[rocm-libraries] ROCm/rocm-libraries#4525 (commit 7f34b22)

[CK] Fix the launch_tests script.

## Motivation

Fix the script that filters the tests.

## Technical Details

There were several places where the paths had to be updated for the
launch_tests script to work correctly.

## Test Plan

<!-- Explain any relevant testing done to verify this PR. -->

## Test Result

<!-- Briefly summarize test outcomes. -->

## Submission Checklist

- [ ] Look over the contributing guidelines at
https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.
This commit is contained in:
Illia Silin
2026-02-12 04:43:27 +00:00
committed by assistant-librarian[bot]
parent e1e2f7ac2e
commit 47c7c034e9
5 changed files with 291 additions and 206 deletions

View File

@@ -16,8 +16,6 @@ import sys
import subprocess
from collections import defaultdict
import json
from concurrent.futures import ThreadPoolExecutor, as_completed
import threading
class EnhancedNinjaDependencyParser:
@@ -32,9 +30,6 @@ class EnhancedNinjaDependencyParser:
self.object_to_all_deps = {} # object -> [all_dependencies]
self.file_to_executables = defaultdict(set) # file -> {executables}
# Thread safety
self.lock = threading.Lock()
def parse_dependencies(self):
"""Main method to parse all dependencies."""
print(f"Parsing ninja dependencies from: {self.build_file_path}")
@@ -92,80 +87,98 @@ class EnhancedNinjaDependencyParser:
print(f"Found {len(self.object_to_source)} object-to-source mappings")
def _extract_object_dependencies(self):
"""Extract detailed dependencies for all object files using ninja -t deps."""
object_files = list(self.object_to_source.keys())
# Process object files in parallel for better performance
"""Extract detailed dependencies for all object files using a single ninja -t deps call.
Previous implementation spawned ninja -t deps per object file (29K+ subprocesses),
each re-parsing the full build.ninja. For large monorepo builds (246MB+ build.ninja),
each call takes 2-14 seconds, making the total ~54 minutes.
This implementation calls ninja -t deps once (no args) to dump ALL deps in ~2 seconds,
then parses the output and filters to only the objects we care about.
"""
object_files = set(self.object_to_source.keys())
if not object_files:
print("No object files found - skipping dependency extraction")
return
max_workers = min(128, len(object_files)) # Limit concurrent processes
print(f"Running single 'ninja -t deps' call for all built objects...")
with ThreadPoolExecutor(max_workers=max_workers) as executor:
# Submit all object files for processing
future_to_obj = {
executor.submit(self._get_object_dependencies, obj): obj
for obj in object_files
}
# Process completed futures
completed = 0
for future in as_completed(future_to_obj):
obj_file = future_to_obj[future]
try:
dependencies = future.result()
with self.lock:
self.object_to_all_deps[obj_file] = dependencies
completed += 1
if completed % 100 == 0:
print(
f"Processed {completed}/{len(object_files)} object files..."
)
except Exception as e:
print(f"Error processing {obj_file}: {e}")
print(
f"Completed dependency extraction for {len(self.object_to_all_deps)} object files"
)
def _get_object_dependencies(self, object_file):
"""Get all dependencies for a single object file using ninja -t deps."""
try:
# Run ninja -t deps for this object file
cmd = [self.ninja_executable, "-t", "deps", object_file]
cmd = [self.ninja_executable, "-t", "deps"]
result = subprocess.run(
cmd, cwd=self.build_dir, capture_output=True, text=True, timeout=30
cmd, cwd=self.build_dir, capture_output=True, text=True, timeout=120
)
if result.returncode != 0:
return []
if result.returncode != 0 and not result.stdout:
print(f"Warning: ninja -t deps returned code {result.returncode}")
if result.stderr:
print(f" stderr: {result.stderr.strip()}")
return
dependencies = []
lines = result.stdout.strip().split("\n")
# Parse the combined output: each block starts with an object name line,
# followed by indented dependency lines
ws_root = getattr(self, "workspace_root", "..")
ws_prefix = ws_root.rstrip("/") + "/"
for line in lines[1:]: # Skip first line with metadata
line = line.strip()
if line and not line.startswith("#"):
# Convert absolute paths to relative paths from workspace root
dep_file = line
ws_root = getattr(self, "workspace_root", "..")
ws_prefix = ws_root.rstrip("/") + "/"
if dep_file.startswith(ws_prefix):
dep_file = dep_file[len(ws_prefix) :]
dependencies.append(dep_file)
current_obj = None
current_deps = []
matched = 0
return dependencies
for line in result.stdout.split("\n"):
if not line:
continue
if not line.startswith(" ") and not line.startswith("\t"):
# This is an object header line like:
# some/path/foo.cpp.o: #deps 42, deps mtime ... (VALID)
# Save the previous block if it was relevant
if current_obj and current_obj in object_files:
self.object_to_all_deps[current_obj] = current_deps
matched += 1
if matched % 100 == 0:
print(f" Matched {matched} objects so far...")
# Parse the new object name (everything before the colon)
colon_pos = line.find(":")
if colon_pos > 0:
current_obj = line[:colon_pos].strip()
current_deps = []
else:
current_obj = None
current_deps = []
else:
# Indented dependency line
if current_obj is not None:
dep_file = line.strip()
if dep_file and not dep_file.startswith("#"):
# Strip workspace root prefix from absolute paths
if dep_file.startswith(ws_prefix):
dep_file = dep_file[len(ws_prefix):]
current_deps.append(dep_file)
# Don't forget the last block
if current_obj and current_obj in object_files:
self.object_to_all_deps[current_obj] = current_deps
matched += 1
except subprocess.TimeoutExpired:
print("Error: ninja -t deps timed out after 120 seconds")
return
except Exception as e:
print(f"Error getting dependencies for {object_file}: {e}")
return []
print(f"Error running ninja -t deps: {e}")
return
print(
f"Completed dependency extraction for {len(self.object_to_all_deps)} "
f"of {len(object_files)} object files"
)
def _build_file_to_executable_mapping(self):
"""Build the final mapping from files to executables."""
print("Building file-to-executable mapping...")
# For monorepo, truncate the path before and including projects/<project_name>
# This regex matches both absolute and relative monorepo paths
self.project = None
rl_regex = rf"rocm-libraries[\\/]+projects[\\/]+([^\\/]+)[\\/]+(.*)"
rl_regex = rf"(?:^|.*[\\/])projects[\\/]+([^\\/]+)[\\/]+(.*)"
for exe, object_files in self.executable_to_objects.items():
for obj_file in object_files:
# Add all dependencies of this object file
@@ -195,34 +208,51 @@ class EnhancedNinjaDependencyParser:
print(f" {f}: {len(exes)} executables")
def _is_project_file(self, file_path):
"""Determine if a file is part of the project (not system files)."""
# Include files that are clearly part of the project
if any(
file_path.startswith(prefix)
for prefix in [
"projects/composablekernel/include/",
"projects/composablekernel/library/",
"projects/composablekernel/test/",
"projects/composablekernel/example/",
"projects/composablekernel/src/",
"projects/composablekernel/profiler/",
"projects/composablekernel/build/include/",
"projects/composablekernel/build/_deps/gtest",
"projects/composablekernel/client_example",
"projects/composablekernel/codegen",
"projects/composablekernel/tile_engine",
]
):
return True
"""Determine if a file is part of the project (not system files).
# Exclude system files
Handles both standalone-style paths (e.g., include/ck/...) and
monorepo-style paths (e.g., projects/composablekernel/include/ck/...).
"""
# Exclude system files first (absolute paths to system dirs)
if any(
file_path.startswith(prefix)
for prefix in ["/usr/", "/opt/rocm", "/lib/", "/system/", "/local/"]
):
return False
# Include files with common source/header extensions
# Project directory prefixes (without monorepo prefix).
# These match paths relative to the CK project root.
project_dirs = [
"include/",
"library/",
"test/",
"example/",
"src/",
"profiler/",
"build/include/",
"build/_deps/gtest",
"client_example",
"codegen",
"tile_engine",
"dispatcher",
"experimental",
"tutorial",
]
# Check both stripped paths (relative to CK root) and
# monorepo-prefixed paths (relative to monorepo root)
if any(file_path.startswith(prefix) for prefix in project_dirs):
return True
# Also check monorepo-style paths (projects/composablekernel/...)
if any(
file_path.startswith(f"projects/composablekernel/{prefix}")
for prefix in project_dirs
):
return True
# Include files with common source/header extensions that weren't
# excluded as system files above
if file_path.endswith(
(".cpp", ".hpp", ".h", ".c", ".cc", ".cxx", ".cu", ".hip", ".inc")
):
@@ -344,7 +374,7 @@ def main():
sys.exit(1)
parser = EnhancedNinjaDependencyParser(build_file, ninja_path)
parser.workspace_root = workspace_root # Attach for use in _get_object_dependencies
parser.workspace_root = workspace_root # Attach for use in _extract_object_dependencies
parser.parse_dependencies()
parser.print_summary()

