https://gcc.gnu.org/bugzilla/show_bug.cgi?id=116498
--- Comment #3 from Alan Fisher <alancf0 at gmail dot com> --- > It seems odd that neither Pmode nor ptr_mode is in the set of integer modes > though. This does seem odd... One would think PSI should sit between HI and SI as here: https://gcc.gnu.org/onlinedocs/gccint/Machine-Modes.html . Again, as someone who knows little about gcc internals maybe this is not that useful but... Upon investigation this appears to be a result of how calc_wider_mode() in genmodes.cc works. From what I can see, - Each mode is iterated through - The objects in each mode are sorted according to some set of criteria (see below) - but the sorting is done only within the mode, not between modes - Following the sorting process of a mode in calc_wider_mode, the 'wider' member of the mode structure is set to the next member in the sorted list - The algorithm for emitting the mode_wider array involves iterating over each mode, printing out each object within, and its 'wider' member. - This means that all of the integer modes are sorted together within their category (2); the single partial integer mode is "sorted" within its category (3); and the latter is emitted in the mode_wider text output below the former - The comparison used to sort the objects within modes involves the 'bytesize' and 'precision' fields, the latter of which is used for partial integers to indicate their bit length. A number of other fields are also used. Most of this code / design seems to have been around untouched for a long time. Perhaps someone else can comment on whether this is correct output for genmodes. All the best, Alan
