From patchwork Mon Mar 9 12:43:41 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Vladimir Medvedkin X-Patchwork-Id: 66454 X-Patchwork-Delegate: david.marchand@redhat.com Return-Path: X-Original-To: patchwork@inbox.dpdk.org Delivered-To: patchwork@inbox.dpdk.org Received: from dpdk.org (dpdk.org [92.243.14.124]) by inbox.dpdk.org (Postfix) with ESMTP id 3260DA052E; Mon, 9 Mar 2020 13:44:01 +0100 (CET) Received: from [92.243.14.124] (localhost [127.0.0.1]) by dpdk.org (Postfix) with ESMTP id C71F71C0AC; Mon, 9 Mar 2020 13:43:55 +0100 (CET) Received: from mga14.intel.com (mga14.intel.com [192.55.52.115]) by dpdk.org (Postfix) with ESMTP id 128A71C045 for ; Mon, 9 Mar 2020 13:43:52 +0100 (CET) X-Amp-Result: SKIPPED(no attachment in message) X-Amp-File-Uploaded: False Received: from fmsmga002.fm.intel.com ([10.253.24.26]) by fmsmga103.fm.intel.com with ESMTP/TLS/DHE-RSA-AES256-GCM-SHA384; 09 Mar 2020 05:43:52 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.70,533,1574150400"; d="scan'208";a="276500009" Received: from silpixa00400072.ir.intel.com ([10.237.222.213]) by fmsmga002.fm.intel.com with ESMTP; 09 Mar 2020 05:43:51 -0700 From: Vladimir Medvedkin To: dev@dpdk.org Cc: konstantin.ananyev@intel.com, bruce.richardson@intel.com Date: Mon, 9 Mar 2020 12:43:41 +0000 Message-Id: <1583757826-375246-2-git-send-email-vladimir.medvedkin@intel.com> X-Mailer: git-send-email 2.7.4 In-Reply-To: <1583757826-375246-1-git-send-email-vladimir.medvedkin@intel.com> References: <1583757826-375246-1-git-send-email-vladimir.medvedkin@intel.com> Subject: [dpdk-dev] [PATCH 1/6] eal: introduce zmm type for AVX 512-bit X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" New data type to manipulate 512 bit AVX values. Signed-off-by: Vladimir Medvedkin --- lib/librte_eal/common/include/arch/x86/rte_vect.h | 20 ++++++++++++++++++++ 1 file changed, 20 insertions(+) diff --git a/lib/librte_eal/common/include/arch/x86/rte_vect.h b/lib/librte_eal/common/include/arch/x86/rte_vect.h index df5a607..09f30e6 100644 --- a/lib/librte_eal/common/include/arch/x86/rte_vect.h +++ b/lib/librte_eal/common/include/arch/x86/rte_vect.h @@ -90,6 +90,26 @@ __extension__ ({ \ }) #endif /* (defined(__ICC) && __ICC < 1210) */ +#ifdef __AVX512F__ + +typedef __m512i zmm_t; + +#define ZMM_SIZE (sizeof(zmm_t)) +#define ZMM_MASK (ZMM_SIZE - 1) + +typedef union rte_zmm { + zmm_t z; + ymm_t y[ZMM_SIZE / sizeof(ymm_t)]; + xmm_t x[ZMM_SIZE / sizeof(xmm_t)]; + uint8_t u8[ZMM_SIZE / sizeof(uint8_t)]; + uint16_t u16[ZMM_SIZE / sizeof(uint16_t)]; + uint32_t u32[ZMM_SIZE / sizeof(uint32_t)]; + uint64_t u64[ZMM_SIZE / sizeof(uint64_t)]; + double pd[ZMM_SIZE / sizeof(double)]; +} rte_zmm_t; + +#endif /* __AVX512F__ */ + #ifdef __cplusplus } #endif From patchwork Mon Mar 9 12:43:42 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Vladimir Medvedkin X-Patchwork-Id: 66455 X-Patchwork-Delegate: david.marchand@redhat.com Return-Path: X-Original-To: patchwork@inbox.dpdk.org Delivered-To: patchwork@inbox.dpdk.org Received: from dpdk.org (dpdk.org [92.243.14.124]) by inbox.dpdk.org (Postfix) with ESMTP id 0D061A052E; Mon, 9 Mar 2020 13:44:10 +0100 (CET) Received: from [92.243.14.124] (localhost [127.0.0.1]) by dpdk.org (Postfix) with ESMTP id DA1AD1C0C0; Mon, 9 Mar 2020 13:43:56 +0100 (CET) Received: from mga14.intel.com (mga14.intel.com [192.55.52.115]) by dpdk.org (Postfix) with ESMTP id 41EA11C045 for ; Mon, 9 Mar 2020 13:43:54 +0100 (CET) X-Amp-Result: SKIPPED(no attachment in message) X-Amp-File-Uploaded: False Received: from fmsmga002.fm.intel.com ([10.253.24.26]) by fmsmga103.fm.intel.com with ESMTP/TLS/DHE-RSA-AES256-GCM-SHA384; 09 Mar 2020 05:43:53 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.70,533,1574150400"; d="scan'208";a="276500015" Received: from silpixa00400072.ir.intel.com ([10.237.222.213]) by fmsmga002.fm.intel.com with ESMTP; 09 Mar 2020 05:43:52 -0700 From: Vladimir Medvedkin To: dev@dpdk.org Cc: konstantin.ananyev@intel.com, bruce.richardson@intel.com Date: Mon, 9 Mar 2020 12:43:42 +0000 Message-Id: <1583757826-375246-3-git-send-email-vladimir.medvedkin@intel.com> X-Mailer: git-send-email 2.7.4 In-Reply-To: <1583757826-375246-1-git-send-email-vladimir.medvedkin@intel.com> References: <1583757826-375246-1-git-send-email-vladimir.medvedkin@intel.com> Subject: [dpdk-dev] [PATCH 2/6] fib: make lookup function type configurable X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" Add type argument to dir24_8_get_lookup_fn() Now it supports 3 different lookup implementations: RTE_FIB_DIR24_8_SCALAR_MACRO RTE_FIB_DIR24_8_SCALAR_INLINE RTE_FIB_DIR24_8_SCALAR_UNI Add new rte_fib_set_lookup_fn() - user can change lookup function type runtime. Signed-off-by: Vladimir Medvedkin --- lib/librte_fib/dir24_8.c | 32 ++++++++++++++++++++------------ lib/librte_fib/dir24_8.h | 2 +- lib/librte_fib/rte_fib.c | 20 +++++++++++++++++++- lib/librte_fib/rte_fib.h | 22 ++++++++++++++++++++++ lib/librte_fib/rte_fib_version.map | 1 + 5 files changed, 