xref: /aosp_15_r20/external/ComputeLibrary/cl_kernels/helpers_asymm.hembed (revision c217d954acce2dbc11938adb493fc0abd69584f3) 
1R"(
2#ifndef ARM_COMPUTE_HELPERS_ASYMM_H
3#define ARM_COMPUTE_HELPERS_ASYMM_H
4
5
6#ifndef ARM_COMPUTE_HELPER_H
7#define ARM_COMPUTE_HELPER_H
8
9
10
11
12#define STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
13    VSTORE(N0)                                                 \
14    (BASENAME##0, 0, (__global DATA_TYPE *)(PTR + 0 * STRIDE_Y + Z##0));
15
16#define STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
17    STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
18    VSTORE(N0)                                                 \
19    (BASENAME##1, 0, (__global DATA_TYPE *)(PTR + 1 * STRIDE_Y + Z##1));
20
21#define STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
22    STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
23    VSTORE(N0)                                                 \
24    (BASENAME##2, 0, (__global DATA_TYPE *)(PTR + 2 * STRIDE_Y + Z##2));
25
26#define STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
27    STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
28    VSTORE(N0)                                                 \
29    (BASENAME##3, 0, (__global DATA_TYPE *)(PTR + 3 * STRIDE_Y + Z##3));
30
31#define STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
32    STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
33    VSTORE(N0)                                                 \
34    (BASENAME##4, 0, (__global DATA_TYPE *)(PTR + 4 * STRIDE_Y + Z##4));
35
36#define STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
37    STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
38    VSTORE(N0)                                                 \
39    (BASENAME##5, 0, (__global DATA_TYPE *)(PTR + 5 * STRIDE_Y + Z##5));
40
41#define STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
42    STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
43    VSTORE(N0)                                                 \
44    (BASENAME##6, 0, (__global DATA_TYPE *)(PTR + 6 * STRIDE_Y + Z##6));
45
46#define STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
47    STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
48    VSTORE(N0)                                                 \
49    (BASENAME##7, 0, (__global DATA_TYPE *)(PTR + 7 * STRIDE_Y + Z##7));
50
51#define STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
52    STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
53    VSTORE(N0)                                                 \
54    (BASENAME##8, 0, (__global DATA_TYPE *)(PTR + 8 * STRIDE_Y + Z##8));
55
56#define STORE_ROW_10(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
57    STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)      \
58    VSTORE(N0)                                                  \
59    (BASENAME##9, 0, (__global DATA_TYPE *)(PTR + 9 * STRIDE_Y + Z##9));
60
61#define STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
62    STORE_ROW_10(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
63    VSTORE(N0)                                                  \
64    (BASENAME##A, 0, (__global DATA_TYPE *)(PTR + 10 * STRIDE_Y + Z##A));
65
66#define STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
67    STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
68    VSTORE(N0)                                                  \
69    (BASENAME##B, 0, (__global DATA_TYPE *)(PTR + 11 * STRIDE_Y + Z##B));
70
71#define STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
72    STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
73    VSTORE(N0)                                                  \
74    (BASENAME##C, 0, (__global DATA_TYPE *)(PTR + 12 * STRIDE_Y + Z##C));
75
76#define STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
77    STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
78    VSTORE(N0)                                                  \
79    (BASENAME##D, 0, (__global DATA_TYPE *)(PTR + 13 * STRIDE_Y + Z##D));
80
81#define STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
82    STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
83    VSTORE(N0)                                                  \
84    (BASENAME##E, 0, (__global DATA_TYPE *)(PTR + 14 * STRIDE_Y + Z##E));
85
86#define STORE_ROW_16(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
87    STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
88    VSTORE(N0)                                                  \
89    (BASENAME##F, 0, (__global DATA_TYPE *)(PTR + 15 * STRIDE_Y + Z##F));
90
91
92
93#define CONVERT_STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
94    VSTORE(N0)                                                         \
95    (CONVERT_SAT((BASENAME##0), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 0 * STRIDE_Y + Z##0));
96
97#define CONVERT_STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
98    CONVERT_STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
99    VSTORE(N0)                                                         \
100    (CONVERT_SAT((BASENAME##1), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 1 * STRIDE_Y + Z##1));
101
102#define CONVERT_STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
103    CONVERT_STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
104    VSTORE(N0)                                                         \
105    (CONVERT_SAT((BASENAME##2), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 2 * STRIDE_Y + Z##2));
106
107#define CONVERT_STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
108    CONVERT_STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
109    VSTORE(N0)                                                         \
110    (CONVERT_SAT((BASENAME##3), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 3 * STRIDE_Y + Z##3));
111
112#define CONVERT_STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
113    CONVERT_STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
114    VSTORE(N0)                                                         \
115    (CONVERT_SAT((BASENAME##4), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 4 * STRIDE_Y + Z##4));
116
117#define CONVERT_STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
118    CONVERT_STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
119    VSTORE(N0)                                                         \
120    (CONVERT_SAT((BASENAME##5), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 5 * STRIDE_Y + Z##5));
121
122#define CONVERT_STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
123    CONVERT_STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
124    VSTORE(N0)                                                         \
125    (CONVERT_SAT((BASENAME##6), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 6 * STRIDE_Y + Z##6));
126
127#define CONVERT_STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
128    CONVERT_STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
129    VSTORE(N0)                                                         \
130    (CONVERT_SAT((BASENAME##7), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 7 * STRIDE_Y + Z##7));
131
132#define CONVERT_STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
133    CONVERT_STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
134    VSTORE(N0)                                                         \
135    (CONVERT_SAT((BASENAME##8), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 8 * STRIDE_Y + Z##8));
136
137#define CONVERT_STORE_ROW_10(N0, DATA, BASENAME, PTR, STRIDE_Y, Z) \
138    CONVERT_STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
139    VSTORE(N0)                                                     \
140    (CONVERT_SAT((BASENAME##9), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 9 * STRIDE_Y + Z##9));
141
142#define CONVERT_STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
143    CONVERT_STORE_ROW_10(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
144    VSTORE(N0)                                                          \
145    (CONVERT_SAT((BASENAME##A), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 10 * STRIDE_Y + Z##A));
146
147#define CONVERT_STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
148    CONVERT_STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
149    VSTORE(N0)                                                          \
150    (CONVERT_SAT((BASENAME##B), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 11 * STRIDE_Y + Z##B));
151
152#define CONVERT_STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
153    CONVERT_STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
154    VSTORE(N0)                                                          \
155    (CONVERT_SAT((BASENAME##C), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 12 * STRIDE_Y + Z##C));
156
157#define CONVERT_STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
158    CONVERT_STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
159    VSTORE(N0)                                                          \
160    (CONVERT_SAT((BASENAME##D), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 13 * STRIDE_Y + Z##D));
161
162#define CONVERT_STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
163    CONVERT_STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
164    VSTORE(N0)                                                          \
165    (CONVERT_SAT((BASENAME##E), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 14 * STRIDE_Y + Z##E));
166
167#define CONVERT_STORE_ROW_16(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
168    CONVERT_STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
169    VSTORE(N0)                                                          \
170    (CONVERT_SAT((BASENAME##F), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 15 * STRIDE_Y + Z##F));
171
172
173
174
175#define STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_ROW_##M0(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
176#define STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
177
178
179
180#define CONVERT_STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) CONVERT_STORE_ROW_##M0(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
181#define CONVERT_STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) CONVERT_STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
182
183
184
185#define STORE_ROW_PARTIAL_1(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
186    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
187    (BASENAME##0, 0, (__global DATA_TYPE *)(PTR + 0 * STRIDE_Y + Z##0));
188
189#define STORE_ROW_PARTIAL_2(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
190    STORE_ROW_PARTIAL_1(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
191    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
192    (BASENAME##1, 0, (__global DATA_TYPE *)(PTR + 1 * STRIDE_Y + Z##1));
193
194#define STORE_ROW_PARTIAL_3(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
195    STORE_ROW_PARTIAL_2(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
196    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
197    (BASENAME##2, 0, (__global DATA_TYPE *)(PTR + 2 * STRIDE_Y + Z##2));
198
199#define STORE_ROW_PARTIAL_4(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
