fortran patch

[Dave Love]

This patch should be included to fix a significant Fortran bug.
Regression test below.

-------------- next part --------------
Date: Fri, 10 Oct 1997 16:31:42 -0400 (EDT)
Message-Id: <199710102031.QAA27219@tarpit.Cadence.COM>
From: Craig Burley <cburley@cadence.com>
To: fortran@gnu.org, masik@nc25.troja.mff.cuni.cz
Subject: g77 patch to fix DO bug -- g77-971010.diff.gz

Here's the patch I *would* just upload to ~fortran/dev/ normally, but
which I can't due to the FSF being (mostly) offline.  Don't worry
about it if you don't have the other two or three patches; only the
documentation patchlet enclosed will be inappropriate in that case.

        tq vm, (burley)


*** g77-new/f/ChangeLog.~1~	Mon Sep 29 16:21:27 1997
--- g77-new/f/ChangeLog	Fri Oct 10 13:07:18 1997
***************
*** 1,2 ****
--- 1,12 ----
+ Fri Oct 10 13:00:48 1997  Craig Burley  <burley@gnu.ai.mit.edu>
+ 
+ 	* ste.c (ffeste_begin_iterdo_): Fix loop setup so iteration
+ 	variable is modified only after the #iterations is calculated;
+ 	otherwise if the iteration variable is aliased to any of the
+ 	operands in the start, end, or increment expressions, the
+ 	wrong #iterations might be calculated.
+ 
+ 	* com.c (ffecom_save_tree): Fix indentation.
+ 
  Mon Sep 29 16:18:21 1997  Craig Burley  <burley@gnu.ai.mit.edu>
  
*** g77-new/f/com.c.~1~	Tue Sep  2 17:25:26 1997
--- g77-new/f/com.c	Fri Oct 10 13:00:43 1997
*************** ffecom_return_expr (ffebld expr)
*** 13269,13273 ****
  
  #if FFECOM_targetCURRENT == FFECOM_targetGCC
! tree ffecom_save_tree (tree t)
  {
    return save_expr (t);
--- 13269,13274 ----
  
  #if FFECOM_targetCURRENT == FFECOM_targetGCC
! tree
! ffecom_save_tree (tree t)
  {
    return save_expr (t);
*** g77-new/f/news.texi.~1~	Tue Sep  9 02:11:37 1997
--- g77-new/f/news.texi	Fri Oct 10 15:12:13 1997
***************
*** 6,10 ****
  @c in the G77 distribution, as well as in the G77 manual.
  
! @c 1997-09-09
  
  @ifclear NEWSONLY
--- 6,10 ----
  @c in the G77 distribution, as well as in the G77 manual.
  
! @c 1997-10-10
  
  @ifclear NEWSONLY
*************** diagnostics, internal improvements, and 
*** 24,27 ****
--- 24,68 ----
  This order is not strict---for example, some items
  involve a combination of these elements.
+ 
+ @heading In 0.5.22:
+ @itemize @bullet
+ @item
+ Fix code generation for iterative @code{DO} loops that
+ have one or more references to the iteration variable,
+ or to aliases of it, in their control expressions.
+ For example, @samp{DO 10 J=2,J} now is compiled correctly.
+ 
+ @item
+ Fix a compiler crash involving optimization of
+ certain complicated expressions involving @code{COMPLEX}
+ arithmetic (especially multiplication).
+ 
+ @item
+ Rename the @code{gcc} keyword @code{restrict} to
+ @code{__restrict__}, to avoid rejecting valid, existing,
+ C programs.
+ Support for @code{restrict} is now more like support
+ for @code{complex}.
+ 
+ @item
+ Fix @code{DTime} intrinsic so as not to truncate
+ results to integer values (on some systems).
+ 
+ @item
+ Improve diagnostic messages from @code{libf2c}
+ so it is more likely that the printing of the
+ active format string is limited to the string,
+ with no trailing garbage being printed.
+ 
+ (Unlike @code{f2c}, @code{g77} does not append
+ a null byte to its compiled form of every
+ format string specified via a @code{FORMAT} statement.
+ However, @code{f2c} would exhibit the problem
+ anyway for a statement like @samp{PRINT '(I)garbage', 1}
+ by printing @samp{(I)garbage} as the format string.)
+ 
+ @item
+ Upgrade to @code{libf2c} as of 1997-09-23.
+ @end itemize
  
  @heading In 0.5.21:
*** g77-new/f/ste.c.~1~	Mon Aug 11 14:53:40 1997
--- g77-new/f/ste.c	Fri Oct 10 13:06:03 1997
*************** ffeste_begin_iterdo_ (ffestw block, tree
*** 396,405 ****
    /* Do the initial assignment into the DO var.  */
  
!   expr = ffecom_modify (void_type_node, tvar, tstart);
!   expand_expr_stmt (expr);
  
    expr = ffecom_2 (MINUS_EXPR, TREE_TYPE (tvar),
  		   tend,
! 		   TREE_CONSTANT (tstart) ? tstart : tvar);
  
    if (!ffe_is_onetrip ())
--- 396,404 ----
    /* Do the initial assignment into the DO var.  */
  
!   tstart = ffecom_save_tree (tstart);
  
    expr = ffecom_2 (MINUS_EXPR, TREE_TYPE (tvar),
  		   tend,
! 		   tstart);
  
    if (!ffe_is_onetrip ())
*************** ffeste_begin_iterdo_ (ffestw block, tree
*** 438,441 ****
--- 437,443 ----
  				FFETARGET_charactersizeNONE, -1, FALSE);
    expr = ffecom_modify (void_type_node, niters, expr);
+   expand_expr_stmt (expr);
+ 
+   expr = ffecom_modify (void_type_node, tvar, tstart);
    expand_expr_stmt (expr);
  


-------------- next part --------------
* Fri Oct 10 13:00:48 1997  Craig Burley  <burley@gnu.ai.mit.edu>
* 
* 	* ste.c (ffeste_begin_iterdo_): Fix loop setup so iteration
* 	variable is modified only after the #iterations is calculated;
* 	otherwise if the iteration variable is aliased to any of the
* 	operands in the start, end, or increment expressions, the
      i=3
      j=0
      do i=i,5
        j = j+i
      end do
      do i=3,i
        j = j+i
      end do
      if (i.ne.7) call abort()
      print *, i,j
      end