File size: 11,971 Bytes
8b7c501 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 |
#!/usr/bin/env python
# Copyright 2019 Google LLC
#
# This source code is licensed under the BSD-style license found in the
# LICENSE file in the root directory of this source tree.
import argparse
import bisect
import codecs
import math
import os
import re
import sys
import yaml
sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
from primes import next_prime
import xngen
import xnncommon
parser = argparse.ArgumentParser(description='XNNPACK generator')
parser.add_argument("-s", "--spec", metavar="FILE", required=True,
help="Spec (YAML) file")
parser.add_argument("-o", "--output", metavar="FILE", required=True,
help='Output (C++ source) file')
parser.set_defaults(defines=list())
def split_ukernel_name(name):
common_name, target_name = name.split("__", 1)
common_parts = common_name.split("_")
param_spec = common_parts[-1]
# New transitional naming convention for unipass microkernels.
m = re.search(r'(\d+)p(\d+)c', param_spec);
assert(m)
primary_tile = 0;
cr = int(m[2])
kr = int(m[1])
arch, isa, assembly = xnncommon.parse_target_name(target_name)
requantization = common_parts[-3]
if requantization not in ["fp32", "rndnu"]:
requantization = None
return primary_tile, cr, kr, requantization, arch, isa, assembly
DWCONV_TEST_CODE = """\
TEST(${TEST_NAME}, c_eq_${CBLOCK}) {
$if ISA_CHECK:
${ISA_CHECK};
DWConvMicrokernelTester()
.channel_tile(${CR})
.kernel_tile(${KR})
.channels(${CBLOCK})
.Test(${", ".join(TEST_ARGS)});
}
$if IS_PIPELINED:
TEST(${TEST_NAME}, c_eq_${CBLOCK * 2}) {
$if ISA_CHECK:
${ISA_CHECK};
DWConvMicrokernelTester()
.channel_tile(${CR})
.kernel_tile(${KR})
.channels(${CBLOCK * 2})
.Test(${", ".join(TEST_ARGS)});
}
$if CBLOCK > 1:
TEST(${TEST_NAME}, c_div_${CBLOCK}) {
$if ISA_CHECK:
${ISA_CHECK};
for (uint32_t channels = ${ADJCBLOCK + CBLOCK}; channels < ${CR * 16}; channels += ${CR * 3}) {
DWConvMicrokernelTester()
.channel_tile(${CR})
.kernel_tile(${KR})
.channels(channels)
.Test(${", ".join(TEST_ARGS)});
}
}
$if ACTIVATION == "MINMAX":
TEST(${TEST_NAME}, c_div_${CBLOCK}_with_qmin) {
$if ISA_CHECK:
${ISA_CHECK};
for (uint32_t channels = ${ADJCBLOCK + CBLOCK}; channels < ${CR * 16}; channels += ${CR * 3}) {
DWConvMicrokernelTester()
.channel_tile(${CR})
.kernel_tile(${KR})
.channels(channels)
.qmin(128)
.Test(${", ".join(TEST_ARGS)});
}
}
TEST(${TEST_NAME}, c_div_${CBLOCK}_with_qmax) {
$if ISA_CHECK:
${ISA_CHECK};
for (uint32_t channels = ${ADJCBLOCK + CBLOCK}; channels < ${CR * 16}; channels += ${CR * 3}) {
DWConvMicrokernelTester()
.channel_tile(${CR})
.kernel_tile(${KR})
.channels(channels)
.qmax(128)
.Test(${", ".join(TEST_ARGS)});
}
}
TEST(${TEST_NAME}, c_lt_${ADJCBLOCK}) {
$if ISA_CHECK:
${ISA_CHECK};
for (uint32_t channels = 1; channels < ${ADJCBLOCK}; channels++) {
DWConvMicrokernelTester()
.channel_tile(${CR})
.kernel_tile(${KR})
.channels(channels)
.Test(${", ".join(TEST_ARGS)});
}
}
TEST(${TEST_NAME}, c_gt_${ADJCBLOCK}) {
$if ISA_CHECK:
${ISA_CHECK};
for (uint32_t channels = ${ADJCBLOCK + 1}; channels < ${10 if CBLOCK == 1 else ADJCBLOCK + CBLOCK}; channels++) {
DWConvMicrokernelTester()
.channel_tile(${CR})
.kernel_tile(${KR})
.channels(channels)
.Test(${", ".join(TEST_ARGS)});
}
}
$if ACTIVATION == "MINMAX":
TEST(${TEST_NAME}, c_gt_${ADJCBLOCK}_with_qmin) {
$if ISA_CHECK:
${ISA_CHECK};
for (uint32_t channels = ${ADJCBLOCK + 1}; channels < ${10 if CBLOCK == 1 else ADJCBLOCK + CBLOCK}; channels++) {
DWConvMicrokernelTester()
.channel_tile(${CR})
.kernel_tile(${KR})
.channels(channels)
.qmin(128)
