AMD Radeon GPU compute update
Memory operations: Vertex buffer read, atomic operations, cacheless write are working, which means that the most problematic issues with compute support
have been taken care of.
It is possible that, there are still problems with thread scheduling, I haven't tested the code with too many threads. But with but relatively small amount of work the code can be made usable, and GPU machine code
can be used.
Next step is some kind of compatibility with GPU binaries compiled from
OpenCL by the AMD compiler.
code repository:
git://github.com/rakadam/HD-Radeon-Compute.git
git://github.com/rakadam/AMD-GPU-Asm-Disasm.git
The code is still a bit buggy (need better makefiles and some refactoring, some libdrm functionality has been duplicated in C++ too)
User login
Latest posts
-
Structured code generation for AMD GPUs are now supported in Minotaurus/LLVM
-
Julia demo can be compiled through LLVM now. We tested the new LLVM - Minotaurus interface today successfully.
-
Julia demo runs on AMD Fusion now. The speedup is around 4x compared to CPU. With the arrival of the Fusion platform we decided to tune the code generation of Minotaurus to support AMD GPUs as well.
-
Please find here our new demo video of Minotaurus. As an actual state of our main project, the system can compile the same C++ code parts to both CPU and GPU. Any comments are welcome! ...
This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement No 767912.