From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mga01.intel.com (mga01.intel.com [192.55.52.88]) by dpdk.org (Postfix) with ESMTP id B2B7A5A7C for ; Thu, 29 Jan 2015 16:19:12 +0100 (CET) Received: from orsmga001.jf.intel.com ([10.7.209.18]) by fmsmga101.fm.intel.com with ESMTP; 29 Jan 2015 07:19:07 -0800 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.09,486,1418112000"; d="scan'208";a="644520896" Received: from irsmsx109.ger.corp.intel.com ([163.33.3.23]) by orsmga001.jf.intel.com with ESMTP; 29 Jan 2015 07:18:49 -0800 Received: from irsmsx105.ger.corp.intel.com ([169.254.7.81]) by IRSMSX109.ger.corp.intel.com ([169.254.13.11]) with mapi id 14.03.0195.001; Thu, 29 Jan 2015 15:17:44 +0000 From: "Ananyev, Konstantin" To: "Wang, Zhihong" , "dev@dpdk.org" Thread-Topic: [dpdk-dev] [PATCH v2 4/4] lib/librte_eal: Optimized memcpy in arch/x86/rte_memcpy.h for both SSE and AVX platforms Thread-Index: AQHQO2zhVuBnoe9lV0q9rFfXffl6zZzXMYTQ Date: Thu, 29 Jan 2015 15:17:44 +0000 Message-ID: <2601191342CEEE43887BDE71AB977258213E085B@irsmsx105.ger.corp.intel.com> References: <1422499127-11689-1-git-send-email-zhihong.wang@intel.com> <1422499127-11689-5-git-send-email-zhihong.wang@intel.com> In-Reply-To: <1422499127-11689-5-git-send-email-zhihong.wang@intel.com> Accept-Language: en-IE, en-US Content-Language: en-US X-MS-Has-Attach: X-MS-TNEF-Correlator: x-originating-ip: [163.33.239.180] Content-Type: text/plain; charset="us-ascii" Content-Transfer-Encoding: quoted-printable MIME-Version: 1.0 Subject: Re: [dpdk-dev] [PATCH v2 4/4] lib/librte_eal: Optimized memcpy in arch/x86/rte_memcpy.h for both SSE and AVX platforms X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: patches and discussions about DPDK List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Thu, 29 Jan 2015 15:19:14 -0000 Hi Zhihong, > -----Original Message----- > From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Zhihong Wang > Sent: Thursday, January 29, 2015 2:39 AM > To: dev@dpdk.org > Subject: [dpdk-dev] [PATCH v2 4/4] lib/librte_eal: Optimized memcpy in ar= ch/x86/rte_memcpy.h for both SSE and AVX platforms >=20 > Main code changes: >=20 > 1. Differentiate architectural features based on CPU flags >=20 > a. Implement separated move functions for SSE/AVX/AVX2 to make full u= tilization of cache bandwidth >=20 > b. Implement separated copy flow specifically optimized for target ar= chitecture >=20 > 2. Rewrite the memcpy function "rte_memcpy" >=20 > a. Add store aligning >=20 > b. Add load aligning based on architectural features >=20 > c. Put block copy loop into inline move functions for better control = of instruction order >=20 > d. Eliminate unnecessary MOVs >=20 > 3. Rewrite the inline move functions >=20 > a. Add move functions for unaligned load cases >=20 > b. Change instruction order in copy loops for better pipeline utiliza= tion >=20 > c. Use intrinsics instead of assembly code >=20 > 4. Remove slow glibc call for constant copies >=20 > Signed-off-by: Zhihong Wang > --- > .../common/include/arch/x86/rte_memcpy.h | 680 +++++++++++++++= ------ > 1 file changed, 509 insertions(+), 171 deletions(-) >=20 > diff --git a/lib/librte_eal/common/include/arch/x86/rte_memcpy.h b/lib/li= brte_eal/common/include/arch/x86/rte_memcpy.h > index fb9eba8..7b2d382 100644 > --- a/lib/librte_eal/common/include/arch/x86/rte_memcpy.h > +++ b/lib/librte_eal/common/include/arch/x86/rte_memcpy.h > @@ -34,166 +34,189 @@ > #ifndef _RTE_MEMCPY_X86_64_H_ > #define _RTE_MEMCPY_X86_64_H_ >=20 > +/** > + * @file > + * > + * Functions for SSE/AVX/AVX2 implementation of memcpy(). > + */ > + > +#include > #include > #include > -#include > +#include >=20 > #ifdef __cplusplus > extern "C" { > #endif >=20 > -#include "generic/rte_memcpy.h" > +/** > + * Copy bytes from one location to another. The locations must not overl= ap. > + * > + * @note This is implemented