Re: [PATCH v3 1/3] rtl: directly handle MEM in gen_highpart [PR102125]

[Richard Earnshaw via Gcc-patches]

On 13/09/2021 10:38, Richard Sandiford via Gcc-patches wrote:
Richard Earnshaw via Gcc-patches <gcc-patches@gcc.gnu.org> writes:
gen_lowpart_general handles forming a lowpart of a MEM by using
adjust_address to rework and validate a new version of the MEM.
Do the same for gen_highpart rather than calling simplify_gen_subreg
for this case.

Looks OK, but what went wrong with the existing code?  Did
simplify_gen_subreg refuse to handle a MEM that you wanted
it to handle, or did the validize_mem go wrong for some reason?

It refused to handle it and simply returned (subreg (mem)) - see the 
discussion on version 1 of the patch series.

R.


Thanks,
Richard

gcc/ChangeLog:

        PR target/102125
        * emit-rtl.c (gen_highpart): Use adjust_address to handle
        MEM rather than calling simplify_gen_subreg.
---
  gcc/emit-rtl.c | 23 +++++++++++++----------
  1 file changed, 13 insertions(+), 10 deletions(-)

diff --git a/gcc/emit-rtl.c b/gcc/emit-rtl.c
index 77ea8948ee8..0ba110879aa 100644
--- a/gcc/emit-rtl.c
+++ b/gcc/emit-rtl.c
@@ -1585,19 +1585,22 @@ gen_highpart (machine_mode mode, rtx x)
    gcc_assert (known_le (msize, (unsigned int) UNITS_PER_WORD)
              || known_eq (msize, GET_MODE_UNIT_SIZE (GET_MODE (x))));
  
-  result = simplify_gen_subreg (mode, x, GET_MODE (x),
-                               subreg_highpart_offset (mode, GET_MODE (x)));
-  gcc_assert (result);
-
-  /* simplify_gen_subreg is not guaranteed to return a valid operand for
-     the target if we have a MEM.  gen_highpart must return a valid operand,
-     emitting code if necessary to do so.  */
-  if (MEM_P (result))
+  /* gen_lowpart_common handles a lot of special cases due to needing to handle
+     paradoxical subregs; it only calls simplify_gen_subreg when certain that
+     it will produce something meaningful.  The only case we need to handle
+     specially here is MEM.  */
+  if (MEM_P (x))
      {
-      result = validize_mem (result);
-      gcc_assert (result);
+      poly_int64 offset = subreg_highpart_offset (mode, GET_MODE (x));
+      return adjust_address (x, mode, offset);
      }
  
+  result = simplify_gen_subreg (mode, x, GET_MODE (x),
+                               subreg_highpart_offset (mode, GET_MODE (x)));
+  /* Since we handle MEM directly above, we should never get a MEM back
+     from simplify_gen_subreg.  */
+  gcc_assert (result && !MEM_P (result));
+
    return result;
  }