On 23/10/13 18:16, Walter Bright wrote:
To precisely control maximum precision, I suggest using inline assembler to use the exact sequence of instructions needed for double-double operations.
Could be relevant here: last year I wrote some code which divided up the closed interval [0, 1] into small increments. If the division was by powers of 10, dmd would wind up mis-calculating; with powers of 2, it was fine.
By contrast GDC and LDC were fine, which makes me think their backends must implement effective decimal support for floating-point within certain limited bounds.
