[Prev][Next][Index]

Sage Problems: U of Michigan: README


Professor Gannon,

	I enjoyed speaking with you at ICS94, and I appreciate your offer to
take a look at what we have been working on here at Michigan. I have included
the following files in a tar'd uuencoded file to follow.

	The top directory is kkinstru.gannon. It contains four directories
and several subfiles as follows:

// This directory contains the entire source instrumentation program. It runs
// with sage++-1.3.1 and I've been running it on a sun4.
	kkinstru:
		Makefile	kkinstru.C

// Currently empty, the source file 'kkinstru' from the directory kkinstru is
// copied here in the 'make'.
	bin:

// This file contains the instrumentation Makefile and four test files to be
// instrumented, and the 'post-instrumentation versions (prefixed by 'kki.').
// Note if you trying running this code, it will ask for parallel region
// numbers as you instrument the code. Use the numbers 1, 2, 3, 4 as you
// run the commands 'instru kki.main.F', 'instru kki.lu.F', 'instru
// kki.matmult.F', and 'instru kki.simple1.F'.
	usr:
		main.F		lu.F		matmult.F	simple1.F
		main.proj	lu.proj		matmult.proj	simple1.proj
		main.lst	lu.lst		matmult.lst	simple1.lst
		kki.main.F	kki.lu.F	kki.matmult.F	kki.simple1.F
		instru		Makefile.instru

// The main problem I have is trying to get the shadow variables to appear
// in the appropriate location upon instrumentation. They should appear 
// immediately after the declaration of the shadowee array. In the following
// directory I supply the 'desired' code. I also supply excerpts from the
// TWO applicable sections in the file kkinstru/kkinstru.C. Note that the
// 'main' procedure and the called subroutines are instrumented by different
// sections of code. Please note in the instrumentation of the 'main' routine,
// I managed to get the shadow arrays to appear ABOVE the shadowee. Putting
// them below introduces an indefinite recursion by the method I used. I've
// had no luck in moving the location of the instrumented shadow array in
// called subroutines.
	desired:
		des.kki.main.F	des.kki.lu.F	des.kki.matmult.F
		des.kki.simple1.F		instru.main.code.extract
		instru.calledsubroutine.code.extract

// Any help you could give would be appreciated. Please let me know if 
// anything is unclear.

//					Thanks again,

//					Eric L. Boyd
//					boyd@eecs.umich.edu

// Other minor problems. If you don't have time to address these, I understand.
// I am mentioning them because I suspect most are a problem with sage.

a) When I updated from sage v1.0.0 to v1.3.1, the original author's kkinstru.C
no longer worked. I had to comment out the following case in a switch statment
in kkinstru.C.

#ifdef I_THINK_ITS_OLD_SAGE_STUFF
        case PROC_COM:
#endif


b) When I compile, I get the following warning message. (This is also appeared
after porting to v1.3.1.) Is there something wrong?

bagabolts [ 104 ] src -: make
g++ -I/a/eiffel/n/eiffel/z/projects/sage/sage++-1.3.1/lib/include -I/a/eiffel/n/
eiffel/z/projects/sage/sage++-1.3.1/h -g  -c kkinstru.C
kkinstru.C: In function `void instrumentFile(class SgFile *)':
kkinstru.C:542: warning: implicit declaration of function `int exit(...)'
g++  kkinstru.o /a/eiffel/n/eiffel/z/projects/sage/sage++-1.3.1/lib/sun4/libSage
++.a /a/eiffel/n/eiffel/z/projects/sage/sage++-1.3.1/lib/sun4/libsage.a /a/eiffe
l/n/eiffel/z/projects/sage/sage++-1.3.1/lib/sun4/libdb.a /a/eiffel/n/eiffel/z/pr
ojects/sage/sage++-1.3.1/lib/sun4/libann.a -o kkinstru

c) Are there plans to get multiple files in a foo.proj file to work?

d) In the file lu.F, if line 15 is replaced from:
c       external iowner_proc
   to:
        external iowner_proc
   then line 96 in file kki.lu.F appears after instrumentation as:
       integer  function iowner_proc ()
   instead of:
       integer  function iowner_proc (i, j, pv, ph, nv, nh)

   Is this a sage bug, a bug in kkinstru.C, or a misinterpreation of Fortran
   language semantics?

e) I believe this problem is an inadequacy of kkinstru.C, but I am not sure
   how to fix it. In kki.matmult.F, line 33, as follows:
                c(i,j) = c(i,j) + a(i,k) * b(k,j)
   is instrumented as follows:
                c(i,j) = c(i,j) + a(i,k) * b(k,j)
                call SH$WRITE(3,ISH$pregion,ISH$_c((loc (c(i,j)) - loc (
     +c) + mod (loc (c),128)) / 128 + 1))
                call SH$READ(3,ISH$pregion,ISH$_c((loc (c(i,j)) - loc (c
     +) + mod (loc (c),128)) / 128 + 1))
                call SH$READ(1,ISH$pregion,ISH$_a((loc (a(i,k)) - loc (a
     +) + mod (loc (a),128)) / 128 + 1))
                call SH$READ(2,ISH$pregion,ISH$_b((loc (b(k,j)) - loc (b
     +) + mod (loc (b),128)) / 128 + 1))
   when it should be instrumented as follows:
                c(i,j) = c(i,j) + a(i,k) * b(k,j)
                call SH$READ(3,ISH$pregion,ISH$_c((loc (c(i,j)) - loc (c
     +) + mod (loc (c),128)) / 128 + 1))
                call SH$READ(1,ISH$pregion,ISH$_a((loc (a(i,k)) - loc (a
     +) + mod (loc (a),128)) / 128 + 1))
                call SH$READ(2,ISH$pregion,ISH$_b((loc (b(k,j)) - loc (b
     +) + mod (loc (b),128)) / 128 + 1))
                call SH$WRITE(3,ISH$pregion,ISH$_c((loc (c(i,j)) - loc (
     +c) + mod (loc (c),128)) / 128 + 1))

   Any ideas?
Follow-Up(s):