BreakPoint
Rube Goldberg Code
EXT has on Intel 32-bit 80-bits. On Sparc it had 128-bit but that was software emulated. On all other platforms including Intel 64-bit, EXT === DBL.
@paul_cardinale wrote:
Right. But on Intel machines, all FP operations are done internally with 80 bits; the extra (guard) bits ensuring that the result will be accurate to the full precision of a 64-bit FP number. When you use EXT format, you have no guard bits, and (depending on the complexity of the calculation) the accumulated truncation errors can yield a result not much more accurate than a 64-bit FP number.
There is very anecdotal evidence that this is not guaranteed, for example in a thread switch these registers get swapped out as 64bits, and the extra precision is irreversibly lost. (I think it is even more noticeable if you use the parallel FOR loop).
@Jacobson-ni wrote:
The 2D array is just a constant after the first value right?
I don't think we can draw any conclusions because even the intent is not clear.
It's not even really a 2D array, just data that would fit equally well into a 1D array caused by the fact that we are autoindexing on the backwards wires creating an array of one element (4, or 3). I am also pretty sure the feedback node was inserted automatically because they created a loop (see also).
