arrayfire / forge Goto Github PK

Steps to reproduce

• install ninja

cmake -Gninja ..
ninja

Error:

ninja: error: 'third_party/freetype/lib/libfreetype.a', needed by 'src/libforge.so', missing and no known rule to make it

Issue originally reported here:
arrayfire/arrayfire#861

Calls such has glTexSubImage2D, glTexImage2D use 4-byte alignment by default. For types such as unsigned char, char, there is possibility that above gl*** calls will result in segmentation faults. Will have to verify and fix it if required.

Useful topics

• GLFW no context hints for window
• SDL2 context creation function
• glfw opengl context objects
• WGL context creation process

OpenCL examples are not working at all

surface

plotting

plot3

histogram

fractal

CPU examples look garbled at default window size, but they work after maximizing the window.

Histogram default window

Histogram maximized window

Fractal default window

Fractal maximized window

The shader parameters with in qualifier are currently using vec[2|3|4] vector types whose base type is single precision floating point numbers. In cases where the vertex buffers are of different types, check if the shader parameters needs to be changed to their corresponding vector types. If yes, make necessary changes.

I'm trying to update the Homebrew formula for arrayfire to version 3.2.2. In the process, I'm using version af3.2.2 of forge. It compiles on OSX 10.9 and 10.11, but not on 10.10. Using clang++, I get the error message:

In file included from /Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX10.11.sdk/System/Library/Frameworks/OpenGL.framework/Headers/CGLCurrent.h:8:
/Applications/Xcode.app/Contents/Developer/Platforms/MacOSX.platform/Developer/SDKs/MacOSX10.11.sdk/System/Library/Frameworks/OpenGL.framework/Headers/CGLTypes.h:75:15: error: missing ',' between enumerators
        kCGLPFAStereo OPENGL_ENUM_DEPRECATED(10_0, 10_11)        =   6,
                     ^
                     , 

You can find the full build log here.
The error above is located here.

Thanks in advance for any clue!

• Plot
• Histogram
• Surface
• Plot3

All most all of the forge objects have the option to let user provide per vertex attributes such as colors and alphas for rendering.

In the case of contour plots, I am not sure if per vertex colours or alphas makes any sense. One possible usage that i can see is per iosline color and alpha values.

@arrayfire/core-devel Any possible use cases of allowing per vertex colours/alphas ? An alternative to these per vertex attributes i thought off is using colormaps with isolines, which is the most common way of coloring contour lines from what i have seen so far.

We can discuss the different possible directions in which we want to implement the current issue. Please post the URLS to other plotting libraries for reference.

Ref arrayfire/arrayfire#719

There are (at least) two issues here with the histogram example:

1. The Perlin distribution should result in a Gaussian-like distribution of noise, whereas the histogram does not display anything remotely close to this, and
2. There appears to be a plotting error wherein every 3n and 4n values are near zero in the histogram.

The Perlin distribution used in the histograms should result in a Gaussian-like noise distribution; however, in the examples, it is anything but Gaussian:

This is even more apparent if one increases the number of bins to around 100:

Also, there appears to be an odd pattern wherein every 3rd and 4th value on the x-axis is near zero.

Usage example

class Surface {
    // ...
};

fg::Surface surf(100, 100, FGType);

fg::Window wnd;
wnd->draw(surf);

Which would render something as below (left hand side in the picture).

In the code for PR #74, z-depth buffering appears to be broken for (at least) surface plots in the CPU and CUDA backends. See the plots below.

The following image is taken from surface_opencl and shows correct z-depth behavior:

The following image is taken from surface_cuda and shows INCORRECT z-depth behavior:

The following image is taken from surface_cpu and shows INCORRECT z-depth behavior:

There are some compilation issues with Visual Studio 2015.

Currently, each character is rendered as is. However, this can pose a problem if the color in which the character is rendered is same as the background ground. If an outline is drawn, it would be visually helpful to easy identify text in situations such as explained above.

The FindGLM.cmake module was removed with commit g-truc/glm@62a7dad from GLM. Thus, setting USE_SYSTEM_GLM=ON results in a failing FIND_PACKAGE call for GLM 0.9.7.0 and later.

As shown below, the plot3_cpu example has a vertical line in the "funnel" whereas the line does not appear in plot3_cuda or plot3_opencl. This appears in code from PR #74 .

C02QH2D7G8WM:opencl userone$ ./histogram_opencl
Fontconfig error: Cannot load default config file
Freetype face initilization(3001)
C02QH2D7G8WM:opencl userone$ ./plot
-bash: ./plot: No such file or directory
C02QH2D7G8WM:opencl userone$ ./plot3_opencl
Fontconfig error: Cannot load default config file
Freetype face initilization(3001)

On Linux, we use fontconfig to do the same. A similar work around needs to be done for Windows platform.

