From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mails.dpdk.org (mails.dpdk.org [217.70.189.124]) by inbox.dpdk.org (Postfix) with ESMTP id 130B7A0C55; Mon, 6 Sep 2021 18:04:11 +0200 (CEST) Received: from [217.70.189.124] (localhost [127.0.0.1]) by mails.dpdk.org (Postfix) with ESMTP id AA38D41100; Mon, 6 Sep 2021 18:04:07 +0200 (CEST) Received: from mga06.intel.com (mga06.intel.com [134.134.136.31]) by mails.dpdk.org (Postfix) with ESMTP id 3299C410ED for ; Mon, 6 Sep 2021 18:04:06 +0200 (CEST) X-IronPort-AV: E=McAfee;i="6200,9189,10099"; a="281011520" X-IronPort-AV: E=Sophos;i="5.85,272,1624345200"; d="scan'208";a="281011520" Received: from fmsmga007.fm.intel.com ([10.253.24.52]) by orsmga104.jf.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 06 Sep 2021 09:04:05 -0700 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.85,272,1624345200"; d="scan'208";a="464093084" Received: from silpixa00400072.ir.intel.com ([10.237.222.213]) by fmsmga007.fm.intel.com with ESMTP; 06 Sep 2021 09:04:03 -0700 From: Vladimir Medvedkin To: dev@dpdk.org Cc: konstantin.ananyev@intel.com, andrey.chilikin@intel.com, yipeng1.wang@intel.com, sameh.gobriel@intel.com, bruce.richardson@intel.com, john.mcnamara@intel.com Date: Mon, 6 Sep 2021 17:03:55 +0100 Message-Id: <1630944239-363648-2-git-send-email-vladimir.medvedkin@intel.com> X-Mailer: git-send-email 2.7.4 In-Reply-To: <1630944239-363648-1-git-send-email-vladimir.medvedkin@intel.com> References: <1630944239-363648-1-git-send-email-vladimir.medvedkin@intel.com> Subject: [dpdk-dev] [PATCH 1/5] hash: add new toeplitz hash implementation X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.29 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" This patch add a new Toeplitz hash implementation using Galios Fields New Instructions (GFNI). Signed-off-by: Vladimir Medvedkin --- doc/api/doxy-api-index.md | 1 + lib/hash/meson.build | 1 + lib/hash/rte_thash.c | 26 ++++++ lib/hash/rte_thash.h | 22 +++++ lib/hash/rte_thash_gfni.h | 229 ++++++++++++++++++++++++++++++++++++++++++++++ lib/hash/version.map | 2 + 6 files changed, 281 insertions(+) create mode 100644 lib/hash/rte_thash_gfni.h diff --git a/doc/api/doxy-api-index.md b/doc/api/doxy-api-index.md index 1992107..7549477 100644 --- a/doc/api/doxy-api-index.md +++ b/doc/api/doxy-api-index.md @@ -139,6 +139,7 @@ The public API headers are grouped by topics: [hash] (@ref rte_hash.h), [jhash] (@ref rte_jhash.h), [thash] (@ref rte_thash.h), + [thash_gfni] (@ref rte_thash_gfni.h), [FBK hash] (@ref rte_fbk_hash.h), [CRC hash] (@ref rte_hash_crc.h) diff --git a/lib/hash/meson.build b/lib/hash/meson.build index 9bc5ef9..40444ac 100644 --- a/lib/hash/meson.build +++ b/lib/hash/meson.build @@ -7,6 +7,7 @@ headers = files( 'rte_hash.h', 'rte_jhash.h', 'rte_thash.h', + 'rte_thash_gfni.h', ) indirect_headers += files('rte_crc_arm64.h') diff --git a/lib/hash/rte_thash.c b/lib/hash/rte_thash.c index d5a95a6..07447f7 100644 --- a/lib/hash/rte_thash.c +++ b/lib/hash/rte_thash.c @@ -11,6 +11,7 @@ #include #include #include +#include #define THASH_NAME_LEN 64 #define TOEPLITZ_HASH_LEN 32 @@ -88,6 +89,23 @@ struct rte_thash_ctx { uint8_t hash_key[0]; }; +uint8_t rte_thash_gfni_supported; + +void +rte_thash_complete_matrix(uint64_t *matrixes, uint8_t *rss_key, int size) +{ + int i, j; + uint8_t *m = (uint8_t *)matrixes; + + for (i = 0; i < size; i++) { + for (j = 0; j < 8; j++) { + m[i * 8 + j] = (rss_key[i] << j)| + (uint8_t)((uint16_t)(rss_key[i + 1]) >> + (8 - j)); + } + } +} + static inline uint32_t get_bit_lfsr(struct thash_lfsr *lfsr) { @@ -759,3 +777,11 @@ rte_thash_adjust_tuple(struct rte_thash_ctx *ctx, return ret; } + +RTE_INIT(rte_thash_gfni_init) +{ +#ifdef __GFNI__ + if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_GFNI)) + rte_thash_gfni_supported = 1; +#endif +} diff --git a/lib/hash/rte_thash.h b/lib/hash/rte_thash.h index 76109fc..e3f1fc6 100644 --- a/lib/hash/rte_thash.h +++ b/lib/hash/rte_thash.h @@ -28,6 +28,7 @@ extern "C" { #include #include #include +#include #if defined(RTE_ARCH_X86) || defined(__ARM_NEON) #include @@ -113,6 +114,8 @@ union rte_thash_tuple { }; #endif +extern uint8_t rte_thash_gfni_supported; + /** * Prepare special converted key to use with rte_softrss_be() * @param orig @@ -223,6 +226,25 @@ rte_softrss_be(uint32_t *input_tuple, uint32_t input_len, return ret; } +/** + * Converts Toeplitz hash key (RSS key) into matrixes required + * for GFNI implementation + * + * @warning + * @b EXPERIMENTAL: this API may change without prior notice. + * + * @param matrixes + * pointer to the memory where matrixes will be writen. + * Note: the size of this memory must be equal to size * 8 + * @param rss_key + * pointer to the Toeplitz hash key + * @param size + * Size of the rss_key in bytes. + */ +__rte_experimental +void +rte_thash_complete_matrix(uint64_t *matrixes, uint8_t *rss_key, int size); + /** @internal Logarithm of minimum size of the RSS ReTa */ #define RTE_THASH_RETA_SZ_MIN 2U /** @internal Logarithm of maximum size of the RSS ReTa */ diff --git a/lib/hash/rte_thash_gfni.h b/lib/hash/rte_thash_gfni.h new file mode 100644 index 0000000..8f89d7d --- /dev/null +++ b/lib/hash/rte_thash_gfni.h @@ -0,0 +1,229 @@ +/* SPDX-License-Identifier: BSD-3-Clause + * Copyright(c) 2021 Intel Corporation + */ + +#ifndef _RTE_THASH_GFNI_H_ +#define _RTE_THASH_GFNI_H_ + +/** + * @file + * + * Optimized Toeplitz hash functions implementation + * using Galois Fields New Instructions. + */ + +#include + +#ifdef __cplusplus +extern "C" { +#endif + +#ifdef __GFNI__ + +#define RTE_THASH_FIRST_ITER_MSK 0x0f0f0f0f0f0e0c08 +#define RTE_THASH_PERM_MSK 0x0f0f0f0f0f0f0f0f +#define RTE_THASH_FIRST_ITER_MSK_2 0xf0f0f0f0f0e0c080 +#define RTE_THASH_PERM_MSK_2 0xf0f0f0f0f0f0f0f0 +#define RTE_THASH_REWIND_MSK 0x0000000000113377 + +__rte_internal +static inline void +__rte_thash_xor_reduce(__m512i xor_acc, uint32_t *val_1, uint32_t *val_2) +{ + __m256i