Open mikrosk opened 2 days ago
I think its not relevant for us (maybe it was based on some amiga example). We don't have any libraries that are compiled with -mbaserel, and even if we have, i guess there will be other places where a4 is not preserved in assembler code.
But does it really cripple performance? Instead of copying 5*44+36
bytes, you could then copy/zero 5*48+16
bytes, but that are still 6 movem.
But does it really cripple performance? Instead of copying 5*44+36 bytes, you could then copy/zero 5*48+16 bytes, but that are still 6 movem.
Hmm, true. I guess this could pay off in case we unroll the loop for 512B blocks.
Both
bzero
andbcopy
contains same comment: https://github.com/freemint/mintlib/blob/dbfab2d80bd265e2c537f179ccf26f1747a73b01/string/bzero.S#L92 https://github.com/freemint/mintlib/blob/dbfab2d80bd265e2c537f179ccf26f1747a73b01/string/bcopy.S#L97 essentially crippling performance with omittinga4
register in the copy loop. Is this still relevant? It has been there since the initial revision.