Unsupported type error for int and uint types

[selcuksert]

Description

Running TornadoVM powered applications on a PC with Turkish locale yields to following error:

Unable to compile task graph.double - computeDouble
The internal error is: [Error during the Task Compilation]: Unsupported type: ınt

Stacktrace: [
[email protected]/uk.ac.manchester.tornado.drivers.opencl.runtime.OCLTornadoDevice.compileTask(OCLTornadoDevice.java:306)
[email protected]/uk.ac.manchester.tornado.drivers.opencl.runtime.OCLTornadoDevice.compileJavaToAccelerator(OCLTornadoDevice.java:342)
[email protected]/uk.ac.manchester.tornado.drivers.opencl.runtime.OCLTornadoDevice.installCode(OCLTornadoDevice.java:469)
[email protected]/uk.ac.manchester.tornado.runtime.interpreter.TornadoVMInterpreter.compileTaskFromBytecodeToBinary(TornadoVMInterpreter.java:670)
[email protected]/uk.ac.manchester.tornado.runtime.interpreter.TornadoVMInterpreter.execute(TornadoVMInterpreter.java:333)
[email protected]/uk.ac.manchester.tornado.runtime.interpreter.TornadoVMInterpreter.execute(TornadoVMInterpreter.java:909)
java.base/java.util.Spliterators$ArraySpliterator.forEachRemaining(Spliterators.java:1024)
java.base/java.util.stream.ReferencePipeline$Head.forEach(ReferencePipeline.java:762)
[email protected]/uk.ac.manchester.tornado.runtime.TornadoVM.executeInterpreterSingleThreaded(TornadoVM.java:127)
[email protected]/uk.ac.manchester.tornado.runtime.TornadoVM.execute(TornadoVM.java:114)
[email protected]/uk.ac.manchester.tornado.runtime.tasks.TornadoTaskGraph.scheduleInner(TornadoTaskGraph.java:892)
[email protected]/uk.ac.manchester.tornado.runtime.tasks.TornadoTaskGraph.execute(TornadoTaskGraph.java:1470)
[email protected]/uk.ac.manchester.tornado.runtime.tasks.TornadoTaskGraph.execute(TornadoTaskGraph.java:1499)
[email protected]/uk.ac.manchester.tornado.api.TaskGraph.execute(TaskGraph.java:759)
[email protected]/uk.ac.manchester.tornado.api.ImmutableTaskGraph.execute(ImmutableTaskGraph.java:50)
[email protected]/uk.ac.manchester.tornado.api.TornadoExecutor.lambda$execute$0(TornadoExecutor.java:47)
java.base/java.util.ArrayList.forEach(ArrayList.java:1596)
[email protected]/uk.ac.manchester.tornado.api.TornadoExecutor.execute(TornadoExecutor.java:47)
[email protected]/uk.ac.manchester.tornado.api.TornadoExecutionPlan.execute(TornadoExecutionPlan.java:163)
com.corp.compute.TornadoCompute.buildAndRun(TornadoCompute.java:33)
com.corp.Main.main(Main.java:30)
]

To overcome the issue following JVM parameters should be added as a workaround:

-Duser.country=US -Duser.language=en

Dec 22, 2024 6:24:03 PM uk.ac.manchester.tornado.runtime.common.TornadoLogger debug
INFO: 	OpenCL compilation status = CL_BUILD_SUCCESS

TornadoVM test reports similar errors. To eliminate the errors, test should be run as follows:

tornado-test -V --jvm="-Duser.country=US -Duser.language=en"

How To Reproduce

Ensure system locale is set as tr-TR and use OpenCL backend:

% defaults read .GlobalPreferences AppleLanguages                              
(
    "tr-TR"
)