63 insertions(+), 14 deletions(-) diff --git a/lib/librte_fib/dir24_8.c b/lib/librte_fib/dir24_8.c index c9dce3c..825d061 100644 --- a/lib/librte_fib/dir24_8.c +++ b/lib/librte_fib/dir24_8.c @@ -45,13 +45,6 @@ struct dir24_8_tbl { #define ROUNDUP(x, y) RTE_ALIGN_CEIL(x, (1 << (32 - y))) -enum lookup_type { - MACRO, - INLINE, - UNI -}; -enum lookup_type test_lookup = MACRO; - static inline void * get_tbl24_p(struct dir24_8_tbl *dp, uint32_t ip, uint8_t nh_sz) { @@ -253,11 +246,18 @@ dir24_8_lookup_bulk_uni(void *p, const uint32_t *ips, } rte_fib_lookup_fn_t -dir24_8_get_lookup_fn(struct rte_fib_conf *fib_conf) +dir24_8_get_lookup_fn(void *p, enum rte_fib_dir24_8_lookup_type type) { - enum rte_fib_dir24_8_nh_sz nh_sz = fib_conf->dir24_8.nh_sz; + enum rte_fib_dir24_8_nh_sz nh_sz; + struct dir24_8_tbl *dp = p; - if (test_lookup == MACRO) { + if (dp == NULL) + return NULL; + + nh_sz = dp->nh_sz; + + switch (type) { + case RTE_FIB_DIR24_8_SCALAR_MACRO: switch (nh_sz) { case RTE_FIB_DIR24_8_1B: return dir24_8_lookup_bulk_1b; @@ -267,8 +267,10 @@ dir24_8_get_lookup_fn(struct rte_fib_conf *fib_conf) return dir24_8_lookup_bulk_4b; case RTE_FIB_DIR24_8_8B: return dir24_8_lookup_bulk_8b; + default: + return NULL; } - } else if (test_lookup == INLINE) { + case RTE_FIB_DIR24_8_SCALAR_INLINE: switch (nh_sz) { case RTE_FIB_DIR24_8_1B: return dir24_8_lookup_bulk_0; @@ -278,9 +280,15 @@ dir24_8_get_lookup_fn(struct rte_fib_conf *fib_conf) return dir24_8_lookup_bulk_2; case RTE_FIB_DIR24_8_8B: return dir24_8_lookup_bulk_3; + default: + return NULL; } - } else + case RTE_FIB_DIR24_8_SCALAR_UNI: return dir24_8_lookup_bulk_uni; + default: + return NULL; + } + return NULL; } diff --git a/lib/librte_fib/dir24_8.h b/lib/librte_fib/dir24_8.h index 1ec437c..53c5dd2 100644 --- a/lib/librte_fib/dir24_8.h +++ b/lib/librte_fib/dir24_8.h @@ -22,7 +22,7 @@ void dir24_8_free(void *p); rte_fib_lookup_fn_t -dir24_8_get_lookup_fn(struct rte_fib_conf *conf); +dir24_8_get_lookup_fn(void *p, enum rte_fib_dir24_8_lookup_type type); int dir24_8_modify(struct rte_fib *fib, uint32_t ip, uint8_t depth, diff --git a/lib/librte_fib/rte_fib.c b/lib/librte_fib/rte_fib.c index e090808..59120b5 100644 --- a/lib/librte_fib/rte_fib.c +++ b/lib/librte_fib/rte_fib.c @@ -107,7 +107,8 @@ init_dataplane(struct rte_fib *fib, __rte_unused int socket_id, fib->dp = dir24_8_create(dp_name, socket_id, conf); if (fib->dp == NULL) return -rte_errno; - fib->lookup = dir24_8_get_lookup_fn(conf); + fib->lookup = dir24_8_get_lookup_fn(fib->dp, + RTE_FIB_DIR24_8_SCALAR_MACRO); fib->modify = dir24_8_modify; return 0; default: @@ -317,3 +318,20 @@ rte_fib_get_rib(struct rte_fib *fib) { return (fib == NULL) ? NULL : fib->rib; } + +int +rte_fib_set_lookup_fn(struct rte_fib *fib, int type) +{ + rte_fib_lookup_fn_t fn; + + switch (fib->type) { + case RTE_FIB_DIR24_8: + fn = dir24_8_get_lookup_fn(fib->dp, type); + if (fn == NULL) + return -EINVAL; + fib->lookup = fn; + return 0; + default: + return -EINVAL; + } +} diff --git a/lib/librte_fib/rte_fib.h b/lib/librte_fib/rte_fib.h index d06c5ef..0e98775 100644 --- a/lib/librte_fib/rte_fib.h +++ b/lib/librte_fib/rte_fib.h @@ -47,6 +47,12 @@ enum rte_fib_dir24_8_nh_sz { RTE_FIB_DIR24_8_8B }; +enum rte_fib_dir24_8_lookup_type { + RTE_FIB_DIR24_8_SCALAR_MACRO, + RTE_FIB_DIR24_8_SCALAR_INLINE, + RTE_FIB_DIR24_8_SCALAR_UNI +}; + /** FIB configuration structure */ struct rte_fib_conf { enum rte_fib_type type; /**< Type of FIB struct */ @@ -185,4 +191,20 @@ __rte_experimental struct rte_rib * rte_fib_get_rib(struct rte_fib *fib); +/** + * Set lookup function based on type + * + * @param fib + * FIB object handle + * @param type + * type of lookup function + * + * @return + * -EINVAL on failure + * 0 on success + */ +__rte_experimental +int +rte_fib_set_lookup_fn(struct rte_fib *fib, int type); + #endif /* _RTE_FIB_H_ */ diff --git a/lib/librte_fib/rte_fib_version.map b/lib/librte_fib/rte_fib_version.map index 9527417..216af66 100644 --- a/lib/librte_fib/rte_fib_version.map +++ b/lib/librte_fib/rte_fib_version.map @@ -9,6 +9,7 @@ EXPERIMENTAL { rte_fib_lookup_bulk; rte_fib_get_dp; rte_fib_get_rib; + rte_fib_set_lookup_fn; rte_fib6_add; rte_fib6_create; From patchwork Mon Mar 9 12:43:43 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Vladimir Medvedkin X-Patchwork-Id: 66456 X-Patchwork-Delegate: david.marchand@redhat.com Return-Path: X-Original-To: patchwork@inbox.dpdk.org Delivered-To: patchwork@inbox.dpdk.org Received: from dpdk.org (dpdk.org [92.243.14.124]) by inbox.dpdk.org (Postfix) with ESMTP id 8BDBBA052E; Mon, 9 Mar 2020 13:44:18 +0100 (CET) Received: from [92.243.14.124] (localhost [127.0.0.1]) by dpdk.org (Postfix) with ESMTP id 129FB1C0CB; Mon, 9 Mar 2020 13:43:58 +0100 (CET) Received: from mga14.intel.com (mga14.intel.com [192.55.52.115]) by dpdk.org (Postfix) with ESMTP id 6983E1C06D for ; Mon, 9 Mar 2020 13:43:55 +0100 (CET) X-Amp-Result: SKIPPED(no attachment in message) X-Amp-File-Uploaded: False Received: from