rtl-optimization/103075 - avoid ICEing on unfolded int-to-float converts

The following avoids asserting in exact_int_to_float_conversion_p that
the argument is not constant which it in fact can be with
-frounding-math and inexact int-to-float conversions.  Say so.

2021-11-04  Richard Biener  <rguenther@suse.de>

	PR rtl-optimization/103075
	* simplify-rtx.c (exact_int_to_float_conversion_p): Return
	false for a VOIDmode operand.

	* gcc.dg/pr103075.c: New testcase.

gcc/simplify-rtx.c

@@ -899,10 +899,12 @@ simplify_context::simplify_unary_operation (rtx_code code, machine_mode mode,
 static bool
 exact_int_to_float_conversion_p (const_rtx op)
 {
-  int out_bits = significand_size (GET_MODE_INNER (GET_MODE (op)));
   machine_mode op0_mode = GET_MODE (XEXP (op, 0));
-  /* Constants shouldn't reach here.  */
-  gcc_assert (op0_mode != VOIDmode);
+  /* Constants can reach here with -frounding-math, if they do then
+     the conversion isn't exact.  */
+  if (op0_mode == VOIDmode)
+    return false;
+  int out_bits = significand_size (GET_MODE_INNER (GET_MODE (op)));
   int in_prec = GET_MODE_UNIT_PRECISION (op0_mode);
   int in_bits = in_prec;
   if (HWI_COMPUTABLE_MODE_P (op0_mode))

gcc/testsuite/gcc.dg/pr103075.c

@@ -0,0 +1,8 @@
+/* { dg-do compile } */
+/* { dg-options "-O -frounding-math" } */
+
+float
+foo (void)
+{
+  return (float) 0x1699925 * 1.1;
+}