Use following computation execution:

package com.corp.compute;

import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

import uk.ac.manchester.tornado.api.TaskGraph;
import uk.ac.manchester.tornado.api.TornadoExecutionPlan;
import uk.ac.manchester.tornado.api.TornadoExecutionResult;
import uk.ac.manchester.tornado.api.annotations.Parallel;
import uk.ac.manchester.tornado.api.enums.DataTransferMode;
import uk.ac.manchester.tornado.api.exceptions.TornadoExecutionPlanException;
import uk.ac.manchester.tornado.api.math.TornadoMath;
import uk.ac.manchester.tornado.api.types.arrays.FloatArray;

public class TornadoCompute {
    private static final Logger logger = LoggerFactory.getLogger(TornadoCompute.class.getName());

    private TornadoCompute() {
        throw new IllegalStateException("%s is a utility class and cannot be instantiated!".formatted(this.getClass().getName()));
    }

    public static void buildAndRun(FloatArray array) {
        //@formatter:off
        TaskGraph taskGraph = new TaskGraph("graph")
                .transferToDevice(DataTransferMode.EVERY_EXECUTION, array)
                .task("double", TornadoCompute::computeDouble, array)
                .task("sqrt", TornadoCompute::computeSqrt, array)
                .transferToHost(DataTransferMode.EVERY_EXECUTION, array);
        //@formatter:on

        try (TornadoExecutionPlan tornadoExecutionPlan = new TornadoExecutionPlan(taskGraph.snapshot())) {
            logger.info("Pre-execution result: {}", array.get(1));
            TornadoExecutionResult result = tornadoExecutionPlan.execute();
            logger.info("Execution time: {}", result.getProfilerResult().getTotalTime());
            logger.info("Post-execution result: {}", array.get(1));
        } catch (TornadoExecutionPlanException e) {
            logger.error("Tornado execution error:", e);
        }
    }

    private static void computeDouble(FloatArray array) {
        for (@Parallel int i = 0; i < array.getSize(); i++) {
            float value = array.get(i);
            array.set(i, value * 2);
        }
    }

    private static void computeSqrt(FloatArray array) {
        for (@Parallel int i = 0; i < array.getSize(); i++) {
            float value = array.get(i);
            array.set(i, TornadoMath.sqrt(value));
        }
    }

}

Invoke TornadoVM computation:

package com.corp;

import java.util.Random;

import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

import com.corp.compute.TornadoCompute;

import uk.ac.manchester.tornado.api.types.arrays.FloatArray;

public class Main {
    private static final Logger logger = LoggerFactory.getLogger(Main.class.getName());

    public static void main(String[] args) {
        Random rand = new Random();
        int noOfElements = 10000;
        FloatArray floatArray = new FloatArray(noOfElements);

        for (int i = 0; i < noOfElements; i++) {
            float randFloat = rand.nextFloat(noOfElements);
            floatArray.set(i, randFloat);
        }

        logger.info("Starting computation");
        TornadoCompute.buildAndRun(floatArray);
        logger.info("Finished computation");
    }
}

Expected behavior

Execution should result in success.

Computing system setup (please complete the following information):

• OS: MacOS 15.2
• OpenCL and Driver versions

    Tornado device=0:1
          OPENCL --  [Apple] -- AMD Radeon Pro 5500M Compute Engine
                  Global Memory Size: 4,0 GB
                  Local Memory Size: 64,0 KB
                  Workgroup Dimensions: 3
                  Total Number of Block Threads: [256]
                  Max WorkGroup Configuration: [256, 256, 256]
                  Device OpenCL C version: OpenCL C 1.2
  

Additional context

From stack trace it seems following invocation turns out to be letter ı in Turkish for Integer type name instead of i:

TornadoVM/tornado-drivers/opencl/src/main/java/uk/ac/manchester/tornado/drivers/opencl/graal/asm/OCLAssembler.java

Line 97 in a3b87c7

String type = input.getPlatformKind().name().toLowerCase();

[jjfumero]

Hi @selcuksert . Thank you for the report. Now I wonder if these options:

-Duser.country=US -Duser.language=en

should be enabled by default. TornadoVM needs the type descriptor names that are passed to the JIT compiler. Let me see if we can find an alternative using JVMCI.