On Fri, Apr 23, 2021 at 1:35 AM H.J. Lu via Gcc-patches
<gcc-patches@gcc.gnu.org> wrote:
>
> For op_by_pieces operations between two areas of memory on non-strict
> alignment target, add -foverlap-op-by-pieces=[off|on|max-memset] to
> generate overlapping operations to minimize number of operations if it
> is not a stack push which must not overlap.
>
> When operating on LENGTH bytes of memory, -foverlap-op-by-pieces=on
> starts with the widest usable integer size, MAX_SIZE, for LENGTH bytes
> and finishes with the smallest usable integer size, MIN_SIZE, for the
> remaining bytes where MAX_SIZE >= MIN_SIZE. If MIN_SIZE > the remaining
> bytes, the last operation is performed on MIN_SIZE bytes of overlapping
> memory from the previous operation.
>
> For memset with non-zero byte, -foverlap-op-by-pieces=max-memset generates
> an overlapping fill with MAX_SIZE if the number of the remaining bytes is
> greater than one.
>
> Tested on Linux/x86-64 with both -foverlap-op-by-pieces enabled and
> disabled by default.
Neither the user documentation nor the patch description tells me what
"generate overlapping operations" does. I _suspect_ it's doing an
offset adjusted read/write of the last piece of a memory region to
avoid doing more than one smaller operations. Thus for a region
of size 7 and 4-byte granular ops you'd do operations at
offset 0 and 3 rather than one at 0, a two-byte at offset 4 and
a one-byte at offset 7.
When the tail is of power-of-two size you still generate non-overlapping
ops?
For memmove there's a correctness issue so you have to make sure
to first load the last two ops before performing the stores which
increases register pressure.
I'm not sure we want a -f option to control this - not all targets will
be able to support this. So I'd use a target hook or rather extend
the existing use_by_pieces_infrastructure_p hook with an alternate
return (some flags bitmask I guess). We do have one extra
target hook, compare_by_pieces_branch_ratio, so by that using
an alternate hook might be also OK.
Adding a -m option in targets that want this user-controllable would
be OK of course.
Richard.
> gcc/
>
> PR middl-end/90773
> * common.opt (-foverlap-op-by-pieces): New.
> * expr.c (by_pieces_ninsns): If -foverlap-op-by-pieces is enabled,
> round up size and alignment to the widest integer mode for maximum
> size
> (op_by_pieces_d): Add get_usable_mode, m_push and
> m_non_zero_memset.
> (op_by_pieces_d::op_by_pieces_d): Add 2 bool arguments to
> initialize m_push and m_non_zero_memset.
> (op_by_pieces_d::get_usable_mode): New.
> (op_by_pieces_d::run): Use get_usable_mode to get the largest
> usable integer mode and generate overlapping operations for
> -foverlap-op-by-pieces.
> (PUSHG_P): New.
> (move_by_pieces_d::move_by_pieces_d): Updated for op_by_pieces_d
> change.
> (store_by_pieces_d::store_by_pieces_d): Likewise.
> (clear_by_pieces): Likewsie.
> * toplev.c (process_options): Issue an error when
> -foverlap-op-by-pieces is used for strict alignment target.
> * doc/invoke.texi: Document -foverlap-op-by-pieces.
>
> gcc/testsuite/
>
> PR middl-end/90773
> * g++.dg/pr90773-1.h: New test.
> * g++.dg/pr90773-1a.C: Likewise.
> * g++.dg/pr90773-1b.C: Likewise.
> * g++.dg/pr90773-1c.C: Likewise.
> * g++.dg/pr90773-1d.C: Likewise.
> * gcc.target/i386/pr90773-1.c: Likewise.
> * gcc.target/i386/pr90773-2.c: Likewise.
> * gcc.target/i386/pr90773-3.c: Likewise.
> * gcc.target/i386/pr90773-4.c: Likewise.
> * gcc.target/i386/pr90773-5.c: Likewise.
> * gcc.target/i386/pr90773-6.c: Likewise.
> * gcc.target/i386/pr90773-7.c: Likewise.
> * gcc.target/i386/pr90773-8.c: Likewise.
> * gcc.target/i386/pr90773-9.c: Likewise.
> * gcc.target/i386/pr90773-10.c: Likewise.
> * gcc.target/i386/pr90773-11.c: Likewise.
