mirror of
https://github.com/vllm-project/vllm.git
synced 2026-10-28 09:45:26 +08:00
debcf22843
Signed-off-by: Andreas Karatzas <andreas.karatzas@protonmail.com> Signed-off-by: Andreas Karatzas <akaratza@amd.com> Signed-off-by: Andreas Karatzas <AndreasKaratzas@users.noreply.github.com> Co-authored-by: Codex <noreply@openai.com> Co-authored-by: Claude <noreply@anthropic.com> Co-authored-by: Andreas Karatzas <AndreasKaratzas@users.noreply.github.com>
116 lines
3.8 KiB
Python
116 lines
3.8 KiB
Python
# SPDX-License-Identifier: Apache-2.0
|
|
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
|
|
|
|
import json
|
|
import shutil
|
|
import subprocess
|
|
import sys
|
|
from pathlib import Path
|
|
|
|
|
|
def _get_cmake_bin() -> str:
|
|
cmake = shutil.which("cmake")
|
|
if cmake:
|
|
return cmake
|
|
venv_cmake = Path(sys.executable).parent / "cmake"
|
|
if venv_cmake.is_file():
|
|
return str(venv_cmake)
|
|
return "cmake"
|
|
|
|
|
|
def test_exact_family_arch_precedes_generic_family_fallback(tmp_path: Path):
|
|
repo_root = Path(__file__).parents[1]
|
|
script = tmp_path / "test_cuda_archs.cmake"
|
|
script.write_text(
|
|
f"""
|
|
cmake_minimum_required(VERSION 3.26)
|
|
include("{repo_root / "cmake" / "utils.cmake"}")
|
|
cuda_archs_loose_intersection(
|
|
actual "10.0f;10.7f" "10.7")
|
|
if(NOT "${{actual}}" STREQUAL "10.7f")
|
|
message(FATAL_ERROR "Expected 10.7f, got '${{actual}}'")
|
|
endif()
|
|
"""
|
|
)
|
|
|
|
subprocess.run([_get_cmake_bin(), "-P", script], check=True)
|
|
|
|
|
|
def test_extract_archs_prefers_sass_target_over_corrupted_virtual_arch(
|
|
tmp_path: Path,
|
|
):
|
|
"""Torch's autodetection can emit a bogus arch=compute_* half (e.g.
|
|
capability 12.1 corrupted to arch=compute_20,code=sm_121); the SASS
|
|
target must win, while PTX-only entries keep the virtual arch."""
|
|
repo_root = Path(__file__).parents[1]
|
|
script = tmp_path / "test_extract_archs.cmake"
|
|
script.write_text(
|
|
f"""
|
|
cmake_minimum_required(VERSION 3.26)
|
|
include("{repo_root / "cmake" / "utils.cmake"}")
|
|
extract_unique_cuda_archs_ascending(actual
|
|
"-gencode arch=compute_20,code=sm_121;\
|
|
-gencode arch=compute_80,code=sm_80;\
|
|
-gencode arch=compute_80,code=compute_80")
|
|
if(NOT "${{actual}}" STREQUAL "8.0;12.1")
|
|
message(FATAL_ERROR "Expected '8.0;12.1', got '${{actual}}'")
|
|
endif()
|
|
"""
|
|
)
|
|
|
|
subprocess.run([_get_cmake_bin(), "-P", script], check=True)
|
|
|
|
|
|
def test_clear_cuda_gencode_flags(tmp_path: Path):
|
|
repo_root = Path(__file__).parents[1]
|
|
script = tmp_path / "test_clear_flags.cmake"
|
|
script.write_text(
|
|
f"""
|
|
cmake_minimum_required(VERSION 3.26)
|
|
include("{repo_root / "cmake" / "utils.cmake"}")
|
|
set(CMAKE_CUDA_FLAGS "-Wall -gencode arch=compute_80,code=sm_80")
|
|
clear_cuda_gencode_flags(CUDA_ARCH_FLAGS)
|
|
if(NOT "${{CMAKE_CUDA_FLAGS}}" STREQUAL "-Wall ")
|
|
message(FATAL_ERROR "Expected '-Wall ', got '${{CMAKE_CUDA_FLAGS}}'")
|
|
endif()
|
|
if(NOT "${{CUDA_ARCH_FLAGS}}" STREQUAL "-gencode arch=compute_80,code=sm_80")
|
|
message(FATAL_ERROR "Expected '-gencode arch=compute_80,code=sm_80', "
|
|
"got '${{CUDA_ARCH_FLAGS}}'")
|
|
endif()
|
|
"""
|
|
)
|
|
|
|
subprocess.run([_get_cmake_bin(), "-P", script], check=True)
|
|
|
|
|
|
def test_hipify_records_original_sources_and_headers_without_modifying_them(tmp_path):
|
|
project = tmp_path / "source"
|
|
output = tmp_path / "build" / "csrc"
|
|
project.mkdir()
|
|
source = project / "probe.cu"
|
|
header = project / "cuda_shared.h"
|
|
source.write_text('#include "cuda_shared.h"\n__global__ void probe() {}\n')
|
|
header.write_text("#include <cuda_runtime.h>\n")
|
|
original = {path.name: path.read_bytes() for path in project.iterdir()}
|
|
script = Path(__file__).parents[1] / "cmake" / "hipify.py"
|
|
subprocess.run(
|
|
[
|
|
sys.executable,
|
|
str(script),
|
|
"-p",
|
|
str(project),
|
|
"-o",
|
|
str(output),
|
|
str(source),
|
|
],
|
|
check=True,
|
|
capture_output=True,
|
|
text=True,
|
|
)
|
|
assert {path.name: path.read_bytes() for path in project.iterdir()} == original
|
|
provenance = json.loads((output / "hipify-source-map.json").read_text())
|
|
assert provenance[str(output / "probe.hip")] == str(source)
|
|
assert provenance[str(output / "cuda_shared.h")] == str(header)
|
|
assert provenance[str(output / "hip_shared.h")] == str(header)
|
|
assert '#include "hip_shared.h"' in (output / "probe.hip").read_text()
|