Philip Turner

If that's the case, each random operation would take on the order of 1000 μs (1 millisecond). After the optimization, it would take either 100 μs or 25 μs. Overhead...

https://github.com/philipturner/swift-colab/issues/16

Converting to a draft because Swift-Colab has planned support for JupyterLab.

Yes, but we'd be reverse-engineering AIR, which is undocumented. Apple could change how it works at any moment and break our code. Also, we don't know how to pack AIR...

I read somewhere that Vulkan SPIR-V puts heavy restrictions on control flow, compared to OpenCL SPIR-V. Perhaps we can clone and modify SPIRV-Cross to translate OpenCL SPIR-V into MSL. We...

I see the source code is located at https://github.com/tadeaustria/llvm. Is it possible to reuse this work in hipSYCL? I'm trying to just dump raw SPIR-V, in any dialect, then analyze...

> primarily regarding control flow and pointers - it's not a straightforward conversion and almost a different language. This is where I originally learned of the control flow constraints of...

> does support some of those features. Are you saying the `goto` and labeled `break` features are supported in OpenCL C? > You can directly use it with -driver-mode=openclc -x...

Thanks so much! Here's steps to reproduce. Create a file called `vecAdd.cl`: ```opencl __kernel void vecAdd( __global float *a, __global float *b, __global float *c, const unsigned int n) {...

Now, here's a kernel with a `goto` statement. I think `goto` is fine because you're lowering down to machine code, which just executes `jump` assembly instructions. When raising into OpenCL...