wcsmbs: Add wcscpy loop unroll option

This allows an architecture to use the old generic implementation
and also set explicit loop unrolling.

Checked on aarch64-linux-gnu.

	* include/loop_unroll.h: New file.
	* wcsmbs/wcscpy (__wcscpy): Add option to use loop unrolling
	besides generic implementation.

ChangeLog

@@ -1,3 +1,9 @@
+2019-04-04  Adhemerval Zanella  <adhemerval.zanella@linaro.org>
+
+	* include/loop_unroll.h: New file.
+	* wcsmbs/wcscpy (__wcscpy): Add option to use loop unrolling
+	besides generic implementation.
+
 2019-04-03  DJ Delorie  <dj@redhat.com>
 
 	* time/tst-strftime3.c (tm_to_printed): Disable warning about

include/loop_unroll.h

@@ -0,0 +1,78 @@
+/* Macro for explicit loop unrolling.
+   Copyright (C) 2019 Free Software Foundation, Inc.
+   This file is part of the GNU C Library.
+
+   The GNU C Library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2.1 of the License, or (at your option) any later version.
+
+   The GNU C Library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with the GNU C Library; if not, see
+   <http://www.gnu.org/licenses/>.  */
+
+#ifndef _LOOP_UNROLL_H
+#define _LOOP_UNROLL_H
+
+/* Loop unroll macro to be used for explicit force loop unrolling with a
+   configurable number of iterations.  The idea is to make the loop unrolling
+   independent of whether the compiler is able to unroll through specific
+   optimizations options (-funroll-loops or -funroll-all-loops).
+
+   For instance, to implement strcpy with SRC being the source input and
+   DEST the destination buffer, it is expected the macro to be used in this
+   way:
+
+     #define ITERATION(index)	\
+       ({ char c = *str++; *dest++ = c; c != '\0' })
+
+     while (1)
+       UNROLL_REPEAT (4, ITERATION)
+
+   The loop will be manually unrolled 4 times.  Another option is to do
+   the index update after the tests:
+
+     #define ITERATION(index)	\
+       ({ char c = *(str + index); *(dest + index) = c; c != '\0' })
+     #define UPDATE(n)		\
+       str += n; dst += n
+
+     while (1)
+       UNROLL_REPEAT_UPDATE (4, ITERATION, UPDATE)
+
+   The loop will be manually unrolled 4 times and the SRC and DEST pointers
+   will be updated only after the last iteration.
+
+   Currently, both macros unroll the loop 8 times at maximum.  */
+
+#define UNROLL_REPEAT_1(X)    if (!X(0)) break;
+#define UNROLL_REPEAT_2(X)    UNROLL_REPEAT_1 (X) if (!X (1)) break;
+#define UNROLL_REPEAT_3(X)    UNROLL_REPEAT_2 (X) if (!X (2)) break;
+#define UNROLL_REPEAT_4(X)    UNROLL_REPEAT_3 (X) if (!X (3)) break;
+#define UNROLL_REPEAT_5(X)    UNROLL_REPEAT_4 (X) if (!X (4)) break;
+#define UNROLL_REPEAT_6(X)    UNROLL_REPEAT_5 (X) if (!X (5)) break;
+#define UNROLL_REPEAT_7(X)    UNROLL_REPEAT_6 (X) if (!X (6)) break;
+#define UNROLL_REPEAT_8(X)    UNROLL_REPEAT_7 (X) if (!X (7)) break;
+
+#define UNROLL_EXPAND(...)    __VA_ARGS__
+
+#define UNROLL_REPEAT__(N, X) UNROLL_EXPAND(UNROLL_REPEAT_ ## N) (X)
+#define UNROLL_REPEAT_(N, X)  UNROLL_REPEAT__ (N, X)
+
+#define UNROLL_REPEAT(N, X)                \
+  (void) ({                                \
+    UNROLL_REPEAT_ (UNROLL_EXPAND(N), X);  \
+  })
+
+#define UNROLL_REPEAT_UPDATE(N, X, U)      \
+  (void) ({                                \
+    UNROLL_REPEAT_ (UNROLL_EXPAND(N), X);  \
+    UPDATE (N);                            \
+  })
+
+#endif

wcsmbs/wcscpy.c

@@ -17,6 +17,7 @@
    <http://www.gnu.org/licenses/>.  */
 
 #include <wchar.h>
+#include <loop_unroll.h>
 
 
 #ifdef WCSCPY
@@ -27,7 +28,25 @@
 wchar_t *
 __wcscpy (wchar_t *dest, const wchar_t *src)
 {
+#ifndef UNROLL_NTIMES
   return __wmemcpy (dest, src, __wcslen (src) + 1);
+#else
+  /* Some architectures might have costly tail function call (powerpc
+     for instance) where wmemcpy call overhead for smalls sizes might
+     be more costly than just unrolling the main loop.  */
+  wchar_t *wcp = dest;
+
+#define ITERATION(index)		\
+  ({					\
+     wchar_t c = *src++;		\
+     *wcp++ = c;			\
+     c != L'\0';			\
+  })
+
+  while (1)
+    UNROLL_REPEAT(UNROLL_NTIMES, ITERATION);
+  return dest;
+#endif
 }
 #ifndef WCSCPY
 weak_alias (__wcscpy, wcscpy)