[PATCH][AArch64] NEON-optimised memchr

[PATCH][AArch64] NEON-optimised memchr

[Kévin PETIT]

Hi,

Here’s an AArch64 NEON-optimised version of memchr. It has been tested on a
Cortex-A53 and Cortex-A57 and is on these cores vastly faster than all other
C
implementations I’ve found when searching more than a few bytes.

2014-07-10  Kévin Petit  <kevin.petit@arm.com>

	* libc/machine/aarch64/memchr.S: New file.
	* libc/machine/aarch64/memchr-stub.c: New file.
	* libc/machine/aarch64/Makefile.am: Add the new files.
	* libc/machine/aarch64/Makefile.in: Regenerated.

Regards,
Kévin

Re: [PATCH][AArch64] NEON-optimised memchr

[Richard Earnshaw]

On 10/07/14 17:55, Kevin Petit wrote:
> Hi,
> 
> Here’s an AArch64 NEON-optimised version of memchr. It has been tested on a
> Cortex-A53 and Cortex-A57 and is on these cores vastly faster than all other
> C
> implementations I’ve found when searching more than a few bytes.
> 
> 2014-07-10  Kévin Petit  <kevin.petit@arm.com>
> 
> 	* libc/machine/aarch64/memchr.S: New file.
> 	* libc/machine/aarch64/memchr-stub.c: New file.
> 	* libc/machine/aarch64/Makefile.am: Add the new files.
> 	* libc/machine/aarch64/Makefile.in: Regenerated.
> 
> Regards,
> Kévin
> 
> 
> 

ENOPATCH?

R.

RE: [PATCH][AArch64] NEON-optimised memchr

[Kévin PETIT]

[-- Attachment #1: Type: text/plain, Size: 139 bytes --]

Hi Richard,

> ENOPATCH?

Thanks for pointing it out. I hate when I do this.

Apologies everyone, here it is.

Regards,
Kévin


[-- Attachment #2: neon-memchr.patch --]
[-- Type: application/octet-stream, Size: 11436 bytes --]

commit 4fef5d546e9d2ab0f5a3f0a6c6b60538ee16347b
Author: Kévin PETIT <kevin.petit@arm.com>
Date:   Thu Jul 10 16:48:42 2014 +0100

    NEON optimised memchr

diff --git a/newlib/libc/machine/aarch64/Makefile.am b/newlib/libc/machine/aarch64/Makefile.am
index 43946ba..b9fa7cb 100644
--- a/newlib/libc/machine/aarch64/Makefile.am
+++ b/newlib/libc/machine/aarch64/Makefile.am
@@ -9,6 +9,8 @@ AM_CCASFLAGS = $(INCLUDES)
 noinst_LIBRARIES = lib.a
 
