The constant addprc is present in most Microsoft BASICs of the era from which this code was ported, and in other references this constant means "additional precision". On the C64, and hence the X16, this constant was set to 1.
For the issue at hand, the X register was set to #252, and then incremented before the store of the first byte on (pre-diff) line 165, which set the index to 253, (or with ZP wraparound, -3) relative to reslo, which is resho, or the first byte of the FP result quintet.
The negative offset was a useful space and cycle saving measure as it allowed the code to skip a compare instruction after incrementing X, but allow a branch for 0 and then for 1. Unfortunately, when attempting to use this in native mode on the 65C816, where the ZP wraparound behavior is absent, you'd cause corruption and garbage writing to the wrong address.
This change begins the X register's role in fdiv at 255, which is incremented to 0 before the first use, which (post-diff) stores the value of A to resho just like before. However, now we have added the compare instruction that allows our checks to branch for the 4th and 5th bytes respectively, just like before, but without depending on ZP wraparound behavior.
The constant
addprcis present in most Microsoft BASICs of the era from which this code was ported, and in other references this constant means "additional precision". On the C64, and hence the X16, this constant was set to 1.For the issue at hand, the X register was set to
#252, and then incremented before the store of the first byte on (pre-diff) line 165, which set the index to 253, (or with ZP wraparound, -3) relative toreslo, which isresho, or the first byte of the FP result quintet.The negative offset was a useful space and cycle saving measure as it allowed the code to skip a compare instruction after incrementing X, but allow a branch for 0 and then for 1. Unfortunately, when attempting to use this in native mode on the 65C816, where the ZP wraparound behavior is absent, you'd cause corruption and garbage writing to the wrong address.
This change begins the X register's role in
fdivat 255, which is incremented to 0 before the first use, which (post-diff) stores the value of A to resho just like before. However, now we have added the compare instruction that allows our checks to branch for the 4th and 5th bytes respectively, just like before, but without depending on ZP wraparound behavior.