https://gcc.gnu.org/bugzilla/show_bug.cgi?id=89226
Jakub Jelinek <jakub at gcc dot gnu.org> changed:
What |Removed |Added
----------------------------------------------------------------------------
Status|UNCONFIRMED |NEW
Last reconfirmed| |2019-02-06
CC| |hjl.tools at gmail dot com,
| |jakub at gcc dot gnu.org,
| |uros at gcc dot gnu.org
Ever confirmed|0 |1
--- Comment #3 from Jakub Jelinek <jakub at gcc dot gnu.org> ---
Seems most of the *by_pieces code actually uses widest_int_mode_for_size which
already handles even the wider modes as long as they have a mov<mode>
instruction. With this completely untested patch I get roughly the same code
with -mavx and better with -mavx512f, just as a drawback for some reason the
functions have frame pointer (dunno if that is caused by the OI/XImode, while
vector modes can be handled or what else). Tried memset with zero too, but
haven't tried other memsets (those could be problematic already) or
comparisons.
Thoughts on this? Not a GCC9 material though. Perhaps it should also depend
on the selected preferred vector width, so that we don't e.g. enable AVX512F if
that is undesirable from power consumption POV.
--- gcc/config/i386/i386.h.jj 2019-01-01 12:37:32.988715207 +0100
+++ gcc/config/i386/i386.h 2019-02-06 21:13:01.047765193 +0100
@@ -1886,7 +1886,9 @@ typedef struct ix86_args {
&& TARGET_SSE2 \
&& TARGET_SSE_UNALIGNED_LOAD_OPTIMAL \
&& TARGET_SSE_UNALIGNED_STORE_OPTIMAL) \
- ? GET_MODE_SIZE (TImode) : UNITS_PER_WORD)
+ ? (TARGET_AVX512F ? GET_MODE_SIZE (XImode) \
+ : TARGET_AVX ? GET_MODE_SIZE (OImode) \
+ : GET_MODE_SIZE (TImode)) : UNITS_PER_WORD)
/* If a memory-to-memory move would take MOVE_RATIO or more simple
move-instruction pairs, we will do a movmem or libcall instead.
--- gcc/expr.c.jj 2019-01-22 10:09:47.304587822 +0100
+++ gcc/expr.c 2019-02-06 21:23:18.041603623 +0100
@@ -713,14 +713,31 @@ convert_modes (machine_mode mode, machin
return temp;
}
+/* Return the widest integer mode that is narrower than SIZE bytes. */
+
+static scalar_int_mode
+widest_int_mode_for_size (unsigned int size)
+{
+ scalar_int_mode result = NARROWEST_INT_MODE;
+
+ gcc_checking_assert (size > 1);
+
+ opt_scalar_int_mode tmode;
+ FOR_EACH_MODE_IN_CLASS (tmode, MODE_INT)
+ if (GET_MODE_SIZE (tmode.require ()) < size)
+ result = tmode.require ();
+
+ return result;
+}
+
/* Return the largest alignment we can use for doing a move (or store)
of MAX_PIECES. ALIGN is the largest alignment we could use. */
static unsigned int
alignment_for_piecewise_move (unsigned int max_pieces, unsigned int align)
{
- scalar_int_mode tmode
- = int_mode_for_size (max_pieces * BITS_PER_UNIT, 1).require ();
+ scalar_int_mode tmode = widest_int_mode_for_size (max_pieces + 1);
+ gcc_assert (GET_MODE_SIZE (tmode) == max_pieces);
if (align >= GET_MODE_ALIGNMENT (tmode))
align = GET_MODE_ALIGNMENT (tmode);
@@ -743,23 +760,6 @@ alignment_for_piecewise_move (unsigned i
return align;
}
-/* Return the widest integer mode that is narrower than SIZE bytes. */
-
-static scalar_int_mode
-widest_int_mode_for_size (unsigned int size)
-{
- scalar_int_mode result = NARROWEST_INT_MODE;
-
- gcc_checking_assert (size > 1);
-
- opt_scalar_int_mode tmode;
- FOR_EACH_MODE_IN_CLASS (tmode, MODE_INT)
- if (GET_MODE_SIZE (tmode.require ()) < size)
- result = tmode.require ();
-
- return result;
-}
-
/* Determine whether an operation OP on LEN bytes with alignment ALIGN can
and should be performed piecewise. */
