Patch Detail

HTTP 200 OK
Allow: GET, PUT, PATCH, HEAD, OPTIONS
Content-Type: application/json
Vary: Accept

{
    "id": 7397,
    "url": "https://patches.dpdk.org/api/patches/7397/?format=api",
    "web_url": "https://patches.dpdk.org/project/dpdk/patch/20151003091246.5877843.29154.1@rehivetech.com/",
    "project": {
        "id": 1,
        "url": "https://patches.dpdk.org/api/projects/1/?format=api",
        "name": "DPDK",
        "link_name": "dpdk",
        "list_id": "dev.dpdk.org",
        "list_email": "dev@dpdk.org",
        "web_url": "http://core.dpdk.org",
        "scm_url": "git://dpdk.org/dpdk",
        "webscm_url": "http://git.dpdk.org/dpdk",
        "list_archive_url": "https://inbox.dpdk.org/dev",
        "list_archive_url_format": "https://inbox.dpdk.org/dev/{}",
        "commit_url_format": ""
    },
    "msgid": "<20151003091246.5877843.29154.1@rehivetech.com>",
    "list_archive_url": "https://inbox.dpdk.org/dev/20151003091246.5877843.29154.1@rehivetech.com",
    "date": "2015-10-03T09:12:46",
    "name": "[dpdk-dev,v1,07/12] eal/arm: vector memcpy for ARM",
    "commit_ref": null,
    "pull_url": null,
    "state": "not-applicable",
    "archived": true,
    "hash": "f4523d0ff81bfb62e21fc6edaa2d486c19dda16e",
    "submitter": {
        "id": 292,
        "url": "https://patches.dpdk.org/api/people/292/?format=api",
        "name": "Jan Viktorin",
        "email": "viktorin@rehivetech.com"
    },
    "delegate": null,
    "mbox": "https://patches.dpdk.org/project/dpdk/patch/20151003091246.5877843.29154.1@rehivetech.com/mbox/",
    "series": [],
    "comments": "https://patches.dpdk.org/api/patches/7397/comments/",
    "check": "pending",
    "checks": "https://patches.dpdk.org/api/patches/7397/checks/",
    "tags": {},
    "related": [],
    "headers": {
        "Return-Path": "<dev-bounces@dpdk.org>",
        "X-Original-To": "patchwork@dpdk.org",
        "Delivered-To": "patchwork@dpdk.org",
        "Received": [
            "from [92.243.14.124] (localhost [IPv6:::1])\n\tby dpdk.org (Postfix) with ESMTP id DBC398E61;\n\tSat,  3 Oct 2015 11:12:51 +0200 (CEST)",
            "from we2-f167.wedos.net (w-smtp-out-7.wedos.net [46.28.106.5])\n\tby dpdk.org (Postfix) with ESMTP id A17388D9F\n\tfor <dev@dpdk.org>; Sat,  3 Oct 2015 11:12:49 +0200 (CEST)",
            "from ([80.11.219.92])\n\tby we2-f167.wedos.net (WEDOS Mail Server mail2) with ASMTP (SSL) id\n\tQXP00047 for <dev@dpdk.org>; Sat, 03 Oct 2015 11:12:47 +0200"
        ],
        "Content-Type": "text/plain; charset=\"utf-8\"",
        "MIME-Version": "1.0",
        "Content-Transfer-Encoding": "quoted-printable",
        "X-Mailer": "BlackBerry Email (10.3.2.2639)",
        "Message-ID": "<20151003091246.5877843.29154.1@rehivetech.com>",
        "Date": "Sat, 03 Oct 2015 10:12:46 +0100",
        "From": "Jan Viktorin <viktorin@rehivetech.com>",
        "To": "dev@dpdk.org",
        "Subject": "Re: [dpdk-dev] [PATCH v1 07/12] eal/arm: vector memcpy for ARM",
