All

 

David Wilkinson has submitted a new solution for #11 with fewer instructions.  See the contest webpage for details:

 

http://www.z390.org/z390_Mainframe_Assemble_Coding_Contest.htm

 

Don Higgins

don@...

don-higgins.net

727-545-4127

 

All

 

I am pleased to announce new #1 solution for problem #5 submitted by David Wilkinson.  Here is a way to translate any number of binary bytes to EBCDIC hex using truncated 2 byte table for TROT.

 

http://z390.sourceforge.net/z390_Mainframe_Assemble_Coding_Contest.htm#What_is_New

 

Sorry for the delay in getting this posted as I got engrossed with ZSTRMAC and getting v1.4.03 out the door today.

 

Don Higgins

don@...

don-higgins.net

727-545-4127

 

All

 

The following 3 new solutions and 1 new problem have been contributed: 

• 08/11/08
• P17DW1.MLC/LOG by David Wilkinson compress and decompress using TRT and CRB for total of 628 instructions (1st place)
• Problem #21 binary search submitted by David Wilkinson.
• 08/08/08
• P3DW1.MLC/LOG by David Wilkinson - unpack and TR using 16 byte table for 4 byte parm
• P4DW1.MLC/LOG by David Wilkinson - improved Quicksort using 1057 instructions (2nd place)

For more information visit:

http://z390.sourceforge.net/z390_Mainframe_Assemble_Coding_Contest.htm

 

 

Don Higgins

mailto:don@...

http://don-higgins.net

Tel. 727-545-4127

 

All

 

• P7EH1.MLC/LOG by John Erhman - has been updated to remove work-around for AW since the latest z390 PTF v1.4.01f now has support for AW and all the HFP unnormalized floating point instructions.
• An email quiz question was posted about the most efficient way to test if the left most bit in any mask is on.  The solution is to shift the mask 1 bit right and compare it to the selected bits AND'd with mask.  If the selected bits are high then the high bit must be on.

For more information on contest, visit:

http://z390.sourceforge.net/z390_Mainframe_Assemble_Coding_Contest.htm

Don Higgins

mailto:don@...

http://don.higgins.net

Tel. 727-545-4127

 

All

 

It’s been a while, but the contest will resume once z390 priority 1 compatibility RPI’s are fixed hopefully in the next day or two with RPI v1.4.01f.

 

I’m working on z390 RPI 844 to fix a problem with TMHH, TMHL, TMLH, and TMLL where z390 was not compatible with z9 or z10 running the z390 regression test TESTINS2.MLC.  The problem was with setting CC1 or CC2 when the selected bits were mixed.  The POP says set CC1 if the left most selected bit is 1 else set CC2 when selected bits are mixed.  The quiz question is simply given register bits in  half word REG and mask bits in half work MASK, what is the most efficient way to determine if high selected bit is on or not.

 

Don Higgins

mailto:don@...

http://don.higgins.net

Tel. 727-545-4127

 

All

 

Congratulations to Werner Rams for coding and testing published random number generator which ran for 5 hours on his PC without any duplications.  To see the source visit the site here:

 

http://z390.sourceforge.net/z390_Mainframe_Assemble_Coding_Contest.htm

 

As Martin Ward pointed out, the USA National Institute of Standards and Technology has various statistical tests for random number generators for those interested:

 

http://csrc.nist.gov/groups/ST/toolkit/rng/index.html

 

I’m too busy with development of a new z390 macro assembler based tool at the moment to think of a good problem #21.  I hope to catch up in a few weeks.  Suggestions welcome.

 

As with all the problems, it has helped me to find and fix another bug in the z390 emulator.  The time limit default of 15 seconds and the override limit with TIME(sec) was not working for the emulator.  It will be fixed in the next release.  The project I’m working on with lots of macros has also uncovered an obscure bug or too.  Since HLASM compatibility and bug fixing take first priority, some other RPI’s have been pushed back a bit.  See updated list of pending RPI’s here:

 

http://z390.sourceforge.net/z390_Support_Request_Log.htm

 

Don Higgins

mailto:don@...

http://don.higgins.net

www.z390.org

 

All

 

Congratulations to Werner Rams for solution to #19 using TRTR.  Honorable mention also goes to Steve R.K. for sending email suggesting TRTR earlier the same day as when Werner’s complete solution arrived.  Visit the contest website to view code and output:

 

http://z390.sourceforge.net/z390_Mainframe_Assemble_Coding_Contest.htm

 

A new problem #20 has been posted to write integer random number generator and test program to determine the longest sequence without duplication that it produces for a given seed number.  The longest sequence of non-repeating pseudo-random numbers wins.