 lib_a_SOURCES =
+lib_a_SOURCES += memchr-stub.c
+lib_a_SOURCES += memchr.S
 lib_a_SOURCES += memcmp-stub.c
 lib_a_SOURCES += memcmp.S
 lib_a_SOURCES += memcpy-stub.c
diff --git a/newlib/libc/machine/aarch64/Makefile.in b/newlib/libc/machine/aarch64/Makefile.in
index d1a15f1..fd37695 100644
--- a/newlib/libc/machine/aarch64/Makefile.in
+++ b/newlib/libc/machine/aarch64/Makefile.in
@@ -69,7 +69,8 @@ LIBRARIES = $(noinst_LIBRARIES)
 ARFLAGS = cru
 lib_a_AR = $(AR) $(ARFLAGS)
 lib_a_LIBADD =
-am_lib_a_OBJECTS = lib_a-memcmp-stub.$(OBJEXT) lib_a-memcmp.$(OBJEXT) \
+am_lib_a_OBJECTS = lib_a-memchr-stub.$(OBJEXT) lib_a-memchr.$(OBJEXT) \
+	lib_a-memcmp-stub.$(OBJEXT) lib_a-memcmp.$(OBJEXT) \
 	lib_a-memcpy-stub.$(OBJEXT) lib_a-memcpy.$(OBJEXT) \
 	lib_a-memmove-stub.$(OBJEXT) lib_a-memmove.$(OBJEXT) \
 	lib_a-memset-stub.$(OBJEXT) lib_a-memset.$(OBJEXT) \
@@ -205,11 +206,11 @@ AUTOMAKE_OPTIONS = cygnus
 INCLUDES = $(NEWLIB_CFLAGS) $(CROSS_CFLAGS) $(TARGET_CFLAGS)
 AM_CCASFLAGS = $(INCLUDES)
 noinst_LIBRARIES = lib.a
-lib_a_SOURCES = memcmp-stub.c memcmp.S memcpy-stub.c memcpy.S \
-	memmove-stub.c memmove.S memset-stub.c memset.S setjmp.S \
-	strchr-stub.c strchr.S strchrnul-stub.c strchrnul.S \
-	strcmp-stub.c strcmp.S strlen-stub.c strlen.S strncmp-stub.c \
-	strncmp.S strnlen-stub.c strnlen.S
+lib_a_SOURCES = memchr-stub.c memchr.S memcmp-stub.c memcmp.S \
+	memcpy-stub.c memcpy.S memmove-stub.c memmove.S memset-stub.c \
+	memset.S setjmp.S strchr-stub.c strchr.S strchrnul-stub.c \
+	strchrnul.S strcmp-stub.c strcmp.S strlen-stub.c strlen.S \
+	strncmp-stub.c strncmp.S strnlen-stub.c strnlen.S
 lib_a_CCASFLAGS = $(AM_CCASFLAGS)
 lib_a_CFLAGS = $(AM_CFLAGS)
 ACLOCAL_AMFLAGS = -I ../../.. -I ../../../..
@@ -272,6 +273,12 @@ distclean-compile:
 .S.obj:
 	$(CPPASCOMPILE) -c -o $@ `$(CYGPATH_W) '$<'`
 
+lib_a-memchr.o: memchr.S
+	$(CCAS) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CCASFLAGS) $(CCASFLAGS) -c -o lib_a-memchr.o `test -f 'memchr.S' || echo '$(srcdir)/'`memchr.S
+
+lib_a-memchr.obj: memchr.S
+	$(CCAS) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CCASFLAGS) $(CCASFLAGS) -c -o lib_a-memchr.obj `if test -f 'memchr.S'; then $(CYGPATH_W) 'memchr.S'; else $(CYGPATH_W) '$(srcdir)/memchr.S'; fi`
+
 lib_a-memcmp.o: memcmp.S
 	$(CCAS) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CCASFLAGS) $(CCASFLAGS) -c -o lib_a-memcmp.o `test -f 'memcmp.S' || echo '$(srcdir)/'`memcmp.S
 
@@ -344,6 +351,12 @@ lib_a-strnlen.obj: strnlen.S
 .c.obj:
 	$(COMPILE) -c `$(CYGPATH_W) '$<'`
 
+lib_a-memchr-stub.o: memchr-stub.c
+	$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-memchr-stub.o `test -f 'memchr-stub.c' || echo '$(srcdir)/'`memchr-stub.c
+
+lib_a-memchr-stub.obj: memchr-stub.c
+	$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-memchr-stub.obj `if test -f 'memchr-stub.c'; then $(CYGPATH_W) 'memchr-stub.c'; else $(CYGPATH_W) '$(srcdir)/memchr-stub.c'; fi`
+
 lib_a-memcmp-stub.o: memcmp-stub.c
 	$(CC) $(DEFS) $(DEFAULT_INCLUDES) $(INCLUDES) $(AM_CPPFLAGS) $(CPPFLAGS) $(lib_a_CFLAGS) $(CFLAGS) -c -o lib_a-memcmp-stub.o `test -f 'memcmp-stub.c' || echo '$(srcdir)/'`memcmp-stub.c
 
