vppinfra: add AVX512 variant of clib_memcpy

Taken from DPDK, also AVX2 variant updated to be in sync with DPDK
version.

Change-Id: I8a42e4141a5a1a8cfbee328b07bd0c9b38a9eb05
Signed-off-by: Damjan Marion <damarion@cisco.com>

4 files changed, 357 insertions, 129 deletions

diff --git a/src/vppinfra/memcpy_avx.h b/src/vppinfra/memcpy_avx2.h
index e987d044b58..da398a4512a 100644
--- a/src/vppinfra/memcpy_avx.h
+++ b/src/vppinfra/memcpy_avx2.h
@@ -45,8 +45,8 @@
  *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 
-#ifndef included_clib_memcpy_avx_h
-#define included_clib_memcpy_avx_h
+#ifndef included_clib_memcpy_avx2_h
+#define included_clib_memcpy_avx2_h
 
 #include <stdint.h>
 #include <x86intrin.h>
@@ -84,65 +84,27 @@ clib_mov128 (u8 * dst, const u8 * src)
 }
 
 static inline void
-clib_mov256 (u8 * dst, const u8 * src)
+clib_mov128blocks (u8 * dst, const u8 * src, size_t n)
 {
-  clib_mov128 ((u8 *) dst + 0 * 128, (const u8 *) src + 0 * 128);
-  clib_mov128 ((u8 *) dst + 1 * 128, (const u8 *) src + 1 * 128);
-}
-
-static inline void
-clib_mov64blocks (u8 * dst, const u8 * src, size_t n)
-{
-  __m256i ymm0, ymm1;
+  __m256i ymm0, ymm1, ymm2, ymm3;
 
-  while (n >= 64)
+  while (n >= 128)
     {
       ymm0 =
 	_mm256_loadu_si256 ((const __m256i *) ((const u8 *) src + 0 * 32));
-      n -= 64;
-      ymm1 =
-	_mm256_loadu_si256 ((const __m256i *) ((const u8 *) src + 1 * 32));
-      src = (const u8 *) src + 64;
-      _mm256_storeu_si256 ((__m256i *) ((u8 *) dst + 0 * 32), ymm0);
-      _mm256_storeu_si256 ((__m256i *) ((u8 *) dst + 1 * 32), ymm1);
-      dst = (u8 *) dst + 64;
-    }
-}
-
-static inline void
-clib_mov256blocks (u8 * dst, const u8 * src, size_t n)
-{
-  __m256i ymm0, ymm1, ymm2, ymm3, ymm4, ymm5, ymm6, ymm7;
-
-  while (n >= 256)
-    {
-      ymm0 =
-	_mm256_loadu_si256 ((const __m256i *) ((const u8 *) src + 0 * 32));
-      n -= 256;
+      n -= 128;
       ymm1 =
 	_mm256_loadu_si256 ((const __m256i *) ((const u8 *) src + 1 * 32));
       ymm2 =
 	_mm256_loadu_si256 ((const __m256i *) ((const u8 *) src + 2 * 32));
       ymm3 =
 	_mm256_loadu_si256 ((const __m256i *) ((const u8 *) src + 3 * 32));
-      ymm4 =
-	_mm256_loadu_si256 ((const __m256i *) ((const u8 *) src + 4 * 32));
-      ymm5 =
-	_mm256_loadu_si256 ((const __m256i *) ((const u8 *) src + 5 * 32));
-      ymm6 =
-	_mm256_loadu_si256 ((const __m256i *) ((const u8 *) src + 6 * 32));
-      ymm7 =
-	_mm256_loadu_si256 ((const __m256i *) ((const u8 *) src + 7 * 32));
-      src = (const u8 *) src + 256;
+      src = (const u8 *) src + 128;
       _mm256_storeu_si256 ((__m256i *) ((u8 *) dst + 0 * 32), ymm0);
       _mm256_storeu_si256 ((__m256i *) ((u8 *) dst + 1 * 32), ymm1);
       _mm256_storeu_si256 ((__m256i *) ((u8 *) dst + 2 * 32), ymm2);
       _mm256_storeu_si256 ((__m256i *) ((u8 *) dst + 3 * 32), ymm3);
-      _mm256_storeu_si256 ((__m256i *) ((u8 *) dst + 4 * 32), ymm4);
-      _mm256_storeu_si256 ((__m256i *) ((u8 *) dst + 5 * 32), ymm5);
-      _mm256_storeu_si256 ((__m256i *) ((u8 *) dst + 6 * 32), ymm6);
-      _mm256_storeu_si256 ((__m256i *) ((u8 *) dst + 7 * 32), ymm7);
-      dst = (u8 *) dst + 256;
+      dst = (u8 *) dst + 128;
     }
 }
 