as a macro, so it's address should not be t= aken > + * and care is needed as parameter expressions may be evaluated multiple= times. > + * > + * @param dst > + * Pointer to the destination of the data. > + * @param src > + * Pointer to the source data. > + * @param n > + * Number of bytes to copy. > + * @return > + * Pointer to the destination data. > + */ > +static inline void * > +rte_memcpy(void *dst, const void *src, size_t n) __attribute__((always_i= nline)); >=20 > -#ifdef __INTEL_COMPILER > -#pragma warning(disable:593) /* Stop unused variable warning (reg_a etc)= . */ > -#endif > +#ifdef RTE_MACHINE_CPUFLAG_AVX2 >=20 > +/** > + * AVX2 implementation below > + */ > + > +/** > + * Copy 16 bytes from one location to another, > + * locations should not overlap. > + */ > static inline void > rte_mov16(uint8_t *dst, const uint8_t *src) > { > - __m128i reg_a; > - asm volatile ( > - "movdqu (%[src]), %[reg_a]\n\t" > - "movdqu %[reg_a], (%[dst])\n\t" > - : [reg_a] "=3Dx" (reg_a) > - : [src] "r" (src), > - [dst] "r"(dst) > - : "memory" > - ); > + __m128i xmm0; > + > + xmm0 =3D _mm_loadu_si128((const __m128i *)src); > + _mm_storeu_si128((__m128i *)dst, xmm0); > } >=20 > +/** > + * Copy 32 bytes from one location to another, > + * locations should not overlap. > + */ > static inline void > rte_mov32(uint8_t *dst, const uint8_t *src) > { > - __m128i reg_a, reg_b; > - asm volatile ( > - "movdqu (%[src]), %[reg_a]\n\t" > - "movdqu 16(%[src]), %[reg_b]\n\t" > - "movdqu %[reg_a], (%[dst])\n\t" > - "movdqu %[reg_b], 16(%[dst])\n\t" > - : [reg_a] "=3Dx" (reg_a), > - [reg_b] "=3Dx" (reg_b) > - : [src] "r" (src), > - [dst] "r"(dst) > - : "memory" > - ); > -} > + __m256i ymm0; >=20 > -static inline void > -rte_mov48(uint8_t *dst, const uint8_t *src) > -{ > - __m128i reg_a, reg_b, reg_c; > - asm volatile ( > - "movdqu (%[src]), %[reg_a]\n\t" > - "movdqu 16(%[src]), %[reg_b]\n\t" > - "movdqu 32(%[src]), %[reg_c]\n\t" > - "movdqu %[reg_a], (%[dst])\n\t" > - "movdqu %[reg_b], 16(%[dst])\n\t" > - "movdqu %[reg_c], 32(%[dst])\n\t" > - : [reg_a] "=3Dx" (reg_a), > - [reg_b] "=3Dx" (reg_b), > - [reg_c] "=3Dx" (reg_c) > - : [src] "r" (src), > - [dst] "r"(dst) > - : "memory" > - ); > + ymm0 =3D _mm256_loadu_si256((const __m256i *)src); > + _mm256_storeu_si256((__m256i *)dst, ymm0); > } >=20 > +/** > + * Copy 64 bytes from one location to another, > + * locations should not overlap. > + */ > static inline void > rte_mov64(uint8_t *dst, const uint8_t *src) > { > - __m128i reg_a, reg_b, reg_c, reg_d; > - asm volatile ( > - "movdqu (%[src]), %[reg_a]\n\t" > - "movdqu 16(%[src]), %[reg_b]\n\t" > - "movdqu 32(%[src]), %[reg_c]\n\t" > - "movdqu 48(%[src]), %[reg_d]\n\t" > - "movdqu %[reg_a], (%[dst])\n\t" > - "movdqu %[reg_b], 16(%[dst])\n\t" > - "movdqu %[reg_c], 32(%[dst])\n\t" > - "movdqu %[reg_d], 48(%[dst])\n\t" > - : [reg_a] "=3Dx" (reg_a), > - [reg_b] "=3Dx" (reg_b), > - [reg_c] "=3Dx" (reg_c), > - [reg_d] "=3Dx" (reg_d) > - : [src] "r" (src), > - [dst] "r"(dst) > - : "memory" > - ); > + rte_mov32((uint8_t *)dst + 0 * 32, (const uint8_t *)src + 0 * 32); > + rte_mov32((uint8_t *)dst + 1 * 32, (const uint8_t *)src + 1 * 32); > } >=20 > +/** > + * Copy 128 bytes from one location to another, > + * locations should not overlap. > + */ > static inline void > rte_mov128(uint8_t *dst, const uint8_t *src) > { > - __m128i reg_a, reg_b, reg_c, reg_d, reg_e, reg_f, reg_g, reg_h; > - asm volatile ( > - "movdqu (%[src]), %[reg_a]\n\t" > - "movdqu 16(%[src]), %[reg_b]\n\t" > - "movdqu 32(%[src]), %[reg_c]\n\t" > - "movdqu 48(%[src]), %[reg_d]\n\t" > - "movdqu 64(%[src]), %[reg_e]\n\t" > - "movdqu 80(%[src]), %[reg_f]\n\t" > - "movdqu 96(%[src]), %[reg_g]\n\t" > - "movdqu 112(%[src]), %[reg_h]\n\t" > - "movdqu %[reg_a], (%[dst])\n\t" > - "movdqu %[reg_b], 16(%[dst])\n\t" > - "movdqu %[reg_c], 32(%[dst])\n\t" > - "movdqu %[reg_d], 48(%[dst])\n\t" > - "movdqu %[reg_e], 64(%[dst])\n\t" > - "movdqu %[reg_f], 80(%[dst])\n\t" > - "movdqu %[reg_g], 96(%[dst])\n\t" > - "movdqu %[reg_h], 112(%[dst])\n\t" > - : [reg_a] "=3Dx" (reg_a), > - [reg_b] "=3Dx" (reg_b), > - [reg_c] "=3Dx" (reg_c), > - [reg_d] "=3Dx" (reg_d), > - [reg_e] "=3Dx" (reg_e), > - [reg_f] "=3Dx" (reg_f), > - [reg_g] "=3Dx" (reg_g), > - [reg_h] "=3Dx" (reg_h) > - : [src] "r" (src), > - [dst] "r"(dst) > - : "memory" > - ); > + rte_mov32((uint8_t *)dst + 0 * 32, (const uint8_t *)src + 0 * 32); > + rte_mov32((uint8_t *)dst + 1 * 32, (const uint8_t *)src + 1 * 32); > + rte_mov32((uint8_t *)dst + 2 * 32, (const uint8_t *)src + 2 * 32); > + rte_mov32((uint8_t *)dst + 3 * 32, (const uint8_t *)src + 3 * 32); > } >=20 > -#ifdef __INTEL_COMPILER > -#pragma warning(enable:593) > -#endif > - > +/** > + * Copy 256 bytes from one location to another, > + * locations should not overlap. > + */ > static inline void > rte_mov256(uint8_t *dst, const uint8_t *src) > { > - rte_mov128(dst, src); > - rte_mov128(dst + 128, src + 128); > + rte_mov32((uint8_t *)dst + 0 * 32, (const uint8_t *)src + 0 * 32); > + rte_mov32((uint8_t *)dst + 1 * 32, (const uint8_t *)src + 1 * 32); > + rte_mov32((uint8_t *)dst + 2 * 32, (const uint8_t *)src + 2 * 32); > + rte_mov32((uint8_t *)dst + 3 * 32, (const uint8_t *)src + 3 * 32); > + rte_mov32((uint8_t *)dst + 4 * 32, (const uint8_t *)src + 4 * 32); > + rte_mov32((uint8_t *)dst + 5 * 32, (const uint8_t *)src + 5 * 32); > + rte_mov32((uint8_t *)dst + 6 * 32, (const uint8_t *)src + 6 * 32); > + rte_mov32((uint8_t *)dst + 7 * 32, (const uint8_t *)src + 7 * 32); > } >=20 > -#define rte_memcpy(dst, src, n) \ > - ({ (__builtin_constant_p(n)) ? \ > - memcpy((dst), (src), (n)) : \ > - rte_memcpy_func((dst), (src), (n)); }) > +/** > + * Copy 64-byte blocks from one location to another, > + * locations should not overlap. > + */ > +static inline void > +rte_mov64blocks(uint8_t *dst, const uint8_t *src, size_t n) > +{ > + __m256i ymm0, ymm1; > + > + while (n >=3D 64) { > + ymm0 =3D _mm256_loadu_si256((const __m256i *)((const uint8_t *)src + 0= * 32)); > + n -=3D 64; > + ymm1 =3D _mm256_loadu_si256((const __m256i *)((const uint8_t *)src + 1= * 32)); > + src =3D (const uint8_t *)src + 64; > + _mm256_storeu_si256((__m256i *)((uint8_t *)dst + 0 * 32), ymm0); > + _mm256_storeu_si256((__m256i *)((uint8_t *)dst + 1 * 32), ymm1); > + dst =3D (uint8_t *)dst + 64; > + } > +} > + > +/** > + * Copy 256-byte blocks from one location to another, > + * locations should not overlap. > + */ > +static inline void > +rte_mov256blocks(uint8_t *dst, const uint8_t *src, size_t n) > +{ > + __m256i ymm0, ymm1, ymm2, ymm3, ymm4, ymm5, ymm6, ymm7; > + > + while (n >=3D 256) { > + ymm0 =3D _mm256_loadu_si256((const __m256i *)((const uint8_t *)src + 0= * 32)); > + n -=3D 256; > + ymm1 =3D _mm256_loadu_si256((const __m256i *)((const uint8_t *)src + 1= * 32)); > + ymm2 =3D _mm256_loadu_si256((const __m256i *)((const uint8_t *)src + 2= * 32)); > + ymm3 =3D _mm256_loadu_si256((const __m256i *)((const uint8_t *)src + 3= * 32)); > + ymm4 =3D _mm256_loadu_si256((const __m256i *)((const uint8_t *)src + 4= * 32)); > + ymm5 =3D _mm256_loadu_si256((const __m256i *)((const uint8_t *)src + 5= * 32)); > + ymm6 =3D _mm256_loadu_si256((const __m256i *)((const uint8_t *)src + 6= * 32)); > + ymm7 =3D _mm256_loadu_si256((const __m256i *)((const uint8_t *)src + 7= * 32)); > + src =3D (const uint8_t *)src + 256; > + _mm256_storeu_si256((__m256i *)((uint8_t *)dst + 0 * 32), ymm0); > + _mm256_storeu_si256((__m256i *)((uint8_t *)dst + 1 * 32), ymm1); > + _mm256_storeu_si256((__m256i *)((uint8_t *)dst + 2 * 32), ymm2); > + _mm256_storeu_si256((__m256i *)((uint8_t *)dst + 3 * 32), ymm3); > + _mm256_storeu_si256((__m256i *)((uint8_t *)dst + 4 * 32), ymm4); > + _mm256_storeu_si256((__m256i *)((uint8_t *)dst + 5 * 32), ymm5); > + _mm256_storeu_si256((__m256i *)((uint8_t *)dst + 6 * 32), ymm6); > + _mm256_storeu_si256((__m256i *)((uint8_t *)dst + 7 * 32), ymm7); > + dst =3D (uint8_t *)dst + 256; > + } > +} >=20 > static inline void * > -rte_memcpy_func(void *dst, const void *src, size_t n) > +rte_memcpy(void *dst, const void *src, size_t n) > { > void *ret =3D dst; > + int dstofss; > + int bits; >=20 > - /* We can't copy < 16 bytes using XMM registers so do it manually. */ > + /** > + * Copy less than 16 bytes > + */ > if (n < 16) { > if (n & 0x01) { > *(uint8_t *)dst =3D *(const uint8_t *)src; > - dst =3D (uint8_t *)dst + 1; > src =3D (const uint8_t *)src + 1; > + dst =3D (uint8_t *)dst + 1; > } > if (n & 0x02) { > *(uint16_t *)dst =3D *(const uint16_t *)src; > - dst =3D (uint16_t *)dst + 1; > src =3D (const uint16_t *)src + 1; > + dst =3D (uint16_t *)dst + 1; > } > if (n & 0x04) { > *(uint32_t *)dst =3D *(const uint32_t *)src; > - dst =3D (uint32_t *)dst + 1; > src =3D (const uint32_t *)src + 1; > + dst =3D (uint32_t *)dst + 1; > } > if (n & 0x08) { > *(uint64_t *)dst =3D *(const uint64_t *)src; > @@ -201,95 +224,410 @@ rte_memcpy_func(void *dst, const void *src, size_t= n) > return ret; > } >=20 > - /* Special fast cases for <=3D 128 bytes */ > + /** > + * Fast way when copy size doesn't exceed 512 bytes > + */ > if (n <=3D 32) { > rte_mov16((uint8_t *)dst, (const uint8_t *)src); > rte_mov16((uint8_t *)dst - 16 + n, (const uint8_t *)src - 16 + n); > return ret; > } > - > if (n <=3D 64) { > rte_mov32((uint8_t *)dst, (const uint8_t *)src); > rte_mov32((uint8_t *)dst - 32 + n, (const uint8_t *)src - 32 + n); > return ret; > } > - > - if (n <=3D 128) { > - rte_mov64((uint8_t *)dst, (const uint8_t *)src); > - rte_mov64((uint8_t *)dst - 64 + n, (const uint8_t *)src - 64 + n); > + if (n <=3D 512) { > + if (n >=3D 256) { > + n -=3D 256; > + rte_mov256((uint8_t *)dst, (const uint8_t *)src); > + src =3D (const uint8_t *)src + 256; > + dst =3D (uint8_t *)dst + 256; > + } > + if (n >=3D 128) { > + n -=3D 128; > + rte_mov128((uint8_t *)dst, (const uint8_t *)src); > + src =3D (const uint8_t *)src + 128; > + dst =3D (uint8_t *)dst + 128; > + } > + if (n >=3D 64) { > + n -=3D 64; > + rte_mov64((uint8_t *)dst, (const uint8_t *)src); > + src =3D (const uint8_t *)src + 64; > + dst =3D (uint8_t *)dst + 64; > + } > +COPY_BLOCK_64_BACK31: > + if (n > 32) { > + rte_mov32((uint8_t *)dst, (const uint8_t *)src); > + rte_mov32((uint8_t *)dst - 32 + n, (const uint8_t *)src - 32 + n); > + return ret; > + } > + if (n > 0) { > + rte_mov32((uint8_t *)dst - 32 + n, (const uint8_t *)src - 32 + n); > + } > return ret; > } >=20 > - /* > - * For large copies > 128 bytes. This combination of 256, 64 and 16 byt= e > - * copies was found to be faster than doing 128 and 32 byte copies as > - * well. > + /** > + * Make store aligned when copy size exceeds 512 bytes > */ > - for ( ; n >=3D 256; n -=3D 256) { > - rte_mov256((uint8_t *)dst, (const uint8_t *)src); > - dst =3D (uint8_t *)dst + 256; > - src =3D (const uint8_t *)src + 256; > + dstofss =3D 32 - (int)((long long)(void *)dst & 0x1F); > + n -=3D dstofss; > + rte_mov32((uint8_t *)dst, (const uint8_t *)src); > + src =3D (const uint8_t *)src + dstofss; > + dst =3D (uint8_t *)dst + dstofss; > + > + /** > + * Copy 256-byte blocks. > + * Use copy block function for better instruction order control, > + * which is important when load is unaligned. > + */ > + rte_mov256blocks((uint8_t *)dst, (const uint8_t *)src, n); > + bits =3D n; > + n =3D n & 255; > + bits -=3D n; > + src =3D (const uint8_t *)src + bits; > + dst =3D (uint8_t *)dst + bits; > + > + /** > + * Copy 64-byte blocks. > + * Use copy block function for better instruction order control, > + * which is important when load is unaligned. > + */ > + if (n >=3D 64) { > + rte_mov64blocks((uint8_t *)dst, (const uint8_t *)src, n); > + bits =3D n; > + n =3D n & 63; > + bits -=3D n; > + src =3D (const