200    STORE_ROW_PARTIAL_3(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
201    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
202    (BASENAME##3, 0, (__global DATA_TYPE *)(PTR + 3 * STRIDE_Y + Z##3));
203
204#define STORE_ROW_PARTIAL_5(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
205    STORE_ROW_PARTIAL_4(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
206    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
207    (BASENAME##4, 0, (__global DATA_TYPE *)(PTR + 4 * STRIDE_Y + Z##4));
208
209#define STORE_ROW_PARTIAL_6(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
210    STORE_ROW_PARTIAL_5(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
211    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
212    (BASENAME##5, 0, (__global DATA_TYPE *)(PTR + 5 * STRIDE_Y + Z##5));
213
214#define STORE_ROW_PARTIAL_7(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
215    STORE_ROW_PARTIAL_6(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
216    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
217    (BASENAME##6, 0, (__global DATA_TYPE *)(PTR + 6 * STRIDE_Y + Z##6));
218
219#define STORE_ROW_PARTIAL_8(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
220    STORE_ROW_PARTIAL_7(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
221    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
222    (BASENAME##7, 0, (__global DATA_TYPE *)(PTR + 7 * STRIDE_Y + Z##7));
223
224#define STORE_ROW_PARTIAL_9(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
225    STORE_ROW_PARTIAL_8(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
226    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
227    (BASENAME##8, 0, (__global DATA_TYPE *)(PTR + 8 * STRIDE_Y + Z##8));
228
229#define STORE_ROW_PARTIAL_10(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
230    STORE_ROW_PARTIAL_9(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)      \
231    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
232    (BASENAME##9, 0, (__global DATA_TYPE *)(PTR + 9 * STRIDE_Y + Z##9));
233
234#define STORE_ROW_PARTIAL_11(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
235    STORE_ROW_PARTIAL_10(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
236    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
237    (BASENAME##A, 0, (__global DATA_TYPE *)(PTR + 10 * STRIDE_Y + Z##A));
238
239#define STORE_ROW_PARTIAL_12(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
240    STORE_ROW_PARTIAL_11(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
241    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
242    (BASENAME##B, 0, (__global DATA_TYPE *)(PTR + 11 * STRIDE_Y + Z##B));
243
244#define STORE_ROW_PARTIAL_13(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
245    STORE_ROW_PARTIAL_12(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
246    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
247    (BASENAME##C, 0, (__global DATA_TYPE *)(PTR + 12 * STRIDE_Y + Z##C));
248
249#define STORE_ROW_PARTIAL_14(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
250    STORE_ROW_PARTIAL_13(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
251    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
252    (BASENAME##D, 0, (__global DATA_TYPE *)(PTR + 13 * STRIDE_Y + Z##D));
253
254#define STORE_ROW_PARTIAL_15(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
255    STORE_ROW_PARTIAL_14(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
256    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
257    (BASENAME##E, 0, (__global DATA_TYPE *)(PTR + 14 * STRIDE_Y + Z##E));
258
259#define STORE_ROW_PARTIAL_16(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
260    STORE_ROW_PARTIAL_15(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
261    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
262    (BASENAME##F, 0, (__global DATA_TYPE *)(PTR + 15 * STRIDE_Y + Z##F));
263
264
265
266#define STORE_BLOCK_PARTIAL_STR(STORE_M0, STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_ROW_PARTIAL_##STORE_M0(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
267#define STORE_BLOCK_PARTIAL(STORE_M0, STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_BLOCK_PARTIAL_STR(STORE_M0, STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
268
269#define STORE_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
270    if(!(PARTIAL_COND_X) && !(PARTIAL_COND_Y))                                                                                                            \
271    {                                                                                                                                                     \
272        STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                                                           \
273    }                                                                                                                                                     \
274    else if((PARTIAL_COND_Y) && !(PARTIAL_COND_X))                                                                                                        \
275    {                                                                                                                                                     \
276        STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                                             \
277    }                                                                                                                                                     \
278    else if(!(PARTIAL_COND_Y) && (PARTIAL_COND_X))                                                                                                        \
279    {                                                                                                                                                     \
280        STORE_BLOCK_PARTIAL(M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                                             \
281    }                                                                                                                                                     \
282    else                                                                                                                                                  \
283    {                                                                                                                                                     \
284        STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                               \
285    }
286
287#define STORE_BLOCK_PARTIAL_IN_X(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_N0, PARTIAL_COND_X) \
288    if(!(PARTIAL_COND_X))                                                                                         \
289    {                                                                                                             \
290        STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                   \
291    }                                                                                                             \
292    else                                                                                                          \
293    {                                                                                                             \
294        STORE_BLOCK_PARTIAL(M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                     \
295    }
296
297#define STORE_BLOCK_PARTIAL_IN_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_COND_Y) \
298    if(!(PARTIAL_COND_Y))                                                                                         \
299    {                                                                                                             \
300        STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                   \
301    }                                                                                                             \
302    else                                                                                                          \
303    {                                                                                                             \
304        STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                     \
305    }
306
307
308#if defined(PARTIAL_STORE_M0) && defined(PARTIAL_STORE_N0)
309
310
311#if PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 == 0
312
313#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
314    STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
315
316#elif PARTIAL_STORE_M0 > 0 && PARTIAL_STORE_N0 == 0
317
318#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
319    STORE_BLOCK_PARTIAL_IN_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_COND_Y)
320
321#elif PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 > 0
322
323#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
324    STORE_BLOCK_PARTIAL_IN_X(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_N0, PARTIAL_COND_X)
325
326#else
327
328#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
329    STORE_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X)
330
331#endif
332
333#endif
334
335
336#if defined(PARTIAL_STORE_M0)
337
338#define COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_M0) \
339    ((uint)(max(0, (int)(y * M0) - (int)((M0 - PARTIAL_STORE_M0) % M0))))
340#else
341#define COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_M0) \
342    ((uint)(y * M0))
343#endif
344
345
346
347#define STORE_VECTOR_SELECT(basename, data_type, ptr, vec_size, leftover, cond) \
348    STORE_BLOCK_PARTIAL_IN_X(1, vec_size, data_type, basename, ptr, 0, 0, leftover, cond)
349
350
351#if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(cl_khr_fp16)
352#pragma OPENCL EXTENSION cl_khr_fp16 : enable
353#endif
354
355#if defined(ARM_COMPUTE_OPENCL_DOT8_ENABLED) && defined(cl_arm_integer_dot_product_int8)
356#pragma OPENCL EXTENSION cl_arm_integer_dot_product_int8 : enable
357#endif
358
359#if defined(ARM_COMPUTE_OPENCL_DOT8_ACC_ENABLED) && defined(cl_arm_integer_dot_product_accumulate_int8)
360#pragma OPENCL EXTENSION cl_arm_integer_dot_product_accumulate_int8 : enable
361#endif
362
363#if defined(ARM_COMPUTE_DEBUG_ENABLED) && defined(cl_arm_printf)
364#pragma OPENCL EXTENSION cl_arm_printf : enable
365#endif
366
367#define GPU_ARCH_MIDGARD 0x100
368#define GPU_ARCH_BIFROST 0x200
369#define GPU_ARCH_VALHALL 0x300
370
371
372#define CONCAT(a, b) a##b
373
374
375#define EXPAND(x) x
376
377
378#define CLAMP(x, min_val, max_val) min(max(x, min_val), max_val)
379
380
381#define REV1(x) ((x))
382#define REV2(x) ((x).s10)
383#define REV3(x) ((x).s210)
384#define REV4(x) ((x).s3210)
385#define REV8(x) ((x).s76543210)
386#define REV16(x) ((x).sFEDCBA9876543210)
387
388
389
390#define REVERSE_STR(x, s) REV##s((x))
391#define REVERSE(x, s) REVERSE_STR(x, s)
392
393
394
395#define ROT1_0(x) ((x))
396#define ROT1_1(x) ((x))
397
398#define ROT2_0(x) ((x))
399#define ROT2_1(x) ((x).s10)
400#define ROT2_2(x) ((x))
401
402#define ROT3_0(x) ((x))
403#define ROT3_1(x) ((x).s201)
404#define ROT3_2(x) ((x).s120)
405#define ROT3_3(x) ((x))
406
407#define ROT4_0(x) ((x))
408#define ROT4_1(x) ((x).s3012)
409#define ROT4_2(x) ((x).s2301)