> ---
> gcc/common.opt | 19 +++
> gcc/doc/invoke.texi | 14 ++
> gcc/expr.c | 159 ++++++++++++++++-----
> gcc/testsuite/g++.dg/pr90773-1.h | 14 ++
> gcc/testsuite/g++.dg/pr90773-1a.C | 13 ++
> gcc/testsuite/g++.dg/pr90773-1b.C | 5 +
> gcc/testsuite/g++.dg/pr90773-1c.C | 5 +
> gcc/testsuite/g++.dg/pr90773-1d.C | 19 +++
> gcc/testsuite/gcc.target/i386/pr90773-1.c | 17 +++
> gcc/testsuite/gcc.target/i386/pr90773-10.c | 13 ++
> gcc/testsuite/gcc.target/i386/pr90773-11.c | 13 ++
> gcc/testsuite/gcc.target/i386/pr90773-2.c | 20 +++
> gcc/testsuite/gcc.target/i386/pr90773-3.c | 23 +++
> gcc/testsuite/gcc.target/i386/pr90773-4.c | 13 ++
> gcc/testsuite/gcc.target/i386/pr90773-5.c | 13 ++
> gcc/testsuite/gcc.target/i386/pr90773-6.c | 11 ++
> gcc/testsuite/gcc.target/i386/pr90773-7.c | 11 ++
> gcc/testsuite/gcc.target/i386/pr90773-8.c | 13 ++
> gcc/testsuite/gcc.target/i386/pr90773-9.c | 13 ++
> gcc/toplev.c | 8 ++
> 20 files changed, 383 insertions(+), 33 deletions(-)
> create mode 100644 gcc/testsuite/g++.dg/pr90773-1.h
> create mode 100644 gcc/testsuite/g++.dg/pr90773-1a.C
> create mode 100644 gcc/testsuite/g++.dg/pr90773-1b.C
> create mode 100644 gcc/testsuite/g++.dg/pr90773-1c.C
> create mode 100644 gcc/testsuite/g++.dg/pr90773-1d.C
> create mode 100644 gcc/testsuite/gcc.target/i386/pr90773-1.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr90773-10.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr90773-11.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr90773-2.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr90773-3.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr90773-4.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr90773-5.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr90773-6.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr90773-7.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr90773-8.c
> create mode 100644 gcc/testsuite/gcc.target/i386/pr90773-9.c
>
> diff --git a/gcc/common.opt b/gcc/common.opt
> index a75b44ee47e..7f5b38c7810 100644
> --- a/gcc/common.opt
> +++ b/gcc/common.opt
> @@ -2123,6 +2123,25 @@ foptimize-sibling-calls
> Common Var(flag_optimize_sibling_calls) Optimization
> Optimize sibling and tail recursive calls.
>
> +foverlap-op-by-pieces
> +Common RejectNegative Alias(foverlap-op-by-pieces=,on)
> +
> +foverlap-op-by-pieces=
> +Common Joined RejectNegative Enum(overlap_op_by_pieces)
> Var(flag_overlap_op_by_pieces) Init(0)
> +-foverlap-op-by-pieces=[off|on|max-memset] Generate overlapping
> operations between two areas of memory.
> +
> +Enum
> +Name(overlap_op_by_pieces) Type(int)
> +
> +EnumValue
> +Enum(overlap_op_by_pieces) String(off) Value(0)
> +
> +EnumValue
> +Enum(overlap_op_by_pieces) String(on) Value(1)
> +
> +EnumValue
> +Enum(overlap_op_by_pieces) String(max-memset) Value(2)
> +
> fpartial-inlining
> Common Var(flag_partial_inlining) Optimization
> Perform partial inlining.
> diff --git a/gcc/doc/invoke.texi b/gcc/doc/invoke.texi
> index e98b0962b9f..dbdd1095216 100644
> --- a/gcc/doc/invoke.texi
> +++ b/gcc/doc/invoke.texi
> @@ -530,6 +530,7 @@ Objective-C and Objective-C++ Dialects}.
> -fno-sched-spec -fno-signed-zeros @gol
> -fno-toplevel-reorder -fno-trapping-math -fno-zero-initialized-in-bss @gol
> -fomit-frame-pointer -foptimize-sibling-calls @gol
> +-foverlap-op-by-pieces=@r{[}off@r{|}on@r{|}max-memset@r{]} @gol
> -fpartial-inlining -fpeel-loops -fpredictive-commoning @gol
> -fprefetch-loop-arrays @gol
> -fprofile-correction @gol
> @@ -10360,6 +10361,19 @@ their @code{_FORTIFY_SOURCE} counterparts into
> faster alternatives.
>
> Enabled at levels @option{-O2}, @option{-O3}.
>
> +@item -foverlap-op-by-pieces=@r{[}off@r{|}on@r{|}max-memset@r{]}
> +@opindex -foverlap-op-by-pieces
> +The value @code{on} tells the compiler to generate overlapping
> +operations between two areas of memory by using the largest integer
> +operation to minimize number of operations if it is not a stack push.
