compute fletcher 4 with avx instructions

Detect if the running CPU supports AVX instruction, and evaluate
Fletcher-4 computation througput and choose the fastest one.

Signed-off-by: Jinshan Xiong <[email protected]>

include/sys/Makefile.am

@@ -49,6 +49,7 @@ COMMON_H = \
 	$(top_srcdir)/include/sys/space_reftree.h \
 	$(top_srcdir)/include/sys/spa.h \
 	$(top_srcdir)/include/sys/spa_impl.h \
+	$(top_srcdir)/include/sys/spa_checksum.h \
 	$(top_srcdir)/include/sys/trace.h \
 	$(top_srcdir)/include/sys/trace_acl.h \
 	$(top_srcdir)/include/sys/trace_arc.h \

include/sys/spa.h

@@ -35,6 +35,7 @@
 #include <sys/sysmacros.h>
 #include <sys/types.h>
 #include <sys/fs/zfs.h>
+#include <sys/spa_checksum.h>
 
 #ifdef	__cplusplus
 extern "C" {
@@ -142,12 +143,6 @@ typedef struct dva {
 	uint64_t	dva_word[2];
 } dva_t;
 
-/*
- * Each block has a 256-bit checksum -- strong enough for cryptographic hashes.
- */
-typedef struct zio_cksum {
-	uint64_t	zc_word[4];
-} zio_cksum_t;
 
 /*
  * Each block is described by its DVAs, time of birth, checksum, etc.
@@ -440,35 +435,9 @@ _NOTE(CONSTCOND) } while (0)
 	DVA_EQUAL(&(bp1)->blk_dva[1], &(bp2)->blk_dva[1]) &&	\
 	DVA_EQUAL(&(bp1)->blk_dva[2], &(bp2)->blk_dva[2]))
 
-#define	ZIO_CHECKSUM_EQUAL(zc1, zc2) \
-	(0 == (((zc1).zc_word[0] - (zc2).zc_word[0]) | \
-	((zc1).zc_word[1] - (zc2).zc_word[1]) | \
-	((zc1).zc_word[2] - (zc2).zc_word[2]) | \
-	((zc1).zc_word[3] - (zc2).zc_word[3])))
-
-#define	ZIO_CHECKSUM_IS_ZERO(zc) \
-	(0 == ((zc)->zc_word[0] | (zc)->zc_word[1] | \
-	(zc)->zc_word[2] | (zc)->zc_word[3]))
-
-#define	ZIO_CHECKSUM_BSWAP(zcp)					\
-{								\
-	(zcp)->zc_word[0] = BSWAP_64((zcp)->zc_word[0]);	\
-	(zcp)->zc_word[1] = BSWAP_64((zcp)->zc_word[1]);	\
-	(zcp)->zc_word[2] = BSWAP_64((zcp)->zc_word[2]);	\
-	(zcp)->zc_word[3] = BSWAP_64((zcp)->zc_word[3]);	\
-}
-
 
 #define	DVA_IS_VALID(dva)	(DVA_GET_ASIZE(dva) != 0)
 
-#define	ZIO_SET_CHECKSUM(zcp, w0, w1, w2, w3)	\
-{						\
-	(zcp)->zc_word[0] = w0;			\
-	(zcp)->zc_word[1] = w1;			\
-	(zcp)->zc_word[2] = w2;			\
-	(zcp)->zc_word[3] = w3;			\
-}
-
 #define	BP_IDENTITY(bp)		(ASSERT(!BP_IS_EMBEDDED(bp)), &(bp)->blk_dva[0])
 #define	BP_IS_GANG(bp)		\
 	(BP_IS_EMBEDDED(bp) ? B_FALSE : DVA_GET_GANG(BP_IDENTITY(bp)))