 

Don Higgins

mailto:don@...

http://don.higgins.net

 

All

 

Following the release of z390 v1.4.01a with new date and time callable routine named DAT that can be used for benchmark testing, I’ve published a first solution to problem #18.  The solution show that the MIP rate dropped 15% when using the new z10 compare and branch code, but the elapsed time in nano-seconds also reduced by 8%.  Visit the web site to see source code including copy of the new DAT routine and the execution log showing results:

 

http://z390.sourceforge.net/z390_Mainframe_Assemble_Coding_Contest.htm

 

I’ve also posted a new problem #18 to write code to find the last non-blank character in an 80 byte text line using the least number of instructions.  Hint there is a z10 instruction that might be the fastest, but I could also be wrong.

 

Enjoy the weekend!

 

Don Higgins

mailto:don@...

http://don.higgins.net

 

All

 

Correction on the number of instructions required for solution to problem 17 - it is 827 and not 30,619.  I had mistakenly published the instructions per second which appears at end of log 672instead of the total which appears on the end of step message displayed on console log, err, and optional sta files.  Note that the instructions per second number is also inaccurate for programs running less than a second due to startup overhead.

 

I’m really glad Werner Rams corrected my error, as I was discussing problems 17 and 18 with Melvyn Maltz which he was visiting here on Friday.  I was commenting that the initial number of instructions seemed too high and it was!

Melvyn is taking a tour of the US following our presentation at SHARE.  Checkout new picture here:

 

http://z390.sourceforge.net/project/index.html

 

Regarding problem 18, I’m currently working on adding the last 30 new z10 instructions to assembler and the new rt\test\TESTINS3.MLC regression test of each new instruction and hope to have z390 v1.4.01 tested and published by Friday 03/14/08.  Note I’ve also installed J2SE 1.6.0_05 update and will be testing it this week.

 

Don Higgins

mailto:don@...

http://don.higgins.net

 

All

 

Congratulations to Werner Rams for the first solution to problem #17.  His solution compresses and decompresses 3 records using CLCL to detect end of duplicate character strings.  This solution executed 30,629 instructions.  To view the source code and the generated log file running on z390 visit:

 

http://z390.sourceforge.net/z390_Mainframe_Assemble_Coding_Contest.htm

 

I’ve added a new problem # 18 to measure the performance gain from using new IBM z10 opcode CIJNE:

Write a benchmark program to calculate the percent  performance improvement to 2 decimal places when replacing the following loop code:

LOOP DS 0H
            BCTR      R1,0
*** APPLICATION CODE COMMENTED OUT FOR TEST ***
            LTR         R1,R1
            JNE         LOOP

with the following optimized loop code using the new z10 compare immediate and branch relative opcode code:

LOOP DS 0H
            BCTR    R1,0
*** APPLICATION CODE COMMENTED OUT FOR TEST ***
            CIJNE  R1,0,LOOP

The performance improvement in this case comes from replacing 2 instruction cycles fetching a total of 6 bytes with a single instruction cycle fetching 6 bytes.  You can use whatever interval timing method is available on your system such as TIME BIN (requires running standalone).  The initial values in R1 must be set to perform enough iterations to reduce the timing error due to interval timer precision etc.  To code and unit test solution on z390 you will need version v1.4.01 with the new z10 opcodes scheduled for release by 03/14/08.  To run the real test, you will need an IBM z10 mainframe and updated HLASM.  My own initial test on pre-release version of z390 v1.4.01 indicates about a 15% improvement but this is measuring J2SE VM emulation overhead of each instruction cycle on Intel Dual-Core chip versus measuring the IBM z10 hardware/Millicode.  I'll be very interested to hear the real results.

 

Don Higgins

mailto:don@...

http://don.higgins.net

 

Congratulations to Werner Rams for a better solution to primes from 3 to 97 using bit table!

 

http://z390.sourceforge.net/z390_Mainframe_Assemble_Coding_Contest.htm

 

Don Higgins

mailto:don@...

http://don.higgins.net

 

John, all

 

My intention in using the work “transparent” was to indicate that all 256 EBCDIC characters are allowed including any kind of operand characters and comments plus continuation character and sequence characters up to 80 byte record length.  So yes the compression and decompression routines will need to handle any special byte codes used in compression which can also appear as valid byte codes in the original source.  This is the sort of compression and decompression routines required to handle any sort of data like ZIP does.