        "X-BeenThere": "dev@dpdk.org",
        "X-Mailman-Version": "2.1.15",
        "Precedence": "list",
        "List-Id": "patches and discussions about DPDK <dev.dpdk.org>",
        "List-Unsubscribe": "<http://dpdk.org/ml/options/dev>,\n\t<mailto:dev-request@dpdk.org?subject=unsubscribe>",
        "List-Archive": "<http://dpdk.org/ml/archives/dev/>",
        "List-Post": "<mailto:dev@dpdk.org>",
        "List-Help": "<mailto:dev-request@dpdk.org?subject=help>",
        "List-Subscribe": "<http://dpdk.org/ml/listinfo/dev>,\n\t<mailto:dev-request@dpdk.org?subject=subscribe>",
        "Errors-To": "dev-bounces@dpdk.org",
        "Sender": "\"dev\" <dev-bounces@dpdk.org>"
    },
    "content": "‎I am missing the percent sign (%) for the measurments...\n\nSo it's eg. by 4.9 %‎ faster then memcpy...\n\nRegards\nJan Viktorin\n  Původní zpráva  \nOd: Jan Viktorin\nOdesláno: sobota, 3. října 2015 9:58\nKomu: dev@dpdk.org\nKopie: Vlastimil Kosar; Jan Viktorin\nPředmět: [PATCH v1 07/12] eal/arm: vector memcpy for ARM\n\nFrom: Vlastimil Kosar <kosar@rehivetech.com>\n\nThe SSE based memory copy in DPDK only support x86. This patch\nadds ARM NEON based memory copy functions for ARM architecture.\n\nThe implementation improves memory copy of short or well aligned\ndata buffers. The following measurements show improvements over\nthe libc memcpy on Cortex CPUs.\n\nLength (B) a15 a7 a9\n1 4.9 15.2 3.2\n7 56.9 48.2 40.3\n8 37.3 39.8 29.6\n9 69.3 38.7 33.9\n15 60.8 35.3 23.7\n16 50.6 35.9 35.0\n17 57.7 35.7 31.1\n31 16.0 23.3 9.0\n32 65.9 13.5 21.4\n33 3.9 10.3 -3.7\n63 2.0 12.9 -2.0\n64 66.5 0.0 16.5\n65 2.7 7.6 -35.6\n127 0.1 4.5 -18.9\n128 66.2 1.5 -51.4\n129 -0.8 3.2 -35.8\n255 -3.1 -0.9 -69.1\n256 67.9 1.2 7.2\n257 -3.6 -1.9 -36.9\n320 67.7 1.4 0.0\n384 66.8 1.4 -14.2\n511 -44.9 -2.3 -41.9\n512 67.3 1.4 -6.8\n513 -41.7 -3.0 -36.2\n1023 -82.4 -2.8 -41.2\n1024 68.3 1.4 -11.6\n1025 -80.1 -3.3 -38.1\n1518 -47.3 -5.0 -38.3\n1522 -48.3 -6.0 -37.9\n1600 65.4 1.3 -27.3\n2048 59.5 1.5 -10.9\n3072 52.3 1.5 -12.2\n4096 45.3 1.4 -12.5\n5120 40.6 1.5 -14.5\n6144 35.4 1.4 -13.4\n7168 32.9 1.4 -13.9\n8192 28.2 1.4 -15.1\n\nSigned-off-by: Vlastimil Kosar <kosar@rehivetech.com>\nSigned-off-by: Jan Viktorin <viktorin@rehivetech.com>\n---\n.../common/include/arch/arm/rte_memcpy.h | 270 +++++++++++++++++++++\n1 file changed, 270 insertions(+)\ncreate mode 100644 lib/librte_eal/common/include/arch/arm/rte_memcpy.h",
    "diff": "diff --git a/lib/librte_eal/common/include/arch/arm/rte_memcpy.h b/lib/librte_eal/common/include/arch/arm/rte_memcpy.h\nnew file mode 100644\nindex 0000000..ac885e9\n--- /dev/null\n+++ b/lib/librte_eal/common/include/arch/arm/rte_memcpy.h\n@@ -0,0 +1,270 @@\n+/*\n+ * BSD LICENSE\n+ *\n+ * Copyright(c) 2015 RehiveTech. All rights reserved.\n+ *\n+ * Redistribution and use in source and binary forms, with or without\n+ * modification, are permitted provided that the following conditions\n+ * are met:\n+ *\n+ * * Redistributions of source code must retain the above copyright\n+ * notice, this list of conditions and the following disclaimer.