Hello world,
the attached patch implements simplification for cshift completely.
It also fixes a bug where compile-time simplification was handled
incorrectly for a negative value. For PR 83650 (the wrong
simplification) for the other branches I suspect a quite simple fix will
be possible, which I will address separately.
Regression-tested. OK for trunk?
Regards
Thomas
2018-01-02 Thomas Koenig <tkoe...@gcc.gnu.org>
PR fortran/45689
PR fortran/83650
* simplify.c (gfc_simplify_cshift): Re-implement to allow full
range of arguments.
2018-01-02 Thomas Koenig <tkoe...@gcc.gnu.org>
PR fortran/45689
PR fortran/83650
* gfortran.dg/simplify_cshift_1.f90: Correct erroneous case.
* gfortran.dg/simplify_cshift_4.f90: New test.
Index: fortran/simplify.c
===================================================================
--- fortran/simplify.c (Revision 255788)
+++ fortran/simplify.c (Arbeitskopie)
@@ -1950,92 +1950,212 @@ gfc_simplify_count (gfc_expr *mask, gfc_expr *dim,
simplify_transformation_to_array (result, mask, dim, mask, gfc_count, NULL);
}
+/* Simplification routine for cshift. This works by copying the array
+ expressions into a one-dimensional array, shuffling the values into another
+ one-dimensional array and creating the new array expression from this. The
+ shuffling part is basically taken from the library routine. */
gfc_expr *
gfc_simplify_cshift (gfc_expr *array, gfc_expr *shift, gfc_expr *dim)
{
- gfc_expr *a, *result;
- int dm;
+ gfc_expr *result;
+ int which;
+ gfc_expr **arrayvec, **resultvec;
+ gfc_expr **rptr, **sptr;
+ mpz_t size;
+ size_t arraysize, shiftsize, i;
+ gfc_constructor *array_ctor, *shift_ctor;
+ ssize_t *shiftvec, *hptr;
+ ssize_t shift_val, len;
+ ssize_t count[GFC_MAX_DIMENSIONS], extent[GFC_MAX_DIMENSIONS],
+ hs_ex[GFC_MAX_DIMENSIONS],
+ hstride[GFC_MAX_DIMENSIONS], sstride[GFC_MAX_DIMENSIONS],
+ a_extent[GFC_MAX_DIMENSIONS], a_stride[GFC_MAX_DIMENSIONS],
+ h_extent[GFC_MAX_DIMENSIONS],
+ ss_ex[GFC_MAX_DIMENSIONS];
+ ssize_t rsoffset;
+ int d, n;
+ bool continue_loop;
+ gfc_expr **src, **dest;
- /* DIM is only useful for rank > 1, but deal with it here as one can
- set DIM = 1 for rank = 1. */
+ if (!is_constant_array_expr (array))
+ return NULL;
+
+ if (shift->rank > 0)
+ gfc_simplify_expr (shift, 1);
+
+ if (!gfc_is_constant_expr (shift))
+ return NULL;
+
+ /* Make dim zero-based. */
if (dim)
{
if (!gfc_is_constant_expr (dim))
return NULL;
- dm = mpz_get_si (dim->value.integer);
+ which = mpz_get_si (dim->value.integer) - 1;
}
else
- dm = 1;
+ which = 0;
- /* Copy array into 'a', simplify it, and then test for a constant array. */
- a = gfc_copy_expr (array);
- gfc_simplify_expr (a, 0);
- if (!is_constant_array_expr (a))
+ gfc_array_size (array, &size);
+ arraysize = mpz_get_ui (size);
+ mpz_clear (size);
+
+ result = gfc_get_array_expr (array->ts.type, array->ts.kind, &array->where);
+ result->shape = gfc_copy_shape (array->shape, array->rank);
+ result->rank = array->rank;
+ result->ts.u.derived = array->ts.u.derived;
+
+ if (arraysize == 0)
+ return result;
+
+ arrayvec = XCNEWVEC (gfc_expr *, arraysize);
