JNA library slower screenshot than robot class?

Question

Since Robot.createScreenCaputure() method is slow, I decided to use native library. I searched and found this forum and find a specific code snipplet which uses JNA Library . It's an old version so that I rewrote the code:

import java.awt.Rectangle;
import java.awt.image.BufferedImage;
import java.awt.image.ColorModel;
import java.awt.image.DataBuffer;
import java.awt.image.DataBufferInt;
import java.awt.image.DataBufferUShort;
import java.awt.image.DirectColorModel;
import java.awt.image.Raster;
import java.awt.image.WritableRaster;

import com.sun.jna.Native;
import com.sun.jna.win32.W32APIOptions;
import com.sun.jna.platform.win32.WinDef;
import com.sun.jna.platform.win32.WinNT;
import com.sun.jna.platform.win32.WinGDI;

public class JNAScreenShot {

    public static BufferedImage getScreenshot(Rectangle bounds) {
        WinDef.HDC windowDC = GDI.GetDC(USER.GetDesktopWindow());
        WinDef.HBITMAP outputBitmap =
                GDI.CreateCompatibleBitmap(windowDC,
                bounds.width, bounds.height);
        try {
            WinDef.HDC blitDC = GDI.CreateCompatibleDC(windowDC);
            try {
                WinNT.HANDLE oldBitmap =
                        GDI.SelectObject(blitDC, outputBitmap);
                try {
                    GDI.BitBlt(blitDC,
                            0, 0, bounds.width, bounds.height,
                            windowDC,
                            bounds.x, bounds.y,
                            GDI32.SRCCOPY);
                } finally {
                    GDI.SelectObject(blitDC, oldBitmap);
                }
                WinGDI.BITMAPINFO bi = new WinGDI.BITMAPINFO(40);
                bi.bmiHeader.biSize = 40;
                boolean ok =
                        GDI.GetDIBits(blitDC, outputBitmap, 0, bounds.height,
                        (byte[]) null, bi, WinGDI.DIB_RGB_COLORS);
                if (ok) {
                    WinGDI.BITMAPINFOHEADER bih = bi.bmiHeader;
                    bih.biHeight = -Math.abs(bih.biHeight);
                    bi.bmiHeader.biCompression = 0;
                    return bufferedImageFromBitmap(blitDC, outputBitmap, bi);
                } else {
                    return null;
                }
            } finally {
                GDI.DeleteObject(blitDC);
            }
        } finally {
            GDI.DeleteObject(outputBitmap);
        }
    }

    private static BufferedImage bufferedImageFromBitmap(WinDef.HDC blitDC,
            WinDef.HBITMAP outputBitmap,
            WinGDI.BITMAPINFO bi) {
        WinGDI.BITMAPINFOHEADER bih = bi.bmiHeader;
        int height = Math.abs(bih.biHeight);
        final ColorModel cm;
        final DataBuffer buffer;
        final WritableRaster raster;
        int strideBits =
                (bih.biWidth * bih.biBitCount);
        int strideBytesAligned =
                (((strideBits - 1) | 0x1F) + 1) >> 3;
        final int strideElementsAligned;
        switch (bih.biBitCount) {
            case 16:
                strideElementsAligned = strideBytesAligned / 2;
                cm = new DirectColorModel(16, 0x7C00, 0x3E0, 0x1F);
                buffer =
                        new DataBufferUShort(strideElementsAligned * height);
                raster =
                        Raster.createPackedRaster(buffer,
                        bih.biWidth, height,
                        strideElementsAligned,
                        ((DirectColorModel) cm).getMasks(),
                        null);
                break;
            case 32:
                strideElementsAligned = strideBytesAligned / 4;
                cm = new DirectColorModel(32, 0xFF0000, 0xFF00, 0xFF);
                buffer =
                        new DataBufferInt(strideElementsAligned * height);
                raster =
                        Raster.createPackedRaster(buffer,
                        bih.biWidth, height,
                        strideElementsAligned,
                        ((DirectColorModel) cm).getMasks(),
                        null);
                break;
            default:
                throw new IllegalArgumentException("Unsupported bit count: " + bih.biBitCount);
        }
        final boolean ok;
        switch (buffer.getDataType()) {
            case DataBuffer.TYPE_INT: {
                int[] pixels = ((DataBufferInt) buffer).getData();
                ok = GDI.GetDIBits(blitDC, outputBitmap, 0, raster.getHeight(), pixels, bi, 0);
            }
            break;
            case DataBuffer.TYPE_USHORT: {
                short[] pixels = ((DataBufferUShort) buffer).getData();
                ok = GDI.GetDIBits(blitDC, outputBitmap, 0, raster.getHeight(), pixels, bi, 0);
            }
            break;
            default:
                throw new AssertionError("Unexpected buffer element type: " + buffer.getDataType());
        }
        if (ok) {
            return new BufferedImage(cm, raster, false, null);
        } else {
            return null;
        }
    }
    private static final User32 USER = User32.INSTANCE;
    private static final GDI32 GDI = GDI32.INSTANCE;
}

