The following message is a courtesy copy of an article that has been posted to bit.listserv.ibm-main as well.
[EMAIL PROTECTED] writes: > The Prin of Operations, programming notes on using the TR with DAT on, > state that there will be a performance hit if the second operand > actually crosses the 4096 line. This is because it will do a 'mock' > execution first. > > Assuming DAT on, is the performance hit related to the possibility > that the following 4096 page is not in virtual memory? way back on 360/67 ... (actually all 360s) TR used to test start & start+255 (end) address of the table ... which met that if it crossed a 4k page ... it would catch both ... aka page fault both pages ... before starting instruction execution. somewhere along the way ... something was raised that TR only uses that much of the table that the input data-stream might used ... for instance, if the translation input stream only had values 0-9 ... and the table was within 256 bytes of the end of an addressable region ... then the instruction might fail (with start+256 precheck) ... even tho it otherwise could succesfully execute. so the TR instruction was "fixed" ... if the table start is within 256 bytes of the end of an addressable boundary... it "pre-executes" the instruction to see if any input stream bytes would index the table across the boundary. this would also theoritically have been a problem with 2k key fetch protect ... and the table was within 256 bytes of a 2k boundary (with the next 2k, fetch protected) and the input data stream never indexed anything (in the table) across the addressable boundary. a past thread that also got into this subject: http://www.garlic.com/~lynn/2005j.html#36 A second look at memory access alignment http://www.garlic.com/~lynn/2005j.html#37 A second look at memory access alignment http://www.garlic.com/~lynn/2005j.html#39 A second look at memory access alignment http://www.garlic.com/~lynn/2005j.html#40 A second look at memory access alignment http://www.garlic.com/~lynn/2005j.html#43 A second look at memory access alignment http://www.garlic.com/~lynn/2005j.html#44 A second look at memory access alignment ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html