@@ -155,9 +117,9 @@ clib_memcpy (void *dst, const void *src, size_t n)
   size_t dstofss;
   size_t bits;
 
-	/**
-         * Copy less than 16 bytes
-         */
+  /**
+   * Copy less than 16 bytes
+   */
   if (n < 16)
     {
       if (n & 0x01)
@@ -168,47 +130,47 @@ clib_memcpy (void *dst, const void *src, size_t n)
 	}
       if (n & 0x02)
 	{
-	  *(uint16_t *) dstu = *(const uint16_t *) srcu;
-	  srcu = (uword) ((const uint16_t *) srcu + 1);
-	  dstu = (uword) ((uint16_t *) dstu + 1);
+	  *(u16 *) dstu = *(const u16 *) srcu;
+	  srcu = (uword) ((const u16 *) srcu + 1);
+	  dstu = (uword) ((u16 *) dstu + 1);
 	}
       if (n & 0x04)
 	{
-	  *(uint32_t *) dstu = *(const uint32_t *) srcu;
-	  srcu = (uword) ((const uint32_t *) srcu + 1);
-	  dstu = (uword) ((uint32_t *) dstu + 1);
+	  *(u32 *) dstu = *(const u32 *) srcu;
+	  srcu = (uword) ((const u32 *) srcu + 1);
+	  dstu = (uword) ((u32 *) dstu + 1);
 	}
       if (n & 0x08)
 	{
-	  *(uint64_t *) dstu = *(const uint64_t *) srcu;
+	  *(u64 *) dstu = *(const u64 *) srcu;
 	}
       return ret;
     }
 
-	/**
-         * Fast way when copy size doesn't exceed 512 bytes
-         */
+  /**
+    * Fast way when copy size doesn't exceed 512 bytes
+    */
   if (n <= 32)
     {
       clib_mov16 ((u8 *) dst, (const u8 *) src);
       clib_mov16 ((u8 *) dst - 16 + n, (const u8 *) src - 16 + n);
       return ret;
     }
+  if (n <= 48)
+    {
+      clib_mov16 ((u8 *) dst, (const u8 *) src);
+      clib_mov16 ((u8 *) dst + 16, (const u8 *) src + 16);
+      clib_mov16 ((u8 *) dst - 16 + n, (const u8 *) src - 16 + n);
+      return ret;
+    }
   if (n <= 64)
     {
       clib_mov32 ((u8 *) dst, (const u8 *) src);
       clib_mov32 ((u8 *) dst - 32 + n, (const u8 *) src - 32 + n);
       return ret;
     }
-  if (n <= 512)
+  if (n <= 256)
     {
-      if (n >= 256)
-	{
-	  n -= 256;
-	  clib_mov256 ((u8 *) dst, (const u8 *) src);
-	  src = (const u8 *) src + 256;
-	  dst = (u8 *) dst + 256;
-	}
       if (n >= 128)
 	{
 	  n -= 128;
@@ -216,6 +178,7 @@ clib_memcpy (void *dst, const void *src, size_t n)
 	  src = (const u8 *) src + 128;
 	  dst = (u8 *) dst + 128;
 	}
+    COPY_BLOCK_128_BACK31:
       if (n >= 64)
 	{
 	  n -= 64;
@@ -223,7 +186,6 @@ clib_memcpy (void *dst, const void *src, size_t n)
 	  src = (const u8 *) src + 64;
 	  dst = (u8 *) dst + 64;
 	}
-    COPY_BLOCK_64_BACK31:
       if (n > 32)
 	{
 	  clib_mov32 ((u8 *) dst, (const u8 *) src);
@@ -237,9 +199,9 @@ clib_memcpy (void *dst, const void *src, size_t n)
       return ret;
     }
 