410#define ROT4_3(x) ((x).s1230)
411#define ROT4_4(x) ((x))
412
413#define ROT8_0(x) ((x))
414#define ROT8_1(x) ((x).s70123456)
415#define ROT8_2(x) ((x).s67012345)
416#define ROT8_3(x) ((x).s56701234)
417#define ROT8_4(x) ((x).s45670123)
418#define ROT8_5(x) ((x).s34567012)
419#define ROT8_6(x) ((x).s23456701)
420#define ROT8_7(x) ((x).s12345670)
421#define ROT8_8(x) ((x))
422
423#define ROT16_0(x) ((x))
424#define ROT16_1(x) ((x).sF0123456789ABCDE)
425#define ROT16_2(x) ((x).sEF0123456789ABCD)
426#define ROT16_3(x) ((x).sDEF0123456789ABC)
427#define ROT16_4(x) ((x).sCDEF0123456789AB)
428#define ROT16_5(x) ((x).sBCDEF0123456789A)
429#define ROT16_6(x) ((x).sABCDEF0123456789)
430#define ROT16_7(x) ((x).s9ABCDEF012345678)
431#define ROT16_8(x) ((x).s89ABCDEF01234567)
432#define ROT16_9(x) ((x).s789ABCDEF0123456)
433#define ROT16_10(x) ((x).s6789ABCDEF012345)
434#define ROT16_11(x) ((x).s56789ABCDEF01234)
435#define ROT16_12(x) ((x).s456789ABCDEF0123)
436#define ROT16_13(x) ((x).s3456789ABCDEF012)
437#define ROT16_14(x) ((x).s23456789ABCDEF01)
438#define ROT16_15(x) ((x).s123456789ABCDEF0)
439#define ROT16_16(x) ((x))
440
441
442
443#define ROTATE_STR(x, s, n) ROT##s##_##n(x)
444#define ROTATE(x, s, n) ROTATE_STR(x, s, n)
445
446
447
448#define V_OFFS1(dt) (dt##1)(0)
449#define V_OFFS2(dt) (dt##2)(0, 1)
450#define V_OFFS3(dt) (dt##3)(0, 1, 2)
451#define V_OFFS4(dt) (dt##4)(0, 1, 2, 3)
452#define V_OFFS8(dt) (dt##8)(0, 1, 2, 3, 4, 5, 6, 7)
453#define V_OFFS16(dt) (dt##16)(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15)
454
455
456
457#define VEC_OFFS_STR(dt, s) V_OFFS##s(dt)
458#define VEC_OFFS(dt, s) VEC_OFFS_STR(dt, s)
459
460
461#define VLOAD_STR(size) vload##size
462#define VLOAD(size) VLOAD_STR(size)
463
464
465#define VLOAD_PARTIAL_STR(size, load_size) vload_partial_##size##_##load_size
466#define VLOAD_PARTIAL(size, load_size) VLOAD_PARTIAL_STR(size, load_size)
467
468#define NO_LOAD(data, offs, ptr) \
469    {                            \
470    }
471
472
473#define vload_partial_1_0 NO_LOAD
474#define vload_partial_1_1 vload1
475#define vload_partial_1_2 NO_LOAD
476#define vload_partial_1_3 NO_LOAD
477#define vload_partial_1_4 NO_LOAD
478#define vload_partial_1_5 NO_LOAD
479#define vload_partial_1_6 NO_LOAD
480#define vload_partial_1_7 NO_LOAD
481#define vload_partial_1_8 NO_LOAD
482#define vload_partial_1_9 NO_LOAD
483#define vload_partial_1_10 NO_LOAD
484#define vload_partial_1_11 NO_LOAD
485#define vload_partial_1_12 NO_LOAD
486#define vload_partial_1_13 NO_LOAD
487#define vload_partial_1_14 NO_LOAD
488#define vload_partial_1_15 NO_LOAD
489#define vload_partial_1_16 NO_LOAD
490
491#define vload_partial_2_0 NO_LOAD
492#define vload_partial_2_1 vload_partial_1
493#define vload_partial_2_2 vload_partial_2
494#define vload_partial_2_3 NO_LOAD
495#define vload_partial_2_4 NO_LOAD
496#define vload_partial_2_5 NO_LOAD
497#define vload_partial_2_6 NO_LOAD
498#define vload_partial_2_7 NO_LOAD
499#define vload_partial_2_8 NO_LOAD
500#define vload_partial_2_9 NO_LOAD
501#define vload_partial_2_10 NO_LOAD
502#define vload_partial_2_11 NO_LOAD
503#define vload_partial_2_12 NO_LOAD
504#define vload_partial_2_13 NO_LOAD
505#define vload_partial_2_14 NO_LOAD
506#define vload_partial_2_15 NO_LOAD
507#define vload_partial_2_16 NO_LOAD
508
509#define vload_partial_3_0 NO_LOAD
510#define vload_partial_3_1 vload_partial_1
511#define vload_partial_3_2 vload_partial_2
512#define vload_partial_3_3 vload_partial_3
513#define vload_partial_3_4 NO_LOAD
514#define vload_partial_3_5 NO_LOAD
515#define vload_partial_3_6 NO_LOAD
516#define vload_partial_3_7 NO_LOAD
517#define vload_partial_3_8 NO_LOAD
518#define vload_partial_3_9 NO_LOAD
519#define vload_partial_3_10 NO_LOAD
520#define vload_partial_3_11 NO_LOAD
521#define vload_partial_3_12 NO_LOAD
522#define vload_partial_3_13 NO_LOAD
523#define vload_partial_3_14 NO_LOAD
524#define vload_partial_3_15 NO_LOAD
525#define vload_partial_3_16 NO_LOAD
526
527#define vload_partial_4_0 NO_LOAD
528#define vload_partial_4_1 vload_partial_1
529#define vload_partial_4_2 vload_partial_2
530#define vload_partial_4_3 vload_partial_3
531#define vload_partial_4_4 vload_partial_4
532#define vload_partial_4_5 NO_LOAD
533#define vload_partial_4_6 NO_LOAD
534#define vload_partial_4_7 NO_LOAD
535#define vload_partial_4_8 NO_LOAD
536#define vload_partial_4_9 NO_LOAD
537#define vload_partial_4_10 NO_LOAD
538#define vload_partial_4_11 NO_LOAD
539#define vload_partial_4_12 NO_LOAD
540#define vload_partial_4_13 NO_LOAD
541#define vload_partial_4_14 NO_LOAD
542#define vload_partial_4_15 NO_LOAD
543#define vload_partial_4_16 NO_LOAD
544
545#define vload_partial_8_0 NO_LOAD
546#define vload_partial_8_1 vload_partial_1
547#define vload_partial_8_2 vload_partial_2
548#define vload_partial_8_3 vload_partial_3
549#define vload_partial_8_4 vload_partial_4
550#define vload_partial_8_5 vload_partial_5
551#define vload_partial_8_6 vload_partial_6
552#define vload_partial_8_7 vload_partial_7
553#define vload_partial_8_8 vload_partial_8
554#define vload_partial_8_9 NO_LOAD
555#define vload_partial_8_10 NO_LOAD
556#define vload_partial_8_11 NO_LOAD
557#define vload_partial_8_12 NO_LOAD
558#define vload_partial_8_13 NO_LOAD
559#define vload_partial_8_14 NO_LOAD
560#define vload_partial_8_15 NO_LOAD
561#define vload_partial_8_16 NO_LOAD
562
563#define vload_partial_16_0 NO_LOAD
564#define vload_partial_16_1 vload_partial_1
565#define vload_partial_16_2 vload_partial_2
566#define vload_partial_16_3 vload_partial_3
567#define vload_partial_16_4 vload_partial_4
568#define vload_partial_16_5 vload_partial_5
569#define vload_partial_16_6 vload_partial_6
570#define vload_partial_16_7 vload_partial_7
571#define vload_partial_16_8 vload_partial_8
572#define vload_partial_16_9 vload_partial_9
573#define vload_partial_16_10 vload_partial_10
574#define vload_partial_16_11 vload_partial_11
575#define vload_partial_16_12 vload_partial_12
576#define vload_partial_16_13 vload_partial_13
577#define vload_partial_16_14 vload_partial_14
578#define vload_partial_16_15 vload_partial_15
579#define vload_partial_16_16 vload_partial_16
580
581
582#define vload_partial_1(DATA, OFFSET, PTR) \
583    DATA.s0 = vload1(OFFSET, PTR);
584
585#define vload_partial_2(DATA, OFFSET, PTR) \
586    DATA.s01 = vload2(OFFSET, PTR);
587
588#define vload_partial_3(DATA, OFFSET, PTR) \
589    DATA.s012 = vload3(OFFSET, PTR);
590
591#define vload_partial_4(DATA, OFFSET, PTR) \
592    DATA.s0123 = vload4(OFFSET, PTR);
593
594#define vload_partial_5(DATA, OFFSET, PTR)    \
595    vload_partial_4(DATA.s0123, OFFSET, PTR); \
596    DATA.s4 = vload1(OFFSET, PTR + 4);
597
598#define vload_partial_6(DATA, OFFSET, PTR)    \
599    vload_partial_4(DATA.s0123, OFFSET, PTR); \
600    vload_partial_2(DATA.s45, OFFSET, PTR + 4);
601
602#define vload_partial_7(DATA, OFFSET, PTR)    \
603    vload_partial_4(DATA.s0123, OFFSET, PTR); \
604    vload_partial_3(DATA.s456, OFFSET, PTR + 4);
605
606#define vload_partial_8(DATA, OFFSET, PTR) \
607    DATA.s01234567 = vload8(OFFSET, PTR);
608
609#define vload_partial_9(DATA, OFFSET, PTR)        \
610    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
611    DATA.s8 = vload1(OFFSET, PTR + 8);
612
613#define vload_partial_10(DATA, OFFSET, PTR)       \
614    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
615    vload_partial_2(DATA.s89, OFFSET, PTR + 8);
616
617#define vload_partial_11(DATA, OFFSET, PTR)       \
618    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
619    vload_partial_3(DATA.s89A, OFFSET, PTR + 8);
620
621#define vload_partial_12(DATA, OFFSET, PTR)       \
622    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
623    vload_partial_4(DATA.s89AB, OFFSET, PTR + 8);
624
625#define vload_partial_13(DATA, OFFSET, PTR)       \
626    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
627    vload_partial_5(DATA.s89ABCDEF, OFFSET, PTR + 8);
628
629#define vload_partial_14(DATA, OFFSET, PTR)       \
630    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
631    vload_partial_6(DATA.s89ABCDEF, OFFSET, PTR + 8);
632
633#define vload_partial_15(DATA, OFFSET, PTR)       \
634    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
635    vload_partial_7(DATA.s89ABCDEF, OFFSET, PTR + 8);
636
637#define vload_partial_16(DATA, OFFSET, PTR) \
638    DATA = vload16(OFFSET, PTR);
639
640
641
642#define PIXEL_UNIT4 1
643#define PIXEL_UNIT8 2
644#define PIXEL_UNIT16 4
645
646
647#define CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT_STR(vec_size) PIXEL_UNIT##vec_size
648#define CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT(vec_size) CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT_STR(vec_size)
649
650
651#define read_image2d_floatx1(img, x_coord, y_coord) (float4)(read_imagef(img, (int2)(x_coord, y_coord)));
652#define read_image2d_floatx2(img, x_coord, y_coord) (float8)(read_imagef(img, (int2)(x_coord, y_coord)), read_imagef(img, (int2)(x_coord + 1, y_coord)));
653#define read_image2d_floatx4(img, x_coord, y_coord) (float16)(read_imagef(img, (int2)(x_coord, y_coord)), read_imagef(img, (int2)(x_coord + 1, y_coord)), read_imagef(img, (int2)(x_coord + 2, y_coord)), read_imagef(img, (int2)(x_coord + 3, y_coord)));
654
655#if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(cl_khr_fp16)
656#define read_image2d_halfx1(img, x_coord, y_coord) (half4)(read_imageh(img, (int2)(x_coord, y_coord)));
657#define read_image2d_halfx2(img, x_coord, y_coord) (half8)(read_imageh(img, (int2)(x_coord, y_coord)), read_imageh(img, (int2)(x_coord + 1, y_coord)));
658#define read_image2d_halfx4(img, x_coord, y_coord) (half16)(read_imageh(img, (int2)(x_coord, y_coord)), read_imageh(img, (int2)(x_coord + 1, y_coord)), read_imageh(img, (int2)(x_coord + 2, y_coord)), read_imageh(img, (int2)(x_coord + 3, y_coord)));
659#endif
660
661#define write_image2d_floatx1(img, x_coord, y_coord, values) (write_imagef(img, (int2)(x_coord, y_coord), values));
662#define write_image2d_floatx2(img, x_coord, y_coord, values) (write_imagef(img, (int2)(x_coord, y_coord), values.s0123), write_imagef(img, (int2)(x_coord + 1, y_coord), values.s4567));
663#define write_image2d_floatx4(img, x_coord, y_coord, values) (write_imagef(img, (int2)(x_coord, y_coord), values.s0123), write_imagef(img, (int2)(x_coord + 1, y_coord), values.s4567), write_imagef(img, (int2)(x_coord + 2, y_coord), values.s89AB), write_imagef(img, (int2)(x_coord + 3, y_coord), values.sCDEF));
664
665#if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(cl_khr_fp16)
666#define write_image2d_halfx1(img, x_coord, y_coord, values) (write_imageh(img, (int2)(x_coord, y_coord), values));
667#define write_image2d_halfx2(img, x_coord, y_coord, values) (write_imageh(img, (int2)(x_coord, y_coord), values.s0123), write_imageh(img, (int2)(x_coord + 1, y_coord), values.s4567));
668#define write_image2d_halfx4(img, x_coord, y_coord, values) (write_imageh(img, (int2)(x_coord, y_coord), values.s0123), write_imageh(img, (int2)(x_coord + 1, y_coord), values.s4567), write_imageh(img, (int2)(x_coord + 2, y_coord), values.s89AB), write_imageh(img, (int2)(x_coord + 3, y_coord), values.sCDEF));
669#endif
670
671
672#define READ_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord) read_image2d_##data_type##x##n0(img, x_coord, y_coord)
673#define READ_IMAGE2D(data_type, n0, img, x_coord, y_coord) READ_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord)
