There are two ways to get a panama foreign branch JDK.
Using foreign function call in Java involves the following three steps:
You will (almost always) need to have Visual Studio installed, since most libraries indirectly depend on Visual Studio runtime libraries and this currently means that jextract needs their header to extract successfully. Windows examples have been tested with Build Tools for Visual Studio 2017.
It is generally a good idea to give jextract a bunch of extra memory since a lot of big system headers are transitively included. The extra memory will make the jextract run significantly faster. Windows support was added only recently, and the memory usage of jextract has not been optimized yet, so this is a workaround. You can give extra memory by passing e.g. -J-Xmx8G to jextract as an additional argument, which in this example gives jextract 8 gigabytes of memory.
Commands are tested in PowerShell.
jextract -l python2.7 \
-L /System/Library/Frameworks/Python.framework/Versions/2.7/lib --record-library-path \
--exclude-symbols .*_FromFormatV\|_.*\|PyOS_vsnprintf\|.*_VaParse.*\|.*_VaBuild.*\|PyBuffer_SizeFromFormat\|vasprintf\|vfprintf\|vprintf\|vsprintf \
-t org.python \
/usr/include/python2.7/Python.h \
-o python.jar
// import java.foreign packages
import java.foreign.Libraries;
import java.foreign.Scope;
import java.foreign.memory.Pointer;
// import jextracted python 'header' classes
import static org.python.Python_h.*;
import static org.python.pythonrun_h.*;
public class PythonMain {
public static void main(String[] args) {
Py_Initialize();
try (Scope s = org.python.Python_h.scope().fork()) {
PyRun_SimpleStringFlags(s.allocateCString(
"print(sum([33, 55, 66])); print('Hello from Python!')\n"),
Pointer.ofNull());
}
Py_Finalize();
}
}
jextract -l python2.7 \
-L /System/Library/Frameworks/Python.framework/Versions/2.7/lib \
--exclude-symbols .*_FromFormatV\|_.*\|PyOS_vsnprintf\|.*_VaParse.*\|.*_VaBuild.*\|PyBuffer_SizeFromFormat\|vasprintf\|vfprintf\|vprintf\|vsprintf \
-t org.python \
/usr/include/python2.7/Python.h \
-o org.python.jmodjdk.compiler and org.python modules are added to the generated (jlinked) JDK
In the following commands, it is assumed that you've put $pythonjdk/bin in your $PATH
jextract -l python2.7 \
-L /usr/lib/python2.7/config-x86_64-linux-gnu --record-library-path \
--exclude-symbols .*_FromFormatV\|_.*\|PyOS_vsnprintf\|.*_VaParse.*\|.*_VaBuild.*\|PyBuffer_SizeFromFormat\|vasprintf\|vfprintf\|vprintf\|vsprintf \
-t org.python \
/usr/include/python2.7/Python.h \
-o python.jarFollow the instructions from the Mac OS section
Where python 2.7 is installed in the C:\Python27 directory:
jextract /usr/include/sqlite3.h -t org.sqlite -lsqlite3 \
-L /usr/lib --record-library-path \
--exclude-symbols sqlite3_vmprintf \
--exclude-symbols sqlite3_vsnprintf \
-o sqlite3.jar
import java.lang.invoke.*;
import java.foreign.*;
import java.foreign.memory.*;
import org.sqlite.sqlite3.*;
import static org.sqlite.sqlite3_h.*;
public class SqliteMain {
public static void main(String[] args) throws Exception {
try (Scope scope = scope().fork()) {
// char* errMsg;
Pointer<Pointer<Byte>> errMsg = scope.allocate(NativeTypes.INT8.pointer());
// sqlite3* db;
Pointer<Pointer<sqlite3>> db = scope.allocate(LayoutType.ofStruct(sqlite3.class).pointer());
int rc = sqlite3_open(scope.allocateCString("employee.db"), db);
if (rc != 0) {
