John - Re: "...do you think that there would not be too many software companies coming out in the open to admit that their softwares are affected by errata #62, "
My interpretation is as follows. Most software is written with commercially available compilers - C, C++, Visual Basic, Visual C++, and a host of others.
Within these compilers are mathematical functions or libraries of functions that actually perform the floating point operations, when called upon to do so by the calling program.
As for C, there is an ANSII C standard that compilers must comply to -
I don't know for sure, but ANSII might dictate range checking and boundary checking before doing a type conversion - that is a Float to an Integer. These of course are done in sofware, NOT HARDWARE.
The core of the issue lies in how these floating point libraries were constructed - from a coding standpoint. Are range checks included, are boundary values checked, are illegal operations trapped, etc.?
If these functions and libraries contain the necessary safeguards to prevent illegal operations, such as converting a large floating point number into an integer that is too small to accept the number, then the chances of this Flag Errata causing a problem are remote - assuming again that the appropriate compilers are used.
Couple this to the small probability of most software actually using the FIST, or FISTP instructions in the first place (float to integer conversion), along with the probability that ONLY LARGE NEGATIVE NUMBERS trigger the Flag Errata, and you are faced to conclude that the probability of encountering such an error is remote - very remote.
Witness the 18+ months that Pentium Pro processors have been on the market and in workstations from Intergraph, IBM, Compaq, NetPower, Amdahl, etc.
Where are the faults? Show me the faults!
Paul |