> +The value @code{max-memset} tells the compiler to generate an
> +overlapping fill with non-zero byte in the maximum single fill size
> +if the last fill size is greater than one. The value @code{off}
> +turns off this optimization.
> +
> +This option is only valid for targets which do not require strict
> +alignment.
> +
> @item -fno-inline
> @opindex fno-inline
> @opindex finline
> diff --git a/gcc/expr.c b/gcc/expr.c
> index a0e19465965..375a5497309 100644
> --- a/gcc/expr.c
> +++ b/gcc/expr.c
> @@ -815,12 +815,27 @@ by_pieces_ninsns (unsigned HOST_WIDE_INT l, unsigned
> int align,
> unsigned int max_size, by_pieces_operation op)
> {
> unsigned HOST_WIDE_INT n_insns = 0;
> + scalar_int_mode mode;
> +
> + if (flag_overlap_op_by_pieces && op != COMPARE_BY_PIECES)
> + {
> + /* NB: Round up L and ALIGN to the widest integer mode for
> + MAX_SIZE. */
> + mode = widest_int_mode_for_size (max_size);
> + if (optab_handler (mov_optab, mode) != CODE_FOR_nothing)
> + {
> + unsigned HOST_WIDE_INT up = ROUND_UP (l, GET_MODE_SIZE (mode));
> + if (up > l)
> + l = up;
> + align = GET_MODE_ALIGNMENT (mode);
> + }
> + }
>
> align = alignment_for_piecewise_move (MOVE_MAX_PIECES, align);
>
> while (max_size > 1 && l > 0)
> {
> - scalar_int_mode mode = widest_int_mode_for_size (max_size);
> + mode = widest_int_mode_for_size (max_size);
> enum insn_code icode;
>
> unsigned int modesize = GET_MODE_SIZE (mode);
> @@ -1041,6 +1056,9 @@ pieces_addr::maybe_postinc (HOST_WIDE_INT size)
>
> class op_by_pieces_d
> {
> + private:
> + scalar_int_mode get_usable_mode (scalar_int_mode mode, unsigned int);
> +
> protected:
> pieces_addr m_to, m_from;
> unsigned HOST_WIDE_INT m_len;
> @@ -1048,6 +1066,10 @@ class op_by_pieces_d
> unsigned int m_align;
> unsigned int m_max_size;
> bool m_reverse;
> + /* True if this is a stash push. */
> + bool m_push;
> + /* True if this memset with non-zero byte. */
> + bool m_non_zero_memset;
>
> /* Virtual functions, overriden by derived classes for the specific
> operation. */
> @@ -1059,7 +1081,7 @@ class op_by_pieces_d
>
> public:
> op_by_pieces_d (rtx, bool, rtx, bool, by_pieces_constfn, void *,
> - unsigned HOST_WIDE_INT, unsigned int);
> + unsigned HOST_WIDE_INT, unsigned int, bool, bool);
> void run ();
> };
>
> @@ -1074,10 +1096,12 @@ op_by_pieces_d::op_by_pieces_d (rtx to, bool to_load,
> by_pieces_constfn from_cfn,
> void *from_cfn_data,
> unsigned HOST_WIDE_INT len,
> - unsigned int align)
> + unsigned int align, bool push,
> + bool non_zero_memset)
> : m_to (to, to_load, NULL, NULL),
> m_from (from, from_load, from_cfn, from_cfn_data),
> - m_len (len), m_max_size (MOVE_MAX_PIECES + 1)
> + m_len (len), m_max_size (MOVE_MAX_PIECES + 1),
> + m_push (push), m_non_zero_memset (non_zero_memset)
> {
> int toi = m_to.get_addr_inc ();
> int fromi = m_from.get_addr_inc ();
> @@ -1108,6 +1132,25 @@ op_by_pieces_d::op_by_pieces_d (rtx to, bool to_load,
> m_align = align;
> }
>
> +/* This function returns the largest usable integer mode for LEN bytes
> + whose size is no bigger than size of MODE. */
> +
> +scalar_int_mode
> +op_by_pieces_d::get_usable_mode (scalar_int_mode mode, unsigned int len)
> +{
> + unsigned int size;
> + do
> + {
> + size = GET_MODE_SIZE (mode);
> + if (len >= size && prepare_mode (mode, m_align))
> + break;
> + /* NB: widest_int_mode_for_size checks SIZE > 1. */
> + mode = widest_int_mode_for_size (size);
> + }
> + while (1);
> + return mode;
> +}
> +
