Benchmark for PencilFFT slower than FFTW #37

musoke · 2021-03-16T02:55:27Z

When benchmarking, an FFT of an array is much slower when it is contained in a PencilGrids than Grids.

mpirun -np 1 julia benchmarks/fft.jl

Is this because the Grids FFT is greedy and uses extra threads?

The text was updated successfully, but these errors were encountered:

Add a script to benchmark different ways of doing a single step. Related: #37

Add a script to benchmark different ways of doing a single step. This makes it clear that something in the implementation of `PencilGrids` is slowing it down: a single step with `PencilGrids` is much slower than `Grids` and using `PencilArrays` directly. Related: #37

Give concrete types to some members of `PencilGrids`. This allows the compiler to infer more and optimize. https://docs.julialang.org/en/v1/manual/performance-tips/#Avoid-fields-with-abstract-type Fixes: #37

Add a script to benchmark different ways of doing a single step. This makes it clear that something in the implementation of `PencilGrids` is slowing it down: a single step with `PencilGrids` is much slower than `Grids` and using `PencilArrays` directly. Related: #37

Give concrete types to some members of `PencilGrids`. This allows the compiler to infer more and optimize. https://docs.julialang.org/en/v1/manual/performance-tips/#Avoid-fields-with-abstract-type Fixes: #37

Add a script to benchmark different ways of doing a single step. This makes it clear that something in the implementation of `PencilGrids` is slowing it down: a single step with `PencilGrids` is much slower than `Grids` and using `PencilArrays` directly. Related: #37

Give concrete types to some members of `PencilGrids`. This allows the compiler to infer more and optimize. https://docs.julialang.org/en/v1/manual/performance-tips/#Avoid-fields-with-abstract-type Fixes: #37

Give concrete types to some members of `PencilGrids`. This allows the compiler to infer more and optimize. https://docs.julialang.org/en/v1/manual/performance-tips/#Avoid-fields-with-abstract-type Partial fix for #37

Add a script to benchmark different ways of doing a single step. This makes it clear that something in the implementation of `PencilGrids` is slowing it down: a single step with `PencilGrids` is much slower than `Grids` and using `PencilArrays` directly. Related: #37

Give concrete types to some members of `PencilGrids`. This allows the compiler to infer more and optimize. https://docs.julialang.org/en/v1/manual/performance-tips/#Avoid-fields-with-abstract-type Partial fix for #37

musoke · 2021-06-22T12:32:55Z

jipolanco/PencilFFTs.jl#35

musoke added this to the 0.1 milestone Apr 28, 2021

musoke added a commit that referenced this issue Jun 17, 2021

bench: Add benchmark file for single step

1725636

Add a script to benchmark different ways of doing a single step. Related: #37

musoke modified the milestones: 0.1, 0.2 Jun 22, 2021

musoke removed this from the 0.2 milestone Oct 11, 2021

musoke closed this as completed Dec 23, 2021

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Benchmark for PencilFFT slower than FFTW #37

Benchmark for PencilFFT slower than FFTW #37

musoke commented Mar 16, 2021

musoke commented Jun 22, 2021

Benchmark for PencilFFT slower than FFTW #37

Benchmark for PencilFFT slower than FFTW #37

Comments

musoke commented Mar 16, 2021

musoke commented Jun 22, 2021