System.err.println("sqlite3_open failed: " + rc);
return;
}
// create a new table
Pointer<Byte> sql = scope.allocateCString(
"CREATE TABLE EMPLOYEE (" +
" ID INT PRIMARY KEY NOT NULL," +
" NAME TEXT NOT NULL," +
" SALARY REAL NOT NULL )"
);
rc = sqlite3_exec(db.get(), sql, Callback.ofNull(), Pointer.ofNull(), errMsg);
if (rc != 0) {
System.err.println("sqlite3_exec failed: " + rc);
System.err.println("SQL error: " + Pointer.toString(errMsg.get()));
sqlite3_free(errMsg.get());
}
// insert two rows
sql = scope.allocateCString(
"INSERT INTO EMPLOYEE (ID,NAME,SALARY) " +
"VALUES (134, 'Xyz', 200000.0); " +
"INSERT INTO EMPLOYEE (ID,NAME,SALARY) " +
"VALUES (333, 'Abc', 100000.0);"
);
rc = sqlite3_exec(db.get(), sql, Callback.ofNull(), Pointer.ofNull(), errMsg);
if (rc != 0) {
System.err.println("sqlite3_exec failed: " + rc);
System.err.println("SQL error: " + Pointer.toString(errMsg.get()));
sqlite3_free(errMsg.get());
}
int[] rowNum = new int[1];
// callback to print rows from SELECT query
Callback<FI1> callback = scope.allocateCallback(FI1.class, (a, argc, argv, columnNames) -> {
System.out.println("Row num: " + rowNum[0]++);
System.out.println("numColumns = " + argc);
for (int i = 0; i < argc; i++) {
String name = Pointer.toString(columnNames.offset(i).get());
String value = Pointer.toString(argv.offset(i).get());
System.out.printf("%s = %s\n", name, value);
}
return 0;
});
// select query
sql = scope.allocateCString("SELECT * FROM EMPLOYEE");
rc = sqlite3_exec(db.get(), sql, callback, Pointer.ofNull(), errMsg);
if (rc != 0) {
System.err.println("sqlite3_exec failed: " + rc);
System.err.println("SQL error: " + Pointer.toString(errMsg.get()));
sqlite3_free(errMsg.get());
}
sqlite3_close(db.get());
}
}
}On Ubuntu (16.04) to install sqlite3 headers and libraries the following command is required:
This should install the sqlite3 header (under /usr/include), as well as the sqlite3 shared library (under /usr/lib/x86_64-linux-gnu).
To extract sqlite, run the following command:
jextract /usr/include/sqlite3.h -t org.sqlite -lsqlite3 \
-L /usr/lib/x86_64-linux-gnu --record-library-path \
--exclude-symbols sqlite3_vmprintf \
--exclude-symbols sqlite3_vsnprintf \
-o sqlite3.jarPlease refer to the Mac OS instructions; once the library as been extracted (as per the instructions above), the sample program shown in that section should work on Ubuntu as well.
BLAS is a popular library that allows fast matrix and vector computation: http://www.netlib.org/blas/.
On Mac, blas is available as part of the OpenBLAS library: https://github.com/xianyi/OpenBLAS/wiki
OpenBLAS is an optimized BLAS library based on GotoBLAS2 1.13 BSD version.
You can install openblas using HomeBrew
It installs include and lib directories under /usr/local/opt/openblas
On Ubuntu, blas is distributed as part of the atlas library: http://math-atlas.sourceforge.net/.
You can install atlas using apt
This command will install include files under /usr/include/atlas and corresponding libraries under /usr/lib/atlas-dev.
The following command can be used to extract cblas.h on MacOs
jextract -C "-D FORCE_OPENBLAS_COMPLEX_STRUCT" \
-L /usr/local/opt/openblas/lib -I /usr/local/opt/openblas \
-l openblas -t blas --record-library-path /usr/local/opt/openblas/include/cblas.h \
-o cblas.jarThe FORCE_OPENBLAS_COMPLEX_STRUCT define is needed because jextract does not yet handle C99 _Complex types. The rest of the options are standard ones.
