vedithal-amd
aa5dfb98f9
* Fix rocprofiler-sdk metrics definition
* Use TCC_EA0_RDREQ_128B instead of TCC_BUBBLE counter for L2 cache to
HBM counters and metrics
* Update MI350 counter definitions
* FETCH_SIZE
* BANDWIDTH_EA
* Update MI350 metrics definitions
* System Speed of Light, L2-Fabric Read BW
* Roofline Plot Points, AI (Arithmetic Intensity) HBM
* Roofline Performance Rates, HBM Bandwidth
* Remove redundant definition for gfx950 and fix BANDWIDTH_EA definition
Test HBM bandwidth metric for memcopy workload
* Add memcopy.cpp workload
* Add metric validation test suite to validate HBM Bandwidth metric for
memcopy workload
* Move gpu_soc() to test_utils.py for better re-usability
* Update TUI analysis config
* Fix hbm bandwidth formula for mi350 in calc_ai_profile
Co-authored-by: Alysa Liu <Alysa.Liu@amd.com>
139 líneas
4.7 KiB
Python
139 líneas
4.7 KiB
Python
##############################################################################
|
|
# MIT License
|
|
#
|
|
# Copyright (c) 2025 Advanced Micro Devices, Inc. All Rights Reserved.
|
|
#
|
|
# 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.
|
|
|
|
##############################################################################
|
|
|
|
import csv
|
|
from pathlib import Path
|
|
|
|
import pytest
|
|
import test_utils
|
|
|
|
config = {}
|
|
config["vseq"] = ["./tests/vsequential_access"]
|
|
config["vrand"] = ["./tests/vrandom_access"]
|
|
config["cleanup"] = True
|
|
config["COUNTER_LOGGING"] = False
|
|
config["METRIC_COMPARE"] = False
|
|
config["METRIC_LOGGING"] = False
|
|
|
|
|
|
def load_metrics(csv_file_path):
|
|
"""
|
|
Reads the CSV file into a dictionary of dictionaries:
|
|
{
|
|
"Metric_1": {
|
|
"Avg": value,
|
|
"Min": value,
|
|
"Max": value,
|
|
"Unit": "unit"
|
|
},
|
|
"Metric_2": { ... },
|
|
...
|
|
}
|
|
"""
|
|
metrics_data = {}
|
|
with open(csv_file_path, newline="") as csvfile:
|
|
reader = csv.DictReader(csvfile) # reads header from first line
|
|
|
|
for row in reader:
|
|
metric_name = row["Metric"].strip()
|
|
metrics_data[metric_name] = {
|
|
"Avg": float(row["Avg"]) if row["Avg"] else None,
|
|
"Min": float(row["Min"]) if row["Min"] else None,
|
|
"Max": float(row["Max"]) if row["Max"] else None,
|
|
"Unit": row["Unit"].strip() if row["Unit"] else None,
|
|
}
|
|
return metrics_data
|
|
|
|
|
|
soc = test_utils.gpu_soc()
|
|
|
|
|
|
@pytest.mark.L1_cache
|
|
def test_L1_cache_counters(
|
|
binary_handler_profile_rocprof_compute, binary_handler_analyze_rocprof_compute
|
|
):
|
|
if not soc or "MI300" not in soc:
|
|
pytest.skip("Skipping L1 cache test for non-mi300 socs.")
|
|
|
|
# set up two apps: sequential and random access
|
|
app_names = ["vseq", "vrand"]
|
|
options = ["-b", "16"]
|
|
|
|
result = {}
|
|
metrics = ["Read Req", "Write Req", "Cache Hit Rate"]
|
|
base = Path(test_utils.get_output_dir())
|
|
|
|
for app_name in app_names:
|
|
workload_dir = f"{base}/{app_name}"
|
|
workload_dir_output = f"{base}_{app_name}"
|
|
|
|
# 1. profile the app
|
|
return_code = binary_handler_profile_rocprof_compute(
|
|
config,
|
|
workload_dir,
|
|
options,
|
|
check_success=False,
|
|
roof=False,
|
|
app_name=app_name,
|
|
)
|
|
assert return_code == 0
|
|
|
|
# 2. analyze the results
|
|
return_code = binary_handler_analyze_rocprof_compute([
|
|
"analyze",
|
|
"--path",
|
|
workload_dir,
|
|
"-b",
|
|
"16.3",
|
|
"--output-format",
|
|
"csv",
|
|
"--output-name",
|
|
workload_dir_output,
|
|
])
|
|
assert return_code == 0
|
|
|
|
# 3. save results in local
|
|
|
|
# FIXME: customize file name to avoid hardcode
|
|
csv_path = workload_dir_output + "/16.3_vL1D_cache_access_metrics.csv"
|
|
data = load_metrics(csv_path)
|
|
|
|
for metric in metrics:
|
|
if app_name not in result or not isinstance(result[app_name], dict):
|
|
result[app_name] = {}
|
|
result[app_name][metric] = data[metric]["Avg"]
|
|
|
|
# 4. clean local output
|
|
test_utils.clean_output_dir(config["cleanup"], workload_dir)
|
|
test_utils.clean_output_dir(config["cleanup"], workload_dir_output)
|
|
test_utils.clean_output_dir(config["cleanup"], base)
|
|
|
|
# 5. check results are expected
|
|
|
|
# FIXME: use a range for comparison to account for different results
|
|
assert result["vseq"]["Cache Hit Rate"] >= result["vrand"]["Cache Hit Rate"]
|
|
assert result["vseq"]["Read Req"] <= result["vrand"]["Read Req"]
|
|
assert result["vseq"]["Write Req"] <= result["vrand"]["Write Req"]
|