Skip to content

Commit

Permalink
#30: Update entrypoint.sh and Markdown code gen
Browse files Browse the repository at this point in the history
  • Loading branch information
@JacobDomagala
JacobDomagala committed May 14, 2024
1 parent 6876544 commit 9ac7103
Show file tree
Hide file tree
Showing 4 changed files with 210 additions and 95 deletions.
17 changes: 1 addition & 16 deletions entrypoint.sh
View file
Original file line number Diff line number Diff line change
Expand Up @@ -104,29 +104,14 @@ tmp_dir=$(mktemp -d -t ci-XXXXXXXXXX)
# Generate graph
python3 /generate_build_graph.py -vt "$vt_build_time" -te "$tests_and_examples_build" -r "$GITHUB_RUN_NUMBER"

# perf_test_files=$(find "$VT_BUILD_FOLDER/tests/" -name "*_mem.csv" | sed 's!.*/!!' | sed -e 's/_mem.csv$//')
cd perf_tests

python3 /generate_perf_graph.py

# for file in $perf_test_files
# do
# # Each test generates both time/mem files
# time_file="${file}_time.csv"
# memory_file="${file}_mem.csv"

# echo "Test files $VT_BUILD_FOLDER/tests/$time_file $VT_BUILD_FOLDER/tests/$memory_file for test: $file"

# python3 /generate_perf_graph.py -time "$VT_BUILD_FOLDER/tests/$time_file"\
# -mem "$VT_BUILD_FOLDER/tests/$memory_file" -r "$GITHUB_RUN_NUMBER" -wiki "$tmp_dir"
# done

cd -

cp "$GITHUB_WORKSPACE/build_result.txt" "$INPUT_BUILD_STATS_OUTPUT"
eval cp "$GITHUB_WORKSPACE/flame_heaptrack*" "./perf_tests/"

python3 /generate_wiki_pages.py -t "$perf_test_files"
python3 /generate_wiki_pages.py

git add .
git commit -m "$INPUT_COMMIT_MESSAGE"
Expand Down
214 changes: 178 additions & 36 deletions generate_perf_graph.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -8,30 +8,81 @@
GRAPH_HEIGHT = 10
NUM_LAST_BUILDS = int(os.getenv("INPUT_NUM_LAST_BUILD", "30")) - 1
VT_BUILD_FOLDER = os.getenv("VT_BUILD_FOLDER", "/build/vt")
RUN_NUM = os.getenv("RUN_NUMBER")
DATE = date.today().strftime("%d %B %Y")
COMMIT_ID = os.getenv("GITHUB_SHA", "")

def generate_bar_graph_for_single_value(test_file_name, title):

def generate_bar_graph_for_single_value(test_file_name, title, hisotry_title):
time_df = pd.read_csv(f"{VT_BUILD_FOLDER}/tests/{test_file_name}.csv")

_, ax = plt.subplots(figsize=(GRAPH_WIDTH, GRAPH_HEIGHT))
x_pos = range(len(time_df))

ax.bar(x=x_pos, height=time_df['mean'], yerr=time_df['stdev'], align='center', alpha=0.7, ecolor='black', capsize=10)
ax.bar(
x=x_pos,
height=time_df["mean"],
yerr=time_df["stdev"],
align="center",
alpha=0.7,
ecolor="black",
capsize=10,
)

plt.title(title)

plt.xticks(x_pos, time_df['name'])
plt.xticks(x_pos, time_df["name"])
plt.xlabel("")

plt.ylabel("Time (ms)")
plt.tight_layout()

plt.savefig(f"{test_file_name}.png")

time_df["commit"] = COMMIT_ID
time_df["run_num"] = RUN_NUM

file_path = f"{test_file_name}_history.csv"
if os.path.exists(file_path):
past_results = pd.read_csv(file_path)
else:
past_results = pd.DataFrame(columns=time_df.columns)
past_results.to_csv(file_path, index=False)

updated_results = pd.concat([past_results, time_df], ignore_index=True)
updated_results.to_csv(file_path, index=False)

last_N_results = updated_results.tail(NUM_LAST_BUILDS)

_, ax = plt.subplots(figsize=(GRAPH_WIDTH, GRAPH_HEIGHT))
x_pos = range(len(last_N_results))

ax.bar(
x=x_pos,
height=last_N_results["mean"],
yerr=last_N_results["stdev"],
align="center",
alpha=0.7,
ecolor="black",
capsize=10,
)

plt.title(hisotry_title)

plt.xticks(x_pos, last_N_results["run_num"])
plt.xlabel("Run numbers")

plt.ylabel("Time (ms)")
plt.tight_layout()

plt.savefig(f"{test_file_name}_history.png")


def ping_pong():
df = pd.read_csv(f"{VT_BUILD_FOLDER}/tests/test_ping_pong_time.csv")

