Martin, first - congrats on being recognized in EE Times.
Of course I'd rather it have been for something else - why do I have the nagging feeling I am somehow responsbile for your "trivial observation."
I still have trouble with this bug as being classified as anything other than an errata. I have played with some code that takes really big numbers and stuffs them into smaller types, where they can't be represented and nothing happens, other than worthless data, as expected. I am doing this on a Sparc, now if I divide by zero - I get a core dump, but nothing when I cause implicit narrowing with a loss of real data. I have not verified under C++ that a catch-able exception occurs under these same conditions, I'll try that next.
We'll all know tomorrow how Intel decides to deal with it, but I fear if it's classified errata - some folks will not be content as the issue is rather contentious and a time tested route to instant fame.
Speaking only as an investor, I hope it's over tomorrow. As a programmer I hope they fix it with a new mask or some other technique quickly. I would like to say I have no ill will towards you, facts is facts, as long as they don't get overblown into hype. I am not in a position to speak for Intel on the severity of this bug or how any other software vendor classifies it. The media on the other hand is like a room full of child screaming for attention armed mostly with hype at this point trying to describe complex problems to what they consider the uneducated masses.
Again, as an investor I am planning on holding some cash in reserve to load up if the problem becomes costly to Intel in the end. I have heard people and myself included say "I wish I had bought Intel during the floating point bug". Well, now may be your chance, keep some powder dry as we may be having that second chance any day now. |
