Does a4 have to be sacrificed in bzero/bcopy?

[mikrosk]

Both bzero and bcopy contains same comment:

mintlib/string/bzero.S

Line 92 in dbfab2d

| exclude a4 because of -mbaserel

mintlib/string/bcopy.S

Line 97 in dbfab2d

| exclude a4 because of -mbaserel

essentially crippling performance with omitting a4 register in the copy loop. Is this still relevant? It has been there since the initial revision.

[th-otto]

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.

[mikrosk]

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.