Comhaid
rocm-systems/tests/rocprofv3/tracing-hip-in-libraries/validate.py
T
Jonathan R. Madsen 1f96593b4f Test using HIP Graphs (#835)
* Test using hip graphs

* Remove assert for api_end < async_end

* Update rocprofv3/tracing-hip-in-libraries::test_api_trace

* Update rocprofv3/tracing-hip-in-libraries::test_api_trace

* Increase rocprofv3-test-trace-hip-in-libraries-validate timeout

* Update rocprofv3/tracing-hip-in-libraries::test_api_trace

* Remove submit retry

* Update rocprofv3/tracing-hip-in-libraries::test_api_trace

* Increase rocprofv3-test-trace-hip-in-libraries-validate timeout

* Update lib/common/container/record_header_buffer.hpp

- minor tweaks

* Update lib/rocprofiler-sdk/buffer.hpp

- tweak ROCPROFILER_BUFFER_POLICY_LOSSLESS flush behavior

* Increase rocprofv3-test-trace-hip-in-libraries-validate timeout

* Update rocprofv3/tracing-hip-in-libraries::test_api_trace

* Revert rocprofv3-test-trace-hip-in-libraries-validate timeout

* Update run-ci.py

- RETRY_COUNT set to zero
2024-05-07 15:10:22 -05:00

261 línte
8.7 KiB
Python

#!/usr/bin/env python3
import re
import sys
import pytest
class dim3(object):
def __init__(self, x, y, z):
self.x = int(x)
self.y = int(y)
self.z = int(z)
def as_tuple(self):
return (self.x, self.y, self.z)
def validate_average(row):
avg_ns = float(row["AverageNs"])
tot_ns = int(row["TotalDurationNs"])
num_cnt = float(row["Calls"])
assert abs(avg_ns - (tot_ns / num_cnt)) < 1.0e-3, f"{row}"
def validate_stats(row):
min_v = int(row["MinNs"])
max_v = int(row["MaxNs"])
avg_v = float(row["AverageNs"])
cnt_v = int(row["Calls"])
stddev_v = float(row["StdDev"])
assert min_v > 0, f"{row}"
assert max_v > 0, f"{row}"
assert min_v < max_v if cnt_v > 1 else min_v == max_v, f"{row}"
assert min_v < avg_v if cnt_v > 1 else min_v == int(avg_v), f"{row}"
assert max_v > avg_v if cnt_v > 1 else max_v == int(avg_v), f"{row}"
assert stddev_v > 0.0 if cnt_v > 1 else int(stddev_v) == 0, f"{row}"
def test_api_trace(
hsa_input_data,
hip_input_data,
kernel_input_data,
memory_copy_input_data,
hip_stats_data,
):
functions = []
hsa_correlation_ids = []
hip_correlation_ids = []
for row in hsa_input_data:
assert row["Domain"] in (
"HSA_CORE_API",
"HSA_AMD_EXT_API",
"HSA_IMAGE_EXT_API",
"HSA_FINALIZE_EXT_API",
)
assert int(row["Process_Id"]) > 0
assert int(row["Thread_Id"]) >= int(row["Process_Id"])
assert int(row["End_Timestamp"]) >= int(row["Start_Timestamp"])
functions.append(row["Function"])
cid = int(row["Correlation_Id"])
hsa_correlation_ids.append(cid)
for row in hip_input_data:
assert row["Domain"] in [
"HIP_RUNTIME_API",
"HIP_COMPILER_API",
]
assert int(row["Process_Id"]) > 0
assert int(row["Thread_Id"]) == 0 or int(row["Thread_Id"]) >= int(
row["Process_Id"]
)
assert int(row["End_Timestamp"]) >= int(row["Start_Timestamp"])
functions.append(row["Function"])
cid = int(row["Correlation_Id"])
hip_correlation_ids.append(cid)
def get_sorted_unique(inp):
return sorted(list(set(inp)))
def diagnose_non_unique(_input_data):
_corr_id_hist = {}
for row in _input_data:
_cid = int(row["Correlation_Id"])
# ensure duplicate does not already exist
assert (
_cid not in _corr_id_hist.keys()
), f"\ncurrent : {row}\nprevious: {_corr_id_hist[_cid]}"
_corr_id_hist[_cid] = row
if len(hsa_correlation_ids) != len(get_sorted_unique(hsa_correlation_ids)):
diagnose_non_unique(hsa_input_data)
if len(hip_correlation_ids) != len(get_sorted_unique(hip_correlation_ids)):
diagnose_non_unique(hip_input_data)
correlation_ids = get_sorted_unique(hsa_correlation_ids + hip_correlation_ids)
# make sure that we have associated API calls for all async ops
for itr in [kernel_input_data, memory_copy_input_data]:
for row in itr:
cid = int(row["Correlation_Id"])
assert (
cid in correlation_ids
), f"[{cid}] {row}\nCorrelation IDs:\n\t{correlation_ids}"
# all correlation ids are unique
if len(correlation_ids) != (len(hsa_input_data) + len(hip_input_data)):
for itr in hsa_input_data:
assert int(itr["Correlation_Id"]) in correlation_ids, f"{itr}"
for itr in hip_input_data:
assert int(itr["Correlation_Id"]) in correlation_ids, f"{itr}"
assert len(correlation_ids) == (len(hsa_input_data) + len(hip_input_data))