> /* This function contains the main loop used for expanding a block
> operation. First move what we can in the largest integer mode,
> then go to successively smaller modes. For every access, call
> @@ -1116,42 +1159,80 @@ op_by_pieces_d::op_by_pieces_d (rtx to, bool to_load,
> void
> op_by_pieces_d::run ()
> {
> - while (m_max_size > 1 && m_len > 0)
> + if (m_len == 0)
> + return;
> +
> + /* NB: widest_int_mode_for_size checks M_MAX_SIZE > 1. */
> + scalar_int_mode mode = widest_int_mode_for_size (m_max_size);
> + mode = get_usable_mode (mode, m_len);
> +
> + do
> {
> - scalar_int_mode mode = widest_int_mode_for_size (m_max_size);
> + unsigned int size = GET_MODE_SIZE (mode);
> + rtx to1 = NULL_RTX, from1;
>
> - if (prepare_mode (mode, m_align))
> + while (m_len >= size)
> {
> - unsigned int size = GET_MODE_SIZE (mode);
> - rtx to1 = NULL_RTX, from1;
> + if (m_reverse)
> + m_offset -= size;
>
> - while (m_len >= size)
> - {
> - if (m_reverse)
> - m_offset -= size;
> + to1 = m_to.adjust (mode, m_offset);
> + from1 = m_from.adjust (mode, m_offset);
>
> - to1 = m_to.adjust (mode, m_offset);
> - from1 = m_from.adjust (mode, m_offset);
> + m_to.maybe_predec (-(HOST_WIDE_INT)size);
> + m_from.maybe_predec (-(HOST_WIDE_INT)size);
>
> - m_to.maybe_predec (-(HOST_WIDE_INT)size);
> - m_from.maybe_predec (-(HOST_WIDE_INT)size);
> + generate (to1, from1, mode);
>
> - generate (to1, from1, mode);
> + m_to.maybe_postinc (size);
> + m_from.maybe_postinc (size);
>
> - m_to.maybe_postinc (size);
> - m_from.maybe_postinc (size);
> + if (!m_reverse)
> + m_offset += size;
>
> - if (!m_reverse)
> - m_offset += size;
> + m_len -= size;
> + }
>
> - m_len -= size;
> - }
> + finish_mode (mode);
>
> - finish_mode (mode);
> - }
> + if (m_len == 0)
> + return;
>
> - m_max_size = GET_MODE_SIZE (mode);
> + if (!m_push && flag_overlap_op_by_pieces)
> + {
> + /* NB: Generate overlapping operations if it is not a stack
> + push since stack push must not overlap. */
> + if (m_len == 1
> + || !m_non_zero_memset
> + || flag_overlap_op_by_pieces < 2)
> + {
> + /* If the remaining length is 1, this is not memset with
> + non-zero byte or max-memset isn't enabled, get the
> + smallest integer mode for M_LEN bytes. */
> + mode = smallest_int_mode_for_size (m_len * BITS_PER_UNIT);
> + mode = get_usable_mode (mode, GET_MODE_SIZE (mode));
> + }
> + int gap = GET_MODE_SIZE (mode) - m_len;
> + if (gap > 0)
> + {
> + /* If size of MODE > M_LEN, generate the last operation
> + in MODE for the remaining bytes with ovelapping memory
> + from the previois operation. */
> + if (m_reverse)
> + m_offset += gap;
> + else
> + m_offset -= gap;
> + m_len += gap;
> + }
> + }
> + else
> + {
> + /* NB: widest_int_mode_for_size checks SIZE > 1. */
> + mode = widest_int_mode_for_size (size);
> + mode = get_usable_mode (mode, m_len);
> + }
> }
> + while (1);
>
> /* The code above should have handled everything. */
> gcc_assert (!m_len);
> @@ -1160,6 +1241,12 @@ op_by_pieces_d::run ()
> /* Derived class from op_by_pieces_d, providing support for block move
> operations. */
>
> +#ifdef PUSH_ROUNDING
> +#define PUSHG_P(to) ((to) == nullptr)
> +#else
> +#define PUSHG_P(to) false
> +#endif
> +
> class move_by_pieces_d : public op_by_pieces_d
> {
> insn_gen_fn m_gen_fun;
> @@ -1169,7 +1256,8 @@ class move_by_pieces_d : public op_by_pieces_d
> public:
> move_by_pieces_d (rtx to, rtx from, unsigned HOST_WIDE_INT len,
> unsigned int align)
> - : op_by_pieces_d (to, false, from, true, NULL, NULL, len, align)
