1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
|
/*
* Copyright (c) 2022-2023 Arm Limited.
*
* SPDX-License-Identifier: MIT
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to
* deal in the Software without restriction, including without limitation the
* rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
* sell copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*/
#include <cstddef>
#include <cstdint>
#if defined(ARM_COMPUTE_ENABLE_SME2)
namespace arm_conv {
namespace depthwise {
void sme2_fp32_nhwc_3x3_s1_output2x2_mla_depthfirst_direct_impl(
const unsigned int n_tile_rows,
const unsigned int n_tile_cols,
const float *inptr,
int64_t ld_input_row,
int64_t ld_input_col,
float *outptr,
int64_t ld_output_row,
int64_t ld_output_col,
const void *params,
unsigned int n_channels,
const float activation_min,
const float activation_max
)
{
struct Args
{
const uint64_t n_tile_rows, n_tile_cols;
const float *inptr;
const uint64_t ld_input_row;
const uint64_t ld_input_col;
float *outptr;
const uint64_t ld_output_row;
const uint64_t ld_output_col;
const void *params;
const float min, max;
uint64_t tile_i = 0, tile_j = 0;
Args(
const unsigned int n_tile_rows,
const unsigned int n_tile_cols,
const float *inptr,
int64_t ld_input_row,
int64_t ld_input_col,
float *outptr,
int64_t ld_output_row,
int64_t ld_output_col,
const void *params,
const float activation_min,
const float activation_max
) : n_tile_rows(n_tile_rows), n_tile_cols(n_tile_cols), inptr(inptr),
ld_input_row(ld_input_row), ld_input_col(ld_input_col), outptr(outptr),
ld_output_row(ld_output_row), ld_output_col(ld_output_col),
params(params), min(activation_min), max(activation_max)
{
}
};
Args params_struct(
n_tile_rows, n_tile_cols,
inptr, ld_input_row, ld_input_col,
outptr, ld_output_row, ld_output_col,
params, activation_min, activation_max
);
__asm__ __volatile__(
".inst 0xd503477f // SMSTART ZA\n"
"ptrue p3.b\n"
".inst 0x25207810 // ptrue pn8.b\n"
"mov x4, #0x0\n"
"mov x5, #0x0\n"
"1:" // Tile loop
"str x4, [%x[params_struct], %[offsetof_args_tile_i]]\n"
"mov x22, #0x2\n"
"str x5, [%x[params_struct], %[offsetof_args_tile_j]]\n"
"ldr x21, [%x[params_struct], %[offsetof_args_ld_input_row]]\n"
"mul x20, x4, x21\n" // offset = tile_i * ld_input_row
"ldr x6, [%x[params_struct], %[offsetof_args_ld_input_col]]\n"
"madd x20, x5, x6, x20\n" // offset += tile_j * ld_input_col
