I just joined the LLVM Dev list.
What is the status of the PTX BE?
thank,
Vince Schuster
I just joined the LLVM Dev list.
What is the status of the PTX BE?
Most of our testing has been on micro-kernels, though I have been able to generate correct (and mostly efficient) code for basic compute kernels. Support for texture fetches are not implemented. Device function calls are a current work-in-progress.
Basically, we support:
- Arithmetic on i1/i16/i32/i64/float/double
- Branching
- Global memory read/write
- Shared memory read/write
- SM 1.0 - SM 2.0
- PTX 2.0 - PTX 2.3
We have not put much time into optimizations yet. We do simple optimizations like folding mul/add pairs into mad, but that’s about it for now.
For some basic usage information with Clang, you can get my collection of examples at https://github.com/jholewinski/llvm-ptx-samples. These double as more robust test cases.
Feel free to contact me directly with any questions.