674
675
676#define WRITE_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord, values) write_image2d_##data_type##x##n0(img, x_coord, y_coord, values)
677#define WRITE_IMAGE2D(data_type, n0, img, x_coord, y_coord, values) WRITE_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord, values)
678
679#define VSTORE_STR(size) vstore##size
680#define VSTORE(size) VSTORE_STR(size)
681
682#define float1 float
683#define half1 half
684#define char1 char
685#define uchar1 uchar
686#define short1 short
687#define ushort1 ushort
688#define int1 int
689#define uint1 uint
690#define long1 long
691#define ulong1 ulong
692#define double1 double
693
694#define vload1(OFFSET, PTR) *(OFFSET + PTR)
695#define vstore1(DATA, OFFSET, PTR) *(OFFSET + PTR) = DATA
696
697
698#define VSTORE_PARTIAL_STR(size, store_size) vstore_partial_##size##_##store_size
699#define VSTORE_PARTIAL(size, store_size) VSTORE_PARTIAL_STR(size, store_size)
700
701#define NO_STORE(data, offs, ptr) \
702    {                             \
703    }
704
705
706#define vstore_partial_1_0 NO_STORE
707#define vstore_partial_1_1 vstore1
708#define vstore_partial_1_2 NO_STORE
709#define vstore_partial_1_3 NO_STORE
710#define vstore_partial_1_4 NO_STORE
711#define vstore_partial_1_5 NO_STORE
712#define vstore_partial_1_6 NO_STORE
713#define vstore_partial_1_7 NO_STORE
714#define vstore_partial_1_8 NO_STORE
715#define vstore_partial_1_9 NO_STORE
716#define vstore_partial_1_10 NO_STORE
717#define vstore_partial_1_11 NO_STORE
718#define vstore_partial_1_12 NO_STORE
719#define vstore_partial_1_13 NO_STORE
720#define vstore_partial_1_14 NO_STORE
721#define vstore_partial_1_15 NO_STORE
722#define vstore_partial_1_16 NO_STORE
723
724#define vstore_partial_2_0 NO_STORE
725#define vstore_partial_2_1 vstore_partial_1
726#define vstore_partial_2_2 vstore_partial_2
727#define vstore_partial_2_3 NO_STORE
728#define vstore_partial_2_4 NO_STORE
729#define vstore_partial_2_5 NO_STORE
730#define vstore_partial_2_6 NO_STORE
731#define vstore_partial_2_7 NO_STORE
732#define vstore_partial_2_8 NO_STORE
733#define vstore_partial_2_9 NO_STORE
734#define vstore_partial_2_10 NO_STORE
735#define vstore_partial_2_11 NO_STORE
736#define vstore_partial_2_12 NO_STORE
737#define vstore_partial_2_13 NO_STORE
738#define vstore_partial_2_14 NO_STORE
739#define vstore_partial_2_15 NO_STORE
740#define vstore_partial_2_16 NO_STORE
741
742#define vstore_partial_3_0 NO_STORE
743#define vstore_partial_3_1 vstore_partial_1
744#define vstore_partial_3_2 vstore_partial_2
745#define vstore_partial_3_3 vstore_partial_3
746#define vstore_partial_3_4 NO_STORE
747#define vstore_partial_3_5 NO_STORE
748#define vstore_partial_3_6 NO_STORE
749#define vstore_partial_3_7 NO_STORE
750#define vstore_partial_3_8 NO_STORE
751#define vstore_partial_3_9 NO_STORE
752#define vstore_partial_3_10 NO_STORE
753#define vstore_partial_3_11 NO_STORE
754#define vstore_partial_3_12 NO_STORE
755#define vstore_partial_3_13 NO_STORE
756#define vstore_partial_3_14 NO_STORE
757#define vstore_partial_3_15 NO_STORE
758#define vstore_partial_3_16 NO_STORE
759
760#define vstore_partial_4_0 NO_STORE
761#define vstore_partial_4_1 vstore_partial_1
762#define vstore_partial_4_2 vstore_partial_2
763#define vstore_partial_4_3 vstore_partial_3
764#define vstore_partial_4_4 vstore_partial_4
765#define vstore_partial_4_5 NO_STORE
766#define vstore_partial_4_6 NO_STORE
767#define vstore_partial_4_7 NO_STORE
768#define vstore_partial_4_8 NO_STORE
769#define vstore_partial_4_9 NO_STORE
770#define vstore_partial_4_10 NO_STORE
771#define vstore_partial_4_11 NO_STORE
772#define vstore_partial_4_12 NO_STORE
773#define vstore_partial_4_13 NO_STORE
774#define vstore_partial_4_14 NO_STORE
775#define vstore_partial_4_15 NO_STORE
776#define vstore_partial_4_16 NO_STORE
777
778#define vstore_partial_8_0 NO_STORE
779#define vstore_partial_8_1 vstore_partial_1
780#define vstore_partial_8_2 vstore_partial_2
781#define vstore_partial_8_3 vstore_partial_3
782#define vstore_partial_8_4 vstore_partial_4
783#define vstore_partial_8_5 vstore_partial_5
784#define vstore_partial_8_6 vstore_partial_6
785#define vstore_partial_8_7 vstore_partial_7
786#define vstore_partial_8_8 vstore_partial_8
787#define vstore_partial_8_9 NO_STORE
788#define vstore_partial_8_10 NO_STORE
789#define vstore_partial_8_11 NO_STORE
790#define vstore_partial_8_12 NO_STORE
791#define vstore_partial_8_13 NO_STORE
792#define vstore_partial_8_14 NO_STORE
793#define vstore_partial_8_15 NO_STORE
794#define vstore_partial_8_16 NO_STORE
795
796#define vstore_partial_16_0 NO_STORE
797#define vstore_partial_16_1 vstore_partial_1
798#define vstore_partial_16_2 vstore_partial_2
799#define vstore_partial_16_3 vstore_partial_3
800#define vstore_partial_16_4 vstore_partial_4
801#define vstore_partial_16_5 vstore_partial_5
802#define vstore_partial_16_6 vstore_partial_6
803#define vstore_partial_16_7 vstore_partial_7
804#define vstore_partial_16_8 vstore_partial_8
805#define vstore_partial_16_9 vstore_partial_9
806#define vstore_partial_16_10 vstore_partial_10
807#define vstore_partial_16_11 vstore_partial_11
808#define vstore_partial_16_12 vstore_partial_12
809#define vstore_partial_16_13 vstore_partial_13
810#define vstore_partial_16_14 vstore_partial_14
811#define vstore_partial_16_15 vstore_partial_15
812#define vstore_partial_16_16 vstore_partial_16
813
814
815#define vstore_partial_1(DATA, OFFSET, PTR) \
816    vstore1(DATA.s0, OFFSET, PTR);
817
818#define vstore_partial_2(DATA, OFFSET, PTR) \
819    vstore2(DATA.s01, OFFSET, PTR);
820
821#define vstore_partial_3(DATA, OFFSET, PTR) \
822    vstore3(DATA.s012, OFFSET, PTR);
823
824#define vstore_partial_4(DATA, OFFSET, PTR) \
825    vstore4(DATA.s0123, OFFSET, PTR);
826
827#define vstore_partial_5(DATA, OFFSET, PTR)    \
828    vstore_partial_4(DATA.s0123, OFFSET, PTR); \
829    vstore1(DATA.s4, OFFSET, PTR + 4);
830
831#define vstore_partial_6(DATA, OFFSET, PTR)    \
832    vstore_partial_4(DATA.s0123, OFFSET, PTR); \
833    vstore_partial_2(DATA.s45, OFFSET, PTR + 4);
834
835#define vstore_partial_7(DATA, OFFSET, PTR)    \
836    vstore_partial_4(DATA.s0123, OFFSET, PTR); \
837    vstore_partial_3(DATA.s456, OFFSET, PTR + 4);
838
839#define vstore_partial_8(DATA, OFFSET, PTR) \
840    vstore8(DATA.s01234567, OFFSET, PTR);
841
842#define vstore_partial_9(DATA, OFFSET, PTR)        \
843    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
844    vstore1(DATA.s8, OFFSET, PTR + 8);
845
846#define vstore_partial_10(DATA, OFFSET, PTR)       \
847    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
848    vstore_partial_2(DATA.s89, OFFSET, PTR + 8);
849
850#define vstore_partial_11(DATA, OFFSET, PTR)       \
851    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
852    vstore_partial_3(DATA.s89a, OFFSET, PTR + 8);
853
854#define vstore_partial_12(DATA, OFFSET, PTR)       \
855    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
856    vstore_partial_4(DATA.s89ab, OFFSET, PTR + 8);
857
858#define vstore_partial_13(DATA, OFFSET, PTR)       \
859    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
860    vstore_partial_5(DATA.s89abcdef, OFFSET, PTR + 8);
861
862#define vstore_partial_14(DATA, OFFSET, PTR)       \
863    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
864    vstore_partial_6(DATA.s89abcdef, OFFSET, PTR + 8);
865
866#define vstore_partial_15(DATA, OFFSET, PTR)       \
867    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
868    vstore_partial_7(DATA.s89abcdef, OFFSET, PTR + 8);
869
870#define vstore_partial_16(DATA, OFFSET, PTR) \
871    vstore16(DATA, OFFSET, PTR);
872
873
874
875
876
877#define convert_float_sat convert_float
878#define convert_float1_sat convert_float
879#define convert_float2_sat convert_float2
880#define convert_float3_sat convert_float3
881#define convert_float4_sat convert_float4
882#define convert_float8_sat convert_float8
883#define convert_float16_sat convert_float16
884#define convert_half_sat convert_float
885#define convert_half1_sat convert_half
886#define convert_half2_sat convert_half2
887#define convert_half3_sat convert_half3
888#define convert_half4_sat convert_half4
889#define convert_half8_sat convert_half8
890#define convert_half16_sat convert_half16
891
892#define convert_float1 convert_float
893#define convert_half1 convert_half
894#define convert_char1 convert_char
895#define convert_uchar1 convert_uchar
896#define convert_short1 convert_short
897#define convert_ushort1 convert_ushort
898#define convert_int1 convert_int
899#define convert_uint1 convert_uint
900#define convert_long1 convert_long
901#define convert_ulong1 convert_ulong
902#define convert_double1 convert_double
903
904#define convert_char1_sat convert_char_sat
905#define convert_uchar1_sat convert_uchar_sat
906#define convert_uchar2_sat convert_uchar2_sat
907#define convert_uchar3_sat convert_uchar3_sat
908#define convert_uchar4_sat convert_uchar4_sat
909#define convert_uchar8_sat convert_uchar8_sat
910#define convert_uchar16_sat convert_uchar16_sat
911#define convert_short1_sat convert_short_sat
912#define convert_ushort1_sat convert_ushort_sat
913#define convert_int1_sat convert_int_sat
914#define convert_uint1_sat convert_uint_sat
915#define convert_long1_sat convert_long_sat
916#define convert_ulong1_sat convert_ulong_sat
917#define convert_double1_sat convert_double_sat
918
919#define VEC_DATA_TYPE_STR(type, size) type##size
920#define VEC_DATA_TYPE(type, size) VEC_DATA_TYPE_STR(type, size)
921
922#define CONVERT_STR(x, type) (convert_##type((x)))
923#define CONVERT(x, type) CONVERT_STR(x, type)
924
925#define CONVERT_SAT_STR(x, type) (convert_##type##_sat((x)))
926#define CONVERT_SAT(x, type) CONVERT_SAT_STR(x, type)
927
928#define CONVERT_SAT_ROUND_STR(x, type, round) (convert_##type##_sat_##round((x)))
929#define CONVERT_SAT_ROUND(x, type, round) CONVERT_SAT_ROUND_STR(x, type, round)
930
931#define select_vec_dt_uchar(size) uchar##size
932#define select_vec_dt_char(size) char##size
933#define select_vec_dt_ushort(size) ushort##size
934#define select_vec_dt_short(size) short##size
935#define select_vec_dt_half(size) short##size
936#define select_vec_dt_uint(size) uint##size
937#define select_vec_dt_int(size) int##size
938#define select_vec_dt_float(size) int##size
939#define select_vec_dt_ulong(size) ulong##size
940#define select_vec_dt_long(size) long##size
941
942#define SELECT_VEC_DATA_TYPE_STR(type, size) select_vec_dt_##type(size)
943#define SELECT_VEC_DATA_TYPE(type, size) SELECT_VEC_DATA_TYPE_STR(type, size)
944#define SELECT_DATA_TYPE(type) SELECT_VEC_DATA_TYPE_STR(type, 1)
945
946#define signed_int_vec_dt_uchar(size) char##size
947#define signed_int_vec_dt_char(size) char##size
948#define signed_int_vec_dt_ushort(size) short##size
949#define signed_int_vec_dt_short(size) short##size
950#define signed_int_vec_dt_half(size) short##size
951#define signed_int_vec_dt_uint(size) int##size
952#define signed_int_vec_dt_int(size) int##size
953#define signed_int_vec_dt_float(size) int##size
954#define signed_int_vec_dt_ulong(size) long##size
955#define signed_int_vec_dt_long(size) long##size
956
957#define SIGNED_INT_VEC_DATA_TYPE_STR(type, size) signed_int_vec_dt_##type(size)
958#define SIGNED_INT_VEC_DATA_TYPE(type, size) SIGNED_INT_VEC_DATA_TYPE_STR(type, size)
959#define SIGNED_INT_DATA_TYPE(type) SIGNED_INT_VEC_DATA_TYPE_STR(type, 1)
960
