : Signed byte elements from mult0 are multiplied with
761 signed byte elements from cnst0 producing a result
762 twice the size of input i.e. signed halfword.
763 Then this multiplication results of adjacent odd-even elements
764 are added together and stored to the out vector
765 (2 signed halfword results)
767 #define DOTP_SB2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1) \
769 out0 = (RTYPE) __msa_dotp_s_h((v16i8) mult0, (v16i8) cnst0); \
770 out1 = (RTYPE) __msa_dotp_s_h((v16i8) mult1, (v16i8) cnst1); \
772 #define DOTP_SB2_SH(...) DOTP_SB2(v8i16, __VA_ARGS__)
774 #define DOTP_SB3(RTYPE, mult0, mult1, mult2, cnst0, cnst1, cnst2, \
777 DOTP_SB2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1); \
778 out2 = (RTYPE) __msa_dotp_s_h((v16i8) mult2, (v16i8) cnst2); \
780 #define DOTP_SB3_SH(...) DOTP_SB3(v8i16, __VA_ARGS__)
782 #define DOTP_SB4(RTYPE, mult0, mult1, mult2, mult3, \
783 cnst0, cnst1, cnst2, cnst3, out0, out1, out2, out3) \
785 DOTP_SB2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1); \
786 DOTP_SB2(RTYPE, mult2, mult3, cnst2, cnst3, out2, out3); \
788 #define DOTP_SB4_SH(...) DOTP_SB4(v8i16, __VA_ARGS__)
790 /* Description : Dot product of halfword vector elements
791 Arguments : Inputs - mult0, mult1
794 Return Type - as per RTYPE
795 Details : Signed halfword elements from mult0 are multiplied with
796 signed halfword elements from cnst0 producing a result
797 twice the size of input i.e. signed word.
798 Then this multiplication results of adjacent odd-even elements
799 are added together and stored to the out vector
800 (2 signed word results)
802 #define DOTP_SH2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1) \
804 out0 = (RTYPE) __msa_dotp_s_w((v8i16) mult0, (v8i16) cnst0); \
805 out1 = (RTYPE) __msa_dotp_s_w((v8i16) mult1, (v8i16) cnst1); \
807 #define DOTP_SH2_SW(...) DOTP_SH2(v4i32, __VA_ARGS__)
809 #define DOTP_SH4(RTYPE, mult0, mult1, mult2, mult3, \
810 cnst0, cnst1, cnst2, cnst3, \
811 out0, out1, out2, out3) \
813 DOTP_SH2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1); \
814 DOTP_SH2(RTYPE, mult2, mult3, cnst2, cnst3, out2, out3); \
816 #define DOTP_SH4_SW(...) DOTP_SH4(v4i32, __VA_ARGS__)
818 /* Description : Dot product & addition of byte vector elements
819 Arguments : Inputs - mult0, mult1
822 Return Type - as per RTYPE
823 Details : Signed byte elements from mult0 are multiplied with
824 signed byte elements from cnst0 producing a result
825 twice the size of input i.e. signed halfword.
826 Then this multiplication results of adjacent odd-even elements
827 are added to the out vector
828 (2 signed halfword results)
830 #define DPADD_SB2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1) \
832 out0 = (RTYPE) __msa_dpadd_s_h((v8i16) out0, \
833 (v16i8) mult0, (v16i8) cnst0); \
834 out1 = (RTYPE) __msa_dpadd_s_h((v8i16) out1, \
835 (v16i8) mult1, (v16i8) cnst1); \
837 #define DPADD_SB2_SH(...) DPADD_SB2(v8i16, __VA_ARGS__)
839 #define DPADD_SB4(RTYPE, mult0, mult1, mult2, mult3, \
840 cnst0, cnst1, cnst2, cnst3, out0, out1, out2, out3) \
842 DPADD_SB2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1); \
843 DPADD_SB2(RTYPE, mult2, mult3, cnst2, cnst3, out2, out3); \
845 #define DPADD_SB4_SH(...) DPADD_SB4(v8i16, __VA_ARGS__)
847 /* Description : Dot product & addition of byte vector elements
848 Arguments : Inputs - mult0, mult1
851 Return Type - as per RTYPE
852 Details : Unsigned byte elements from mult0 are multiplied with
853 unsigned byte elements from cnst0 producing a result
854 twice the size of input i.e. unsigned halfword.
855 Then this multiplication results of adjacent odd-even elements
856 are added to the out vector
857 (2 unsigned halfword results)
859 #define DPADD_UB2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1) \
861 out0 = (RTYPE) __msa_dpadd_u_h((v8u16) out0, \
862 (v16u8) mult0, (v16u8) cnst0); \
863 out1 = (RTYPE) __msa_dpadd_u_h((v8u16) out1, \
864 (v16u8) mult1, (v16u8) cnst1); \
866 #define DPADD_UB2_UH(...) DPADD_UB2(v8u16, __VA_ARGS__)
868 /* Description : Dot product & addition of halfword vector elements
869 Arguments : Inputs - mult0, mult1
872 Return Type - as per RTYPE
873 Details : Signed halfword elements from mult0 are multiplied with
874 signed halfword elements from cnst0 producing a result
875 twice the size of input i.e. signed word.
876 Then this multiplication results of adjacent odd-even elements
877 are added to the out vector
878 (2 signed word results)
880 #define DPADD_SH2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1) \
882 out0 = (RTYPE) __msa_dpadd_s_w((v4i32) out0, \
883 (v8i16) mult0, (v8i16) cnst0); \
884 out1 = (RTYPE) __msa_dpadd_s_w((v4i32) out1, \
885 (v8i16) mult1, (v8i16) cnst1); \
887 #define DPADD_SH2_SW(...) DPADD_SH2(v4i32, __VA_ARGS__)
889 #define DPADD_SH4(RTYPE, mult0, mult1, mult2, mult3, \
890 cnst0, cnst1, cnst2, cnst3, out0, out1, out2, out3) \
892 DPADD_SH2(RTYPE, mult0, mult1, cnst0, cnst1, out0, out1); \
893 DPADD_SH2(RTYPE, mult2, mult3, cnst2, cnst3, out2, out3); \
895 #define DPADD_SH4_SW(...) DPADD_SH4(v4i32, __VA_ARGS__)
897 /* Description : Minimum values between unsigned elements of
898 either vector are copied to the output vector
899 Arguments : Inputs - in0, in1, min_vec
900 Outputs - in0, in1, (in place)
901 Return Type - as per RTYPE
902 Details : Minimum of unsigned halfword element values from 'in0' and
903 'min_value' are written to output vector 'in0'
905 #define MIN_UH2(RTYPE, in0, in1, min_vec) \
907 in0 = (RTYPE) __msa_min_u_h((v8u16) in0, min_vec); \
908 in1 = (RTYPE) __msa_min_u_h((v8u16) in1, min_vec); \
910 #define MIN_UH2_UH(...) MIN_UH2(v8u16, __VA_ARGS__)
912 #define MIN_UH4(RTYPE, in0, in1, in2, in3, min_vec) \
914 MIN_UH2(RTYPE, in0, in1, min_vec); \
915 MIN_UH2(RTYPE, in2, in3, min_vec); \
917 #define MIN_UH4_UH(...) MIN_UH4(v8u16, __VA_ARGS__)
919 /* Description : Clips all halfword elements of input vector between min & max
920 out = ((in) < (min)) ? (min) : (((in) > (max)) ? (max) : (in))
921 Arguments : Inputs - in (input vector)
922 - min (min threshold)
923 - max (max threshold)
924 Outputs - in (output vector with clipped elements)
925 Return Type - signed halfword
927 #define CLIP_SH(in, min, max) \
929 in = __msa_max_s_h((v8i16) min, (v8i16) in); \
930 in = __msa_min_s_h((v8i16) max, (v8i16) in); \
933 /* Description : Clips all signed halfword elements of input vector
935 Arguments : Inputs - in (input vector)
936 Outputs - in (output vector with clipped elements)
937 Return Type - signed halfwords
939 #define CLIP_SH_0_255(in) \
941 in = __msa_maxi_s_h((v8i16) in, 0); \
942 in = (v8i16) __msa_sat_u_h((v8u16) in, 7); \
945 #define CLIP_SH2_0_255(in0, in1) \
947 CLIP_SH_0_255(in0); \
948 CLIP_SH_0_255(in1); \
951 #define CLIP_SH4_0_255(in0, in1, in2, in3) \
953 CLIP_SH2_0_255(in0, in1); \
954 CLIP_SH2_0_255(in2, in3); \
957 #define CLIP_SH8_0_255(in0, in1, in2, in3, \
958 in4, in5, in6, in7) \
960 CLIP_SH4_0_255(in0, in1, in2, in3); \
961 CLIP_SH4_0_255(in4, in5, in6, in7); \
964 /* Description : Clips all signed word elements of input vector
966 Arguments : Inputs - in (input vector)
967 Outputs - in (output vector with clipped elements)
968 Return Type - signed word
970 #define CLIP_SW_0_255(in) \
972 in = __msa_maxi_s_w((v4i32) in, 0); \
973 in = (v4i32) __msa_sat_u_w((v4u32) in, 7); \
976 #define CLIP_SW2_0_255(in0, in1) \
978 CLIP_SW_0_255(in0); \
979 CLIP_SW_0_255(in1); \
982 #define CLIP_SW4_0_255(in0, in1, in2, in3) \
984 CLIP_SW2_0_255(in0, in1); \
985 CLIP_SW2_0_255(in2, in3); \
988 #define CLIP_SW8_0_255(in0, in1, in2, in3, \
989 in4, in5, in6, in7) \
991 CLIP_SW4_0_255(in0, in1, in2, in3); \
992 CLIP_SW4_0_255(in4, in5, in6, in7); \
995 /* Description : Addition of 4 signed word elements
996 4 signed word elements of input vector are added together and
997 resulted integer sum is returned
998 Arguments : Inputs - in (signed word vector)
999 Outputs - sum_m (i32 sum)
1000 Return Type - signed word
1002 #define HADD_SW_S32(in) \
1004 v2i64 res0_m, res1_m; \
1007 res0_m = __msa_hadd_s_d((v4i32) in, (v4i32) in); \
1008 res1_m = __msa_splati_d(res0_m, 1); \
1010 sum_m = __msa_copy_s_w((v4i32) res0_m, 0); \
1014 /* Description : Addition of 8 unsigned halfword elements
1015 8 unsigned halfword elements of input vector are added
1016 together and resulted integer sum is returned
1017 Arguments : Inputs - in (unsigned halfword vector)
1018 Outputs - sum_m (u32 sum)
1019 Return Type - unsigned word
1021 #define HADD_UH_U32(in) \
1024 v2u64 res0_m, res1_m; \
1027 res_m = __msa_hadd_u_w((v8u16) in, (v8u16) in); \
1028 res0_m = __msa_hadd_u_d(res_m, res_m); \
1029 res1_m = (v2u64) __msa_splati_d((v2i64) res0_m, 1); \
1031 sum_m = __msa_copy_u_w((v4i32) res0_m, 0); \
1035 /* Description : Horizontal addition of signed byte vector elements
1036 Arguments : Inputs - in0, in1
1037 Outputs - out0, out1
1038 Return Type - as per RTYPE
1039 Details : Each signed odd byte element from 'in0' is added to
1040 even signed byte element from 'in0' (pairwise) and the
1041 halfword result is stored in 'out0'
1043 #define HADD_SB2(RTYPE, in0, in1, out0, out1) \
1045 out0 = (RTYPE) __msa_hadd_s_h((v16i8) in0, (v16i8) in0); \
1046 out1 = (RTYPE) __msa_hadd_s_h((v16i8) in1, (v16i8) in1); \
1048 #define HADD_SB2_SH(...) HADD_SB2(v8i16, __VA_ARGS__)
1050 #define HADD_SB4(RTYPE, in0, in1, in2, in3, out0, out1, out2, out3) \
1052 HADD_SB2(RTYPE, in0, in1, out0, out1); \
1053 HADD_SB2(RTYPE, in2, in3, out2, out3); \
1055 #define HADD_SB4_UH(...) HADD_SB4(v8u16, __VA_ARGS__)
1056 #define HADD_SB4_SH(...) HADD_SB4(v8i16, __VA_ARGS__)
1058 /* Description : Horizontal addition of unsigned byte vector elements
1059 Arguments : Inputs - in0, in1
1060 Outputs - out0, out1
1061 Return Type - as per RTYPE
1062 Details : Each unsigned odd byte element from 'in0' is added to
1063 even unsigned byte element from 'in0' (pairwise) and the
1064 halfword result is stored in 'out0'
1066 #define HADD_UB2(RTYPE, in0, in1, out0, out1) \
1068 out0 = (RTYPE) __msa_hadd_u_h((v16u8) in0, (v16u8) in0); \
1069 out1 = (RTYPE) __msa_hadd_u_h((v16u8) in1, (v16u8) in1); \
1071 #define HADD_UB2_UH(...) HADD_UB2(v8u16, __VA_ARGS__)
1073 #define HADD_UB3(RTYPE, in0, in1, in2, out0, out1, out2) \
1075 HADD_UB2(RTYPE, in0, in1, out0, out1); \
1076 out2 = (RTYPE) __msa_hadd_u_h((v16u8) in2, (v16u8) in2); \
1078 #define HADD_UB3_UH(...) HADD_UB3(v8u16, __VA_ARGS__)
1080 #define HADD_UB4(RTYPE, in0, in1, in2, in3, out0, out1, out2, out3) \
1082 HADD_UB2(RTYPE, in0, in1, out0, out1); \
1083 HADD_UB2(RTYPE, in2, in3, out2, out3); \
1085 #define HADD_UB4_UB(...) HADD_UB4(v16u8, __VA_ARGS__)
1086 #define HADD_UB4_UH(...) HADD_UB4(v8u16, __VA_ARGS__)
1087 #define HADD_UB4_SH(...) HADD_UB4(v8i16, __VA_ARGS__)
1089 /* Description : Horizontal subtraction of unsigned byte vector elements
1090 Arguments : Inputs - in0, in1
1091 Outputs - out0, out1
1092 Return Type - as per RTYPE
1093 Details : Each unsigned odd byte element from 'in0' is subtracted from
1094 even unsigned byte element from 'in0' (pairwise) and the
1095 halfword result is stored in 'out0'
1097 #define HSUB_UB2(RTYPE, in0, in1, out0, out1) \
1099 out0 = (RTYPE) __msa_hsub_u_h((v16u8) in0, (v16u8) in0); \
1100 out1 = (RTYPE) __msa_hsub_u_h((v16u8) in1, (v16u8) in1); \
1102 #define HSUB_UB2_UH(...) HSUB_UB2(v8u16, __VA_ARGS__)
1103 #define HSUB_UB2_SH(...) HSUB_UB2(v8i16, __VA_ARGS__)
1105 #define HSUB_UB4(RTYPE, in0, in1, in2, in3, out0, out1, out2, out3) \
1107 HSUB_UB2(RTYPE, in0, in1, out0, out1); \
1108 HSUB_UB2(RTYPE, in2, in3, out2, out3); \
1110 #define HSUB_UB4_UH(...) HSUB_UB4(v8u16, __VA_ARGS__)
1111 #define HSUB_UB4_SH(...) HSUB_UB4(v8i16, __VA_ARGS__)
1113 /* Description : SAD (Sum of Absolute Difference)
1114 Arguments : Inputs - in0, in1, ref0, ref1 (unsigned byte src & ref)
1115 Outputs - sad_m (halfword vector with sad)
1116 Return Type - unsigned halfword
1117 Details : Absolute difference of all the byte elements from 'in0' with
1118 'ref0' is calculated and preserved in 'diff0'. From the 16
1119 unsigned absolute diff values, even-odd pairs are added
1120 together to generate 8 halfword results.
1123 #define SAD_UB2_UH(in0, in1, ref0, ref1) \
1125 v8u16 sad_m = { 0 }; \
1126 sad_m += __builtin_msa2_sad_adj2_u_w2x_b((v16u8) in0, (v16u8) ref0); \
1127 sad_m += __builtin_msa2_sad_adj2_u_w2x_b((v16u8) in1, (v16u8) ref1); \
1131 #define SAD_UB2_UH(in0, in1, ref0, ref1) \
1133 v16u8 diff0_m, diff1_m; \
1134 v8u16 sad_m = { 0 }; \
1136 diff0_m = __msa_asub_u_b((v16u8) in0, (v16u8) ref0); \
1137 diff1_m = __msa_asub_u_b((v16u8) in1, (v16u8) ref1); \
1139 sad_m += __msa_hadd_u_h((v16u8) diff0_m, (v16u8) diff0_m); \
1140 sad_m += __msa_hadd_u_h((v16u8) diff1_m, (v16u8) diff1_m); \
1144 #endif // #if HAVE_MSA2
1146 /* Description : Insert specified word elements from input vectors to 1
1148 Arguments : Inputs - in0, in1, in2, in3 (4 input vectors)
1149 Outputs - out (output vector)
1150 Return Type - as per RTYPE
1152 #define INSERT_W2(RTYPE, in0, in1, out) \
1154 out = (RTYPE) __msa_insert_w((v4i32) out, 0, in0); \
1155 out = (RTYPE) __msa_insert_w((v4i32) out, 1, in1); \
1157 #define INSERT_W2_UB(...) INSERT_W2(v16u8, __VA_ARGS__)
1158 #define INSERT_W2_SB(...) INSERT_W2(v16i8, __VA_ARGS__)
1160 #define INSERT_W4(RTYPE, in0, in1, in2, in3, out) \
1162 out = (RTYPE) __msa_insert_w((v4i32) out, 0, in0); \
1163 out = (RTYPE) __msa_insert_w((v4i32) out, 1, in1); \
1164 out = (RTYPE) __msa_insert_w((v4i32) out, 2, in2); \
1165 out = (RTYPE) __msa_insert_w((v4i32) out, 3, in3); \
1167 #define INSERT_W4_UB(...) INSERT_W4(v16u8, __VA_ARGS__)
1168 #define INSERT_W4_SB(...) INSERT_W4(v16i8, __VA_ARGS__)
1169 #define INSERT_W4_SH(...) INSERT_W4(v8i16, __VA_ARGS__)
1170 #define INSERT_W4_SW(...) INSERT_W4(v4i32, __VA_ARGS__)
1172 /* Description : Insert specified double word elements from input vectors to 1
1174 Arguments : Inputs - in0, in1 (2 input vectors)
1175 Outputs - out (output vector)
1176 Return Type - as per RTYPE
1178 #define INSERT_D2(RTYPE, in0, in1, out) \
1180 out = (RTYPE) __msa_insert_d((v2i64) out, 0, in0); \
1181 out = (RTYPE) __msa_insert_d((v2i64) out, 1, in1); \
1183 #define INSERT_D2_UB(...) INSERT_D2(v16u8, __VA_ARGS__)
1184 #define INSERT_D2_SB(...) INSERT_D2(v16i8, __VA_ARGS__)
1185 #define INSERT_D2_SH(...) INSERT_D2(v8i16, __VA_ARGS__)
1186 #define INSERT_D2_SD(...) INSERT_D2(v2i64, __VA_ARGS__)
1188 /* Description : Interleave even byte elements from vectors
1189 Arguments : Inputs - in0, in1, in2, in3
1190 Outputs - out0, out1
1191 Return Type - as per RTYPE
1192 Details : Even byte elements of 'in0' and even byte
1193 elements of 'in1' are interleaved and copied to 'out0'
1194 Even byte elements of 'in2' and even byte
1195 elements of 'in3' are interleaved and copied to 'out1'
1197 #define ILVEV_B2(RTYPE, in0, in1, in2, in3, out0, out1) \
1199 out0 = (RTYPE) __msa_ilvev_b((v16i8) in1, (v16i8) in0); \
1200 out1 = (RTYPE) __msa_ilvev_b((v16i8) in3, (v16i8) in2); \
1202 #define ILVEV_B2_UB(...) ILVEV_B2(v16u8, __VA_ARGS__)
1203 #define ILVEV_B2_SB(...) ILVEV_B2(v16i8, __VA_ARGS__)
1204 #define ILVEV_B2_SH(...) ILVEV_B2(v8i16, __VA_ARGS__)
1205 #define ILVEV_B2_SD(...) ILVEV_B2(v2i64, __VA_ARGS__)
1207 /* Description : Interleave even halfword elements from vectors
1208 Arguments : Inputs - in0, in1, in2, in3
1209 Outputs - out0, out1
1210 Return Type - as per RTYPE
1211 Details : Even halfword elements of 'in0' and even halfword
1212 elements of 'in1' are interleaved and copied to 'out0'
1213 Even halfword elements of 'in2' and even halfword
1214 elements of 'in3' are interleaved and copied to 'out1'
1216 #define ILVEV_H2(RTYPE, in0, in1, in2, in3, out0, out1) \
1218 out0 = (RTYPE) __msa_ilvev_h((v8i16) in1, (v8i16) in0); \
1219 out1 = (RTYPE) __msa_ilvev_h((v8i16) in3, (v8i16) in2); \
1221 #define ILVEV_H2_UB(...) ILVEV_H2(v16u8, __VA_ARGS__)
1222 #define ILVEV_H2_SH(...) ILVEV_H2(v8i16, __VA_ARGS__)
1223 #define ILVEV_H2_SW(...) ILVEV_H2(v4i32, __VA_ARGS__)
1225 /* Description : Interleave even word elements from vectors
1226 Arguments : Inputs - in0, in1, in2, in3
1227 Outputs - out0, out1
1228 Return Type - as per RTYPE
1229 Details : Even word elements of 'in0' and even word
1230 elements of 'in1' are interleaved and copied to 'out0'
1231 Even word elements of 'in2' and even word
1232 elements of 'in3' are interleaved and copied to 'out1'
1234 #define ILVEV_W2(RTYPE, in0, in1, in2, in3, out0, out1) \
1236 out0 = (RTYPE) __msa_ilvev_w((v4i32) in1, (v4i32) in0); \
1237 out1 = (RTYPE) __msa_ilvev_w((v4i32) in3, (v4i32) in2); \
1239 #define ILVEV_W2_UB(...) ILVEV_W2(v16u8, __VA_ARGS__)
1240 #define ILVEV_W2_SB(...) ILVEV_W2(v16i8, __VA_ARGS__)
1241 #define ILVEV_W2_UH(...) ILVEV_W2(v8u16, __VA_ARGS__)
1242 #define ILVEV_W2_SD(...) ILVEV_W2(v2i64, __VA_ARGS__)
1244 /* Description : Interleave even double word elements from vectors
1245 Arguments : Inputs - in0, in1, in2, in3
1246 Outputs - out0, out1
1247 Return Type - as per RTYPE
1248 Details : Even double word elements of 'in0' and even double word
1249 elements of 'in1' are interleaved and copied to 'out0'
1250 Even double word elements of 'in2' and even double word
1251 elements of 'in3' are interleaved and copied to 'out1'
1253 #define ILVEV_D2(RTYPE, in0, in1, in2, in3, out0, out1) \
1255 out0 = (RTYPE) __msa_ilvev_d((v2i64) in1, (v2i64) in0); \
1256 out1 = (RTYPE) __msa_ilvev_d((v2i64) in3, (v2i64) in2); \
1258 #define ILVEV_D2_UB(...) ILVEV_D2(v16u8, __VA_ARGS__)
1259 #define ILVEV_D2_SB(...) ILVEV_D2(v16i8, __VA_ARGS__)
1260 #define ILVEV_D2_SW(...) ILVEV_D2(v4i32, __VA_ARGS__)
1262 /* Description : Interleave left half of byte elements from vectors
1263 Arguments : Inputs - in0, in1, in2, in3
1264 Outputs - out0, out1
1265 Return Type - as per RTYPE
1266 Details : Left half of byte elements of in0 and left half of byte
1267 elements of in1 are interleaved and copied to out0.
1268 Left half of byte elements of in2 and left half of byte
1269 elements of in3 are interleaved and copied to out1.
1271 #define ILVL_B2(RTYPE, in0, in1, in2, in3, out0, out1) \
1273 out0 = (RTYPE) __msa_ilvl_b((v16i8) in0, (v16i8) in1); \
1274 out1 = (RTYPE) __msa_ilvl_b((v16i8) in2, (v16i8) in3); \
1276 #define ILVL_B2_UB(...) ILVL_B2(v16u8, __VA_ARGS__)
1277 #define ILVL_B2_SB(...) ILVL_B2(v16i8, __VA_ARGS__)
1278 #define ILVL_B2_UH(...) ILVL_B2(v8u16, __VA_ARGS__)
1279 #define ILVL_B2_SH(...) ILVL_B2(v8i16, __VA_ARGS__)
1281 #define ILVL_B4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, \
1282 out0, out1, out2, out3) \
1284 ILVL_B2(RTYPE, in0, in1, in2, in3, out0, out1); \
1285 ILVL_B2(RTYPE, in4, in5, in6, in7, out2, out3); \
1287 #define ILVL_B4_UB(...) ILVL_B4(v16u8, __VA_ARGS__)
1288 #define ILVL_B4_SB(...) ILVL_B4(v16i8, __VA_ARGS__)
1289 #define ILVL_B4_UH(...) ILVL_B4(v8u16, __VA_ARGS__)
1290 #define ILVL_B4_SH(...) ILVL_B4(v8i16, __VA_ARGS__)
1292 /* Description : Interleave left half of halfword elements from vectors
1293 Arguments : Inputs - in0, in1, in2, in3
1294 Outputs - out0, out1
1295 Return Type - as per RTYPE
1296 Details : Left half of halfword elements of in0 and left half of halfword
1297 elements of in1 are interleaved and copied to out0.
1298 Left half of halfword elements of in2 and left half of halfword
1299 elements of in3 are interleaved and copied to out1.
1301 #define ILVL_H2(RTYPE, in0, in1, in2, in3, out0, out1) \
1303 out0 = (RTYPE) __msa_ilvl_h((v8i16) in0, (v8i16) in1); \
1304 out1 = (RTYPE) __msa_ilvl_h((v8i16) in2, (v8i16) in3); \
1306 #define ILVL_H2_SH(...) ILVL_H2(v8i16, __VA_ARGS__)
1307 #define ILVL_H2_SW(...) ILVL_H2(v4i32, __VA_ARGS__)
1309 #define ILVL_H4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, \
1310 out0, out1, out2, out3) \
1312 ILVL_H2(RTYPE, in0, in1, in2, in3, out0, out1); \
1313 ILVL_H2(RTYPE, in4, in5, in6, in7, out2, out3); \
1315 #define ILVL_H4_SH(...) ILVL_H4(v8i16, __VA_ARGS__)
1316 #define ILVL_H4_SW(...) ILVL_H4(v4i32, __VA_ARGS__)
1318 /* Description : Interleave left half of word elements from vectors
1319 Arguments : Inputs - in0, in1, in2, in3
1320 Outputs - out0, out1
1321 Return Type - as per RTYPE
1322 Details : Left half of word elements of in0 and left half of word
1323 elements of in1 are interleaved and copied to out0.