\n+ * * Redistributions in binary form must reproduce the above copyright\n+ * notice, this list of conditions and the following disclaimer in\n+ * the documentation and/or other materials provided with the\n+ * distribution.\n+ * * Neither the name of RehiveTech nor the names of its\n+ * contributors may be used to endorse or promote products derived\n+ * from this software without specific prior written permission.\n+ *\n+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS\n+ * \"AS IS\" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT\n+ * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR\n+ * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT\n+ * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,\n+ * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT\n+ * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,\n+ * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY\n+ * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT\n+ * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE\n+ * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.\n+ */\n+\n+#ifndef _RTE_MEMCPY_ARM_H_\n+#define _RTE_MEMCPY_ARM_H_\n+\n+#include <stdint.h>\n+#include <string.h>\n+/* ARM NEON Intrinsics are used to copy data */\n+#include <arm_neon.h>\n+\n+#ifdef __cplusplus\n+extern \"C\" {\n+#endif\n+\n+#include \"generic/rte_memcpy.h\"\n+\n+static inline void\n+rte_mov16(uint8_t *dst, const uint8_t *src)\n+{\n+\tvst1q_u8(dst, vld1q_u8(src));\n+}\n+\n+static inline void\n+rte_mov32(uint8_t *dst, const uint8_t *src)\n+{\n+\tasm volatile (\"vld1.8 {d0-d3}, [%[src]]\\n\\t\"\n+\t \"vst1.8 {d0-d3}, [%[dst]]\\n\\t\"\n+\t : [src] \"+r\" (src), [dst] \"+r\" (dst)\n+\t : : \"memory\", \"d0\", \"d1\", \"d2\", \"d3\");\n+}\n+\n+static inline void\n+rte_mov48(uint8_t *dst, const uint8_t *src)\n+{\n+\tasm volatile (\"vld1.8 {d0-d3}, [%[src]]!\\n\\t\"\n+\t \"vld1.8 {d4-d5}, [%[src]]\\n\\t\"\n+\t \"vst1.8 {d0-d3}, [%[dst]]!\\n\\t\"\n+\t \"vst1.8 {d4-d5}, [%[dst]]\\n\\t\"\n+\t : [src] \"+r\" (src), [dst] \"+r\" (dst)\n+\t : : \"memory\", \"d0\", \"d1\", \"d2\", \"d3\", \"d4\", \"d5\");\n+}\n+\n+static inline void\n+rte_mov64(uint8_t *dst, const uint8_t *src)\n+{\n+\tasm volatile (\"vld1.8 {d0-d3}, [%[src]]!\\n\\t\"\n+\t \"vld1.8 {d4-d7}, [%[src]]\\n\\t\"\n+\t \"vst1.8 {d0-d3}, [%[dst]]!\\n\\t\"\n+\t \"vst1.8 {d4-d7}, [%[dst]]\\n\\t\"\n+\t : [src] \"+r\" (src), [dst] \"+r\" (dst)\n+\t : : \"memory\", \"d0\", \"d1\", \"d2\", \"d3\", \"d4\", \"d5\", \"d6\", \"d7\");\n+}\n+\n+static inline void\n+rte_mov128(uint8_t *dst, const uint8_t *src)\n+{\n+\tasm volatile (\"pld [%[src], #64]\" : : [src] \"r\" (src));\n+\tasm volatile (\"vld1.8 {d0-d3}, [%[src]]!\\n\\t\"\n+\t \"vld1.8 {d4-d7}, [%[src]]!\\n\\t\"\n+\t \"vld1.8 {d8-d11}, [%[src]]!\\n\\t\"\n+\t \"vld1.8 {d12-d15}, [%[src]]\\n\\t\"\n+\t \"vst1.8 {d0-d3}, [%[dst]]!\\n\\t\"\n+\t \"vst1.8 {d4-d7}, [%[dst]]!