-	/**
-         * Make store aligned when copy size exceeds 512 bytes
-         */
+    /**
+      * Make store aligned when copy size exceeds 256 bytes
+      */
   dstofss = (uword) dst & 0x1F;
   if (dstofss > 0)
     {
@@ -250,41 +212,24 @@ clib_memcpy (void *dst, const void *src, size_t n)
       dst = (u8 *) dst + dstofss;
     }
 
-	/**
-         * Copy 256-byte blocks.
-         * Use copy block function for better instruction order control,
-         * which is important when load is unaligned.
-         */
-  clib_mov256blocks ((u8 *) dst, (const u8 *) src, n);
+  /**
+    * Copy 128-byte blocks.
+    */
+  clib_mov128blocks ((u8 *) dst, (const u8 *) src, n);
   bits = n;
-  n = n & 255;
+  n = n & 127;
   bits -= n;
   src = (const u8 *) src + bits;
   dst = (u8 *) dst + bits;
 
-	/**
-         * Copy 64-byte blocks.
-         * Use copy block function for better instruction order control,
-         * which is important when load is unaligned.
-         */
-  if (n >= 64)
-    {
-      clib_mov64blocks ((u8 *) dst, (const u8 *) src, n);
-      bits = n;
-      n = n & 63;
-      bits -= n;
-      src = (const u8 *) src + bits;
-      dst = (u8 *) dst + bits;
-    }
-
-	/**
-         * Copy whatever left
-         */
-  goto COPY_BLOCK_64_BACK31;
+  /**
+   * Copy whatever left
+   */
+  goto COPY_BLOCK_128_BACK31;
 }
 
 
-#endif /* included_clib_mamcpy_avx_h */
+#endif /* included_clib_mamcpy_avx2_h */
 
