1 /*
2 * Copyright (c) 2013 The WebM project authors. All Rights Reserved.
3 *
4 * Use of this source code is governed by a BSD-style license
5 * that can be found in the LICENSE file in the root of the source
6 * tree. An additional intellectual property rights grant can be found
7 * in the file PATENTS. All contributing project authors may
8 * be found in the AUTHORS file in the root of the source tree.
9 */
10
11 #include <assert.h>
12 #include <stdio.h>
13
14 #include "./vpx_dsp_rtcd.h"
15 #include "vpx_dsp/mips/convolve_common_dspr2.h"
16 #include "vpx_dsp/vpx_convolve.h"
17 #include "vpx_dsp/vpx_dsp_common.h"
18 #include "vpx_ports/mem.h"
19
20 #if HAVE_DSPR2
convolve_bi_avg_vert_4_dspr2(const uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int16_t * filter_y,int32_t w,int32_t h)21 static void convolve_bi_avg_vert_4_dspr2(const uint8_t *src, int32_t src_stride,
22 uint8_t *dst, int32_t dst_stride,
23 const int16_t *filter_y, int32_t w,
24 int32_t h) {
25 int32_t x, y;
26 const uint8_t *src_ptr;
27 uint8_t *dst_ptr;
28 uint8_t *cm = vpx_ff_cropTbl;
29 uint32_t vector4a = 64;
30 uint32_t load1, load2;
31 uint32_t p1, p2;
32 uint32_t scratch1, scratch2;
33 uint32_t store1, store2;
34 int32_t Temp1, Temp2;
35 const int16_t *filter = &filter_y[3];
36 uint32_t filter45;
37
38 filter45 = ((const int32_t *)filter)[0];
39
40 for (y = h; y--;) {
41 /* prefetch data to cache memory */
42 prefetch_store(dst + dst_stride);
43
44 for (x = 0; x < w; x += 4) {
45 src_ptr = src + x;
46 dst_ptr = dst + x;
47
48 __asm__ __volatile__(
49 "ulw %[load1], 0(%[src_ptr]) \n\t"
50 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
51 "ulw %[load2], 0(%[src_ptr]) \n\t"
52
53 "mtlo %[vector4a], $ac0 \n\t"
54 "mtlo %[vector4a], $ac1 \n\t"
55 "mtlo %[vector4a], $ac2 \n\t"
56 "mtlo %[vector4a], $ac3 \n\t"
57 "mthi $zero, $ac0 \n\t"
58 "mthi $zero, $ac1 \n\t"
59 "mthi $zero, $ac2 \n\t"
60 "mthi $zero, $ac3 \n\t"
61
62 "preceu.ph.qbr %[scratch1], %[load1] \n\t"
63 "preceu.ph.qbr %[p1], %[load2] \n\t"
64 "precrq.ph.w %[p2], %[p1], %[scratch1] \n\t" /* pixel 2 */
65 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
66
67 "dpa.w.ph $ac0, %[p1], %[filter45] \n\t"
68 "dpa.w.ph $ac1, %[p2], %[filter45] \n\t"
69
70 "preceu.ph.qbl %[scratch1], %[load1] \n\t"
71 "preceu.ph.qbl %[p1], %[load2] \n\t"
72 "precrq.ph.w %[p2], %[p1], %[scratch1] \n\t" /* pixel 2 */
73 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
74
75 "dpa.w.ph $ac2, %[p1], %[filter45] \n\t"
76 "dpa.w.ph $ac3, %[p2], %[filter45] \n\t"
77
78 "extp %[Temp1], $ac0, 31 \n\t"
79 "extp %[Temp2], $ac1, 31 \n\t"
80
81 "lbu %[scratch1], 0(%[dst_ptr]) \n\t"
82 "lbu %[scratch2], 1(%[dst_ptr]) \n\t"
83
84 "lbux %[store1], %[Temp1](%[cm]) \n\t"
85 "addqh_r.w %[store1], %[store1], %[scratch1] \n\t" /* pixel 1 */
86 "extp %[Temp1], $ac2, 31 \n\t"
