Allow larger copies when not slow_unaligned_access and no padding.

This allows copy_blkmode_to_reg to perform larger copies when it is safe to do so by calculating
the bitsize per iteration doing the maximum copy allowed that does not read more
than the amount of bits left to copy.

Strictly speaking, this copying is only done if:

  1. the target supports fast unaligned access
  2. no padding is being used.

This should avoid the issues of the first patch (PR85123) but still work for targets that are safe
to do so.

Original patch https://gcc.gnu.org/ml/gcc-patches/2017-11/msg01088.html
Previous respin https://gcc.gnu.org/ml/gcc-patches/2018-04/msg00239.html

gcc/
2018-08-16  Tamar Christina  <tamar.christina@arm.com>

	* expr.c (copy_blkmode_to_reg): Perform larger copies when safe.

git-svn-id: svn+ssh://gcc.gnu.org/svn/gcc/trunk@263594 138bc75d-0d04-0410-961f-82ee72b054a4

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index d913d7d9c0108e17c9a0663b4a9b1b5a2abf45c0..62969666e092a783d8e5415b2dea911df4658ce5 100644 (file)
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,7 @@
+2018-08-16  Tamar Christina  <tamar.christina@arm.com>
+
+       * expr.c (copy_blkmode_to_reg): Perform larger copies when safe.
+
 2018-08-16  Matthew Malcomson  <matthew.malcomson@arm.com>
 
        * doc/rtl.texi: Replace old RTX class names with new names.

diff --git a/gcc/expr.c b/gcc/expr.c
index de6709defd601b8dfb9d49f45b4e19fcd97504f2..e8a4f8831288ddc7a89151313515b0471587db22 100644 (file)
--- a/gcc/expr.c
+++ b/gcc/expr.c
@@ -2767,6 +2767,7 @@ copy_blkmode_to_reg (machine_mode mode_in, tree src)
   /* No current ABI uses variable-sized modes to pass a BLKmnode type.  */
   fixed_size_mode mode = as_a <fixed_size_mode> (mode_in);
   fixed_size_mode dst_mode;
+  scalar_int_mode min_mode;
 
   gcc_assert (TYPE_MODE (TREE_TYPE (src)) == BLKmode);
 
@@ -2796,6 +2797,7 @@ copy_blkmode_to_reg (machine_mode mode_in, tree src)
   n_regs = (bytes + UNITS_PER_WORD - 1) / UNITS_PER_WORD;
   dst_words = XALLOCAVEC (rtx, n_regs);
   bitsize = MIN (TYPE_ALIGN (TREE_TYPE (src)), BITS_PER_WORD);
+  min_mode = smallest_int_mode_for_size (bitsize);
 
   /* Copy the structure BITSIZE bits at a time.  */
   for (bitpos = 0, xbitpos = padding_correction;
@@ -2816,6 +2818,25 @@ copy_blkmode_to_reg (machine_mode mode_in, tree src)
          emit_move_insn (dst_word, CONST0_RTX (word_mode));
        }
 
+      /* Find the largest integer mode that can be used to copy all or as
+        many bits as possible of the structure if the target supports larger
+        copies.  There are too many corner cases here w.r.t to alignments on
+        the read/writes.  So if there is any padding just use single byte
+        operations.  */
+      opt_scalar_int_mode mode_iter;
+      if (padding_correction == 0 && !STRICT_ALIGNMENT)
+       {
+         FOR_EACH_MODE_FROM (mode_iter, min_mode)
+           {
+             unsigned int msize = GET_MODE_BITSIZE (mode_iter.require ());
+             if (msize <= ((bytes * BITS_PER_UNIT) - bitpos)
+                 && msize <= BITS_PER_WORD)
+               bitsize = msize;
+             else
+               break;
+           }
+       }
+
       /* We need a new source operand each time bitpos is on a word
         boundary.  */
       if (bitpos % BITS_PER_WORD == 0)