DPDK patches and discussions
 help / color / mirror / Atom feed
From: Konstantin Ananyev <konstantin.ananyev@intel.com>
To: dev@dpdk.org
To: dev@dpdk.org
Subject: [dpdk-dev] [PATCH 1/2] lpm: Introduce rte_lpm_lookupx4
Date: Thu, 22 May 2014 17:55:41 +0100	[thread overview]
Message-ID: <1400777742-498-2-git-send-email-konstantin.ananyev@intel.com> (raw)
In-Reply-To: <1400777742-498-1-git-send-email-konstantin.ananyev@intel.com>

Introduce rte_lpm_lookupx4():
 - Allows to lookup four IP addresses in an LPM table.
 - Uses SSE instrincts.

Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
---
 app/test/test_lpm.c                             |   70 ++++++++++++++
 lib/librte_eal/common/Makefile                  |    1 +
 lib/librte_eal/common/include/rte_common_vect.h |   93 ++++++++++++++++++
 lib/librte_lpm/rte_lpm.h                        |  117 +++++++++++++++++++++++
 4 files changed, 281 insertions(+), 0 deletions(-)
 create mode 100644 lib/librte_eal/common/include/rte_common_vect.h

diff --git a/app/test/test_lpm.c b/app/test/test_lpm.c
index ffed766..1b2f32a 100644
--- a/app/test/test_lpm.c
+++ b/app/test/test_lpm.c
@@ -310,6 +310,8 @@ test6(void)
 int32_t
 test7(void)
 {
+	__m128i ipx4;
+	uint16_t hop[4];
 	struct rte_lpm *lpm = NULL;
 	uint32_t ip = IPv4(0, 0, 0, 0);
 	uint8_t depth = 32, next_hop_add = 100, next_hop_return = 0;
@@ -324,6 +326,13 @@ test7(void)
 	status = rte_lpm_lookup(lpm, ip, &next_hop_return);
 	TEST_LPM_ASSERT((status == 0) && (next_hop_return == next_hop_add));
 
+	ipx4 = _mm_set_epi32(ip, ip + 0x100, ip - 0x100, ip);
+	rte_lpm_lookupx4(lpm, ipx4, hop, UINT16_MAX);
+	TEST_LPM_ASSERT(hop[0] == next_hop_add);
+	TEST_LPM_ASSERT(hop[1] == UINT16_MAX);
+	TEST_LPM_ASSERT(hop[2] == UINT16_MAX);
+	TEST_LPM_ASSERT(hop[3] == next_hop_add);
+
 	status = rte_lpm_delete(lpm, ip, depth);
 	TEST_LPM_ASSERT(status == 0);
 
@@ -347,6 +356,8 @@ test7(void)
 int32_t
 test8(void)
 {
+	__m128i ipx4;
+	uint16_t hop[4];
 	struct rte_lpm *lpm = NULL;
 	uint32_t ip1 = IPv4(127, 255, 255, 255), ip2 = IPv4(128, 0, 0, 0);
 	uint8_t depth, next_hop_add, next_hop_return;
@@ -370,6 +381,13 @@ test8(void)
 		status = rte_lpm_lookup(lpm, ip2, &next_hop_return);
 		TEST_LPM_ASSERT((status == 0) &&
 			(next_hop_return == next_hop_add));
+
+		ipx4 = _mm_set_epi32(ip2, ip1, ip2, ip1);
+		rte_lpm_lookupx4(lpm, ipx4, hop, UINT16_MAX);
+		TEST_LPM_ASSERT(hop[0] == UINT16_MAX);
+		TEST_LPM_ASSERT(hop[1] == next_hop_add);
+		TEST_LPM_ASSERT(hop[2] == UINT16_MAX);
+		TEST_LPM_ASSERT(hop[3] == next_hop_add);
 	}
 
 	/* Loop with rte_lpm_delete. */
@@ -391,6 +409,18 @@ test8(void)
 
 		status = rte_lpm_lookup(lpm, ip1, &next_hop_return);
 		TEST_LPM_ASSERT(status == -ENOENT);
+
+		ipx4 = _mm_set_epi32(ip1, ip1, ip2, ip2);
+		rte_lpm_lookupx4(lpm, ipx4, hop, UINT16_MAX);
+		if (depth != 1) {
+			TEST_LPM_ASSERT(hop[0] == next_hop_add);
+			TEST_LPM_ASSERT(hop[1] == next_hop_add);
+		} else {
+			TEST_LPM_ASSERT(hop[0] == UINT16_MAX);
+			TEST_LPM_ASSERT(hop[1] == UINT16_MAX);
+		}
+		TEST_LPM_ASSERT(hop[2] == UINT16_MAX);
+		TEST_LPM_ASSERT(hop[3] == UINT16_MAX);
 	}
 
