Preallocate imfilter kernel memory (#45)

* Preallocate imfilter kernel memory in GPU * Update tests * Refact krn buffer optimization * Revert preallocate imfilter kernel * Add CUDA.zeros to pad kernel
JuliaEarth · Aug 26, 2022 · d55da6c · d55da6c
1 parent 522472a
commit d55da6c
Showing 1 changed file with 2 additions and 2 deletions.
diff --git a/src/imfilter.jl b/src/imfilter.jl
@@ -12,8 +12,8 @@ function imfilter_gpu(img, krn)
   T = eltype(img)
 
   # pad kernel to common size with image
-  padsize = size(img) .- size(krn)
-  padkrn  = padarray(krn, Fill(zero(T), ntuple(i->0, N), padsize))
+  padkrn = CUDA.zeros(size(img))
+  copyto!(padkrn, CartesianIndices(krn), CuArray(krn), CartesianIndices(krn))
 
   # perform ifft(fft(img) .* conj.(fft(krn)))
   fftimg = img |> CUFFT.fft