2 * Copyright (c) 2015 Shivraj Patil (Shivraj.Patil@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 #include "libavutil/mips/generic_macros_msa.h"
22 #include "h264dsp_mips.h"
24 static void intra_predict_vert_8x8_msa(uint8_t *src, uint8_t *dst,
28 uint32_t src_data1, src_data2;
31 src_data2 = LW(src + 4);
33 for (row = 8; row--;) {
35 SW(src_data2, (dst + 4));
40 static void intra_predict_vert_16x16_msa(uint8_t *src, uint8_t *dst,
48 for (row = 16; row--;) {
54 static void intra_predict_horiz_8x8_msa(uint8_t *src, int32_t src_stride,
55 uint8_t *dst, int32_t dst_stride)
57 uint64_t out0, out1, out2, out3, out4, out5, out6, out7;
59 out0 = src[0 * src_stride] * 0x0101010101010101;
60 out1 = src[1 * src_stride] * 0x0101010101010101;
61 out2 = src[2 * src_stride] * 0x0101010101010101;
62 out3 = src[3 * src_stride] * 0x0101010101010101;
63 out4 = src[4 * src_stride] * 0x0101010101010101;
64 out5 = src[5 * src_stride] * 0x0101010101010101;
65 out6 = src[6 * src_stride] * 0x0101010101010101;
66 out7 = src[7 * src_stride] * 0x0101010101010101;
68 SD4(out0, out1, out2, out3, dst, dst_stride);
69 dst += (4 * dst_stride);
70 SD4(out4, out5, out6, out7, dst, dst_stride);
73 static void intra_predict_horiz_16x16_msa(uint8_t *src, int32_t src_stride,
74 uint8_t *dst, int32_t dst_stride)
77 uint8_t inp0, inp1, inp2, inp3;
78 v16u8 src0, src1, src2, src3;
80 for (row = 4; row--;) {
90 src0 = (v16u8) __msa_fill_b(inp0);
91 src1 = (v16u8) __msa_fill_b(inp1);
92 src2 = (v16u8) __msa_fill_b(inp2);
93 src3 = (v16u8) __msa_fill_b(inp3);
95 ST_UB4(src0, src1, src2, src3, dst, dst_stride);
96 dst += (4 * dst_stride);
100 static void intra_predict_dc_8x8_msa(uint8_t *src_top, uint8_t *src_left,
101 int32_t src_stride_left,
102 uint8_t *dst, int32_t dst_stride,
103 uint8_t is_above, uint8_t is_left)
106 uint32_t out, addition = 0;
107 v16u8 src_above, store;
112 if (is_left && is_above) {
113 src_above = LD_UB(src_top);
115 sum_above = __msa_hadd_u_h(src_above, src_above);
116 sum_top = __msa_hadd_u_w(sum_above, sum_above);
117 sum = __msa_hadd_u_d(sum_top, sum_top);
118 addition = __msa_copy_u_w((v4i32) sum, 0);
120 for (row = 0; row < 8; row++) {
121 addition += src_left[row * src_stride_left];
124 addition = (addition + 8) >> 4;
125 store = (v16u8) __msa_fill_b(addition);
126 } else if (is_left) {
127 for (row = 0; row < 8; row++) {
128 addition += src_left[row * src_stride_left];
131 addition = (addition + 4) >> 3;
132 store = (v16u8) __msa_fill_b(addition);
133 } else if (is_above) {
134 src_above = LD_UB(src_top);
136 sum_above = __msa_hadd_u_h(src_above, src_above);
137 sum_top = __msa_hadd_u_w(sum_above, sum_above);
138 sum = __msa_hadd_u_d(sum_top, sum_top);
139 sum = (v2u64) __msa_srari_d((v2i64) sum, 3);
140 store = (v16u8) __msa_splati_b((v16i8) sum, 0);
142 store = (v16u8) __msa_ldi_b(128);
145 out = __msa_copy_u_w((v4i32) store, 0);
147 for (row = 8; row--;) {
154 static void intra_predict_dc_16x16_msa(uint8_t *src_top, uint8_t *src_left,
155 int32_t src_stride_left,
156 uint8_t *dst, int32_t dst_stride,
157 uint8_t is_above, uint8_t is_left)