"mul x20, x20, x22\n" // offset *= kernel_stride * output_size
"ldr x7, [%x[params_struct], %[offsetof_args_inptr]]\n"
"add x7, x7, x20, LSL #2\n" // inptr[0] += offset * sizeof(float)
"add x8, x7, x21, LSL #2\n"
"add x17, x8, x21, LSL #2\n"
"add x16, x6, x6\n"
"ldr x15, [%x[params_struct], %[offsetof_args_params]]\n"
"add x14, x17, x21, LSL #2\n"
"add x13, x16, x6\n"
"cbnz x5, 2f\n"
"ldr x20, [%x[params_struct], %[offsetof_args_n_tile_cols]]\n"
"sub x21, x20, x5\n"
"sub x21, x21, #0x1\n"
"lsl x12, %x[n_channels], #0x2\n"
"mov x20, #0x8\n"
"and x21, x21, #0x3fffff\n"
"mul x20, x20, x6\n"
"orr x12, x12, x21, LSL #22\n"
"orr x12, x12, x20, LSL #38\n"
"add x11, x8, x6, LSL #2\n"
"add x10, x7, x13, LSL #2\n"
"add x9, x8, x16, LSL #2\n"
"add x28, x17, x6, LSL #2\n"
"add x27, x14, x13, LSL #2\n"
"add x26, x7, x6, LSL #2\n"
"add x25, x7, x16, LSL #2\n"
"add x24, x17, x16, LSL #2\n"
"add x23, x8, x13, LSL #2\n"
"add x22, x17, x13, LSL #2\n"
"add x21, x14, x6, LSL #2\n"
"add x20, x14, x16, LSL #2\n"
".inst 0xf8ac497a // rprfm pldonce, x12, [x11]\n"
".inst 0xf8ac48fa // rprfm pldonce, x12, [x7]\n"
".inst 0xf8ac495a // rprfm pldonce, x12, [x10]\n"
".inst 0xf8ac493a // rprfm pldonce, x12, [x9]\n"
".inst 0xf8ac4b9a // rprfm pldonce, x12, [x28]\n"
".inst 0xf8ac49da // rprfm pldonce, x12, [x14]\n"
".inst 0xf8ac4b7a // rprfm pldonce, x12, [x27]\n"
".inst 0xf8ac4b5a // rprfm pldonce, x12, [x26]\n"
".inst 0xf8ac4b3a // rprfm pldonce, x12, [x25]\n"
".inst 0xf8ac4b1a // rprfm pldonce, x12, [x24]\n"
".inst 0xf8ac491a // rprfm pldonce, x12, [x8]\n"
".inst 0xf8ac4afa // rprfm pldonce, x12, [x23]\n"
".inst 0xf8ac4a3a // rprfm pldonce, x12, [x17]\n"
".inst 0xf8ac4ada // rprfm pldonce, x12, [x22]\n"
".inst 0xf8ac4aba // rprfm pldonce, x12, [x21]\n"
".inst 0xf8ac4a9a // rprfm pldonce, x12, [x20]\n"
"2:" // Tile loop: Prefetch input rows: End
"ldr x22, [%x[params_struct], %[offsetof_args_ld_output_row]]\n"
"mul x21, x4, x22\n" // offset = tile_i * ld_output_row
"mov x20, #0x2\n"
"ld1w { z22.s }, p3/Z, [x15]\n"
"ldr x25, [%x[params_struct], %[offsetof_args_ld_output_col]]\n"
"madd x21, x5, x25, x21\n" // offset += tile_j * ld_output_col
"addvl x15, x15, #1\n"
".inst 0xa040c1e0 // ld1w { z0.s-z3.s }, pn8.b/Z, [x15]\n"
"ldr x24, [%x[params_struct], %[offsetof_args_outptr]]\n"
"mul x21, x21, x20\n" // offset *= output_tile_size
"cntw x23\n"
"ld1rw { z21.s }, p3/Z, [%x[params_struct], %[offsetof_args_min]]\n"
"addvl x15, x15, #4\n"