> + : op_by_pieces_d (to, false, from, true, NULL, NULL, len, align,
> + PUSHG_P (to), false)
> {
> }
> rtx finish_retmode (memop_ret);
> @@ -1263,8 +1351,10 @@ class store_by_pieces_d : public op_by_pieces_d
>
> public:
> store_by_pieces_d (rtx to, by_pieces_constfn cfn, void *cfn_data,
> - unsigned HOST_WIDE_INT len, unsigned int align)
> - : op_by_pieces_d (to, false, NULL_RTX, true, cfn, cfn_data, len, align)
> + unsigned HOST_WIDE_INT len, unsigned int align,
> + bool non_zero_memset)
> + : op_by_pieces_d (to, false, NULL_RTX, true, cfn, cfn_data, len,
> + align, false, non_zero_memset)
> {
> }
> rtx finish_retmode (memop_ret);
> @@ -1411,7 +1501,8 @@ store_by_pieces (rtx to, unsigned HOST_WIDE_INT len,
> memsetp ? SET_BY_PIECES : STORE_BY_PIECES,
> optimize_insn_for_speed_p ()));
>
> - store_by_pieces_d data (to, constfun, constfundata, len, align);
> + store_by_pieces_d data (to, constfun, constfundata, len, align,
> + memsetp);
> data.run ();
>
> if (retmode != RETURN_BEGIN)
> @@ -1438,7 +1529,8 @@ clear_by_pieces (rtx to, unsigned HOST_WIDE_INT len,
> unsigned int align)
> if (len == 0)
> return;
>
> - store_by_pieces_d data (to, clear_by_pieces_1, NULL, len, align);
> + store_by_pieces_d data (to, clear_by_pieces_1, NULL, len, align,
> + false);
> data.run ();
> }
>
> @@ -1460,7 +1552,8 @@ class compare_by_pieces_d : public op_by_pieces_d
> compare_by_pieces_d (rtx op0, rtx op1, by_pieces_constfn op1_cfn,
> void *op1_cfn_data, HOST_WIDE_INT len, int align,
> rtx_code_label *fail_label)
> - : op_by_pieces_d (op0, true, op1, true, op1_cfn, op1_cfn_data, len,
> align)
> + : op_by_pieces_d (op0, true, op1, true, op1_cfn, op1_cfn_data, len,
> + align, false, false)
> {
> m_fail_label = fail_label;
> }
> diff --git a/gcc/testsuite/g++.dg/pr90773-1.h
> b/gcc/testsuite/g++.dg/pr90773-1.h
> new file mode 100644
> index 00000000000..abdb78b078b
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/pr90773-1.h
> @@ -0,0 +1,14 @@
> +class fixed_wide_int_storage {
> +public:
> + long val[10];
> + int len;
> + fixed_wide_int_storage ()
> + {
> + len = sizeof (val) / sizeof (val[0]);
> + for (int i = 0; i < len; i++)
> + val[i] = i;
> + }
> +};
> +
> +extern void foo (fixed_wide_int_storage);
> +extern int record_increment(void);
> diff --git a/gcc/testsuite/g++.dg/pr90773-1a.C
> b/gcc/testsuite/g++.dg/pr90773-1a.C
> new file mode 100644
> index 00000000000..3ab8d929f74
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/pr90773-1a.C
> @@ -0,0 +1,13 @@
> +// { dg-do compile }
> +// { dg-options "-O2" }
> +// { dg-additional-options "-mno-avx -msse2 -mtune=skylake" { target {
> i?86-*-* x86_64-*-* } } }
> +
> +#include "pr90773-1.h"
> +
> +int
> +record_increment(void)
> +{
> + fixed_wide_int_storage x;
> + foo (x);
> + return 0;
> +}
> diff --git a/gcc/testsuite/g++.dg/pr90773-1b.C
> b/gcc/testsuite/g++.dg/pr90773-1b.C
> new file mode 100644
> index 00000000000..9713b2dd612
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/pr90773-1b.C
> @@ -0,0 +1,5 @@
> +// { dg-do compile }
> +// { dg-options "-O2" }
> +// { dg-additional-options "-mno-avx512f -march=skylake" { target { i?86-*-*
> x86_64-*-* } } }
> +
> +#include "pr90773-1a.C"
> diff --git a/gcc/testsuite/g++.dg/pr90773-1c.C
> b/gcc/testsuite/g++.dg/pr90773-1c.C
> new file mode 100644
> index 00000000000..699357a88dc
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/pr90773-1c.C
> @@ -0,0 +1,5 @@
