[PATCH, i386]: Fix PR 66648, incorrect memcpy expansion with unrolled_loop strategy
Uros Bizjak
ubizjak@gmail.com
Fri Jun 26 12:08:00 GMT 2015
Hello!
As explained in the PR [1], we have to emit check for main loop
execution UNLESS we guarantee that min_size is at least equal to
size_needed. In this testcase, VRP declared minimum copying size, and
after unaligned prologue adjustment, min_size was still non-zero. The
!min_size check didn't account for the case that min_size can be
non-zero, but still less than size_needed.
2015-06-26 Uros Bizjak <ubizjak@gmail.com>
PR target/66648
* config/i386/i386.c (ix86_expand_set_or_movmem): Emit main loop
execution guard when min_size is less than size_needed.
testsuite/ChangeLog:
2015-06-26 Uros Bizjak <ubizjak@gmail.com>
PR target/66648
* gcc.target/i386/pr66648.c: New test.
Bootstrapped and regression tested on x86_64-linux-gnu {,-m32}.
Honza, does the patch look OK to you?
[1] https://gcc.gnu.org/bugzilla/show_bug.cgi?id=66648
Uros.
-------------- next part --------------
Index: config/i386/i386.c
===================================================================
--- config/i386/i386.c (revision 224993)
+++ config/i386/i386.c (working copy)
@@ -25008,7 +25008,8 @@ ix86_expand_set_or_movmem (rtx dst, rtx src, rtx c
dst = change_address (dst, BLKmode, destreg);
set_mem_align (dst, desired_align * BITS_PER_UNIT);
epilogue_size_needed = 0;
- if (need_zero_guard && !min_size)
+ if (need_zero_guard
+ && min_size < (unsigned HOST_WIDE_INT) size_needed)
{
/* It is possible that we copied enough so the main loop will not
execute. */
@@ -25140,7 +25141,7 @@ ix86_expand_set_or_movmem (rtx dst, rtx src, rtx c
max_size -= align_bytes;
}
if (need_zero_guard
- && !min_size
+ && min_size < (unsigned HOST_WIDE_INT) size_needed
&& (count < (unsigned HOST_WIDE_INT) size_needed
|| (align_bytes == 0
&& count < ((unsigned HOST_WIDE_INT) size_needed
Index: testsuite/gcc.target/pr66648.c
===================================================================
--- testsuite/gcc.target/pr66648.c (revision 0)
+++ testsuite/gcc.target/pr66648.c (working copy)
@@ -0,0 +1,33 @@
+/* { dg-do run } */
+/* { dg-options "-O2 -mstringop-strategy=unrolled_loop -mtune=nocona" } */
+
+#define PATTERN 0xdeadbeef
+#define SIZE 32
+
+struct S { int i; char str[SIZE]; int j; };
+
+void __attribute__((noclone, noinline))
+my_memcpy (char *, const char *, unsigned int);
+
+void
+my_memcpy (char *dst, const char *src, unsigned int len)
+{
+ if (len < 8)
+ __builtin_abort ();
+
+ __builtin_memcpy (dst, src, len);
+}
+
+int
+main (void)
+{
+ const char str[SIZE]= "1234567890123456789012345678901";
+ struct S *s = __builtin_malloc (sizeof (struct S));
+
+ s->j = PATTERN;
+ my_memcpy (s->str, str, SIZE);
+ if (s->j != PATTERN)
+ __builtin_abort ();
+
+ return 0;
+}
More information about the Gcc-patches
mailing list