x86 descriptor base fetching performance opportunities

From: cutaway
Date: Fri Jun 17 2005 - 08:33:21 EST

This is something I've just started investigating, but preliminary
benchmarks of trial instruction sequences look VERY favorable for
conditional compilation of different code for those CPU's that can do a
BSWAP instruction.

The Linux src appears not conditionalized in any way in its handling of
descriptor get/set operations at the moment.

The fetching of descriptor bases is a necessarily clumsy affair on 386
resulting in several shift/rotate/masking operations because the upper 16
bytes of the base are split in a rather un-handy way in the 2nd dword of a

However, for 486 and better CPU's this un-handy layout looks like it can be
mitigated by BSWAP. ex(pseudo-asm)

movl 4(des_ptr),eax // Take hi-dword of descriptor
rol 8,eax // AL=des-hi8, AH=des-hi-mid8
bswap eax // Now, EAX high 16 == high 16 of descriptor
movw 2(des_ptr),ax // Fill in the low 16 bits

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/