uint8_t *)src + bits; > + dst =3D (uint8_t *)dst + bits; > } >=20 > - /* > - * We split the remaining bytes (which will be less than 256) into > - * 64byte (2^6) chunks. > - * Using incrementing integers in the case labels of a switch statement > - * enourages the compiler to use a jump table. To get incrementing > - * integers, we shift the 2 relevant bits to the LSB position to first > - * get decrementing integers, and then subtract. > + /** > + * Copy whatever left > */ > - switch (3 - (n >> 6)) { > - case 0x00: > - rte_mov64((uint8_t *)dst, (const uint8_t *)src); > - n -=3D 64; > - dst =3D (uint8_t *)dst + 64; > - src =3D (const uint8_t *)src + 64; /* fallthrough */ > - case 0x01: > - rte_mov64((uint8_t *)dst, (const uint8_t *)src); > - n -=3D 64; > - dst =3D (uint8_t *)dst + 64; > - src =3D (const uint8_t *)src + 64; /* fallthrough */ > - case 0x02: > - rte_mov64((uint8_t *)dst, (const uint8_t *)src); > - n -=3D 64; > - dst =3D (uint8_t *)dst + 64; > - src =3D (const uint8_t *)src + 64; /* fallthrough */ > - default: > - ; > + goto COPY_BLOCK_64_BACK31; > +} > + > +#else /* RTE_MACHINE_CPUFLAG_AVX2 */ > + > +/** > + * SSE & AVX implementation below > + */ > + > +/** > + * Copy 16 bytes from one location to another, > + * locations should not overlap. > + */ > +static inline void > +rte_mov16(uint8_t *dst, const uint8_t *src) > +{ > + __m128i xmm0; > + > + xmm0 =3D _mm_loadu_si128((const __m128i *)(const __m128i *)src); > + _mm_storeu_si128((__m128i *)dst, xmm0); > +} > + > +/** > + * Copy 32 bytes from one location to another, > + * locations should not overlap. > + */ > +static inline void > +rte_mov32(uint8_t *dst, const uint8_t *src) > +{ > + rte_mov16((uint8_t *)dst + 0 * 16, (const uint8_t *)src + 0 * 16); > + rte_mov16((uint8_t *)dst + 1 * 16, (const uint8_t *)src + 1 * 16); > +} > + > +/** > + * Copy 64 bytes from one location to another, > + * locations should not overlap. > + */ > +static inline void > +rte_mov64(uint8_t *dst, const uint8_t *src) > +{ > + rte_mov16((uint8_t *)dst + 0 * 16, (const uint8_t *)src + 0 * 16); > + rte_mov16((uint8_t *)dst + 1 * 16, (const uint8_t *)src + 1 * 16); > + rte_mov16((uint8_t *)dst + 2 * 16, (const uint8_t *)src + 2 * 16); > + rte_mov16((uint8_t *)dst + 3 * 16, (const uint8_t *)src + 3 * 16); > +} > + > +/** > + * Copy 128 bytes from one location to another, > + * locations should not overlap. > + */ > +static inline void > +rte_mov128(uint8_t *dst, const uint8_t *src) > +{ > + rte_mov16((uint8_t *)dst + 0 * 16, (const uint8_t *)src + 0 * 16); > + rte_mov16((uint8_t *)dst + 1 * 16, (const uint8_t *)src + 1 * 16); > + rte_mov16((uint8_t *)dst + 2 * 16, (const uint8_t *)src + 2 * 16); > + rte_mov16((uint8_t *)dst + 3 * 16, (const uint8_t *)src + 3 * 16); > + rte_mov16((uint8_t *)dst + 4 * 16, (const uint8_t *)src + 4 * 16); > + rte_mov16((uint8_t *)dst + 5 * 16, (const uint8_t *)src + 5 * 16); > + rte_mov16((uint8_t *)dst + 6 * 16, (const uint8_t *)src + 6 * 16); > + rte_mov16((uint8_t *)dst + 7 * 16, (const uint8_t *)src + 7 * 16); > +} > + > +/** > + * Copy 256 bytes from one location to another, > + * locations should not overlap. > + */ > +static inline void > +rte_mov256(uint8_t *dst, const uint8_t *src) > +{ > + rte_mov16((uint8_t *)dst + 0 * 16, (const uint8_t *)src + 0 * 16); > + rte_mov16((uint8_t *)dst + 1 * 16, (const uint8_t *)src + 1 * 16); > + rte_mov16((uint8_t *)dst + 2 * 16, (const uint8_t *)src + 2 * 16); > + rte_mov16((uint8_t *)dst + 3 * 16, (const uint8_t *)src + 3 * 16); > + rte_mov16((uint8_t *)dst + 4 * 16, (const uint8_t *)src + 4 * 16); > + rte_mov16((uint8_t *)dst + 5 * 16, (const uint8_t *)src + 5 * 16); > + rte_mov16((uint8_t *)dst + 6 * 16, (const uint8_t *)src + 6 * 16); > + rte_mov16((uint8_t *)dst + 7 * 16, (const uint8_t *)src + 7 * 16); > + rte_mov16((uint8_t *)dst + 8 * 16, (const uint8_t *)src + 8 * 16); > + rte_mov16((uint8_t *)dst + 9 * 16, (const uint8_t *)src + 9 * 16); > + rte_mov16((uint8_t *)dst + 10 * 16, (const uint8_t *)src + 10 * 16); > + rte_mov16((uint8_t *)dst + 11 * 16, (const uint8_t *)src + 11 * 16); > + rte_mov16((uint8_t *)dst + 12 * 16, (const uint8_t *)src + 12 * 16); > + rte_mov16((uint8_t *)dst + 13 * 16, (const uint8_t *)src + 13 * 16); > + rte_mov16((uint8_t *)dst + 14 * 16, (const uint8_t *)src + 14 * 16); > + rte_mov16((uint8_t *)dst + 15 * 16, (const uint8_t *)src + 15 * 16); > +} > + > +/** > + * Macro for copying unaligned block from one location to another with c= onstant load offset, > + * 47 bytes leftover maximum, > + * locations should not overlap. > + * Requirements: > + * - Store is aligned > + * - Load offset is , which must be immediate value within [1, 1= 5] > + * - For , make sure bit backwards & <16 - offset> bit for= wards are available for loading > + * - , , must be variables > + * - __m128i ~ must be pre-defined > + */ > +#define MOVEUNALIGNED_LEFT47_IMM(dst, src, len, offset) = \ > +({ = \ > + int tmp; = \ > + while (len >=3D 128 + 16 - offset) { = \ > + xmm0 =3D _mm_loadu_si128((const __m128i *)((const uint8_t *)src = - offset + 0 * 16)); \ > + len -=3D 128; = \ > + xmm1 =3D _mm_loadu_si128((const __m128i *)((const uint8_t *)src = - offset + 1 * 16)); \ > + xmm2 =3D _mm_loadu_si128((const __m128i *)((const uint8_t *)src = - offset + 2 * 16)); \ > + xmm3 =3D _mm_loadu_si128((const __m128i *)((const uint8_t *)src = - offset + 3 * 16)); \ > + xmm4 =3D _mm_loadu_si128((const __m128i *)((const uint8_t *)src = - offset + 4 * 16)); \ > + xmm5 =3D _mm_loadu_si128((const __m128i *)((const uint8_t *)src = - offset + 5 * 16)); \ > + xmm6 =3D _mm_loadu_si128((const __m128i *)((const uint8_t *)src = - offset + 6 * 16)); \ > + xmm7 =3D _mm_loadu_si128((const __m128i *)((const uint8_t *)src = - offset + 7 * 16)); \ > + xmm8 =3D _mm_loadu_si128((const __m128i *)((const uint8_t *)src = - offset + 8 * 16)); \ > + src =3D (const uint8_t *)src + 128; = \ > + _mm_storeu_si128((__m128i *)((uint8_t *)dst + 0 * 16), _mm_align= r_epi8(xmm1, xmm0, offset)); \ > + _mm_storeu_si128((__m128i *)((uint8_t *)dst + 1 * 16), _mm_align= r_epi8(xmm2, xmm1, offset)); \ > + _mm_storeu_si128((__m128i *)((uint8_t *)dst + 2 * 16), _mm_align= r_epi8(xmm3, xmm2, offset)); \ > + _mm_storeu_si128((__m128i *)((uint8_t *)dst + 3 * 16), _mm_align= r_epi8(xmm4, xmm3, offset)); \ > + _mm_storeu_si128((__m128i *)((uint8_t *)dst + 4 * 16), _mm_align= r_epi8(xmm5, xmm4, offset)); \ > + _mm_storeu_si128((__m128i *)((uint8_t *)dst + 5 * 16), _mm_align= r_epi8(xmm6, xmm5, offset)); \ > + _mm_storeu_si128((__m128i *)((uint8_t *)dst + 6 * 16), _mm_align= r_epi8(xmm7, xmm6, offset)); \ > + _mm_storeu_si128((__m128i *)((uint8_t *)dst + 7 * 16), _mm_align= r_epi8(xmm8, xmm7, offset)); \ > + dst =3D (uint8_t *)dst + 128; = \ > + } = \ > + tmp =3D len; = \ > + len =3D ((len - 16 + offset) & 127) + 16 - offset; = \ > + tmp -=3D len; = \ > + src =3D (const uint8_t *)src + tmp; = \ > + dst =3D (uint8_t *)dst + tmp; = \ > + if (len >=3D 32 + 16 - offset) { = \ > + while (len >=3D 32 + 16 - offset) { = \ > + xmm0 =3D _mm_loadu_si128((const __m128i *)((const uint8_t *)= src - offset + 0 * 16)); \ > + len -=3D 32; = \ > + xmm1 =3D _mm_loadu_si128((const __m128i *)((const uint8_t *)= src - offset + 1 * 16)); \ > + xmm2 =3D _mm_loadu_si128((const __m128i *)((const uint8_t *)= src - offset + 2 * 16)); \ > + src =3D (const uint8_t *)src + 32; = \ > + _mm_storeu_si128((__m128i *)((uint8_t *)dst + 0 * 16), _mm_a= lignr_epi8(xmm1, xmm0, offset)); \ > + _mm_storeu_si128((__m128i *)((uint8_t *)dst + 1 * 16), _mm_a= lignr_epi8(xmm2, xmm1, offset)); \ > + dst =3D (uint8_t *)dst + 32; = \ > + } = \ > + tmp =3D len; = \ > + len =3D ((len - 16 + offset) & 31) + 16 - offset; = \ > + tmp -=3D len; = \ > + src =3D (const uint8_t *)src + tmp; = \ > + dst =3D (uint8_t *)dst + tmp; = \ > + } = \ > +}) > + > +/** > + * Macro for copying unaligned block from one location to another, > + * 47 bytes leftover maximum, > + * locations should not overlap. > + * Use switch here because the aligning instruction requires immediate v= alue for shift count. > + * Requirements: > + * - Store is aligned > + * - Load offset is , which must be within [1, 15] > + * - For , make sure bit backwards & <16 - offset> bit for= wards are available for loading > + * - , , must be variables > + * - __m128i ~ used in MOVEUNALIGNED_LEFT47_IMM must be pr= e-defined > + */ > +#define MOVEUNALIGNED_LEFT47(dst, src, len, offset) \ > +({ \ > + switch (offset) { \ > + case 0x01: MOVEUNALIGNED_LEFT47_IMM(dst, src, n, 0x01); break; \ > + case 0x02: MOVEUNALIGNED_LEFT47_IMM(dst, src, n, 0x02); break; \ > + case 0x03: MOVEUNALIGNED_LEFT47_IMM(dst, src, n, 0x03); break; \ > + case 0x04: MOVEUNALIGNED_LEFT47_IMM(dst, src, n, 0x04); break; \ > + case 0x05: MOVEUNALIGNED_LEFT47_IMM(dst, src, n, 0x05); break; \ > + case 0x06: MOVEUNALIGNED_LEFT47_IMM(dst, src, n, 0x06); break; \ > + case 0x07: MOVEUNALIGNED_LEFT47_IMM(dst, src, n, 0x07); break; \ > + case 0x08: MOVEUNALIGNED_LEFT47_IMM(dst, src, n, 0x08); break; \ > + case 0x09: MOVEUNALIGNED_LEFT47_IMM(dst, src, n, 0x09); break; \ > + case 0x0A: MOVEUNALIGNED_LEFT47_IMM(dst, src, n, 0x0A); break; \ > + case 0x0B: MOVEUNALIGNED_LEFT47_IMM(dst, src, n, 0x0B); break; \ > + case 0x0C: MOVEUNALIGNED_LEFT47_IMM(dst, src, n, 0x0C); break; \ > + case 0x0D: MOVEUNALIGNED_LEFT47_IMM(dst, src, n, 0x0D); break; \ > + case 0x0E: MOVEUNALIGNED_LEFT47_IMM(dst, src, n, 0x0E); break; \ > + case 0x0F: MOVEUNALIGNED_LEFT47_IMM(dst, src, n, 0x0F); break; \ > + default:; \ > + } \ > +}) We probably didn't understand each other properly. My thought was to do something like: #define MOVEUNALIGNED_128(offset) do { \ _mm_storeu_si128((__m128i *)((uint8_t *)dst + 0 * 16), _mm_alignr_epi8(xmm= 1, xmm0, offset)); \ _mm_storeu_si128((__m128i *)((uint8_t *)dst + 1 * 16), _mm_alignr_epi8(xmm= 2, xmm1, offset)); \ _mm_storeu_si128((__m128i *)((uint8_t *)dst + 2 * 16), _mm_alignr_epi8(xmm= 3, xmm2, offset)); \ _mm_storeu_si128((__m128i *)((uint8_t *)dst + 3 * 16), _mm_alignr_epi8(xmm= 4, xmm3, offset)); \ _mm_storeu_si128((__m128i *)((uint8_t *)dst + 4 * 16), _mm_alignr_epi8(xmm= 5, xmm4, offset)); \ _mm_storeu_si128((__m128i *)((uint8_t *)dst + 5 * 16), _mm_alignr_epi8(xmm= 6, xmm5, offset)); \ _mm_storeu_si128((__m128i *)((uint8_t *)dst + 6 * 16), _mm_alignr_epi8(xmm= 7, xmm6, offset)); \ _mm_storeu_si128((__m128i *)((uint8_t *)dst + 7 * 16), _mm_alignr_epi8(xmm= 8, xmm7, offset)); \ =20 } while(0); Then at MOVEUNALIGNED_LEFT47_IMM: .... xmm7 =3D _mm_loadu_si128((const __m128i *)((const uint8_t *)src - offset + = 7 * 16)); \ xmm8 =3D _mm_loadu_si128((const __m128i *)((const uint8_t *)src - offset + = 8 * 16)); \ src =3D (const uint8_t *)src + 128; switch(offset) { case 0x01: MOVEUNALIGNED_128(1); break; =20 ... case 0x0f: MOVEUNALIGNED_128(f); break; } ... dst =3D (uint8_t *)dst + 128 An then in rte_memcpy() you don't need a switch for MOVEUNALIGNED_LEFT47_IM= M itself. Thought it might help to generate smaller code for rte_ememcpy() while keep= ing performance reasonably high. Konstantin > + > +static inline void * > +rte_memcpy(void *dst, const void *src, size_t n) > +{ > + __m128i xmm0, xmm1, xmm2, xmm3, xmm4, xmm5, xmm6, xmm7, xmm8; > + void *ret =3D dst; > + int dstofss; > + int srcofs; > + > + /** > + * Copy less than 16 bytes > + */ > + if (n < 16) { > + if (n & 0x01) { > + *(uint8_t *)dst =3D *(const uint8_t *)src; > + src =3D (const uint8_t *)src + 1; > + dst =3D (uint8_t *)dst + 1; > + } > + if (n & 0x02) { > + *(uint16_t *)dst =3D *(const uint16_t *)src; > + src =3D (const uint16_t *)src + 1; > + dst =3D (uint16_t *)dst + 1; > + } > + if (n & 0x04) { > + *(uint32_t *)dst =3D *(const uint32_t *)src; > + src =3D (const uint32_t *)src + 1; > + dst =3D (uint32_t *)dst + 1; > + } > + if (n & 0x08) { > + *(uint64_t *)dst =3D *(const uint64_t *)src; > + } > + return ret; > } >=20 > - /* > - * We split the remaining bytes (which will be less than 64) into > - * 16byte (2^4) chunks, using the same switch structure as above. > + /** > + * Fast way when copy size doesn't exceed 512 bytes > */ > - switch (3 - (n >> 4)) { > - case 0x00: > - rte_mov16((uint8_t *)dst, (const uint8_t *)src); > - n -=3D 16; > - dst =3D (uint8_t *)dst + 16; > - src =3D (const uint8_t *)src + 16; /* fallthrough */ > - case 0x01: > - rte_mov16((uint8_t *)dst, (const uint8_t *)src); > - n -=3D 16; > - dst =3D (uint8_t *)dst + 16; > - src =3D (const uint8_t *)src + 16; /* fallthrough */ > - case 0x02: > + if (n <=3D 32) { > rte_mov16((uint8_t *)dst, (const uint8_t *)src); > - n -=3D 16; > - dst =3D (uint8_t *)dst + 16; > - src =3D (const uint8_t *)src + 16; /* fallthrough */ > - default: > - ; > + rte_mov16((uint8_t *)dst - 16 + n, (const uint8_t *)src - 16 + n); > + return ret; > } > - > - /* Copy any remaining bytes, without going beyond end of buffers */ > - if (n !=3D 0) { > + if (n <=3D 48) { > + rte_mov32((uint8_t *)dst, (const uint8_t *)src); > + rte_mov16((uint8_t *)dst - 16 + n, (const uint8_t *)src - 16 + n); > + return ret; > + } > + if (n <=3D 64) { > + rte_mov32((uint8_t *)dst, (const uint8_t *)src); > + rte_mov16((uint8_t *)dst + 32, (const uint8_t *)src + 32); > rte_mov16((uint8_t *)dst - 16 + n, (const uint8_t *)src - 16 + n); > + return ret; > + } > + if (n <=3D 128) { > + goto COPY_BLOCK_128_BACK15; > } > - return ret; > + if (n <=3D 512) { > + if (n >=3D 256) { > + n -=3D 256; > + rte_mov128((uint8_t *)dst, (const uint8_t *)src); > + rte_mov128((uint8_t *)dst + 128, (const uint8_t *)src + 128); > + src =3D (const uint8_t *)src + 256; > + dst =3D (uint8_t *)dst + 256; > + } > +COPY_BLOCK_255_BACK15: > + if (n >=3D 128) { > + n -=3D 128; > + rte_mov128((uint8_t *)dst, (const uint8_t *)src); > + src =3D (const uint8_t *)src + 128; > + dst =3D (uint8_t *)dst + 128; > + } > +COPY_BLOCK_128_BACK15: > + if (n >=3D 64) { > + n -=3D 64; > + rte_mov64((uint8_t *)dst, (const uint8_t *)src); > + src =3D (const uint8_t *)src + 64; > + dst =3D (uint8_t *)dst + 64; > + } > +COPY_BLOCK_64_BACK15: > + if (n >=3D 32) { > + n -=3D 32; > + rte_mov32((uint8_t *)dst, (const uint8_t *)src); > + src =3D (const uint8_t *)src + 32; > + dst =3D (uint8_t *)dst + 32; > + } > + if (n > 16) { > + rte_mov16((uint8_t *)dst, (const uint8_t *)src); > + rte_mov16((uint8_t *)dst - 16 + n, (const uint8_t *)src - 16 + n); > + return ret; > + } > + if (n > 0) { > + rte_mov16((uint8_t *)dst - 16 + n, (const uint8_t *)src - 16 + n); > + } > + return ret; > + } > + > + /** > + * Make store aligned when copy size exceeds 512 bytes, > + * and make sure the first 15 bytes are copied, because > + * unaligned copy functions require up to 15 bytes > + * backwards access. > + */ > + dstofss =3D 16 - (int)((long long)(void *)dst & 0x0F) + 16; > + n -=3D dstofss; > + rte_mov32((uint8_t *)dst, (const uint8_t *)src); > + src =3D (const uint8_t *)src + dstofss; > + dst =3D (uint8_t *)dst + dstofss; > + srcofs =3D (int)((long long)(const void *)src & 0x0F); > + > + /** > + * For aligned copy > + */ > + if (srcofs =3D=3D 0) { > + /** > + * Copy 256-byte blocks > + */ > + for (; n >=3D 256; n -=3D 256) { > + rte_mov256((uint8_t *)dst, (const uint8_t *)src); > + dst =3D (uint8_t *)dst + 256; > + src =3D (const uint8_t *)src + 256; > + } > + > + /** > + * Copy whatever left > + */ > + goto COPY_BLOCK_255_BACK15; > + } > + > + /** > + * For copy with unaligned load > + */ > + MOVEUNALIGNED_LEFT47(dst, src, n, srcofs); > + > + /** > + * Copy whatever left > + */ > + goto COPY_BLOCK_64_BACK15; > } >=20 > +#endif /* RTE_MACHINE_CPUFLAG_AVX2 */ > + > #ifdef __cplusplus > } > #endif > -- > 1.9.3