"add x24, x24, x21, LSL #2\n" // outptrs[0] += offset * sizeof(float)
".inst 0xa040c1e4 // ld1w { z4.s-z7.s }, pn8.b/Z, [x15]\n"
"whilelt p2.s, XZR, %x[n_channels]\n"
"addvl x15, x15, #4\n"
"ld1rw { z14.s }, p3/Z, [%x[params_struct], %[offsetof_args_max]]\n"
"cmp x23, %x[n_channels]\n"
"add x22, x24, x22, LSL #2\n"
"ld1w { z8.s }, p3/Z, [x15]\n"
"mov x21, #0x0\n"
"sub x20, XZR, x23\n"
"ld1w { z9.s }, p2/Z, [x8, x6, LSL #2]\n"
"ld1w { z10.s }, p2/Z, [x7]\n"
"addvl x15, x15, #1\n"
"ld1w { z11.s }, p2/Z, [x7, x13, LSL #2]\n"
"ld1w { z12.s }, p2/Z, [x8, x16, LSL #2]\n"
"ld1w { z13.s }, p2/Z, [x17, x6, LSL #2]\n"
"bge 4f\n"
"3:" // Tile loop: Channel loop
"movprfx z28, z22\n fmla z28.s, p3/M, z4.s, z9.s\n"
"movprfx z29, z22\n fmla z29.s, p3/M, z3.s, z9.s\n"
"whilelt p1.s, x23, %x[n_channels]\n"
"incw x21\n"
"movprfx z30, z22\n fmla z30.s, p3/M, z1.s, z9.s\n"
"movprfx z31, z22\n fmla z31.s, p3/M, z0.s, z9.s\n"
"ld1w { z18.s }, p2/Z, [x14]\n"
"incw x23\n"
"fmla z28.s, p3/M, z0.s, z10.s\n"
"fmla z29.s, p3/M, z2.s, z11.s\n"
"ld1w { z17.s }, p2/Z, [x14, x13, LSL #2]\n"
"mov p0.b, p2.b\n"
"fmla z30.s, p3/M, z2.s, z12.s\n"
"fmla z31.s, p3/M, z1.s, z12.s\n"
"ld1w { z16.s }, p2/Z, [x17, x16, LSL #2]\n"
"incw x20\n"
"fmla z28.s, p3/M, z5.s, z12.s\n"
"fmla z29.s, p3/M, z4.s, z12.s\n"
"ld1w { z11.s }, p2/Z, [x7, x6, LSL #2]\n"
"fmla z30.s, p3/M, z6.s, z18.s\n"
"fmla z31.s, p3/M, z3.s, z13.s\n"
"ld1w { z10.s }, p2/Z, [x7, x16, LSL #2]\n"
"addvl x7, x7, #1\n"
"fmla z28.s, p3/M, z7.s, z13.s\n"
"fmla z29.s, p3/M, z6.s, z13.s\n"
"ld1w { z22.s }, p3/Z, [x15]\n"
"addvl x15, x15, #1\n"
"fmla z30.s, p3/M, z4.s, z13.s\n"
"fmla z31.s, p3/M, z8.s, z17.s\n"
"ld1w { z9.s }, p2/Z, [x8]\n"
"fmla z28.s, p3/M, z1.s, z11.s\n"
"fmla z29.s, p3/M, z0.s, z11.s\n"
"ld1w { z19.s }, p2/Z, [x8, x13, LSL #2]\n"
"addvl x8, x8, #1\n"
"fmla z30.s, p3/M, z5.s, z16.s\n"
"fmla z31.s, p3/M, z4.s, z16.s\n"
"fmla z28.s, p3/M, z2.s, z10.s\n"
"fmla z29.s, p3/M, z1.s, z10.s\n"
"ld1w { z18.s }, p2/Z, [x17]\n"
"fmla z30.s, p3/M, z0.s, z9.s\n"
"fmla z31.s, p3/M, z2.s, z19.s\n"
"fmla z28.s, p3/M, z8.s, z16.s\n"
"fmla z29.s, p3/M, z7.s, z16.s\n"
"ld1w { z17.s }, p2/Z, [x17, x13, LSL #2]\n"
"addvl x17, x17, #1\n"
"fmla z30.s, p3/M, z3.s, z18.s\n"
"fmla z31.s, p3/M, z5.s, z17.s\n"
"ld1w { z13.s }, p1/Z, [x17, x6, LSL #2]\n"
"fmla z28.s, p3/M, z3.s, z9.s\n"
"ld1w { z16.s }, p2/Z, [x14, x6, LSL #2]\n"
"fmla z29.s, p3/M, z5.s, z19.s\n"
