From: Richard Biener <rguenther@suse.de>
To: gcc-patches@gcc.gnu.org
Cc: kyrylo.tkachov@arm.com
Subject: [PATCH] Allow non-NULL offset for store-merging bases
Date: Wed, 02 Nov 2016 13:15:00 -0000 [thread overview]
Message-ID: <alpine.LSU.2.11.1611021413321.5294@t29.fhfr.qr> (raw)
The following teaches store-merging to handle non-NULL offset if the
base is already addressable (otherwise introducing new pointers to
a non-addressable base invalidates points-to information, see a comment
in the patch how we could avoid this in theory).
Bootstrap and regtest running on x86_64-unknown-linux-gnu.
Richard.
2016-11-02 Richard Biener <rguenther@suse.de>
* gimple-ssa-store-merging.c: Include gimplify-me.h.
(imm_store_chain_info::output_merged_stores): Force base_addr
to be proper GIMPLE for a MEM_REF address.
(pass_store_merging::execute): Restrict negative bitpos
handling to non-MEM_REF bases. Remove TREE_THIS_VOLATILE
check. Take into account non-NULL_TREE offset if the base
is already addressable.
* gcc.dg/store_merging_8.c: New testcase.
Index: gcc/gimple-ssa-store-merging.c
===================================================================
--- gcc/gimple-ssa-store-merging.c (revision 241789)
+++ gcc/gimple-ssa-store-merging.c (working copy)
@@ -125,6 +125,7 @@
#include "tree-cfg.h"
#include "tree-eh.h"
#include "target.h"
+#include "gimplify-me.h"
/* The maximum size (in bits) of the stores this pass should generate. */
#define MAX_STORE_BITSIZE (BITS_PER_WORD)
@@ -1127,6 +1128,8 @@ imm_store_chain_info::output_merged_stor
unsigned int i;
bool fail = false;
+ tree addr = force_gimple_operand_1 (unshare_expr (base_addr), &seq,
+ is_gimple_mem_ref_addr, NULL_TREE);
FOR_EACH_VEC_ELT (split_stores, i, split_store)
{
unsigned HOST_WIDE_INT try_size = split_store->size;
@@ -1137,7 +1140,7 @@ imm_store_chain_info::output_merged_stor
tree int_type = build_nonstandard_integer_type (try_size, UNSIGNED);
int_type = build_aligned_type (int_type, align);
- tree dest = fold_build2 (MEM_REF, int_type, base_addr,
+ tree dest = fold_build2 (MEM_REF, int_type, addr,
build_int_cst (offset_type, try_pos));
tree src = native_interpret_expr (int_type,
@@ -1366,15 +1369,10 @@ pass_store_merging::execute (function *f
&unsignedp, &reversep, &volatilep);
/* As a future enhancement we could handle stores with the same
base and offset. */
- bool invalid = offset || reversep || bitpos < 0
+ bool invalid = reversep
|| ((bitsize > MAX_BITSIZE_MODE_ANY_INT)
&& (TREE_CODE (rhs) != INTEGER_CST))
- || !rhs_valid_for_store_merging_p (rhs)
- /* An access may not be volatile itself but base_addr may be
- a volatile decl i.e. MEM[&volatile-decl]. The hashing for
- tree_operand_hash won't consider such stores equal to each
- other so we can't track chains on them. */
- || TREE_THIS_VOLATILE (base_addr);
+ || !rhs_valid_for_store_merging_p (rhs);
/* We do not want to rewrite TARGET_MEM_REFs. */
if (TREE_CODE (base_addr) == TARGET_MEM_REF)
@@ -1398,7 +1396,32 @@ pass_store_merging::execute (function *f
/* get_inner_reference returns the base object, get at its
address now. */
else
- base_addr = build_fold_addr_expr (base_addr);
+ {
+ if (bitpos < 0)
+ invalid = true;
+ base_addr = build_fold_addr_expr (base_addr);
+ }
+
+ if (! invalid
+ && offset != NULL_TREE)
+ {
+ /* If the access is variable offset then a base
+ decl has to be address-taken to be able to
+ emit pointer-based stores to it.
+ ??? We might be able to get away with
+ re-using the original base up to the first
+ variable part and then wrapping that inside
+ a BIT_FIELD_REF. */
+ tree base = get_base_address (base_addr);
+ if (! base
+ || (DECL_P (base)
+ && ! TREE_ADDRESSABLE (base)))
+ invalid = true;
+ else
+ base_addr = build2 (POINTER_PLUS_EXPR,
+ TREE_TYPE (base_addr),
+ base_addr, offset);
+ }
struct imm_store_chain_info **chain_info
= m_stores.get (base_addr);
Index: gcc/testsuite/gcc.dg/store_merging_8.c
===================================================================
--- gcc/testsuite/gcc.dg/store_merging_8.c (revision 0)
+++ gcc/testsuite/gcc.dg/store_merging_8.c (working copy)
@@ -0,0 +1,38 @@
+/* { dg-do compile } */
+/* { dg-require-effective-target non_strict_align } */
+/* { dg-options "-O2 -fdump-tree-store-merging" } */
+
+struct baz {
+ struct bar {
+ int a;
+ char b;
+ char c;
+ char d;
+ char e;
+ char f;
+ char g;
+ } a[4];
+} x;
+struct baz *xx = &x;
+
+void
+foo1 (int i)
+{
+ x.a[i].b = 0;
+ x.a[i].a = 0;
+ x.a[i].c = 0;
+ x.a[i].d = 0;
+ x.a[i].e = 0;
+}
+
+void
+foo2 (int i)
+{
+ x.a[i].b = 0;
+ x.a[i].a = 0;
+ x.a[i].c = 1;
+ x.a[i].d = 0;
+ x.a[i].e = 0;
+}
+
+/* { dg-final { scan-tree-dump-times "Merging successful" 2 "store-merging" } } */
reply other threads:[~2016-11-02 13:15 UTC|newest]
Thread overview: [no followups] expand[flat|nested] mbox.gz Atom feed
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=alpine.LSU.2.11.1611021413321.5294@t29.fhfr.qr \
--to=rguenther@suse.de \
--cc=gcc-patches@gcc.gnu.org \
--cc=kyrylo.tkachov@arm.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).