public inbox for gcc-patches@gcc.gnu.org
 help / color / mirror / Atom feed
* [PATCH] Alias analysis of zero sized arrays
@ 2017-04-26  6:34 Steve Ellcey
  2017-04-26 10:05 ` Richard Biener
  0 siblings, 1 reply; 6+ messages in thread
From: Steve Ellcey @ 2017-04-26  6:34 UTC (permalink / raw)
  To: gcc-patches; +Cc: Richard Biener

[-- Attachment #1: Type: text/plain, Size: 1255 bytes --]

This patch changes get_ref_base_and_extent to treat zero sized arrays
like C99 flexible arrays and assume that references to zero sized
arrays can also be made beyond the 'end' of the array.  C99 flexible
arrays are recognized by not having an INTEGER_CST limit/size and the
routine then sets maxsize to -1 for special handling.  This patch
checks for a value of 0 when we do have an INTEGER_CST limit/size and
handles it like a flexible array.

Tested with a bootstrap on aarch64 and a GCC test run with no
regressions.

I did not create a testcase because the test I have (attached) is not
runnable.  If you compile this test case for aarch64 with
'-UFLEX -O2 -fno-strict-aliasing' you will get two loads, then two
stores in the main loop.  In the original large program that this came
from, that generated bad code.  If compiled with -DFLEX instead of
-UFLEX, you get load, store, load, store and that was working.  With
this patch, both versions (-UFLEX and -DFLEX) generate the load, store,
load, store sequence.

OK for checkin?

Steve Ellcey
sellcey@cavium.com


2017-04-25  Steve Ellcey  <sellcey@cavium.com>

	* tree-dfa.c (get_ref_base_and_extent): Treat zero size array like
	a C99 flexible array.


[-- Attachment #2: alias.patch --]
[-- Type: text/x-patch, Size: 977 bytes --]

diff --git a/gcc/tree-dfa.c b/gcc/tree-dfa.c
index 8ee46dc..79c3489 100644
--- a/gcc/tree-dfa.c
+++ b/gcc/tree-dfa.c
@@ -438,6 +438,10 @@ get_ref_base_and_extent (tree exp, HOST_WIDE_INT *poffset,
       && TREE_CODE (size_tree) == INTEGER_CST)
     bitsize = wi::to_offset (size_tree);
 
+  /* If zero sized array, treat it like a flexible array.  */
+  if (bitsize == 0)
+    bitsize = -1;
+
   /* Initially, maxsize is the same as the accessed element size.
      In the following it will only grow (or become -1).  */
   maxsize = bitsize;
@@ -504,7 +508,13 @@ get_ref_base_and_extent (tree exp, HOST_WIDE_INT *poffset,
 		if (maxsize != -1
 		    && csize
 		    && TREE_CODE (csize) == INTEGER_CST)
-		  maxsize = wi::to_offset (csize) - bit_offset;
+		  {
+		    maxsize = wi::to_offset (csize) - bit_offset;
+
+		    /* If zero sized array, treat it like a flexible array.  */
+		    if (maxsize == 0)
+		      maxsize = -1;
+		  }
 		else
 		  maxsize = -1;
 	      }

[-- Attachment #3: test.c --]
[-- Type: text/x-csrc, Size: 430 bytes --]

struct q {
	int b;
};
struct r {
   int n;
   struct q slot[0];
};
struct s {
   int n;
#ifdef FLEX
 long int o[];
#else
 long int o[0];
#endif
};
extern int x, y, m;
extern struct s *a;
extern struct r *b;
extern void bar();
int foo() {
   int i,j;
   for (i = 0; i < m; i++) {
   	a->o[i] = sizeof(*a);
   	b = ((struct r *)(((char *)a) + a->o[a->n]));
	for (j = 0; j < 10; j++) {
		b->slot[j].b = 0;
   	}
        bar();
  }
}

^ permalink raw reply	[flat|nested] 6+ messages in thread
end of thread, other threads:[~2017-04-26 19:09 UTC | newest]

Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2017-04-26  6:34 [PATCH] Alias analysis of zero sized arrays Steve Ellcey
2017-04-26 10:05 ` Richard Biener
2017-04-26 10:07   ` Richard Biener
2017-04-26 19:09     ` Steve Ellcey
2017-04-26 20:08       ` Richard Biener
2017-04-26 21:59         ` Steve Ellcey

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).