• Box plots
• Pie Charts

The values can be seen here:
https://gist.github.com/pavanky/93a0106d92400ccb9b53

Following are the effected Renderable objects:

• Image
• Plot
• Histogram
• Plot3
• Surface

This has been observed only on AMD APU machine so far.

At present X-axis tick marks have the left side immediately below the tick mark. The labels should be centered on the tick mark.

• Documentation
• Vendor Neutral Copy Headers
• Improve Mouse based rotation interaction
• [ ] Fix rendering bugs on AMD GPUs
• 2D Vector Fields
• Test Examples on Primary Platforms(Linux, OSX & WIndows)
• Test Examples on Multiple GPU Vendors
• Update cl.hpp in examples to use cl2.hpp and add that as external project for examples.

Optionally

• glbinding switch
• 3D Vector Fields

Early version maintained individual textures for each character glyph, which was very inefficient.

Pack the glyphs into a single texture.

The histogram_opencl example appears to have two bugs:

1. The color of the bars is always yellow and
2. Between successive frames only the last bar ever updates

See screenshots below.

Appearance of histogram_cuda shows bars that are in different colors and they all update their value every frame. This is the correct behavior.

whereas histogram_opencl shows bars that are all yellow in color and only the last bar updates between successive frames:

On OSX El Capitan (10.11)
Using [email protected]
and clang:

forge (master)‣ clang --version
Apple LLVM version 7.0.0 (clang-700.1.76)
Target: x86_64-apple-darwin15.0.0
Thread model: posix

after cloning forge and running through the build process

I'm seeing

[ 69%] Building CXX object examples/CMakeFiles/example_histogram_opencl.dir/opencl/histogram.cpp.o
/Users/tmpvar/work/gfx/tmp/forge/examples/opencl/histogram.cpp:283:16: error: use of undeclared identifier 'fmod'
            if(fmod(persistance, 0.5f) < 0.01)
               ^
1 error generated.

after applying this patch

I was able to get it to build, but upon running the surface_opencl demo I see:

forge (master)‣ examples/opencl/surface_opencl
InfoLog:
WARNING: Output of vertex shader 'hpoint' not read by fragment shader

OpenGL Shader linking failed(2001)

I am currently working on an Gentoo ebuild for forge. I noticed that the CMake stuff is only installed if -DBUILD_EXAMPLES=ON is passed:

-- Installing: /var/tmp/portage/sci-visualization/forge-9999/image/usr/share/Forge/examples/CMakeLists.txt
-- Installing: /var/tmp/portage/sci-visualization/forge-9999/image/usr/share/Forge/examples/CMakeModules
-- Installing: /var/tmp/portage/sci-visualization/forge-9999/image/usr/share/Forge/examples/CMakeModules/FindOpenCL.cmake
-- Installing: /var/tmp/portage/sci-visualization/forge-9999/image/usr/share/Forge/examples/CMakeModules/FindForge.cmake
-- Installing: /var/tmp/portage/sci-visualization/forge-9999/image/usr/share/Forge/examples/CMakeModules/FindGLEWmx.cmake

Otherwise, only the forge library and headers get installed. Initally I expected that only FindForge.cmake would be installed. FindOpenGL.cmake and a FindGLEW.cmake are provided by the system.

Furthermore, I tried to change the install destination with -DFG_INSTALL_CMAKE_DIR="${ED}/usr/share/cmake/Modules" which shows no effect. Maybe you can give me a hint.

When displaying a plot which has values that cross zero, the value zero should appear in the plot and all labels should be equally spaced about zero.

I experienced that f64, c32 and c64 causes an error. Will they be supported later on?

• Add option to use different type of markers for plots
• Add option to render different type of plots eg. scatter etc. Probably add the ones listed here as we progress forward. But definitely add scatter plot.

Keep it off by default and turn on when asked for.

It is necessary to add support for drawing of computer vision features.

A circle/cross/X can be drawn around the coordinates where a feature is located. Feature matching can be drawn as lines between matching features in two different images.

See image below for example.

Currently considering the following alternatives for GLFW.

• SDL2
• Qt

Sometimes you have to display visualisations over a web interface. Wondering if you can add this functionality.

As reference following web rendering frameworks look cool so it would be nice if you have feature parity with them:

• https://plot.ly/
• https://juliaplots.github.io/
• http://p5js.org/
• http://visjs.org/
• http://threejs.org/
• https://www.blend4web.com/en/
• http://stack.gl/

Will event handlers, like mouse click, mouse drag, scrolling, etc. be integrated to ArrayFire?

• 3d line plot
• 3d scatter plot

These would be almost similar to what 2d plotting was doing.