\\n\\t\"\n+\t \"vst1.8 {d8-d11}, [%[dst]]!\\n\\t\"\n+\t \"vst1.8 {d12-d15}, [%[dst]]\\n\\t\"\n+\t : [src] \"+r\" (src), [dst] \"+r\" (dst)\n+\t : : \"memory\", \"d0\", \"d1\", \"d2\", \"d3\", \"d4\", \"d5\", \"d6\", \"d7\",\n+\t \"d8\", \"d9\", \"d10\", \"d11\", \"d12\", \"d13\", \"d14\", \"d15\");\n+}\n+\n+static inline void\n+rte_mov256(uint8_t *dst, const uint8_t *src)\n+{\n+\tasm volatile (\"pld [%[src], #64]\" : : [src] \"r\" (src));\n+\tasm volatile (\"pld [%[src], #128]\" : : [src] \"r\" (src));\n+\tasm volatile (\"pld [%[src], #192]\" : : [src] \"r\" (src));\n+\tasm volatile (\"pld [%[src], #256]\" : : [src] \"r\" (src));\n+\tasm volatile (\"pld [%[src], #320]\" : : [src] \"r\" (src));\n+\tasm volatile (\"pld [%[src], #384]\" : : [src] \"r\" (src));\n+\tasm volatile (\"pld [%[src], #448]\" : : [src] \"r\" (src));\n+\tasm volatile (\"vld1.8 {d0-d3}, [%[src]]!\\n\\t\"\n+\t \"vld1.8 {d4-d7}, [%[src]]!\\n\\t\"\n+\t \"vld1.8 {d8-d11}, [%[src]]!\\n\\t\"\n+\t \"vld1.8 {d12-d15}, [%[src]]!\\n\\t\"\n+\t \"vld1.8 {d16-d19}, [%[src]]!\\n\\t\"\n+\t \"vld1.8 {d20-d23}, [%[src]]!\\n\\t\"\n+\t \"vld1.8 {d24-d27}, [%[src]]!\\n\\t\"\n+\t \"vld1.8 {d28-d31}, [%[src]]\\n\\t\"\n+\t \"vst1.8 {d0-d3}, [%[dst]]!\\n\\t\"\n+\t \"vst1.8 {d4-d7}, [%[dst]]!\\n\\t\"\n+\t \"vst1.8 {d8-d11}, [%[dst]]!\\n\\t\"\n+\t \"vst1.8 {d12-d15}, [%[dst]]!\\n\\t\"\n+\t \"vst1.8 {d16-d19}, [%[dst]]!\\n\\t\"\n+\t \"vst1.8 {d20-d23}, [%[dst]]!\\n\\t\"\n+\t \"vst1.8 {d24-d27}, [%[dst]]!\\n\\t\"\n+\t \"vst1.8 {d28-d31}, [%[dst]]!\\n\\t\"\n+\t : [src] \"+r\" (src), [dst] \"+r\" (dst)\n+\t : : \"memory\", \"d0\", \"d1\", \"d2\", \"d3\", \"d4\", \"d5\", \"d6\", \"d7\",\n+\t \"d8\", \"d9\", \"d10\", \"d11\", \"d12\", \"d13\", \"d14\", \"d15\",\n+\t \"d16\", \"d17\", \"d18\", \"d19\", \"d20\", \"d21\", \"d22\", \"d23\",\n+\t \"d24\", \"d25\", \"d26\", \"d27\", \"d28\", \"d29\", \"d30\", \"d31\");\n+}\n+\n+#define rte_memcpy(dst, src, n) \\\n+\t({ (__builtin_constant_p(n)) ? \\\n+\tmemcpy((dst), (src), (n)) : \\\n+\trte_memcpy_func((dst), (src), (n)); })\n+\n+static inline void *\n+rte_memcpy_func(void *dst, const void *src, size_t n)\n+{\n+\tvoid *ret = dst;\n+\n+\t/* We can't copy < 16 bytes using XMM registers so do it manually. */\n+\tif (n < 16) {\n+\t if (n & 0x01) {\n+\t *(uint8_t *)dst = *(const uint8_t *)src;\n+\t dst = (uint8_t *)dst + 1;\n+\t src = (const uint8_t *)src + 1;\n+\t }\n+\t if (n & 0x02) {\n+\t *(uint16_t *)dst = *(const uint16_t *)src;\n+\t dst = (uint16_t *)dst + 1;\n+\t src = (const uint16_t *)src + 1;\n+\t }\n+\t if (n & 0x04) {\n+\t *(uint32_t *)dst = *(const uint32_t *)src;\n+\t dst = (uint32_t *)dst + 1;\n+\t src = (const uint32_t *)src + 1;\n+\t }\n+\t if (n & 0x08) {\n+\t /* ARMv7 can not handle unaligned access to long long\n+\t * (uint64_t). Therefore two uint32_t operations are used.\n+\t * TODO: use NEON too?