 	rte_lpm_free(lpm);
@@ -822,6 +852,8 @@ test11(void)
 int32_t
 test12(void)
 {
+	__m128i ipx4;
+	uint16_t hop[4];
 	struct rte_lpm *lpm = NULL;
 	uint32_t ip, i;
 	uint8_t depth, next_hop_add, next_hop_return;
@@ -842,6 +874,13 @@ test12(void)
 		TEST_LPM_ASSERT((status == 0) &&
 				(next_hop_return == next_hop_add));
 
+		ipx4 = _mm_set_epi32(ip, ip + 1, ip, ip - 1);
+		rte_lpm_lookupx4(lpm, ipx4, hop, UINT16_MAX);
+		TEST_LPM_ASSERT(hop[0] == UINT16_MAX);
+		TEST_LPM_ASSERT(hop[1] == next_hop_add);
+		TEST_LPM_ASSERT(hop[2] == UINT16_MAX);
+		TEST_LPM_ASSERT(hop[3] == next_hop_add);
+
 		status = rte_lpm_delete(lpm, ip, depth);
 		TEST_LPM_ASSERT(status == 0);
 
@@ -1237,6 +1276,37 @@ perf_test(void)
 			(double)total_time / ((double)ITERATIONS * BATCH_SIZE),
 			(count * 100.0) / (double)(ITERATIONS * BATCH_SIZE));
 
+	/* Measure LookupX4 */
+	total_time = 0;
+	count = 0;
+	for (i = 0; i < ITERATIONS; i ++) {
+		static uint32_t ip_batch[BATCH_SIZE];
+		uint16_t next_hops[4];
+
+		/* Create array of random IP addresses */
+		for (j = 0; j < BATCH_SIZE; j ++)
+			ip_batch[j] = rte_rand();
+
+		/* Lookup per batch */
+		begin = rte_rdtsc();
+		for (j = 0; j < BATCH_SIZE; j += RTE_DIM(next_hops)) {
+			unsigned k;
+			__m128i ipx4;
+
+			ipx4 = _mm_loadu_si128((__m128i *)(ip_batch + j));
+			ipx4 = *(__m128i *)(ip_batch + j);
+			rte_lpm_lookupx4(lpm, ipx4, next_hops, UINT16_MAX);
+			for (k = 0; k < RTE_DIM(next_hops); k++)
+				if (unlikely (next_hops[k] == UINT16_MAX))
+					count++;
+		}
+
+		total_time += rte_rdtsc() - begin;
+	}
+	printf("LPM LookupX4: %.1f cycles (fails = %.1f%%)\n",
+			(double)total_time / ((double)ITERATIONS * BATCH_SIZE),
+			(count * 100.0) / (double)(ITERATIONS * BATCH_SIZE));
+
 	/* Delete */
 	status = 0;
 	begin = rte_rdtsc();
diff --git a/lib/librte_eal/common/Makefile b/lib/librte_eal/common/Makefile
index 0016fc5..3103019 100644
--- a/lib/librte_eal/common/Makefile
+++ b/lib/librte_eal/common/Makefile
@@ -39,6 +39,7 @@ INC += rte_rwlock.h rte_spinlock.h rte_tailq.h rte_interrupts.h rte_alarm.h
 INC += rte_string_fns.h rte_cpuflags.h rte_version.h rte_tailq_elem.h
 INC += rte_eal_memconfig.h rte_malloc_heap.h
 INC += rte_hexdump.h rte_devargs.h rte_dev.h
+INC += rte_common_vect.h
 
