Address #13: Use sympy.codegen.rewriting.optimize

It's really comfortable to write optimizations in terms of sympy.codegen.rewrite.RewriteOptim:

    # Evaluates all constant terms
    evaluate_constant_terms = ReplaceOptim(
        lambda e: hasattr(e, 'is_constant') and e.is_constant,
        lambda p: p.evalf()
    )

This PR adds a parameter sympy_optimizations to the create_*_kernel functions that applies the list of optimizations to the assignments before creating the AST.

sympy.codegen.rewrite already has some optimizations. Some similar to the optimizations of pystencils. For example create_expand_pow_optimization(limit) is really similar to the logic in CustomSympyPrinter._print_Pow.

See #13 (closed)

Problem: old versions of sympy (e.g. from ubuntu CI) don't have sympy.codegen.rewrite. The optimizations are skipped in that case. test_and_coverage applies all optimizations.

We could also try to implement a fma-optimization (fused-multipy add) with that and sympy.Wild.

changed title from Address #13 (closed): Use codegen.rewriting.optimize to Address #13 (closed): Use sympy.codegen.rewriting.optimize

changed the description

added 1 commit

• 7e32d75c - Do not evalf() if expression is_integer

Compare with previous version

added 1 commit

• bc55cde3 - Do not evalf() if expression is_integer

Compare with previous version

added 1 commit

• 328ea64f - Add assumption functions to cast_func

Compare with previous version

added 1 commit

• 0cc6af3d - Add assumption is_integer == True to IntegerFunctionTwoArgsMixIn

Compare with previous version

Looks good! I didn't know about the sympy optimization rules. One small thing though: Is it necessary to call this in all the create_kernel functions? I'd prefer that the user calls the optimization explicitly, or create a simplification strategy. At least this is the way all the other optimization currently work. The user would then separately do

assignments = optimize(assignments)
ast = create_kernel(assignments)

added 45 commits

• 0cc6af3d...64734915 - 33 commits from branch pycodegen:master
• 938e0de0 - Let iterator of AssignmentCollection iterate over all_assignments (not only main_assignments)
• 280f28f9 - Add pystencils.optimizations
• 7ab81857 - Apply default optimizations in `create_kernel` routines
• 0c671ea0 - Add expm1 test
• f6b324a5 - Add sympy optimization: evaluate_constant_terms
• 38bea5c9 - Regard case that pystencils.astnodes.Conditional is in assignments or...
• 47f9d708 - Rename: pystencils.optimizations -> pystencils.math_optimizations
• 873969f9 - Avoid optimizations of IntegerFunctionTwoArgsMixIn
• d8327971 - Rename in create_kernel: optimizations -> sympy_optimizations
• b0aa8134 - Do not evalf() if expression is_integer
• 58558da9 - Add assumption functions to cast_func
• 4ed5b6a6 - Add assumption is_integer == True to IntegerFunctionTwoArgsMixIn

Compare with previous version

added 16 commits

• 4ed5b6a6...ddbe84da - 4 commits from branch pycodegen:master
• 268f3e26 - Let iterator of AssignmentCollection iterate over all_assignments (not only main_assignments)
• 3dd6f491 - Add pystencils.optimizations
• b731bc49 - Apply default optimizations in `create_kernel` routines
• 1286f042 - Add expm1 test
• 03fe9ab0 - Add sympy optimization: evaluate_constant_terms
• 3531885b - Regard case that pystencils.astnodes.Conditional is in assignments or...
• 2e23b82e - Rename: pystencils.optimizations -> pystencils.math_optimizations
• dfc5428b - Avoid optimizations of IntegerFunctionTwoArgsMixIn
• b3af8449 - Rename in create_kernel: optimizations -> sympy_optimizations
• 5cdf85c4 - Do not evalf() if expression is_integer
• 06964419 - Add assumption functions to cast_func
• c9c9ca55 - Add assumption is_integer == True to IntegerFunctionTwoArgsMixIn

Compare with previous version

added 1 commit

• e8a92a9e - Suppress evaluation of constants in test_sum_prod

Compare with previous version

mentioned in merge request !47 (closed)

added 24 commits

• e8a92a9e...71b8767b - 20 commits from branch pycodegen:master
• 8479aa27 - Let iterator of AssignmentCollection iterate over all_assignments (not only main_assignments)
• da5cfa52 - Add dtype assumptions to cast_func
• c3a4fb73 - Add property is_integer -> True to IntegerFunctionTwoArgsMixIn
• 8d610054 - Add pystencils.math_optimizations

Compare with previous version

mentioned in merge request !54 (closed)

@bauer I made this PR less controversial by removing all changes to kernelcreation.

I also didn't want to add another parameter to create_kernel since it already has a lot. However, I want pystencils to always optimize and still give the user the chance to add own optimizations or remove some. I thought this concept is pretty abstract to include custom optimizations for sympy.Pow (no in CustomSympyPrinter and by that steer the behavior of pystencils.

So this would be a parameter to create_kernel to avoid parameters to create_kernel in future.

Created a follow-up PR with these changes: !54 (closed)

merged

mentioned in commit 97307848