\n+\t */\n+\t *(uint32_t *)dst = *(const uint32_t *)src;\n+\t dst = (uint32_t *)dst + 1;\n+\t src = (const uint32_t *)src + 1;\n+\t *(uint32_t *)dst = *(const uint32_t *)src;\n+\t }\n+\t return ret;\n+\t}\n+\n+\t/* Special fast cases for <= 128 bytes */\n+\tif (n <= 32) {\n+\t rte_mov16((uint8_t *)dst, (const uint8_t *)src);\n+\t rte_mov16((uint8_t *)dst - 16 + n,\n+\t (const uint8_t *)src - 16 + n);\n+\t return ret;\n+\t}\n+\n+\tif (n <= 64) {\n+\t rte_mov32((uint8_t *)dst, (const uint8_t *)src);\n+\t rte_mov32((uint8_t *)dst - 32 + n,\n+\t (const uint8_t *)src - 32 + n);\n+\t return ret;\n+\t}\n+\n+\tif (n <= 128) {\n+\t rte_mov64((uint8_t *)dst, (const uint8_t *)src);\n+\t rte_mov64((uint8_t *)dst - 64 + n,\n+\t (const uint8_t *)src - 64 + n);\n+\t return ret;\n+\t}\n+\n+\t/*\n+\t * For large copies > 128 bytes. This combination of 256, 64 and 16 byte\n+\t * copies was found to be faster than doing 128 and 32 byte copies as\n+\t * well.\n+\t */\n+\tfor ( ; n >= 256; n -= 256) {\n+\t rte_mov256((uint8_t *)dst, (const uint8_t *)src);\n+\t dst = (uint8_t *)dst + 256;\n+\t src = (const uint8_t *)src + 256;\n+\t}\n+\n+\t/*\n+\t * We split the remaining bytes (which will be less than 256) into\n+\t * 64byte (2^6) chunks.\n+\t * Using incrementing integers in the case labels of a switch statement\n+\t * enourages the compiler to use a jump table. To get incrementing\n+\t * integers, we shift the 2 relevant bits to the LSB position to first\n+\t * get decrementing integers, and then subtract.\n+\t */\n+\tswitch (3 - (n >> 6)) {\n+\tcase 0x00:\n+\t rte_mov64((uint8_t *)dst, (const uint8_t *)src);\n+\t n -= 64;\n+\t dst = (uint8_t *)dst + 64;\n+\t src = (const uint8_t *)src + 64; /* fallthrough */\n+\tcase 0x01:\n+\t rte_mov64((uint8_t *)dst, (const uint8_t *)src);\n+\t n -= 64;\n+\t dst = (uint8_t *)dst + 64;\n+\t src = (const uint8_t *)src + 64; /* fallthrough */\n+\tcase 0x02:\n+\t rte_mov64((uint8_t *)dst, (const uint8_t *)src);\n+\t n -= 64;\n+\t dst = (uint8_t *)dst + 64;\n+\t src = (const uint8_t *)src + 64; /* fallthrough */\n+\tdefault:\n+\t ;\n+\t}\n+\n+\t/*\n+\t * We split the remaining bytes (which will be less than 64) into\n+\t * 16byte (2^4) chunks, using the same switch structure as above.\n+\t */\n+\tswitch (3 - (n >> 4)) {\n+\tcase 0x00:\n+\t rte_mov16((uint8_t *)dst, (const uint8_t *)src);\n+\t n -= 16;\n+\t dst = (uint8_t *)dst + 16;\n+\t src = (const uint8_t *)src + 16; /* fallthrough */\n+\tcase 0x01:\n+\t rte_mov16((uint8_t *)dst, (const uint8_t *)src);\n+\t n -= 16;\n+\t dst = (uint8_t *)dst + 16;\n+\t src = (const uint8_t *)src + 16; /* fallthrough */\n+\tcase 0x02:\n+\t rte_mov16((uint8_t *)dst, (const uint8_t *)src);\n+\t n -= 16;\n+\t dst = (uint8_t *)dst + 16;\n+\t src = (const uint8_t *)src + 16; /* fallthrough */\n+\tdefault:\n+\t ;\n+\t}\n+\n+\t/* Copy any remaining bytes, without going beyond end of buffers */\n+\tif (n != 0)\n+\t rte_mov16((uint8_t *)dst - 16 + n,\n+\t (const uint8_t *)src - 16 + n);\n+\treturn ret;\n+}\n+\n+#ifdef __cplusplus\n+}\n+#endif\n+\n+#endif /* _RTE_MEMCPY_ARM_H_ */\n",
    "prefixes": [
        "dpdk-dev",
        "v1",
        "07/12"
    ]
}