> +// { dg-do compile }
> +// { dg-options "-O2" }
> +// { dg-additional-options "-march=skylake-avx512" { target { i?86-*-*
> x86_64-*-* } } }
> +
> +#include "pr90773-1a.C"
> diff --git a/gcc/testsuite/g++.dg/pr90773-1d.C
> b/gcc/testsuite/g++.dg/pr90773-1d.C
> new file mode 100644
> index 00000000000..bf9d8543c1b
> --- /dev/null
> +++ b/gcc/testsuite/g++.dg/pr90773-1d.C
> @@ -0,0 +1,19 @@
> +// { dg-do run }
> +// { dg-options "-O2" }
> +// { dg-additional-options "-march=native" { target { i?86-*-* x86_64-*-* }
> } }
> +// { dg-additional-sources "pr90773-1a.C" }
> +
> +#include "pr90773-1.h"
> +
> +void
> +foo (fixed_wide_int_storage x)
> +{
> + for (int i = 0; i < x.len; i++)
> + if (x.val[i] != i)
> + __builtin_abort ();
> +}
> +
> +int main ()
> +{
> + return record_increment ();
> +}
> diff --git a/gcc/testsuite/gcc.target/i386/pr90773-1.c
> b/gcc/testsuite/gcc.target/i386/pr90773-1.c
> new file mode 100644
> index 00000000000..86fec27dad0
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr90773-1.c
> @@ -0,0 +1,17 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -mtune=generic -foverlap-op-by-pieces" } */
> +
> +extern char *dst, *src;
> +
> +void
> +foo (void)
> +{
> + __builtin_memcpy (dst, src, 15);
> +}
> +
> +/* { dg-final { scan-assembler-times "movq\[\\t \]+\\(%\[\^,\]+\\)," 1 {
> target { ! ia32 } } } } */
> +/* { dg-final { scan-assembler-times "movq\[\\t \]+7\\(%\[\^,\]+\\)," 1 {
> target { ! ia32 } } } } */
> +/* { dg-final { scan-assembler-times "movl\[\\t \]+\\(%\[\^,\]+\\)," 1 {
> target ia32 } } } */
> +/* { dg-final { scan-assembler-times "movl\[\\t \]+4\\(%\[\^,\]+\\)," 1 {
> target ia32 } } } */
> +/* { dg-final { scan-assembler-times "movl\[\\t \]+8\\(%\[\^,\]+\\)," 1 {
> target ia32 } } } */
> +/* { dg-final { scan-assembler-times "movl\[\\t \]+11\\(%\[\^,\]+\\)," 1 {
> target ia32 } } } */
> diff --git a/gcc/testsuite/gcc.target/i386/pr90773-10.c
> b/gcc/testsuite/gcc.target/i386/pr90773-10.c
> new file mode 100644
> index 00000000000..5985877cc10
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr90773-10.c
> @@ -0,0 +1,13 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -mtune=generic -foverlap-op-by-pieces=max-memset" } */
> +
> +extern char *dst;
> +
> +void
> +foo (int c)
> +{
> + __builtin_memset (dst, c, 5);
> +}
> +
> +/* { dg-final { scan-assembler-times "movl\[\\t \]+.+, \\(%\[\^,\]+\\)" 1 }
> } */
> +/* { dg-final { scan-assembler-times "movb\[\\t \]+.+, 4\\(%\[\^,\]+\\)" 1 }
> } */
> diff --git a/gcc/testsuite/gcc.target/i386/pr90773-11.c
> b/gcc/testsuite/gcc.target/i386/pr90773-11.c
> new file mode 100644
> index 00000000000..9bf57aa3a44
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr90773-11.c
> @@ -0,0 +1,13 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -mtune=generic -foverlap-op-by-pieces=max-memset" } */
> +
> +extern char *dst;
> +
> +void
> +foo (int c)
> +{
> + __builtin_memset (dst, c, 6);
> +}
> +
> +/* { dg-final { scan-assembler-times "movl\[\\t \]+.+, \\(%\[\^,\]+\\)" 1 }
> } */
> +/* { dg-final { scan-assembler-times "movl\[\\t \]+.+, 2\\(%\[\^,\]+\\)" 1 }
> } */
> diff --git a/gcc/testsuite/gcc.target/i386/pr90773-2.c
> b/gcc/testsuite/gcc.target/i386/pr90773-2.c
> new file mode 100644
> index 00000000000..ebdf9dac6e8
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr90773-2.c
> @@ -0,0 +1,20 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -mtune=generic -foverlap-op-by-pieces" } */
