1
0
mirror of https://github.com/ARM-software/workload-automation.git synced 2025-01-31 10:11:17 +00:00

workloads: add memcpy workload

Ported memcpy workload from WA2.
This commit is contained in:
Sergei Trofimov 2017-10-05 16:50:13 +01:00
parent c93b1a0939
commit 0d54c5f070
5 changed files with 211 additions and 0 deletions

View File

@ -0,0 +1,81 @@
# Copyright 2013-2015 ARM Limited
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
# pylint: disable=E1101,W0201
import os
import re
from wa import Workload, Parameter, Executable
THIS_DIR = os.path.dirname(__file__)
RESULT_REGEX = re.compile('Total time: ([\d.]+) s.*Bandwidth: ([\d.]+) MB/s', re.S)
class Memcpy(Workload):
name = 'memcpy'
description = """
Runs memcpy in a loop.
This will run memcpy in a loop for a specified number of times on a buffer
of a specified size. Additionally, the affinity of the test can be set to
one or more specific cores.
This workload is single-threaded. It genrates no scores or metrics by
itself.
"""
parameters = [
Parameter('buffer_size', kind=int, default=1024 * 1024 * 5,
description='''
Specifies the size, in bytes, of the buffer to be copied.
'''),
Parameter('iterations', kind=int, default=1000,
description='''
Specfies the number of iterations that will be performed.
'''),
Parameter('cpus', kind=list,
description='''
A list of integers specifying ordinals of cores to which the
affinity of the test process should be set. If not specified,
all avaiable cores will be used.
'''),
]
def initialize(self, context):
self.binary_name = 'memcpy'
resource = Executable(self, self.target.abi, self.binary_name)
host_binary = context.resolver.get(resource)
Memcpy.target_exe = self.target.install_if_needed(host_binary)
def setup(self, context):
self.command = '{} -i {} -s {}'.format(self.target_exe, self.iterations, self.buffer_size)
for c in (self.cpus or []):
self.command += ' -c {}'.format(c)
self.result = None
def run(self, context):
self.result = self.target.execute(self.command, timeout=300)
def extract_results(self, context):
if self.result:
match = RESULT_REGEX.search(self.result)
context.add_metric('time', float(match.group(1)), 'seconds', lower_is_better=True)
context.add_metric('bandwidth', float(match.group(2)), 'MB/s')

Binary file not shown.

Binary file not shown.

View File

@ -0,0 +1,16 @@
# Copyright 2013-2015 ARM Limited
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
${CROSS_COMPILE}gcc -static memcopy.c -o memcopy

View File

@ -0,0 +1,114 @@
/* Copyright 2013-2015 ARM Limited
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
#define _GNU_SOURCE
#include <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <sched.h>
#include <unistd.h>
#include <sys/syscall.h>
#include <pthread.h>
#include <time.h>
const int MAX_CPUS = 8;
const int DEFAULT_ITERATIONS = 1000;
const int DEFAULT_BUFFER_SIZE = 1024 * 1024 * 5;
int set_affinity(size_t cpus_size, int* cpus)
{
int i;
int mask = 0;
for(i = 0; i < cpus_size; ++i)
{
mask |= 1 << cpus[i];
}
return syscall(__NR_sched_setaffinity, 0, sizeof(mask), &mask);
}
int main(int argc, char** argv)
{
int cpus[MAX_CPUS];
int next_cpu = 0;
int iterations = DEFAULT_ITERATIONS;
int buffer_size = DEFAULT_BUFFER_SIZE;
int c;
while ((c = getopt(argc, argv, "i:c:s:")) != -1)
switch (c)
{
case 'c':
cpus[next_cpu++] = atoi(optarg);
if (next_cpu == MAX_CPUS)
{
fprintf(stderr, "Max CPUs exceeded.");
abort();
}
break;
case 'i':
iterations = atoi(optarg);
break;
case 's':
buffer_size = atoi(optarg);
break;
default:
abort();
break;
}
int ret;
if (next_cpu != 0)
if (ret = set_affinity(next_cpu, cpus))
{
fprintf(stderr, "sched_setaffinity returnred %i.", ret);
abort();
}
char* source = malloc(buffer_size);
char* dest = malloc(buffer_size);
struct timespec before, after;
if (clock_gettime(CLOCK_MONOTONIC, &before))
{
fprintf(stderr, "Could not get start time.");
abort();
}
int i;
for (i = 0; i < iterations; ++i)
{
memcpy(dest, source, buffer_size);
}
if (clock_gettime(CLOCK_MONOTONIC, &after))
{
fprintf(stderr, "Could not get end time.");
abort();
}
free(dest);
free(source);
long delta_sec = (long)(after.tv_sec - before.tv_sec);
long delta_nsec = after.tv_nsec - before.tv_nsec;
double delta = (double)delta_sec + delta_nsec / 1e9;
printf("Total time: %f s\n", delta);
printf("Bandwidth: %f MB/s\n", buffer_size / delta * iterations / 1e6);
return 0;
}