On 8/30/22, Undiscussed Groomed for Male Slavery, One Victim of Many <gmkarl@gmail.com> wrote:
1234 ET
i'm engaging a bug in neural network training plans in gneural networks. the call to memcpy was backward, copying the new unallocated memory over the data that was set. have not yet verified that i am interpreting the bug correctly.
more detail: there are a handful of parallel functions that put a structure on the stack as well as a pointer to it. the structure is filled via this pointer, rather than direct access to the stack object. after the structure is filled, the pointer is reseated to point to memory on the heap. a copy is then performed from the heap memory, to the stack. (the thing that seems a big error to me) then the unallocated heap memory is returned to the program. in two of the functions it looks to me like the copy is made from a different buffer with a similar name that has unrelated content. the error strikes me because it is similar to errors i struggle with in day to day activities, with a reversal that trashes the intent maybe paired with a use of the reversed thing before or under its reversal this thing i am scared of is not the main issue, as far as i can tell, it's still failing when i swap it to look right, so i may be describing my thinking issues rather than the code, who knows