 

 

 

Can we have some clarification on Problem 17 in as far as any constraints on the operand field.

 

For example, compressing:

 

L1    CLI    F1,64

 

Is much easier than compressing:

 

L1    CLI    F1,C’ ‘

 

If quoted operands are to be allowed, the recognition and compression algorithms become much more difficult.

 

Also, can we assume that neither comments or sequence numbers (73-80) are to be permitted?

 

John P. Baker

All

 

Visit the ZMFACC contest web site for links to source code and output logs plus new problems:

 

http://z390.sourceforge.net/z390_Mainframe_Assemble_Coding_Contest.htm

 

1.       Congratulations to Werner Rams for first solution to problem 15 requiring rounding of DP result!  Werner used SRP to solve the problem. 

2.      I have also posted a second solution for problem 15 which uses AP and CP to accomplish the rounding.

3.      A new problem 17 has been posted requiring transparent compression and decompression of spaces.

4.      I’ve posted 2 additional solutions for problem 12 calculation of standard deviation.  Now there are 3 solutions; one using BFP, one using HFP, and one using just DFP instructions.  The DFP solution proved to be more challenging than expected as the latest POP (at least for today) is missing the opcode SQXTR corresponding to the existing SQXBR and SQXR used in the other solutions.  As a result I have coded a substitute SQXTR macro which calls standalone SQXTR CSECT which uses DFP instructions to perform the square root function using Newton Raphson method.  For the current problem it converged to 34 digit solution in 7 iterations.  Suggestions for improving the first guess will help.  Note all 3 solutions produce the same result to 33+ decimal places.

5.      I posted a first solution to problem 16 using a SETBIT macro and a TESTBIT macro to access bit table used to record prime numbers from 3 to 97 and then retrieve and print them.

 

All feedback welcome!.

 

Don Higgins

mailto:don@...

http://don.higgins.net

 

All

 

Today a friend living in Finland sent me this fascinating reference on hash tables by Bob Jenkins:

 

http://burtleburtle.net/bob/hash/doobs.html

 

It demonstrates to me that just when you think you know something about a subject, you find out you don’t know much at all.  For links to other interesting math subjects also visit Bob’s website (Thanks Bob):

 

http://burtleburtle.net/bob/

 

Now that I’ve gotten past VSAM insert support using AVL trees, and I don’t want to start anything too big before next week’s SHARE conference, I plan to submit solutions to the ZMFACC problems with no solutions yet, and post the next problem on best transparent compression/decompression routines.

 

Don Higgins

mailto:don@...

http://don.higgins.net

 

All 

 

Congratulations to Werner Hams for submitting new #1 solution to problem #11 hash table lookup using linked list for duplicates.  This solution uses a few more instructions but uses significantly less memory for tables (approximately 12 * the number entries in table):

 

http://z390.sourceforge.net/z390_Mainframe_Assemble_Coding_Contest.htm

 

Don Higgins

mailto:don@...

http://don.higgins.net

 

All

 

I’ve posted correction to comments on Martin Ward’s solution to #14.  It supports up to 31 digit solutions.

 

I’ve also posted new current problem category index.  Thanks to those contributing to the categories.

 

http://z390.sourceforge.net/z390_Mainframe_Assemble_Coding_Contest.htm#What_is_New

 

Don Higgins

mailto:don@...

http://don.higgins.net

 

All

 

Congratulations to Martin Ward for a very impressive solution to problem #14 using packed decimal to handle higher value Ackerman functions with up to 15 digit solutions.   I also posted my recursive macro code only solution which only handles up to 32 bit integer solutions.

 

A new problem #16 requiring routines to store and fetch bits to/from bit array representing numbers submitted by Jim Connelley has been posted.  Thanks Jim.   For the old timers this may remind you of how cool it was when HASP (Alias JES2) came out to speed up OS/PCP/MFT/MVT spooling using a bit array in memory to handle allocation of spool tracks as opposed to updating every sysout file in VTOC.

