[PATCH][v5] GIMPLE store merging pass

[Kyrill Tkachov]

Hi all,

This is another revision of the pass addressing Richard's feedback [1]
I believe I've addressed all of it and added more comments to the code where
needed.

The output_merged_store function now uses the new split_group helper to break
up the merged store into multiple regular-sized stores.

The apply_stores function that splats the stores in a group together can now
return a bool to indicate failure and is used to reject quickly one-store groups
and other store groups that we cannot output.

One thing I've been struggling with is reimplementing encode_tree_to_bitpos,
the function that applies a tree constant to the merged byte array.
I've tried to reimplement it by writing the constant to a byte array with native_encode_expr
and manipulating the bytes directly to insert them into the appropriate bit position without
constructing an intermediate wide_int.  This works, but only for little-endian.
On big-endian it generated wrong code.
So this patch doesn't include that implementation but rather uses the previous one that uses
a wide_int but is correct on both endiannesses.
Richard, I am sending out a patch that implements the cheaper algorithm separately if you
want to help debug it.

This has been bootstrapped and tested on arm, aarch64, aarch64_be, x86_64.
Besides the encode_tree_to_bitpos reimplementation (which will have its own thread)
does this version look good?

Thanks,
Kyrill

[1] https://gcc.gnu.org/ml/gcc-patches/2016-09/msg02225.html

2016-10-10  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>

     PR middle-end/22141
     * Makefile.in (OBJS): Add gimple-ssa-store-merging.o.
     * common.opt (fstore-merging): New Optimization option.
     * opts.c (default_options_table): Add entry for
     OPT_ftree_store_merging.
     * fold-const.h (can_native_encode_type_p): Declare prototype.
     * fold-const.c (can_native_encode_type_p): Define.
     * params.def (PARAM_STORE_MERGING_ALLOW_UNALIGNED): Define.
     * passes.def: Insert pass_tree_store_merging.
     * tree-pass.h (make_pass_store_merging): Declare extern
     prototype.
     * gimple-ssa-store-merging.c: New file.
     * doc/invoke.texi (Optimization Options): Document
     -fstore-merging.

2016-10-10  Kyrylo Tkachov  <kyrylo.tkachov@arm.com>
             Jakub Jelinek  <jakub@redhat.com>
             Andrew Pinski  <pinskia@gmail.com>

     PR middle-end/22141
     PR rtl-optimization/23684
     * gcc.c-torture/execute/pr22141-1.c: New test.
     * gcc.c-torture/execute/pr22141-2.c: Likewise.
     * gcc.target/aarch64/ldp_stp_1.c: Adjust for -fstore-merging.
     * gcc.target/aarch64/ldp_stp_4.c: Likewise.
     * gcc.dg/store_merging_1.c: New test.
     * gcc.dg/store_merging_2.c: Likewise.
     * gcc.dg/store_merging_3.c: Likewise.
     * gcc.dg/store_merging_4.c: Likewise.
     * gcc.dg/store_merging_5.c: Likewise.
     * gcc.dg/store_merging_6.c: Likewise.
     * gcc.dg/store_merging_7.c: Likewise.
     * gcc.target/i386/pr22141.c: Likewise.
     * gcc.target/i386/pr34012.c: Add -fno-store-merging to dg-options.
     * g++.dg/init/new17.C: Likewise.
-------------- next part --------------
A non-text attachment was scrubbed...
Name: store-merging.patch
Type: text/x-patch
Size: 62491 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/gcc-patches/attachments/20161010/ddd63afc/attachment.bin>