1324 Left half of word elements of in2 and left half of word
1325 elements of in3 are interleaved and copied to out1.
1327 #define ILVL_W2(RTYPE, in0, in1, in2, in3, out0, out1) \
1329 out0 = (RTYPE) __msa_ilvl_w((v4i32) in0, (v4i32) in1); \
1330 out1 = (RTYPE) __msa_ilvl_w((v4i32) in2, (v4i32) in3); \
1332 #define ILVL_W2_UB(...) ILVL_W2(v16u8, __VA_ARGS__)
1333 #define ILVL_W2_SB(...) ILVL_W2(v16i8, __VA_ARGS__)
1334 #define ILVL_W2_SH(...) ILVL_W2(v8i16, __VA_ARGS__)
1336 /* Description : Interleave right half of byte elements from vectors
1337 Arguments : Inputs - in0, in1, in2, in3, in4, in5, in6, in7
1338 Outputs - out0, out1, out2, out3
1339 Return Type - as per RTYPE
1340 Details : Right half of byte elements of in0 and right half of byte
1341 elements of in1 are interleaved and copied to out0.
1342 Right half of byte elements of in2 and right half of byte
1343 elements of in3 are interleaved and copied to out1.
1344 Similar for other pairs
1346 #define ILVR_B2(RTYPE, in0, in1, in2, in3, out0, out1) \
1348 out0 = (RTYPE) __msa_ilvr_b((v16i8) in0, (v16i8) in1); \
1349 out1 = (RTYPE) __msa_ilvr_b((v16i8) in2, (v16i8) in3); \
1351 #define ILVR_B2_UB(...) ILVR_B2(v16u8, __VA_ARGS__)
1352 #define ILVR_B2_SB(...) ILVR_B2(v16i8, __VA_ARGS__)
1353 #define ILVR_B2_UH(...) ILVR_B2(v8u16, __VA_ARGS__)
1354 #define ILVR_B2_SH(...) ILVR_B2(v8i16, __VA_ARGS__)
1355 #define ILVR_B2_SW(...) ILVR_B2(v4i32, __VA_ARGS__)
1357 #define ILVR_B3(RTYPE, in0, in1, in2, in3, in4, in5, out0, out1, out2) \
1359 ILVR_B2(RTYPE, in0, in1, in2, in3, out0, out1); \
1360 out2 = (RTYPE) __msa_ilvr_b((v16i8) in4, (v16i8) in5); \
1362 #define ILVR_B3_UB(...) ILVR_B3(v16u8, __VA_ARGS__)
1363 #define ILVR_B3_SB(...) ILVR_B3(v16i8, __VA_ARGS__)
1364 #define ILVR_B3_UH(...) ILVR_B3(v8u16, __VA_ARGS__)
1365 #define ILVR_B3_SH(...) ILVR_B3(v8i16, __VA_ARGS__)
1367 #define ILVR_B4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, \
1368 out0, out1, out2, out3) \
1370 ILVR_B2(RTYPE, in0, in1, in2, in3, out0, out1); \
1371 ILVR_B2(RTYPE, in4, in5, in6, in7, out2, out3); \
1373 #define ILVR_B4_UB(...) ILVR_B4(v16u8, __VA_ARGS__)
1374 #define ILVR_B4_SB(...) ILVR_B4(v16i8, __VA_ARGS__)
1375 #define ILVR_B4_UH(...) ILVR_B4(v8u16, __VA_ARGS__)
1376 #define ILVR_B4_SH(...) ILVR_B4(v8i16, __VA_ARGS__)
1377 #define ILVR_B4_SW(...) ILVR_B4(v4i32, __VA_ARGS__)
1379 #define ILVR_B8(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, \
1380 in8, in9, in10, in11, in12, in13, in14, in15, \
1381 out0, out1, out2, out3, out4, out5, out6, out7) \
1383 ILVR_B4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, \
1384 out0, out1, out2, out3); \
1385 ILVR_B4(RTYPE, in8, in9, in10, in11, in12, in13, in14, in15, \
1386 out4, out5, out6, out7); \
1388 #define ILVR_B8_UH(...) ILVR_B8(v8u16, __VA_ARGS__)
1389 #define ILVR_B8_SW(...) ILVR_B8(v4i32, __VA_ARGS__)
1391 /* Description : Interleave right half of halfword elements from vectors
1392 Arguments : Inputs - in0, in1, in2, in3, in4, in5, in6, in7
1393 Outputs - out0, out1, out2, out3
1394 Return Type - as per RTYPE
1395 Details : Right half of halfword elements of in0 and right half of
1396 halfword elements of in1 are interleaved and copied to out0.
1397 Right half of halfword elements of in2 and right half of
1398 halfword elements of in3 are interleaved and copied to out1.
1399 Similar for other pairs
1401 #define ILVR_H2(RTYPE, in0, in1, in2, in3, out0, out1) \
1403 out0 = (RTYPE) __msa_ilvr_h((v8i16) in0, (v8i16) in1); \
1404 out1 = (RTYPE) __msa_ilvr_h((v8i16) in2, (v8i16) in3); \
1406 #define ILVR_H2_SH(...) ILVR_H2(v8i16, __VA_ARGS__)
1407 #define ILVR_H2_SW(...) ILVR_H2(v4i32, __VA_ARGS__)
1409 #define ILVR_H3(RTYPE, in0, in1, in2, in3, in4, in5, out0, out1, out2) \
1411 ILVR_H2(RTYPE, in0, in1, in2, in3, out0, out1); \
1412 out2 = (RTYPE) __msa_ilvr_h((v8i16) in4, (v8i16) in5); \
1414 #define ILVR_H3_SH(...) ILVR_H3(v8i16, __VA_ARGS__)
1416 #define ILVR_H4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, \
1417 out0, out1, out2, out3) \
1419 ILVR_H2(RTYPE, in0, in1, in2, in3, out0, out1); \
1420 ILVR_H2(RTYPE, in4, in5, in6, in7, out2, out3); \
1422 #define ILVR_H4_SH(...) ILVR_H4(v8i16, __VA_ARGS__)
1423 #define ILVR_H4_SW(...) ILVR_H4(v4i32, __VA_ARGS__)
1425 #define ILVR_W2(RTYPE, in0, in1, in2, in3, out0, out1) \
1427 out0 = (RTYPE) __msa_ilvr_w((v4i32) in0, (v4i32) in1); \
1428 out1 = (RTYPE) __msa_ilvr_w((v4i32) in2, (v4i32) in3); \
1430 #define ILVR_W2_UB(...) ILVR_W2(v16u8, __VA_ARGS__)
1431 #define ILVR_W2_SB(...) ILVR_W2(v16i8, __VA_ARGS__)
1432 #define ILVR_W2_SH(...) ILVR_W2(v8i16, __VA_ARGS__)
1434 #define ILVR_W4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, \
1435 out0, out1, out2, out3) \
1437 ILVR_W2(RTYPE, in0, in1, in2, in3, out0, out1); \
1438 ILVR_W2(RTYPE, in4, in5, in6, in7, out2, out3); \
1440 #define ILVR_W4_SB(...) ILVR_W4(v16i8, __VA_ARGS__)
1441 #define ILVR_W4_UB(...) ILVR_W4(v16u8, __VA_ARGS__)
1443 /* Description : Interleave right half of double word elements from vectors
1444 Arguments : Inputs - in0, in1, in2, in3, in4, in5, in6, in7
1445 Outputs - out0, out1, out2, out3
1446 Return Type - as per RTYPE
1447 Details : Right half of double word elements of in0 and right half of
1448 double word elements of in1 are interleaved and copied to out0.
1449 Right half of double word elements of in2 and right half of
1450 double word elements of in3 are interleaved and copied to out1.
1452 #define ILVR_D2(RTYPE, in0, in1, in2, in3, out0, out1) \
1454 out0 = (RTYPE) __msa_ilvr_d((v2i64) in0, (v2i64) in1); \
1455 out1 = (RTYPE) __msa_ilvr_d((v2i64) in2, (v2i64) in3); \
1457 #define ILVR_D2_UB(...) ILVR_D2(v16u8, __VA_ARGS__)
1458 #define ILVR_D2_SB(...) ILVR_D2(v16i8, __VA_ARGS__)
1459 #define ILVR_D2_SH(...) ILVR_D2(v8i16, __VA_ARGS__)
1461 #define ILVR_D3(RTYPE, in0, in1, in2, in3, in4, in5, out0, out1, out2) \
1463 ILVR_D2(RTYPE, in0, in1, in2, in3, out0, out1); \
1464 out2 = (RTYPE) __msa_ilvr_d((v2i64) in4, (v2i64) in5); \
1466 #define ILVR_D3_SB(...) ILVR_D3(v16i8, __VA_ARGS__)
1468 #define ILVR_D4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, \
1469 out0, out1, out2, out3) \
1471 ILVR_D2(RTYPE, in0, in1, in2, in3, out0, out1); \
1472 ILVR_D2(RTYPE, in4, in5, in6, in7, out2, out3); \
1474 #define ILVR_D4_SB(...) ILVR_D4(v16i8, __VA_ARGS__)
1475 #define ILVR_D4_UB(...) ILVR_D4(v16u8, __VA_ARGS__)
1477 /* Description : Interleave left half of double word elements from vectors
1478 Arguments : Inputs - in0, in1, in2, in3
1479 Outputs - out0, out1
1480 Return Type - as per RTYPE
1481 Details : Left half of double word elements of in0 and left half of
1482 double word elements of in1 are interleaved and copied to out0.
1483 Left half of double word elements of in2 and left half of
1484 double word elements of in3 are interleaved and copied to out1.