include/sys/spa_checksum.h

@@ -0,0 +1,72 @@
+/*
+ * CDDL HEADER START
+ *
+ * The contents of this file are subject to the terms of the
+ * Common Development and Distribution License (the "License").
+ * You may not use this file except in compliance with the License.
+ *
+ * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
+ * or http://www.opensolaris.org/os/licensing.
+ * See the License for the specific language governing permissions
+ * and limitations under the License.
+ *
+ * When distributing Covered Code, include this CDDL HEADER in each
+ * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
+ * If applicable, add the following below this CDDL HEADER, with the
+ * fields enclosed by brackets "[]" replaced with your own identifying
+ * information: Portions Copyright [yyyy] [name of copyright owner]
+ *
+ * CDDL HEADER END
+ */
+/*
+ * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
+ * Use is subject to license terms.
+ */
+
+#ifndef	_SPA_CHECKSUM_H
+#define	_SPA_CHECKSUM_H
+
+#include <sys/types.h>
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+/*
+ * Each block has a 256-bit checksum -- strong enough for cryptographic hashes.
+ */
+typedef struct zio_cksum {
+	uint64_t	zc_word[4];
+} zio_cksum_t;
+
+#define	ZIO_SET_CHECKSUM(zcp, w0, w1, w2, w3)	\
+{						\
+	(zcp)->zc_word[0] = w0;			\
+	(zcp)->zc_word[1] = w1;			\
+	(zcp)->zc_word[2] = w2;			\
+	(zcp)->zc_word[3] = w3;			\
+}
+
+#define	ZIO_CHECKSUM_EQUAL(zc1, zc2) \
+	(0 == (((zc1).zc_word[0] - (zc2).zc_word[0]) | \
+	((zc1).zc_word[1] - (zc2).zc_word[1]) | \
+	((zc1).zc_word[2] - (zc2).zc_word[2]) | \
+	((zc1).zc_word[3] - (zc2).zc_word[3])))
+
+#define	ZIO_CHECKSUM_IS_ZERO(zc) \
+	(0 == ((zc)->zc_word[0] | (zc)->zc_word[1] | \
+	(zc)->zc_word[2] | (zc)->zc_word[3]))
+
+#define	ZIO_CHECKSUM_BSWAP(zcp)					\
+{								\
+	(zcp)->zc_word[0] = BSWAP_64((zcp)->zc_word[0]);	\
+	(zcp)->zc_word[1] = BSWAP_64((zcp)->zc_word[1]);	\
+	(zcp)->zc_word[2] = BSWAP_64((zcp)->zc_word[2]);	\
+	(zcp)->zc_word[3] = BSWAP_64((zcp)->zc_word[3]);	\
+}
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif

include/zfs_fletcher.h

@@ -27,7 +27,7 @@
 #define	_ZFS_FLETCHER_H
 
 #include <sys/types.h>
-#include <sys/spa.h>
+#include <sys/spa_checksum.h>
 
 #ifdef	__cplusplus
 extern "C" {
@@ -45,6 +45,19 @@ void fletcher_4_incremental_native(const void *, uint64_t,
     zio_cksum_t *);
 void fletcher_4_incremental_byteswap(const void *, uint64_t,
     zio_cksum_t *);
+void fletcher_4_init(void);
+
+/*
+ * fletcher checksum struct
+ */
+struct fletcher_4_calls {
+	void (*init)(zio_cksum_t *);
+	void (*fini)(zio_cksum_t *);
+	void (*compute)(const void *, uint64_t, zio_cksum_t *);
+	void (*compute_byteswap)(const void *, uint64_t, zio_cksum_t *);
+	boolean_t (*valid)(void);
+	const char *name;
+};
 
 #ifdef	__cplusplus
 }

module/zcommon/Makefile.in

@@ -15,3 +15,5 @@ $(MODULE)-objs += zfs_comutil.o
 $(MODULE)-objs += zfs_fletcher.o
 $(MODULE)-objs += zfs_uio.o
 $(MODULE)-objs += zpool_prop.o
+
+$(MODULE)-$(CONFIG_X86) += zfs_fletcher_intel.o

module/zcommon/zfs_fletcher.c

@@ -128,8 +128,8 @@
 #include <sys/types.h>
 #include <sys/sysmacros.h>
 #include <sys/byteorder.h>
-#include <sys/zio.h>
 #include <sys/spa.h>
+#include <zfs_fletcher.h>
 
 void
 fletcher_2_native(const void *buf, uint64_t size, zio_cksum_t *zcp)
@@ -165,43 +165,13 @@ fletcher_2_byteswap(const void *buf, uint64_t size, zio_cksum_t *zcp)
 	ZIO_SET_CHECKSUM(zcp, a0, a1, b0, b1);
 }
 
-void
-fletcher_4_native(const void *buf, uint64_t size, zio_cksum_t *zcp)
+static void fletcher_4_generic_init(zio_cksum_t *zcp)
 {
-	const uint32_t *ip = buf;
-	const uint32_t *ipend = ip + (size / sizeof (uint32_t));
-	uint64_t a, b, c, d;
-
-	for (a = b = c = d = 0; ip < ipend; ip++) {
-		a += ip[0];
-		b += a;
-		c += b;
-		d += c;
-	}
-
-	ZIO_SET_CHECKSUM(zcp, a, b, c, d);
-}
-
-void
-fletcher_4_byteswap(const void *buf, uint64_t size, zio_cksum_t *zcp)
-{
-	const uint32_t *ip = buf;
-	const uint32_t *ipend = ip + (size / sizeof (uint32_t));
-	uint64_t a, b, c, d;
-
-	for (a = b = c = d = 0; ip < ipend; ip++) {
-		a += BSWAP_32(ip[0]);
-		b += a;
-		c += b;
-		d += c;
-	}
-
-	ZIO_SET_CHECKSUM(zcp, a, b, c, d);
+	ZIO_SET_CHECKSUM(zcp, 0, 0, 0, 0);
 }
 