 

Don Higgins

mailto:don@...

http://don.higgins.net

 

All

 

Problem #15 has been posted.  Neither #14 or #15 have any solutions posted yet.  Visit website for details:

 

http://z390.sourceforge.net/z390_Mainframe_Assemble_Coding_Contest.htm

 

A friend asked me about the potential application of techniques used in the contest problems.  In response here is an initial list of applications by category with my first thoughts:

 

1.      Swapping fields (#1,#2) – used in sorting, file buffering

2.      Conversion to display characters (#3,#6,#9,#10) – used in dump, error display, or report formatting

3.      Sorting records (#4) – used in preparing for sequential processing and reporting

4.      Conversion of display characters to binary (#5) – used in decoding input data for processing

5.      Floating point calculations (#7,#8,#12) – used to calculate statistics with required precision

6.      Table lookup (#11) – used to access data tables required for processes such as validating records

7.      Decimal calculations (#13,#15) – used to calculate currency precisely in base 10

8.      Recursive functions (#14) – used in sorting routines such as Quicksort, compiling languages

 

Going forward, I’d like to build a list of categories for future problems.  My initial list of possible additions include:

 

1.      Boolean logic

2.      Branch logic

3.      Comparisons

4.      Compression and de-compression (someone just asked me if z390 supports CMPSC yet.  The answer was not yet but I did go read the 9 page description in the latest POP which includes 3 page diagram)

5.      Encryption and decryption

6.      File access methods

7.      Heuristics

8.      Totally useless just for fun

 

Do you have some categories or specific questions you would like to see added.  All feedback welcome.

 

Don Higgins

mailto:don@...

http://don.higgins.net

 

All

 

ZMFACC Assembler Coding Contest - A new problem #14 - code the Ackerman function to calc a(4,1)=65533.

 

Code a macro assembler program to calculate the value of the Ackerman function a(4,1) = 65533.  The Ackerman function a(m,n) is  a recursively defined function as follows:

1.      If m = 0, then a(m, n) = n+1

2.      if m > 0 and n = 0, then a(m, n) = a(m-1,1)

3.      if m> 0 and n > 0, then a(m,n) = a(m-1,a(m,n-1))

For more information on the Ackerman function visit:

http://en.wikipedia.org/wiki/Ackermann_function

http://z390.sourceforge.net/z390_Mainframe_Assemble_Coding_Contest.htm

 

Don Higgins

mailto:don@...

http://don.higgins.net

 

All

 

I’ve updated the problem statement for #13 per posting on Mainframe Assembler List indicating it was incorrect.  I’ve also posted a new solution to #13 using 7 DFP instructions to perform the required division and rounding to 2 decimal places.  Calculating the number of significant digits to request for rounding to 2 decimal places was not exactly trivial since the division produced an irrational fraction with repeating sequence of 270270… which resulted in 16 significant digit result with corresponding exponent set to required position versus preferred position.  Visit the web site to see source and log listings:

 

http://z390.sourceforge.net/z390_Mainframe_Assemble_Coding_Contest.htm

 

Note running this solution on z390 requires update to PTF level v1400a which can now be downloaded in InstallShield format for easy installation from www.z390.org.  This PTF fixes trap in RRDTR and RRXTR under some conditions and includes trailing zeros as significant digits in ESDTR and ESXTR instructions.  Also the traces for several DFP instructions have been corrected to show correct GPR versus FPR registers.

 

Although figuring out how to do this in DFP was fun, I don’t recommend changing existing PD routines like Steve’s P13SC1.MLC unless there is a well defined business application requirement such as the need for an industry standard method that can be used on different hardware and software platforms producing exactly the same result.

 

Don Higgins

mailto:don@...

http://don.higgins.net

 

All

 

Congratulations to Steve Comstock with http://www.trainersfriend.com/  for posting first solution to problem #13 using 5 packed decimal instructions.  See the source and generated output here:

 

http://z390.sourceforge.net/z390_Mainframe_Assemble_Coding_Contest.htm

 

Some follow-up questions and comments after reading all the posts today regarding the ZMFACC contest on the http://www.listserv.uga.edu/archives/asm370.html:

 

1.       In response to the question about the purpose of the contest and the trade-off between fewest instructions, fastest execution, least memory, and ease of maintenance, I have updated the web site to say the primary goals are to have some fun and learn more about mainframe assembler.  Hopefully in the process folks will also share what they think makes for maintainable assembler code.  Certainly I agree that when writing assembler code for a project where the code will live on indefinitely and probably be maintained by someone else, the first priority may well be making the code as understandable and portable as possible including such things as defining the derivation and limits of any constants, avoiding tricks which may be machine dependant and may make future upgrades and/or porting to other languages much more difficult.