160 uint32_t addition = 0;
161 v16u8 src_above, store;
166 if (is_left && is_above) {
167 src_above = LD_UB(src_top);
169 sum_above = __msa_hadd_u_h(src_above, src_above);
170 sum_top = __msa_hadd_u_w(sum_above, sum_above);
171 sum = __msa_hadd_u_d(sum_top, sum_top);
172 sum_top = (v4u32) __msa_pckev_w((v4i32) sum, (v4i32) sum);
173 sum = __msa_hadd_u_d(sum_top, sum_top);
174 addition = __msa_copy_u_w((v4i32) sum, 0);
176 for (row = 0; row < 16; row++) {
177 addition += src_left[row * src_stride_left];
180 addition = (addition + 16) >> 5;
181 store = (v16u8) __msa_fill_b(addition);
182 } else if (is_left) {
183 for (row = 0; row < 16; row++) {
184 addition += src_left[row * src_stride_left];
187 addition = (addition + 8) >> 4;
188 store = (v16u8) __msa_fill_b(addition);
189 } else if (is_above) {
190 src_above = LD_UB(src_top);
192 sum_above = __msa_hadd_u_h(src_above, src_above);
193 sum_top = __msa_hadd_u_w(sum_above, sum_above);
194 sum = __msa_hadd_u_d(sum_top, sum_top);
195 sum_top = (v4u32) __msa_pckev_w((v4i32) sum, (v4i32) sum);
196 sum = __msa_hadd_u_d(sum_top, sum_top);
197 sum = (v2u64) __msa_srari_d((v2i64) sum, 4);
198 store = (v16u8) __msa_splati_b((v16i8) sum, 0);
200 store = (v16u8) __msa_ldi_b(128);
203 for (row = 16; row--;) {
209 #define INTRA_PREDICT_VALDC_8X8_MSA(val) \
210 static void intra_predict_##val##dc_8x8_msa(uint8_t *dst, \
211 int32_t dst_stride) \
216 store = __msa_ldi_b(val); \
217 out = __msa_copy_u_w((v4i32) store, 0); \
219 for (row = 8; row--;) { \
221 SW(out, (dst + 4)); \
226 INTRA_PREDICT_VALDC_8X8_MSA(127);
227 INTRA_PREDICT_VALDC_8X8_MSA(129);
229 #define INTRA_PREDICT_VALDC_16X16_MSA(val) \
230 static void intra_predict_##val##dc_16x16_msa(uint8_t *dst, \
231 int32_t dst_stride) \
236 store = (v16u8) __msa_ldi_b(val); \
238 for (row = 16; row--;) { \
244 INTRA_PREDICT_VALDC_16X16_MSA(127);
245 INTRA_PREDICT_VALDC_16X16_MSA(129);
247 static void intra_predict_plane_8x8_msa(uint8_t *src, int32_t stride)
250 int32_t res, res0, res1, res2, res3;
252 v16i8 shf_mask = { 3, 5, 2, 6, 1, 7, 0, 8, 3, 5, 2, 6, 1, 7, 0, 8 };
253 v8i16 short_multiplier = { 1, 2, 3, 4, 1, 2, 3, 4 };
254 v4i32 int_multiplier = { 0, 1, 2, 3 };
256 v8i16 vec9, vec10, vec11;
257 v4i32 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, vec8;
260 src_top = LD_UB(src - (stride + 1));
261 src_top = (v16u8) __msa_vshf_b(shf_mask, (v16i8) src_top, (v16i8) src_top);
263 vec9 = __msa_hsub_u_h(src_top, src_top);
264 vec9 *= short_multiplier;
265 vec8 = __msa_hadd_s_w(vec9, vec9);
266 sum = __msa_hadd_s_d(vec8, vec8);
268 res0 = __msa_copy_s_w((v4i32) sum, 0);
270 res1 = (src[4 * stride - 1] - src[2 * stride - 1]) +
271 2 * (src[5 * stride - 1] - src[stride - 1]) +
272 3 * (src[6 * stride - 1] - src[-1]) +
273 4 * (src[7 * stride - 1] - src[-stride - 1]);
277 res0 = (res0 + 16) >> 5;
278 res1 = (res1 + 16) >> 5;
280 res3 = 3 * (res0 + res1);
281 res2 = 16 * (src[7 * stride - 1] + src[-stride + 7] + 1);
284 vec8 = __msa_fill_w(res0);
285 vec4 = __msa_fill_w(res);
286 vec2 = __msa_fill_w(res1);
287 vec5 = vec8 * int_multiplier;
