Bohan Hou
859498dc01
## Summary This PR adds the initial TIRx support needed for low-level programming of Blackwell-class GPU architectures. As part of the ongoing TIRx refactor, it introduces TVMScript support for directly scripting advanced hardware features without relying on scheduling as the primary programming interface. The change keeps existing `s_tir` script support intact while making direct scripting a first-class path for TIRx programs. ## Main Changes - Add TIRx operator dispatch and layout infrastructure. - Add TVMScript support for new low-level TIRx operations. - Add analysis, transform, and lowering support for TIRx IR nodes. - Add CUDA/Blackwell-oriented codegen and intrinsic coverage. - Add Python and C++ integration points for TIRx scripting and runtime support. ## Validation - `pre-commit run --all-files` - `ninja -C build -j32` - `CUDA_VISIBLE_DEVICES=2 pytest tests/python/tirx/ -n 16` - `1723 passed, 47 skipped, 32 warnings` - `CUDA_VISIBLE_DEVICES=2 python -m pytest -v tests/python/all-platform-minimal-test` - `37 passed, 105 skipped` - `TVM_TEST_TARGETS=llvm python -m pytest -v tests/python/tirx-analysis tests/python/tirx-base tests/python/tirx-transform -n 16` - `664 passed, 25 skipped, 9 xfailed, 1 xpassed` ## Local CI Notes Some full CI-equivalent jobs were not locally reproducible because this machine is missing parts of the Apache TVM CI environment, including `llvm-config-15/17`, Vulkan, ROCm, Maven, Sphinx, Doxygen, Emscripten, and ARM/QEMU cross-toolchain components. Metal-specific tests were skipped locally because no Metal runtime is available.
264 lines
8.0 KiB
Python
264 lines
8.0 KiB
Python
# Licensed to the Apache Software Foundation (ASF) under one
|
|
# or more contributor license agreements. See the NOTICE file
|
|
# distributed with this work for additional information
|
|
# regarding copyright ownership. The ASF licenses this file
|
|
# to you under the Apache License, Version 2.0 (the
|
|
# "License"); you may not use this file except in compliance
|
|
# with the License. You may obtain a copy of the License at
|
|
#
|
|
# http://www.apache.org/licenses/LICENSE-2.0
|
|
#
|
|
# Unless required by applicable law or agreed to in writing,
|
|
# software distributed under the License is distributed on an
|
|
# "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
|
|
# KIND, either express or implied. See the License for the
|
|
# specific language governing permissions and limitations
|
|
# under the License.
|
|
"""Tests for the Executable class."""
|
|
|
|
import os
|
|
import tempfile
|
|
|
|
import numpy as np
|
|
|
|
import tvm
|
|
import tvm.testing
|
|
from tvm.runtime import Executable
|
|
from tvm.script import tirx as T
|
|
|
|
|
|
@tvm.script.ir_module
|
|
class MyModule:
|
|
@T.prim_func(s_tir=True)
|
|
def add(
|
|
A: T.Buffer((10,), "float32"),
|
|
B: T.Buffer((10,), "float32"),
|
|
C: T.Buffer((10,), "float32"),
|
|
):
|
|
for i in range(10):
|
|
C[i] = A[i] + B[i]
|
|
|
|
|
|
def test_executable_init():
|
|
"""Test initialization of Executable class."""
|
|
lib = tvm.tirx.build(MyModule, target="llvm")
|
|
executable = Executable(lib)
|
|
|
|
assert executable.mod is lib
|
|
assert executable._jitted_mod is None
|
|
|
|
|
|
def test_executable_getitem():
|
|
"""Test __getitem__ method of Executable class."""
|
|
lib = tvm.tirx.build(MyModule, target="llvm")
|
|
executable = Executable(lib)
|
|
|
|
# Jit the module first
|
|
executable.jit()
|
|
|
|
# Test __getitem__
|
|
add_func = executable["add"]
|
|
|
|
# Verify the function works
|
|
a = tvm.runtime.tensor(np.array([1.0] * 10, dtype="float32"))
|
|
b = tvm.runtime.tensor(np.array([2.0] * 10, dtype="float32"))
|
|
c = tvm.runtime.tensor(np.array([0.0] * 10, dtype="float32"))
|
|
|
|
add_func(a, b, c)
|
|
|
|
# Check results
|
|
tvm.testing.assert_allclose(c.numpy(), np.array([3.0] * 10, dtype="float32"))
|
|
|
|
|
|
def test_executable_jit_already_jitted():
|
|
"""Test jit method when module is already jitted."""