87
88 "lbux %[store2], %[Temp2](%[cm]) \n\t"
89 "addqh_r.w %[store2], %[store2], %[scratch2] \n\t" /* pixel 2 */
90 "extp %[Temp2], $ac3, 31 \n\t"
91 "lbu %[scratch1], 2(%[dst_ptr]) \n\t"
92
93 "sb %[store1], 0(%[dst_ptr]) \n\t"
94 "sb %[store2], 1(%[dst_ptr]) \n\t"
95 "lbu %[scratch2], 3(%[dst_ptr]) \n\t"
96
97 "lbux %[store1], %[Temp1](%[cm]) \n\t"
98 "lbux %[store2], %[Temp2](%[cm]) \n\t"
99 "addqh_r.w %[store1], %[store1], %[scratch1] \n\t" /* pixel 3 */
100 "addqh_r.w %[store2], %[store2], %[scratch2] \n\t" /* pixel 4 */
101
102 "sb %[store1], 2(%[dst_ptr]) \n\t"
103 "sb %[store2], 3(%[dst_ptr]) \n\t"
104
105 : [load1] "=&r"(load1), [load2] "=&r"(load2), [p1] "=&r"(p1),
106 [p2] "=&r"(p2), [scratch1] "=&r"(scratch1),
107 [scratch2] "=&r"(scratch2), [Temp1] "=&r"(Temp1),
108 [Temp2] "=&r"(Temp2), [store1] "=&r"(store1),
109 [store2] "=&r"(store2), [src_ptr] "+r"(src_ptr)
110 : [filter45] "r"(filter45), [vector4a] "r"(vector4a),
111 [src_stride] "r"(src_stride), [cm] "r"(cm), [dst_ptr] "r"(dst_ptr));
112 }
113
114 /* Next row... */
115 src += src_stride;
116 dst += dst_stride;
117 }
118 }
119
convolve_bi_avg_vert_64_dspr2(const uint8_t * src,int32_t src_stride,uint8_t * dst,int32_t dst_stride,const int16_t * filter_y,int32_t h)120 static void convolve_bi_avg_vert_64_dspr2(const uint8_t *src,
121 int32_t src_stride, uint8_t *dst,
122 int32_t dst_stride,
123 const int16_t *filter_y, int32_t h) {
124 int32_t x, y;
125 const uint8_t *src_ptr;
126 uint8_t *dst_ptr;
127 uint8_t *cm = vpx_ff_cropTbl;
128 uint32_t vector4a = 64;
129 uint32_t load1, load2;
130 uint32_t p1, p2;
131 uint32_t scratch1, scratch2;
132 uint32_t store1, store2;
133 int32_t Temp1, Temp2;
134 const int16_t *filter = &filter_y[3];
135 uint32_t filter45;
136
137 filter45 = ((const int32_t *)filter)[0];
138
139 for (y = h; y--;) {
140 /* prefetch data to cache memory */
141 prefetch_store(dst + dst_stride);
142 prefetch_store(dst + dst_stride + 32);
143
144 for (x = 0; x < 64; x += 4) {
145 src_ptr = src + x;
146 dst_ptr = dst + x;
147
148 __asm__ __volatile__(
149 "ulw %[load1], 0(%[src_ptr]) \n\t"
150 "add %[src_ptr], %[src_ptr], %[src_stride] \n\t"
151 "ulw %[load2], 0(%[src_ptr]) \n\t"
152
153 "mtlo %[vector4a], $ac0 \n\t"
154 "mtlo %[vector4a], $ac1 \n\t"
155 "mtlo %[vector4a], $ac2 \n\t"
156 "mtlo %[vector4a], $ac3 \n\t"
157 "mthi $zero, $ac0 \n\t"
158 "mthi $zero, $ac1 \n\t"
159 "mthi $zero, $ac2 \n\t"
160 "mthi $zero, $ac3 \n\t"
161
162 "preceu.ph.qbr %[scratch1], %[load1] \n\t"
163 "preceu.ph.qbr %[p1], %[load2] \n\t"
164 "precrq.ph.w %[p2], %[p1], %[scratch1] \n\t" /* pixel 2 */
165 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
166
167 "dpa.w.ph $ac0, %[p1], %[filter45] \n\t"
168 "dpa.w.ph $ac1, %[p2], %[filter45] \n\t"
169
170 "preceu.ph.qbl %[scratch1], %[load1] \n\t"
171 "preceu.ph.qbl %[p1], %[load2] \n\t"