290 for (lpcnt = 4; lpcnt--;) {
299 SRA_4V(vec0, vec1, vec6, vec7, 5);
300 PCKEV_H2_SH(vec1, vec0, vec7, vec6, vec10, vec11);
301 CLIP_SH2_0_255(vec10, vec11);
302 PCKEV_B2_SH(vec10, vec10, vec11, vec11, vec10, vec11);
304 out0 = __msa_copy_s_d((v2i64) vec10, 0);
305 out1 = __msa_copy_s_d((v2i64) vec11, 0);
315 static void intra_predict_plane_16x16_msa(uint8_t *src, int32_t stride)
318 int32_t res0, res1, res2, res3;
319 uint64_t load0, load1;
320 v16i8 shf_mask = { 7, 8, 6, 9, 5, 10, 4, 11, 3, 12, 2, 13, 1, 14, 0, 15 };
321 v8i16 short_multiplier = { 1, 2, 3, 4, 5, 6, 7, 8 };
322 v4i32 int_multiplier = { 0, 1, 2, 3 };
323 v16u8 src_top = { 0 };
325 v4i32 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, vec8, res_add;
327 load0 = LD(src - (stride + 1));
328 load1 = LD(src - (stride + 1) + 9);
330 INSERT_D2_UB(load0, load1, src_top);
332 src_top = (v16u8) __msa_vshf_b(shf_mask, (v16i8) src_top, (v16i8) src_top);
334 vec9 = __msa_hsub_u_h(src_top, src_top);
335 vec9 *= short_multiplier;
336 vec8 = __msa_hadd_s_w(vec9, vec9);
337 res_add = (v4i32) __msa_hadd_s_d(vec8, vec8);
339 res0 = __msa_copy_s_w(res_add, 0) + __msa_copy_s_w(res_add, 2);
341 res1 = (src[8 * stride - 1] - src[6 * stride - 1]) +
342 2 * (src[9 * stride - 1] - src[5 * stride - 1]) +
343 3 * (src[10 * stride - 1] - src[4 * stride - 1]) +
344 4 * (src[11 * stride - 1] - src[3 * stride - 1]) +
345 5 * (src[12 * stride - 1] - src[2 * stride - 1]) +
346 6 * (src[13 * stride - 1] - src[stride - 1]) +
347 7 * (src[14 * stride - 1] - src[-1]) +
348 8 * (src[15 * stride - 1] - src[-1 * stride - 1]);
352 res0 = (res0 + 32) >> 6;
353 res1 = (res1 + 32) >> 6;
355 res3 = 7 * (res0 + res1);
356 res2 = 16 * (src[15 * stride - 1] + src[-stride + 15] + 1);
359 vec8 = __msa_fill_w(res0);
360 vec4 = __msa_fill_w(res2);
361 vec5 = __msa_fill_w(res1);
363 vec7 = vec8 * int_multiplier;
365 for (lpcnt = 16; lpcnt--;) {
372 SRA_4V(vec0, vec1, vec2, vec3, 5);
373 PCKEV_H2_SH(vec1, vec0, vec3, vec2, vec9, vec10);
374 CLIP_SH2_0_255(vec9, vec10);
375 PCKEV_ST_SB(vec9, vec10, src);
382 static void intra_predict_dc_4blk_8x8_msa(uint8_t *src, int32_t stride)
385 uint32_t src0, src1, src3, src2 = 0;
386 uint32_t out0, out1, out2, out3;
391 src_top = LD_UB(src - stride);
392 add = __msa_hadd_u_h((v16u8) src_top, (v16u8) src_top);
393 sum = __msa_hadd_u_w(add, add);
394 src0 = __msa_copy_u_w((v4i32) sum, 0);
395 src1 = __msa_copy_u_w((v4i32) sum, 1);
397 for (lp_cnt = 0; lp_cnt < 4; lp_cnt++) {
398 src0 += src[lp_cnt * stride - 1];
399 src2 += src[(4 + lp_cnt) * stride - 1];
402 src0 = (src0 + 4) >> 3;
403 src3 = (src1 + src2 + 4) >> 3;
404 src1 = (src1 + 2) >> 2;
405 src2 = (src2 + 2) >> 2;
406 out0 = src0 * 0x01010101;
407 out1 = src1 * 0x01010101;
408 out2 = src2 * 0x01010101;
409 out3 = src3 * 0x01010101;
411 for (lp_cnt = 4; lp_cnt--;) {
414 SW(out2, (src + 4 * stride));
415 SW(out3, (src + 4 * stride + 4));
420 static void intra_predict_hor_dc_8x8_msa(uint8_t *src, int32_t stride)
423 uint32_t src0 = 0, src1 = 0;
426 for (lp_cnt = 0; lp_cnt < 4; lp_cnt++) {