fmsmga002.fm.intel.com ([10.253.24.26]) by fmsmga103.fm.intel.com with ESMTP/TLS/DHE-RSA-AES256-GCM-SHA384; 09 Mar 2020 05:43:55 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.70,533,1574150400"; d="scan'208";a="276500021" Received: from silpixa00400072.ir.intel.com ([10.237.222.213]) by fmsmga002.fm.intel.com with ESMTP; 09 Mar 2020 05:43:54 -0700 From: Vladimir Medvedkin To: dev@dpdk.org Cc: konstantin.ananyev@intel.com, bruce.richardson@intel.com Date: Mon, 9 Mar 2020 12:43:43 +0000 Message-Id: <1583757826-375246-4-git-send-email-vladimir.medvedkin@intel.com> X-Mailer: git-send-email 2.7.4 In-Reply-To: <1583757826-375246-1-git-send-email-vladimir.medvedkin@intel.com> References: <1583757826-375246-1-git-send-email-vladimir.medvedkin@intel.com> Subject: [dpdk-dev] [PATCH 3/6] fib: introduce AVX512 lookup X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" Add new lookup implementation for DIR24_8 algorithm using AVX512 instruction set Signed-off-by: Vladimir Medvedkin --- lib/librte_fib/dir24_8.c | 71 ++++++++++++++++++++++++ lib/librte_fib/dir24_8_avx512.h | 116 ++++++++++++++++++++++++++++++++++++++++ lib/librte_fib/rte_fib.h | 3 +- 3 files changed, 189 insertions(+), 1 deletion(-) create mode 100644 lib/librte_fib/dir24_8_avx512.h diff --git a/lib/librte_fib/dir24_8.c b/lib/librte_fib/dir24_8.c index 825d061..9f51dfc 100644 --- a/lib/librte_fib/dir24_8.c +++ b/lib/librte_fib/dir24_8.c @@ -245,6 +245,62 @@ dir24_8_lookup_bulk_uni(void *p, const uint32_t *ips, } } +#ifdef __AVX512F__ + +#include "dir24_8_avx512.h" + +static void +rte_dir24_8_vec_lookup_bulk_1b(void *p, const uint32_t *ips, + uint64_t *next_hops, const unsigned int n) +{ + uint32_t i; + for (i = 0; i < (n / 16); i++) + dir24_8_vec_lookup_x16(p, ips + i * 16, next_hops + i * 16, + sizeof(uint8_t)); + + dir24_8_lookup_bulk_1b(p, ips + i * 16, next_hops + i * 16, + n - i * 16); +} + +static void +rte_dir24_8_vec_lookup_bulk_2b(void *p, const uint32_t *ips, + uint64_t *next_hops, const unsigned int n) +{ + uint32_t i; + for (i = 0; i < (n / 16); i++) + dir24_8_vec_lookup_x16(p, ips + i * 16, next_hops + i * 16, + sizeof(uint16_t)); + + dir24_8_lookup_bulk_2b(p, ips + i * 16, next_hops + i * 16, + n - i * 16); +} + +static void +rte_dir24_8_vec_lookup_bulk_4b(void *p, const uint32_t *ips, + uint64_t *next_hops, const unsigned int n) +{ + uint32_t i; + for (i = 0; i < (n / 16); i++) + dir24_8_vec_lookup_x16(p, ips + i * 16, next_hops + i * 16, + sizeof(uint32_t)); + + dir24_8_lookup_bulk_4b(p, ips + i * 16, next_hops + i * 16, + n - i * 16); +} + +static void +rte_dir24_8_vec_lookup_bulk_8b(void *p, const uint32_t *ips, + uint64_t *next_hops, const unsigned int n) +{ + uint32_t i; + for (i = 0; i < (n / 8); i++) + dir24_8_vec_lookup_x8_8b(p, ips + i * 8, next_hops + i * 8); + + dir24_8_lookup_bulk_8b(p, ips + i * 8, next_hops + i * 8, n - i * 8); +} + +#endif /* __AVX512F__ */ + rte_fib_lookup_fn_t dir24_8_get_lookup_fn(void *p, enum rte_fib_dir24_8_lookup_type type) { @@ -285,6 +341,21 @@ dir24_8_get_lookup_fn(void *p, enum rte_fib_dir24_8_lookup_type type) } case RTE_FIB_DIR24_8_SCALAR_UNI: return dir24_8_lookup_bulk_uni; +#ifdef __AVX512F__ + case RTE_FIB_DIR24_8_VECTOR: + switch (nh_sz) { + case RTE_FIB_DIR24_8_1B: + return rte_dir24_8_vec_lookup_bulk_1b; + case RTE_FIB_DIR24_8_2B: + return rte_dir24_8_vec_lookup_bulk_2b; + case RTE_FIB_DIR24_8_4B: + return rte_dir24_8_vec_lookup_bulk_4b; + case RTE_FIB_DIR24_8_8B: + return rte_dir24_8_vec_lookup_bulk_8b; + default: + return NULL; + } +#endif default: return NULL; } diff --git a/lib/librte_fib/dir24_8_avx512.h b/lib/librte_fib/dir24_8_avx512.h new file mode 100644 index 0000000..3b6680c --- /dev/null +++ b/lib/librte_fib/dir24_8_avx512.h @@ -0,0 +1,116 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2020 Intel Corporation + */ + +#ifndef _DIR248_AVX512_H_ +#define _DIR248_AVX512_H_ + +#include + +static __rte_always_inline void +dir24_8_vec_lookup_x16(void *p, const uint32_t *ips, + uint64_t *next_hops, int size) +{ + struct dir24_8_tbl *dp = (struct dir24_8_tbl *)p; + __mmask16 msk_ext; + __mmask16 exp_msk = 0x5555; + __m512i ip_vec, idxes, res, bytes; + const __m512i zero = _mm512_set1_epi32(0); + const __m512i lsb = _mm512_set1_epi32(1); + const __m512i lsbyte_msk = _mm512_set1_epi32(0xff); + __m512i tmp1, tmp2, res_msk; + __m256i tmp256; + /* used to mask gather values if size is 1/2 (8/16 bit next hops) */ + if (size == sizeof(uint8_t)) + res_msk = _mm512_set1_epi32(UINT8_MAX); + else if (size == sizeof(uint16_t)) + res_msk = _mm512_set1_epi32(UINT16_MAX); + + ip_vec = _mm512_loadu_si512(ips); + /* mask 24 most significant bits */ + idxes = _mm512_srli_epi32(ip_vec, 8); + + /** + * lookup in tbl24 + * Put it inside branch to make compiller happy with -O0 + */ + if (size == sizeof(uint8_t)) { + res = _mm512_i32gather_epi32(idxes, (const int *)dp->tbl24, 1); + res = _mm512_and_epi32(res, res_msk); + } else if (size == sizeof(uint16_t)) { + res = _mm512_i32gather_epi32(idxes, (const int *)dp->tbl24, 2); + res = _mm512_and_epi32(res, res_msk); + } else + res = _mm512_i32gather_epi32(idxes, (const int *)dp->tbl24, 4); + + /* get extended entries indexes */ + msk_ext = _mm512_test_epi32_mask(res, lsb); + + if (msk_ext != 0) { + idxes = _mm512_srli_epi32(res, 1); + idxes = _mm512_slli_epi32(idxes, 8); + bytes = _mm512_and_epi32(ip_vec, lsbyte_msk); + idxes = _mm512_maskz_add_epi32(msk_ext, idxes, bytes); + if (size == sizeof(uint8_t)) { + idxes = _mm512_mask_i32gather_epi32(zero, msk_ext, + idxes, (const int *)dp->tbl8, 1); + idxes = _mm512_and_epi32(idxes, res_msk); + } else if (size == sizeof(uint16_t)) { + idxes = _mm512_mask_i32gather_epi32(zero, msk_ext, + idxes, (const int *)dp->tbl8, 2); + idxes = _mm512_and_epi32(idxes, res_msk); + } else + idxes = _mm512_mask_i32gather_epi32(zero, msk_ext, + idxes, (const int *)dp->tbl8, 4); + + res = _mm512_mask_blend_epi32(msk_ext, res, idxes); + } + + res = _mm512_srli_epi32(res, 1); + tmp1 = _mm512_maskz_expand_epi32(exp_msk, res); + tmp256 = _mm512_extracti32x8_epi32(res, 1); + tmp2 = _mm512_maskz_expand_epi32(exp_msk, + _mm512_castsi256_si512(tmp256)); + _mm512_storeu_si512(next_hops, tmp1); + _mm512_storeu_si512(next_hops + 8, tmp2); +} + +static __rte_always_inline void +dir24_8_vec_lookup_x8_8b(void *p, const uint32_t *ips, + uint64_t *next_hops) +{ + struct dir24_8_tbl *dp = (struct dir24_8_tbl *)p; + const __m512i zero = _mm512_set1_epi32(0); + const __m512i lsbyte_msk = _mm512_set1_epi64(0xff); + const __m512i lsb = _mm512_set1_epi64(1); + __m512i res, idxes, bytes; + __m256i idxes_256, ip_vec; + __mmask8 msk_ext; + + ip_vec = _mm256_loadu_si256((const void *)ips); + /* mask 24 most significant bits */ + idxes_256 = _mm256_srli_epi32(ip_vec, 8); + + /* lookup in tbl24 */ + res = _mm512_i32gather_epi64(idxes_256, (const void *)dp->tbl24, 8); + + /* get extended entries indexes */ + msk_ext = _mm512_test_epi64_mask(res, lsb); + + if (msk_ext != 0) { + bytes = _mm512_cvtepi32_epi64(ip_vec); + idxes = _mm512_srli_epi64(res, 1); + idxes = _mm512_slli_epi64(idxes, 8); + bytes = _mm512_and_epi64(bytes, lsbyte_msk); + idxes = _mm512_maskz_add_epi64(msk_ext, idxes, bytes); + idxes = _mm512_mask_i64gather_epi64(zero, msk_ext, idxes, + (const void *)dp->tbl8, 8); + + res = _mm512_mask_blend_epi64(msk_ext, res, idxes); + } + + res = _mm512_srli_epi64(res, 1); + _mm512_storeu_si512(next_hops, res); +} + +#endif /* _DIR248_AVX512_H_ */ diff --git a/lib/librte_fib/rte_fib.h b/lib/librte_fib/rte_fib.h index 0e98775..89d0f12 100644 --- a/lib/librte_fib/rte_fib.h +++ b/lib/librte_fib/rte_fib.h @@ -50,7 +50,8 @@ enum rte_fib_dir24_8_nh_sz { enum rte_fib_dir24_8_lookup_type { RTE_FIB_DIR24_8_SCALAR_MACRO, RTE_FIB_DIR24_8_SCALAR_INLINE, - RTE_FIB_DIR24_8_SCALAR_UNI + RTE_FIB_DIR24_8_SCALAR_UNI, + RTE_FIB_DIR24_8_VECTOR }; /** FIB configuration structure */ From patchwork Mon Mar 9 12:43:44 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Vladimir Medvedkin X-Patchwork-Id: 66457 X-Patchwork-Delegate: david.marchand@redhat.com Return-Path: X-Original-To: patchwork@inbox.dpdk.org Delivered-To: patchwork@inbox.dpdk.org Received: from dpdk.org (dpdk.org [92.243.14.124]) by inbox.dpdk.org (Postfix) with ESMTP id B5C7BA052E; Mon, 9 Mar 2020 13:44:30 +0100 (CET) Received: from [92.243.14.124] (localhost [127.0.0.1]) by dpdk.org (Postfix) with ESMTP id B1B941C0DC; Mon, 9 Mar 2020 13:43:59 +0100 (CET) Received: from mga14.intel.com (mga14.intel.com [192.55.52.115]) by dpdk.org (Postfix) with ESMTP id 925D01C0B7 for ; Mon, 9 Mar 2020 13:43:56 +0100 (CET) X-Amp-Result: SKIPPED(no attachment in message) X-Amp-File-Uploaded: False Received: from fmsmga002.fm.intel.com ([10.253.24.26]) by fmsmga103.fm.intel.com with ESMTP/TLS/DHE-RSA-AES256-GCM-SHA384; 09 Mar 2020 05:43:56 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.70,533,1574150400"; d="scan'208";a="276500035" Received: from silpixa00400072.ir.intel.com ([10.237.222.213]) by fmsmga002.fm.intel.com with ESMTP; 09 Mar 2020 05:43:55 -0700 From: Vladimir Medvedkin To: dev@dpdk.org Cc: konstantin.ananyev@intel.com, bruce.richardson@intel.com Date: Mon, 9 Mar 2020 12:43:44 +0000 Message-Id: <1583757826-375246-5-git-send-email-vladimir.medvedkin@intel.com> X-Mailer: git-send-email 2.7.4 In-Reply-To: <1583757826-375246-1-git-send-email-vladimir.medvedkin@intel.com> References: <1583757826-375246-1-git-send-email-vladimir.medvedkin@intel.com> Subject: [dpdk-dev] [PATCH 4/6] fib6: make lookup function type configurable X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" Add type argument to trie_get_lookup_fn() Now it only supports RTE_FIB6_TRIE_SCALAR Add new rte_fib6_set_lookup_fn() - user can change lookup function type runtime. Signed-off-by: Vladimir Medvedkin --- lib/librte_fib/rte_fib6.c | 19 ++++++++++++++++++- lib/librte_fib/rte_fib6.h | 20 ++++++++++++++++++++ lib/librte_fib/rte_fib_version.map | 1 + lib/librte_fib/trie.c | 25 ++++++++++++++----------- lib/librte_fib/trie.h | 2 +- 5 files changed, 54 insertions(+), 13 deletions(-) diff --git a/lib/librte_fib/rte_fib6.c b/lib/librte_fib/rte_fib6.c index a1f0db8..9eff712 100644 --- a/lib/librte_fib/rte_fib6.c +++ b/lib/librte_fib/rte_fib6.c @@ -107,7 +107,7 @@ init_dataplane(struct rte_fib6 *fib, __rte_unused int socket_id, fib->dp = trie_create(dp_name, socket_id, conf); if (fib->dp == NULL) return -rte_errno; - fib->lookup = rte_trie_get_lookup_fn(conf); + fib->lookup = trie_get_lookup_fn(fib->dp, RTE_FIB6_TRIE_SCALAR); fib->modify = trie_modify; return 0; default: @@ -319,3 +319,20 @@ rte_fib6_get_rib(struct rte_fib6 *fib) { return (fib == NULL) ? NULL : fib->rib; } + +int +rte_fib6_set_lookup_fn(struct rte_fib6 *fib, int type) +{ + rte_fib6_lookup_fn_t fn; + + switch (fib->type) { + case RTE_FIB6_TRIE: + fn = trie_get_lookup_fn(fib->dp, type); + if (fn == NULL) + return -EINVAL; + fib->lookup = fn; + return 0; + default: + return -EINVAL; + } +} diff --git a/lib/librte_fib/rte_fib6.h b/lib/librte_fib/rte_fib6.h index 4268704..7d2fd9f 100644 --- a/lib/librte_fib/rte_fib6.h +++ b/lib/librte_fib/rte_fib6.h @@ -48,6 +48,10 @@ enum rte_fib_trie_nh_sz { RTE_FIB6_TRIE_8B }; +enum rte_fib_trie_lookup_type { + RTE_FIB6_TRIE_SCALAR +}; + /** FIB configuration structure */ struct rte_fib6_conf { enum rte_fib6_type type; /**< Type of FIB struct */ @@ -190,4 +194,20 @@ __rte_experimental struct rte_rib6 * rte_fib6_get_rib(struct rte_fib6 *fib); +/** + * Set lookup function based on type + * + * @param fib + * FIB object handle + * @param type + * type of lookup function + * + * @return + * -EINVAL on failure + * 0 on success + */ +__rte_experimental +int +rte_fib6_set_lookup_fn(struct rte_fib6 *fib, int type); + #endif /* _RTE_FIB6_H_ */ diff --git a/lib/librte_fib/rte_fib_version.map b/lib/librte_fib/rte_fib_version.map index 216af66..9d1e181 100644 --- a/lib/librte_fib/rte_fib_version.map +++ b/lib/librte_fib/rte_fib_version.map @@ -19,6 +19,7 @@ EXPERIMENTAL { rte_fib6_lookup_bulk; rte_fib6_get_dp; rte_fib6_get_rib; + rte_fib6_set_lookup_fn; local: *; }; diff --git a/lib/librte_fib/trie.c b/lib/librte_fib/trie.c index 2ae2add..63c519a 100644 --- a/lib/librte_fib/trie.c +++ b/lib/librte_fib/trie.c @@ -59,13 +59,6 @@ enum edge { REDGE }; -enum lookup_type { - MACRO, - INLINE, - UNI -}; -static enum lookup_type test_lookup = MACRO; - static inline uint32_t get_tbl24_idx(const uint8_t *ip) { @@ -154,11 +147,18 @@ LOOKUP_FUNC(4b, uint32_t, 2) LOOKUP_FUNC(8b, uint64_t, 3) rte_fib6_lookup_fn_t -rte_trie_get_lookup_fn(struct rte_fib6_conf *conf) +trie_get_lookup_fn(void *p, enum rte_fib_trie_lookup_type type) { - enum rte_fib_trie_nh_sz nh_sz = conf->trie.nh_sz; + enum rte_fib_trie_nh_sz nh_sz; + struct rte_trie_tbl *dp = p; - if (test_lookup == MACRO) { + if (dp == NULL) + return NULL; + + nh_sz = dp->nh_sz; + + switch (type) { + case RTE_FIB6_TRIE_SCALAR: switch (nh_sz) { case RTE_FIB6_TRIE_2B: return rte_trie_lookup_bulk_2b; @@ -166,9 +166,12 @@ rte_trie_get_lookup_fn(struct rte_fib6_conf *conf) return rte_trie_lookup_bulk_4b; case RTE_FIB6_TRIE_8B: return rte_trie_lookup_bulk_8b; + default: + return NULL; } + default: + return NULL; } - return NULL; } diff --git a/lib/librte_fib/trie.h b/lib/librte_fib/trie.h index bb750c5..0d5ef9a 100644 --- a/lib/librte_fib/trie.h +++ b/lib/librte_fib/trie.h @@ -22,7 +22,7 @@ void trie_free(void *p); rte_fib6_lookup_fn_t -rte_trie_get_lookup_fn(struct rte_fib6_conf *fib_conf); +trie_get_lookup_fn(void *p, enum rte_fib_trie_lookup_type type); int trie_modify(struct rte_fib6 *fib, const uint8_t ip[RTE_FIB6_IPV6_ADDR_SIZE], From patchwork Mon Mar 9 12:43:45 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Vladimir Medvedkin X-Patchwork-Id: 66458 X-Patchwork-Delegate: david.marchand@redhat.com Return-Path: X-Original-To: patchwork@inbox.dpdk.org Delivered-To: patchwork@inbox.dpdk.org Received: from dpdk.org (dpdk.org [92.243.14.124]) by inbox.dpdk.org (Postfix) with ESMTP id BA4EAA052E; Mon, 9 Mar 2020 13:44:41 +0100 (CET) Received: from [92.243.14.124] (localhost [127.0.0.1]) by dpdk.org (Postfix) with ESMTP id D25E91C114; Mon, 9 Mar 2020 13:44:00 +0100 (CET) Received: from mga14.intel.com (mga14.intel.com [192.55.52.115]) by dpdk.org (Postfix) with ESMTP id E45631C0C7 for ; Mon, 9 Mar 2020 13:43:57 +0100 (CET) X-Amp-Result: SKIPPED(no attachment in message) X-Amp-File-Uploaded: False Received: from fmsmga002.fm.intel.com ([10.253.24.26]) by fmsmga103.fm.intel.com with ESMTP/TLS/DHE-RSA-AES256-GCM-SHA384; 09 Mar 2020 05:43:57 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.70,533,1574150400"; d="scan'208";a="276500050" Received: from silpixa00400072.ir.intel.com ([10.237.222.213]) by fmsmga002.fm.intel.com with ESMTP; 09 Mar 2020 05:43:56 -0700 From: Vladimir Medvedkin To: dev@dpdk.org Cc: konstantin.ananyev@intel.com, bruce.richardson@intel.com Date: Mon, 9 Mar 2020 12:43:45 +0000 Message-Id: <1583757826-375246-6-git-send-email-vladimir.medvedkin@intel.com> X-Mailer: git-send-email 