The interruptible instructions (usually) only get interrupted if the string is longer than 256 bytes, which is not usually the case. We just need a simple function for operating on short strings, very often. Some of the suggestions to allocate lots of extra pages were somewhat missing the point!
-----Original Message----- From: IBM Mainframe Discussion List <[email protected]> On Behalf Of David Crayford Sent: Friday, April 28, 2023 7:41 PM To: [email protected] Subject: Re: Inexplicable 0C4! On 28/4/23 19:57, Robin Atwood wrote: > I need to know the length of the string, preferably without needing a > new parameter and altering a lot of code. At which point I remembered > that the XL/C compiler generates inline code for the strlen(s) function. For > those who are interested, this is: Why don't you just use TRT like the C/C++ compiler does for it's inline string handling optimizations? In our experience those interruptible instructions may seem like syntactic sugar but they don't perform better. The only noticeable difference was the introduction of the SIMD instructions. > > LR R4,R2 copy string ptr > XR R0,R0 search for zero byte > SRST R0,R4 search > BO *-4 resume search > SLR R0,R2 get length > > I assume some magic causes the SRST not to "recognize an access > exception" because normally the first register > (R0 here) contains an address limiting the search. Well, that was an > interesting exercise! > > Thanks > > On Fri, Apr 28, 2023 at 6:33 PM Robin Atwood <[email protected]> wrote: > >> Thanks, Michael, that is the problem. The next page is available, but >> the one after that isn't. Obviously, I had read the entry for TRE in >> the PoOps but didn't take in the significance of the Access >> Exceptions clause. (In fact, I still don't understand it: >> "may or may not"!) >> >> The problem is I maintain legacy C code originally developed with >> SAS/C, which had a strupr(s) function. There is no length parameter, >> so I was at a loss what to set R1+1 to for the TRE instruction. >> I could search for the zero bytes using SRST but presumably that >> would hit the same problem going off the end of a page. >> Although, the SRST will wrap at top of storage - what does that imply? >> I could invent the strnupr(s,n) function but that will involve many >> code changes. ☹ >> >> @Ed Jaffe: thanks for the tip about the TEA, but where do you find it? >> IPCS STATUS FAILDATA displays it but, in my dumps, PSA+90 and PDS+A8 >> both have zero in the address part. >> >> Thanks to all who replied. >> >> -----Original Message----- >> From: IBM Mainframe Discussion List <[email protected]> On >> Behalf Of Michael Stein >> Sent: Thursday, April 27, 2023 10:20 PM >> To: [email protected] >> Subject: Re: Inexplicable 0C4! >> >> On Thu, Apr 27, 2023 at 07:58:23PM +0700, Robin Atwood wrote: >>> I have had two of these during the course of two months, so it's >>> getting serious! >>> >>> The abend happens in my implementation of a C strupr() function when >>> the TRE instruction is executed: >>> >>> TST TRE R2,R1 fold string >>> >>> R0 R1 R2 R3 PSW >>> >>> 00000000 0009C5EC 1AA748F8 7FFFFFFF 078D04008009BD14 >> from principles of operation TRE: >> >> Access exceptions for the portion of the first operand to the right of >> the last byte processed may or may not be recognized. For an operand >> longer than 4K bytes, access exceptions are not recognized for locations >> more than 4K bytes beyond the last byte processed. >> >> Access exceptions for all 256 bytes of the second operand may be >> recognized, even if not all bytes are used. >> >> The first operand is > 4K bytes, R2 is 1AA748F8, what's the status of >> the next page? >> >> R1 looks ok as the 2nd operand only needs to clear 256 bytes. >> > ---------------------------------------------------------------------- > 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