427 src0 += src[lp_cnt * stride - 1];
428 src1 += src[(4 + lp_cnt) * stride - 1];
431 src0 = (src0 + 2) >> 2;
432 src1 = (src1 + 2) >> 2;
433 out0 = src0 * 0x0101010101010101;
434 out1 = src1 * 0x0101010101010101;
436 for (lp_cnt = 4; lp_cnt--;) {
438 SD(out1, (src + 4 * stride));
443 static void intra_predict_vert_dc_8x8_msa(uint8_t *src, int32_t stride)
446 uint32_t out0 = 0, out1 = 0;
452 src_top = LD_UB(src - stride);
453 add = __msa_hadd_u_h(src_top, src_top);
454 sum = __msa_hadd_u_w(add, add);
455 sum = (v4u32) __msa_srari_w((v4i32) sum, 2);
456 res0 = (v4i32) __msa_splati_b((v16i8) sum, 0);
457 res1 = (v4i32) __msa_splati_b((v16i8) sum, 4);
458 out0 = __msa_copy_u_w(res0, 0);
459 out1 = __msa_copy_u_w(res1, 0);
461 for (lp_cnt = 8; lp_cnt--;) {
468 static void intra_predict_mad_cow_dc_l0t_8x8_msa(uint8_t *src, int32_t stride)
471 uint32_t src0, src1, src2 = 0;
472 uint32_t out0, out1, out2;
477 src_top = LD_UB(src - stride);
478 add = __msa_hadd_u_h(src_top, src_top);
479 sum = __msa_hadd_u_w(add, add);
480 src0 = __msa_copy_u_w((v4i32) sum, 0);
481 src1 = __msa_copy_u_w((v4i32) sum, 1);
483 for (lp_cnt = 0; lp_cnt < 4; lp_cnt++) {
484 src2 += src[lp_cnt * stride - 1];
486 src2 = (src0 + src2 + 4) >> 3;
487 src0 = (src0 + 2) >> 2;
488 src1 = (src1 + 2) >> 2;
489 out0 = src0 * 0x01010101;
490 out1 = src1 * 0x01010101;
491 out2 = src2 * 0x01010101;
493 for (lp_cnt = 4; lp_cnt--;) {
496 SW(out0, src + stride * 4);
497 SW(out1, src + stride * 4 + 4);
502 static void intra_predict_mad_cow_dc_0lt_8x8_msa(uint8_t *src, int32_t stride)
505 uint32_t src0, src1, src2 = 0, src3;
506 uint32_t out0, out1, out2, out3;
511 src_top = LD_UB(src - stride);
512 add = __msa_hadd_u_h(src_top, src_top);
513 sum = __msa_hadd_u_w(add, add);
514 src0 = __msa_copy_u_w((v4i32) sum, 0);
515 src1 = __msa_copy_u_w((v4i32) sum, 1);
517 for (lp_cnt = 0; lp_cnt < 4; lp_cnt++) {
518 src2 += src[(4 + lp_cnt) * stride - 1];
521 src0 = (src0 + 2) >> 2;
522 src3 = (src1 + src2 + 4) >> 3;
523 src1 = (src1 + 2) >> 2;
524 src2 = (src2 + 2) >> 2;
526 out0 = src0 * 0x01010101;
527 out1 = src1 * 0x01010101;
528 out2 = src2 * 0x01010101;
529 out3 = src3 * 0x01010101;
531 for (lp_cnt = 4; lp_cnt--;) {
534 SW(out2, src + stride * 4);
535 SW(out3, src + stride * 4 + 4);
540 static void intra_predict_mad_cow_dc_l00_8x8_msa(uint8_t *src, int32_t stride)
546 for (lp_cnt = 0; lp_cnt < 4; lp_cnt++) {
547 src0 += src[lp_cnt * stride - 1];
550 src0 = (src0 + 2) >> 2;
551 out0 = src0 * 0x0101010101010101;
552 out1 = 0x8080808080808080;
554 for (lp_cnt = 4; lp_cnt--;) {
556 SD(out1, src + stride * 4);
561 static void intra_predict_mad_cow_dc_0l0_8x8_msa(uint8_t *src, int32_t stride)
567 for (lp_cnt = 0; lp_cnt < 4; lp_cnt++) {
568 src0 += src[(4 + lp_cnt) * stride - 1];
571 src0 = (src0 + 2) >> 2;
573 out0 = 0x8080808080808080;
574 out1 = src0 * 0x0101010101010101;
576 for (lp_cnt = 4; lp_cnt--;) {
578 SD(out1, src + stride * 4);
583 void ff_h264_intra_predict_plane_8x8_msa(uint8_t *src, ptrdiff_t stride)
585 intra_predict_plane_8x8_msa(src, stride);
588 void ff_h264_intra_predict_dc_4blk_8x8_msa(uint8_t *src, ptrdiff_t stride)
