blob: b25164383fbed0372020c61d6051b570937f9bd4 (
plain)
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
|
#!/bin/sh -e
. ../../lib/sh-test-lib
OUTPUT="$(pwd)/output"
RESULT_FILE="${OUTPUT}/result.txt"
export RESULT_FILE
bandwidth_test() {
test_list="rd wr rdwr cp frd fwr fcp bzero bcopy"
for test in ${test_list}; do
# bw_mem use MB/s as units.
# shellcheck disable=SC2154
./bin/"${abi}"/bw_mem 512m "$test" 2>&1 \
| awk -v test_case="memory-${test}-bandwidth" \
'{printf("%s pass %s MB/s\n", test_case, $2)}' \
| tee -a "${RESULT_FILE}"
done
}
latency_test() {
# Set memory size to 256M to make sure that main memory will be measured.
lat_output="${OUTPUT}/lat-mem-rd.txt"
./bin/"${abi}"/lat_mem_rd 256m 128 2>&1 | tee "${lat_output}"
# According to lmbench manual:
# Only data accesses are measured; the instruction cache is not measured.
# L1: Try stride of 128 and array size of .00098.
# L2: Try stride of 128 and array size of .125.
grep "^0.00098" "${lat_output}" \
| awk '{printf("l1-read-latency pass %s ns\n", $2)}' \
| tee -a "${RESULT_FILE}"
grep "^0.125" "${lat_output}" \
| awk '{printf("l2-read-latency pass %s ns\n", $2)}' \
| tee -a "${RESULT_FILE}"
# Main memory: the last line.
grep "^256" "${lat_output}" \
| awk '{printf("main-memory-read-latency pass %s ns\n", $2)}' \
| tee -a "${RESULT_FILE}"
}
# Test run.
create_out_dir "${OUTPUT}"
detect_abi
bandwidth_test
latency_test
|