+ array_ctor = gfc_constructor_first (array->value.constructor);
+ for (i = 0; i < arraysize; i++)
{
- gfc_free_expr (a);
- return NULL;
+ arrayvec[i] = array_ctor->expr;
+ array_ctor = gfc_constructor_next (array_ctor);
}
- if (a->rank == 1)
+ resultvec = XCNEWVEC (gfc_expr *, arraysize);
+
+ extent[0] = 1;
+ count[0] = 0;
+
+ for (d=0; d < array->rank; d++)
{
- gfc_constructor *ca, *cr;
- mpz_t size;
- int i, j, shft, sz;
+ a_extent[d] = mpz_get_si (array->shape[d]);
+ a_stride[d] = d == 0 ? 1 : a_stride[d-1] * a_extent[d-1];
+ }
- if (!gfc_is_constant_expr (shift))
+ if (shift->rank > 0)
+ {
+ gfc_array_size (shift, &size);
+ shiftsize = mpz_get_ui (size);
+ mpz_clear (size);
+ shiftvec = XCNEWVEC (ssize_t, shiftsize);
+ shift_ctor = gfc_constructor_first (shift->value.constructor);
+ for (d = 0; d < shift->rank; d++)
{
- gfc_free_expr (a);
- return NULL;
+ h_extent[d] = mpz_get_si (shift->shape[d]);
+ hstride[d] = d == 0 ? 1 : hstride[d-1] * h_extent[d-1];
}
+ }
+ else
+ shiftvec = NULL;
+
+ /* Shut up compiler */
+ len = 1;
+ rsoffset = 1;
- shft = mpz_get_si (shift->value.integer);
+ n = 0;
+ for (d=0; d < array->rank; d++)
+ {
+ if (d == which)
+ {
+ rsoffset = a_stride[d];
+ len = a_extent[d];
+ }
+ else
+ {
+ count[n] = 0;
+ extent[n] = a_extent[d];
+ sstride[n] = a_stride[d];
+ ss_ex[n] = sstride[n] * extent[n];
+ if (shiftvec)
+ hs_ex[n] = hstride[n] * extent[n];
+ n++;
+ }
+ }
- /* Case (i): If ARRAY has rank one, element i of the result is
- ARRAY (1 + MODULO (i + SHIFT - 1, SIZE (ARRAY))). */
-
- mpz_init (size);
- gfc_array_size (a, &size);
- sz = mpz_get_si (size);
- mpz_clear (size);
-
- /* Adjust shft to deal with right or left shifts. */
- shft = shft < 0 ? 1 - shft : shft;
-
- /* Special case: Shift to the original order! */
- if (sz == 0 || shft % sz == 0)
- return a;
-
- result = gfc_copy_expr (a);
- cr = gfc_constructor_first (result->value.constructor);
- for (i = 0; i < sz; i++, cr = gfc_constructor_next (cr))
+ if (shiftvec)
+ {
+ for (i = 0; i < shiftsize; i++)
{
- j = (i + shft) % sz;
- ca = gfc_constructor_first (a->value.constructor);
- while (j-- > 0)
- ca = gfc_constructor_next (ca);
- cr->expr = gfc_copy_expr (ca->expr);
+ ssize_t val;
+ val = mpz_get_si (shift_ctor->expr->value.integer);
+ val = val % len;
+ if (val < 0)
+ val += len;
+ shiftvec[i] = val;
+ shift_ctor = gfc_constructor_next (shift_ctor);
}
-
- gfc_free_expr (a);
- return result;
+ shift_val = 0;
}
else
{
- /* FIXME: Deal with rank > 1 arrays. For now, don't leak memory. */
+ shift_val = mpz_get_si (shift->value.integer);
+ shift_val = shift_val % len;
+ if (shift_val < 0)
+ shift_val += len;
+ }
- /* GCC bootstrap is too stupid to realize that the above code for dm
- is correct. First, dim can be specified for a rank 1 array. It is
- not needed in this nor used here. Second, the code is simply waiting
- for someone to implement rank > 1 simplification. For now, add a
- pessimization to the code that has a zero valid reason to be here. */
- if (dm > array->rank)
- gcc_unreachable ();
+ continue_loop = true;
+ d = array->rank;
+ rptr = resultvec;
+ sptr = arrayvec;