The following command can be used to extract cblas.h on Ubuntu
jextract -L /usr/lib/atlas-base -I /usr/include/atlas/ \
-l cblas -t blas --record-library-path \
/usr/include/atlas/cblas.h -o cblas.jar
import blas.cblas;
import static blas.cblas_h.*;
import static blas.cblas_h.CBLAS_ORDER.*;
import static blas.cblas_h.CBLAS_TRANSPOSE.*;
import java.foreign.NativeTypes;
import java.foreign.Scope;
import java.foreign.memory.Array;
public class TestBlas {
public static void main(String[] args) {
@cblas.CBLAS_ORDER int Layout;
@cblas.CBLAS_TRANSPOSE int transa;
double alpha, beta;
int m, n, lda, incx, incy, i;
Layout = CblasColMajor;
transa = CblasNoTrans;
m = 4; /* Size of Column ( the number of rows ) */
n = 4; /* Size of Row ( the number of columns ) */
lda = 4; /* Leading dimension of 5 * 4 matrix is 5 */
incx = 1;
incy = 1;
alpha = 1;
beta = 0;
try (Scope sc = scope().fork()){
Array<Double> a = sc.allocateArray(NativeTypes.DOUBLE, m * n);
Array<Double> x = sc.allocateArray(NativeTypes.DOUBLE, n);
Array<Double> y = sc.allocateArray(NativeTypes.DOUBLE, n);
/* The elements of the first column */
a.set(0, 1.0);
a.set(1, 2.0);
a.set(2, 3.0);
a.set(3, 4.0);
/* The elements of the second column */
a.set(m, 1.0);
a.set(m + 1, 1.0);
a.set(m + 2, 1.0);
a.set(m + 3, 1.0);
/* The elements of the third column */
a.set(m * 2, 3.0);
a.set(m * 2 + 1, 4.0);
a.set(m * 2 + 2, 5.0);
a.set(m * 2 + 3, 6.0);
/* The elements of the fourth column */
a.set(m * 3, 5.0);
a.set(m * 3 + 1, 6.0);
a.set(m * 3 + 2, 7.0);
a.set(m * 3 + 3, 8.0);
/* The elemetns of x and y */
x.set(0, 1.0);
x.set(1, 2.0);
x.set(2, 1.0);
x.set(3, 1.0);
y.set(0, 0.0);
y.set(1, 0.0);
y.set(2, 0.0);
y.set(3, 0.0);
cblas_dgemv(Layout, transa, m, n, alpha, a.elementPointer(), lda, x.elementPointer(), incx, beta,
y.elementPointer(), incy);
/* Print y */
for (i = 0; i < n; i++)
System.out.print(String.format(" y%d = %f\n", i, y.get(i)));
}
}
}On Ubuntu, the same steps used to install the blas (via atlas) library also install headers and libraries for the LAPACK library, a linear algebra computation library built on top of blas.
The following command can be used to extract the LAPACK header:
jextract -L /usr/lib/atlas-base/atlas -I /usr/include/atlas/ \
-l lapack -t lapack --record-library-path /usr/include/atlas/clapack.h -o clapack.jarimport java.foreign.NativeTypes;
import java.foreign.Scope;
import java.foreign.memory.Array;
import static lapack.clapack_h.*;
import static lapack.cblas_h.*;
public class TestLapack {
public static void main(String[] args) {
/* Locals */
try (Scope sc = lapack.clapack_h.scope().fork()) {
Array<Double> A = sc.allocateArray(NativeTypes.DOUBLE, new double[]{
1, 2, 3, 4, 5, 1, 3, 5, 2, 4, 1, 4, 2, 5, 3
});
Array<Double> b = sc.allocateArray(NativeTypes.DOUBLE, new double[]{
-10, 12, 14, 16, 18, -3, 14, 12, 16, 16
});
int info, m, n, lda, ldb, nrhs;
/* Initialization */
m = 5;
n = 3;
nrhs = 2;
lda = 5;
ldb = 5;
/* Print Entry Matrix */
print_matrix_colmajor("Entry Matrix A", m, n, A, lda );
/* Print Right Rand Side */
print_matrix_colmajor("Right Hand Side b", n, nrhs, b, ldb );
System.out.println();
/* Executable statements */
// printf( "LAPACKE_dgels (col-major, high-level) Example Program Results\n" );
/* Solve least squares problem*/
info = clapack_dgels(CblasColMajor, CblasNoTrans, m, n, nrhs, A.elementPointer(), lda, b.elementPointer(), ldb);
/* Print Solution */
print_matrix_colmajor("Solution", n, nrhs, b, ldb );
System.out.println();
System.exit(info);
}
}
static void print_matrix_colmajor(String msg, int m, int n, Array<Double> mat, int ldm) {
int i, j;
System.out.printf("\n %s\n", msg);
for( i = 0; i < m; i++ ) {
for( j = 0; j < n; j++ ) System.out.printf(" %6.2f", mat.get(i+j*ldm));
System.out.printf( "\n" );
}
}
}On Mac OS, lapack is installed under /usr/local/opt/lapack directory.