961#define sum_reduce_1(x) (x)
962#define sum_reduce_2(x) ((x).s0) + ((x).s1)
963#define sum_reduce_3(x) sum_reduce_2((x).s01) + ((x).s2)
964#define sum_reduce_4(x) sum_reduce_2((x).s01) + sum_reduce_2((x).s23)
965#define sum_reduce_8(x) sum_reduce_4((x).s0123) + sum_reduce_4((x).s4567)
966#define sum_reduce_16(x) sum_reduce_8((x).s01234567) + sum_reduce_8((x).s89ABCDEF)
967
968#define SUM_REDUCE_STR(x, size) sum_reduce_##size(x)
969#define SUM_REDUCE(x, size) SUM_REDUCE_STR(x, size)
970
971#define prod_reduce_1(x) (x)
972#define prod_reduce_2(x) ((x).s0) * ((x).s1)
973#define prod_reduce_3(x) prod_reduce_2((x).s01) * ((x).s2)
974#define prod_reduce_4(x) prod_reduce_2((x).s01) * prod_reduce_2((x).s23)
975#define prod_reduce_8(x) prod_reduce_4((x).s0123) * prod_reduce_4((x).s4567)
976#define prod_reduce_16(x) prod_reduce_8((x).s01234567) * prod_reduce_8((x).s89ABCDEF)
977
978#define PROD_REDUCE_STR(x, size) prod_reduce_##size(x)
979#define PROD_REDUCE(x, size) PROD_REDUCE_STR(x, size)
980
981#define max_reduce_1(x) (x)
982#define max_reduce_2(x) max(((x).s0), ((x).s1))
983#define max_reduce_3(x) max(max_reduce_2((x).s01), ((x).s2))
984#define max_reduce_4(x) max(max_reduce_2((x).s01), max_reduce_2((x).s23))
985#define max_reduce_8(x) max(max_reduce_4((x).s0123), max_reduce_4((x).s4567))
986#define max_reduce_16(x) max(max_reduce_8((x).s01234567), max_reduce_8((x).s89ABCDEF))
987
988#define MAX_REDUCE_STR(x, size) max_reduce_##size(x)
989#define MAX_REDUCE(x, size) MAX_REDUCE_STR(x, size)
990
991#define VECTOR_DECLARATION(name)     \
992    __global uchar *name##_ptr,      \
993    uint        name##_stride_x, \
994    uint        name##_step_x,   \
995    uint        name##_offset_first_element_in_bytes
996
997#define IMAGE_DECLARATION(name)      \
998    __global uchar *name##_ptr,      \
999    uint        name##_stride_x, \
1000    uint        name##_step_x,   \
1001    uint        name##_stride_y, \
1002    uint        name##_step_y,   \
1003    uint        name##_offset_first_element_in_bytes
1004
1005#define TENSOR3D_DECLARATION(name)   \
1006    __global uchar *name##_ptr,      \
1007    uint        name##_stride_x, \
1008    uint        name##_step_x,   \
1009    uint        name##_stride_y, \
1010    uint        name##_step_y,   \
1011    uint        name##_stride_z, \
1012    uint        name##_step_z,   \
1013    uint        name##_offset_first_element_in_bytes
1014
1015#define TENSOR4D_DECLARATION(name)   \
1016    __global uchar *name##_ptr,      \
1017    uint        name##_stride_x, \
1018    uint        name##_step_x,   \
1019    uint        name##_stride_y, \
1020    uint        name##_step_y,   \
1021    uint        name##_stride_z, \
1022    uint        name##_step_z,   \
1023    uint        name##_stride_w, \
1024    uint        name##_step_w,   \
1025    uint        name##_offset_first_element_in_bytes
1026
1027#define TENSOR5D_DECLARATION(name)   \
1028    __global uchar *name##_ptr,      \
1029    uint        name##_stride_x, \
1030    uint        name##_step_x,   \
1031    uint        name##_stride_y, \
1032    uint        name##_step_y,   \
1033    uint        name##_stride_z, \
1034    uint        name##_step_z,   \
1035    uint        name##_stride_w, \
1036    uint        name##_step_w,   \
1037    uint        name##_stride_v, \
1038    uint        name##_step_v,   \
1039    uint        name##_offset_first_element_in_bytes
1040
1041#define CONVERT_TO_VECTOR_STRUCT(name) \
1042    update_vector_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x)
1043
1044#define CONVERT_TO_VECTOR_STRUCT_NO_STEP(name) \
1045    update_vector_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0)
1046
1047#define CONVERT_TO_IMAGE_STRUCT(name) \
1048    update_image_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y)
1049
1050#define CONVERT_TO_IMAGE_STRUCT_NO_STEP(name) \
1051    update_image_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0)
1052
1053#define CONVERT_TENSOR3D_TO_IMAGE_STRUCT(name) \
1054    update_image_from_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, name##_stride_z, name##_step_z)
1055
1056#define CONVERT_TENSOR3D_TO_IMAGE_STRUCT_NO_STEP(name) \
1057    update_image_from_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0, name##_stride_z, name##_step_z)
1058
1059#define CONVERT_TENSOR3D_TO_IMAGE_STRUCT(name) \
1060    update_image_from_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, name##_stride_z, name##_step_z)
1061
1062#define CONVERT_TO_TENSOR3D_STRUCT(name)                                                                                                           \
1063    update_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, \
1064                                 name##_stride_z, name##_step_z)
1065
1066#define CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(name) \
1067    update_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0, name##_stride_z, 0)
1068
1069#define CONVERT_TO_TENSOR4D_STRUCT(name, mod_size)                                                                                                 \
1070    update_tensor4D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, \
1071                                 name##_stride_z, name##_step_z, name##_stride_w, name##_step_w, mod_size)
1072
1073#define CONVERT_TO_TENSOR4D_STRUCT_NO_STEP(name, mod_size) \
1074    update_tensor4D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0, name##_stride_z, 0, name##_stride_w, 0, mod_size)
1075
1076#define CONVERT_TO_TENSOR3D_STRUCT_NO_UPDATE_PTR(name)                                                                                       \
1077    tensor3D_ptr_no_update(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, \
1078                           name##_stride_z, name##_step_z)
1079
1080
1081typedef struct Vector
1082{
1083    __global uchar *ptr;
1084    int             offset_first_element_in_bytes;
1085    int             stride_x;
1086} Vector;
1087
1088
1089typedef struct Image
1090{
1091    __global uchar *ptr;
1092    int             offset_first_element_in_bytes;
1093    int             stride_x;
1094    int             stride_y;
1095} Image;
1096
1097
1098typedef struct Tensor3D
1099{
1100    __global uchar *ptr;
1101    int             offset_first_element_in_bytes;
1102    int             stride_x;
1103    int             stride_y;
1104    int             stride_z;
1105} Tensor3D;
1106
1107
1108typedef struct Tensor4D
1109{
1110    __global uchar *ptr;
1111    int             offset_first_element_in_bytes;
1112    int             stride_x;
1113    int             stride_y;
1114    int             stride_z;
1115    int             stride_w;
1116} Tensor4D;
1117
1118
1119inline Vector update_vector_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x)
1120{
1121    Vector vector =
1122    {
1123        .ptr                           = ptr,
1124        .offset_first_element_in_bytes = offset_first_element_in_bytes,
1125        .stride_x                      = stride_x,
1126    };
1127    vector.ptr += vector.offset_first_element_in_bytes + get_global_id(0) * step_x;
1128    return vector;
1129}
1130
1131
1132inline Image update_image_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y)
1133{
1134    Image img =
1135    {
1136        .ptr                           = ptr,
1137        .offset_first_element_in_bytes = offset_first_element_in_bytes,
1138        .stride_x                      = stride_x,
1139        .stride_y                      = stride_y
1140    };
1141    img.ptr += img.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y;
1142    return img;
1143}
1144
1145
1146inline Image update_image_from_tensor3D_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z)
1147{
1148    Image img =
1149    {
1150        .ptr                           = ptr,
1151        .offset_first_element_in_bytes = offset_first_element_in_bytes,
1152        .stride_x                      = stride_x,
1153        .stride_y                      = stride_y
1154    };
1155    img.ptr += img.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y + get_global_id(2) * step_z;
1156    return img;
1157}
1158
1159
1160inline Tensor3D update_tensor3D_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z)
1161{
1162    Tensor3D tensor =
1163    {
1164        .ptr                           = ptr,
1165        .offset_first_element_in_bytes = offset_first_element_in_bytes,
1166        .stride_x                      = stride_x,
1167        .stride_y                      = stride_y,
1168        .stride_z                      = stride_z
1169    };
1170    tensor.ptr += tensor.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y + get_global_id(2) * step_z;
1171    return tensor;
1172}
1173
1174
1175inline Tensor3D tensor3D_ptr_no_update(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z)
1176{
1177    Tensor3D tensor =
1178    {
1179        .ptr                           = ptr,
1180        .offset_first_element_in_bytes = offset_first_element_in_bytes,
1181        .stride_x                      = stride_x,
1182        .stride_y                      = stride_y,
1183        .stride_z                      = stride_z
1184    };
1185    return tensor;
1186}
1187
1188inline Tensor4D update_tensor4D_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z, uint stride_w,
1189                                             uint step_w,
1190                                             uint mod_size)
1191{
1192    Tensor4D tensor =
1193    {
1194        .ptr                           = ptr,
1195        .offset_first_element_in_bytes = offset_first_element_in_bytes,
1196        .stride_x                      = stride_x,
1197        .stride_y                      = stride_y,
1198        .stride_z                      = stride_z,
1199        .stride_w                      = stride_w
1200    };
1201
1202    tensor.ptr += tensor.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y + (get_global_id(2) % mod_size) * step_z + (get_global_id(2) / mod_size) * step_w;
1203    return tensor;
1204}
1205
1206
1207inline __global const uchar *vector_offset(const Vector *vec, int x)
1208{
1209    return vec->ptr + x * vec->stride_x;
1210}
1211
1212
1213inline __global uchar *offset(const Image *img, int x, int y)
1214{
1215    return img->ptr + x * img->stride_x + y * img->stride_y;
1216}
1217
1218
1219inline __global const uchar *tensor3D_offset(const Tensor3D *tensor, int x, int y, int z)
1220{
1221    return tensor->ptr + x * tensor->stride_x + y * tensor->stride_y + z * tensor->stride_z;
1222}
1223
1224
1225inline __global const uchar *tensor4D_offset(const Tensor4D *tensor, int x, int y, int z, int w)
1226{
1227    return tensor->ptr + x * tensor->stride_x + y * tensor->stride_y + z * tensor->stride_z + w * tensor->stride_w;
1228}
1229
1230
1231inline __global const uchar *tensor3D_index2ptr(const Tensor3D *tensor, uint width, uint height, uint depth, uint index)
1232{
1233    uint num_elements = width * height;
1234
1235    const uint z = index / num_elements;
1236
1237    index %= num_elements;
1238
1239    const uint y = index / width;
1240
1241    index %= width;
1242
1243    const uint x = index;
1244
1245    return tensor->ptr + x * tensor->stride_x + y * tensor->stride_y + z * tensor->stride_z + tensor->offset_first_element_in_bytes;
1246}
1247
1248#endif
1249
1250
1251#define CONVERT_DOWN_RTE_STR(x, type) (convert_##type##_rte((x)))