diff --git a/newlib/libc/machine/aarch64/memchr-stub.c b/newlib/libc/machine/aarch64/memchr-stub.c
new file mode 100644
index 0000000..dc550aa
--- /dev/null
+++ b/newlib/libc/machine/aarch64/memchr-stub.c
@@ -0,0 +1,31 @@
+/* Copyright (c) 2013, Linaro Limited
+   All rights reserved.
+
+   Redistribution and use in source and binary forms, with or without
+   modification, are permitted provided that the following conditions are met:
+       * Redistributions of source code must retain the above copyright
+         notice, this list of conditions and the following disclaimer.
+       * Redistributions in binary form must reproduce the above copyright
+         notice, this list of conditions and the following disclaimer in the
+         documentation and/or other materials provided with the distribution.
+       * Neither the name of the Linaro nor the
+         names of its contributors may be used to endorse or promote products
+         derived from this software without specific prior written permission.
+
+   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+   HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. */
+
+#if (defined (__OPTIMIZE_SIZE__) || defined (PREFER_SIZE_OVER_SPEED))
+# include "../../string/memchr.c"
+#else
+/* See memchr.S  */
+#endif
diff --git a/newlib/libc/machine/aarch64/memchr.S b/newlib/libc/machine/aarch64/memchr.S
new file mode 100644
index 0000000..eb59a5a
--- /dev/null
+++ b/newlib/libc/machine/aarch64/memchr.S
@@ -0,0 +1,170 @@
+/*
+ * memchr - find a character in a memory zone
+ *
+ * Copyright (c) 2014, ARM Limited
+ * All rights Reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in the
+ *       documentation and/or other materials provided with the distribution.
+ *     * Neither the name of the company nor the names of its contributors
+ *       may be used to endorse or promote products derived from this
+ *       software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#if (defined (__OPTIMIZE_SIZE__) || defined (PREFER_SIZE_OVER_SPEED))
+/* See memchr-stub.c  */
+#else
+/* Assumptions:
+ *
+ * ARMv8-a, AArch64
+ * Neon Available.
+ */
+
+/* Arguments and results.  */
+#define srcin		x0
+#define chrin		w1
+#define cntin		x2
+
+#define result		x0
+
+#define src		x3
+#define	tmp		x4
+#define wtmp2		w5
+#define synd		x6
+#define soff		x9
+#define cntrem		x10
+
+#define vrepchr		v0
+#define vdata1		v1
+#define vdata2		v2
+#define vhas_chr1	v3
+#define vhas_chr2	v4
+#define vrepmask	v5
+#define vend		v6
+
+/*
+ * Core algorithm:
+ *
+ * For each 32-byte chunk we calculate a 64-bit syndrome value, with two bits
+ * per byte. For each tuple, bit 0 is set if the relevant byte matched the
+ * requested character and bit 1 is not used (faster than using a 32bit
+ * syndrome). Since the bits in the syndrome reflect exactly the order in which
+ * things occur in the original string, counting trailing zeros allows to
+ * identify exactly which byte has matched.
+ */
+
+	.macro def_fn f p2align=0
+	.text
+	.p2align \p2align