590 intra_predict_dc_4blk_8x8_msa(src, stride);
593 void ff_h264_intra_predict_hor_dc_8x8_msa(uint8_t *src, ptrdiff_t stride)
595 intra_predict_hor_dc_8x8_msa(src, stride);
598 void ff_h264_intra_predict_vert_dc_8x8_msa(uint8_t *src, ptrdiff_t stride)
600 intra_predict_vert_dc_8x8_msa(src, stride);
603 void ff_h264_intra_predict_mad_cow_dc_l0t_8x8_msa(uint8_t *src,
606 intra_predict_mad_cow_dc_l0t_8x8_msa(src, stride);
609 void ff_h264_intra_predict_mad_cow_dc_0lt_8x8_msa(uint8_t *src,
612 intra_predict_mad_cow_dc_0lt_8x8_msa(src, stride);
615 void ff_h264_intra_predict_mad_cow_dc_l00_8x8_msa(uint8_t *src,
618 intra_predict_mad_cow_dc_l00_8x8_msa(src, stride);
621 void ff_h264_intra_predict_mad_cow_dc_0l0_8x8_msa(uint8_t *src,
624 intra_predict_mad_cow_dc_0l0_8x8_msa(src, stride);
627 void ff_h264_intra_predict_plane_16x16_msa(uint8_t *src, ptrdiff_t stride)
629 intra_predict_plane_16x16_msa(src, stride);
632 void ff_h264_intra_pred_vert_8x8_msa(uint8_t *src, ptrdiff_t stride)
636 intra_predict_vert_8x8_msa(src - stride, dst, stride);
639 void ff_h264_intra_pred_horiz_8x8_msa(uint8_t *src, ptrdiff_t stride)
643 intra_predict_horiz_8x8_msa(src - 1, stride, dst, stride);
646 void ff_h264_intra_pred_dc_16x16_msa(uint8_t *src, ptrdiff_t stride)
648 uint8_t *src_top = src - stride;
649 uint8_t *src_left = src - 1;
652 intra_predict_dc_16x16_msa(src_top, src_left, stride, dst, stride, 1, 1);
655 void ff_h264_intra_pred_vert_16x16_msa(uint8_t *src, ptrdiff_t stride)
659 intra_predict_vert_16x16_msa(src - stride, dst, stride);
662 void ff_h264_intra_pred_horiz_16x16_msa(uint8_t *src, ptrdiff_t stride)
666 intra_predict_horiz_16x16_msa(src - 1, stride, dst, stride);
669 void ff_h264_intra_pred_dc_left_16x16_msa(uint8_t *src, ptrdiff_t stride)
671 uint8_t *src_top = src - stride;
672 uint8_t *src_left = src - 1;
675 intra_predict_dc_16x16_msa(src_top, src_left, stride, dst, stride, 0, 1);
678 void ff_h264_intra_pred_dc_top_16x16_msa(uint8_t *src, ptrdiff_t stride)
680 uint8_t *src_top = src - stride;
681 uint8_t *src_left = src - 1;
684 intra_predict_dc_16x16_msa(src_top, src_left, stride, dst, stride, 1, 0);
687 void ff_h264_intra_pred_dc_128_8x8_msa(uint8_t *src, ptrdiff_t stride)
689 uint8_t *src_top = src - stride;
690 uint8_t *src_left = src - 1;
693 intra_predict_dc_8x8_msa(src_top, src_left, stride, dst, stride, 0, 0);
696 void ff_h264_intra_pred_dc_128_16x16_msa(uint8_t *src, ptrdiff_t stride)
698 uint8_t *src_top = src - stride;
699 uint8_t *src_left = src - 1;
702 intra_predict_dc_16x16_msa(src_top, src_left, stride, dst, stride, 0, 0);
705 void ff_vp8_pred8x8_127_dc_8_msa(uint8_t *src, ptrdiff_t stride)
707 intra_predict_127dc_8x8_msa(src, stride);
710 void ff_vp8_pred8x8_129_dc_8_msa(uint8_t *src, ptrdiff_t stride)
712 intra_predict_129dc_8x8_msa(src, stride);
715 void ff_vp8_pred16x16_127_dc_8_msa(uint8_t *src, ptrdiff_t stride)
717 intra_predict_127dc_16x16_msa(src, stride);
720 void ff_vp8_pred16x16_129_dc_8_msa(uint8_t *src, ptrdiff_t stride)
722 intra_predict_129dc_16x16_msa(src, stride);