> +/* { dg-additional-options "-mno-avx -msse2" { target { ! ia32 } } } */
> +/* { dg-additional-options "-mno-sse" { target ia32 } } */
> +
> +extern char *dst, *src;
> +
> +void
> +foo (void)
> +{
> + __builtin_memcpy (dst, src, 19);
> +}
> +
> +/* { dg-final { scan-assembler-times "movdqu\[\\t \]+\\(%\[\^,\]+\\)," 1 {
> target { ! ia32 } } } } */
> +/* { dg-final { scan-assembler-times "movl\[\\t \]+15\\(%\[\^,\]+\\)," 1 {
> target { ! ia32 } } } } */
> +/* { dg-final { scan-assembler-times "movl\[\\t \]+\\(%\[\^,\]+\\)," 1 {
> target ia32 } } } */
> +/* { dg-final { scan-assembler-times "movl\[\\t \]+4\\(%\[\^,\]+\\)," 1 {
> target ia32 } } } */
> +/* { dg-final { scan-assembler-times "movl\[\\t \]+8\\(%\[\^,\]+\\)," 1 {
> target ia32 } } } */
> +/* { dg-final { scan-assembler-times "movl\[\\t \]+12\\(%\[\^,\]+\\)," 1 {
> target ia32 } } } */
> +/* { dg-final { scan-assembler-times "movl\[\\t \]+15\\(%\[\^,\]+\\)," 1 {
> target ia32 } } } */
> diff --git a/gcc/testsuite/gcc.target/i386/pr90773-3.c
> b/gcc/testsuite/gcc.target/i386/pr90773-3.c
> new file mode 100644
> index 00000000000..d876f878f60
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr90773-3.c
> @@ -0,0 +1,23 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -mtune=generic -foverlap-op-by-pieces" } */
> +/* { dg-additional-options "-mno-avx -msse2" { target { ! ia32 } } } */
> +/* { dg-additional-options "-mno-sse" { target ia32 } } */
> +
> +extern char *dst, *src;
> +
> +void
> +foo (void)
> +{
> + __builtin_memcpy (dst, src, 31);
> +}
> +
> +/* { dg-final { scan-assembler-times "movdqu\[\\t \]+\\(%\[\^,\]+\\)," 1 {
> target { ! ia32 } } } } */
> +/* { dg-final { scan-assembler-times "movdqu\[\\t \]+15\\(%\[\^,\]+\\)," 1 {
> target { ! ia32 } } } } */
> +/* { dg-final { scan-assembler-times "movl\[\\t \]+\\(%\[\^,\]+\\)," 1 {
> target ia32 } } } */
> +/* { dg-final { scan-assembler-times "movl\[\\t \]+4\\(%\[\^,\]+\\)," 1 {
> target ia32 } } } */
> +/* { dg-final { scan-assembler-times "movl\[\\t \]+8\\(%\[\^,\]+\\)," 1 {
> target ia32 } } } */
> +/* { dg-final { scan-assembler-times "movl\[\\t \]+12\\(%\[\^,\]+\\)," 1 {
> target ia32 } } } */
> +/* { dg-final { scan-assembler-times "movl\[\\t \]+16\\(%\[\^,\]+\\)," 1 {
> target ia32 } } } */
> +/* { dg-final { scan-assembler-times "movl\[\\t \]+20\\(%\[\^,\]+\\)," 1 {
> target ia32 } } } */
> +/* { dg-final { scan-assembler-times "movl\[\\t \]+24\\(%\[\^,\]+\\)," 1 {
> target ia32 } } } */
> +/* { dg-final { scan-assembler-times "movl\[\\t \]+27\\(%\[\^,\]+\\)," 1 {
> target ia32 } } } */
> diff --git a/gcc/testsuite/gcc.target/i386/pr90773-4.c
> b/gcc/testsuite/gcc.target/i386/pr90773-4.c
> new file mode 100644
> index 00000000000..0df1b2fc247
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr90773-4.c
> @@ -0,0 +1,13 @@
> +/* { dg-do compile { target { ! ia32 } } } */
> +/* { dg-options "-O2 -mno-avx -msse2 -mtune=generic -foverlap-op-by-pieces"
> } */
> +
> +extern char *dst;
> +
> +void
> +foo (void)
> +{
> + __builtin_memset (dst, 0, 31);
> +}
> +
> +/* { dg-final { scan-assembler-times "movups\[\\t \]+%xmm\[0-9\]+,
> \\(%\[\^,\]+\\)" 1 } } */
> +/* { dg-final { scan-assembler-times "movups\[\\t \]+%xmm\[0-9\]+,
> 15\\(%\[\^,\]+\\)" 1 } } */
> diff --git a/gcc/testsuite/gcc.target/i386/pr90773-5.c
> b/gcc/testsuite/gcc.target/i386/pr90773-5.c
> new file mode 100644
> index 00000000000..65c9fe88696
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr90773-5.c
> @@ -0,0 +1,13 @@