 ifeq ($(CONFIG_RTE_INSECURE_FUNCTION_WARNING),y)
 INC += rte_warnings.h
diff --git a/lib/librte_eal/common/include/rte_common_vect.h b/lib/librte_eal/common/include/rte_common_vect.h
new file mode 100644
index 0000000..065fc52
--- /dev/null
+++ b/lib/librte_eal/common/include/rte_common_vect.h
@@ -0,0 +1,93 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
+ *   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 Intel Corporation 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
+ *   OWNER 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.
+ */
+
+#ifndef _RTE_COMMON_VECT_H_
+#define _RTE_COMMON_VECT_H_
+
+/**
+ * @file
+ *
+ * RTE SSE/AVX related header.
+ */
+
+#if (defined(__ICC) || (__GNUC__ == 4 &&  __GNUC_MINOR__ < 4))
+
+#ifdef __SSE__
+#include <xmmintrin.h>
+#endif
+
+#ifdef __SSE2__
+#include <emmintrin.h>
+#endif
+
+#if defined (__SSE4_2__) || defined (__SSE4_1__)
+#include <smmintrin.h>
+#endif
+
+#else
+
+#include <x86intrin.h>
+
+#endif
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef __m128i xmm_t;
+
+#define	XMM_SIZE	(sizeof (xmm_t))
+#define	XMM_MASK	(XMM_SIZE - 1)
+
+typedef union rte_xmm {
+    xmm_t    m;
+    uint8_t  u8[XMM_SIZE / sizeof (uint8_t)];
+    uint16_t u16[XMM_SIZE / sizeof (uint16_t)];
+    uint32_t u32[XMM_SIZE / sizeof (uint32_t)];
+    uint64_t u64[XMM_SIZE / sizeof (uint64_t)];
+    double   pd[XMM_SIZE / sizeof (double)];
+} rte_xmm_t;
+
+#ifdef RTE_ARCH_I686
+#define _mm_cvtsi128_si64(a) ({ \
+	rte_xmm_t m;            \
+	m.m = (a);              \
+	(m.u64[0]);             \
+})
+#endif
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* _RTE_COMMON__VECT_H_ */
diff --git a/lib/librte_lpm/rte_lpm.h b/lib/librte_lpm/rte_lpm.h
index 033f542..be8aa1d 100644
--- a/lib/librte_lpm/rte_lpm.h
+++ b/lib/librte_lpm/rte_lpm.h
@@ -45,6 +45,8 @@
 #include <stdlib.h>
 #include <rte_branch_prediction.h>
 #include <rte_memory.h>
+#include <rte_common.h>
+#include <rte_common_vect.h>
 
 #ifdef __cplusplus
 extern "C" {
@@ -331,6 +333,121 @@ rte_lpm_lookup_bulk_func(const struct rte_lpm *lpm, const uint32_t * ips,
 	return 0;
 }
 