.Test(${", ".join(TEST_ARGS)});
}
}
TEST(${TEST_NAME}, c_gt_${ADJCBLOCK}_with_qmax) {
$if ISA_CHECK:
${ISA_CHECK};
for (uint32_t channels = ${ADJCBLOCK + 1}; channels < ${10 if CBLOCK == 1 else ADJCBLOCK + CBLOCK}; channels++) {
DWConvMicrokernelTester()
.channel_tile(${CR})
.kernel_tile(${KR})
.channels(channels)
.qmax(128)
.Test(${", ".join(TEST_ARGS)});
}
}
TEST(${TEST_NAME}, multipixel) {
$if ISA_CHECK:
${ISA_CHECK};
for (size_t channels = 1; channels <= ${CBLOCK * 5}; channels += ${max(1, CBLOCK - 1)}) {
DWConvMicrokernelTester()
.channel_tile(${CR})
.kernel_tile(${KR})
.channels(channels)
.width(3)
.Test(${", ".join(TEST_ARGS)});
}
}
TEST(${TEST_NAME}, multipixel_with_step) {
$if ISA_CHECK:
${ISA_CHECK};
for (size_t channels = 1; channels <= ${CBLOCK * 5}; channels += ${max(1, CBLOCK - 1)}) {
for (size_t step = 2; step <= ${KR}; step++) {
DWConvMicrokernelTester()
.channel_tile(${CR})
.kernel_tile(${KR})
.channels(channels)
.width(3)
.step(step)
.Test(${", ".join(TEST_ARGS)});
}
}
}
TEST(${TEST_NAME}, multipixel_with_output_stride) {
$if ISA_CHECK:
${ISA_CHECK};
for (size_t channels = 1; channels <= ${CBLOCK * 5}; channels += ${max(1, CBLOCK - 1)}) {
DWConvMicrokernelTester()
.channel_tile(${CR})
.kernel_tile(${KR})
.channels(channels)
.width(5)
.output_stride(${next_prime(CR * 5 + 1)})
.Test(${", ".join(TEST_ARGS)});
}
}
$if ACTIVATION == "MINMAX":
TEST(${TEST_NAME}, multipixel_with_qmin) {
$if ISA_CHECK:
${ISA_CHECK};
for (size_t channels = 1; channels <= ${CBLOCK * 5}; channels += ${max(1, CBLOCK - 1)}) {
DWConvMicrokernelTester()
.channel_tile(${CR})
.kernel_tile(${KR})
.channels(channels)
.width(3)
.qmin(128)
.Test(${", ".join(TEST_ARGS)});
}
}
TEST(${TEST_NAME}, multipixel_with_qmax) {
$if ISA_CHECK:
${ISA_CHECK};
for (size_t channels = 1; channels <= ${CBLOCK * 5}; channels += ${max(1, CBLOCK - 1)}) {
DWConvMicrokernelTester()
.channel_tile(${CR})
.kernel_tile(${KR})
.channels(channels)
.width(3)
.qmax(128)
.Test(${", ".join(TEST_ARGS)});
}
}
$if DATATYPE == "qu8":
TEST(${TEST_NAME}, input_zero_point_only) {
$if ISA_CHECK:
${ISA_CHECK};
for (size_t channels = 1; channels <= ${CBLOCK * 5}; channels += ${max(1, CBLOCK - 1)}) {
DWConvMicrokernelTester()
.channel_tile(${CR})
.kernel_tile(${KR})
.channels(channels)
.width(3)
.input_zero_point(255)
.kernel_zero_point(0)
.Test(${", ".join(TEST_ARGS)});
}
}
TEST(${TEST_NAME}, kernel_zero_point_only) {
$if ISA_CHECK:
${ISA_CHECK};
for (size_t channels = 1; channels <= ${CBLOCK * 5}; channels += ${max(1, CBLOCK - 1)}) {
DWConvMicrokernelTester()
.channel_tile(${CR})
.kernel_tile(${KR})
.channels(channels)
.width(3)
.input_zero_point(0)
.kernel_zero_point(255)
.Test(${", ".join(TEST_ARGS)});
}
}
TEST(${TEST_NAME}, input_offset) {
$if ISA_CHECK:
${ISA_CHECK};
for (uint32_t channels = ${ADJCBLOCK + CBLOCK}; channels < ${CR * 16}; channels += ${CR * 3}) {
DWConvMicrokernelTester()
.channel_tile(${CR})
.kernel_tile(${KR})
.channels(channels)
.input_offset(${next_prime(CR + 1) * 16})
.Test(${", ".join(TEST_ARGS)});
}
}
TEST(${TEST_NAME}, zero) {
$if ISA_CHECK:
${ISA_CHECK};
for (uint32_t mz = 0; mz < ${KR}; mz++) {
for (uint32_t channels = ${ADJCBLOCK + CBLOCK}; channels < ${CR * 16}; channels += ${CR * 3}) {
DWConvMicrokernelTester()
.channel_tile(${CR})
.kernel_tile(${KR})
.channels(channels)
.input_offset(${next_prime(CR + 1) * 16})
.zero_index(mz)
.Test(${", ".join(TEST_ARGS)});
}
}
}
"""
def generate_test_cases(ukernel, primary_tile, cr, kr, c_block,
init_fn, requantization, is_pipelined, isa):
"""Generates all tests cases for a DWCONV micro-kernel.