1252#define CONVERT_DOWN_RTE(x, type) CONVERT_DOWN_RTE_STR(x, type)
1253
1254
1255inline uchar quantize_qasymm8(float input, float offset, float scale)
1256{
1257    float out_f32 = input / scale + offset;
1258    uchar res_u8  = CONVERT_SAT(CONVERT_DOWN_RTE(out_f32, int), uchar);
1259    return res_u8;
1260}
1261
1262
1263inline float dequantize_qasymm8(uchar input, float offset, float scale)
1264{
1265    return ((float)input - offset) * scale;
1266}
1267
1268
1269inline float dequantize_qasymm8_signed(char input, float offset, float scale)
1270{
1271    return ((float)input - offset) * scale;
1272}
1273
1274
1275#define QUANTIZE_IMPL(type, size)                                                                                       \
1276    inline VEC_DATA_TYPE(type, size) quantize_##type##size(VEC_DATA_TYPE(float, size) input, float offset, float scale) \
1277    {                                                                                                                   \
1278        VEC_DATA_TYPE(float, size)                                                                                      \
1279        out_f32 = input / (VEC_DATA_TYPE(float, size))(scale) + (VEC_DATA_TYPE(float, size))(offset);                   \
1280        VEC_DATA_TYPE(type, size)                                                                                       \
1281        res = CONVERT_SAT(CONVERT_DOWN_RTE(out_f32, VEC_DATA_TYPE(int, size)), VEC_DATA_TYPE(type, size));              \
1282        return res;                                                                                                     \
1283    }
1284
1285
1286#define DEQUANTIZE_IMPL(type, size)                                                                                       \
1287    inline VEC_DATA_TYPE(float, size) dequantize_##type##size(VEC_DATA_TYPE(type, size) input, float offset, float scale) \
1288    {                                                                                                                     \
1289        return (CONVERT(input, VEC_DATA_TYPE(float, size)) - offset) * scale;                                             \
1290    }
1291
1292
1293#define ASYMM_ROUNDING_DIVIDE_BY_POW2_IMPL(size)                                                                                        \
1294    inline VEC_DATA_TYPE(int, size) asymm_rounding_divide_by_POW2_##size(VEC_DATA_TYPE(int, size) x, VEC_DATA_TYPE(int, size) exponent) \
1295    {                                                                                                                                   \
1296        const VEC_DATA_TYPE(int, size)                                                                                                  \
1297        zero = (VEC_DATA_TYPE(int, size))0;                                                                                         \
1298        const VEC_DATA_TYPE(int, size)                                                                                                  \
1299        one = (VEC_DATA_TYPE(int, size))1;                                                                                          \
1300        VEC_DATA_TYPE(int, size)                                                                                                        \
1301        mask = (one << exponent) - one;                                                                                                 \
1302        VEC_DATA_TYPE(int, size)                                                                                                        \
1303        threshold = (mask >> 1) + select(zero, one, (SELECT_VEC_DATA_TYPE(int, size))(x < 0));                                          \
1304        return (x >> exponent) + select(zero, one, (SELECT_VEC_DATA_TYPE(int, size))((x & mask) > threshold));                          \
1305    }
1306
1307
1308#define ASYMM_MULT_IMPL(size)                                                                                \
1309    inline VEC_DATA_TYPE(int, size) asymm_mult##size(VEC_DATA_TYPE(int, size) a, VEC_DATA_TYPE(int, size) b) \
1310    {                                                                                                        \
1311        VEC_DATA_TYPE(int, size)                                                                             \
1312        overflow = a == b && a == INT_MIN;                                                                   \
1313        VEC_DATA_TYPE(long, size)                                                                            \
1314        a_64 = convert_long##size(a);                                                                        \
1315        VEC_DATA_TYPE(long, size)                                                                            \
1316        b_64 = convert_long##size(b);                                                                        \
1317        VEC_DATA_TYPE(long, size)                                                                            \
1318        ab_64 = a_64 * b_64;                                                                                 \
1319                                                                                      \
1320        VEC_DATA_TYPE(long, size)                                                                            \
1321        mask1 = 1 << 30;                                                                                     \
1322        VEC_DATA_TYPE(long, size)                                                                            \
1323        mask2 = 1 - (1 << 30);                                                                               \
1324        VEC_DATA_TYPE(long, size)                                                                            \
1325        is_positive_or_zero = ab_64 >= 0;                                                                    \
1326        VEC_DATA_TYPE(long, size)                                                                            \
1327        nudge = select(mask2, mask1, (SELECT_VEC_DATA_TYPE(long, size))(is_positive_or_zero));               \
1328        VEC_DATA_TYPE(long, size)                                                                            \
1329        mask = 1ll << 31;                                                                                    \
1330        VEC_DATA_TYPE(int, size)                                                                             \
1331        ab_x2_high32 = convert_int##size((ab_64 + nudge) / mask);                                            \
1332        return select(ab_x2_high32, INT_MAX, (SELECT_VEC_DATA_TYPE(int, size))(overflow));                   \
1333    }
1334
1335
1336#define ASYMM_EXP_ON_INTERVAL_BETWEEN_NEGATIVE_ONE_QUARTER_AND_0_EXCL_IMPL(size)                                                    \
1337    inline VEC_DATA_TYPE(int, size) asymm_exp_on_interval_between_negative_one_quarter_and_0_excl##size(VEC_DATA_TYPE(int, size) a) \
1338    {                                                                                                                               \
1339        const VEC_DATA_TYPE(int, size) constant_term     = 1895147668;                                                              \
1340        const VEC_DATA_TYPE(int, size) constant_1_over_3 = 715827883;                                                               \
1341        const int k_fractional_bits = 31;                                                                                           \
1342        VEC_DATA_TYPE(int, size)                                                                                                    \
1343        x = a + (1 << (k_fractional_bits - 3));                                                                                     \
1344        VEC_DATA_TYPE(int, size)                                                                                                    \
1345        x2 = ASYMM_MULT(x, x, size);                                                                                                \
1346        VEC_DATA_TYPE(int, size)                                                                                                    \
1347        x3 = ASYMM_MULT(x2, x, size);                                                                                               \
1348        VEC_DATA_TYPE(int, size)                                                                                                    \
1349        x4 = ASYMM_MULT(x2, x2, size);                                                                                              \
1350        VEC_DATA_TYPE(int, size)                                                                                                    \
1351        x4_over_4 = ASYMM_ROUNDING_DIVIDE_BY_POW2(x4, 2, size);                                                                     \
1352        VEC_DATA_TYPE(int, size)                                                                                                    \
1353        x4_over_24_plus_x3_over_6_plus_x2 = ASYMM_MULT((x4_over_4 + x3), constant_1_over_3, size) + x2;                             \
1354        VEC_DATA_TYPE(int, size)                                                                                                    \
1355        x4_over_24_plus_x3_over_6_plus_x2_over_2 = ASYMM_ROUNDING_DIVIDE_BY_POW2(x4_over_24_plus_x3_over_6_plus_x2, 1, size);       \
1356        return constant_term + ASYMM_MULT(constant_term, x + x4_over_24_plus_x3_over_6_plus_x2_over_2, size);                       \
1357    }
1358
1359
1360#define ASYMM_SELECT_USING_MASK_IMPL(size)                                                                                                                                \
1361    inline VEC_DATA_TYPE(int, size) asymm_select_using_mask##size(VEC_DATA_TYPE(int, size) if_mask, VEC_DATA_TYPE(int, size) then_val, VEC_DATA_TYPE(int, size) else_val) \
1362    {                                                                                                                                                                     \
1363        return (if_mask & then_val) ^ (~if_mask & else_val);                                                                                                              \
1364    }
1365
1366
1367#define ASYMM_MASK_IF_ZERO_IMPL(size)                                                    \
1368    inline VEC_DATA_TYPE(int, size) asymm_mask_if_zero##size(VEC_DATA_TYPE(int, size) a) \
1369    {                                                                                    \
1370        const VEC_DATA_TYPE(int, size) all_zeros = 0;                                    \
1371        const VEC_DATA_TYPE(int, size) all_ones  = ~0;                                   \
1372        return select(all_zeros, all_ones, (SELECT_VEC_DATA_TYPE(int, size))(a == 0));   \
1373    }
1374
1375
1376#define ASYMM_MASK_IF_NON_ZERO_IMPL(size)                                                    \
1377    inline VEC_DATA_TYPE(int, size) asymm_mask_if_non_zero##size(VEC_DATA_TYPE(int, size) a) \
1378    {                                                                                        \
1379        const VEC_DATA_TYPE(int, size) all_zeros = 0;                                        \