1486 #define ILVL_D2(RTYPE, in0, in1, in2, in3, out0, out1) \
1488 out0 = (RTYPE) __msa_ilvl_d((v2i64) in0, (v2i64) in1); \
1489 out1 = (RTYPE) __msa_ilvl_d((v2i64) in2, (v2i64) in3); \
1491 #define ILVL_D2_UB(...) ILVL_D2(v16u8, __VA_ARGS__)
1492 #define ILVL_D2_SB(...) ILVL_D2(v16i8, __VA_ARGS__)
1493 #define ILVL_D2_SH(...) ILVL_D2(v8i16, __VA_ARGS__)
1495 /* Description : Interleave both left and right half of input vectors
1496 Arguments : Inputs - in0, in1
1497 Outputs - out0, out1
1498 Return Type - as per RTYPE
1499 Details : Right half of byte elements from 'in0' and 'in1' are
1500 interleaved and stored to 'out0'
1501 Left half of byte elements from 'in0' and 'in1' are
1502 interleaved and stored to 'out1'
1504 #define ILVRL_B2(RTYPE, in0, in1, out0, out1) \
1506 out0 = (RTYPE) __msa_ilvr_b((v16i8) in0, (v16i8) in1); \
1507 out1 = (RTYPE) __msa_ilvl_b((v16i8) in0, (v16i8) in1); \
1509 #define ILVRL_B2_UB(...) ILVRL_B2(v16u8, __VA_ARGS__)
1510 #define ILVRL_B2_SB(...) ILVRL_B2(v16i8, __VA_ARGS__)
1511 #define ILVRL_B2_UH(...) ILVRL_B2(v8u16, __VA_ARGS__)
1512 #define ILVRL_B2_SH(...) ILVRL_B2(v8i16, __VA_ARGS__)
1513 #define ILVRL_B2_SW(...) ILVRL_B2(v4i32, __VA_ARGS__)
1515 #define ILVRL_H2(RTYPE, in0, in1, out0, out1) \
1517 out0 = (RTYPE) __msa_ilvr_h((v8i16) in0, (v8i16) in1); \
1518 out1 = (RTYPE) __msa_ilvl_h((v8i16) in0, (v8i16) in1); \
1520 #define ILVRL_H2_UB(...) ILVRL_H2(v16u8, __VA_ARGS__)
1521 #define ILVRL_H2_SB(...) ILVRL_H2(v16i8, __VA_ARGS__)
1522 #define ILVRL_H2_SH(...) ILVRL_H2(v8i16, __VA_ARGS__)
1523 #define ILVRL_H2_SW(...) ILVRL_H2(v4i32, __VA_ARGS__)
1525 #define ILVRL_W2(RTYPE, in0, in1, out0, out1) \
1527 out0 = (RTYPE) __msa_ilvr_w((v4i32) in0, (v4i32) in1); \
1528 out1 = (RTYPE) __msa_ilvl_w((v4i32) in0, (v4i32) in1); \
1530 #define ILVRL_W2_UB(...) ILVRL_W2(v16u8, __VA_ARGS__)
1531 #define ILVRL_W2_SH(...) ILVRL_W2(v8i16, __VA_ARGS__)
1532 #define ILVRL_W2_SW(...) ILVRL_W2(v4i32, __VA_ARGS__)
1534 /* Description : Maximum values between signed elements of vector and
1535 5-bit signed immediate value are copied to the output vector
1536 Arguments : Inputs - in0, in1, in2, in3, max_val
1537 Outputs - in0, in1, in2, in3 (in place)
1538 Return Type - as per RTYPE
1539 Details : Maximum of signed halfword element values from 'in0' and
1540 'max_val' are written to output vector 'in0'
1542 #define MAXI_SH2(RTYPE, in0, in1, max_val) \
1544 in0 = (RTYPE) __msa_maxi_s_h((v8i16) in0, max_val); \
1545 in1 = (RTYPE) __msa_maxi_s_h((v8i16) in1, max_val); \
1547 #define MAXI_SH2_UH(...) MAXI_SH2(v8u16, __VA_ARGS__)
1548 #define MAXI_SH2_SH(...) MAXI_SH2(v8i16, __VA_ARGS__)
1550 #define MAXI_SH4(RTYPE, in0, in1, in2, in3, max_val) \
1552 MAXI_SH2(RTYPE, in0, in1, max_val); \
1553 MAXI_SH2(RTYPE, in2, in3, max_val); \
1555 #define MAXI_SH4_UH(...) MAXI_SH4(v8u16, __VA_ARGS__)
1556 #define MAXI_SH4_SH(...) MAXI_SH4(v8i16, __VA_ARGS__)
1558 #define MAXI_SH8(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, max_val) \
1560 MAXI_SH4(RTYPE, in0, in1, in2, in3, max_val); \
1561 MAXI_SH4(RTYPE, in4, in5, in6, in7, max_val); \
1563 #define MAXI_SH8_UH(...) MAXI_SH8(v8u16, __VA_ARGS__)
1564 #define MAXI_SH8_SH(...) MAXI_SH8(v8i16, __VA_ARGS__)
1566 /* Description : Saturate the halfword element values to the max
1567 unsigned value of (sat_val+1 bits)
1568 The element data width remains unchanged
1569 Arguments : Inputs - in0, in1, in2, in3, sat_val
1570 Outputs - in0, in1, in2, in3 (in place)
1571 Return Type - as per RTYPE
1572 Details : Each unsigned halfword element from 'in0' is saturated to the
1573 value generated with (sat_val+1) bit range
1574 Results are in placed to original vectors
1576 #define SAT_UH2(RTYPE, in0, in1, sat_val) \
1578 in0 = (RTYPE) __msa_sat_u_h((v8u16) in0, sat_val); \
1579 in1 = (RTYPE) __msa_sat_u_h((v8u16) in1, sat_val); \
1581 #define SAT_UH2_UH(...) SAT_UH2(v8u16, __VA_ARGS__)
1582 #define SAT_UH2_SH(...) SAT_UH2(v8i16, __VA_ARGS__)
1584 #define SAT_UH4(RTYPE, in0, in1, in2, in3, sat_val) \
1586 SAT_UH2(RTYPE, in0, in1, sat_val); \
1587 SAT_UH2(RTYPE, in2, in3, sat_val); \
1589 #define SAT_UH4_UH(...) SAT_UH4(v8u16, __VA_ARGS__)
1590 #define SAT_UH4_SH(...) SAT_UH4(v8i16, __VA_ARGS__)
1592 #define SAT_UH8(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, sat_val) \
1594 SAT_UH4(RTYPE, in0, in1, in2, in3, sat_val); \
1595 SAT_UH4(RTYPE, in4, in5, in6, in7, sat_val); \
1597 #define SAT_UH8_UH(...) SAT_UH8(v8u16, __VA_ARGS__)
1598 #define SAT_UH8_SH(...) SAT_UH8(v8i16, __VA_ARGS__)
1600 /* Description : Saturate the halfword element values to the max
1601 unsigned value of (sat_val+1 bits)
1602 The element data width remains unchanged
1603 Arguments : Inputs - in0, in1, in2, in3, sat_val
1604 Outputs - in0, in1, in2, in3 (in place)
1605 Return Type - as per RTYPE
1606 Details : Each unsigned halfword element from 'in0' is saturated to the
1607 value generated with (sat_val+1) bit range
1608 Results are in placed to original vectors
1610 #define SAT_SH2(RTYPE, in0, in1, sat_val) \
1612 in0 = (RTYPE) __msa_sat_s_h((v8i16) in0, sat_val); \
1613 in1 = (RTYPE) __msa_sat_s_h((v8i16) in1, sat_val); \
1615 #define SAT_SH2_SH(...) SAT_SH2(v8i16, __VA_ARGS__)
1617 #define SAT_SH3(RTYPE, in0, in1, in2, sat_val) \
1619 SAT_SH2(RTYPE, in0, in1, sat_val); \
1620 in2 = (RTYPE) __msa_sat_s_h((v8i16) in2, sat_val); \
1622 #define SAT_SH3_SH(...) SAT_SH3(v8i16, __VA_ARGS__)
1624 #define SAT_SH4(RTYPE, in0, in1, in2, in3, sat_val) \
1626 SAT_SH2(RTYPE, in0, in1, sat_val); \
1627 SAT_SH2(RTYPE, in2, in3, sat_val); \
1629 #define SAT_SH4_SH(...) SAT_SH4(v8i16, __VA_ARGS__)
1631 /* Description : Saturate the word element values to the max
1632 unsigned value of (sat_val+1 bits)
1633 The element data width remains unchanged
1634 Arguments : Inputs - in0, in1, in2, in3, sat_val
1635 Outputs - in0, in1, in2, in3 (in place)
1636 Return Type - as per RTYPE
1637 Details : Each unsigned word element from 'in0' is saturated to the
1638 value generated with (sat_val+1) bit range
1639 Results are in placed to original vectors
1641 #define SAT_SW2(RTYPE, in0, in1, sat_val) \
1643 in0 = (RTYPE) __msa_sat_s_w((v4i32) in0, sat_val); \
1644 in1 = (RTYPE) __msa_sat_s_w((v4i32) in1, sat_val); \
1646 #define SAT_SW2_SW(...) SAT_SW2(v4i32, __VA_ARGS__)
1648 #define SAT_SW4(RTYPE, in0, in1, in2, in3, sat_val) \
1650 SAT_SW2(RTYPE, in0, in1, sat_val); \
1651 SAT_SW2(RTYPE, in2, in3, sat_val); \
1653 #define SAT_SW4_SW(...) SAT_SW4(v4i32, __VA_ARGS__)
1655 /* Description : Indexed halfword element values are replicated to all
1656 elements in output vector
1657 Arguments : Inputs - in, idx0, idx1
1658 Outputs - out0, out1
1659 Return Type - as per RTYPE
1660 Details : 'idx0' element value from 'in' vector is replicated to all
1661 elements in 'out0' vector
1662 Valid index range for halfword operation is 0-7
1664 #define SPLATI_H2(RTYPE, in, idx0, idx1, out0, out1) \
1666 out0 = (RTYPE) __msa_splati_h((v8i16) in, idx0); \
1667 out1 = (RTYPE) __msa_splati_h((v8i16) in, idx1); \
1669 #define SPLATI_H2_SB(...) SPLATI_H2(v16i8, __VA_ARGS__)
1670 #define SPLATI_H2_SH(...) SPLATI_H2(v8i16, __VA_ARGS__)
1672 #define SPLATI_H3(RTYPE, in, idx0, idx1, idx2, \
1675 SPLATI_H2(RTYPE, in, idx0, idx1, out0, out1); \
1676 out2 = (RTYPE) __msa_splati_h((v8i16) in, idx2); \
1678 #define SPLATI_H3_SB(...) SPLATI_H3(v16i8, __VA_ARGS__)
1679 #define SPLATI_H3_SH(...) SPLATI_H3(v8i16, __VA_ARGS__)
1681 #define SPLATI_H4(RTYPE, in, idx0, idx1, idx2, idx3, \
1682 out0, out1, out2, out3) \
1684 SPLATI_H2(RTYPE, in, idx0, idx1, out0, out1); \
1685 SPLATI_H2(RTYPE, in, idx2, idx3, out2, out3); \
1687 #define SPLATI_H4_SB(...) SPLATI_H4(v16i8, __VA_ARGS__)
1688 #define SPLATI_H4_SH(...) SPLATI_H4(v8i16, __VA_ARGS__)
1690 /* Description : Indexed word element values are replicated to all
1691 elements in output vector
1692 Arguments : Inputs - in, stidx
1693 Outputs - out0, out1
1694 Return Type - as per RTYPE
1695 Details : 'stidx' element value from 'in' vector is replicated to all
1696 elements in 'out0' vector
1697 'stidx + 1' element value from 'in' vector is replicated to all
1698 elements in 'out1' vector
1699 Valid index range for halfword operation is 0-3
1701 #define SPLATI_W2(RTYPE, in, stidx, out0, out1) \
1703 out0 = (RTYPE) __msa_splati_w((v4i32) in, stidx); \
1704 out1 = (RTYPE) __msa_splati_w((v4i32) in, (stidx+1)); \
1706 #define SPLATI_W2_SH(...) SPLATI_W2(v8i16, __VA_ARGS__)
1707 #define SPLATI_W2_SW(...) SPLATI_W2(v4i32, __VA_ARGS__)
1709 #define SPLATI_W4(RTYPE, in, out0, out1, out2, out3) \
1711 SPLATI_W2(RTYPE, in, 0, out0, out1); \
1712 SPLATI_W2(RTYPE, in, 2, out2, out3); \
1714 #define SPLATI_W4_SH(...) SPLATI_W4(v8i16, __VA_ARGS__)
1715 #define SPLATI_W4_SW(...) SPLATI_W4(v4i32, __VA_ARGS__)
1717 /* Description : Pack even byte elements of vector pairs
1718 Arguments : Inputs - in0, in1, in2, in3
1719 Outputs - out0, out1
1720 Return Type - as per RTYPE
1721 Details : Even byte elements of in0 are copied to the left half of
1722 out0 & even byte elements of in1 are copied to the right
1724 Even byte elements of in2 are copied to the left half of
1725 out1 & even byte elements of in3 are copied to the right
1728 #define PCKEV_B2(RTYPE, in0, in1, in2, in3, out0, out1) \
1730 out0 = (RTYPE) __msa_pckev_b((v16i8) in0, (v16i8) in1); \
1731 out1 = (RTYPE) __msa_pckev_b((v16i8) in2, (v16i8) in3); \
1733 #define PCKEV_B2_SB(...) PCKEV_B2(v16i8, __VA_ARGS__)
1734 #define PCKEV_B2_UB(...) PCKEV_B2(v16u8, __VA_ARGS__)
1735 #define PCKEV_B2_SH(...) PCKEV_B2(v8i16, __VA_ARGS__)
1736 #define PCKEV_B2_SW(...) PCKEV_B2(v4i32, __VA_ARGS__)
1738 #define PCKEV_B3(RTYPE, in0, in1, in2, in3, in4, in5, out0, out1, out2) \
1740 PCKEV_B2(RTYPE, in0, in1, in2, in3, out0, out1); \
1741 out2 = (RTYPE) __msa_pckev_b((v16i8) in4, (v16i8) in5); \
1743 #define PCKEV_B3_UB(...) PCKEV_B3(v16u8, __VA_ARGS__)
1744 #define PCKEV_B3_SB(...) PCKEV_B3(v16i8, __VA_ARGS__)
1746 #define PCKEV_B4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, \
1747 out0, out1, out2, out3) \
1749 PCKEV_B2(RTYPE, in0, in1, in2, in3, out0, out1); \
1750 PCKEV_B2(RTYPE, in4, in5, in6, in7, out2, out3); \
1752 #define PCKEV_B4_SB(...) PCKEV_B4(v16i8, __VA_ARGS__)
1753 #define PCKEV_B4_UB(...) PCKEV_B4(v16u8, __VA_ARGS__)
1754 #define PCKEV_B4_SH(...) PCKEV_B4(v8i16, __VA_ARGS__)
1755 #define PCKEV_B4_SW(...) PCKEV_B4(v4i32, __VA_ARGS__)
1757 /* Description : Pack even halfword elements of vector pairs
1758 Arguments : Inputs - in0, in1, in2, in3