"fmla z30.s, p3/M, z7.s, z16.s\n"
"fmla z31.s, p3/M, z6.s, z16.s\n"
"ld1w { z16.s }, p2/Z, [x14, x16, LSL #2]\n"
"whilelt p2.s, x21, %x[n_channels]\n"
"fmla z28.s, p3/M, z6.s, z18.s\n"
"fmla z29.s, p3/M, z8.s, z17.s\n"
".inst 0xa040c1e0 // ld1w { z0.s-z3.s }, pn8.b/Z, [x15]\n"
"addvl x15, x15, #4\n"
"fmla z30.s, p3/M, z8.s, z16.s\n"
"fmla z31.s, p3/M, z7.s, z16.s\n"
".inst 0xa040c1e4 // ld1w { z4.s-z7.s }, pn8.b/Z, [x15]\n"
"addvl x15, x15, #4\n"
"cmp x23, %x[n_channels]\n"
".inst 0xc1aecabc // fclamp { z28.s-z31.s }, z21.s, z14.s\n"
"addvl x14, x14, #1\n"
"ld1w { z9.s }, p1/Z, [x8, x6, LSL #2]\n"
"ld1w { z10.s }, p1/Z, [x7]\n"
"st1w { z28.s }, p0, [x24]\n"
"ld1w { z11.s }, p1/Z, [x7, x13, LSL #2]\n"
"st1w { z29.s }, p0, [x24, x25, LSL #2]\n"
"addvl x24, x24, #1\n"
"ld1w { z12.s }, p1/Z, [x8, x16, LSL #2]\n"
"st1w { z30.s }, p0, [x22]\n"
"st1w { z31.s }, p0, [x22, x25, LSL #2]\n"
"addvl x22, x22, #1\n"
"ld1w { z8.s }, p3/Z, [x15]\n"
"addvl x15, x15, #1\n"
"blt 3b\n"
"4:" // Tile loop: Channel tail
"movprfx z24, z22\n fmla z24.s, p3/M, z4.s, z9.s\n"
"movprfx z25, z22\n fmla z25.s, p3/M, z3.s, z9.s\n"
"ldr x5, [%x[params_struct], %[offsetof_args_tile_j]]\n"
"add x5, x5, #0x1\n"
"movprfx z26, z22\n fmla z26.s, p3/M, z1.s, z9.s\n"
"movprfx z27, z22\n fmla z27.s, p3/M, z0.s, z9.s\n"
"ld1w { z17.s }, p2/Z, [x14]\n"
"ldr x4, [%x[params_struct], %[offsetof_args_tile_i]]\n"
"fmla z24.s, p3/M, z0.s, z10.s\n"
"fmla z25.s, p3/M, z2.s, z11.s\n"
"ld1w { z16.s }, p2/Z, [x14, x13, LSL #2]\n"
"ldr x20, [%x[params_struct], %[offsetof_args_n_tile_cols]]\n"
"fmla z26.s, p3/M, z2.s, z12.s\n"
"fmla z27.s, p3/M, z1.s, z12.s\n"
"ld1w { z20.s }, p2/Z, [x17, x16, LSL #2]\n"
"ldr x21, [%x[params_struct], %[offsetof_args_n_tile_rows]]\n"
"fmla z24.s, p3/M, z5.s, z12.s\n"
"fmla z25.s, p3/M, z4.s, z12.s\n"
"ld1w { z18.s }, p2/Z, [x7, x6, LSL #2]\n"
"cmp x5, x20\n"
"fmla z26.s, p3/M, z6.s, z17.s\n"
"fmla z27.s, p3/M, z3.s, z13.s\n"
"ld1w { z17.s }, p2/Z, [x7, x16, LSL #2]\n"
"add x20, x4, #0x1\n"
"fmla z24.s, p3/M, z7.s, z13.s\n"
"fmla z25.s, p3/M, z6.s, z13.s\n"
"csel x4, x4, x20, LT\n"
"mov p0.b, p2.b\n"
"fmla z26.s, p3/M, z4.s, z13.s\n"
"fmla z27.s, p3/M, z8.s, z16.s\n"
"ld1w { z16.s }, p2/Z, [x8]\n"
"csel x5, x5, XZR, LT\n"
"fmla z24.s, p3/M, z1.s, z18.s\n"
"fmla z25.s, p3/M, z0.s, z18.s\n"
"ld1w { z19.s }, p2/Z, [x8, x13, LSL #2]\n"
"cmp x4, x21\n"
"fmla z26.s, p3/M, z5.s, z20.s\n"
