This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

[patch, fortran] power.f90 FAIL in testsuite


I'm looking for commentary on this proposed patch, which fixes the
incorrect calculation of integer powers when the base is 1. The patch
identifies power operations with an integer base of 1, and replace them
with a constant 1. The conversion takes place in eval_intrinsic().

Such a transform is logical, but I'm not certain I'm doing it in the
most appropriate place. I picked something small, so my glaring mistakes could be easily identified... ;)


Then, of course, we get into the issue of whether or not other known transforms should be applied, such as for real and complex types.

This patch also modifies the test suite slightly, by removing the complex power tests from power.f90, and creating a new cpower.f90 test that, at this time, fails, likely due to the complex-specific PR 17603.

The test suite is not very comprehensive, testing a few specific cases but not general correctness. Combing integer, real, and complex tests can cloud results. I'm willing to propose (as an example) test case examples for ** operator tests, providing better coverage and granularity.

..Scott

2004-11-01 Scott Robert Ladd <scott.ladd@coyotegulch.com>

    * gcc/fortran/arith.c (eval_intrinsic): Added transform from power
      of integer 1 to a constant of 1

    * gcc/testsuite/gfortran.fortran-torture/execute/power.f90:
      Removed complex power tests to cpower.f90

    * gcc/testsuite/gfortran.fortran-torture/execute/cpower.f90:
      Complex power tests


Index: MAINTAINERS
===================================================================
RCS file: /cvs/gcc/gcc/MAINTAINERS,v
retrieving revision 1.380
diff -u -3 -p -r1.380 MAINTAINERS
--- MAINTAINERS	20 Oct 2004 07:50:06 -0000	1.380
+++ MAINTAINERS	1 Nov 2004 19:37:03 -0000
@@ -247,7 +247,11 @@ Matthias Klose					doko@debian.org
 Jeff Knaggs					jknaggs@redhat.com
 Michael Koch					konqueror@gmx.de
 Matt Kraai					kraai@alumni.cmu.edu
+<<<<<<< MAINTAINERS
+Scott Robert Ladd				scott.ladd@coyotegulch.com
+=======
 Aaron W. LaFramboise				aaronavay62@aaronwl.com
+>>>>>>> 1.380
 Marc Lehmann					pcg@goof.com
 Alan Lehotsky					apl@alum.mit.edu
 James Lemke					jim@wasabisystems.com
Index: gcc/fortran/arith.c
===================================================================
RCS file: /cvs/gcc/gcc/gcc/fortran/arith.c,v
retrieving revision 1.17
diff -u -3 -p -r1.17 arith.c
--- gcc/fortran/arith.c	28 Oct 2004 21:43:46 -0000	1.17
+++ gcc/fortran/arith.c	1 Nov 2004 19:37:09 -0000
@@ -1633,6 +1633,19 @@ eval_intrinsic (gfc_intrinsic_op operato
       gfc_internal_error ("eval_intrinsic(): Bad operator");
     }
 
+  /* Simplify power operation to constant if first op1 is 1 */
+  if (operator == INTRINSIC_POWER && op1->ts.type == BT_INTEGER)
+    {
+      if (mpz_cmp_si (op1->value.integer,1) == 0)
+        {
+          result = gfc_get_expr ();
+          *result = *op1;
+          gfc_free_expr (op1);
+          gfc_free_expr (op2);
+          return result;
+        }
+    }
+
   /* Try to combine the operators.  */
   if (operator == INTRINSIC_POWER && op2->ts.type != BT_INTEGER)
     goto runtime;
Index: gcc/testsuite/gfortran.fortran-torture/execute/cpower.f90
===================================================================
RCS file: gcc/testsuite/gfortran.fortran-torture/execute/cpower.f90
diff -N gcc/testsuite/gfortran.fortran-torture/execute/cpower.f90
--- /dev/null	1 Jan 1970 00:00:00 -0000
+++ gcc/testsuite/gfortran.fortran-torture/execute/cpower.f90	1 Nov 2004 19:37:23 -0000
@@ -0,0 +1,19 @@
+! Program to test the power (**) operator
+program testpow
+   implicit none
+   real(kind=4)    :: two
+   complex(kind=4) :: c, z
+   real, parameter :: del = 0.0001
+
+   c = (2.0, 3.0)
+   z = c ** 2
+   if (abs(z - (-5.0, 12.0)) .gt. del) call abort
+
+   z = c ** 7
+   if (abs(z - (6554.0, 4449.0)) .gt. del) call abort
+
+   two = 2.0
+   c = (2.0, 3.0)
+   c = c ** two
+   if (abs(c - (-5.0, 12.0)) .gt. del) call abort
+end program
Index: gcc/testsuite/gfortran.fortran-torture/execute/power.f90
===================================================================
RCS file: /cvs/gcc/gcc/gcc/testsuite/gfortran.fortran-torture/execute/power.f90,v
retrieving revision 1.3
diff -u -3 -p -r1.3 power.f90
--- gcc/testsuite/gfortran.fortran-torture/execute/power.f90	19 May 2004 00:34:57 -0000	1.3
+++ gcc/testsuite/gfortran.fortran-torture/execute/power.f90	1 Nov 2004 19:37:24 -0000
@@ -3,7 +3,6 @@ program testpow
    implicit none
    real(kind=4) r, s, two
    real(kind=8) :: q
-   complex(kind=4) :: c, z
    real, parameter :: del = 0.0001
    integer i, j
 
@@ -33,12 +32,6 @@ program testpow
    j = i ** (-11)
    if (abs (j - (-1)) .gt. del) call abort
 
-   c = (2.0, 3.0)
-   z = c ** 2
-   if (abs(z - (-5.0, 12.0)) .gt. del) call abort
-   z = c ** 7
-   if (abs(z - (6554.0, 4449.0)) .gt. del) call abort
-
    two = 2.0
 
    r = two ** 1
@@ -69,7 +62,4 @@ program testpow
    i = -3
    r = two ** i
    if (abs (r - 0.125) .gt. del) call abort
-   c = (2.0, 3.0)
-   c = c ** two
-   if (abs(c - (-5.0, 12.0)) .gt. del) call abort
 end program

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]