Now that you mention it, I remember that the C/C++ compiler has a architecture option to control the instructions generated. I should have known that the PL/X compiler would too. I didn't know that they both share the same back-end. I wish that the COBOL compiler did. I am constantly amazed at the amount of code generate by a simpe:
ADD +1 TO WS-INTEGER. when WS-INTEGER is defined as PIC S9(9) BINARY or NATIVE. COBOL seems to have an inordinate love for PACKED-DECIMAL. Someone once said it was due to ANSI standards compliance. Might be worth it, in CPU saved, to license the C compiler and port the OpenCOBOL Unless I somehow have the wrong compile parameters. -- John McKown Systems Engineer IV IT Administrative Services Group HealthMarkets(r) 9151 Boulevard 26 * N. Richland Hills * TX 76010 (817) 255-3225 phone * [email protected] * www.HealthMarkets.com Confidentiality Notice: This e-mail message may contain confidential or proprietary information. If you are not the intended recipient, please contact the sender by reply e-mail and destroy all copies of the original message. HealthMarkets(r) is the brand name for products underwritten and issued by the insurance subsidiaries of HealthMarkets, Inc. -The Chesapeake Life Insurance Company(r), Mid-West National Life Insurance Company of TennesseeSM and The MEGA Life and Health Insurance Company.SM > -----Original Message----- > From: IBM Mainframe Discussion List > [mailto:[email protected]] On Behalf Of David Crayford > Sent: Thursday, April 12, 2012 11:04 AM > To: [email protected] > Subject: Re: Modernizing the BCP code ? > > AFAIK, the PL/X compiler shares a back-end with the other code > optimizers, so should produce excellent code. The compiler team is in > Toronto and I was lucky enough to talk to a few at Share in Orlando. > As it stands z/OS needs to run on machines as old as a z/800. So the > architecture level (ARCH(n)) of BCP code has to support old > hardware. As > Bob Shannon mentioned, with z/OS going to a new version > IBM can safely recompile the BCP PL/X code for not so ancient boxes. > > On 12/04/2012 11:14 PM, McKown, John wrote: > > First thing to wonder about is: "Has IBM upgraded the > PL/AS(current name?) compiler to emit Relative and Immediate > instructions?". Most of the BCP is supposedly written in > PL/AS, not HLASM. Second, for those modules which are written > in HLASM, is it cost effective to IBM to systematically > update the source from the older instructions? Why would it > be? The new instructions may (or may not) be less resource > efficient that the older versions.I doubt that many are more > efficient (LHI being more efficient seems likely). Many of > the R&I instructions are longer than the corresponding > Base+Displacement instruction. The simpliest thing for IBM > would be to use the IEABRCX macro to change all the branch > instructions to branch relative instructions. But you can't > do that type of OPSYN method to do something like replacing > the LA with an LARL. So replacing LA with LARL would require > looking at every LA instruction to see if it could be > replaced by an LARL. The same with replacing selec! > te! > > d L and LH with LHI, which would verifying that the value > being loaded is indeed never modified. And I don't see a > general way to do it if they are inside a macro. And an LARL > may not be the best idea simply because it is 6 bytes instead > of the LA instruction's 4 bytes. So, at the very least, it > will take up more space in the I-cache hardware. Other > immediate instructions may also require more bytes that the > replaced instruction. So, IMO, a command such as "Replace all > base+displacement instructions in a program with the > corresponding R&I instruction, whenever possible." is not a good idea. > > > > IMO, the only time to use a relative instruction is in > "baseless" code. Only use "baseless" coding when you are > either out of registers or you need too many base registers > due to the size of the program. Which, in HLASM programs, > usually means the program is too big and should be split up somehow. > > > > Again, the only time to use immediate instructions is when > the value is a constant and will fit into a halfword. Why a > halfword? Because the LHI is the same number of bytes as the > LH or L instruction, taking up the same I-cache space. And > because the immediate instructions to load a fullword are > larger than the base+displacement load instruction, and so > use more I-cache space. Space which may be better spend > buffering more instructions and let the D-cache hold the > value instead of the I-cache holding the equivalent of the > instruction and data, and not using the D-cache. This is > especially true if there are two or more immediate > instructions in the I-cache that are loading the same value. > In this case, assuming the duplicate instruction is something > like "L r,=F'10'", the F'10' would be held in the D-cache and > so be likely be used almost as fast, if not equally fast, as > it would be by being encoded in the instruction in the I-cache. > > > > However, I will freely admit that I use R&I instructions in > preference to B+D instructions in most of my HLASM programs. > > > > Oh, and one other reason to use at least one Branch > instruction in your program instead of a Brance Relative. If > the start of a program starts with the sequence: > > > > USING *,15 > > B AROUND > > DC C'some character string' > > AROUND DS 0H > > > > Then the SYSUDUMP will print "some character string" in the > save area trace portion of the dump. In most cases, this > works because GPR15 points to the B instruction. > > > > -- > > John McKown > > Systems Engineer IV > > IT > > > > Administrative Services Group > > > > HealthMarkets(r) > > > > 9151 Boulevard 26 * N. Richland Hills * TX 76010 > > (817) 255-3225 phone * > > [email protected] * www.HealthMarkets.com > > > > Confidentiality Notice: This e-mail message may contain > confidential or proprietary information. If you are not the > intended recipient, please contact the sender by reply e-mail > and destroy all copies of the original message. > HealthMarkets(r) is the brand name for products underwritten > and issued by the insurance subsidiaries of HealthMarkets, > Inc. -The Chesapeake Life Insurance Company(r), Mid-West > National Life Insurance Company of TennesseeSM and The MEGA > Life and Health Insurance Company.SM > > > > > > > >> -----Original Message----- > >> From: IBM Mainframe Discussion List > >> [mailto:[email protected]] On Behalf Of Miklos Szigetvari > >> Sent: Thursday, April 12, 2012 9:16 AM > >> To: [email protected] > >> Subject: Modernizing the BCP code ? > >> > >> Hi > >> > >> We try to modernize our code here, with relative > instructions, long > >> displacements , immediate s etc etc > >> What about the control program ? > >> Just got some REXX IRXINIT dumps, and seems to me the code > >> is not very > >> modern. > >> > >> > ---------------------------------------------------------------------- > >> For IBM-MAIN subscribe / signoff / archive access instructions, > >> send email to [email protected] with the message: INFO IBM-MAIN > >> > >> > > > ---------------------------------------------------------------------- > > For IBM-MAIN subscribe / signoff / archive access instructions, > > send email to [email protected] with the message: INFO IBM-MAIN > > ---------------------------------------------------------------------- > For IBM-MAIN subscribe / signoff / archive access instructions, > send email to [email protected] with the message: INFO IBM-MAIN > > ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO IBM-MAIN