1759 Outputs - out0, out1
1760 Return Type - as per RTYPE
1761 Details : Even halfword elements of in0 are copied to the left half of
1762 out0 & even halfword elements of in1 are copied to the right
1764 Even halfword elements of in2 are copied to the left half of
1765 out1 & even halfword elements of in3 are copied to the right
1768 #define PCKEV_H2(RTYPE, in0, in1, in2, in3, out0, out1) \
1770 out0 = (RTYPE) __msa_pckev_h((v8i16) in0, (v8i16) in1); \
1771 out1 = (RTYPE) __msa_pckev_h((v8i16) in2, (v8i16) in3); \
1773 #define PCKEV_H2_SH(...) PCKEV_H2(v8i16, __VA_ARGS__)
1774 #define PCKEV_H2_SW(...) PCKEV_H2(v4i32, __VA_ARGS__)
1776 #define PCKEV_H4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, \
1777 out0, out1, out2, out3) \
1779 PCKEV_H2(RTYPE, in0, in1, in2, in3, out0, out1); \
1780 PCKEV_H2(RTYPE, in4, in5, in6, in7, out2, out3); \
1782 #define PCKEV_H4_SH(...) PCKEV_H4(v8i16, __VA_ARGS__)
1783 #define PCKEV_H4_SW(...) PCKEV_H4(v4i32, __VA_ARGS__)
1785 /* Description : Pack even double word elements of vector pairs
1786 Arguments : Inputs - in0, in1, in2, in3
1787 Outputs - out0, out1
1788 Return Type - as per RTYPE
1789 Details : Even double elements of in0 are copied to the left half of
1790 out0 & even double elements of in1 are copied to the right
1792 Even double elements of in2 are copied to the left half of
1793 out1 & even double elements of in3 are copied to the right
1796 #define PCKEV_D2(RTYPE, in0, in1, in2, in3, out0, out1) \
1798 out0 = (RTYPE) __msa_pckev_d((v2i64) in0, (v2i64) in1); \
1799 out1 = (RTYPE) __msa_pckev_d((v2i64) in2, (v2i64) in3); \
1801 #define PCKEV_D2_UB(...) PCKEV_D2(v16u8, __VA_ARGS__)
1802 #define PCKEV_D2_SB(...) PCKEV_D2(v16i8, __VA_ARGS__)
1803 #define PCKEV_D2_SH(...) PCKEV_D2(v8i16, __VA_ARGS__)
1805 #define PCKEV_D4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, \
1806 out0, out1, out2, out3) \
1808 PCKEV_D2(RTYPE, in0, in1, in2, in3, out0, out1); \
1809 PCKEV_D2(RTYPE, in4, in5, in6, in7, out2, out3); \
1811 #define PCKEV_D4_UB(...) PCKEV_D4(v16u8, __VA_ARGS__)
1813 /* Description : Pack odd double word elements of vector pairs
1814 Arguments : Inputs - in0, in1
1815 Outputs - out0, out1
1816 Return Type - as per RTYPE
1817 Details : As operation is on same input 'in0' vector, index 1 double word
1818 element is overwritten to index 0 and result is written to out0
1819 As operation is on same input 'in1' vector, index 1 double word
1820 element is overwritten to index 0 and result is written to out1
1822 #define PCKOD_D2(RTYPE, in0, in1, in2, in3, out0, out1) \
1824 out0 = (RTYPE) __msa_pckod_d((v2i64) in0, (v2i64) in1); \
1825 out1 = (RTYPE) __msa_pckod_d((v2i64) in2, (v2i64) in3); \
1827 #define PCKOD_D2_UB(...) PCKOD_D2(v16u8, __VA_ARGS__)
1828 #define PCKOD_D2_SH(...) PCKOD_D2(v8i16, __VA_ARGS__)
1829 #define PCKOD_D2_SD(...) PCKOD_D2(v2i64, __VA_ARGS__)
1831 /* Description : Each byte element is logically xor'ed with immediate 128
1832 Arguments : Inputs - in0, in1
1833 Outputs - in0, in1 (in-place)
1834 Return Type - as per RTYPE
1835 Details : Each unsigned byte element from input vector 'in0' is
1836 logically xor'ed with 128 and result is in-place stored in
1838 Each unsigned byte element from input vector 'in1' is
1839 logically xor'ed with 128 and result is in-place stored in
1841 Similar for other pairs
1843 #define XORI_B2_128(RTYPE, in0, in1) \
1845 in0 = (RTYPE) __msa_xori_b((v16u8) in0, 128); \
1846 in1 = (RTYPE) __msa_xori_b((v16u8) in1, 128); \
1848 #define XORI_B2_128_UB(...) XORI_B2_128(v16u8, __VA_ARGS__)
1849 #define XORI_B2_128_SB(...) XORI_B2_128(v16i8, __VA_ARGS__)
1850 #define XORI_B2_128_SH(...) XORI_B2_128(v8i16, __VA_ARGS__)
1852 #define XORI_B3_128(RTYPE, in0, in1, in2) \
1854 XORI_B2_128(RTYPE, in0, in1); \
1855 in2 = (RTYPE) __msa_xori_b((v16u8) in2, 128); \
1857 #define XORI_B3_128_SB(...) XORI_B3_128(v16i8, __VA_ARGS__)
1859 #define XORI_B4_128(RTYPE, in0, in1, in2, in3) \
1861 XORI_B2_128(RTYPE, in0, in1); \
1862 XORI_B2_128(RTYPE, in2, in3); \
1864 #define XORI_B4_128_UB(...) XORI_B4_128(v16u8, __VA_ARGS__)
1865 #define XORI_B4_128_SB(...) XORI_B4_128(v16i8, __VA_ARGS__)
1866 #define XORI_B4_128_SH(...) XORI_B4_128(v8i16, __VA_ARGS__)
1868 #define XORI_B5_128(RTYPE, in0, in1, in2, in3, in4) \
1870 XORI_B3_128(RTYPE, in0, in1, in2); \
1871 XORI_B2_128(RTYPE, in3, in4); \
1873 #define XORI_B5_128_SB(...) XORI_B5_128(v16i8, __VA_ARGS__)
1875 #define XORI_B6_128(RTYPE, in0, in1, in2, in3, in4, in5) \
1877 XORI_B4_128(RTYPE, in0, in1, in2, in3); \
1878 XORI_B2_128(RTYPE, in4, in5); \
1880 #define XORI_B6_128_SB(...) XORI_B6_128(v16i8, __VA_ARGS__)
1882 #define XORI_B7_128(RTYPE, in0, in1, in2, in3, in4, in5, in6) \
1884 XORI_B4_128(RTYPE, in0, in1, in2, in3); \
1885 XORI_B3_128(RTYPE, in4, in5, in6); \
1887 #define XORI_B7_128_SB(...) XORI_B7_128(v16i8, __VA_ARGS__)
1889 #define XORI_B8_128(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7) \
1891 XORI_B4_128(RTYPE, in0, in1, in2, in3); \
1892 XORI_B4_128(RTYPE, in4, in5, in6, in7); \
1894 #define XORI_B8_128_SB(...) XORI_B8_128(v16i8, __VA_ARGS__)
1895 #define XORI_B8_128_UB(...) XORI_B8_128(v16u8, __VA_ARGS__)
1897 /* Description : Addition of signed halfword elements and signed saturation
1898 Arguments : Inputs - in0, in1, in2, in3
1899 Outputs - out0, out1
1900 Return Type - as per RTYPE
1901 Details : Signed halfword elements from 'in0' are added to signed
1902 halfword elements of 'in1'. The result is then signed saturated
1903 between -32768 to +32767 (as per halfword data type)
1904 Similar for other pairs
1906 #define ADDS_SH2(RTYPE, in0, in1, in2, in3, out0, out1) \
1908 out0 = (RTYPE) __msa_adds_s_h((v8i16) in0, (v8i16) in1); \
1909 out1 = (RTYPE) __msa_adds_s_h((v8i16) in2, (v8i16) in3); \
1911 #define ADDS_SH2_SH(...) ADDS_SH2(v8i16, __VA_ARGS__)
1913 #define ADDS_SH4(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, \
1914 out0, out1, out2, out3) \
1916 ADDS_SH2(RTYPE, in0, in1, in2, in3, out0, out1); \
1917 ADDS_SH2(RTYPE, in4, in5, in6, in7, out2, out3); \
1919 #define ADDS_SH4_UH(...) ADDS_SH4(v8u16, __VA_ARGS__)
1920 #define ADDS_SH4_SH(...) ADDS_SH4(v8i16, __VA_ARGS__)
1922 /* Description : Shift left all elements of vector (generic for all data types)
1923 Arguments : Inputs - in0, in1, in2, in3, shift
1924 Outputs - in0, in1, in2, in3 (in place)
1925 Return Type - as per input vector RTYPE
1926 Details : Each element of vector 'in0' is left shifted by 'shift' and
1927 result is in place written to 'in0'
1928 Similar for other pairs
1930 #define SLLI_2V(in0, in1, shift) \
1932 in0 = in0 << shift; \
1933 in1 = in1 << shift; \
1935 #define SLLI_4V(in0, in1, in2, in3, shift) \
1937 in0 = in0 << shift; \
1938 in1 = in1 << shift; \
1939 in2 = in2 << shift; \
1940 in3 = in3 << shift; \
1943 /* Description : Arithmetic shift right all elements of vector
1944 (generic for all data types)
1945 Arguments : Inputs - in0, in1, in2, in3, shift
1946 Outputs - in0, in1, in2, in3 (in place)
1947 Return Type - as per input vector RTYPE
1948 Details : Each element of vector 'in0' is right shifted by 'shift' and
1949 result is in place written to 'in0'
1950 Here, 'shift' is GP variable passed in
1951 Similar for other pairs
1953 #define SRA_4V(in0, in1, in2, in3, shift) \
1955 in0 = in0 >> shift; \
1956 in1 = in1 >> shift; \
1957 in2 = in2 >> shift; \
1958 in3 = in3 >> shift; \
1961 /* Description : Shift right logical all halfword elements of vector
1962 Arguments : Inputs - in0, in1, in2, in3, shift
1963 Outputs - in0, in1, in2, in3 (in place)
1964 Return Type - as per RTYPE
1965 Details : Each element of vector 'in0' is shifted right logical by
1966 number of bits respective element holds in vector 'shift' and
1967 result is in place written to 'in0'
1968 Here, 'shift' is a vector passed in
1969 Similar for other pairs
1971 #define SRL_H4(RTYPE, in0, in1, in2, in3, shift) \
1973 in0 = (RTYPE) __msa_srl_h((v8i16) in0, (v8i16) shift); \
1974 in1 = (RTYPE) __msa_srl_h((v8i16) in1, (v8i16) shift); \
1975 in2 = (RTYPE) __msa_srl_h((v8i16) in2, (v8i16) shift); \
1976 in3 = (RTYPE) __msa_srl_h((v8i16) in3, (v8i16) shift); \
1978 #define SRL_H4_UH(...) SRL_H4(v8u16, __VA_ARGS__)
1980 #define SRLR_H4(RTYPE, in0, in1, in2, in3, shift) \
1982 in0 = (RTYPE) __msa_srlr_h((v8i16) in0, (v8i16) shift); \
1983 in1 = (RTYPE) __msa_srlr_h((v8i16) in1, (v8i16) shift); \
1984 in2 = (RTYPE) __msa_srlr_h((v8i16) in2, (v8i16) shift); \
1985 in3 = (RTYPE) __msa_srlr_h((v8i16) in3, (v8i16) shift); \
1987 #define SRLR_H4_UH(...) SRLR_H4(v8u16, __VA_ARGS__)
1988 #define SRLR_H4_SH(...) SRLR_H4(v8i16, __VA_ARGS__)
1990 #define SRLR_H8(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, shift) \
1992 SRLR_H4(RTYPE, in0, in1, in2, in3, shift); \
1993 SRLR_H4(RTYPE, in4, in5, in6, in7, shift); \
1995 #define SRLR_H8_UH(...) SRLR_H8(v8u16, __VA_ARGS__)
1996 #define SRLR_H8_SH(...) SRLR_H8(v8i16, __VA_ARGS__)
1998 /* Description : Shift right arithmetic rounded halfwords
1999 Arguments : Inputs - in0, in1, shift
2000 Outputs - in0, in1, (in place)
2001 Return Type - as per RTYPE
2002 Details : Each element of vector 'in0' is shifted right arithmetic by
2003 number of bits respective element holds in vector 'shift'.
2004 The last discarded bit is added to shifted value for rounding
2005 and the result is in place written to 'in0'
2006 Here, 'shift' is a vector passed in
2007 Similar for other pairs
2009 #define SRAR_H2(RTYPE, in0, in1, shift) \
2011 in0 = (RTYPE) __msa_srar_h((v8i16) in0, (v8i16) shift); \
2012 in1 = (RTYPE) __msa_srar_h((v8i16) in1, (v8i16) shift); \
2014 #define SRAR_H2_UH(...) SRAR_H2(v8u16, __VA_ARGS__)
2015 #define SRAR_H2_SH(...) SRAR_H2(v8i16, __VA_ARGS__)
2017 #define SRAR_H3(RTYPE, in0, in1, in2, shift) \
2019 SRAR_H2(RTYPE, in0, in1, shift) \
2020 in2 = (RTYPE) __msa_srar_h((v8i16) in2, (v8i16) shift); \
2022 #define SRAR_H3_SH(...) SRAR_H3(v8i16, __VA_ARGS__)
2024 #define SRAR_H4(RTYPE, in0, in1, in2, in3, shift) \
2026 SRAR_H2(RTYPE, in0, in1, shift) \
2027 SRAR_H2(RTYPE, in2, in3, shift) \
2029 #define SRAR_H4_UH(...) SRAR_H4(v8u16, __VA_ARGS__)
2030 #define SRAR_H4_SH(...) SRAR_H4(v8i16, __VA_ARGS__)
2032 /* Description : Shift right arithmetic rounded words
2033 Arguments : Inputs - in0, in1, shift
2034 Outputs - in0, in1, (in place)
2035 Return Type - as per RTYPE
2036 Details : Each element of vector 'in0' is shifted right arithmetic by
2037 number of bits respective element holds in vector 'shift'.
