As we are not going to write an OpenCL or Cuda compiler, which already exists, we will start working from an IR of the kernels to run. As Vulkan is around the corner and will use SPIR-V as input for shaders and kernels, and it is (will be) possible to generate SPIR-V out of Cuda kernels, we will use SPIR-V as our input IR.
pmoreau claimed this task.
pmoreau raised the priority of this task from to High.
pmoreau changed the visibility from "All Users" to "Public (No Login Required)".
pmoreau removed subtasks: T80: Implement decorations, T76: Fix FunctionCall, T62: Parse group decorations, T61: `getIndirect()` returns `nullptr` in `CodeEmitterNV50::emitLoad()`, T59: Fix Fermi+ support.Comment Actions
Use Compute (SPIR-V 1.0) instead