+/* Mask four results. */
+#define	 RTE_LPM_MASKX4_RES	UINT64_C(0x00ff00ff00ff00ff)
+
+/**
+ * Lookup four IP addresses in an LPM table.
+ *
+ * @param lpm
+ *   LPM object handle
+ * @param ip
+ *   Four IPs to be looked up in the LPM table
+ * @param hop
+ *   Next hop of the most specific rule found for IP (valid on lookup hit only).
+ *   This is an 4 elements array of two byte values.
+ *   If the lookup was succesfull for the given IP, then least significant byte
+ *   of the corresponding element is the  actual next hop and the most
+ *   significant byte is zero.
+ *   If the lookup for the given IP failed, then corresponding element would
+ *   contain default value, see description of then next parameter.
+ * @param defv
+ *   Default value to populate into corresponding element of hop[] array,
+ *   if lookup would fail.
+ */
+static inline void
+rte_lpm_lookupx4(const struct rte_lpm *lpm, __m128i ip, uint16_t hop[4],
+	uint16_t defv)
+{
+	__m128i i24;
+	rte_xmm_t i8;
+	uint16_t tbl[4];
+	uint64_t idx, pt;
+
+	const __m128i mask8 =
+		_mm_set_epi32(UINT8_MAX, UINT8_MAX, UINT8_MAX, UINT8_MAX);
+
+	/*
+	 * RTE_LPM_VALID_EXT_ENTRY_BITMASK for 4 LPM entries
+	 * as one 64-bit value (0x0300030003000300).
+	 */
+	const uint64_t mask_xv =
+		((uint64_t)RTE_LPM_VALID_EXT_ENTRY_BITMASK |
+		(uint64_t)RTE_LPM_VALID_EXT_ENTRY_BITMASK << 16 |
+		(uint64_t)RTE_LPM_VALID_EXT_ENTRY_BITMASK << 32 |
+		(uint64_t)RTE_LPM_VALID_EXT_ENTRY_BITMASK << 48);
+
+	/*
+	 * RTE_LPM_LOOKUP_SUCCESS for 4 LPM entries
+	 * as one 64-bit value (0x0100010001000100).
+	 */
+	const uint64_t mask_v =
+		((uint64_t)RTE_LPM_LOOKUP_SUCCESS |
+		(uint64_t)RTE_LPM_LOOKUP_SUCCESS << 16 |
+		(uint64_t)RTE_LPM_LOOKUP_SUCCESS << 32 |
+		(uint64_t)RTE_LPM_LOOKUP_SUCCESS << 48);
+
+	/* get 4 indexes for tbl24[]. */
+	i24 = _mm_srli_epi32(ip, CHAR_BIT);
+
+	/* extract values from tbl24[] */
+	idx = _mm_cvtsi128_si64(i24);
+	i24 = _mm_srli_si128(i24, sizeof (uint64_t));
+
+	tbl[0] = *(const uint16_t *)&lpm->tbl24[(uint32_t)idx];
+	tbl[1] = *(const uint16_t *)&lpm->tbl24[idx >> 32];
+
+	idx = _mm_cvtsi128_si64(i24);
+
+	tbl[2] = *(const uint16_t *)&lpm->tbl24[(uint32_t)idx];
+	tbl[3] = *(const uint16_t *)&lpm->tbl24[idx >> 32];
+
+	/* get 4 indexes for tbl8[]. */
+	i8.m = _mm_and_si128(ip, mask8);
+
+	pt = (uint64_t)tbl[0] |
+		(uint64_t)tbl[1] << 16 |
+		(uint64_t)tbl[2] << 32 |
+		(uint64_t)tbl[3] << 48;
+
+	/* search successfully finished for all 4 IP addresses. */
+	if (likely((pt & mask_xv) == mask_v)) {
+		uintptr_t ph = (uintptr_t)hop;
+		*(uint64_t *)ph = pt & RTE_LPM_MASKX4_RES;
+		return;
+	}
+
+	if (unlikely((pt & RTE_LPM_VALID_EXT_ENTRY_BITMASK) ==
+			RTE_LPM_VALID_EXT_ENTRY_BITMASK)) {
+		i8.u32[0] = i8.u32[0] +
+			(uint8_t)tbl[0] * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
+		tbl[0] = *(const uint16_t *)&lpm->tbl8[i8.u32[0]];
+	}
+	if (unlikely((pt >> 16 & RTE_LPM_VALID_EXT_ENTRY_BITMASK) ==
+			RTE_LPM_VALID_EXT_ENTRY_BITMASK)) {
+		i8.u32[1] = i8.u32[1] +
+			(uint8_t)tbl[1] * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
+		tbl[1] = *(const uint16_t *)&lpm->tbl8[i8.u32[1]];
+	}
+	if (unlikely((pt >> 32 & RTE_LPM_VALID_EXT_ENTRY_BITMASK) ==
+			RTE_LPM_VALID_EXT_ENTRY_BITMASK)) {
+		i8.u32[2] = i8.u32[2] +
+			(uint8_t)tbl[2] * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
+		tbl[2] = *(const uint16_t *)&lpm->tbl8[i8.u32[2]];
+	}
+	if (unlikely((pt >> 48 & RTE_LPM_VALID_EXT_ENTRY_BITMASK) ==
+			RTE_LPM_VALID_EXT_ENTRY_BITMASK)) {
+		i8.u32[3] = i8.u32[3] +
+			(uint8_t)tbl[3] * RTE_LPM_TBL8_GROUP_NUM_ENTRIES;
+		tbl[3] = *(const uint16_t *)&lpm->tbl8[i8.u32[3]];
+	}
+
+	hop[0] = (tbl[0] & RTE_LPM_LOOKUP_SUCCESS) ? (uint8_t)tbl[0] : defv;
+	hop[1] = (tbl[1] & RTE_LPM_LOOKUP_SUCCESS) ? (uint8_t)tbl[1] : defv;
+	hop[2] = (tbl[2] & RTE_LPM_LOOKUP_SUCCESS) ? (uint8_t)tbl[2] : defv;
+	hop[3] = (tbl[3] & RTE_LPM_LOOKUP_SUCCESS) ? (uint8_t)tbl[3] : defv;
+}
+
 #ifdef __cplusplus
 }
 #endif
-- 
1.7.7.6

  reply	other threads:[~2014-05-22 16:56 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2014-05-22 16:55 [dpdk-dev] [PATCH 0/2] L3FWD sample optimisation Konstantin Ananyev
2014-05-22 16:55 ` Konstantin Ananyev [this message]
2014-05-22 16:55 ` [dpdk-dev] [PATCH 2/2] l3fwd: reorganise and optimize l3fwd LPM code path Konstantin Ananyev
2014-05-23  8:05 ` [dpdk-dev] [PATCH 0/2] L3FWD sample optimisation Thomas Monjalon
2014-06-04 13:47 ` Cao, Waterman
2014-06-06  8:26 ` De Lara Guarch, Pablo

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=1400777742-498-2-git-send-email-konstantin.ananyev@intel.com \
    --to=konstantin.ananyev@intel.com \
    --cc=dev@dpdk.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
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 NNTP newsgroup(s).