2038 The last discarded bit is added to shifted value for rounding
2039 and the result is in place written to 'in0'
2040 Here, 'shift' is a vector passed in
2041 Similar for other pairs
2043 #define SRAR_W2(RTYPE, in0, in1, shift) \
2045 in0 = (RTYPE) __msa_srar_w((v4i32) in0, (v4i32) shift); \
2046 in1 = (RTYPE) __msa_srar_w((v4i32) in1, (v4i32) shift); \
2048 #define SRAR_W2_SW(...) SRAR_W2(v4i32, __VA_ARGS__)
2050 #define SRAR_W4(RTYPE, in0, in1, in2, in3, shift) \
2052 SRAR_W2(RTYPE, in0, in1, shift) \
2053 SRAR_W2(RTYPE, in2, in3, shift) \
2055 #define SRAR_W4_SW(...) SRAR_W4(v4i32, __VA_ARGS__)
2057 /* Description : Shift right arithmetic rounded (immediate)
2058 Arguments : Inputs - in0, in1, in2, in3, shift
2059 Outputs - in0, in1, in2, in3 (in place)
2060 Return Type - as per RTYPE
2061 Details : Each element of vector 'in0' is shifted right arithmetic by
2063 The last discarded bit is added to shifted value for rounding
2064 and the result is in place written to 'in0'
2065 Similar for other pairs
2067 #define SRARI_H2(RTYPE, in0, in1, shift) \
2069 in0 = (RTYPE) __msa_srari_h((v8i16) in0, shift); \
2070 in1 = (RTYPE) __msa_srari_h((v8i16) in1, shift); \
2072 #define SRARI_H2_UH(...) SRARI_H2(v8u16, __VA_ARGS__)
2073 #define SRARI_H2_SH(...) SRARI_H2(v8i16, __VA_ARGS__)
2075 #define SRARI_H4(RTYPE, in0, in1, in2, in3, shift) \
2077 SRARI_H2(RTYPE, in0, in1, shift); \
2078 SRARI_H2(RTYPE, in2, in3, shift); \
2080 #define SRARI_H4_UH(...) SRARI_H4(v8u16, __VA_ARGS__)
2081 #define SRARI_H4_SH(...) SRARI_H4(v8i16, __VA_ARGS__)
2083 /* Description : Shift right arithmetic rounded (immediate)
2084 Arguments : Inputs - in0, in1, shift
2085 Outputs - in0, in1 (in place)
2086 Return Type - as per RTYPE
2087 Details : Each element of vector 'in0' is shifted right arithmetic by
2089 The last discarded bit is added to shifted value for rounding
2090 and the result is in place written to 'in0'
2091 Similar for other pairs
2093 #define SRARI_W2(RTYPE, in0, in1, shift) \
2095 in0 = (RTYPE) __msa_srari_w((v4i32) in0, shift); \
2096 in1 = (RTYPE) __msa_srari_w((v4i32) in1, shift); \
2098 #define SRARI_W2_SW(...) SRARI_W2(v4i32, __VA_ARGS__)
2100 #define SRARI_W4(RTYPE, in0, in1, in2, in3, shift) \
2102 SRARI_W2(RTYPE, in0, in1, shift); \
2103 SRARI_W2(RTYPE, in2, in3, shift); \
2105 #define SRARI_W4_SH(...) SRARI_W4(v8i16, __VA_ARGS__)
2106 #define SRARI_W4_SW(...) SRARI_W4(v4i32, __VA_ARGS__)
2108 /* Description : Multiplication of pairs of vectors
2109 Arguments : Inputs - in0, in1, in2, in3
2110 Outputs - out0, out1
2111 Details : Each element from 'in0' is multiplied with elements from 'in1'
2112 and result is written to 'out0'
2113 Similar for other pairs
2115 #define MUL2(in0, in1, in2, in3, out0, out1) \
2120 #define MUL4(in0, in1, in2, in3, in4, in5, in6, in7, out0, out1, out2, out3) \
2122 MUL2(in0, in1, in2, in3, out0, out1); \
2123 MUL2(in4, in5, in6, in7, out2, out3); \
2126 /* Description : Addition of 2 pairs of vectors
2127 Arguments : Inputs - in0, in1, in2, in3
2128 Outputs - out0, out1
2129 Details : Each element from 2 pairs vectors is added and 2 results are
2132 #define ADD2(in0, in1, in2, in3, out0, out1) \
2137 #define ADD4(in0, in1, in2, in3, in4, in5, in6, in7, out0, out1, out2, out3) \
2139 ADD2(in0, in1, in2, in3, out0, out1); \
2140 ADD2(in4, in5, in6, in7, out2, out3); \
2143 /* Description : Subtraction of 2 pairs of vectors
2144 Arguments : Inputs - in0, in1, in2, in3
2145 Outputs - out0, out1
2146 Details : Each element from 2 pairs vectors is subtracted and 2 results
2149 #define SUB2(in0, in1, in2, in3, out0, out1) \
2154 #define SUB4(in0, in1, in2, in3, in4, in5, in6, in7, out0, out1, out2, out3) \
2162 /* Description : Sign extend byte elements from right half of the vector
2163 Arguments : Input - in (byte vector)
2164 Output - out (sign extended halfword vector)
2165 Return Type - signed halfword
2166 Details : Sign bit of byte elements from input vector 'in' is
2167 extracted and interleaved with same vector 'in' to generate
2168 8 halfword elements keeping sign intact
2170 #define UNPCK_R_SB_SH(in, out) \
2174 sign_m = __msa_clti_s_b((v16i8) in, 0); \
2175 out = (v8i16) __msa_ilvr_b(sign_m, (v16i8) in); \
2178 /* Description : Sign extend halfword elements from right half of the vector
2179 Arguments : Inputs - in (input halfword vector)
2180 Outputs - out (sign extended word vectors)
2181 Return Type - signed word
2182 Details : Sign bit of halfword elements from input vector 'in' is
2183 extracted and interleaved with same vector 'in0' to generate
2184 4 word elements keeping sign intact
2187 #define UNPCK_R_SH_SW(in, out) \
2189 out = (v4i32) __builtin_msa2_w2x_lo_s_h((v8i16) in); \
2192 #define UNPCK_R_SH_SW(in, out) \
2196 sign_m = __msa_clti_s_h((v8i16) in, 0); \
2197 out = (v4i32) __msa_ilvr_h(sign_m, (v8i16) in); \
2199 #endif // #if HAVE_MSA2
2201 /* Description : Sign extend byte elements from input vector and return
2202 halfword results in pair of vectors
2203 Arguments : Inputs - in (1 input byte vector)
2204 Outputs - out0, out1 (sign extended 2 halfword vectors)
2205 Return Type - signed halfword
2206 Details : Sign bit of byte elements from input vector 'in' is
2207 extracted and interleaved right with same vector 'in0' to
2208 generate 8 signed halfword elements in 'out0'
2209 Then interleaved left with same vector 'in0' to
2210 generate 8 signed halfword elements in 'out1'
2213 #define UNPCK_SB_SH(in, out0, out1) \
2215 out0 = (v4i32) __builtin_msa2_w2x_lo_s_b((v16i8) in); \
2216 out1 = (v4i32) __builtin_msa2_w2x_hi_s_b((v16i8) in); \
2219 #define UNPCK_SB_SH(in, out0, out1) \
2223 tmp_m = __msa_clti_s_b((v16i8) in, 0); \
2224 ILVRL_B2_SH(tmp_m, in, out0, out1); \
2226 #endif // #if HAVE_MSA2
2228 /* Description : Zero extend unsigned byte elements to halfword elements
2229 Arguments : Inputs - in (1 input unsigned byte vector)
2230 Outputs - out0, out1 (unsigned 2 halfword vectors)
2231 Return Type - signed halfword
2232 Details : Zero extended right half of vector is returned in 'out0'
2233 Zero extended left half of vector is returned in 'out1'
2235 #define UNPCK_UB_SH(in, out0, out1) \
2237 v16i8 zero_m = { 0 }; \
2239 ILVRL_B2_SH(zero_m, in, out0, out1); \
2242 /* Description : Sign extend halfword elements from input vector and return
2243 result in pair of vectors
2244 Arguments : Inputs - in (1 input halfword vector)
2245 Outputs - out0, out1 (sign extended 2 word vectors)
2246 Return Type - signed word
2247 Details : Sign bit of halfword elements from input vector 'in' is
2248 extracted and interleaved right with same vector 'in0' to
2249 generate 4 signed word elements in 'out0'
2250 Then interleaved left with same vector 'in0' to
2251 generate 4 signed word elements in 'out1'
2254 #define UNPCK_SH_SW(in, out0, out1) \
2256 out0 = (v4i32) __builtin_msa2_w2x_lo_s_h((v8i16) in); \
2257 out1 = (v4i32) __builtin_msa2_w2x_hi_s_h((v8i16) in); \
2260 #define UNPCK_SH_SW(in, out0, out1) \
2264 tmp_m = __msa_clti_s_h((v8i16) in, 0); \
2265 ILVRL_H2_SW(tmp_m, in, out0, out1); \
2267 #endif // #if HAVE_MSA2
2269 /* Description : Swap two variables
2270 Arguments : Inputs - in0, in1
2271 Outputs - in0, in1 (in-place)
2272 Details : Swapping of two input variables using xor
2274 #define SWAP(in0, in1) \
2281 /* Description : Butterfly of 4 input vectors
2282 Arguments : Inputs - in0, in1, in2, in3
2283 Outputs - out0, out1, out2, out3
2284 Details : Butterfly operation
2286 #define BUTTERFLY_4(in0, in1, in2, in3, out0, out1, out2, out3) \
2295 /* Description : Butterfly of 8 input vectors
2296 Arguments : Inputs - in0 ... in7
2297 Outputs - out0 .. out7
2298 Details : Butterfly operation
2300 #define BUTTERFLY_8(in0, in1, in2, in3, in4, in5, in6, in7, \
2301 out0, out1, out2, out3, out4, out5, out6, out7) \
2314 /* Description : Butterfly of 16 input vectors
2315 Arguments : Inputs - in0 ... in15
2316 Outputs - out0 .. out15
2317 Details : Butterfly operation
2319 #define BUTTERFLY_16(in0, in1, in2, in3, in4, in5, in6, in7, \
2320 in8, in9, in10, in11, in12, in13, in14, in15, \
2321 out0, out1, out2, out3, out4, out5, out6, out7, \
2322 out8, out9, out10, out11, out12, out13, out14, out15) \
2324 out0 = in0 + in15; \
2325 out1 = in1 + in14; \
2326 out2 = in2 + in13; \
2327 out3 = in3 + in12; \
2328 out4 = in4 + in11; \
2329 out5 = in5 + in10; \
2335 out10 = in5 - in10; \
2336 out11 = in4 - in11; \
2337 out12 = in3 - in12; \
2338 out13 = in2 - in13; \
2339 out14 = in1 - in14; \
2340 out15 = in0 - in15; \
2343 /* Description : Transposes input 4x4 byte block
2344 Arguments : Inputs - in0, in1, in2, in3 (input 4x4 byte block)
2345 Outputs - out0, out1, out2, out3 (output 4x4 byte block)
2346 Return Type - unsigned byte
2349 #define TRANSPOSE4x4_UB_UB(in0, in1, in2, in3, out0, out1, out2, out3) \
2351 v16i8 zero_m = { 0 }; \
2352 v16i8 s0_m, s1_m, s2_m, s3_m; \
2354 ILVR_D2_SB(in1, in0, in3, in2, s0_m, s1_m); \
2355 ILVRL_B2_SB(s1_m, s0_m, s2_m, s3_m); \
2357 out0 = (v16u8) __msa_ilvr_b(s3_m, s2_m); \
2358 out1 = (v16u8) __msa_sldi_b(zero_m, (v16i8) out0, 4); \
2359 out2 = (v16u8) __msa_sldi_b(zero_m, (v16i8) out1, 4); \
2360 out3 = (v16u8) __msa_sldi_b(zero_m, (v16i8) out2, 4); \
2363 /* Description : Transposes input 8x4 byte block into 4x8
2364 Arguments : Inputs - in0, in1, in2, in3 (input 8x4 byte block)
2365 Outputs - out0, out1, out2, out3 (output 4x8 byte block)
2366 Return Type - as per RTYPE
2369 #define TRANSPOSE8x4_UB(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, \
2370 out0, out1, out2, out3) \
2372 v16i8 tmp0_m, tmp1_m, tmp2_m, tmp3_m; \
2374 ILVEV_W2_SB(in0, in4, in1, in5, tmp0_m, tmp1_m); \
2375 tmp2_m = __msa_ilvr_b(tmp1_m, tmp0_m); \
2376 ILVEV_W2_SB(in2, in6, in3, in7, tmp0_m, tmp1_m); \
2378 tmp3_m = __msa_ilvr_b(tmp1_m, tmp0_m); \