jextract -L /usr/local/opt/lapack/lib -I /usr/local/opt/lapack/ \
-l lapacke -t lapack --record-library-path /usr/local/opt/lapack/include/lapacke.h -o clapack.jar
import java.foreign.NativeTypes;
import java.foreign.Scope;
import java.foreign.memory.Array;
import static lapack.lapacke_h.*;
public class TestLapack {
public static void main(String[] args) {
/* Locals */
try (Scope sc = scope().fork()) {
Array<Double> A = sc.allocateArray(NativeTypes.DOUBLE, new double[]{
1, 2, 3, 4, 5, 1, 3, 5, 2, 4, 1, 4, 2, 5, 3
});
Array<Double> b = sc.allocateArray(NativeTypes.DOUBLE, new double[]{
-10, 12, 14, 16, 18, -3, 14, 12, 16, 16
});
int info, m, n, lda, ldb, nrhs;
/* Initialization */
m = 5;
n = 3;
nrhs = 2;
lda = 5;
ldb = 5;
/* Print Entry Matrix */
print_matrix_colmajor("Entry Matrix A", m, n, A, lda );
/* Print Right Rand Side */
print_matrix_colmajor("Right Hand Side b", n, nrhs, b, ldb );
System.out.println();
/* Executable statements */
// printf( "LAPACKE_dgels (col-major, high-level) Example Program Results\n" );
/* Solve least squares problem*/
info = LAPACKE_dgels(LAPACK_COL_MAJOR, (byte)'N', m, n, nrhs, A.elementPointer(), lda, b.elementPointer(), ldb);
/* Print Solution */
print_matrix_colmajor("Solution", n, nrhs, b, ldb );
System.out.println();
System.exit(info);
}
}
static void print_matrix_colmajor(String msg, int m, int n, Array<Double> mat, int ldm) {
int i, j;
System.out.printf("\n %s\n", msg);
for( i = 0; i < m; i++ ) {
for( j = 0; j < n; j++ ) System.out.printf(" %6.2f", mat.get(i+j*ldm));
System.out.printf( "\n" );
}
}
}jextract -t org.unix -lproc -L /usr/lib --record-library-path -o libproc.jar /usr/include/libproc.h
import java.foreign.*;
import java.foreign.memory.*;
import static org.unix.libproc_h.*;
public class LibprocMain {
private static final int NAME_BUF_MAX = 256;
public static void main(String[] args) {
// Scope for native allocations
try (Scope s = scope().fork()) {
// get the number of processes
int numPids = proc_listallpids(Pointer.ofNull(), 0);
// allocate an array
Array<Integer> pids = s.allocateArray(NativeTypes.INT32, numPids);
// list all the pids into the native array
proc_listallpids(pids.elementPointer(), numPids);
// convert native array to java array
int[] jpids = pids.toArray(num -> new int[num]);
// buffer for process name
Pointer<Byte> nameBuf = s.allocate(NativeTypes.INT8, NAME_BUF_MAX);
for (int i = 0; i < jpids.length; i++) {
int pid = jpids[i];
// get the process name
proc_name(pid, nameBuf, NAME_BUF_MAX);
String procName = Pointer.toString(nameBuf);
// print pid and process name
System.out.printf("%d %s\n", pid, procName);
}
}
}
}
jextract -l readline -L /usr/local/opt/readline/lib/ --record-library-path \
-t org.unix \
/usr/include/readline/readline.h \
--exclude-symbol readline_echoing_p -o readline.jar
import java.foreign.*;
import java.foreign.memory.*;
import static org.unix.readline_h.*;
public class Readline {
public static void main(String[] args) {
// Scope for native allocations
try (Scope s = scope().fork()) {
// allocate C memory initialized with Java string content
var pstr = s.allocateCString("name? ");
// call "readline" API
var p = readline(pstr);
// print char* as is
System.out.println(p);
// convert char* ptr from readline as Java String & print it
System.out.println(Pointer.toString(p));
}
}
}
javac -cp readline.jar Readline.java
java -cp readline.jar:. Readline
import java.lang.invoke.*;
import java.foreign.*;
import java.foreign.memory.*;
import org.unix.curl.*;
import org.unix.curl_h;
import static org.unix.curl_h.*;
import static org.unix.easy_h.*;
public class CurlMain {
public static void main(String[] args) {
try (Scope s = curl_h.scope().fork()) {
curl_global_init(CURL_GLOBAL_DEFAULT);
Pointer<Void> curl = curl_easy_init();
if(!curl.isNull()) {
Pointer<Byte> url = s.allocateCString(args[0]);
curl_easy_setopt(curl, CURLOPT_URL, url);
int res = curl_easy_perform(curl);
if (res != CURLE_OK) {
curl_easy_cleanup(curl);
}
}
curl_global_cleanup();
}
}
}
import java.foreign.*;
import java.lang.invoke.*;
import org.unix.unistd;
public class Getpid {
public static void main(String[] args) {
// bind unistd interface
var u = Libraries.bind(MethodHandles.lookup(), unistd.class);
// call getpid from the unistd.h
System.out.println(u.getpid());
// check process id from Java API!
