[PATCH] Fix PR36691, wrong-code with wrong # of iterations

[Richard Guenther]

On Mon, 4 Aug 2008, Zdenek Dvorak wrote:

> Hi,
> 
> > This fixes PR36691 where we wrongly skip generation of assumptions
> > for non-wrapping of the induction variable.
> > 
> > Bootstrapped and tested on x86_64-unknown-linux-gnu - Zdenek, does this
> > look ok?
> 
> yes, thanks for fixing this.  Although,
> 
> > *************** number_of_iterations_lt_to_ne (tree type
> > *** 692,703 ****
> >     mpz_set_double_int (mmod, tree_to_double_int (mod), true);
> >     mpz_neg (mmod, mmod);
> >   
> >     if (integer_nonzerop (iv0->step))
> >       {
> >         /* The final value of the iv is iv1->base + MOD, assuming that this
> >   	 computation does not overflow, and that
> >   	 iv0->base <= iv1->base + MOD.  */
> > !       if (!iv1->no_overflow && !integer_zerop (mod))
> >   	{
> >   	  bound = fold_build2 (MINUS_EXPR, type,
> >   			       TYPE_MAX_VALUE (type1), tmod);
> 
> it might be better to just fix the tests and check iv0->no_overflow here
> (since iv1->step == 0, it does not make sense to check whether it
> overflows),
> 
> > *************** number_of_iterations_lt_to_ne (tree type
> > *** 719,725 ****
> >         /* The final value of the iv is iv0->base - MOD, assuming that this
> >   	 computation does not overflow, and that
> >   	 iv0->base - MOD <= iv1->base. */
> > !       if (!iv0->no_overflow && !integer_zerop (mod))
> >   	{
> >   	  bound = fold_build2 (PLUS_EXPR, type1,
> >   			       TYPE_MIN_VALUE (type1), tmod);
> 
> and iv1->no_overflow here.

Ok.  I didn't realize one of them is always zero.

So the following is what I applied after re-testing.

Thanks,
Richard.

2008-08-04  Richard Guenther  <rguenther@suse.de>

	PR middle-end/36691
	* tree-ssa-loop-niter.c (number_of_iterations_lt_to_ne): Correctly
	check for no_overflow.

	* gcc.c-torture/execute/pr36691.c: New testcase.

Index: gcc/testsuite/gcc.c-torture/execute/pr36691.c
===================================================================
*** gcc/testsuite/gcc.c-torture/execute/pr36691.c	(revision 0)
--- gcc/testsuite/gcc.c-torture/execute/pr36691.c	(revision 0)
***************
*** 0 ****
--- 1,17 ----
+ unsigned char g_5;
+ 
+ void func_1 (void)
+ {
+   for (g_5 = 9; g_5 >= 4; g_5 -= 5)
+     ;
+ }
+ 
+ extern void abort (void);
+ int main (void)
+ {
+   func_1 ();
+   if (g_5 != 0)
+     abort ();
+   return 0;
+ }
+ 
Index: gcc/tree-ssa-loop-niter.c
===================================================================
*** gcc/tree-ssa-loop-niter.c	(revision 138620)
--- gcc/tree-ssa-loop-niter.c	(working copy)
*************** number_of_iterations_lt_to_ne (tree type
*** 697,703 ****
        /* The final value of the iv is iv1->base + MOD, assuming that this
  	 computation does not overflow, and that
  	 iv0->base <= iv1->base + MOD.  */
!       if (!iv1->no_overflow && !integer_zerop (mod))
  	{
  	  bound = fold_build2 (MINUS_EXPR, type,
  			       TYPE_MAX_VALUE (type1), tmod);
--- 697,703 ----
        /* The final value of the iv is iv1->base + MOD, assuming that this
  	 computation does not overflow, and that
  	 iv0->base <= iv1->base + MOD.  */
!       if (!iv0->no_overflow && !integer_zerop (mod))
  	{
  	  bound = fold_build2 (MINUS_EXPR, type,
  			       TYPE_MAX_VALUE (type1), tmod);
*************** number_of_iterations_lt_to_ne (tree type
*** 719,725 ****
        /* The final value of the iv is iv0->base - MOD, assuming that this
  	 computation does not overflow, and that
  	 iv0->base - MOD <= iv1->base. */
!       if (!iv0->no_overflow && !integer_zerop (mod))
  	{
  	  bound = fold_build2 (PLUS_EXPR, type1,
  			       TYPE_MIN_VALUE (type1), tmod);
--- 719,725 ----
        /* The final value of the iv is iv0->base - MOD, assuming that this
  	 computation does not overflow, and that
  	 iv0->base - MOD <= iv1->base. */
!       if (!iv1->no_overflow && !integer_zerop (mod))
  	{
  	  bound = fold_build2 (PLUS_EXPR, type1,
  			       TYPE_MIN_VALUE (type1), tmod);