2379 ILVRL_H2_SB(tmp3_m, tmp2_m, tmp0_m, tmp1_m); \
2381 ILVRL_W2(RTYPE, tmp1_m, tmp0_m, out0, out2); \
2382 out1 = (RTYPE) __msa_ilvl_d((v2i64) out2, (v2i64) out0); \
2383 out3 = (RTYPE) __msa_ilvl_d((v2i64) out0, (v2i64) out2); \
2385 #define TRANSPOSE8x4_UB_UB(...) TRANSPOSE8x4_UB(v16u8, __VA_ARGS__)
2386 #define TRANSPOSE8x4_UB_UH(...) TRANSPOSE8x4_UB(v8u16, __VA_ARGS__)
2388 /* Description : Transposes input 8x8 byte block
2389 Arguments : Inputs - in0, in1, in2, in3, in4, in5, in6, in7
2390 (input 8x8 byte block)
2391 Outputs - out0, out1, out2, out3, out4, out5, out6, out7
2392 (output 8x8 byte block)
2393 Return Type - as per RTYPE
2396 #define TRANSPOSE8x8_UB(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, \
2397 out0, out1, out2, out3, out4, out5, out6, out7) \
2399 v16i8 tmp0_m, tmp1_m, tmp2_m, tmp3_m; \
2400 v16i8 tmp4_m, tmp5_m, tmp6_m, tmp7_m; \
2401 v16i8 zeros = { 0 }; \
2403 ILVR_B4_SB(in2, in0, in3, in1, in6, in4, in7, in5, \
2404 tmp0_m, tmp1_m, tmp2_m, tmp3_m); \
2405 ILVRL_B2_SB(tmp1_m, tmp0_m, tmp4_m, tmp5_m); \
2406 ILVRL_B2_SB(tmp3_m, tmp2_m, tmp6_m, tmp7_m); \
2407 ILVRL_W2(RTYPE, tmp6_m, tmp4_m, out0, out2); \
2408 ILVRL_W2(RTYPE, tmp7_m, tmp5_m, out4, out6); \
2409 SLDI_B4(RTYPE, zeros, out0, zeros, out2, zeros, out4, zeros, out6, \
2410 8, out1, out3, out5, out7); \
2412 #define TRANSPOSE8x8_UB_UB(...) TRANSPOSE8x8_UB(v16u8, __VA_ARGS__)
2413 #define TRANSPOSE8x8_UB_UH(...) TRANSPOSE8x8_UB(v8u16, __VA_ARGS__)
2415 /* Description : Transposes 16x4 block into 4x16 with byte elements in vectors
2416 Arguments : Inputs - in0, in1, in2, in3, in4, in5, in6, in7,
2417 in8, in9, in10, in11, in12, in13, in14, in15
2418 Outputs - out0, out1, out2, out3
2419 Return Type - unsigned byte
2422 #define TRANSPOSE16x4_UB_UB(in0, in1, in2, in3, in4, in5, in6, in7, \
2423 in8, in9, in10, in11, in12, in13, in14, in15, \
2424 out0, out1, out2, out3) \
2426 v2i64 tmp0_m, tmp1_m, tmp2_m, tmp3_m; \
2428 ILVEV_W2_SD(in0, in4, in8, in12, tmp0_m, tmp1_m); \
2429 out1 = (v16u8) __msa_ilvev_d(tmp1_m, tmp0_m); \
2431 ILVEV_W2_SD(in1, in5, in9, in13, tmp0_m, tmp1_m); \
2432 out3 = (v16u8) __msa_ilvev_d(tmp1_m, tmp0_m); \
2434 ILVEV_W2_SD(in2, in6, in10, in14, tmp0_m, tmp1_m); \
2436 tmp2_m = __msa_ilvev_d(tmp1_m, tmp0_m); \
2437 ILVEV_W2_SD(in3, in7, in11, in15, tmp0_m, tmp1_m); \
2439 tmp3_m = __msa_ilvev_d(tmp1_m, tmp0_m); \
2440 ILVEV_B2_SD(out1, out3, tmp2_m, tmp3_m, tmp0_m, tmp1_m); \
2441 out0 = (v16u8) __msa_ilvev_h((v8i16) tmp1_m, (v8i16) tmp0_m); \
2442 out2 = (v16u8) __msa_ilvod_h((v8i16) tmp1_m, (v8i16) tmp0_m); \
2444 tmp0_m = (v2i64) __msa_ilvod_b((v16i8) out3, (v16i8) out1); \
2445 tmp1_m = (v2i64) __msa_ilvod_b((v16i8) tmp3_m, (v16i8) tmp2_m); \
2446 out1 = (v16u8) __msa_ilvev_h((v8i16) tmp1_m, (v8i16) tmp0_m); \
2447 out3 = (v16u8) __msa_ilvod_h((v8i16) tmp1_m, (v8i16) tmp0_m); \
2450 /* Description : Transposes 16x8 block into 8x16 with byte elements in vectors
2451 Arguments : Inputs - in0, in1, in2, in3, in4, in5, in6, in7,
2452 in8, in9, in10, in11, in12, in13, in14, in15
2453 Outputs - out0, out1, out2, out3, out4, out5, out6, out7
2454 Return Type - unsigned byte
2457 #define TRANSPOSE16x8_UB_UB(in0, in1, in2, in3, in4, in5, in6, in7, \
2458 in8, in9, in10, in11, in12, in13, in14, in15, \
2459 out0, out1, out2, out3, out4, out5, out6, out7) \
2461 v16u8 tmp0_m, tmp1_m, tmp2_m, tmp3_m; \
2462 v16u8 tmp4_m, tmp5_m, tmp6_m, tmp7_m; \
2464 ILVEV_D2_UB(in0, in8, in1, in9, out7, out6); \
2465 ILVEV_D2_UB(in2, in10, in3, in11, out5, out4); \
2466 ILVEV_D2_UB(in4, in12, in5, in13, out3, out2); \
2467 ILVEV_D2_UB(in6, in14, in7, in15, out1, out0); \
2469 tmp0_m = (v16u8) __msa_ilvev_b((v16i8) out6, (v16i8) out7); \
2470 tmp4_m = (v16u8) __msa_ilvod_b((v16i8) out6, (v16i8) out7); \
2471 tmp1_m = (v16u8) __msa_ilvev_b((v16i8) out4, (v16i8) out5); \
2472 tmp5_m = (v16u8) __msa_ilvod_b((v16i8) out4, (v16i8) out5); \
2473 out5 = (v16u8) __msa_ilvev_b((v16i8) out2, (v16i8) out3); \
2474 tmp6_m = (v16u8) __msa_ilvod_b((v16i8) out2, (v16i8) out3); \
2475 out7 = (v16u8) __msa_ilvev_b((v16i8) out0, (v16i8) out1); \
2476 tmp7_m = (v16u8) __msa_ilvod_b((v16i8) out0, (v16i8) out1); \
2478 ILVEV_H2_UB(tmp0_m, tmp1_m, out5, out7, tmp2_m, tmp3_m); \
2479 out0 = (v16u8) __msa_ilvev_w((v4i32) tmp3_m, (v4i32) tmp2_m); \
2480 out4 = (v16u8) __msa_ilvod_w((v4i32) tmp3_m, (v4i32) tmp2_m); \
2482 tmp2_m = (v16u8) __msa_ilvod_h((v8i16) tmp1_m, (v8i16) tmp0_m); \
2483 tmp3_m = (v16u8) __msa_ilvod_h((v8i16) out7, (v8i16) out5); \
2484 out2 = (v16u8) __msa_ilvev_w((v4i32) tmp3_m, (v4i32) tmp2_m); \
2485 out6 = (v16u8) __msa_ilvod_w((v4i32) tmp3_m, (v4i32) tmp2_m); \
2487 ILVEV_H2_UB(tmp4_m, tmp5_m, tmp6_m, tmp7_m, tmp2_m, tmp3_m); \
2488 out1 = (v16u8) __msa_ilvev_w((v4i32) tmp3_m, (v4i32) tmp2_m); \
2489 out5 = (v16u8) __msa_ilvod_w((v4i32) tmp3_m, (v4i32) tmp2_m); \
2491 tmp2_m = (v16u8) __msa_ilvod_h((v8i16) tmp5_m, (v8i16) tmp4_m); \
2492 tmp3_m = (v16u8) __msa_ilvod_h((v8i16) tmp7_m, (v8i16) tmp6_m); \
2493 out3 = (v16u8) __msa_ilvev_w((v4i32) tmp3_m, (v4i32) tmp2_m); \
2494 out7 = (v16u8) __msa_ilvod_w((v4i32) tmp3_m, (v4i32) tmp2_m); \
2497 /* Description : Transposes 4x4 block with half word elements in vectors
2498 Arguments : Inputs - in0, in1, in2, in3
2499 Outputs - out0, out1, out2, out3
2500 Return Type - signed halfword
2503 #define TRANSPOSE4x4_SH_SH(in0, in1, in2, in3, out0, out1, out2, out3) \
2507 ILVR_H2_SH(in1, in0, in3, in2, s0_m, s1_m); \
2508 ILVRL_W2_SH(s1_m, s0_m, out0, out2); \
2509 out1 = (v8i16) __msa_ilvl_d((v2i64) out0, (v2i64) out0); \
2510 out3 = (v8i16) __msa_ilvl_d((v2i64) out0, (v2i64) out2); \
2513 /* Description : Transposes 8x8 block with half word elements in vectors
2514 Arguments : Inputs - in0, in1, in2, in3, in4, in5, in6, in7
2515 Outputs - out0, out1, out2, out3, out4, out5, out6, out7
2516 Return Type - as per RTYPE
2519 #define TRANSPOSE8x8_H(RTYPE, in0, in1, in2, in3, in4, in5, in6, in7, \
2520 out0, out1, out2, out3, out4, out5, out6, out7) \
2523 v8i16 tmp0_m, tmp1_m, tmp2_m, tmp3_m; \
2524 v8i16 tmp4_m, tmp5_m, tmp6_m, tmp7_m; \
2526 ILVR_H2_SH(in6, in4, in7, in5, s0_m, s1_m); \
2527 ILVRL_H2_SH(s1_m, s0_m, tmp0_m, tmp1_m); \
2528 ILVL_H2_SH(in6, in4, in7, in5, s0_m, s1_m); \
2529 ILVRL_H2_SH(s1_m, s0_m, tmp2_m, tmp3_m); \
2530 ILVR_H2_SH(in2, in0, in3, in1, s0_m, s1_m); \
2531 ILVRL_H2_SH(s1_m, s0_m, tmp4_m, tmp5_m); \
2532 ILVL_H2_SH(in2, in0, in3, in1, s0_m, s1_m); \
2533 ILVRL_H2_SH(s1_m, s0_m, tmp6_m, tmp7_m); \
2534 PCKEV_D4(RTYPE, tmp0_m, tmp4_m, tmp1_m, tmp5_m, tmp2_m, tmp6_m, \
2535 tmp3_m, tmp7_m, out0, out2, out4, out6); \
2536 out1 = (RTYPE) __msa_pckod_d((v2i64) tmp0_m, (v2i64) tmp4_m); \
2537 out3 = (RTYPE) __msa_pckod_d((v2i64) tmp1_m, (v2i64) tmp5_m); \
2538 out5 = (RTYPE) __msa_pckod_d((v2i64) tmp2_m, (v2i64) tmp6_m); \
2539 out7 = (RTYPE) __msa_pckod_d((v2i64) tmp3_m, (v2i64) tmp7_m); \
2541 #define TRANSPOSE8x8_UH_UH(...) TRANSPOSE8x8_H(v8u16, __VA_ARGS__)
2542 #define TRANSPOSE8x8_SH_SH(...) TRANSPOSE8x8_H(v8i16, __VA_ARGS__)
2544 /* Description : Transposes 4x4 block with word elements in vectors
2545 Arguments : Inputs - in0, in1, in2, in3
2546 Outputs - out0, out1, out2, out3
2547 Return Type - signed word
2550 #define TRANSPOSE4x4_SW_SW(in0, in1, in2, in3, out0, out1, out2, out3) \
2552 v4i32 s0_m, s1_m, s2_m, s3_m; \
2554 ILVRL_W2_SW(in1, in0, s0_m, s1_m); \
2555 ILVRL_W2_SW(in3, in2, s2_m, s3_m); \
2557 out0 = (v4i32) __msa_ilvr_d((v2i64) s2_m, (v2i64) s0_m); \
2558 out1 = (v4i32) __msa_ilvl_d((v2i64) s2_m, (v2i64) s0_m); \
2559 out2 = (v4i32) __msa_ilvr_d((v2i64) s3_m, (v2i64) s1_m); \
2560 out3 = (v4i32) __msa_ilvl_d((v2i64) s3_m, (v2i64) s1_m); \
2563 /* Description : Average byte elements from pair of vectors and store 8x4 byte
2564 block in destination memory
2565 Arguments : Inputs - in0, in1, in2, in3, in4, in5, in6, in7, pdst, stride
2566 Details : Each byte element from input vector pair 'in0' and 'in1' are
2567 averaged (a + b)/2 and stored in 'tmp0_m'
2568 Each byte element from input vector pair 'in2' and 'in3' are
2569 averaged (a + b)/2 and stored in 'tmp1_m'
2570 Each byte element from input vector pair 'in4' and 'in5' are
2571 averaged (a + b)/2 and stored in 'tmp2_m'
2572 Each byte element from input vector pair 'in6' and 'in7' are
2573 averaged (a + b)/2 and stored in 'tmp3_m'
2574 The half vector results from all 4 vectors are stored in
2575 destination memory as 8x4 byte block
2577 #define AVE_ST8x4_UB(in0, in1, in2, in3, in4, in5, in6, in7, pdst, stride) \
2579 uint64_t out0_m, out1_m, out2_m, out3_m; \
2580 v16u8 tmp0_m, tmp1_m, tmp2_m, tmp3_m; \
2582 tmp0_m = __msa_ave_u_b((v16u8) in0, (v16u8) in1); \
2583 tmp1_m = __msa_ave_u_b((v16u8) in2, (v16u8) in3); \
2584 tmp2_m = __msa_ave_u_b((v16u8) in4, (v16u8) in5); \
2585 tmp3_m = __msa_ave_u_b((v16u8) in6, (v16u8) in7); \
2587 out0_m = __msa_copy_u_d((v2i64) tmp0_m, 0); \
2588 out1_m = __msa_copy_u_d((v2i64) tmp1_m, 0); \
2589 out2_m = __msa_copy_u_d((v2i64) tmp2_m, 0); \
2590 out3_m = __msa_copy_u_d((v2i64) tmp3_m, 0); \
2591 SD4(out0_m, out1_m, out2_m, out3_m, pdst, stride); \
2594 /* Description : Average byte elements from pair of vectors and store 16x4 byte
2595 block in destination memory
2596 Arguments : Inputs - in0, in1, in2, in3, in4, in5, in6, in7, pdst, stride
2597 Details : Each byte element from input vector pair 'in0' and 'in1' are
2598 averaged (a + b)/2 and stored in 'tmp0_m'
2599 Each byte element from input vector pair 'in2' and 'in3' are
2600 averaged (a + b)/2 and stored in 'tmp1_m'
2601 Each byte element from input vector pair 'in4' and 'in5' are
2602 averaged (a + b)/2 and stored in 'tmp2_m'
2603 Each byte element from input vector pair 'in6' and 'in7' are
2604 averaged (a + b)/2 and stored in 'tmp3_m'
2605 The results from all 4 vectors are stored in destination