+	.global \f
+	.type \f, %function
+\f:
+	.endm
+
+def_fn memchr
+	/*
+	 * Magic constant 0x40100401 allows us to identify which lane matches
+	 * the requested byte.
+	 */
+	mov	wtmp2, #0x0401
+	movk	wtmp2, #0x4010, lsl #16
+	dup	vrepchr.16b, chrin
+	/* Work with aligned 32-byte chunks */
+	bic	src, srcin, #31
+	dup	vrepmask.4s, wtmp2
+	ands	soff, srcin, #31
+	and	cntrem, cntin, #31
+	b.eq	.Lloop
+
+	/*
+	 * Input string is not 32-byte aligned. We calculate the syndrome
+	 * value for the aligned 32 bytes block containing the first bytes
+	 * and mask the irrelevant part.
+	 */
+
+	ld1	{vdata1.16b, vdata2.16b}, [src], #32
+	sub	tmp, soff, #32
+	adds	cntin, cntin, tmp
+	cmeq	vhas_chr1.16b, vdata1.16b, vrepchr.16b
+	cmeq	vhas_chr2.16b, vdata2.16b, vrepchr.16b
+	and	vhas_chr1.16b, vhas_chr1.16b, vrepmask.16b
+	and	vhas_chr2.16b, vhas_chr2.16b, vrepmask.16b
+	addp	vend.16b, vhas_chr1.16b, vhas_chr2.16b		/* 256->128 */
+	addp	vend.16b, vend.16b, vend.16b			/* 128->64 */
+	mov	synd, vend.2d[0]
+	/* Clear the soff*2 lower bits */
+	lsl	tmp, soff, #1
+	lsr	synd, synd, tmp
+	lsl	synd, synd, tmp
+	/* The first block can also be the last */
+	b.ls	.Lmasklast
+	/* Have we found something already? */
+	cbnz	synd, .Ltail
+
+.Lloop:
+	ld1	{vdata1.16b, vdata2.16b}, [src], #32
+	subs	cntin, cntin, #32
+	cmeq	vhas_chr1.16b, vdata1.16b, vrepchr.16b
+	cmeq	vhas_chr2.16b, vdata2.16b, vrepchr.16b
+	/* If we're out of data we finish regardless of the result */
+	b.ls	.Lend
+	/* Use a fast check for the termination condition */
+	orr	vend.16b, vhas_chr1.16b, vhas_chr2.16b
+	addp	vend.2d, vend.2d, vend.2d
+	mov	synd, vend.2d[0]
+	/* We're not out of data, loop if we haven't found the character */
+	cbz	synd, .Lloop
+
+.Lend:
+	/* Termination condition found, let's calculate the syndrome value */
+	and	vhas_chr1.16b, vhas_chr1.16b, vrepmask.16b
+	and	vhas_chr2.16b, vhas_chr2.16b, vrepmask.16b
+	addp	vend.16b, vhas_chr1.16b, vhas_chr2.16b		/* 256->128 */
+	addp	vend.16b, vend.16b, vend.16b			/* 128->64 */
+	mov	synd, vend.2d[0]
+	/* Only do the clear for the last possible block */
+	b.hi	.Ltail
+
+.Lmasklast:
+	/* Clear the (32 - ((cntrem + soff) % 32)) * 2 upper bits */
+	add	tmp, cntrem, soff
+	and	tmp, tmp, #31
+	sub	tmp, tmp, #32
+	neg	tmp, tmp, lsl #1
+	lsl	synd, synd, tmp
+	lsr	synd, synd, tmp
+
+.Ltail:
+	/* Count the trailing zeros using bit reversing */
+	rbit	synd, synd
+	/* Compensate the last post-increment */
+	sub	src, src, #32
+	/* Check that we have found a character */
+	cmp	synd, #0
+	/* And count the leading zeros */
+	clz	synd, synd
+	/* Compute the potential result */
+	add	result, src, synd, lsr #1
+	/* Select result or NULL */
+	csel	result, xzr, result, eq
+	ret
+
+	.size	memchr, . - memchr
+#endif

Re: [PATCH][AArch64] NEON-optimised memchr

[Richard Earnshaw]

On 10/07/14 19:27, Kevin Petit wrote:
> Hi Richard,
> 
>> ENOPATCH?
> 
> Thanks for pointing it out. I hate when I do this.
> 
> Apologies everyone, here it is.
> 
> Regards,
> Kévin
> 
> 
> 

Thanks.

Committed.

R.