PR 70687 reports a case where combine.c mishandles integer modes
wider than unsigned HOST_WIDE_INT. I don't have a testcase since
the PR is just pointing out the hole.
Also, I think a ZERO_EXTEND of a vector mode could in principle satisfy
the subreg condition but wouldn't be equivalent to an AND. E.g.:
(zero_extend:V4DI (subreg:V4SI (reg:V4DI R) 0))
Tested on x86_64-linux-gnu.
gcc/
PR rtl-optimization/70687
* combine.c (change_zero_ext): Check for scalar modes. Use wide_int
instead of unsigned HOST_WIDE_INT.
git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@235811
138bc75d-0d04-0410-961f-
82ee72b054a4
+2016-05-03 Richard Sandiford <richard.sandiford@arm.com>
+
+ PR rtl-optimization/70687
+ * combine.c (change_zero_ext): Check for scalar modes. Use wide_int
+ instead of unsigned HOST_WIDE_INT.
+
2016-05-03 Bernd Schmidt <bschmidt@redhat.com>
PR rtl-optimization/44281
XEXP (x, 0), GEN_INT (start));
}
else if (GET_CODE (x) == ZERO_EXTEND
+ && SCALAR_INT_MODE_P (mode)
&& GET_CODE (XEXP (x, 0)) == SUBREG
&& GET_MODE (SUBREG_REG (XEXP (x, 0))) == mode
&& subreg_lowpart_p (XEXP (x, 0)))
else
continue;
- unsigned HOST_WIDE_INT mask = 1;
- mask <<= size;
- mask--;
-
- x = gen_rtx_AND (mode, x, GEN_INT (mask));
+ wide_int mask = wi::mask (size, false, GET_MODE_PRECISION (mode));
+ x = gen_rtx_AND (mode, x, immed_wide_int_const (mask, mode));
SUBST (**iter, x);
changed = true;