Andreas Klöckner

Unnecessary CUDA API calls in the CUDA backend

1

Watching some CL program (gory details [here](https://notes.tiker.net/s/3D3-VhuSf)) execute on pocl's CUDA backend via `nvprof` (Nvidia's profiling tool) gives me ``` API calls: 52.24% 1.20894s 72145 16.757us 7.6450us 22.126ms cuLaunchKernel 19.36%...

Core affinity for pthreads?

36

Flipping through the pthreads backend, it didn't seem as though core affinity for the created threads was getting enabled anywhere. If that's indeed the case, that might be an easy...

Valgrind "invalid write" on rank-1 matrix fill kernel

6

I've come across [another example](https://gist.github.com/inducer/8d1ad78c548bf053a85eabc015804484) of a kernel that appears to be miscompiled, resulting in a segfault or complaints about heap corruption. When I run that script on top of...

Outerloop vectorizer fail

4

This kernel here: https://gist.github.com/inducer/8f7cd72829c85acc1d3fcb9c4a5dae05 gets vectorized into full-width 256-bit vectors without a problem by both Intel CL and ispc. (Note how the workgroup size already conveniently matches the expected vector...

Loop variable use in closure not considered

1

Running ``` flake8==3.9.2 flake8-bugbear==21.4.3 flake8-polyfill==1.0.2 flake8-quotes==3.2.0 ``` on ``` def main(): def f(x): print(i, x) for i in [1, 2, 3]: f('HI') ``` gives me ``` bugbearbug.py:5:9: B007 Loop control...

Error on alias (but no name) in `CITATION.cff`

1

For PyOpenCL's [2022.1.6 release](https://github.com/inducer/pyopencl/releases/tag/v2022.1.6), [`CITATION.cff`](https://github.com/inducer/pyopencl/blob/6b68b79e9400e802755735f366eda221973cb606/CITATION.cff) contained the name of an individual (@gw0) for whom only an alias, but no first and family names are known. I represented this in `CITATION.cff`...

This was a dead-end while debugging #77. No need for it currently, but who knows? Maybe we'll need it sometime.

The analogous issue to https://github.com/inducer/pytato/issues/163 also exists in pymbolic. cc @alexfikl

Stop auto-flattening

3

FP arithmetic isn't associative, after all.

cc @kaushikcfd Prototype remap: https://gist.github.com/inducer/b293430efb27c50d0048cd278540f038