2606 memory as 16x4 byte block
2608 #define AVE_ST16x4_UB(in0, in1, in2, in3, in4, in5, in6, in7, pdst, stride) \
2610 v16u8 tmp0_m, tmp1_m, tmp2_m, tmp3_m; \
2612 tmp0_m = __msa_ave_u_b((v16u8) in0, (v16u8) in1); \
2613 tmp1_m = __msa_ave_u_b((v16u8) in2, (v16u8) in3); \
2614 tmp2_m = __msa_ave_u_b((v16u8) in4, (v16u8) in5); \
2615 tmp3_m = __msa_ave_u_b((v16u8) in6, (v16u8) in7); \
2617 ST_UB4(tmp0_m, tmp1_m, tmp2_m, tmp3_m, pdst, stride); \
2620 /* Description : Average rounded byte elements from pair of vectors and store
2621 8x4 byte block in destination memory
2622 Arguments : Inputs - in0, in1, in2, in3, in4, in5, in6, in7, pdst, stride
2623 Details : Each byte element from input vector pair 'in0' and 'in1' are
2624 average rounded (a + b + 1)/2 and stored in 'tmp0_m'
2625 Each byte element from input vector pair 'in2' and 'in3' are
2626 average rounded (a + b + 1)/2 and stored in 'tmp1_m'
2627 Each byte element from input vector pair 'in4' and 'in5' are
2628 average rounded (a + b + 1)/2 and stored in 'tmp2_m'
2629 Each byte element from input vector pair 'in6' and 'in7' are
2630 average rounded (a + b + 1)/2 and stored in 'tmp3_m'
2631 The half vector results from all 4 vectors are stored in
2632 destination memory as 8x4 byte block
2634 #define AVER_ST8x4_UB(in0, in1, in2, in3, in4, in5, in6, in7, pdst, stride) \
2636 uint64_t out0_m, out1_m, out2_m, out3_m; \
2637 v16u8 tp0_m, tp1_m, tp2_m, tp3_m; \
2639 AVER_UB4_UB(in0, in1, in2, in3, in4, in5, in6, in7, \
2640 tp0_m, tp1_m, tp2_m, tp3_m); \
2642 out0_m = __msa_copy_u_d((v2i64) tp0_m, 0); \
2643 out1_m = __msa_copy_u_d((v2i64) tp1_m, 0); \
2644 out2_m = __msa_copy_u_d((v2i64) tp2_m, 0); \
2645 out3_m = __msa_copy_u_d((v2i64) tp3_m, 0); \
2646 SD4(out0_m, out1_m, out2_m, out3_m, pdst, stride); \
2649 /* Description : Average rounded byte elements from pair of vectors and store
2650 16x4 byte block in destination memory
2651 Arguments : Inputs - in0, in1, in2, in3, in4, in5, in6, in7, pdst, stride
2652 Details : Each byte element from input vector pair 'in0' and 'in1' are
2653 average rounded (a + b + 1)/2 and stored in 'tmp0_m'
2654 Each byte element from input vector pair 'in2' and 'in3' are
2655 average rounded (a + b + 1)/2 and stored in 'tmp1_m'
2656 Each byte element from input vector pair 'in4' and 'in5' are
2657 average rounded (a + b + 1)/2 and stored in 'tmp2_m'
2658 Each byte element from input vector pair 'in6' and 'in7' are
2659 average rounded (a + b + 1)/2 and stored in 'tmp3_m'
2660 The vector results from all 4 vectors are stored in
2661 destination memory as 16x4 byte block
2663 #define AVER_ST16x4_UB(in0, in1, in2, in3, in4, in5, in6, in7, pdst, stride) \
2665 v16u8 t0_m, t1_m, t2_m, t3_m; \
2667 AVER_UB4_UB(in0, in1, in2, in3, in4, in5, in6, in7, \
2668 t0_m, t1_m, t2_m, t3_m); \
2669 ST_UB4(t0_m, t1_m, t2_m, t3_m, pdst, stride); \
2672 /* Description : Average rounded byte elements from pair of vectors,
2673 average rounded with destination and store 8x4 byte block
2674 in destination memory
2675 Arguments : Inputs - in0, in1, in2, in3, in4, in5, in6, in7, pdst, stride
2676 Details : Each byte element from input vector pair 'in0' and 'in1' are
2677 average rounded (a + b + 1)/2 and stored in 'tmp0_m'
2678 Each byte element from input vector pair 'in2' and 'in3' are
2679 average rounded (a + b + 1)/2 and stored in 'tmp1_m'
2680 Each byte element from input vector pair 'in4' and 'in5' are
2681 average rounded (a + b + 1)/2 and stored in 'tmp2_m'
2682 Each byte element from input vector pair 'in6' and 'in7' are
2683 average rounded (a + b + 1)/2 and stored in 'tmp3_m'
2684 The half vector results from all 4 vectors are stored in
2685 destination memory as 8x4 byte block
2687 #define AVER_DST_ST8x4_UB(in0, in1, in2, in3, in4, in5, in6, in7, \
2690 v16u8 tmp0_m, tmp1_m, tmp2_m, tmp3_m; \
2691 v16u8 dst0_m, dst1_m, dst2_m, dst3_m; \
2693 LD_UB4(pdst, stride, dst0_m, dst1_m, dst2_m, dst3_m); \
2694 AVER_UB4_UB(in0, in1, in2, in3, in4, in5, in6, in7, \
2695 tmp0_m, tmp1_m, tmp2_m, tmp3_m); \
2696 AVER_ST8x4_UB(dst0_m, tmp0_m, dst1_m, tmp1_m, \
2697 dst2_m, tmp2_m, dst3_m, tmp3_m, pdst, stride); \
2700 /* Description : Average rounded byte elements from pair of vectors,
2701 average rounded with destination and store 16x4 byte block
2702 in destination memory
2703 Arguments : Inputs - in0, in1, in2, in3, in4, in5, in6, in7, pdst, stride
2704 Details : Each byte element from input vector pair 'in0' and 'in1' are
2705 average rounded (a + b + 1)/2 and stored in 'tmp0_m'
2706 Each byte element from input vector pair 'in2' and 'in3' are
2707 average rounded (a + b + 1)/2 and stored in 'tmp1_m'
2708 Each byte element from input vector pair 'in4' and 'in5' are
2709 average rounded (a + b + 1)/2 and stored in 'tmp2_m'
2710 Each byte element from input vector pair 'in6' and 'in7' are
2711 average rounded (a + b + 1)/2 and stored in 'tmp3_m'
2712 The vector results from all 4 vectors are stored in
2713 destination memory as 16x4 byte block
2715 #define AVER_DST_ST16x4_UB(in0, in1, in2, in3, in4, in5, in6, in7, \
2718 v16u8 tmp0_m, tmp1_m, tmp2_m, tmp3_m; \
2719 v16u8 dst0_m, dst1_m, dst2_m, dst3_m; \
2721 LD_UB4(pdst, stride, dst0_m, dst1_m, dst2_m, dst3_m); \
2722 AVER_UB4_UB(in0, in1, in2, in3, in4, in5, in6, in7, \
2723 tmp0_m, tmp1_m, tmp2_m, tmp3_m); \
2724 AVER_ST16x4_UB(dst0_m, tmp0_m, dst1_m, tmp1_m, \
2725 dst2_m, tmp2_m, dst3_m, tmp3_m, pdst, stride); \
2728 /* Description : Add block 4x4
2729 Arguments : Inputs - in0, in1, in2, in3, pdst, stride
2730 Details : Least significant 4 bytes from each input vector are added to
2731 the destination bytes, clipped between 0-255 and then stored.
2733 #define ADDBLK_ST4x4_UB(in0, in1, in2, in3, pdst, stride) \
2735 uint32_t src0_m, src1_m, src2_m, src3_m; \
2736 uint32_t out0_m, out1_m, out2_m, out3_m; \
2737 v8i16 inp0_m, inp1_m, res0_m, res1_m; \
2738 v16i8 dst0_m = { 0 }; \
2739 v16i8 dst1_m = { 0 }; \
2740 v16i8 zero_m = { 0 }; \
2742 ILVR_D2_SH(in1, in0, in3, in2, inp0_m, inp1_m) \
2743 LW4(pdst, stride, src0_m, src1_m, src2_m, src3_m); \
2744 INSERT_W2_SB(src0_m, src1_m, dst0_m); \
2745 INSERT_W2_SB(src2_m, src3_m, dst1_m); \
2746 ILVR_B2_SH(zero_m, dst0_m, zero_m, dst1_m, res0_m, res1_m); \
2747 ADD2(res0_m, inp0_m, res1_m, inp1_m, res0_m, res1_m); \
2748 CLIP_SH2_0_255(res0_m, res1_m); \
2749 PCKEV_B2_SB(res0_m, res0_m, res1_m, res1_m, dst0_m, dst1_m); \
2751 out0_m = __msa_copy_u_w((v4i32) dst0_m, 0); \
2752 out1_m = __msa_copy_u_w((v4i32) dst0_m, 1); \
2753 out2_m = __msa_copy_u_w((v4i32) dst1_m, 0); \
2754 out3_m = __msa_copy_u_w((v4i32) dst1_m, 1); \
2755 SW4(out0_m, out1_m, out2_m, out3_m, pdst, stride); \
2758 /* Description : Dot product and addition of 3 signed halfword input vectors
2759 Arguments : Inputs - in0, in1, in2, coeff0, coeff1, coeff2
2761 Return Type - signed halfword
2762 Details : Dot product of 'in0' with 'coeff0'
2763 Dot product of 'in1' with 'coeff1'
2764 Dot product of 'in2' with 'coeff2'
2765 Addition of all the 3 vector results
2767 out0_m = (in0 * coeff0) + (in1 * coeff1) + (in2 * coeff2)
2769 #define DPADD_SH3_SH(in0, in1, in2, coeff0, coeff1, coeff2) \
2773 out0_m = __msa_dotp_s_h((v16i8) in0, (v16i8) coeff0); \
2774 out0_m = __msa_dpadd_s_h(out0_m, (v16i8) in1, (v16i8) coeff1); \
2775 out0_m = __msa_dpadd_s_h(out0_m, (v16i8) in2, (v16i8) coeff2); \
2780 /* Description : Pack even elements of input vectors & xor with 128
2781 Arguments : Inputs - in0, in1
2783 Return Type - unsigned byte
2784 Details : Signed byte even elements from 'in0' and 'in1' are packed
2785 together in one vector and the resulted vector is xor'ed with
2786 128 to shift the range from signed to unsigned byte
2788 #define PCKEV_XORI128_UB(in0, in1) \
2791 out_m = (v16u8) __msa_pckev_b((v16i8) in1, (v16i8) in0); \
2792 out_m = (v16u8) __msa_xori_b((v16u8) out_m, 128); \
2796 /* Description : Converts inputs to unsigned bytes, interleave, average & store
2797 as 8x4 unsigned byte block
2798 Arguments : Inputs - in0, in1, in2, in3, dst0, dst1, pdst, stride
2800 #define CONVERT_UB_AVG_ST8x4_UB(in0, in1, in2, in3, \
2801 dst0, dst1, pdst, stride) \
2803 v16u8 tmp0_m, tmp1_m; \
2804 uint8_t *pdst_m = (uint8_t *) (pdst); \
2806 tmp0_m = PCKEV_XORI128_UB(in0, in1); \
2807 tmp1_m = PCKEV_XORI128_UB(in2, in3); \
2808 AVER_UB2_UB(tmp0_m, dst0, tmp1_m, dst1, tmp0_m, tmp1_m); \
2809 ST_D4(tmp0_m, tmp1_m, 0, 1, 0, 1, pdst_m, stride); \
2812 /* Description : Pack even byte elements, extract 0 & 2 index words from pair
2813 of results and store 4 words in destination memory as per
2815 Arguments : Inputs - in0, in1, in2, in3, pdst, stride
2817 #define PCKEV_ST4x4_UB(in0, in1, in2, in3, pdst, stride) \
2819 uint32_t out0_m, out1_m, out2_m, out3_m; \
2820 v16i8 tmp0_m, tmp1_m; \
2822 PCKEV_B2_SB(in1, in0, in3, in2, tmp0_m, tmp1_m); \
2824 out0_m = __msa_copy_u_w((v4i32) tmp0_m, 0); \
2825 out1_m = __msa_copy_u_w((v4i32) tmp0_m, 2); \
2826 out2_m = __msa_copy_u_w((v4i32) tmp1_m, 0); \
2827 out3_m = __msa_copy_u_w((v4i32) tmp1_m, 2); \
2829 SW4(out0_m, out1_m, out2_m, out3_m, pdst, stride); \
2832 /* Description : Pack even byte elements and store byte vector in destination
2834 Arguments : Inputs - in0, in1, pdst
2836 #define PCKEV_ST_SB(in0, in1, pdst) \
2839 tmp_m = __msa_pckev_b((v16i8) in1, (v16i8) in0); \
2840 ST_SB(tmp_m, (pdst)); \
2843 /* Description : Horizontal 2 tap filter kernel code
2844 Arguments : Inputs - in0, in1, mask, coeff, shift
2846 #define HORIZ_2TAP_FILT_UH(in0, in1, mask, coeff, shift) \
2851 tmp0_m = __msa_vshf_b((v16i8) mask, (v16i8) in1, (v16i8) in0); \
2852 tmp1_m = __msa_dotp_u_h((v16u8) tmp0_m, (v16u8) coeff); \
2853 tmp1_m = (v8u16) __msa_srari_h((v8i16) tmp1_m, shift); \
2854 tmp1_m = __msa_sat_u_h(tmp1_m, shift); \
2858 #endif /* AVUTIL_MIPS_GENERIC_MACROS_MSA_H */