System.out.println(ProcessHandle.current().pid());
}
}OpenGL is a popular portable graphic library: https://www.opengl.org/
Installing relevant OpenGL headers and libraries can be a bit tricky, as it depends on what graphic card is installed on the target platform. The following instruction assume that the standard version of OpenGL is used (e.g. mesa), rather than a proprietary one (Nvidia or AMD), although the changes to get these working are rather small.
OpenGL is always coupled with a bunch of other libraries, namely GLU and glut. You can install all those libraries using apt, as follows:
If the installation was successful, OpenGL headers can be found under /usr/include/GL, while libraries can be found in the folder /usr/lib/x86_64-linux-gnu/.
To extract the opengl libraries the following command suffices:
jextract -L /usr/lib/x86_64-linux-gnu -l glut -l GLU -l GL --record-library-path -t opengl -o opengl.jar /usr/include/GL/glut.hSince glut depends on the other libraries (GLU and GL), it is not necessary to give additional headers to jextract.
import java.foreign.NativeTypes;
import java.foreign.Scope;
import java.foreign.memory.Array;
import java.foreign.memory.Pointer;
import static opengl.gl_h.*;
import static opengl.freeglut_std_h.*;
public class Teapot {
float rot = 0;
Teapot(Scope sc) {
// Misc Parameters
Array<Float> pos = sc.allocateArray(NativeTypes.FLOAT, new float[] {0.0f, 15.0f, -15.0f, 0});
Array<Float> spec = sc.allocateArray(NativeTypes.FLOAT, new float[] {1, 1, 1, 0});
Array<Float> shini = sc.allocateArray(NativeTypes.FLOAT, new float[] {113});
// Reset Background
glClearColor(0, 0, 0, 0);
// Setup Lighting
glShadeModel(GL_SMOOTH);
glLightfv(GL_LIGHT0, GL_POSITION, pos.elementPointer());
glLightfv(GL_LIGHT0, GL_AMBIENT, spec.elementPointer());
glLightfv(GL_LIGHT0, GL_DIFFUSE, spec.elementPointer());
glLightfv(GL_LIGHT0, GL_SPECULAR, spec.elementPointer());
glMaterialfv(GL_FRONT, GL_SHININESS, shini.elementPointer());
glEnable(GL_LIGHTING);
glEnable(GL_LIGHT0);
glEnable(GL_DEPTH_TEST);
}
void display() {
glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);
glPushMatrix();
glRotatef(-20, 1, 1, 0);
glRotatef(rot, 0, 1, 0);
glutSolidTeapot(0.5);
glPopMatrix();
glutSwapBuffers();
}
void onIdle() {
rot += 0.1;
glutPostRedisplay();
}
public static void main(String[] args) {
try (Scope sc = opengl.gl_h.scope().fork()) {
Pointer<Integer> argc = sc.allocate(NativeTypes.INT32);
argc.set(0);
glutInit(argc, Pointer.ofNull());
glutInitDisplayMode(GLUT_DOUBLE | GLUT_RGBA | GLUT_DEPTH);
glutInitWindowSize(900, 900);
glutCreateWindow(sc.allocateCString("Hello Panama!"));
Teapot teapot = new Teapot(sc);
glutDisplayFunc(sc.allocateCallback(teapot::display));
glutIdleFunc(sc.allocateCallback(teapot::onIdle));
glutMainLoop();
}
}
}Download the freeglut package for MSVC at https://www.transmissionzero.co.uk/software/freeglut-devel/
Extract the freeglut zip.
