On Thu, Jan 12, 2012 at 6:59 PM, Farley, Peter x23353
<[email protected]> wrote:
> EX is indeed expensive, but my guess (untested) is that an EXecuted MVC for
> small lengths (not only under 256 but even less) is probably still more
> efficient than MVCL for those lengths, and *definitely* more efficient than
> MVCLE. My prior experiences in replacing MVCL/CLCL's with multiple MVC/CLC's
> and even MVC/CLC loops for "small" areas (FSVO "small") is that MVCL/CLCL
> loses almost every time.
I have a regular need for a CLC over 0..8 byte (a kind of "pseudo
wildcard" pattern where a trailing * specifies matching of the
preceding substring).
Since it's already Friday here, this is what I came up with after some
experiments (forgive my SPM accent, I trust the intentions are clear).
I'm open for ideas...
LA R1,OP2+L'OP2 Beyond string in case no spaces
TRT OP2,SPACE Find first ' ' in pattern
CR R1,R3
COND NOTEQUAL,DECR,R1 Point at last non-blank, if any
CLI 0(R1),C'*'
IF NOTEQUAL
CLC OP1,OP2 No '*' - just compare them
ELSE ,
SR R1,R3 Compute length before '*'
IF NOTZERO
DECR R1
INLINEX R1,CLC,OP1(0),OP2
FI ,
FI ,
The INLINEX is my macro and generates (in case of the CLC) this
INLINEX R1,CLC,OP1(0),OP2
+INLINEX0338 CLC OP1(0),OP2
+INLINEY0338 EX R1,INLINEX0338
>From what we could tell, the CLC is still warm when EX hits it.
Slower alternative was a computed branch to do the 9 cases. With our
usage mix, the extra test to use a plain CLC over 8 byte seems to pay
off.
