[fix comments on alternatives to shufpd
peter@cordes.ca**20080318201949] {
hunk ./rshift.asm 141
+C	movdqa	%xmm3, %xmm7
hunk ./rshift.asm 150
-C good:  pshufd works, and gave 2.0 cycles/limb on conroe, but 2.5 on Harpertown. (?!).  3.0 on K8
+C good:  2.5c/l on Conroe.  w/pshufd ..,%xmm7, %xmm6 before the load, 2.0 cycles/limb on conroe, but 2.5 on Harpertown. (?!), and  3.0 on K8.
hunk ./rshift.asm 152
+C	punpcklqdq %xmm7, %xmm6		C dest=dest[0],src[0]
hunk ./rshift.asm 154
-C the best alternative for getting limb1 in the low part of a register
hunk ./rshift.asm 160
+C the best alternative for getting limb1 in the low part of a register
}