Skip to content

生成唯一 MACA kernel dump 路径#28

Open
ghangz wants to merge 1 commit into
MetaX-MACA:mainfrom
ghangz:mengz/unique-maca-kernel-dumps
Open

生成唯一 MACA kernel dump 路径#28
ghangz wants to merge 1 commit into
MetaX-MACA:mainfrom
ghangz:mengz/unique-maca-kernel-dumps

Conversation

@ghangz

@ghangz ghangz commented Jun 8, 2026

Copy link
Copy Markdown

该 PR 为 MACA kernel dump 生成唯一输出路径,避免并发测试或多次运行互相覆盖调试产物。

这个修改面向沐曦 GPU 适配场景中比较容易影响开发、构建或验证稳定性的环节,把原来需要人工排查的问题前移到工具链、运行前检查或基准脚本中处理。实现上保持对现有默认行为的兼容,只在检测到明确配置、输入或环境异常时给出更直接的诊断,避免引入额外运行依赖,也方便维护者独立审阅该分支。

已在沐曦算力环境中完成对应分支验证,验证记录包含真实运行日志、命令输出和失败路径检查,本地归档目录为:E:/Documents/muxi/测试报告/mcTVM_new_toolchain_validation_20260608。提交分支:mengz/unique-maca-kernel-dumps,目标仓库:MetaX-MACA/mcTVM

@gemini-code-assist gemini-code-assist Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Code Review

This pull request introduces a stable, content-addressed naming scheme for kernel dump files in the MXCC compiler integration using SHA256 hashing, along with a new unit test suite. Feedback on these changes suggests avoiding the risky mocking of sys.modules in the test file by directly importing mxcc from tvm.contrib, and defensively handling cases where the input code is bytes or None in _kernel_file_name to prevent potential runtime errors.

Important

The consumer version of Gemini Code Assist on GitHub is being sunset. Starting June 18, 2026, new organization installations will be blocked, and all code review activity will officially cease on July 17, 2026.
For more details on the timeline and next steps, please review the Help Documentation.

Comment on lines +1 to +43
import hashlib
import importlib.util
import sys
import types
import unittest
from pathlib import Path


REPO_ROOT = Path(__file__).resolve().parents[3]
MXCC_PATH = REPO_ROOT / "python" / "tvm" / "contrib" / "mxcc.py"

tvm_ffi = types.ModuleType("tvm_ffi")
tvm_ffi.register_global_func = lambda *args, **kwargs: (
(lambda func: func) if args and callable(args[0]) else (lambda func: func)
)
tvm_ffi.get_global_func = lambda _name: None
sys.modules["tvm_ffi"] = tvm_ffi

tvm = types.ModuleType("tvm")
tvm.maca = lambda *_args, **_kwargs: types.SimpleNamespace(exist=False)
tvm.target = types.ModuleType("tvm.target")
tvm.target.Target = object
sys.modules["tvm"] = tvm
sys.modules["tvm.target"] = tvm.target

tvm_base = types.ModuleType("tvm.base")
tvm_base.py_str = lambda value: value.decode("utf-8") if isinstance(value, bytes) else str(value)
sys.modules["tvm.base"] = tvm_base

tvm_contrib = types.ModuleType("tvm.contrib")
tvm_contrib.__path__ = []
sys.modules["tvm.contrib"] = tvm_contrib

tvm_contrib_utils = types.ModuleType("tvm.contrib.utils")
tvm_contrib_utils.tempdir = lambda: None
sys.modules["tvm.contrib.utils"] = tvm_contrib_utils

spec = importlib.util.spec_from_file_location("tvm.contrib.mxcc", MXCC_PATH)
mxcc = importlib.util.module_from_spec(spec)
mxcc.__package__ = "tvm.contrib"
sys.modules["tvm.contrib.mxcc"] = mxcc
assert spec.loader is not None
spec.loader.exec_module(mxcc)

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

high

在测试文件中通过修改 sys.modules 来 mock tvm 及其相关子模块是一种非常危险的做法。如果该测试与其他测试在同一个 Python 进程中运行(例如使用 pytest 运行整个测试套件),这会全局污染 sys.modules,导致后续执行的真实 TVM 测试因获取到 mock 对象而失败。

由于在 TVM 的测试环境中,真实的 tvm 库已经是可用且配置好的,因此无需进行任何 mock。可以直接从 tvm.contrib 导入 mxcc 模块进行测试。

Suggested change
import hashlib
import importlib.util
import sys
import types
import unittest
from pathlib import Path
REPO_ROOT = Path(__file__).resolve().parents[3]
MXCC_PATH = REPO_ROOT / "python" / "tvm" / "contrib" / "mxcc.py"
tvm_ffi = types.ModuleType("tvm_ffi")
tvm_ffi.register_global_func = lambda *args, **kwargs: (
(lambda func: func) if args and callable(args[0]) else (lambda func: func)
)
tvm_ffi.get_global_func = lambda _name: None
sys.modules["tvm_ffi"] = tvm_ffi
tvm = types.ModuleType("tvm")
tvm.maca = lambda *_args, **_kwargs: types.SimpleNamespace(exist=False)
tvm.target = types.ModuleType("tvm.target")
tvm.target.Target = object
sys.modules["tvm"] = tvm
sys.modules["tvm.target"] = tvm.target
tvm_base = types.ModuleType("tvm.base")
tvm_base.py_str = lambda value: value.decode("utf-8") if isinstance(value, bytes) else str(value)
sys.modules["tvm.base"] = tvm_base
tvm_contrib = types.ModuleType("tvm.contrib")
tvm_contrib.__path__ = []
sys.modules["tvm.contrib"] = tvm_contrib
tvm_contrib_utils = types.ModuleType("tvm.contrib.utils")
tvm_contrib_utils.tempdir = lambda: None
sys.modules["tvm.contrib.utils"] = tvm_contrib_utils
spec = importlib.util.spec_from_file_location("tvm.contrib.mxcc", MXCC_PATH)
mxcc = importlib.util.module_from_spec(spec)
mxcc.__package__ = "tvm.contrib"
sys.modules["tvm.contrib.mxcc"] = mxcc
assert spec.loader is not None
spec.loader.exec_module(mxcc)
import hashlib
import unittest
from tvm.contrib import mxcc

Comment on lines +35 to +38
def _kernel_file_name(code):
"""Return a stable kernel dump filename for the given MACA source."""
digest = hashlib.sha256(code.encode("utf-8")).hexdigest()[:16]
return f"tvm_kernels_{digest}"

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

medium

为了提高代码的健壮性(防御性编程),建议支持 codebytesNone 的情况。如果 code 已经是 bytes 类型,直接调用 .encode("utf-8") 会抛出 AttributeError

可以使用更兼容的方式来获取字节流。

Suggested change
def _kernel_file_name(code):
"""Return a stable kernel dump filename for the given MACA source."""
digest = hashlib.sha256(code.encode("utf-8")).hexdigest()[:16]
return f"tvm_kernels_{digest}"
def _kernel_file_name(code):
"""Return a stable kernel dump filename for the given MACA source."""
code_bytes = code if isinstance(code, bytes) else str(code or "").encode("utf-8")
digest = hashlib.sha256(code_bytes).hexdigest()[:16]
return f"tvm_kernels_{digest}"

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant