How to write cross GLProfile compatible shader using JOGL
Note: To be refined / edited!
In order to create cross platform GLSL shaders the following three points have to be taken into account:
GLSL Version Directive
Some OpenGL implementations require the GLSL version directive #version <number>
to be explicitly included in the shader source as the first directive.
The GLSL version directive shall match the OpenGL version of the context, see GLContext's GLSLVersionNumber.
If any of above requirements are not fulfilled the shader code compilation might fail on certain platforms and OpenGL version.
Users may compose their shader at runtime to comply with the GLSL version directive utilizing JOGL's features:
- GLContext's getGLSLVersionNumber(), which returns the GLSL
VersionNumber
matching the GLContext's OpenGL version. - GLContext's getGLSLVersionString(), which returns the GLSL version directive matching the GLContext's OpenGL version.
- ShaderCode's addGLSLVersion(..), which adds the matching GLSL version directive to the shader code.
GLSL Precision Directive
In case you target ES2, ES3 or >= GL3 OpenGL contexts one needs to add a precision qualifier to your variable declarations or use a shader wide default precision qualifier directive.
Users may either add such precision directive in their code in a fine grained manner adjusting precision to their tasks, or they may utilize a JOGL's feature:
- ShaderCode's addDefaultShaderPrecision(..), which adds default precision directive to the shader code.
GLSL Version & Default Precision Directive
The GLSL version and default precision directives maybe performed by the ShaderCode class, which can load your code from an input stream (URI source) or CharSequence array and customize it appropriately via defaultShaderCustomization(..).
ShaderCode
also allows users to perform further custom editing on the fly.
Have a look at how GLSL shaders are loaded inside the jogamp jogl junit tests using ShaderCode
for cross platform GLSL use:
RedSquareES2.
final ShaderCode vp0 = ShaderCode.create(gl, GL2ES2.GL_VERTEX_SHADER, this.getClass(), "shader", "shader/bin", "RedSquareShader", true); final ShaderCode fp0 = ShaderCode.create(gl, GL2ES2.GL_FRAGMENT_SHADER, this.getClass(), "shader", "shader/bin", "RedSquareShader", true); vp0.defaultShaderCustomization(gl, true, true); fp0.defaultShaderCustomization(gl, true, true); final ShaderProgram sp0 = new ShaderProgram(); sp0.add(gl, vp0, System.err); sp0.add(gl, fp0, System.err);
GLSL Version Compatibility
GLSL Version >= 130
GLSL version >= 130 uses in and out keywords instead of attributes and varying. it also requires the fragment shader to explicitly define the out variable. Compatibility with previous GLSL versions can be reached by adding some GLSL pre-processor defines at the beginning of the shaders.
Vertex shader:
#if __VERSION__ >= 130 #define attribute in #define varying out #endif
Fragment shader:
#if __VERSION__ >= 130 #define varying in out vec4 mgl_FragColor; #define texture2D texture #else #define mgl_FragColor gl_FragColor #endif
Look at the Cross platform GLSL shaders used by the JogAmp JOGL junit tests how to add the pre-processor defines:
JOGL Unit Test Shader.
GLSL ES Version 1.0
Attribute Types
ES2. aka GLSL ES version 1.0, has restrictions on the types that can be used as attributes. Specifically:
GLSL ES version 1.0 spec, section 4.3.3:
The attribute qualifier can be used only with the data types float, vec2, vec3, vec4, mat2, mat3, and mat4. Attribute variables cannot be declared as arrays or structures.
Multiple Render Targets (MRT)
Also, GLSL ES version 1.0 doesn't support Multiple Render Targets (MRT).
GLSL version 1.30 MRT code snippet:
out vec4 mgl_FragData[2]; varying vec4 frontColor; void main (void) { mgl_FragData[0] = vec4( frontColor.r, 0.0, 0.0, 1.0 ); mgl_FragData[1] = vec4( 0.0, frontColor.g, 0.0, 1.0 ); }
GLSL ES version 1.0 restricts the number of possible framebuffer color attachments to one, hence any shader using multiple outputs need to be rewritten utilizing multiple shader passes.
To be refined, maybe remove lower part.
In GLSL version <= 1.20, writing to multiple fragment shader outputs meant writing to gl_FragData[0 .. n]. This will still work in GLSL ES version 1.0, but the gl_FragData array will consist of only one element. In 1.30, outputs have to be declared as "out" parameters and then associated with a specific "location" with glBindFragDataLocation. In GLSL 1.40 and above, "out" parameters can be assigned specific locations with the location directive. For example: "layout(location = 0) out vec4 output". In ES3, the situation is the same as GLSL version 1.40 and above.
Texture Formats
glTexImage* requires that the "format" and "internalformat" parameters must match, and can only be one of GL_ALPHA, GL_LUMINANCE, GL_LUMINANCE_ALPHA, GL_RGB, GL_RGBA. This is incompatible with every other OpenGL version.
No read Framebuffer
To be refined A complete lack of any concept of "read" framebuffers; the result is actually a subset of 3.0 and the four framebuffer extensions JOGL uses to give equivalent behaviour on 2.1 (GL_ARB_framebuffer_object | GL_EXT_framebuffer_object, GL_EXT_framebuffer_multisample, GL_EXT_framebuffer_blit, GL_EXT_packed_depth_stencil).
An amusing one is the stencil buffer: The majority of implementations don't support separate depth and stencil framebuffer attachments, so you're required to use GL_OES_packed_depth_stencil to actually get a stencil buffer at all (you're certainly not going to be doing any significant work rendering without a depth buffer).