Navigate to the root directory of the extracted zip and run the following commands:
$inc = "C:\Program Files (x86)\Windows Kits\10\Include\10.0.17134.0"
jextract -L C:\Windows\System32\ -L .\freeglut\bin\x64\ -l opengl32 -l freeglut -t opengl -o opengl.jar --package-map "$inc\um\gl=opengl" --record-library-path .\freeglut\include\GL\glut.hThe directory that is assigned to $inc is an example, and is system dependent. Make sure that the build number at the end of the path (in this case 10.0.17134.0) is the latest one found in the parent folder (C:\Program Files (x86)\Windows Kits\10\Include\).
There are a bunch of warnings generated, but as long as the jar file is generated in the working directory the extraction was successful.
This is the same as in the Ubuntu section
Quoted from https://www.tensorflow.org/install/lang_c
"TensorFlow provides a C API that can be used to build bindings for other languages. The API is defined in c_api.h and designed for simplicity and uniformity rather than convenience."
You can follow the setup procedure as described in the above page.
Alternatively, on Mac, you can install libtensorflow using HomeBrew
Tensorflow ship the libtensorflow with an .so extension, this doesn't work well for java on MacOS as java expect .dylib extension. To work around this, create a symbolic link.
The following command can be used to extract c_api.h.
jextract -C -x -C c++ \
-L /usr/local/lib -l tensorflow --record-library-path \
-o tf.jar -t org.tensorflow.panama \
/usr/local/include/tensorflow/c/c_api.hThe caveat to extract tensorflow C API is that it declare function prototype without argument in C++ style, for example, TF_Version(), which is considered incomplete C function prototype instead of C style as in TF_Version(void). An incomplete function prototype will become vararg funciton. To avoid that, we need to pass clang '-x c++' options to jextract with '-C -x -C c++'
import java.foreign.NativeTypes;
import java.foreign.Scope;
import java.foreign.memory.Array;
import java.foreign.memory.LayoutType;
import java.foreign.memory.Pointer;
import org.tensorflow.panama.c_api.TF_DataType;
import org.tensorflow.panama.c_api.TF_Graph;
import org.tensorflow.panama.c_api.TF_Operation;
import org.tensorflow.panama.c_api.TF_OperationDescription;
import org.tensorflow.panama.c_api.TF_Output;
import org.tensorflow.panama.c_api.TF_Session;
import org.tensorflow.panama.c_api.TF_SessionOptions;
import org.tensorflow.panama.c_api.TF_Status;
import org.tensorflow.panama.c_api.TF_Tensor;
import static org.tensorflow.panama.c_api_h.*;
import static org.tensorflow.panama.c_api_h.TF_DataType.*;
public class TensorFlowExample {
static Pointer<TF_Operation> PlaceHolder(Pointer<TF_Graph> graph, Pointer<TF_Status> status,
@TF_DataType int dtype, String name) {
try (var s = scope().fork()) {
Pointer<TF_OperationDescription> desc = TF_NewOperation(graph,
s.allocateCString("Placeholder"), s.allocateCString(name));
TF_SetAttrType(desc, s.allocateCString("dtype"), TF_FLOAT);
return TF_FinishOperation(desc, status);
}
}
static Pointer<TF_Operation> ConstValue(Pointer<TF_Graph> graph, Pointer<TF_Status> status,
Pointer<TF_Tensor> tensor, String name) {
try (var s = scope().fork()) {
Pointer<TF_OperationDescription> desc = TF_NewOperation(graph,
s.allocateCString("Const"), s.allocateCString(name));
TF_SetAttrTensor(desc, s.allocateCString("value"), tensor, status);
TF_SetAttrType(desc, s.allocateCString("dtype"), TF_TensorType(tensor));
return TF_FinishOperation(desc, status);
}
}
static Pointer<TF_Operation> Add(Pointer<TF_Graph> graph, Pointer<TF_Status> status,
Pointer<TF_Operation> one, Pointer<TF_Operation> two,
String name) {
try (var s = scope().fork()) {
Pointer<TF_OperationDescription> desc = TF_NewOperation(graph,