tmp_256_1, tmp_256_2; + __m128i tmp128_1, tmp128_2; + uint64_t tmp_1, tmp_2; + + tmp_256_1 = _mm512_castsi512_si256(xor_acc); + tmp_256_2 = _mm512_extracti32x8_epi32(xor_acc, 1); + tmp_256_1 = _mm256_xor_si256(tmp_256_1, tmp_256_2); + + tmp128_1 = _mm256_castsi256_si128(tmp_256_1); + tmp128_2 = _mm256_extracti32x4_epi32(tmp_256_1, 1); + tmp128_1 = _mm_xor_si128(tmp128_1, tmp128_2); + + tmp_1 = _mm_extract_epi64(tmp128_1, 0); + tmp_2 = _mm_extract_epi64(tmp128_1, 1); + tmp_1 ^= tmp_2; + + *val_1 = (uint32_t)tmp_1; + *val_2 = (uint32_t)(tmp_1 >> 32); +} + +__rte_internal +static inline __m512i +__rte_thash_gfni(uint64_t *mtrx, uint8_t *tuple, uint8_t *secondary_tuple, + int len) +{ + __m512i permute_idx = _mm512_set_epi8(7, 6, 5, 4, 7, 6, 5, 4, + 6, 5, 4, 3, 6, 5, 4, 3, + 5, 4, 3, 2, 5, 4, 3, 2, + 4, 3, 2, 1, 4, 3, 2, 1, + 3, 2, 1, 0, 3, 2, 1, 0, + 2, 1, 0, -1, 2, 1, 0, -1, + 1, 0, -1, -2, 1, 0, -1, -2, + 0, -1, -2, -3, 0, -1, -2, -3); + + const __m512i rewind_idx = _mm512_set_epi8(0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 0, 0, 0, 0, 0, + 0, 0, 0, 59, 0, 0, 0, 59, + 0, 0, 59, 58, 0, 0, 59, 58, + 0, 59, 58, 57, 0, 59, 58, 57); + const __mmask64 rewind_mask = RTE_THASH_REWIND_MSK; + const __m512i shift_8 = _mm512_set1_epi8(8); + __m512i xor_acc = _mm512_setzero_si512(); + __m512i perm_bytes = _mm512_setzero_si512(); + __m512i vals, matrixes, tuple_bytes, tuple_bytes_2; + __mmask64 load_mask, permute_mask, permute_mask_2; + int chunk_len = 0, i = 0; + uint8_t mtrx_msk; + const int prepend = 3; + + for (; len > 0; len -= 64, tuple += 64) { + if (i == 8) + perm_bytes = _mm512_maskz_permutexvar_epi8(rewind_mask, + rewind_idx, perm_bytes); + + permute_mask = RTE_THASH_FIRST_ITER_MSK; + load_mask = (len >= 64) ? UINT64_MAX : ((1ULL << len) - 1); + tuple_bytes = _mm512_maskz_loadu_epi8(load_mask, tuple); + if (secondary_tuple) { + permute_mask_2 = RTE_THASH_FIRST_ITER_MSK_2; + tuple_bytes_2 = _mm512_maskz_loadu_epi8(load_mask, + secondary_tuple); + } + + chunk_len = __builtin_popcountll(load_mask); + for (i = 0; i < ((chunk_len + prepend) / 8); i++, mtrx += 8) { + perm_bytes = _mm512_mask_permutexvar_epi8(perm_bytes, + permute_mask, permute_idx, tuple_bytes); + + if (secondary_tuple) + perm_bytes = + _mm512_mask_permutexvar_epi8(perm_bytes, + permute_mask_2, permute_idx, + tuple_bytes_2); + + matrixes = _mm512_maskz_loadu_epi64(UINT8_MAX, mtrx); + vals = _mm512_gf2p8affine_epi64_epi8(perm_bytes, + matrixes, 0); + + xor_acc = _mm512_xor_si512(xor_acc, vals); + permute_idx = _mm512_add_epi8(permute_idx, shift_8); + permute_mask = RTE_THASH_PERM_MSK; + if (secondary_tuple) + permute_mask_2 = RTE_THASH_PERM_MSK_2; + } + } + + int rest_len = (chunk_len + prepend) % 8; + if (rest_len != 0) { + mtrx_msk = (1 << (rest_len % 8)) - 1; + matrixes = _mm512_maskz_loadu_epi64(mtrx_msk, mtrx); + if (i == 8) { + perm_bytes = _mm512_maskz_permutexvar_epi8(rewind_mask, + rewind_idx, perm_bytes); + } else { + perm_bytes = _mm512_mask_permutexvar_epi8(perm_bytes, + permute_mask, permute_idx, tuple_bytes); + + if (secondary_tuple) + perm_bytes = + _mm512_mask_permutexvar_epi8( + perm_bytes, permute_mask_2, + permute_idx, tuple_bytes_2); + } + + vals = _mm512_gf2p8affine_epi64_epi8(perm_bytes, matrixes, 0); + xor_acc = _mm512_xor_si512(xor_acc, vals); + } + + return xor_acc; +} + +/** + * Calculate Toeplitz hash. + * + * @warning + * @b EXPERIMENTAL: this API may change without prior notice. + * + * @param m + * Pointer to the matrices generated from the corresponding + * RSS hash key using rte_thash_complete_matrix(). + * @param tuple + * Pointer to the data to be hashed. Data must be in network byte order. + * @param len + * Length of the data to be hashed. + * @return + * Calculated Toeplitz hash value. + */ +__rte_experimental +static inline uint32_t +rte_thash_gfni(uint64_t *m, uint8_t *tuple, int len) +{ + uint32_t val, val_zero; + + __m512i xor_acc = __rte_thash_gfni(m, tuple, NULL, len); + __rte_thash_xor_reduce(xor_acc, &val, &val_zero); + + return val; +} + +/** + * Calculate Toeplitz hash for two independent data buffers. + * + * @warning + * @b EXPERIMENTAL: this API may change without prior notice. + * + * @param m + * Pointer to the matrices generated from the corresponding + * RSS hash key using rte_thash_complete_matrix(). + * @param tuple_1 + * Pointer to the data to be hashed. Data must be in network byte order. + * @param tuple_2 + * Pointer to the data to be hashed. Data must be in network byte order. + * @param len + * Length of the largest data buffer to be hashed. + * @param val_1 + * Pointer to uint32_t where to put calculated Toeplitz hash value for + * the first tuple. + * @param val_2 + * Pointer to uint32_t where to put calculated Toeplitz hash value for + * the second tuple. + */ +__rte_experimental +static inline void +rte_thash_gfni_x2(uint64_t *mtrx, uint8_t *tuple_1, uint8_t *tuple_2, int len, + uint32_t *val_1, uint32_t *val_2) +{ + __m512i xor_acc = __rte_thash_gfni(mtrx, tuple_1, tuple_2, len); + __rte_thash_xor_reduce(xor_acc, val_1, val_2); +} + +#else /* __GFNI__ */ + +static inline uint32_t +rte_thash_gfni(uint64_t *mtrx __rte_unused, uint8_t *key __rte_unused, + int len __rte_unused) +{ + return 0; +} + +static inline void +rte_thash_gfni_x2(uint64_t *mtrx __rte_unused, uint8_t *tuple_1 __rte_unused, + uint8_t *tuple_2 __rte_unused, int len __rte_unused, + uint32_t *val_1 __rte_unused, uint32_t *val_2 __rte_unused) +{ + +} + +#endif + +#ifdef __cplusplus +} +#endif + +#endif /* _RTE_THASH_GFNI_H_ */ diff --git a/lib/hash/version.map b/lib/hash/version.map index ce4309a..cecf922 100644 --- a/lib/hash/version.map +++ b/lib/hash/version.map @@ -39,10 +39,12 @@ EXPERIMENTAL { rte_hash_rcu_qsbr_add; rte_thash_add_helper; rte_thash_adjust_tuple; + rte_thash_complete_matrix; rte_thash_find_existing; rte_thash_free_ctx; rte_thash_get_complement; rte_thash_get_helper; rte_thash_get_key; + rte_thash_gfni_supported; rte_thash_init_ctx; }; -- 2.7.4