|
|
lib = tvm.tirx.build(MyModule, target="llvm")
|
|
executable = Executable(lib)
|
|
|
|
# First jit call
|
|
jitted_mod1 = executable.jit()
|
|
|
|
# Second jit call should return the cached jitted module
|
|
jitted_mod2 = executable.jit()
|
|
assert jitted_mod2 is jitted_mod1
|
|
|
|
# Test with force_recompile
|
|
jitted_mod3 = executable.jit(force_recompile=True)
|
|
# The module might be different after force recompilation
|
|
|
|
# Verify both modules work correctly
|
|
a = tvm.runtime.tensor(np.array([1.0] * 10, dtype="float32"))
|
|
b = tvm.runtime.tensor(np.array([2.0] * 10, dtype="float32"))
|
|
c1 = tvm.runtime.tensor(np.array([0.0] * 10, dtype="float32"))
|
|
c2 = tvm.runtime.tensor(np.array([0.0] * 10, dtype="float32"))
|
|
|
|
jitted_mod1["add"](a, b, c1)
|
|
jitted_mod3["add"](a, b, c2)
|
|
|
|
tvm.testing.assert_allclose(c1.numpy(), np.array([3.0] * 10, dtype="float32"))
|
|
tvm.testing.assert_allclose(c2.numpy(), np.array([3.0] * 10, dtype="float32"))
|
|
|
|
|
|
def test_executable_export_library():
|
|
"""Test export_library method."""
|
|
lib = tvm.tirx.build(MyModule, target="llvm")
|
|
executable = Executable(lib)
|
|
|
|
# Create a temporary directory for the library
|
|
temp_dir = tempfile.mkdtemp()
|
|
try:
|
|
lib_path = os.path.join(temp_dir, "test_lib.so")
|
|
executable.export_library(lib_path)
|
|
|
|
# Verify the library was created
|
|
assert os.path.exists(lib_path)
|
|
|
|
# Load the library back
|
|
loaded_mod = tvm.runtime.load_module(lib_path)
|
|
assert loaded_mod is not None
|
|
|
|
# Test the loaded module
|
|
a = tvm.runtime.tensor(np.array([1.0] * 10, dtype="float32"))
|
|
b = tvm.runtime.tensor(np.array([2.0] * 10, dtype="float32"))
|
|
c = tvm.runtime.tensor(np.array([0.0] * 10, dtype="float32"))
|
|
|
|
loaded_mod["add"](a, b, c)
|
|
|
|
# Check results
|
|
tvm.testing.assert_allclose(c.numpy(), np.array([3.0] * 10, dtype="float32"))
|
|
finally:
|
|
# Clean up
|
|
if os.path.exists(temp_dir):
|
|
import shutil
|
|
|
|
shutil.rmtree(temp_dir)
|
|
|
|
|
|
def test_executable_export_library_with_workspace():
|
|
"""Test export_library method with workspace_dir."""
|
|
lib = tvm.tirx.build(MyModule, target="llvm")
|
|
executable = Executable(lib)
|
|
|
|
# Create temporary directories
|
|
temp_dir = tempfile.mkdtemp()
|
|
workspace_dir = tempfile.mkdtemp()
|
|
|
|
try:
|
|
lib_path = os.path.join(temp_dir, "test_lib.so")
|
|
executable.export_library(lib_path, workspace_dir=workspace_dir)
|
|
|
|
# Verify the library was created
|
|
assert os.path.exists(lib_path)
|
|
|
|
# Load the library back
|
|
loaded_mod = tvm.runtime.load_module(lib_path)
|
|
assert loaded_mod is not None
|
|
|
|
# Test the loaded module
|
|
a = tvm.runtime.tensor(np.array([1.0] * 10, dtype="float32"))
|
|
b = tvm.runtime.tensor(np.array([2.0] * 10, dtype="float32"))
|
|
c = tvm.runtime.tensor(np.array([0.0] * 10, dtype="float32"))
|
|
|
|
loaded_mod["add"](a, b, c)
|
|
|
|
# Check results
|
|
tvm.testing.assert_allclose(c.numpy(), np.array([3.0] * 10, dtype="float32"))
|
|
finally:
|
|
# Clean up
|
|
for directory in [temp_dir, workspace_dir]:
|
|
if os.path.exists(directory):
|
|
import shutil
|
|
|
|
shutil.rmtree(directory)
|
|
|
|
|
|
def test_executable_integration():
|
|
"""Integration test for Executable with a simple TVM module."""