+ hptr = shiftvec;
- gfc_free_expr (a);
+ while (continue_loop)
+ {
+ ssize_t sh;
+ if (shiftvec)
+ sh = *hptr;
+ else
+ sh = shift_val;
+
+ src = &sptr[sh * rsoffset];
+ dest = rptr;
+ for (n = 0; n < len - sh; n++)
+ {
+ *dest = *src;
+ dest += rsoffset;
+ src += rsoffset;
+ }
+ src = sptr;
+ for ( n = 0; n < sh; n++)
+ {
+ *dest = *src;
+ dest += rsoffset;
+ src += rsoffset;
+ }
+ rptr += sstride[0];
+ sptr += sstride[0];
+ if (shiftvec)
+ hptr += hstride[0];
+ count[0]++;
+ n = 0;
+ while (count[n] == extent[n])
+ {
+ count[n] = 0;
+ rptr -= ss_ex[n];
+ sptr -= ss_ex[n];
+ if (shiftvec)
+ hptr -= hs_ex[n];
+ n++;
+ if (n >= d - 1)
+ {
+ continue_loop = false;
+ break;
+ }
+ else
+ {
+ count[n]++;
+ rptr += sstride[n];
+ sptr += sstride[n];
+ if (shiftvec)
+ hptr += hstride[n];
+ }
+ }
}
- return NULL;
+ for (i = 0; i < arraysize; i++)
+ {
+ gfc_constructor_append_expr (&result->value.constructor,
+ gfc_copy_expr (resultvec[i]),
+ NULL);
+ }
+ return result;
}
Index: testsuite/gfortran.dg/simplify_cshift_1.f90
===================================================================
--- testsuite/gfortran.dg/simplify_cshift_1.f90 (Revision 255788)
+++ testsuite/gfortran.dg/simplify_cshift_1.f90 (Arbeitskopie)
@@ -23,12 +23,12 @@ program foo
v = cshift(c, 2)
if (any(b /= v)) call abort
- ! Special cases shift = 0, size(a), 1-size(a)
+ ! Special cases shift = 0, size(a), -size(a)
b = cshift([1, 2, 3, 4, 5], 0)
if (any(b /= a)) call abort
b = cshift([1, 2, 3, 4, 5], size(a))
if (any(b /= a)) call abort
- b = cshift([1, 2, 3, 4, 5], 1-size(a))
+ b = cshift([1, 2, 3, 4, 5], -size(a))
if (any(b /= a)) call abort
! simplification of array arg.
! { dg-do run }
program main
implicit none
integer :: i
integer, parameter, dimension(3,3) :: a = &
reshape([1,2,3,4,5,6,7,8,9], shape(a))
integer, dimension(3,3) :: b
integer, parameter, dimension(3,4,5) :: c = &
reshape([(i**2,i=1,3*4*5)],shape(c))
integer, dimension(3,4,5) :: d
integer, dimension(4,5), parameter :: sh1 =&
reshape([(i**3-12*i**2,i=1,4*5)],shape(sh1))
integer, dimension(3,5), parameter :: sh2 = &
reshape([(i**3-7*i**2,i=1,3*5)], shape(sh2))
integer, dimension(3,4), parameter :: sh3 = &
reshape([(i**3-3*i**2,i=1,3*4)], shape(sh3))
integer, parameter, dimension(3,4,5) :: c1 = cshift(c,shift=sh1,dim=1)
integer, parameter, dimension(3,4,5) :: c2 = cshift(c,shift=sh2,dim=2)
integer, parameter, dimension(3,4,5) :: c3 = cshift(c,shift=sh3,dim=3)
b = a
if (any(cshift(a,1) /= cshift(b,1))) call abort
if (any(cshift(a,2) /= cshift(b,2))) call abort
if (any(cshift(a,1,dim=2) /= cshift(b,1,dim=2))) call abort
d = c
if (any(cshift(c,1) /= cshift(d,1))) call abort
if (any(cshift(c,2) /= cshift(d,2))) call abort
if (any(cshift(c,3) /= cshift(d,3))) call abort
if (any(cshift(c,1,dim=2) /= cshift(d,1,dim=2))) call abort
if (any(cshift(c,2,dim=2) /= cshift(d,2,dim=2))) call abort
if (any(cshift(c,3,dim=3) /= cshift(d,3,dim=3))) call abort
if (any(cshift(d,shift=sh1,dim=1) /= c1)) call abort
if (any(cshift(d,shift=sh2,dim=2) /= c2)) call abort
if (any(cshift(d,shift=sh3,dim=3) /= c3)) call abort
end program main
