ASV Benchmarks Integration

[vchamarthi]

Adds a complete ASV benchmark suite for mkl_fft and wires it into the internal Jenkins CI pipeline.

Benchmarks added (benchmarks/benchmarks/):

bench_fft1d.py — 1-D C2C and R2C/C2R, power-of-two and non-power-of-two
bench_fftnd.py — 2-D and N-D C2C and R2C/C2R, square, non-square, non-power-of-two
bench_numpy_fft.py — full coverage of mkl_fft.interfaces.numpy_fft
bench_scipy_fft.py — full coverage of mkl_fft.interfaces.scipy_fft including Hermitian 2-D/N-D
bench_memory.py — peak RSS for 1-D, 2-D, and 3-D transforms
__init__.py — pins MKL_NUM_THREADS=4 on machines with ≥4 physical cores for consistent cross-machine results (A hack to keep the results consistent on random nodes until CI finds a stable benchmarking machine)
Config (benchmarks/asv.conf.json)

Docs (benchmarks/README.md): structure, coverage, threading rationale, local run commands, CI flow.

[Copilot]

Pull request overview

Adds an ASV benchmark suite for mkl_fft (root API + NumPy/SciPy interfaces + memory) along with configuration, docs, and ignore rules to support running and storing benchmark artifacts in CI.

Changes:

• Introduces ASV benchmark modules covering 1-D, 2-D, and N-D FFTs across multiple dtypes and shapes (including Hermitian variants for SciPy).
• Adds ASV configuration (asv.conf.json) and documentation for local/CI execution.
• Adds benchmark package initialization that pins thread counts via environment variables, plus .gitignore entries for ASV artifacts.

Reviewed changes

Copilot reviewed 8 out of 9 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
benchmarks/benchmarks/bench_scipy_fft.py	Adds full-coverage ASV benchmarks for mkl_fft.interfaces.scipy_fft including Hermitian 2-D/N-D.
benchmarks/benchmarks/bench_numpy_fft.py	Adds full-coverage ASV benchmarks for mkl_fft.interfaces.numpy_fft including Hermitian 1-D.
benchmarks/benchmarks/bench_memory.py	Adds ASV peak-RSS benchmarks for core mkl_fft transforms across 1-D/2-D/3-D.
benchmarks/benchmarks/bench_fftnd.py	Adds 2-D and N-D root-API benchmarks including non-square and non-power-of-two cases.
benchmarks/benchmarks/bench_fft1d.py	Adds 1-D root-API benchmarks for power-of-two and non-power-of-two sizes (C2C and R2C/C2R).
benchmarks/benchmarks/__init__.py	Adds thread pinning logic via MKL_NUM_THREADS/OMP_NUM_THREADS/OPENBLAS_NUM_THREADS.
benchmarks/asv.conf.json	Adds ASV project configuration (dirs, branches, timeouts, regression thresholds).
benchmarks/README.md	Documents benchmark structure, coverage, threading rationale, and local/CI run commands.
.gitignore	Ignores ASV artifact directories.

[ndgrigorian]

should we be manipulating anything but the MKL threads? I'm not sure one way or the other, just curious what the reasoning would be or if it's expected of users

[ndgrigorian]

Suggested change

	class TimeFFT1D:
	class BenchFFT1D:

aligns with the ASV benchmarks in mkl_umath

[ndgrigorian]

similar to in umath, I think we should refactor the common aspects into a base class like BenchFFT and if we need specialized params, just subclass them

means it will be easier to maintain in the future too

[ndgrigorian]

don't think we need this

[ndgrigorian]

I would add this to the table instead, so it's clear which benchmarks have which types

[ndgrigorian]

I would restructure this to focus first on how the user can override MKL_NUM_THREADS (i.e., "The number of threads can be set by setting MKL_NUM_THREADS in the environment [...]") and then discussing why we chose the defaults we did and what they are.

[ndgrigorian]

don't think we need this, we can just refer to build and installation instructions

[ndgrigorian]

drop this, it directly contradicts the previous lines which assume the benchmarks folder is the cwd anyway, and the user should already understand it

[ndgrigorian]

we don't need any of this and shouldn't reference innersource anywhere, external/open source users will just be confused