Build status for all platforms: 
Commercial support:
This directory contains the JavaCPP Presets module for:
- TVM 0.10.0 http://tvm.apache.org/
Please refer to the parent README.md file for more detailed information about the JavaCPP Presets.
Java API documentation is available here:
∗ Call Loader.load(org.bytedeco.tvm.presets.tvm_runtime.class) before using the API in the org.apache.tvm package.
∗ Call Py_Initialize(org.bytedeco.tvm.presets.tvm.cachePackages()) instead of just Py_Initialize().
Here is a simple example of TVM ported to Java from these source files in Python and C++:
We can use Maven 3 to download and install automatically all the class files as well as the native binaries. To run this sample code, after creating the pom.xml and HowtoDeploy.java source files below, simply execute on the command line:
$ mvn compile exec:java<project>
<modelVersion>4.0.0</modelVersion>
<groupId>org.bytedeco.tvm</groupId>
<artifactId>howtodeploy</artifactId>
<version>1.5.8</version>
<properties>
<exec.mainClass>HowtoDeploy</exec.mainClass>
</properties>
<dependencies>
<dependency>
<groupId>org.bytedeco</groupId>
<artifactId>tvm-platform</artifactId>
<version>0.10.0-1.5.8</version>
</dependency>
<!-- Additional dependencies required to use CUDA and cuDNN -->
<dependency>
<groupId>org.bytedeco</groupId>
<artifactId>tvm-platform-gpu</artifactId>
<version>0.10.0-1.5.8</version>
</dependency>
<!-- Additional dependencies to use bundled CUDA and cuDNN -->
<dependency>
<groupId>org.bytedeco</groupId>
<artifactId>cuda-platform-redist</artifactId>
<version>11.8-8.6-1.5.8</version>
</dependency>
<!-- Additional dependencies to use bundled full version of MKL -->
<dependency>
<groupId>org.bytedeco</groupId>
<artifactId>mkl-platform-redist</artifactId>
<version>2022.2-1.5.8</version>
</dependency>
</dependencies>
<build>
<sourceDirectory>.</sourceDirectory>
</build>
</project>/*
* 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.
*/
/*!
* \brief Example code on load and run TVM module.s
* \file cpp_deploy.cc
*/
import org.bytedeco.javacpp.*;
import org.bytedeco.cpython.*;
import org.bytedeco.numpy.*;
import org.bytedeco.tvm.*;
import org.bytedeco.tvm.Module;
import static org.bytedeco.cpython.global.python.*;
import static org.bytedeco.numpy.global.numpy.*;
import static org.bytedeco.tvm.global.tvm_runtime.*;
public class HowtoDeploy {
static void PrepareTestLibs() throws Exception {
String clang = Loader.load(org.bytedeco.llvm.program.clang.class).replace('\\', '/');
String clangPath = clang.substring(0, clang.lastIndexOf('/'));
Py_Initialize(org.bytedeco.tvm.presets.tvm.cachePackages());
if (_import_array() < 0) {
System.err.println("numpy.core.multiarray failed to import");
PyErr_Print();
System.exit(-1);
}
PyObject globals = PyModule_GetDict(PyImport_AddModule("__main__"));
PyRun_StringFlags("\"\"\"Script to prepare test_addone.so\"\"\"\n"
+ "import tvm\n"
+ "import numpy as np\n"
+ "from tvm import te\n"
+ "from tvm import relay\n"
+ "import os\n"
+ "def prepare_test_libs(base_path):\n"
+ " n = te.var(\"n\")\n"
+ " A = te.placeholder((n,), name=\"A\")\n"
+ " B = te.compute(A.shape, lambda *i: A(*i) + 1.0, name=\"B\")\n"
+ " s = te.create_schedule(B.op)\n"
+ " # Compile library as dynamic library\n"
+ " fadd_dylib = tvm.build(s, [A, B], \"llvm\", name=\"addone\")\n"
+ " dylib_path = os.path.join(base_path, \"test_addone_dll.so\")\n"
+ " fadd_dylib.export_library(dylib_path)\n"
+ " # Compile library in system library mode\n"
+ " fadd_syslib = tvm.build(s, [A, B], \"llvm\", name=\"addonesys\")\n"
+ " syslib_path = os.path.join(base_path, \"test_addone_sys.o\")\n"
+ " fadd_syslib.save(syslib_path)\n"
+ "def prepare_graph_lib(base_path):\n"
+ " x = relay.var(\"x\", shape=(2, 2), dtype=\"float32\")\n"
+ " y = relay.var(\"y\", shape=(2, 2), dtype=\"float32\")\n"
+ " params = {\"y\": np.ones((2, 2), dtype=\"float32\")}\n"
+ " mod = tvm.IRModule.from_expr(relay.Function([x, y], x + y))\n"
+ " # build a module\n"
+ " compiled_lib = relay.build(mod, tvm.target.create(\"llvm\"), params=params)\n"
+ " # export it as a shared library\n"
+ " # If you are running cross compilation, you can also consider export\n"
+ " # to tar and invoke host compiler later.\n"
+ " dylib_path = os.path.join(base_path, \"test_relay_add.so\")\n"
+ " compiled_lib.export_library(dylib_path)\n"
+ "if __name__ == \"__main__\":\n"
+ " lib_path = os.path.join(os.getcwd(), \"lib\")\n"
+ " os.makedirs(lib_path, exist_ok = True)\n"
+ " os.environ[\"PATH\"] += os.pathsep + \"" + clangPath + "\"\n"
+ " prepare_test_libs(lib_path)\n"
+ " prepare_graph_lib(lib_path)\n",
Py_file_input, globals, globals, null);
if (PyErr_Occurred() != null) {
System.err.println("Python error occurred");
PyErr_Print();
System.exit(-1);
}
}
static void Verify(Module mod, String fname) {
// Get the function from the module.