interface GDI32 extends com.sun.jna.platform.win32.GDI32,
        com.sun.jna.platform.win32.WinGDI,
        com.sun.jna.platform.win32.WinDef {

    GDI32 INSTANCE =
            (GDI32) Native.loadLibrary(GDI32.class);

    boolean BitBlt(HDC hdcDest, int nXDest, int nYDest,
            int nWidth, int nHeight, HDC hdcSrc,
            int nXSrc, int nYSrc, int dwRop);

    HDC GetDC(HWND hWnd);

    boolean GetDIBits(HDC dc, HBITMAP bmp, int startScan, int scanLines,
            byte[] pixels, BITMAPINFO bi, int usage);

    boolean GetDIBits(HDC dc, HBITMAP bmp, int startScan, int scanLines,
            short[] pixels, BITMAPINFO bi, int usage);

    boolean GetDIBits(HDC dc, HBITMAP bmp, int startScan, int scanLines,
            int[] pixels, BITMAPINFO bi, int usage);
    int SRCCOPY = 0xCC0020;
}

interface User32 extends com.sun.jna.platform.win32.User32 {

    User32 INSTANCE = (User32) Native.loadLibrary(User32.class, W32APIOptions.UNICODE_OPTIONS);

    com.sun.jna.platform.win32.WinDef.HWND GetDesktopWindow();
}

And a test code to see how much it faster faster than Robot Class:

import java.awt.AWTException;
import java.awt.Rectangle;
import java.awt.Robot;
import java.awt.image.BufferedImage;

public class testClass {

    public static void main(String[] args) {

        BufferedImage bi = null, bj = null;
        Rectangle rect = new Rectangle(0, 0, 810, 384);
        long startTime, finishTime;

        startTime = System.currentTimeMillis();
        for (int i = 0; i < 10; i++) {
            bi = JNAScreenShot.getScreenshot(rect);
        }
        finishTime = System.currentTimeMillis();

        System.out.println("With JNA Library: " + (finishTime - startTime)/10);

        Robot robo = null;

        startTime = System.currentTimeMillis();
        try {
            robo = new Robot();
        } catch (AWTException a) {
        }
        for (int i = 0; i < 10; i++) {
            bj = robo.createScreenCapture(rect);
        }
        finishTime = System.currentTimeMillis();
        System.out.println("With Robot Class " + (finishTime - startTime)/10);
    }
}

And the result is

With JNA Library: 77
With Robot Class 37

Guys, please someone explain why is that and how can I fasten it?

Answer 1

Don't try to optimise too early. Create a sensible interface to get the data you want (A screenshot) then create the implementation you want based on either Robot, JNA or JNI.

I would guess that the different implementations give completely different results based on the environment that it is running.

Rule One of Programming: First make it work. Then profile, find a bottleneck and remove or mitigate the effect of the bottleneck.

Answer 2

First, check if the native library is actually faster than your code. It might not be the case.

Assuming you already did check that, I would say the problem here is that calls with JNA are really slow. To skip the one call per cycle problem I would suggest writing a C function like this:

    void callWithJNAfunction(int rectPosX, int rectPosY, rectSideX, rectSideY,int numberOfCycles) {
    for (int i = 0; i < numberOfCycles; i++) {
        //code in C that calls the library
    }

Now compile this code and call the callWithJNAfunction(...) with JNA.

If the problem is the slowness of the JNA calls, it will get faster.

Answer 3

JNA调用需要很长时间，而JNI直接使用c ++。