A Soft Implementation Of ARB_gpu_shader_fp64 Is Slowly Being Tackled For Mesa

Last year there was a Google Summer of Code student working on a library to implement double-precision operations (FP64) in pure GLSL 1.30 as a benefit to older GPUs not having native FP64 capabilities. While that work didn't materialize as a solution in 2017 for those wanting "soft" ARB_gpu_shader_fp64 for being able to potentially expose OpenGL 4.0 on more R600g era GPUs, the work is ongoing.

Back in November was the release of libSoftFloat 1.0, the library being developed for emulating the FP64 support. That work accomplished last year was just as its own library and without any direct integration in Mesa.

A few Phoronix readers pointed out this week that the student (and now software engineer at Collabora) behind it, Elie Tournier, has started a Mesa_GL_ARB_gpu_shader_fp64 branch. That branch was worked on as of earlier this month and is working towards his soft FP64 aspirations but is so far incomplete and hasn't been touched in two weeks.

Meanwhile the libSoftFloat library received just a typo fix days ago as the first work since the v1.0 milestone last year.

There seem to be a lot of R600g era folks hoping to see GL_ARB_gpu_shader_fp64 for them, so hopefully this work will manage to materialize this year.