PackedFunc f = mod.GetFunction(fname);
assert f != null;
// Allocate the DLPack data structures.
//
// Note that we use TVM runtime API to allocate the DLTensor in this example.
// TVM accept DLPack compatible DLTensors, so function can be invoked
// as long as we pass correct pointer to DLTensor array.
//
// For more information please refer to dlpack.
// One thing to notice is that DLPack contains alignment requirement for
// the data pointer and TVM takes advantage of that.
// If you plan to use your customized data container, please
// make sure the DLTensor you pass in meet the alignment requirement.
//
DLTensor x = new DLTensor(null);
DLTensor y = new DLTensor(null);
int ndim = 1;
int dtype_code = kDLFloat;
int dtype_bits = 32;
int dtype_lanes = 1;
int device_type = kDLCPU;
int device_id = 0;
long[] shape = {10};
TVMArrayAlloc(shape, ndim, dtype_code, dtype_bits, dtype_lanes, device_type, device_id, x);
TVMArrayAlloc(shape, ndim, dtype_code, dtype_bits, dtype_lanes, device_type, device_id, y);
FloatPointer xdata = new FloatPointer(x.data());
FloatPointer ydata = new FloatPointer(y.data());
for (long i = 0; i < shape[0]; ++i) {
xdata.put(i, i);
}
// Invoke the function
// PackedFunc is a function that can be invoked via positional argument.
// The signature of the function is specified in tvm.build
TVMValue values = new TVMValue(2);
IntPointer codes = new IntPointer(2);
TVMArgsSetter setter = new TVMArgsSetter(values, codes);
setter.apply(0, x);
setter.apply(1, y);
f.CallPacked(new TVMArgs(values, codes, 2), null);
// Print out the output
for (long i = 0; i < shape[0]; ++i) {
float yi = ydata.get(i);
System.out.println(yi);
assert yi == i + 1.0f;
}
System.out.println("Finish verification...");
TVMArrayFree(x);
TVMArrayFree(y);
}
static void DeploySingleOp() {
// Normally we can directly
Module mod_dylib = Module.LoadFromFile("lib/test_addone_dll.so");
System.out.println("Verify dynamic loading from test_addone_dll.so");
Verify(mod_dylib, "addone");
// For libraries that are directly packed as system lib and linked together with the app
// We can directly use GetSystemLib to get the system wide library.
System.out.println("Verify load function from system lib");
TVMRetValue rv = new TVMRetValue();
Registry.Get("runtime.SystemLib").CallPacked(new TVMArgs((TVMValue)null, (IntPointer)null, 0), rv);
Module mod_syslib = rv.asModule();
// Verify(mod_syslib, "addonesys");
}
static void DeployGraphExecutor() {
System.out.println("Running graph executor...");
// load in the library
DLDevice dev = new DLDevice().device_type(kDLCPU).device_id(0);
Module mod_factory = Module.LoadFromFile("lib/test_relay_add.so");
// create the graph executor module
TVMValue values = new TVMValue(2);
IntPointer codes = new IntPointer(2);
TVMArgsSetter setter = new TVMArgsSetter(values, codes);
setter.apply(0, dev);
TVMRetValue rv = new TVMRetValue();
mod_factory.GetFunction("default").CallPacked(new TVMArgs(values, codes, 1), rv);
Module gmod = rv.asModule();
PackedFunc set_input = gmod.GetFunction("set_input");
PackedFunc get_output = gmod.GetFunction("get_output");
PackedFunc run = gmod.GetFunction("run");
// Use the C++ API
NDArray x = NDArray.Empty(new ShapeTuple(2, 2), new DLDataType().code((byte)kDLFloat).bits((byte)32).lanes((short)1), dev);
NDArray y = NDArray.Empty(new ShapeTuple(2, 2), new DLDataType().code((byte)kDLFloat).bits((byte)32).lanes((short)1), dev);
FloatPointer xdata = new FloatPointer(x.accessDLTensor().data());
FloatPointer ydata = new FloatPointer(y.accessDLTensor().data());
for (int i = 0; i < 2; ++i) {
for (int j = 0; j < 2; ++j) {
xdata.put(i * 2 + j, i * 2 + j);
}
}
// set the right input
setter.apply(0, new BytePointer("x"));
setter.apply(1, x);
set_input.CallPacked(new TVMArgs(values, codes, 2), rv);
// run the code
run.CallPacked(new TVMArgs(values, codes, 0), rv);
// get the output
setter.apply(0, 0);
setter.apply(1, y);
get_output.CallPacked(new TVMArgs(values, codes, 2), rv);
for (int i = 0; i < 2; ++i) {
for (int j = 0; j < 2; ++j) {
float yi = ydata.get(i * 2 + j);
System.out.println(yi);
assert yi == i * 2 + j + 1;
}
}
}
public static void main(String[] args) throws Exception {
/* try to use MKL when available */
System.setProperty("org.bytedeco.openblas.load", "mkl");
PrepareTestLibs();
DeploySingleOp();
DeployGraphExecutor();
System.exit(0);
}
}