Args:
ukernel: C name of the micro-kernel function.
cr: CR parameter of the DWCONV micro-kernel.
kr: KR parameter of the DWCONV micro-kernel.
k_block: Number of C values processed per one iteration of the main loop of
the micro-kernel.
init_fn: C name of the function to initialize microkernel parameters.
requantization: name of the requantization scheme used by the microkernel.
is_pipelined: Indicates if the micro-kernel is implemented with software
pipelining. Additional test cases are generated for software
pipelined micro-kernels to separately test prologue + epiloque
of the pipelined loop and iteration of the pipelined loop.
isa: instruction set required to run the micro-kernel. Generated unit test
will skip execution if the host processor doesn't support this ISA.
Returns:
Code for the test case.
"""
_, test_name = ukernel.split("_", 1)
_, datatype, ukernel_type, activation, _ = ukernel.split("_", 4)
if activation == "ukernel":
activation = "linear"
test_args = [ukernel]
if init_fn:
test_args.append(init_fn)
if requantization:
requantization_datatype = {"qc8": "qs8"}.get(datatype, datatype)
test_args.append("xnn_%s_requantize_%s" %
(requantization_datatype, requantization))
return xngen.preprocess(DWCONV_TEST_CODE, {
"TEST_NAME": test_name.upper().replace("UKERNEL_", ""),
"TEST_ARGS": test_args,
"UKERNEL_TYPE": ukernel_type.upper(),
"DATATYPE": datatype,
"ACTIVATION": activation.upper(),
"PRIMARY_TILE": primary_tile,
"CR": cr,
"KR": kr,
"CBLOCK": c_block,
"ADJCBLOCK": 2 * c_block if is_pipelined else c_block,
"IS_PIPELINED": is_pipelined,
"ISA_CHECK": xnncommon.generate_isa_check_macro(isa),
"next_prime": next_prime,
"sqrt": math.sqrt,
})
def main(args):
options = parser.parse_args(args)
with codecs.open(options.spec, "r", encoding="utf-8") as spec_file:
spec_yaml = yaml.safe_load(spec_file)
if not isinstance(spec_yaml, list):
raise ValueError("expected a list of micro-kernels in the spec")
tests = """\
// Copyright (c) Facebook, Inc. and its affiliates.
// All rights reserved.
//
// Copyright 2019 Google LLC
//
// This source code is licensed under the BSD-style license found in the
// LICENSE file in the root directory of this source tree.
//
// Auto-generated file. Do not edit!
// Specification: {specification}
// Generator: {generator}
#include <gtest/gtest.h>
#include <xnnpack/common.h>
#include <xnnpack/isa-checks.h>
#include <xnnpack/dwconv.h>
#include "dwconv-microkernel-tester.h"
""".format(specification=options.spec, generator=sys.argv[0])
for ukernel_spec in spec_yaml:
name = ukernel_spec["name"]
init_fn = ukernel_spec.get("init")
pipelined = bool(ukernel_spec.get("pipelined", False))
primary_tile, cr, kr, requantization, arch, isa, assembly = \
split_ukernel_name(name)
test_case = generate_test_cases(
name, primary_tile, cr, kr, cr, init_fn, requantization, pipelined, isa)
tests += "\n\n" + xnncommon.postprocess_test_case(test_case, arch, isa, assembly)
txt_changed = True
if os.path.exists(options.output):
with codecs.open(options.output, "r", encoding="utf-8") as output_file:
txt_changed = output_file.read() != tests
if txt_changed:
with codecs.open(options.output, "w", encoding="utf-8") as output_file:
output_file.write(tests)
if __name__ == "__main__":
main(sys.argv[1:])
|