1380        const VEC_DATA_TYPE(int, size) all_ones  = ~0;                                       \
1381        return select(all_zeros, all_ones, (SELECT_VEC_DATA_TYPE(int, size))(a != 0));       \
1382    }
1383
1384#define EXP_BARREL_SHIFTER_IMPL(size)                                                                                                                                                                         \
1385    inline VEC_DATA_TYPE(int, size) exp_barrel_shifter##size(VEC_DATA_TYPE(int, size) result, int exponent, int fp_multiplier, int k_integer_bits, int k_fractional_bits, VEC_DATA_TYPE(int, size) remainder) \
1386    {                                                                                                                                                                                                         \
1387        if(k_integer_bits > exponent)                                                                                                                                                                         \
1388        {                                                                                                                                                                                                     \
1389            const int k_shift_amount = k_integer_bits > exponent ? k_fractional_bits + exponent : 0;                                                                                                          \
1390            return ASYMM_SELECT_USING_MASK(                                                                                                                                                                   \
1391                    ASYMM_MASK_IF_NON_ZERO(remainder & (1 << k_shift_amount), size),                                                                                                                              \
1392                    ASYMM_MULT(result, fp_multiplier, size), result, size);                                                                                                                                       \
1393        }                                                                                                                                                                                                     \
1394        \
1395        return result;                                                                                                                                                                                        \
1396    }
1397
1398
1399#define ASYMM_EXP_ON_NEGATIVE_VALUES_IMPL(size)                                                                               \
1400    inline VEC_DATA_TYPE(int, size) asymm_exp_on_negative_values##size(VEC_DATA_TYPE(int, size) a, int k_integer_bits)        \
1401    {                                                                                                                         \
1402        const int k_fractional_bits = 31 - k_integer_bits;                                                                    \
1403        VEC_DATA_TYPE(int, size)                                                                                              \
1404        k_one_quarter = 1 << (k_fractional_bits - 2);                                                                         \
1405        VEC_DATA_TYPE(int, size)                                                                                              \
1406        mask = k_one_quarter - 1;                                                                                             \
1407        VEC_DATA_TYPE(int, size)                                                                                              \
1408        a_mod_quarter_minus_one_quarter = (a & mask) - k_one_quarter;                                                         \
1409        VEC_DATA_TYPE(int, size)                                                                                              \
1410        a_mod_quarter_minus_one_quarter_scaled = a_mod_quarter_minus_one_quarter << k_integer_bits;                           \
1411        VEC_DATA_TYPE(int, size)                                                                                              \
1412        result = ASYMM_EXP_ON_INTERVAL_BETWEEN_NEGATIVE_ONE_QUARTER_AND_0_EXCL(a_mod_quarter_minus_one_quarter_scaled, size); \
1413        VEC_DATA_TYPE(int, size)                                                                                              \
1414        remainder = a_mod_quarter_minus_one_quarter - a;                                                                      \
1415        \
1416        result = EXP_BARREL_SHIFTER(result, -2, 1672461947, k_integer_bits, k_fractional_bits, remainder, size);              \
1417        result = EXP_BARREL_SHIFTER(result, -1, 1302514674, k_integer_bits, k_fractional_bits, remainder, size);              \
1418        result = EXP_BARREL_SHIFTER(result, +0, 790015084, k_integer_bits, k_fractional_bits, remainder, size);               \
1419        result = EXP_BARREL_SHIFTER(result, +1, 290630308, k_integer_bits, k_fractional_bits, remainder, size);               \
1420        result = EXP_BARREL_SHIFTER(result, +2, 39332535, k_integer_bits, k_fractional_bits, remainder, size);                \
1421        result = EXP_BARREL_SHIFTER(result, +3, 720401, k_integer_bits, k_fractional_bits, remainder, size);                  \
1422        result = EXP_BARREL_SHIFTER(result, +4, 242, k_integer_bits, k_fractional_bits, remainder, size);                     \
1423        \
1424        if(k_integer_bits > 5)                                                                                                \
1425        {                                                                                                                     \
1426            const VEC_DATA_TYPE(int, size) clamp = -(1 << (k_fractional_bits + 5));                                           \
1427            result = ASYMM_SELECT_USING_MASK(ASYMM_MASK_IF_NON_ZERO(a < clamp, size), 0, result, size);                       \
1428        }                                                                                                                     \
1429        \
1430        const VEC_DATA_TYPE(int, size) Q0_one = INT_MAX;                                                                      \
1431        return ASYMM_SELECT_USING_MASK(ASYMM_MASK_IF_ZERO(a, size), Q0_one, result, size);                                    \
1432    }
1433
1434
1435#define ASYMM_SATURATING_ROUNDING_MULT_BY_POW2_IMPL(size)                                                                  \
1436    inline VEC_DATA_TYPE(int, size) asymm_saturating_rounding_mult_by_pow2##size(VEC_DATA_TYPE(int, size) x, int exponent) \
1437    {                                                                                                                      \
1438        if(exponent < 0)                                                                                                   \
1439        {                                                                                                                  \
1440            return ASYMM_ROUNDING_DIVIDE_BY_POW2(x, -exponent, size);                                                      \
1441        }                                                                                                                  \
1442        \
1443        const VEC_DATA_TYPE(int, size) min = INT_MIN;                                                                      \
1444        const VEC_DATA_TYPE(int, size) max = INT_MAX;                                                                      \
1445        int threshold = ((1 << (31 - exponent)) - 1);                                                                      \
1446        VEC_DATA_TYPE(int, size)                                                                                           \
1447        positive_mask = ASYMM_MASK_IF_NON_ZERO(x > threshold, size);                                                       \
1448        VEC_DATA_TYPE(int, size)                                                                                           \
1449        negative_mask = ASYMM_MASK_IF_NON_ZERO(x < -threshold, size);                                                      \
1450        VEC_DATA_TYPE(int, size)                                                                                           \
1451        result = x << exponent;                                                                                            \
1452        result = ASYMM_SELECT_USING_MASK(positive_mask, max, result, size);                                                \
1453        result = ASYMM_SELECT_USING_MASK(negative_mask, min, result, size);                                                \
1454        return result;                                                                                                     \
1455    }
1456
1457
1458#define ASYMM_ROUNDING_HALF_SUM_IMPL(size)                                                                                \
1459    inline VEC_DATA_TYPE(int, size) asymm_rounding_half_sum##size(VEC_DATA_TYPE(int, size) a, VEC_DATA_TYPE(int, size) b) \
1460    {                                                                                                                     \
1461        VEC_DATA_TYPE(long, size)                                                                                         \
1462        a64 = convert_long##size(a);                                                                                      \
1463        VEC_DATA_TYPE(long, size)                                                                                         \
1464        b64 = convert_long##size(b);                                                                                      \
1465        VEC_DATA_TYPE(long, size)                                                                                         \
1466        sum = a64 + b64;                                                                                                  \
1467        const VEC_DATA_TYPE(long, size) one       = 1;                                                                    \
1468        const VEC_DATA_TYPE(long, size) minus_one = -1;                                                                   \
1469        VEC_DATA_TYPE(long, size)                                                                                         \
1470        sign = select(minus_one, one, (SELECT_VEC_DATA_TYPE(long, size))(sum >= 0));                                      \
1471        return convert_int##size((sum + sign) / 2);                                                                       \
1472    }
1473
1474
1475#define ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1_IMPL(size)                                                    \
1476    inline VEC_DATA_TYPE(int, size) asymm_one_over_one_plus_x_for_x_in_0_1##size(VEC_DATA_TYPE(int, size) a) \
1477    {                                                                                                        \
1478        const VEC_DATA_TYPE(int, size) Q0_one = INT_MAX;                                                     \
