loopy

loopy copied to clipboard

OpenCL targets aren't aware of some math built-ins

1

I see the new function interface is much more careful about what's in the namespace (which is great!), but I think a number of the [built-in math functions](https://www.khronos.org/registry/OpenCL/specs/3.0-unified/html/OpenCL_C.html#math-functions) aren't currently...

zachjweiner

`offset` has no business being in the IR. It's a concern that's specific entrypoints in OpenCL and *nothing* else. So it should be a concern for *just that* target. How...

inducer

[loopy.match] Targeting reduction expression in an instruction

3

In the following kernel: ```python knl = lp.make_kernel( ["{[i]: 0

kaushikcfd

Cf. this FIXME: https://github.com/inducer/loopy/blob/3325b829b82e37a3a411c979280acb31d0725d0c/loopy/check.py#L687-L688

inducer

TODO: - [ ] Port `lp.split_iname` so that it uses this interface

kaushikcfd

`lp.privatize_temporaries_with_inames` fails for multiple inames sweeping the same temporary

8

For the kernel: ```python knl = lp.make_kernel( "{[i,j]: 0

kaushikcfd

Loopy's host-side shape inference cannot solve system of equations

1

```python knl = lp.make_kernel( "{[i, j]: 0

kaushikcfd

For example: https://github.com/inducer/pytential/pull/94#discussion_r653094760

inducer

Loopy's codegenerator fails with non-deterministic error

3

For the quite simple kernel ```python import loopy as lp knl = lp.make_kernel( ["{ [i_outer, i_inner] : 4i_outer

kaushikcfd

conj() generates broken code for non-complex types

1

Working code for complex conj(): ```c #define lid(N) ((int) get_local_id(N)) #define gid(N) ((int) get_group_id(N)) #if __OPENCL_C_VERSION__ < 120 #pragma OPENCL EXTENSION cl_khr_fp64: enable #endif #define PYOPENCL_DEFINE_CDOUBLE #include __kernel void __attribute__...

matthiasdiener