Dear Albert ARIBAUD,
In message 20130418082027.4b5ea191@lilith you wrote:
#ifdef USE_HOSTCC crc = htobe32(crc); memcpy(output, &crc, sizeof(crc)); #else crc = cpu_to_be32(crc); put_unaligned(crc, (uint32_t *)output); #endif
This produces the same code as my original patch. If this is acceptable then I will do that, although it doesn't really seem any better.
Wolfgang may not like it any more than put_unaligned_be32() as it builds upon it (and the disk patch could have used the be32 version as
Indeed.
well). Personally, I think we should allow and use these...
... and work on optimizing their implementation for ARM; we should be able to reduce such put()s and get()s to a few instructions. And to
OK - and what about the other architectures that suffer from the same issues?
avoid any misunderstanding, yes, I volunteer for the optimizing work. :)
I really dislike introducing such custom functions when we have standard functions available that implement the same purposes.
Checking Linux code (as U-Boot is not representative here):
-> find * -name '*.c' | wc -l 18362 -> find * -name '*.c' | xargs fgrep -l put_unaligned | wc -l 136
i. e. just 0.75% of the source files actually use any of the "put_unaligned*()" variants - it is a highly exotic function.
htobe32() is even worse - just a single source file in the whole Lnux tree uses it (arch/um/drivers/cow_user.c).
Can we not stick to standard functions? Instead of htobe32() we should use htonl() which is available both in U-Boot and in the host environment.
Best regards,
Wolfgang Denk