Contiguous row-major native kernel; initialise before accessing.
Writable view of (2x+1)(2*y+1) floats.
Rml::ConvolutionFilter::operator[] as a checked shared row.
indexZero-based row index.
Row retaining its native allocation.
Rml::ConvolutionFilter::Initialise with square or rectangular radii.
radiiInteger radius, or integer x/y radii.
operationSum or Dilation.
False for negative radii, preserving the previous kernel.
Rml::ConvolutionFilter::Run; only destination alpha is changed.
destinationWritable A8 or RGBA8 bytes.
dimensionsDestination region size.
strideDestination bytes per row.
formatDestination color format.
sourceSource alpha or RGBA bytes.
sourceDimensionsTightly packed source size.
sourceOffsetSource origin relative to destination.
sourceFormatSource color format.
Buffers must be live, nonoverlapping Uint8Array views without concurrent shared backing. Source rows are tightly packed. Finite kernel weights are checked before writes. Negative Sum weights require nonnegative output and incur a native preflight; values above 255 saturate as upstream specifies.
StaticallocateAllocate zeroed Native-owned bytes for filter sources or destinations.
byteLengthNonnegative byte count within native int bounds.
Writable view retaining only the native allocation.
Native Rml::ConvolutionFilter with a writable shared float32 kernel.
Remarks
Each run crosses into Native once. Native owns the kernel; exported views retain their allocation after reinitialisation, collection and shutdown.