 
 /*
diff --git a/src/vppinfra/memcpy_avx512.h b/src/vppinfra/memcpy_avx512.h
new file mode 100644
index 00000000000..003908026c3
--- /dev/null
+++ b/src/vppinfra/memcpy_avx512.h
@@ -0,0 +1,277 @@
+/*
+ * Copyright (c) 2016 Cisco and/or its affiliates.
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at:
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
+ *   All rights reserved.
+ *
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
+ *   are met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in
+ *       the documentation and/or other materials provided with the
+ *       distribution.
+ *     * Neither the name of Intel Corporation nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#ifndef included_clib_memcpy_avx512_h
+#define included_clib_memcpy_avx512_h
+
+#include <stdint.h>
+#include <x86intrin.h>
+
+static inline void
+clib_mov16 (u8 * dst, const u8 * src)
+{
+  __m128i xmm0;
+
+  xmm0 = _mm_loadu_si128 ((const __m128i *) src);
+  _mm_storeu_si128 ((__m128i *) dst, xmm0);
+}
+
+static inline void
+clib_mov32 (u8 * dst, const u8 * src)
+{
+  __m256i ymm0;
+
+  ymm0 = _mm256_loadu_si256 ((const __m256i *) src);
+  _mm256_storeu_si256 ((__m256i *) dst, ymm0);
+}
+
+static inline void
+clib_mov64 (u8 * dst, const u8 * src)
+{
+  __m512i zmm0;
+
+  zmm0 = _mm512_loadu_si512 ((const void *) src);
+  _mm512_storeu_si512 ((void *) dst, zmm0);
+}
+
+static inline void
+clib_mov128 (u8 * dst, const u8 * src)
+{
+  clib_mov64 (dst + 0 * 64, src + 0 * 64);
+  clib_mov64 (dst + 1 * 64, src + 1 * 64);
+}
+
+static inline void
+clib_mov256 (u8 * dst, const u8 * src)
+{
+  clib_mov128 (dst + 0 * 128, src + 0 * 128);
+  clib_mov128 (dst + 1 * 128, src + 1 * 128);
+}
+
+static inline void
+clib_mov128blocks (u8 * dst, const u8 * src, size_t n)
+{
+  __m512i zmm0, zmm1;
+
+  while (n >= 128)
+    {
+      zmm0 = _mm512_loadu_si512 ((const void *) (src + 0 * 64));
+      n -= 128;
+      zmm1 = _mm512_loadu_si512 ((const void *) (src + 1 * 64));
+      src = src + 128;
+      _mm512_storeu_si512 ((void *) (dst + 0 * 64), zmm0);
+      _mm512_storeu_si512 ((void *) (dst + 1 * 64), zmm1);
+      dst = dst + 128;
+    }
+}
+
+static inline void
+clib_mov512blocks (u8 * dst, const u8 * src, size_t n)
+{
+  __m512i zmm0, zmm1, zmm2, zmm3, zmm4, zmm5, zmm6, zmm7;
+
+  while (n >= 512)
+    {
+      zmm0 = _mm512_loadu_si512 ((const void *) (src + 0 * 64));
+      n -= 512;
+      zmm1 = _mm512_loadu_si512 ((const void *) (src + 1 * 64));
+      zmm2 = _mm512_loadu_si512 ((const void *) (src + 2 * 64));
+      zmm3 = _mm512_loadu_si512 ((const void *) (src + 3 * 64));
+      zmm4 = _mm512_loadu_si512 ((const void *) (src + 4 * 64));
+      zmm5 = _mm512_loadu_si512 ((const void *) (src + 5 * 64));
+      zmm6 = _mm512_loadu_si512 ((const void *) (src + 6 * 64));
+      zmm7 = _mm512_loadu_si512 ((const void *) (src + 7 * 64));
+      src = src + 512;
+      _mm512_storeu_si512 ((void *) (dst + 0 * 64), zmm0);
+      _mm512_storeu_si512 ((void *) (dst + 1 * 64), zmm1);
+      _mm512_storeu_si512 ((void *) (dst + 2 * 64), zmm2);
+      _mm512_storeu_si512 ((void *) (dst + 3 * 64), zmm3);
+      _mm512_storeu_si512 ((void *) (dst + 4 * 64), zmm4);
+      _mm512_storeu_si512 ((void *) (dst + 5 * 64), zmm5);
+      _mm512_storeu_si512 ((void *) (dst + 6 * 64), zmm6);
+      _mm512_storeu_si512 ((void *) (dst + 7 * 64), zmm7);
+      dst = dst + 512;
+    }
+}
+
+static inline void *
+clib_memcpy (void *dst, const void *src, size_t n)
+{
+  uword dstu = (uword) dst;
+  uword srcu = (uword) src;
+  void *ret = dst;
+  size_t dstofss;
+  size_t bits;
+
+	/**
+         * Copy less than 16 bytes
+         */
+  if (n < 16)
+    {
+      if (n & 0x01)
+	{
+	  *(u8 *) dstu = *(const u8 *) srcu;
+	  srcu = (uword) ((const u8 *) srcu + 1);
+	  dstu = (uword) ((u8 *) dstu + 1);
+	}
+      if (n & 0x02)
+	{
+	  *(u16 *) dstu = *(const u16 *) srcu;
+	  srcu = (uword) ((const u16 *) srcu + 1);
+	  dstu = (uword) ((u16 *) dstu + 1);
+	}
+      if (n & 0x04)
+	{
+	  *(u32 *) dstu = *(const u32 *) srcu;
+	  srcu = (uword) ((const u32 *) srcu + 1);
+	  dstu = (uword) ((u32 *) dstu + 1);
+	}
+      if (n & 0x08)
+	*(u64 *) dstu = *(const u64 *) srcu;
+      return ret;
+    }
+
+	/**
+         * Fast way when copy size doesn't exceed 512 bytes
+         */
+  if (n <= 32)
+    {
+      clib_mov16 ((u8 *) dst, (const u8 *) src);
+      clib_mov16 ((u8 *) dst - 16 + n, (const u8 *) src - 16 + n);
+      return ret;
+    }
+  if (n <= 64)
+    {
+      clib_mov32 ((u8 *) dst, (const u8 *) src);
+      clib_mov32 ((u8 *) dst - 32 + n, (const u8 *) src - 32 + n);
+      return ret;
+    }
+  if (n <= 512)
+    {
+      if (n >= 256)
+	{
+	  n -= 256;
+	  clib_mov256 ((u8 *) dst, (const u8 *) src);
+	  src = (const u8 *) src + 256;
+	  dst = (u8 *) dst + 256;
+	}
+      if (n >= 128)
+	{
+	  n -= 128;
+	  clib_mov128 ((u8 *) dst, (const u8 *) src);
+	  src = (const u8 *) src + 128;
+	  dst = (u8 *) dst + 128;
+	}
+    COPY_BLOCK_128_BACK63:
+      if (n > 64)
+	{
+	  clib_mov64 ((u8 *) dst, (const u8 *) src);
+	  clib_mov64 ((u8 *) dst - 64 + n, (const u8 *) src - 64 + n);
+	  return ret;
+	}
+      if (n > 0)
+	clib_mov64 ((u8 *) dst - 64 + n, (const u8 *) src - 64 + n);
+      return ret;
+    }
+
+	/**
+         * Make store aligned when copy size exceeds 512 bytes
+         */
+  dstofss = (uword) dst & 0x3F;
+  if (dstofss > 0)
+    {
+      dstofss = 64 - dstofss;
+      n -= dstofss;
+      clib_mov64 ((u8 *) dst, (const u8 *) src);
+      src = (const u8 *) src + dstofss;
+      dst = (u8 *) dst + dstofss;
+    }
+
+	/**
+         * Copy 512-byte blocks.
+         * Use copy block function for better instruction order control,
+         * which is important when load is unaligned.
+         */
+  clib_mov512blocks ((u8 *) dst, (const u8 *) src, n);
+  bits = n;
+  n = n & 511;
+  bits -= n;
+  src = (const u8 *) src + bits;
+  dst = (u8 *) dst + bits;
+
+	/**
+         * Copy 128-byte blocks.
+         * Use copy block function for better instruction order control,
+         * which is important when load is unaligned.
+         */
+  if (n >= 128)
+    {
+      clib_mov128blocks ((u8 *) dst, (const u8 *) src, n);
+      bits = n;
+      n = n & 127;
+      bits -= n;
+      src = (const u8 *) src + bits;
+      dst = (u8 *) dst + bits;
+    }
+
+	/**
+         * Copy whatever left
+         */
+  goto COPY_BLOCK_128_BACK63;
+}
+
+
+#endif /* included_clib_mamcpy_avx512_h */
+
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */
diff --git a/src/vppinfra/memcpy_sse3.h b/src/vppinfra/memcpy_sse3.h
index f61396c8922..5e4bf7d79c5 100644
--- a/src/vppinfra/memcpy_sse3.h
+++ b/src/vppinfra/memcpy_sse3.h
@@ -222,9 +222,9 @@ clib_memcpy (void *dst, const void *src, size_t n)
       return ret;
     }
 
