Kaushik Kulkarni

Implements a Loop Fusion Transformation

3

Loopy-flavored loop-fusion transformation corresponding to .

Provided test case fails on `main`. Draft because: - Fix for #600 must land first as it has broken out CI.

lp.precompute does not emit the correct `precompute_inames`

1

MWE: ```python import loopy as lp knl = lp.make_kernel( [ "{[iel, iface, idof, ifacedof]:" " 0

- [x] Depends on #293 (commits included here) - [ ] Depends on #350 (commits included here)

A desirable test case for this would be: ```python out[x, e, i] = sum([r, j], sum([j], Jac[x, e, r] * Rdiff[r, i, j] * u[e, j])) ``` should be transformed...

lp.buffer_array allocates more memory than necessary

1

Consider the kernel: ```python import loopy as lp knl = lp.make_kernel( "{[e, i, j]: 0

`lp.extract_subst` does not search for subexpressions of `Product` or `Sum`

3

Consider the kernel: ```python import loopy as lp knl = lp.make_einsum("k,ij,ij->i", arg_names=("a", "b", "c")) knl = lp.extract_subst(knl, "tmp_hoist", "b[i, j] * c[i, j]") ``` raises the following exception: ```python RuntimeError:...

`lp.assignment_to_subst`: Should it support 1-long axes as a special case?

1

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

[precompute] Makes the compute insn's deps. precise

1

Depend on an instruction only if it writes to the variables that the compute instruction reads.