René's URL Explorer Experiment

{"@context":"https://schema.org","@type":"DiscussionForumPosting","headline":"ComplexMultiply redundant operations","articleBody":"In the fftconvolve.hpp (and perhaps elsewhere) for CUDA (arrayfire/src/backend/cuda/kernel/fftconvolve.hpp) complex multiplies are performed like so:\r\n\r\n```\r\n        out.ptr[ridx2].x = c1.x*c2.x - c1.y*c2.y;\r\n        out.ptr[ridx2].y = (c1.x+c1.y) * (c2.x+c2.y) - c1.x*c2.x - c1.y*c2.y;\r\n```\r\n\r\nIs there a reason for this form for the multiply? It would appear to have several redundant operations that may reduce performance and do increase floating point propagation error.\r\n\r\nThe multiplies should be:\r\n\r\n```\r\n        out.ptr[ridx2].x = c1.x*c2.x - c1.y*c2.y;\r\n        out.ptr[ridx2].y = c1.x*c2.y + c1.y*c2.x;\r\n```\r\n","author":{"url":"https://github.com/ewanbarr","@type":"Person","name":"ewanbarr"},"datePublished":"2016-12-19T13:57:15.000Z","interactionStatistic":{"@type":"InteractionCounter","interactionType":"https://schema.org/CommentAction","userInteractionCount":2},"url":"https://github.com/1679/arrayfire/issues/1679"}