On Mon, 9 Jun 2025 22:24:08 GMT, Andy Goryachev <ango...@openjdk.org> wrote:
> the `MaskMarlinAlphaConsumer` field in `RendererContext` is created/resized > in DMarlinRasterizer:102 > > perhaps we could refactor this code to hide the public field and ensure the > single threaded access? That would get into the sort of changes I'd prefer not to make, at least not as part of the initial Unsafe --> FFM conversion, although Laurent could comment on that. Other possibilities include: using thread-local storage for the ALPHA_MAP_XXX off heap arrays (in practice, they are only used from the Quantum renderer thread, so there would only ever be one copy) or dispensing with off-heap arrays entirely for the ALPHA_MAP, and using plain Java arrays. Given how these static arrays are used, I'm not sure using off-heap memory is really buying much in terms of performance (as opposed to the dynamic edges array, which is the primary use of off-heap memory). Really, though, the simplest thing to do for now is to use shared (rather than confined) access for these two arrays and file a follow-up to optimize it if needed. ------------- PR Comment: https://git.openjdk.org/jfx/pull/1814#issuecomment-2957225458
