GLES2 and D3D11 port

attilaz edited this page Aug 16, 2013 · 24 revisions
Clone this wiki locally

Documentation

Changes in API/Pipeline

  • bool h3dInit(void* device)
  h3dInit now has a parameter. On D3D11 pass a valid ID3D11Device device. On other platform pass 0x0.
  • new API: int h3dGetCaps( H3DCaps::List param ) and H3DCaps::List enum to query certain device capabilities: compressed texture format support, depth/shadow texture support, floating point texture support, mrt support, ... . Based on this the appropriate texture format can be loaded from assets or a supported pipeline can be selected at runtime.

  • Removed ResourceFlags::NoTexCompression, EngineOptions::TexCompression from the API and Sampler's allowCompression flag in material. Texture compression may be done in a preprocessing step or moved to a separate utility library that creates compressed textures on the fly and pass those to horde. Note: The compression wasn't working in the current OpenGL backend.

  • Changed TextureFormat H3DFormats::TEX_BGRA8 to H3DFormats::TEX_RGBA8 (and TextureFormats::BGRA8888 to TextureFormats::RGBA8888). Rationale: ES2 has no BGRA8888 format. Extensions like EXT_texture_format_BGRA8888 and GL_APPLE_texture_format_BGRA8888 adds support for this (but this is missing on some platform). I benchmarked rgba/bgra texture upload on iOS (powervr sgx 535,543), android (powervr sgx 540, mali400, adreno 205/220/320, tegra3) and there was no noticable difference in speed. Changing to rgba8 makes stbi texture upload faster (there is no need for CPU swizzling of texture). Volker on desktopGL: "I never noticed any measurable difference in speed and for texture streaming - between bgra8 vs rgba8 - maybe PBOs would make a bigger difference, but are currently not used by Horde3D."

  • Added PVRTCI and ETC1 texture format support

  • Added PVR,KTX file format support

D3D11 specific changes

  • D3D11 attribute->semantic mapping. I can't get the name of the attribute/vertex input element name in D3D11. So h3d attribute names are mapped to d3d11 semantics. vertPos -> POSITION0, texCoords0 -> TEXCOORD0, texCoords1 -> TEXCOORD1, normal -> NORMAL0, tangent -> TANGENT0, joints -> BLENDINDICES0, weights -> BLENDWEIGHTS0, parIdx -> BLENDINDICES0. So for example
     //GL definition
     attribute vec3 vertPos;
     //d3d11 definition 
     float3 vertPos : POSITION0; 
     // name of variable ('vertPos' in previous definition) doesn't matter, so one can use something like this too but I keep it to ease porting the shaders
     float3 vertexPosition  : POSITION0;
  • D3D11 samplers/textures are separate objects. I use name convention to link them: "sampler_" and "texture_" prefix for the names. For example
    //definition and sampling in GL
    uniform sampler2D albedoMap;
    vec4 albedo = texture2D( albedoMap, newCoords.st );
    // definition and sampling in D3D11
    Texture2D texture_albedoMap;
    SamplerState sampler_albedoMap;
    float4 albedo = texture_albedoMap.Sample( sampler_albedoMap, newCoords.xy );

TODO

  • add query for current renderer backend
  • Testing, Testing, Testing,... Create a Regression test app
  • Add formats etc2_atc, astc, bc4-7
  • SkipMipLevel option
  • GLES2: add multisampling using APPLE_framebuffer_multisample.
  • GLES2: add tex/rt float support with use OES_texture_float_linear, OES_texture_half_float_linear, OES_texture_float, OES_texture_half_float, EXT_color_buffer_half_float, EXT_color_buffer_float
  • GLES2: add non-power-of-two support with GL_OES_texture_npot
  • GLES2: srgb support with EXT_sRGB
  • GLES2: add EXT_disjoint_timer_query support