Hi!
The following testcase ICEs, because store_expr is not prepared to handle
VCE-like assignments (while store_field can handle that).
We already had same mode check for the store_expr when size covers the whole
complex to_rtx, this patch adds a similar check when storing just one half
of it (first or second). Richard Sandiford recently added
gcc_checking_assert (COMPLEX_MODE_P (to_mode));
so the patch also removes a now dead && COMPLEX_MODE_P (GET_MODE (to_rtx))
check, and uses the to_mode temporary to simplify the code.
Bootstrapped/regtested on x86_64-linux and i686-linux, ok for trunk/8.2?
2018-06-15 Jakub Jelinek <ja...@redhat.com>
PR middle-end/85878
* expr.c (expand_assignment): Remove now redundant COMPLEX_MODE_P
check from first store_expr, use to_mode instead of GET_MODE (to_rtx).
Only call store_expr for halves if the mode is the same.
* gfortran.fortran-torture/compile/pr85878.f90: New test.
--- gcc/expr.c.jj 2018-06-13 10:05:30.000000000 +0200
+++ gcc/expr.c 2018-06-15 16:29:34.904387548 +0200
@@ -5150,12 +5150,12 @@ expand_assignment (tree to, tree from, b
gcc_checking_assert (COMPLEX_MODE_P (to_mode));
poly_int64 mode_bitsize = GET_MODE_BITSIZE (to_mode);
unsigned short inner_bitsize = GET_MODE_UNIT_BITSIZE (to_mode);
- if (TYPE_MODE (TREE_TYPE (from)) == GET_MODE (to_rtx)
- && COMPLEX_MODE_P (GET_MODE (to_rtx))
+ if (TYPE_MODE (TREE_TYPE (from)) == to_mode
&& known_eq (bitpos, 0)
&& known_eq (bitsize, mode_bitsize))
result = store_expr (from, to_rtx, false, nontemporal, reversep);
- else if (known_eq (bitsize, inner_bitsize)
+ else if (TYPE_MODE (TREE_TYPE (from)) == GET_MODE_INNER (to_mode)
+ && known_eq (bitsize, inner_bitsize)
&& (known_eq (bitpos, 0)
|| known_eq (bitpos, inner_bitsize)))
result = store_expr (from, XEXP (to_rtx, maybe_ne (bitpos, 0)),
--- gcc/testsuite/gfortran.fortran-torture/compile/pr85878.f90.jj
2018-06-15 16:50:53.825639713 +0200
+++ gcc/testsuite/gfortran.fortran-torture/compile/pr85878.f90 2018-06-15
16:49:20.483548544 +0200
@@ -0,0 +1,8 @@
+! PR middle-end/85878
+
+program pr85878
+ real :: a
+ complex :: c = (2.0, 3.0)
+ print *, c
+ print *, transfer (a, c)
+end
Jakub