|
|
# Create target and build
|
|
target = tvm.target.Target("llvm")
|
|
lib = tvm.tirx.build(MyModule, target=target)
|
|
|
|
# Create an executable
|
|
executable = Executable(lib)
|
|
|
|
# Test jit
|
|
jitted_mod = executable.jit()
|
|
assert jitted_mod is not None
|
|
|
|
# Test __getitem__
|
|
add_func = executable["add"]
|
|
assert add_func is not None
|
|
|
|
# Test the function works
|
|
a = tvm.runtime.tensor(np.array([1.0] * 10, dtype="float32"))
|
|
b = tvm.runtime.tensor(np.array([2.0] * 10, dtype="float32"))
|
|
c = tvm.runtime.tensor(np.array([0.0] * 10, dtype="float32"))
|
|
|
|
add_func(a, b, c)
|
|
|
|
# Check results
|
|
tvm.testing.assert_allclose(c.numpy(), np.array([3.0] * 10, dtype="float32"))
|
|
|
|
# Test export_library
|
|
temp_dir = tempfile.mkdtemp()
|
|
try:
|
|
lib_path = os.path.join(temp_dir, "test_lib.so")
|
|
executable.export_library(lib_path)
|
|
|
|
# Verify the library was created
|
|
assert os.path.exists(lib_path)
|
|
|
|
# Load the library back
|
|
loaded_mod = tvm.runtime.load_module(lib_path)
|
|
assert loaded_mod is not None
|
|
|
|
# Test the loaded module
|
|
loaded_add = loaded_mod["add"]
|
|
c_loaded = tvm.runtime.tensor(np.array([0.0] * 10, dtype="float32"))
|
|
loaded_add(a, b, c_loaded)
|
|
|
|
# Check results
|
|
tvm.testing.assert_allclose(c_loaded.numpy(), np.array([3.0] * 10, dtype="float32"))
|
|
|
|
finally:
|
|
# Clean up
|
|
if os.path.exists(temp_dir):
|
|
import shutil
|
|
|
|
shutil.rmtree(temp_dir)
|
|
|
|
|
|
def test_executable_jit_force_recompile():
|
|
"""Test jit method with force_recompile=True."""
|
|
# Create target and build
|
|
target = tvm.target.Target("c")
|
|
lib = tvm.tirx.build(MyModule, target=target)
|
|
|
|
# Create an executable
|
|
executable = Executable(lib)
|
|
|
|
# First jit call
|
|
jitted_mod1 = executable.jit()
|
|
|
|
# Second jit call without force_recompile should return the same module
|
|
jitted_mod2 = executable.jit()
|
|
assert jitted_mod1 is jitted_mod2
|
|
|
|
# Third jit call with force_recompile should return a new module
|
|
jitted_mod3 = executable.jit(force_recompile=True)
|
|
assert jitted_mod3 is not jitted_mod1
|
|
|
|
# Test the function works
|
|
a = tvm.runtime.tensor(np.array([1.0] * 10, dtype="float32"))
|
|
b = tvm.runtime.tensor(np.array([2.0] * 10, dtype="float32"))
|
|
c = tvm.runtime.tensor(np.array([0.0] * 10, dtype="float32"))
|
|
|
|
jitted_mod3["add"](a, b, c)
|
|
|
|
# Check results
|
|
tvm.testing.assert_allclose(c.numpy(), np.array([3.0] * 10, dtype="float32"))
|
|
|
|
|
|
if __name__ == "__main__":
|
|
tvm.testing.main()
|