2.7.4 In-Reply-To: <1583757826-375246-1-git-send-email-vladimir.medvedkin@intel.com> References: <1583757826-375246-1-git-send-email-vladimir.medvedkin@intel.com> Subject: [dpdk-dev] [PATCH 5/6] fib6: introduce AVX512 lookup X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" Add new lookup implementation for FIB6 trie algorithm using AVX512 instruction set Signed-off-by: Vladimir Medvedkin --- lib/librte_fib/rte_fib6.h | 3 +- lib/librte_fib/trie.c | 58 +++++++++++ lib/librte_fib/trie_avx512.h | 231 +++++++++++++++++++++++++++++++++++++++++++ 3 files changed, 291 insertions(+), 1 deletion(-) create mode 100644 lib/librte_fib/trie_avx512.h diff --git a/lib/librte_fib/rte_fib6.h b/lib/librte_fib/rte_fib6.h index 7d2fd9f..f1f8f38 100644 --- a/lib/librte_fib/rte_fib6.h +++ b/lib/librte_fib/rte_fib6.h @@ -49,7 +49,8 @@ enum rte_fib_trie_nh_sz { }; enum rte_fib_trie_lookup_type { - RTE_FIB6_TRIE_SCALAR + RTE_FIB6_TRIE_SCALAR, + RTE_FIB6_TRIE_VECTOR }; /** FIB configuration structure */ diff --git a/lib/librte_fib/trie.c b/lib/librte_fib/trie.c index 63c519a..b983b3a 100644 --- a/lib/librte_fib/trie.c +++ b/lib/librte_fib/trie.c @@ -146,6 +146,51 @@ LOOKUP_FUNC(2b, uint16_t, 1) LOOKUP_FUNC(4b, uint32_t, 2) LOOKUP_FUNC(8b, uint64_t, 3) +#ifdef __AVX512F__ + +#include "trie_avx512.h" + +static void +rte_trie_vec_lookup_bulk_2b(void *p, uint8_t ips[][RTE_FIB6_IPV6_ADDR_SIZE], + uint64_t *next_hops, const unsigned int n) +{ + uint32_t i; + for (i = 0; i < (n / 16); i++) { + trie_vec_lookup_x16(p, (uint8_t (*)[16])&ips[i * 16][0], + next_hops + i * 16, sizeof(uint16_t)); + } + rte_trie_lookup_bulk_2b(p, (uint8_t (*)[16])&ips[i * 16][0], + next_hops + i * 16, n - i * 16); +} + +static void +rte_trie_vec_lookup_bulk_4b(void *p, uint8_t ips[][RTE_FIB6_IPV6_ADDR_SIZE], + uint64_t *next_hops, const unsigned int n) +{ + uint32_t i; + for (i = 0; i < (n / 16); i++) { + trie_vec_lookup_x16(p, (uint8_t (*)[16])&ips[i * 16][0], + next_hops + i * 16, sizeof(uint32_t)); + } + rte_trie_lookup_bulk_4b(p, (uint8_t (*)[16])&ips[i * 16][0], + next_hops + i * 16, n - i * 16); +} + +static void +rte_trie_vec_lookup_bulk_8b(void *p, uint8_t ips[][RTE_FIB6_IPV6_ADDR_SIZE], + uint64_t *next_hops, const unsigned int n) +{ + uint32_t i; + for (i = 0; i < (n / 8); i++) { + trie_vec_lookup_x8_8b(p, (uint8_t (*)[16])&ips[i * 8][0], + next_hops + i * 8); + } + rte_trie_lookup_bulk_8b(p, (uint8_t (*)[16])&ips[i * 8][0], + next_hops + i * 8, n - i * 8); +} + +#endif /* __AVX512F__ */ + rte_fib6_lookup_fn_t trie_get_lookup_fn(void *p, enum rte_fib_trie_lookup_type type) { @@ -169,6 +214,19 @@ trie_get_lookup_fn(void *p, enum rte_fib_trie_lookup_type type) default: return NULL; } +#ifdef __AVX512F__ + case RTE_FIB6_TRIE_VECTOR: + switch (nh_sz) { + case RTE_FIB6_TRIE_2B: + return rte_trie_vec_lookup_bulk_2b; + case RTE_FIB6_TRIE_4B: + return rte_trie_vec_lookup_bulk_4b; + case RTE_FIB6_TRIE_8B: + return rte_trie_vec_lookup_bulk_8b; + default: + return NULL; + } +#endif /* __AVX512F__ */ default: return NULL; } diff --git a/lib/librte_fib/trie_avx512.h b/lib/librte_fib/trie_avx512.h new file mode 100644 index 0000000..f60bbd2 --- /dev/null +++ b/lib/librte_fib/trie_avx512.h @@ -0,0 +1,231 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2020 Intel Corporation + */ + +#ifndef _TRIE_AVX512_H_ +#define _TRIE_AVX512_H_ + +#include + +static __rte_always_inline void +transpose_x16(uint8_t ips[16][RTE_FIB6_IPV6_ADDR_SIZE], + __m512i *first, __m512i *second, __m512i *third, __m512i *fourth) +{ + __m512i tmp1, tmp2, tmp3, tmp4; + __m512i tmp5, tmp6, tmp7, tmp8; + const rte_zmm_t perm_idxes = { + .u32 = { 0, 4, 8, 12, 2, 6, 10, 14, + 1, 5, 9, 13, 3, 7, 11, 15 + }, + }; + + /* load all ip addresses */ + tmp1 = _mm512_loadu_si512(&ips[0][0]); + tmp2 = _mm512_loadu_si512(&ips[4][0]); + tmp3 = _mm512_loadu_si512(&ips[8][0]); + tmp4 = _mm512_loadu_si512(&ips[12][0]); + + /* transpose 4 byte chunks of 16 ips */ + tmp5 = _mm512_unpacklo_epi32(tmp1, tmp2); + tmp7 = _mm512_unpackhi_epi32(tmp1, tmp2); + tmp6 = _mm512_unpacklo_epi32(tmp3, tmp4); + tmp8 = _mm512_unpackhi_epi32(tmp3, tmp4); + + tmp1 = _mm512_unpacklo_epi32(tmp5, tmp6); + tmp3 = _mm512_unpackhi_epi32(tmp5, tmp6); + tmp2 = _mm512_unpacklo_epi32(tmp7, tmp8); + tmp4 = _mm512_unpackhi_epi32(tmp7, tmp8); + + /* first 4-byte chunks of ips[] */ + *first = _mm512_permutexvar_epi32(perm_idxes.z, tmp1); + /* second 4-byte chunks of ips[] */ + *second = _mm512_permutexvar_epi32(perm_idxes.z, tmp3); + /* third 4-byte chunks of ips[] */ + *third = _mm512_permutexvar_epi32(perm_idxes.z, tmp2); + /* fourth 4-byte chunks of ips[] */ + *fourth = _mm512_permutexvar_epi32(perm_idxes.z, tmp4); +} + +static __rte_always_inline void +transpose_x8(uint8_t ips[8][RTE_FIB6_IPV6_ADDR_SIZE], + __m512i *first, __m512i *second) +{ + __m512i tmp1, tmp2, tmp3, tmp4; + const rte_zmm_t perm_idxes = { + .u64 = { 0, 2, 4, 6, 1, 3, 5, 7 + }, + }; + + tmp1 = _mm512_loadu_si512(&ips[0][0]); + tmp2 = _mm512_loadu_si512(&ips[4][0]); + + tmp3 = _mm512_unpacklo_epi64(tmp1, tmp2); + *first = _mm512_permutexvar_epi64(perm_idxes.z, tmp3); + tmp4 = _mm512_unpackhi_epi64(tmp1, tmp2); + *second = _mm512_permutexvar_epi64(perm_idxes.z, tmp4); +} + +static __rte_always_inline void +trie_vec_lookup_x16(void *p, uint8_t ips[16][RTE_FIB6_IPV6_ADDR_SIZE], + uint64_t *next_hops, int size) +{ + struct rte_trie_tbl *dp = (struct rte_trie_tbl *)p; + const __m512i zero = _mm512_set1_epi32(0); + const __m512i lsb = _mm512_set1_epi32(1); + const __m512i two_lsb = _mm512_set1_epi32(3); + __m512i first, second, third, fourth; /*< IPv6 four byte chunks */ + __m512i idxes, res, shuf_idxes; + __m512i tmp, tmp2, bytes, byte_chunk, base_idxes; + /* used to mask gather values if size is 2 (16 bit next hops) */ + const __m512i res_msk = _mm512_set1_epi32(UINT16_MAX); + const rte_zmm_t bswap = { + .u8 = { 2, 1, 0, 255, 6, 5, 4, 255, + 10, 9, 8, 255, 14, 13, 12, 255, + 2, 1, 0, 255, 6, 5, 4, 255, + 10, 9, 8, 255, 14, 13, 12, 255, + 2, 1, 0, 255, 6, 5, 4, 255, + 10, 9, 8, 255, 14, 13, 12, 255, + 2, 1, 0, 255, 6, 5, 4, 255, + 10, 9, 8, 255, 14, 13, 12, 255 + }, + }; + const __mmask64 k = 0x1111111111111111; + int i = 3; + __mmask16 msk_ext, new_msk; + __mmask16 exp_msk = 0x5555; + + transpose_x16(ips, &first, &second, &third, &fourth); + + /* get_tbl24_idx() for every 4 byte chunk */ + idxes = _mm512_shuffle_epi8(first, bswap.z); + + /** + * lookup in tbl24 + * Put it inside branch to make compiller happy with -O0 + */ + if (size == sizeof(uint16_t)) { + res = _mm512_i32gather_epi32(idxes, (const int *)dp->tbl24, 2); + res = _mm512_and_epi32(res, res_msk); + } else + res = _mm512_i32gather_epi32(idxes, (const int *)dp->tbl24, 4); + + + /* get extended entries indexes */ + msk_ext = _mm512_test_epi32_mask(res, lsb); + + tmp = _mm512_srli_epi32(res, 1); + + /* idxes to retrieve bytes */ + shuf_idxes = _mm512_setr_epi32(3, 7, 11, 15, + 19, 23, 27, 31, + 35, 39, 43, 47, + 51, 55, 59, 63); + + base_idxes = _mm512_setr_epi32(0, 4, 8, 12, + 16, 20, 24, 28, + 32, 36, 40, 44, + 48, 52, 56, 60); + + /* traverse down the trie */ + while (msk_ext) { + idxes = _mm512_maskz_slli_epi32(msk_ext, tmp, 8); + byte_chunk = (i < 8) ? + ((i >= 4) ? second : first) : + ((i >= 12) ? fourth : third); + bytes = _mm512_maskz_shuffle_epi8(k, byte_chunk, shuf_idxes); + idxes = _mm512_maskz_add_epi32(msk_ext, idxes, bytes); + if (size == sizeof(uint16_t)) { + tmp = _mm512_mask_i32gather_epi32(zero, msk_ext, + idxes, (const int *)dp->tbl8, 2); + tmp = _mm512_and_epi32(tmp, res_msk); + } else + tmp = _mm512_mask_i32gather_epi32(zero, msk_ext, + idxes, (const int *)dp->tbl8, 4); + new_msk = _mm512_test_epi32_mask(tmp, lsb); + res = _mm512_mask_blend_epi32(msk_ext ^ new_msk, res, tmp); + tmp = _mm512_srli_epi32(tmp, 1); + msk_ext = new_msk; + + shuf_idxes = _mm512_maskz_add_epi8(k, shuf_idxes, lsb); + shuf_idxes = _mm512_and_epi32(shuf_idxes, two_lsb); + shuf_idxes = _mm512_maskz_add_epi8(k, shuf_idxes, base_idxes); + i++; + } + + res = _mm512_srli_epi32(res, 1); + tmp = _mm512_maskz_expand_epi32(exp_msk, res); + __m256i tmp256; + tmp256 = _mm512_extracti32x8_epi32(res, 1); + tmp2 = _mm512_maskz_expand_epi32(exp_msk, + _mm512_castsi256_si512(tmp256)); + _mm512_storeu_si512(next_hops, tmp); + _mm512_storeu_si512(next_hops + 8, tmp2); +} + +static void +trie_vec_lookup_x8_8b(void *p, uint8_t ips[8][RTE_FIB6_IPV6_ADDR_SIZE], + uint64_t *next_hops) +{ + struct rte_trie_tbl *dp = (struct rte_trie_tbl *)p; + const __m512i zero = _mm512_set1_epi32(0); + const __m512i lsb = _mm512_set1_epi32(1); + const __m512i three_lsb = _mm512_set1_epi32(7); + __m512i first, second; /*< IPv6 eight byte chunks */ + __m512i idxes, res, shuf_idxes; + __m512i tmp, bytes, byte_chunk, base_idxes; + const rte_zmm_t bswap = { + .u8 = { 2, 1, 0, 255, 255, 255, 255, 255, + 10, 9, 8, 255, 255, 255, 255, 255, + 2, 1, 0, 255, 255, 255, 255, 255, + 10, 9, 8, 255, 255, 255, 255, 255, + 2, 1, 0, 255, 255, 255, 255, 255, + 10, 9, 8, 255, 255, 255, 255, 255, + 2, 1, 0, 255, 255, 255, 255, 255, + 10, 9, 8, 255, 255, 255, 255, 255 + }, + }; + const __mmask64 k = 0x101010101010101; + int i = 3; + __mmask8 msk_ext, new_msk; + + transpose_x8(ips, &first, &second); + + /* get_tbl24_idx() for every 4 byte chunk */ + idxes = _mm512_shuffle_epi8(first, bswap.z); + + /* lookup in tbl24 */ + res = _mm512_i64gather_epi64(idxes, (const void *)dp->tbl24, 8); + /* get extended entries indexes */ + msk_ext = _mm512_test_epi64_mask(res, lsb); + + tmp = _mm512_srli_epi64(res, 1); + + /* idxes to retrieve bytes */ + shuf_idxes = _mm512_setr_epi64(3, 11, 19, 27, 35, 43, 51, 59); + + base_idxes = _mm512_setr_epi64(0, 8, 16, 24, 32, 40, 48, 56); + + /* traverse down the trie */ + while (msk_ext) { + idxes = _mm512_maskz_slli_epi64(msk_ext, tmp, 8); + byte_chunk = (i < 8) ? first : second; + bytes = _mm512_maskz_shuffle_epi8(k, byte_chunk, shuf_idxes); + idxes = _mm512_maskz_add_epi64(msk_ext, idxes, bytes); + tmp = _mm512_mask_i64gather_epi64(zero, msk_ext, + idxes, (const void *)dp->tbl8, 8); + new_msk = _mm512_test_epi64_mask(tmp, lsb); + res = _mm512_mask_blend_epi64(msk_ext ^ new_msk, res, tmp); + tmp = _mm512_srli_epi64(tmp, 1); + msk_ext = new_msk; + + shuf_idxes = _mm512_maskz_add_epi8(k, shuf_idxes, lsb); + shuf_idxes = _mm512_and_epi64(shuf_idxes, three_lsb); + shuf_idxes = _mm512_maskz_add_epi8(k, shuf_idxes, base_idxes); + i++; + } + + res = _mm512_srli_epi64(res, 1); + _mm512_storeu_si512(next_hops, res); +} + +#endif /* _TRIE_AVX512_H_ */ From patchwork Mon Mar 9 12:43:46 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Vladimir Medvedkin X-Patchwork-Id: 66459 X-Patchwork-Delegate: david.marchand@redhat.com Return-Path: X-Original-To: patchwork@inbox.dpdk.org Delivered-To: patchwork@inbox.dpdk.org Received: from dpdk.org (dpdk.org [92.243.14.124]) by inbox.dpdk.org (Postfix) with ESMTP id 57958A052E; Mon, 9 Mar 2020 13:44:53 +0100 (CET) Received: from [92.243.14.124] (localhost [127.0.0.1]) by dpdk.org (Postfix) with ESMTP id 0AE641C119; Mon, 9 Mar 2020 13:44:02 +0100 (CET) Received: from mga14.intel.com (mga14.intel.com [192.55.52.115]) by dpdk.org (Postfix) with ESMTP id 1C6451C0C7 for ; Mon, 9 Mar 2020 13:43:58 +0100 (CET) X-Amp-Result: SKIPPED(no attachment in message) X-Amp-File-Uploaded: False Received: from fmsmga002.fm.intel.com ([10.253.24.26]) by fmsmga103.fm.intel.com with ESMTP/TLS/DHE-RSA-AES256-GCM-SHA384; 09 Mar 2020 05:43:58 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.70,533,1574150400"; d="scan'208";a="276500058" Received: from silpixa00400072.ir.intel.com ([10.237.222.213]) by fmsmga002.fm.intel.com with ESMTP; 09 Mar 2020 05:43:57 -0700 From: Vladimir Medvedkin To: dev@dpdk.org Cc: konstantin.ananyev@intel.com, bruce.richardson@intel.com Date: Mon, 9 Mar 2020 12:43:46 +0000 Message-Id: <1583757826-375246-7-git-send-email-vladimir.medvedkin@intel.com> X-Mailer: git-send-email 2.7.4 In-Reply-To: <1583757826-375246-1-git-send-email-vladimir.medvedkin@intel.com> References: <1583757826-375246-1-git-send-email-vladimir.medvedkin@intel.com> Subject: [dpdk-dev] [PATCH 6/6] app/testfib: add support for different lookup functions X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" Added -v option to switch between different lookup implementations to measure their performance and correctness. Signed-off-by: Vladimir Medvedkin --- app/test-fib/main.c | 58 ++++++++++++++++++++++++++++++++++++++++++++++++++--- 1 file changed, 55 insertions(+), 3 deletions(-) diff --git a/app/test-fib/main.c b/app/test-fib/main.c index 5fb67f3..926b59a 100644 --- a/app/test-fib/main.c +++ b/app/test-fib/main.c @@ -99,6 +99,7 @@ static struct { uint8_t ent_sz; uint8_t rnd_lookup_ips_ratio; uint8_t print_fract; + uint8_t lookup_fn; } config = { .routes_file = NULL, .lookup_ips_file = NULL, @@ -110,7 +111,8 @@ static struct { .tbl8 = DEFAULT_LPM_TBL8, .ent_sz = 4, .rnd_lookup_ips_ratio = 0, - .print_fract = 10 + .print_fract = 10, + .lookup_fn = 0 }; struct rt_rule_4 { @@ -636,7 +638,11 @@ print_usage(void) "1/2/4/8 (default 4)>]\n" "[-g ]\n" "[-w ]\n" - "[-u ]\n", + "[-u ]\n" + "[-v ]\n", config.prgname); } @@ -679,7 +685,7 @@ parse_opts(int argc, char **argv) int opt; char *endptr; - while ((opt = getopt(argc, argv, "f:t:n:d:l:r:c6ab:e:g:w:u:s")) != + while ((opt = getopt(argc, argv, "f:t:n:d:l:r:c6ab:e:g:w:u:sv:")) != -1) { switch (opt) { case 'f': @@ -767,6 +773,22 @@ parse_opts(int argc, char **argv) rte_exit(-EINVAL, "Invalid option -g\n"); } break; + case 'v': + if ((strcmp(optarg, "s1") == 0) || + (strcmp(optarg, "s") == 0)) + break; + else if (strcmp(optarg, "v") == 0) { + config.lookup_fn = 1; + break; + } else if (strcmp(optarg, "s2") == 0) { + config.lookup_fn = 2; + break; + } else if (strcmp(optarg, "s3") == 0) { + config.lookup_fn = 3; + break; + } + print_usage(); + rte_exit(-EINVAL, "Invalid option -v %s\n", optarg); default: print_usage(); rte_exit(-EINVAL, "Invalid options\n"); @@ -844,6 +866,24 @@ run_v4(void) return -rte_errno; } + if (config.lookup_fn != 0) { + if (config.lookup_fn == 1) + ret = rte_fib_set_lookup_fn(fib, + RTE_FIB_DIR24_8_VECTOR); + else if (config.lookup_fn == 2) + ret = rte_fib_set_lookup_fn(fib, + RTE_FIB_DIR24_8_SCALAR_INLINE); + else if (config.lookup_fn == 3) + ret = rte_fib_set_lookup_fn(fib, + RTE_FIB_DIR24_8_SCALAR_UNI); + else + ret = -EINVAL; + if (ret != 0) { + printf("Can not init lookup function\n"); + return ret; + } + } + for (k = config.print_fract, i = 0; k > 0; k--) { start = rte_rdtsc_precise(); for (j = 0; j < (config.nb_routes - i) / k; j++) { @@ -1023,6 +1063,18 @@ run_v6(void) return -rte_errno; } + if (config.lookup_fn != 0) { + if (config.lookup_fn == 1) + ret = rte_fib6_set_lookup_fn(fib, + RTE_FIB6_TRIE_VECTOR); + else + ret = -EINVAL; + if (ret != 0) { + printf("Can not init lookup function\n"); + return ret; + } + } + for (k = config.print_fract, i = 0; k > 0; k--) { start = rte_rdtsc_precise(); for (j = 0; j < (config.nb_routes - i) / k; j++) {