Understanding the performance-accuracy tradeoffs of floating-point arithmetic on GPUs | IEEE Conference Publication | IEEE Xplore