# Split data by nodes
num_nodes = df['node'].nunique()
num_nodes = df["node"].nunique()
time_data = [df[df["node"] == node] for node in range(num_nodes)]

# Create the plot
Expand All @@ -47,7 +98,7 @@ def ping_pong():
]

for node in range(1, num_nodes):
bar_positions.append([x + bar_width for x in bar_positions[node-1]])
bar_positions.append([x + bar_width for x in bar_positions[node - 1]])

for node in range(num_nodes):
ax_1.bar(
Expand All @@ -72,7 +123,7 @@ def ping_pong():

# Set y-axis label and scale
ax_1.set_ylabel("Time (ms)")
ax_1.set_yscale('log')
ax_1.set_yscale("log")

# Customize y-ticks
y_ticks = [0.03, 1, 5, 40]
Expand All @@ -83,69 +134,161 @@ def ping_pong():
plt.xticks(rotation=85)
plt.tight_layout()

plt.savefig("ping_pong_time.png")
plt.savefig("test_ping_pong_time.png")

memory_df = pd.read_csv(f"{VT_BUILD_FOLDER}/tests/test_ping_pong_mem.csv")
generate_memory_graph("ping_pong", memory_df)


def ping_pong_am():
generate_bar_graph_for_single_value("test_ping_pong_am_time", "Time for sending message (ping-pong) 1000 times")
generate_bar_graph_for_single_value(
"test_ping_pong_am_time",
"Time for sending message (ping-pong) 1000 times",
"Past runs of ping_pong_am",
)


def make_runnable_micro():
generate_bar_graph_for_single_value("test_make_runnable_micro_time", "Time for calling makeRunnable 1000 times")
generate_bar_graph_for_single_value(
"test_make_runnable_micro_time",
"Time for calling makeRunnable 1000 times",
"Past runs of make_runnable_micro",
)


def objgroup_local_send():
generate_bar_graph_for_single_value("test_objgroup_local_send_time", "Time for ObjectGroup Local Send (1000 Iterations)")
generate_bar_graph_for_single_value(
"test_objgroup_local_send_time",
"Time for ObjectGroup Local Send (1000 Iterations)",
"Past runs of objgroup_local_send",
)


def collection_local_send():
time_df = pd.read_csv(f"{VT_BUILD_FOLDER}/tests/test_collection_local_send_time.csv")
time_prealloc_df = pd.read_csv(f"{VT_BUILD_FOLDER}/tests/test_collection_local_send_preallocate_time.csv")
# Read data
time_df = pd.read_csv(
f"{VT_BUILD_FOLDER}/tests/test_collection_local_send_time.csv"
)
time_prealloc_df = pd.read_csv(
f"{VT_BUILD_FOLDER}/tests/test_collection_local_send_preallocate_time.csv"
)

time_df["name"] = "allocate"
time_prealloc_df["name"] = "preallocate"

combined_df = pd.concat([time_df, time_prealloc_df], axis=0)

_, ax = plt.subplots()
# Plot current data
_, ax = plt.subplots(figsize=(GRAPH_WIDTH, GRAPH_HEIGHT))
x_pos = range(len(combined_df))

ax.bar(x=x_pos, height=combined_df['mean'], yerr=combined_df['stdev'], align='center', alpha=0.7, ecolor='black', capsize=10)

ax.bar(
x=x_pos,
height=combined_df["mean"],
yerr=combined_df["stdev"],
align="center",
alpha=0.7,
ecolor="black",
capsize=10,
)
plt.title("Time for Collection Local Send (1000 Iterations)")
plt.xticks(x_pos, combined_df["name"])
plt.xlabel("Type")
plt.ylabel("Time (ms)")
plt.tight_layout()
plt.savefig("test_collection_local_send_time.png")

plt.xticks(x_pos, combined_df['name'])
plt.xlabel("")
### Create historic graph

plt.ylabel("Time (ms)")
combined_df["commit"] = COMMIT_ID
combined_df["run_num"] = RUN_NUM

file_path = "test_collection_local_send_time_history.csv"
if os.path.exists(file_path):
past_results = pd.read_csv(file_path)
else:
past_results = pd.DataFrame(columns=combined_df.columns)

# Append new results and save
updated_results = pd.concat([past_results, combined_df], ignore_index=True)
updated_results.to_csv(file_path, index=False)

# Get last N results
last_N_results = updated_results.tail(NUM_LAST_BUILDS)

# Split data by type
time_data = {
"allocate": last_N_results[last_N_results["name"] == "allocate"],
"preallocate": last_N_results[last_N_results["name"] == "preallocate"],
}