"fmla z27.s, p3/M, z4.s, z20.s\n"
"fmla z24.s, p3/M, z2.s, z17.s\n"
"fmla z25.s, p3/M, z1.s, z17.s\n"
"ld1w { z18.s }, p2/Z, [x17]\n"
"fmla z26.s, p3/M, z0.s, z16.s\n"
"fmla z27.s, p3/M, z2.s, z19.s\n"
"fmla z24.s, p3/M, z8.s, z20.s\n"
"fmla z25.s, p3/M, z7.s, z20.s\n"
"ld1w { z17.s }, p2/Z, [x17, x13, LSL #2]\n"
"fmla z26.s, p3/M, z3.s, z18.s\n"
"fmla z27.s, p3/M, z5.s, z17.s\n"
"fmla z24.s, p3/M, z3.s, z16.s\n"
"ld1w { z16.s }, p2/Z, [x14, x6, LSL #2]\n"
"fmla z25.s, p3/M, z5.s, z19.s\n"
"fmla z26.s, p3/M, z7.s, z16.s\n"
"fmla z27.s, p3/M, z6.s, z16.s\n"
"ld1w { z16.s }, p2/Z, [x14, x16, LSL #2]\n"
"fmla z24.s, p3/M, z6.s, z18.s\n"
"fmla z25.s, p3/M, z8.s, z17.s\n"
"fmla z26.s, p3/M, z8.s, z16.s\n"
"fmla z27.s, p3/M, z7.s, z16.s\n"
".inst 0xc1aecab8 // fclamp { z24.s-z27.s }, z21.s, z14.s\n"
"st1w { z24.s }, p0, [x24]\n"
"st1w { z25.s }, p0, [x24, x25, LSL #2]\n"
"st1w { z26.s }, p0, [x22]\n"
"st1w { z27.s }, p0, [x22, x25, LSL #2]\n"
"blt 1b\n"
".inst 0xd503467f // SMSTOP\n"
:
: [n_channels] "r" ((unsigned long) n_channels), [offsetof_args_inptr] "I" (offsetof(Args, inptr)), [offsetof_args_ld_input_col] "I" (offsetof(Args, ld_input_col)), [offsetof_args_ld_input_row] "I" (offsetof(Args, ld_input_row)), [offsetof_args_ld_output_col] "I" (offsetof(Args, ld_output_col)), [offsetof_args_ld_output_row] "I" (offsetof(Args, ld_output_row)), [offsetof_args_max] "I" (offsetof(Args, max)), [offsetof_args_min] "I" (offsetof(Args, min)), [offsetof_args_n_tile_cols] "I" (offsetof(Args, n_tile_cols)), [offsetof_args_n_tile_rows] "I" (offsetof(Args, n_tile_rows)), [offsetof_args_outptr] "I" (offsetof(Args, outptr)), [offsetof_args_params] "I" (offsetof(Args, params)), [offsetof_args_tile_i] "I" (offsetof(Args, tile_i)), [offsetof_args_tile_j] "I" (offsetof(Args, tile_j)), [params_struct] "r" (¶ms_struct)
: "cc", "memory", "p0", "p1", "p2", "p3", "p4", "p5", "p6", "p7", "p8", "p9", "p10", "p11", "p12", "p13", "p14", "p15", "x4", "x5", "x6", "x7", "x8", "x9", "x10", "x11", "x12", "x13", "x14", "x15", "x16", "x17", "x20", "x21", "x22", "x23", "x24", "x25", "x26", "x27", "x28", "z0", "z1", "z2", "z3", "z4", "z5", "z6", "z7", "z8", "z9", "z10", "z11", "z12", "z13", "z14", "z15", "z16", "z17", "z18", "z19", "z20", "z21", "z22", "z23", "z24", "z25", "z26", "z27", "z28", "z29", "z30", "z31"
);
}
} // namespace depthwise
} // namespace arm_conv
#endif // defined(ARM_COMPUTE_ENABLE_SME2)
|