172 "precrq.ph.w %[p2], %[p1], %[scratch1] \n\t" /* pixel 2 */
173 "append %[p1], %[scratch1], 16 \n\t" /* pixel 1 */
174
175 "dpa.w.ph $ac2, %[p1], %[filter45] \n\t"
176 "dpa.w.ph $ac3, %[p2], %[filter45] \n\t"
177
178 "extp %[Temp1], $ac0, 31 \n\t"
179 "extp %[Temp2], $ac1, 31 \n\t"
180
181 "lbu %[scratch1], 0(%[dst_ptr]) \n\t"
182 "lbu %[scratch2], 1(%[dst_ptr]) \n\t"
183
184 "lbux %[store1], %[Temp1](%[cm]) \n\t"
185 "addqh_r.w %[store1], %[store1], %[scratch1] \n\t" /* pixel 1 */
186 "extp %[Temp1], $ac2, 31 \n\t"
187
188 "lbux %[store2], %[Temp2](%[cm]) \n\t"
189 "addqh_r.w %[store2], %[store2], %[scratch2] \n\t" /* pixel 2 */
190 "extp %[Temp2], $ac3, 31 \n\t"
191 "lbu %[scratch1], 2(%[dst_ptr]) \n\t"
192
193 "sb %[store1], 0(%[dst_ptr]) \n\t"
194 "sb %[store2], 1(%[dst_ptr]) \n\t"
195 "lbu %[scratch2], 3(%[dst_ptr]) \n\t"
196
197 "lbux %[store1], %[Temp1](%[cm]) \n\t"
198 "lbux %[store2], %[Temp2](%[cm]) \n\t"
199 "addqh_r.w %[store1], %[store1], %[scratch1] \n\t" /* pixel 3 */
200 "addqh_r.w %[store2], %[store2], %[scratch2] \n\t" /* pixel 4 */
201
202 "sb %[store1], 2(%[dst_ptr]) \n\t"
203 "sb %[store2], 3(%[dst_ptr]) \n\t"
204
205 : [load1] "=&r"(load1), [load2] "=&r"(load2), [p1] "=&r"(p1),
206 [p2] "=&r"(p2), [scratch1] "=&r"(scratch1),
207 [scratch2] "=&r"(scratch2), [Temp1] "=&r"(Temp1),
208 [Temp2] "=&r"(Temp2), [store1] "=&r"(store1),
209 [store2] "=&r"(store2), [src_ptr] "+r"(src_ptr)
210 : [filter45] "r"(filter45), [vector4a] "r"(vector4a),
211 [src_stride] "r"(src_stride), [cm] "r"(cm), [dst_ptr] "r"(dst_ptr));
212 }
213
214 /* Next row... */
215 src += src_stride;
216 dst += dst_stride;
217 }
218 }
219
vpx_convolve2_avg_vert_dspr2(const uint8_t * src,ptrdiff_t src_stride,uint8_t * dst,ptrdiff_t dst_stride,const InterpKernel * filter,int x0_q4,int32_t x_step_q4,int y0_q4,int y_step_q4,int w,int h)220 void vpx_convolve2_avg_vert_dspr2(const uint8_t *src, ptrdiff_t src_stride,
221 uint8_t *dst, ptrdiff_t dst_stride,
222 const InterpKernel *filter, int x0_q4,
223 int32_t x_step_q4, int y0_q4, int y_step_q4,
224 int w, int h) {
225 const int16_t *const filter_y = filter[y0_q4];
226 uint32_t pos = 38;
227
228 assert(y_step_q4 == 16);
229
230 /* bit positon for extract from acc */
231 __asm__ __volatile__("wrdsp %[pos], 1 \n\t"
232 :
233 : [pos] "r"(pos));
234
235 prefetch_store(dst);
236
237 switch (w) {
238 case 4:
239 case 8:
240 case 16:
241 case 32:
242 convolve_bi_avg_vert_4_dspr2(src, src_stride, dst, dst_stride, filter_y,
243 w, h);
244 break;
245 case 64:
246 prefetch_store(dst + 32);
247 convolve_bi_avg_vert_64_dspr2(src, src_stride, dst, dst_stride, filter_y,
248 h);
249 break;
250 default:
251 vpx_convolve8_avg_vert_c(src, src_stride, dst, dst_stride, filter, x0_q4,
252 x_step_q4, y0_q4, y_step_q4, w, h);
253 break;
254 }
255 }
256 #endif
257