# Create the plot for historical data
_, ax_1 = plt.subplots(figsize=(GRAPH_WIDTH, GRAPH_HEIGHT))
ax_1.set_title("Past Runs of Collection Local Send")

num_iter = list(range(len(time_data["allocate"])))
bar_width = 0.4

bar_positions = {
"allocate": [i - bar_width / 2 for i in num_iter],
"preallocate": [i + bar_width / 2 for i in num_iter],
}

for name in ["allocate", "preallocate"]:
ax_1.bar(
bar_positions[name],
time_data[name]["mean"],
yerr=time_data[name]["stdev"],
label=f"{name}",
width=bar_width,
align="center",
alpha=0.9,
ecolor="black",
capsize=5.0,
)

ax_1.grid(True, which="both", ls="--", linewidth=0.5)
ax_1.set_xlabel("Run Number")
ax_1.set_xticks(num_iter)
ax_1.set_xticklabels(time_data["allocate"]["run_num"].astype(str))
ax_1.set_ylabel("Time (ms)")
ax_1.legend()
plt.tight_layout()
plt.savefig("test_collection_local_send_time_history.png")

plt.savefig("./test_collection_local_send_time.png")

def reduce():
time_df = pd.read_csv(f"{VT_BUILD_FOLDER}/tests/test_reduce_time.csv")
memory_df = pd.read_csv(f"{VT_BUILD_FOLDER}/tests/test_reduce_mem.csv")

# Extract iteration number from 'name'
time_df['iteration'] = time_df['name'].apply(lambda x: int(x.split()[0]))
time_df["iteration"] = time_df["name"].apply(lambda x: int(x.split()[0]))

_, ax = plt.subplots()
for node in time_df['node'].unique():
node_data = time_df[time_df['node'] == node]
_, caps, bars = ax.errorbar(node_data['iteration'], node_data['mean'], yerr=node_data['stdev'], fmt='-', label=f'Node {node}')
_, ax = plt.subplots(figsize=(GRAPH_WIDTH, GRAPH_HEIGHT))
for node in time_df["node"].unique():
node_data = time_df[time_df["node"] == node]
_, caps, bars = ax.errorbar(
node_data["iteration"],
node_data["mean"],
yerr=node_data["stdev"],
fmt="-",
label=f"Node {node}",
)

# loop through bars and caps and set the alpha value
[bar.set_alpha(0.3) for bar in bars]
[cap.set_alpha(0.3) for cap in caps]

ax.set_xlabel('Iteration')
ax.set_ylabel('Time (ms)')
ax.set_title('Reduce times over 100 iterations')
ax.set_xlabel("Iteration")
ax.set_ylabel("Time (ms)")
ax.set_title("Reduce times over 100 iterations")
ax.legend()
plt.tight_layout()
plt.savefig("test_reduce_time.png")

generate_memory_graph("reduce", memory_df)


def prepare_data():
"""Parse the input data, read CSV file and append the new results"""

Expand Down Expand Up @@ -239,32 +382,32 @@ def set_graph_properties():
plt.rc("legend", fontsize=small_size)
plt.rc("figure", titlesize=big_size)


def generate_memory_graph(test_name, memory_data):
_, ax1 = plt.subplots(figsize=(GRAPH_WIDTH, GRAPH_HEIGHT))

ax1.set_title(f"{test_name} Memory Usage")
plt.xlabel("Iteration")
plt.ylabel("Size (MiB)")

num_nodes = memory_data['node'].max() + 1
num_nodes = memory_data["node"].max() + 1

for node in range(num_nodes):
node_data = memory_data[memory_data['node'] == node]
num_iter = list(range(len(node_data))) # Ensure num_iter matches the length of node_data
node_data = memory_data[memory_data["node"] == node]
num_iter = list(
range(len(node_data))
) # Ensure num_iter matches the length of node_data

ax1.plot(
num_iter,
node_data['mem'] / 1024 / 1024,
label=f"Node {node}",
linewidth=4
num_iter, node_data["mem"] / 1024 / 1024, label=f"Node {node}", linewidth=4
)

ax1.xaxis.get_major_locator().set_params(integer=True)
ax1.legend()
ax1.grid(True)

plt.tight_layout()
plt.savefig(f"{test_name}_memory.png")
plt.savefig(f"test_{test_name}_mem.png")


def generate_historic_graph(test_name, num_nodes, dataframe):
Expand Down Expand Up @@ -324,4 +467,3 @@ def generate_historic_graph(test_name, num_nodes, dataframe):
make_runnable_micro()
ping_pong_am()
ping_pong()

Loading

0 comments on commit 9ac7103

Please sign in to comment.