-void
-fletcher_4_incremental_native(const void *buf, uint64_t size,
-    zio_cksum_t *zcp)
+static void
+fletcher_4_generic(const void *buf, uint64_t size, zio_cksum_t *zcp)
 {
 	const uint32_t *ip = buf;
 	const uint32_t *ipend = ip + (size / sizeof (uint32_t));
@@ -222,9 +192,8 @@ fletcher_4_incremental_native(const void *buf, uint64_t size,
 	ZIO_SET_CHECKSUM(zcp, a, b, c, d);
 }
 
-void
-fletcher_4_incremental_byteswap(const void *buf, uint64_t size,
-    zio_cksum_t *zcp)
+static void
+fletcher_4_generic_byteswap(const void *buf, uint64_t size, zio_cksum_t *zcp)
 {
 	const uint32_t *ip = buf;
 	const uint32_t *ipend = ip + (size / sizeof (uint32_t));
@@ -245,7 +214,107 @@ fletcher_4_incremental_byteswap(const void *buf, uint64_t size,
 	ZIO_SET_CHECKSUM(zcp, a, b, c, d);
 }
 
+static const struct fletcher_4_calls fletcher_4_generic_calls = {
+	.init = fletcher_4_generic_init,
+	.compute = fletcher_4_generic,
+	.compute_byteswap = fletcher_4_generic_byteswap,
+	.name = "generic"
+};
+
+static const struct fletcher_4_calls *chosen = &fletcher_4_generic_calls;
+
+void
+fletcher_4_native(const void *buf, uint64_t size, zio_cksum_t *zcp)
+{
+	chosen->init(zcp);
+	chosen->compute(buf, size, zcp);
+	if (chosen->fini != NULL)
+		chosen->fini(zcp);
+}
+
+void
+fletcher_4_byteswap(const void *buf, uint64_t size, zio_cksum_t *zcp)
+{
+	chosen->init(zcp);
+	chosen->compute_byteswap(buf, size, zcp);
+	if (chosen->fini != NULL)
+		chosen->fini(zcp);
+}
+
+void
+fletcher_4_incremental_native(const void *buf, uint64_t size,
+    zio_cksum_t *zcp)
+{
+	fletcher_4_generic(buf, size, zcp);
+}
+
+void
+fletcher_4_incremental_byteswap(const void *buf, uint64_t size,
+    zio_cksum_t *zcp)
+{
+	fletcher_4_generic_byteswap(buf, size, zcp);
+}
+
+
 #if defined(_KERNEL) && defined(HAVE_SPL)
+
+extern struct fletcher_4_calls fletcher_4_avx2_calls;
+
+static const struct fletcher_4_calls *fletcher_4_algos[] = {
+	&fletcher_4_generic_calls,
+#if defined(HAVE_AVX) && defined(HAVE_AVX2)
+	&fletcher_4_avx2_calls,
+#endif
+};
+
+#define	BENCH_SIZE 4096
+#define	ZFS_ARRAY_SIZE(a) (sizeof (a) / sizeof (*a))
+#define	kernel_cpu_relax() do {} while (0)
+
+/* cant use allocation methods from zfs module! */
+static char databuf[BENCH_SIZE];
+
+void
+fletcher_4_init(void)
+{
+	unsigned long bestperf = 0;
+	const unsigned int bits = 4;
+	int i;
+
+	for (i = 0; i < ZFS_ARRAY_SIZE(fletcher_4_algos); i++) {
+		const struct fletcher_4_calls *algo = fletcher_4_algos[i];
+		unsigned long perf = 0;
+		clock_t j0, j1;
+		zio_cksum_t zc;
+
+		if (algo->valid != NULL && !algo->valid())
+			continue;
+
+		kpreempt_disable();
+		j0 = ddi_get_lbolt();
+		while ((j1 = ddi_get_lbolt()) == j0) {
+			kernel_cpu_relax();
+		}
+
+		algo->init(&zc);
+		while (ddi_time_before(ddi_get_lbolt(), j1 + (1 << bits))) {
+			algo->compute(databuf, PAGE_SIZE, &zc);
+			perf++;
+		}
+		if (algo->fini != NULL)
+			algo->fini(&zc);
+		kpreempt_enable();
+
+		if (perf > bestperf) {
+			bestperf = perf;
+			chosen = algo;
+		}
+		cmn_err(CE_NOTE, "fletcher-4: %-8s %5ld MB/s",
+		    algo->name, SEC_TO_TICK(perf) >> (20 - 16 + bits));
+	}
+}
+
+EXPORT_SYMBOL(fletcher_4_init);
 EXPORT_SYMBOL(fletcher_2_native);
 EXPORT_SYMBOL(fletcher_2_byteswap);
 EXPORT_SYMBOL(fletcher_4_native);