On 2015-10-04 6:52 AM, Brent Longborough wrote:
Oops!
Clearly it's desirable to step to the next item each time round
the loop. Whether to use AHI or LA is probably not significant
in a modern pipelined engine, but probably has supporters in
both camps.
I would never use JXLE here (or, indeed, anywhere else) for a
number of complementary reasons ...
In your example R4 is loaded with LA. Ignoring subtle performance
differences I'd use AHI only if the incr > 4095 . If it were "L
R4,TBLADDR" I'd use LA for sure because the address could have the
VL-bit set !
For this type of loop I use R0, R1, R14 and R15. Although I still agree
about BXLE/BXH. I used to optimize the loop using BASR(/BALR) / BCTR:
LA R4,FUNCTBL Start of table
USING BTAB,R4
LHI R0,TBLENT Item count
BASR R14,0 *** LOOP ENTRY ***
CLC FUNC,BFUN This one?
JE FOUND_IT Yes - handle
LA R4,BNXT No - try next
BCTR R0,R14 *** Loop Until done ***
J ENDOFDAYS Not found
... and on today's processors the (1) extra BASR costs more than 'n' BCTRs
save. Likely the same also applies to the BXLE / BXH alternative.
Andreas Geissbuehler