# correlation ids are numbered from 1 to N
assert correlation_ids[0] == 1, f"{correlation_ids}"
assert correlation_ids[-1] == len(correlation_ids), f"{correlation_ids}"
functions = list(set(functions))
for itr in (
"hsa_amd_memory_async_copy_on_engine",
"hsa_agent_get_info",
"hsa_agent_iterate_isas",
"hsa_signal_create",
"hsa_agent_get_info",
"hsa_executable_symbol_get_info",
):
assert itr in functions
if hip_input_data:
for itr in (
"hipGetLastError",
"hipLaunchKernel",
"hipStreamSynchronize",
"hipMemcpyAsync",
"hipFree",
"hipStreamDestroy",
"hipDeviceSynchronize",
"hipDeviceReset",
"hipSetDevice",
):
assert itr in functions
for row in hip_stats_data:
assert int(row["TotalDurationNs"]) > 0
assert int(row["Calls"]) > 0
validate_average(row)
assert float(row["Percentage"]) > 0.0
validate_stats(row)
def test_kernel_trace(kernel_input_data, kernel_stats_data):
valid_kernel_names = sorted(
[
"(anonymous namespace)::transpose(int const*, int*, int, int)",
"void (anonymous namespace)::addition_kernel<float>(float*, float const*, float const*, int, int)",
"void (anonymous namespace)::divide_kernel<float>(float*, float const*, float const*, int, int)",
"void (anonymous namespace)::multiply_kernel<float>(float*, float const*, float const*, int, int)",
"void (anonymous namespace)::subtract_kernel<float>(float*, float const*, float const*, int, int)",
]
)
kernels = []
for row in kernel_input_data:
kernel_name = row["Kernel_Name"]
if re.search(r"__amd_rocclr_.*", kernel_name):
continue
assert row["Kind"] == "KERNEL_DISPATCH"
assert int(row["Agent_Id"]) > 0
assert int(row["Queue_Id"]) > 0
assert int(row["Kernel_Id"]) > 0
assert int(row["Correlation_Id"]) > 0
assert kernel_name in valid_kernel_names
if kernel_name not in kernels:
kernels.append(kernel_name)
workgrp_size = dim3(
row["Workgroup_Size_X"], row["Workgroup_Size_Y"], row["Workgroup_Size_Z"]
)
grid_size = dim3(row["Grid_Size_X"], row["Grid_Size_Y"], row["Grid_Size_Z"])
if kernel_name == "__amd_rocclr_fillBufferAligned":
assert workgrp_size.as_tuple() > (1, 1, 1)
assert grid_size.as_tuple() > (1, 1, 1)
elif "transpose" in kernel_name:
assert workgrp_size.as_tuple() == (32, 32, 1)
assert grid_size.as_tuple() == (9920, 9920, 1)
else:
assert workgrp_size.as_tuple() == (64, 1, 1)
assert grid_size.as_tuple() == (4096, 2048, 1)
assert int(row["End_Timestamp"]) >= int(row["Start_Timestamp"])
kernels = sorted(list(set(kernels)))
assert kernels == valid_kernel_names
for row in kernel_stats_data:
assert int(row["TotalDurationNs"]) > 0
assert int(row["Calls"]) > 0
validate_average(row)
assert float(row["Percentage"]) > 0.0
validate_stats(row)
def test_memory_copy_trace(
agent_info_input_data,
memory_copy_input_data,
hsa_input_data,
hsa_stats_data,
memory_copy_stats_data,
):
def get_agent(node_id):
for row in agent_info_input_data:
if row["Logical_Node_Id"] == node_id:
return row
return None
for row in memory_copy_input_data:
assert row["Kind"] == "MEMORY_COPY"
assert row["Direction"] in ("HOST_TO_DEVICE", "DEVICE_TO_HOST")
src_agent = get_agent(row["Source_Agent_Id"])
dst_agent = get_agent(row["Destination_Agent_Id"])
assert src_agent is not None and dst_agent is not None, f"{agent_info_input_data}"
if row["Direction"] == "HOST_TO_DEVICE":
assert src_agent["Agent_Type"] == "CPU"
assert dst_agent["Agent_Type"] == "GPU"
elif row["Direction"] == "DEVICE_TO_HOST":
assert src_agent["Agent_Type"] == "GPU"
assert dst_agent["Agent_Type"] == "CPU"
assert int(row["Correlation_Id"]) > 0
assert int(row["End_Timestamp"]) >= int(row["Start_Timestamp"])
valid_length = 0
for row in hsa_input_data:
if re.search(r".*memory_async_copy.*", row["Function"]):
valid_length += 1
assert len(memory_copy_input_data) == valid_length
for row in hsa_stats_data:
assert int(row["TotalDurationNs"]) > 0
assert int(row["Calls"]) > 0
validate_average(row)
assert float(row["Percentage"]) > 0.0
validate_stats(row)
for row in memory_copy_stats_data:
assert int(row["TotalDurationNs"]) > 0
assert int(row["Calls"]) > 0
validate_average(row)
assert float(row["Percentage"]) > 0.0
validate_stats(row)
if __name__ == "__main__":
exit_code = pytest.main(["-x", __file__] + sys.argv[1:])
sys.exit(exit_code)