1479        const VEC_DATA_TYPE(int, size) Q2_one = 1 << (31 - 2);                                               \
1480        VEC_DATA_TYPE(int, size)                                                                             \
1481        half_denominator = ASYMM_ROUNDING_HALF_SUM(a, Q0_one, size);                                         \
1482        const VEC_DATA_TYPE(int, size) Q2_48_over_17     = 1515870810;                                       \
1483        const VEC_DATA_TYPE(int, size) Q2_neg_32_over_17 = -1010580540;                                      \
1484        VEC_DATA_TYPE(int, size)                                                                             \
1485        x = Q2_48_over_17 + ASYMM_MULT(half_denominator, Q2_neg_32_over_17, size);                           \
1486        for(int i = 0; i < 3; i++)                                                                           \
1487        {                                                                                                    \
1488            VEC_DATA_TYPE(int, size)                                                                         \
1489            half_denominator_times_x = ASYMM_MULT(half_denominator, x, size);                                \
1490            VEC_DATA_TYPE(int, size)                                                                         \
1491            one_minus_half_denominator_times_x = Q2_one - half_denominator_times_x;                          \
1492            VEC_DATA_TYPE(int, size)                                                                         \
1493            tmp = ASYMM_MULT(x, one_minus_half_denominator_times_x, size);                                   \
1494            x   = x + ASYMM_SATURATING_ROUNDING_MULT_BY_POW2(tmp, 2, size);                                  \
1495        }                                                                                                    \
1496        return ASYMM_SATURATING_ROUNDING_MULT_BY_POW2(x, 1, size);                                           \
1497    }
1498
1499
1500#define ASYMM_RESCALE_IMPL(size)                                                                                                    \
1501    inline VEC_DATA_TYPE(int, size) asymm_rescale##size(VEC_DATA_TYPE(int, size) value, int src_integer_bits, int dst_integer_bits) \
1502    {                                                                                                                               \
1503        int exponent = src_integer_bits - dst_integer_bits;                                                                         \
1504        return ASYMM_SATURATING_ROUNDING_MULT_BY_POW2(value, exponent, size);                                                       \
1505    }
1506
1507#define QUANTIZE_STR(input, offset, scale, type, size) quantize_##type##size(input, offset, scale)
1508#define QUANTIZE(input, offset, scale, type, size) QUANTIZE_STR(input, offset, scale, type, size)
1509#define DEQUANTIZE_STR(input, offset, scale, type, size) dequantize_##type##size(input, offset, scale)
1510#define DEQUANTIZE(input, offset, scale, type, size) DEQUANTIZE_STR(input, offset, scale, type, size)
1511
1512#define ASYMM_ROUNDING_DIVIDE_BY_POW2_STR(x, exponent, size) asymm_rounding_divide_by_POW2_##size(x, exponent)
1513#define ASYMM_ROUNDING_DIVIDE_BY_POW2(x, exponent, size) ASYMM_ROUNDING_DIVIDE_BY_POW2_STR(x, exponent, size)
1514#define ASYMM_MULT_STR(a, b, size) asymm_mult##size(a, b)
1515#define ASYMM_MULT(a, b, size) ASYMM_MULT_STR(a, b, size)
1516#define ASYMM_MULT_BY_QUANT_MULTIPLIER_GREATER_THAN_ONE(x, quantized_multiplier, left_shift, size) \
1517    ASYMM_MULT(x *((VEC_DATA_TYPE(int, size))(1) << (-left_shift)), quantized_multiplier, size)
1518#define ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE(x, quantized_multiplier, right_shift, size) \
1519    ASYMM_ROUNDING_DIVIDE_BY_POW2(ASYMM_MULT(x, quantized_multiplier, size), right_shift, size)
1520#define ASYMM_EXP_ON_INTERVAL_BETWEEN_NEGATIVE_ONE_QUARTER_AND_0_EXCL(a, size) asymm_exp_on_interval_between_negative_one_quarter_and_0_excl##size(a)
1521#define ASYMM_SELECT_USING_MASK(if_mask, then_val, else_val, size) asymm_select_using_mask##size(if_mask, then_val, else_val)
1522#define ASYMM_MASK_IF_ZERO(a, size) asymm_mask_if_zero##size(a)
1523#define ASYMM_MASK_IF_NON_ZERO(a, size) asymm_mask_if_non_zero##size(a)
1524#define EXP_BARREL_SHIFTER(result, exponent, fp_multiplier, k_integer_bits, k_fractional_bits, remainder, size) exp_barrel_shifter##size(result, exponent, fp_multiplier, k_integer_bits, k_fractional_bits, remainder)
1525#define ASYMM_EXP_ON_NEGATIVE_VALUES_STR(a, k_integer_bits, size) asymm_exp_on_negative_values##size(a, k_integer_bits)
1526#define ASYMM_EXP_ON_NEGATIVE_VALUES(a, k_integer_bits, size) ASYMM_EXP_ON_NEGATIVE_VALUES_STR(a, k_integer_bits, size)
1527#define ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1_STR(a, size) asymm_one_over_one_plus_x_for_x_in_0_1##size(a)
1528#define ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1(a, size) ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1_STR(a, size)
1529#define ASYMM_SATURATING_ROUNDING_MULT_BY_POW2(x, exponent, size) asymm_saturating_rounding_mult_by_pow2##size(x, exponent)
1530#define ASYMM_ROUNDING_HALF_SUM(a, b, size) asymm_rounding_half_sum##size(a, b)
1531#define ASYMM_RESCALE_STR(value, src_integer_bits, dst_integer_bits, size) asymm_rescale##size(value, src_integer_bits, dst_integer_bits)
1532#define ASYMM_RESCALE(value, src_integer_bits, dst_integer_bits, size) ASYMM_RESCALE_STR(value, src_integer_bits, dst_integer_bits, size)
1533
1534#define MULTIPLY_BY_QUANTIZED_MULTIPLIER_IMPL(size)                                                                             \
1535    inline VEC_DATA_TYPE(int, size) multiply_by_quantized_multiplier##size(VEC_DATA_TYPE(int, size) input, int qmul, int shift) \
1536    {                                                                                                                           \
1537        const int left_shift  = shift > 0 ? shift : 0;                                                                          \
1538        const int right_shift = shift > 0 ? 0 : -shift;                                                                         \
1539        return ASYMM_ROUNDING_DIVIDE_BY_POW2(ASYMM_MULT(input * (1 << left_shift), qmul, size), right_shift, size);             \
1540    }
1541#define MULTIPLY_BY_QUANTIZED_MULTIPLIER(input, qmul, shift, size) multiply_by_quantized_multiplier##size(input, qmul, shift)
1542
1543QUANTIZE_IMPL(uchar, 1)
1544QUANTIZE_IMPL(char, 1)
1545QUANTIZE_IMPL(uint, 1)
1546QUANTIZE_IMPL(int, 1)
1547QUANTIZE_IMPL(uchar, 2)
1548QUANTIZE_IMPL(char, 2)
1549QUANTIZE_IMPL(uint, 2)
1550QUANTIZE_IMPL(int, 2)
1551QUANTIZE_IMPL(uchar, 3)
1552QUANTIZE_IMPL(char, 3)
1553QUANTIZE_IMPL(uint, 3)
1554QUANTIZE_IMPL(int, 3)
1555QUANTIZE_IMPL(uchar, 4)
1556QUANTIZE_IMPL(ushort, 4)
1557QUANTIZE_IMPL(short, 4)
1558QUANTIZE_IMPL(int, 4)
1559QUANTIZE_IMPL(uchar, 8)
1560QUANTIZE_IMPL(char, 8)
1561QUANTIZE_IMPL(uint, 8)
1562QUANTIZE_IMPL(int, 8)
1563QUANTIZE_IMPL(uchar, 16)
1564QUANTIZE_IMPL(char, 16)
1565QUANTIZE_IMPL(ushort, 16)
1566QUANTIZE_IMPL(short, 16)
1567QUANTIZE_IMPL(uint, 16)
1568QUANTIZE_IMPL(int, 16)
1569
1570DEQUANTIZE_IMPL(uchar, 1)
1571DEQUANTIZE_IMPL(char, 1)
1572DEQUANTIZE_IMPL(uint, 1)
1573DEQUANTIZE_IMPL(int, 1)
1574DEQUANTIZE_IMPL(uchar, 2)
1575DEQUANTIZE_IMPL(char, 2)
1576DEQUANTIZE_IMPL(uint, 2)
1577DEQUANTIZE_IMPL(int, 2)
1578DEQUANTIZE_IMPL(uchar, 3)
1579DEQUANTIZE_IMPL(char, 3)
1580DEQUANTIZE_IMPL(uint, 3)
1581DEQUANTIZE_IMPL(int, 3)
1582DEQUANTIZE_IMPL(uchar, 4)
1583DEQUANTIZE_IMPL(ushort, 4)
1584DEQUANTIZE_IMPL(short, 4)
1585DEQUANTIZE_IMPL(int, 4)
1586DEQUANTIZE_IMPL(uchar, 8)
1587DEQUANTIZE_IMPL(char, 8)
1588DEQUANTIZE_IMPL(uint, 8)
1589DEQUANTIZE_IMPL(int, 8)
1590DEQUANTIZE_IMPL(uchar, 16)
1591DEQUANTIZE_IMPL(char, 16)
1592DEQUANTIZE_IMPL(ushort, 16)
1593DEQUANTIZE_IMPL(short, 16)
1594DEQUANTIZE_IMPL(uint, 16)
1595DEQUANTIZE_IMPL(int, 16)
1596
1597ASYMM_ROUNDING_DIVIDE_BY_POW2_IMPL(1)
1598ASYMM_ROUNDING_DIVIDE_BY_POW2_IMPL(2)
1599ASYMM_ROUNDING_DIVIDE_BY_POW2_IMPL(3)
1600ASYMM_ROUNDING_DIVIDE_BY_POW2_IMPL(4)
1601ASYMM_ROUNDING_DIVIDE_BY_POW2_IMPL(8)
1602ASYMM_ROUNDING_DIVIDE_BY_POW2_IMPL(16)
1603
1604ASYMM_MULT_IMPL(1)
1605ASYMM_MULT_IMPL(2)
1606ASYMM_MULT_IMPL(3)
1607ASYMM_MULT_IMPL(4)
1608ASYMM_MULT_IMPL(8)
1609ASYMM_MULT_IMPL(16)
1610
1611ASYMM_EXP_ON_INTERVAL_BETWEEN_NEGATIVE_ONE_QUARTER_AND_0_EXCL_IMPL(1)
1612ASYMM_EXP_ON_INTERVAL_BETWEEN_NEGATIVE_ONE_QUARTER_AND_0_EXCL_IMPL(2)
1613ASYMM_EXP_ON_INTERVAL_BETWEEN_NEGATIVE_ONE_QUARTER_AND_0_EXCL_IMPL(3)
1614ASYMM_EXP_ON_INTERVAL_BETWEEN_NEGATIVE_ONE_QUARTER_AND_0_EXCL_IMPL(4)
1615ASYMM_EXP_ON_INTERVAL_BETWEEN_NEGATIVE_ONE_QUARTER_AND_0_EXCL_IMPL(8)
1616ASYMM_EXP_ON_INTERVAL_BETWEEN_NEGATIVE_ONE_QUARTER_AND_0_EXCL_IMPL(16)
1617
1618ASYMM_SELECT_USING_MASK_IMPL(1)
1619ASYMM_SELECT_USING_MASK_IMPL(2)
1620ASYMM_SELECT_USING_MASK_IMPL(3)
1621ASYMM_SELECT_USING_MASK_IMPL(4)
1622ASYMM_SELECT_USING_MASK_IMPL(8)
1623ASYMM_SELECT_USING_MASK_IMPL(16)
1624
1625ASYMM_MASK_IF_ZERO_IMPL(1)
1626ASYMM_MASK_IF_ZERO_IMPL(2)
1627ASYMM_MASK_IF_ZERO_IMPL(3)
1628ASYMM_MASK_IF_ZERO_IMPL(4)
1629ASYMM_MASK_IF_ZERO_IMPL(8)
1630ASYMM_MASK_IF_ZERO_IMPL(16)
1631
1632ASYMM_MASK_IF_NON_ZERO_IMPL(1)
1633ASYMM_MASK_IF_NON_ZERO_IMPL(2)
1634ASYMM_MASK_IF_NON_ZERO_IMPL(3)
1635ASYMM_MASK_IF_NON_ZERO_IMPL(4)
1636ASYMM_MASK_IF_NON_ZERO_IMPL(8)
1637ASYMM_MASK_IF_NON_ZERO_IMPL(16)
1638
1639EXP_BARREL_SHIFTER_IMPL(1)
1640EXP_BARREL_SHIFTER_IMPL(2)
1641EXP_BARREL_SHIFTER_IMPL(3)
1642EXP_BARREL_SHIFTER_IMPL(4)
1643EXP_BARREL_SHIFTER_IMPL(8)
1644EXP_BARREL_SHIFTER_IMPL(16)
1645
1646ASYMM_EXP_ON_NEGATIVE_VALUES_IMPL(1)
1647ASYMM_EXP_ON_NEGATIVE_VALUES_IMPL(2)
1648ASYMM_EXP_ON_NEGATIVE_VALUES_IMPL(3)
1649ASYMM_EXP_ON_NEGATIVE_VALUES_IMPL(4)
1650ASYMM_EXP_ON_NEGATIVE_VALUES_IMPL(8)
1651ASYMM_EXP_ON_NEGATIVE_VALUES_IMPL(16)
1652
1653ASYMM_SATURATING_ROUNDING_MULT_BY_POW2_IMPL(1)
1654ASYMM_SATURATING_ROUNDING_MULT_BY_POW2_IMPL(2)
1655ASYMM_SATURATING_ROUNDING_MULT_BY_POW2_IMPL(3)
1656ASYMM_SATURATING_ROUNDING_MULT_BY_POW2_IMPL(4)
1657ASYMM_SATURATING_ROUNDING_MULT_BY_POW2_IMPL(8)
1658ASYMM_SATURATING_ROUNDING_MULT_BY_POW2_IMPL(16)
1659
1660ASYMM_ROUNDING_HALF_SUM_IMPL(1)
1661ASYMM_ROUNDING_HALF_SUM_IMPL(2)
1662ASYMM_ROUNDING_HALF_SUM_IMPL(3)
1663ASYMM_ROUNDING_HALF_SUM_IMPL(4)
1664ASYMM_ROUNDING_HALF_SUM_IMPL(8)
1665ASYMM_ROUNDING_HALF_SUM_IMPL(16)
1666
1667ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1_IMPL(1)
1668ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1_IMPL(2)
1669ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1_IMPL(3)
1670ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1_IMPL(4)
1671ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1_IMPL(8)
1672ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1_IMPL(16)
1673
1674ASYMM_RESCALE_IMPL(1)
1675ASYMM_RESCALE_IMPL(2)
1676ASYMM_RESCALE_IMPL(3)
1677ASYMM_RESCALE_IMPL(4)
1678ASYMM_RESCALE_IMPL(8)
1679ASYMM_RESCALE_IMPL(16)
1680
1681MULTIPLY_BY_QUANTIZED_MULTIPLIER_IMPL(1)
1682MULTIPLY_BY_QUANTIZED_MULTIPLIER_IMPL(2)
1683MULTIPLY_BY_QUANTIZED_MULTIPLIER_IMPL(3)
1684MULTIPLY_BY_QUANTIZED_MULTIPLIER_IMPL(4)
1685MULTIPLY_BY_QUANTIZED_MULTIPLIER_IMPL(8)
1686MULTIPLY_BY_QUANTIZED_MULTIPLIER_IMPL(16)
1687
1688#endif  )"