View File

@@ -34,15 +34,12 @@ import os
def get_changed_files(ref1, ref2, project: str = None):
"""Return a set of files changed between two git refs."""
try:
cmd = ["git", "diff", "--name-only", ref1, ref2]
if project:
# Scope git diff to only this project's subtree for efficiency
cmd += ["--", f"projects/{project}/"]
result = subprocess.run(
["pwd"],
capture_output=True,
text=True,
check=True,
)
print("cwd:", result.stdout)
result = subprocess.run(
["git", "diff", "--name-only", ref1, ref2],
cmd,
capture_output=True,
text=True,
check=True,
@@ -105,7 +102,7 @@ def main():
if not os.path.exists(depmap_json):
print(f"Dependency map JSON not found: {depmap_json}")
sys.exit(1)
file_to_executables = load_depmap(depmap_json)
file_to_executables, _ = load_depmap(depmap_json)
for f, exes in file_to_executables.items():
print(f"{f}: {', '.join(exes)}")
print(f"Total files: {len(file_to_executables)}")
@@ -122,7 +119,7 @@ def main():
if not os.path.exists(depmap_json):
print(f"Dependency map JSON not found: {depmap_json}")
sys.exit(1)
file_to_executables = load_depmap(depmap_json)
file_to_executables, _ = load_depmap(depmap_json)
affected_executables = set()
for f in changed_files:
if f in file_to_executables: