[gouarin]

Checklist

• The title and commit message(s) are descriptive.
• Small commits made to fix your PR have been squashed to avoid history pollution.
• Tests have been added for new features or bug fixes.
• API of new functions and classes are documented (no public API changes).

Description

xreducer_stepper::aggregate_impl had a second overload for keep_dims. It walked every dimension of the expression, looked each one up in m_axes with std::find, and recursed through the dimensions that are not reduced without doing anything there. The other overload walks m_axes directly.

Both apply the same steps to the underlying stepper, since keep_dims only changes the shape of the result. The extra size != 0 test of the keep_dims overload is redundant too: aggregate() already returns the initial value for empty expressions. This PR removes the keep_dims overload.

GCC 13 at -O3 inlined that recursion and warned about m_e.shape()[dim] with dim == dimension() (-Warray-bounds, 18 occurrences in test_xmath). The axis checks in the constructor rule that path out, and the warning goes away with the overload.

xreducer.keep_dims_lazy_all_axes compares lazy keep_dims sums with immediate ones for every set of axes of a 4-D array, in row- and column-major order and with an initial value. It passes before and after the change, also under AddressSanitizer and UBSan.

[codspeed]

Merging this PR will regress 1 benchmark

1 warning

⚠️ Unable to measure the execution of 1 benchmark

This benchmark measured no execution time, so its result is not comparable and counts as unchanged. This usually happens when the compiler removes the code under test.

Prevent compiler optimizations

See affected benchmarks

⚡ 4 improved benchmarks
❌ 1 regressed benchmark
✅ 250 untouched benchmarks

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
❌	xshape_initializer[std\:\:array<std\:\:size_t, 4>]	151.7 ns	180.8 ns	-16.13%
⚡	xshape_access[std\:\:array<std\:\:size_t, 4>]	154.4 ns	125.3 ns	+23.28%
⚡	reducer_reducer[100000x10 / axis 0]	12.7 ms	11.2 ms	+12.88%
⚡	reducer_reducer[10x100000 / axis 0]	12.7 ms	11.2 ms	+12.88%
⚡	reducer_reducer[10x100000 / axis 1]	8.1 ms	7.3 ms	+11.48%

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing gouarin:simplify-reducer-aggregate (30bc8ea) with master (57f3d07)}