s.allocateCString("AddN"), s.allocateCString(name));
Array<TF_Output> add_inputs = s.allocateArray(
LayoutType.ofStruct(TF_Output.class),2);
add_inputs.get(0).oper$set(one);
add_inputs.get(0).index$set(0);
add_inputs.get(1).oper$set(two);
add_inputs.get(1).index$set(0);
TF_AddInputList(desc, add_inputs.elementPointer(), 2);
return TF_FinishOperation(desc, status);
}
}
public static void main(String... args) {
System.out.println("TensorFlow C library version: " + Pointer.toString(TF_Version()));
Pointer<TF_Graph> graph = TF_NewGraph();
Pointer<TF_SessionOptions> options = TF_NewSessionOptions();
Pointer<TF_Status> status = TF_NewStatus();
Pointer<TF_Session> session = TF_NewSession(graph, options, status);
float in_val_one = 4.0f;
float const_two = 2.0f;
Pointer<TF_Tensor> tensor_in = TF_AllocateTensor(TF_FLOAT, Pointer.ofNull(), 0, Float.BYTES);
TF_TensorData(tensor_in).cast(NativeTypes.FLOAT).set(in_val_one);
Pointer<TF_Tensor> tensor_const_two = TF_AllocateTensor(TF_FLOAT, Pointer.ofNull(), 0, Float.BYTES);
TF_TensorData(tensor_const_two).cast(NativeTypes.FLOAT).set(const_two);
// Operations
Pointer<TF_Operation> feed = PlaceHolder(graph, status, TF_FLOAT, "feed");
Pointer<TF_Operation> two = ConstValue(graph, status, tensor_const_two, "const");
Pointer<TF_Operation> add = Add(graph, status, feed, two, "add");
try (var s = scope().fork()) {
var ltPtrTensor = LayoutType.ofStruct(TF_Tensor.class).pointer();
// Session Inputs
TF_Output input_operations = s.allocateStruct(TF_Output.class);
input_operations.oper$set(feed);
input_operations.index$set(0);
Pointer<Pointer<TF_Tensor>> input_tensors = s.allocate(ltPtrTensor);
input_tensors.set(tensor_in);
// Session Outputs
TF_Output output_operations = s.allocateStruct(TF_Output.class);
output_operations.oper$set(add);
output_operations.index$set(0);
Pointer<Pointer<TF_Tensor>> output_tensors = s.allocate(ltPtrTensor);
TF_SessionRun(session, Pointer.ofNull(),
// Inputs
input_operations.ptr(), input_tensors, 1,
// Outputs
output_operations.ptr(), output_tensors, 1,
// Target operations
Pointer.ofNull(), 0, Pointer.ofNull(),
status);
System.out.println(String.format("Session Run Status: %d - %s",
TF_GetCode(status), Pointer.toString(TF_Message(status))));
Pointer<TF_Tensor> tensor_out = output_tensors.get();
System.out.println("Output Tensor Type: " + TF_TensorType(tensor_out));
float outval = TF_TensorData(tensor_out).cast(NativeTypes.FLOAT).get();
System.out.println("Output Tensor Value: " + outval);
TF_CloseSession(session, status);
TF_DeleteSession(session, status);
TF_DeleteSessionOptions(options);
TF_DeleteGraph(graph);
TF_DeleteTensor(tensor_in);
TF_DeleteTensor(tensor_out);
TF_DeleteTensor(tensor_const_two);
TF_DeleteStatus(status);
}
}
}You can download a binary distribution from https://www.tensorflow.org/install/lang_c
Extract the zip file.
The problem outlined in the Mac OS instruction w.r.t. incorrect function prototypes still exists (though it has been solved in the Tensorflow github repository, this change has not yet made it into the binary distributions). On Windows there is however no jextract command that works around this, so the only way to extract the \include\c\c_api.h header successfully is to first manually fix the incorrect function prototypes:
TF_Version() -> TF_Version(void)
TF_NewGraph() -> TF_NewGraph(void)
TF_NewSessionOptions() -> TF_NewSessionOptions(void)
TF_NewStatus() -> TF_NewStatus(void)
TF_NewBuffer() -> TF_NewBuffer(void)
TF_NewImportGraphDefOptions() -> TF_NewImportGraphDefOptions(void)
TF_GetAllOpList() -> TF_GetAllOpList(void)Once you've done this you can use the following jextract command from the libtensorflow root directory:
This is the same as for the Mac OS section.