> +/* { dg-do compile { target { ! ia32 } } } */
> +/* { dg-options "-O2 -mno-avx -msse2 -mtune=generic -foverlap-op-by-pieces"
> } */
> +
> +extern char *dst;
> +
> +void
> +foo (void)
> +{
> + __builtin_memset (dst, 0, 21);
> +}
> +
> +/* { dg-final { scan-assembler-times "movups\[\\t \]+%xmm\[0-9\]+,
> \\(%\[\^,\]+\\)" 1 } } */
> +/* { dg-final { scan-assembler-times "movq\[\\t \]+\\\$0+,
> 13\\(%\[\^,\]+\\)" 1 } } */
> diff --git a/gcc/testsuite/gcc.target/i386/pr90773-6.c
> b/gcc/testsuite/gcc.target/i386/pr90773-6.c
> new file mode 100644
> index 00000000000..0c84d492974
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr90773-6.c
> @@ -0,0 +1,11 @@
> +/* { dg-do compile { target { ! ia32 } } } */
> +/* { dg-options "-O2 -mno-avx -msse2 -mtune=generic -foverlap-op-by-pieces"
> } */
> +
> +void
> +foo (char *dst, char *src)
> +{
> + __builtin_memcpy (dst, src, 255);
> +}
> +
> +/* { dg-final { scan-assembler-times "movdqu\[\\t
> \]+\[0-9\]*\\(%\[\^,\]+\\)," 16 } } */
> +/* { dg-final { scan-assembler-not "mov\[bwlq\]" } } */
> diff --git a/gcc/testsuite/gcc.target/i386/pr90773-7.c
> b/gcc/testsuite/gcc.target/i386/pr90773-7.c
> new file mode 100644
> index 00000000000..732b4d3d992
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr90773-7.c
> @@ -0,0 +1,11 @@
> +/* { dg-do compile { target { ! ia32 } } } */
> +/* { dg-options "-O2 -mno-avx -msse2 -mtune=skylake -foverlap-op-by-pieces"
> } */
> +
> +void
> +foo (char *dst)
> +{
> + __builtin_memset (dst, 0, 255);
> +}
> +
> +/* { dg-final { scan-assembler-times "movups\[\\t \]+%xmm\[0-9\]+,
> \[0-9\]*\\(%\[\^,\]+\\)" 16 } } */
> +/* { dg-final { scan-assembler-not "mov\[bwlq\]" } } */
> diff --git a/gcc/testsuite/gcc.target/i386/pr90773-8.c
> b/gcc/testsuite/gcc.target/i386/pr90773-8.c
> new file mode 100644
> index 00000000000..7ff5ba12daf
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr90773-8.c
> @@ -0,0 +1,13 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -mtune=generic -foverlap-op-by-pieces=max-memset" } */
> +
> +extern char *dst;
> +
> +void
> +foo (void)
> +{
> + __builtin_memset (dst, 0, 5);
> +}
> +
> +/* { dg-final { scan-assembler-times "movl\[\\t \]+.+, \\(%\[\^,\]+\\)" 1 }
> } */
> +/* { dg-final { scan-assembler-times "movb\[\\t \]+.+, 4\\(%\[\^,\]+\\)" 1 }
> } */
> diff --git a/gcc/testsuite/gcc.target/i386/pr90773-9.c
> b/gcc/testsuite/gcc.target/i386/pr90773-9.c
> new file mode 100644
> index 00000000000..c2fc3ba59a7
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/i386/pr90773-9.c
> @@ -0,0 +1,13 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -mtune=generic -foverlap-op-by-pieces=max-memset" } */
> +
> +extern char *dst;
> +
> +void
> +foo (void)
> +{
> + __builtin_memset (dst, 0, 6);
> +}
> +
> +/* { dg-final { scan-assembler-times "movl\[\\t \]+.+, \\(%\[\^,\]+\\)" 1 }
> } */
> +/* { dg-final { scan-assembler-times "movw\[\\t \]+.+, 4\\(%\[\^,\]+\\)" 1 }
> } */
> diff --git a/gcc/toplev.c b/gcc/toplev.c
> index d8cc254adef..23c88c788a2 100644
> --- a/gcc/toplev.c
> +++ b/gcc/toplev.c
> @@ -1323,6 +1323,14 @@ process_options (void)
> }
> }
>
> + if (flag_overlap_op_by_pieces && STRICT_ALIGNMENT)
> + {
> + error_at (UNKNOWN_LOCATION,
> + "%<-foverlap-op-by-pieces%> is not supported for "
> + "strict alignment target");
> + flag_overlap_op_by_pieces = 0;
> + }
> +
> /* One region RA really helps to decrease the code size. */
> if (flag_ira_region == IRA_REGION_AUTODETECT)
> flag_ira_region
> --
> 2.30.2
>