2.      In response to the statement that floating point should not be used in the solution to this problem, I have a question.  What method of rounding is “correct” for this problem?  In Steve’s packed decimal solution he only used 1 extra decimal place to determine rounding.  Using the new packed decimal instructions, it is possible to use any number of “extra” digits to perform rounding up to 34 digits using the 128 bit extended format.   If additional digits are included and there is a carry into the 2 decimal digits required, the result is different.  Perhaps one of the most common methods for performing this function is using a COBOL statement like DIVIDE TOTAL-PRICE BY QUANTITY GIVING UNIT-PRICE ROUNDED.  What kind of code does this generate and does the COBOL standard say exactly how many extra digits or bits will be used in performing the rounding.  I can remember in my distant past, having issues with this sort of thing when simply upgrading from one COBOL compiler version to another on the same machine let along switching between platforms with different hardware and software.  If I am reading this document correctly it would appear that the COBOL “Standard” calls for using 128 bit decimal floating point to handle ROUNDED in a way that is portable across machines and compilers: http://www.cobolstandard.info/j4/files/05-0063.doc

 

Don Higgins

mailto:don@...

http://don.higgins.net

 

I’ve posted a second solution to the hash table problem which allows up to 2 duplicate hash key searches per entry resulting in a reduction in the hash table size from 48011 down to 3473 at the cost of a 10% increase in instructions.

This solution executes 7 instructions if the first entry is the key, 12 instructions if the second entry is key, and 17 instructions if the third entry is the key.

 

To see the source and execution log listings for this and other solutions visit the ZMFACC site here:

 

http://z390.sourceforge.net/z390_Mainframe_Assemble_Coding_Contest.htm

 

I’m still looking for a better solution to this problem that reduces both size and instructions.  Any and all comments welcome.

 

Don Higgins

mailto:don@...

http://don.higgins.net

 

Given a decimal number with 2 decimal places representing the total cost of an item and another decimal number representing the quantify, calculate the unit price with 2 decimal places rounded half up?  Problem was derived from question posted on IBM Mainframe Assembler-List by Ludmila Koganer.

 

I think the most straight forward solution using the least instructions may be done using the new Decimal Floating Point (DFP) instructions.  However the most efficient may be done using basic packed decimal instructions and no floating point instructions.  For more information on ZMFACC problems and solutions visit the web site here:

 

http://z390.sourceforge.net/z390_Mainframe_Assemble_Coding_Contest.htm

 

Don Higgins

mailto:don@...

http://don.higgins.net

 

All

 

I’ve posted a solution to problem #11 which asks for the fastest hash routine to retrieve address of table entry given 8 character key for a table of known opcodes:

 

http://z390.sourceforge.net/z390_Mainframe_Assemble_Coding_Contest.htm

 

The solution was developed using another program P11FIND.MLC which I’ve also posted which searches for minimum hash table with no duplicate keys for the given primary table using the given hash routine which is a 64 bit divide and then using positive remainder as index.  The divisor it found was 48011 resulting in a 172k+ hash table which will return address of primary table entry using 6 instructions including the BR return.

 

Surely there are faster routines using less storage and I’d like to see a few.

 

Don Higgins

mailto:don@...

http://don.higgins.net

 

All

 

This was posted on Hercules email group today by Jay Maynard, and I thought I’d pass it on for those interested in using it to extend older assemblers to include many of the z/Architecture instructions which the z390 already supports:. 

 

Stephen Powell's macro library for z/Architecture instructions for use with z/VM ASMFX and other older assemblers:

 

http://www.jaymoseley.com/hercules/compiling/spopcodes.htm

 

With these macro extensions you may be able to use MVS 3.8 IFX00, z/VM Assembler XF, ASMG etc. to run z390 Mainframe Assembler Coding Contest solutions that depend on some of the newer instructions for 64 bit GPR’s, 16 64 bit FPR’s, extended translate instructions such as TROT, etc.  I don’t see the decimal floating point instructions yet but perhaps they are coming too.

 

Don Higgins

mailto:don@...

http://don.higgins.net

 

All

 

Two new solutions have been posted

 

1.      P10MB1.MLC/LOG by Mats Broberg at SEB using fewer instr. and single ED (Note execution of this solution on z390 requires latest z390 v1.3.08h PTF to fix overflow bug)

2.      P12DSH1.MLC/LOG by Don Higgins -using BFP  to calc standard. deviation for (1, 2, 3, 6) = 1,87

 

There is still no solution posted yet for the hash table problem #12.  I will post a solution soon if none is forthcoming.  How about a new problem that can best be solved using the new decimal floating point?  After spending a good deal of time on DFP recently, one quiz problem that comes to mind is how many different ways can you represent the exact value of 1 using DFP short, double, or extended formats?  The answer may surprise you/   

 

Don Higgins

mailto:don@...

http://don.higgins.net