This set of patches implements the minimal packing algorithm required by GLES 2.0 specification (see http://www.khronos.org/registry/gles/specs/2.0/GLSL_ES_Specification_1.0.17.pdf p111, thank to Paul Berry for pointing this).
Currently any vector varying are occupying a full vec4 slot in output buffer. This means that drivers can only honor their MAX_VARYING_FLOATS if each varying is of vec4 type. The following piglit test shows this : http://lists.freedesktop.org/archives/piglit/2012-February/001850.html For instance, if hardware reports a 32 varyings support, shaders won't compile if there is more than 8 varyings, even if they are floats. The minimal varying algorithm required by the spec has several interesting properties : - If there is enough vec4 slot to store varyings, it won't attempt to pack anything (like currently) - Arrays access are left "untouched", meaning there is no swizzling hack to implement driver side (also meaning that array of varying size is limited to MAX_VARYING_FLOATS/4). - If there are only vec4, vec2 and scalar varyings (not in array), MAX_VARYING_FLOATS can always be honored. Of course more sophisticated algorithms can be implemented, however reshaping arrays/overlaping register (vec3/vec3/vec2 inside 2 vec4s) can be costly for simd/vector architectures (and packing algorithms are NP hard). It might be possible to do optimal packing for scalar architecture, when tgsi opcode support it. Thank for review. Regards, Vincent _______________________________________________ mesa-dev mailing list mesa-dev@lists.freedesktop.org http://lists.freedesktop.org/mailman/listinfo/mesa-dev
