[ROCm/rocprofiler-compute commit: 6cbda3f1d8]
Cette révision appartient à :
Fei Zheng
2025-07-31 08:20:49 -06:00
révisé par GitHub
Parent 94ce792f39
révision 666f00a07f
6 fichiers modifiés avec 32 ajouts et 69 suppressions
+1 -1
Voir le fichier
@@ -212,4 +212,4 @@ extlinks = {
}
# Uncomment if facing rate limit exceed issue with local build
external_projects_remote_repository = ""
external_projects_remote_repository = ""
+1 -1
Voir le fichier
@@ -32,4 +32,4 @@ HIDDEN_COLUMNS = ["coll_level"]
HIDDEN_COLUMNS_CLI = ["Description", "coll_level"]
HIDDEN_SECTIONS = [400, 1900, 2000]
TIME_UNITS = {"s": 10 ** 9, "ms": 10 ** 6, "us": 10 ** 3, "ns": 1}
TIME_UNITS = {"s": 10**9, "ms": 10**6, "us": 10**3, "ns": 1}
+14 -56
Voir le fichier
@@ -259,25 +259,11 @@ def calc_ai(mspec, sort_type, ret_df):
df = df.sort_values(by=["Kernel_Name"])
df = df.reset_index(drop=True)
total_flops = (
valu_flops
) = (
mfma_flops_f6f4
) = (
mfma_flops_f8
) = (
mfma_flops_bf16
) = (
total_flops = valu_flops = mfma_flops_f6f4 = mfma_flops_f8 = mfma_flops_bf16 = (
mfma_flops_f16
) = (
mfma_iops_i8
) = (
mfma_flops_f32
) = (
mfma_flops_f64
) = (
lds_data
) = L1cache_data = L2cache_data = hbm_data = calls = totalDuration = avgDuration = 0.0
) = mfma_iops_i8 = mfma_flops_f32 = mfma_flops_f64 = lds_data = L1cache_data = (
L2cache_data
) = hbm_data = calls = totalDuration = avgDuration = 0.0
kernelName = ""
@@ -498,27 +484,13 @@ def calc_ai(mspec, sort_type, ret_df):
kernelName, idx, calls
)
)
total_flops = (
valu_flops
) = (
mfma_flops_f6f4
) = (
mfma_flops_f8
) = (
total_flops = valu_flops = mfma_flops_f6f4 = mfma_flops_f8 = (
mfma_flops_bf16
) = (
mfma_flops_f16
) = (
mfma_iops_i8
) = (
mfma_flops_f32
) = (
mfma_flops_f64
) = (
) = mfma_flops_f16 = mfma_iops_i8 = mfma_flops_f32 = mfma_flops_f64 = (
lds_data
) = (
L1cache_data
) = L2cache_data = hbm_data = calls = totalDuration = avgDuration = 0.0
) = L1cache_data = L2cache_data = hbm_data = calls = totalDuration = (
avgDuration
) = 0.0
if sort_type == "dispatches":
myList.append(
@@ -542,27 +514,13 @@ def calc_ai(mspec, sort_type, ret_df):
avgDuration,
)
)
total_flops = (
valu_flops
) = (
mfma_flops_f6f4
) = (
mfma_flops_f8
) = (
total_flops = valu_flops = mfma_flops_f6f4 = mfma_flops_f8 = (
mfma_flops_bf16
) = (
mfma_flops_f16
) = (
mfma_iops_i8
) = (
mfma_flops_f32
) = (
mfma_flops_f64
) = (
) = mfma_flops_f16 = mfma_iops_i8 = mfma_flops_f32 = mfma_flops_f64 = (
lds_data
) = (
L1cache_data
) = L2cache_data = hbm_data = calls = totalDuration = avgDuration = 0.0
) = L1cache_data = L2cache_data = hbm_data = calls = totalDuration = (
avgDuration
) = 0.0
myList.sort(key=lambda x: x.totalDuration, reverse=True)
+4 -4
Voir le fichier
@@ -42,7 +42,7 @@ indirs = [
"tests/workloads/vcopy/MI350",
]
time_units = {"s": 10 ** 9, "ms": 10 ** 6, "us": 10 ** 3, "ns": 1}
time_units = {"s": 10**9, "ms": 10**6, "us": 10**3, "ns": 1}
@pytest.mark.misc
@@ -1308,9 +1308,9 @@ def test_mathematical_correctness_all_units(sample_time_data, original_ns_values
from utils.tty import convert_time_columns
test_cases = [
("s", 10 ** 9), # 1 second = 10^9 nanoseconds
("ms", 10 ** 6), # 1 millisecond = 10^6 nanoseconds
("us", 10 ** 3), # 1 microsecond = 10^3 nanoseconds
("s", 10**9), # 1 second = 10^9 nanoseconds
("ms", 10**6), # 1 millisecond = 10^6 nanoseconds
("us", 10**3), # 1 microsecond = 10^3 nanoseconds
("ns", 1), # 1 nanosecond = 1 nanosecond
]
+3 -3
Voir le fichier
@@ -209,9 +209,9 @@ class TestDatabaseConnector:
with patch.object(connector, "prep_import") as mock_prep:
mock_prep.return_value = None
connector.connection_info[
"db"
] = "rocprofiler-compute_test_team_test_workload_MI100"
connector.connection_info["db"] = (
"rocprofiler-compute_test_team_test_workload_MI100"
)
connector.db_import()
+9 -4
Voir le fichier
@@ -6,9 +6,9 @@
# Read utils/unified_config.yaml and split it into metric tables per documentation section
# WARNING: This script will overwrite existing docs/data/metrics_description.yaml
import copy
import hashlib
import re
import copy
from pathlib import Path
import yaml
@@ -34,7 +34,10 @@ def update_analysis_config():
new_panel_config = {"Panel Config": {}}
new_panel_config["Panel Config"]["id"] = panel_config["id"]
new_panel_config["Panel Config"]["title"] = panel_config["title"]
new_panel_config["Panel Config"]["metrics_description"] = {key: value["plain"] for key, value in panel_config.get("metrics_description", {}).items()}
new_panel_config["Panel Config"]["metrics_description"] = {
key: value["plain"]
for key, value in panel_config.get("metrics_description", {}).items()
}
# Convert int into str with 4 digits
panel_id = str(panel_config["id"]).zfill(4)
# Replace parentehsis, hyphen, slash and space with underscore
@@ -57,7 +60,9 @@ def update_analysis_config():
for data_source_config in panel_config["data source"]:
data_source_config = copy.deepcopy(data_source_config)
if "metric_table" in data_source_config:
data_source_config["metric_table"]["metric"] = data_source_config["metric_table"]["metric"][gfx_version]
data_source_config["metric_table"]["metric"] = data_source_config[
"metric_table"
]["metric"][gfx_version]
new_panel_config["Panel Config"]["data source"].append(data_source_config)
# Write panel config to file
filename = Path(
@@ -126,7 +131,7 @@ def update_documentation():
"rst": panel_config["metrics_description"][key]["rst"],
"unit": panel_config["metrics_description"][key]["unit"],
}
panel_metric_map[data_source["metric_table"]["id"]] = metrics_info
panel_metric_map[data_source["metric_table"]["id"]] = metrics_info
# Merge panel_metric_map with section_panel_map
section_metric_map = {}