-	/**
-	 * Fast way when copy size doesn't exceed 512 bytes
-	 */
+  /**
+   * Fast way when copy size doesn't exceed 512 bytes
+   */
   if (n <= 32)
     {
       clib_mov16 ((u8 *) dst, (const u8 *) src);
@@ -295,27 +295,31 @@ clib_memcpy (void *dst, const void *src, size_t 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 = 16 - ((uword) dst & 0x0F) + 16;
-  n -= dstofss;
-  clib_mov32 ((u8 *) dst, (const u8 *) src);
-  src = (const u8 *) src + dstofss;
-  dst = (u8 *) dst + dstofss;
+  /**
+   * 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 = (uword) dst & 0x0F;
+  if (dstofss > 0)
+    {
+      dstofss = 16 - dstofss + 16;
+      n -= dstofss;
+      clib_mov32 ((u8 *) dst, (const u8 *) src);
+      src = (const u8 *) src + dstofss;
+      dst = (u8 *) dst + dstofss;
+    }
   srcofs = ((uword) src & 0x0F);
 
-	/**
-	 * For aligned copy
-	 */
+  /**
+   * For aligned copy
+   */
   if (srcofs == 0)
     {
-		/**
-		 * Copy 256-byte blocks
-		 */
+      /**
+       * Copy 256-byte blocks
+       */
       for (; n >= 256; n -= 256)
 	{
 	  clib_mov256 ((u8 *) dst, (const u8 *) src);
@@ -323,20 +327,20 @@ clib_memcpy (void *dst, const void *src, size_t n)
 	  src = (const u8 *) src + 256;
 	}
 
-		/**
-		 * Copy whatever left
-		 */
+      /**
+       * Copy whatever left
+       */
       goto COPY_BLOCK_255_BACK15;
     }
 
-	/**
-	 * For copy with unaligned load
-	 */
+  /**
+   * For copy with unaligned load
+   */
   CLIB_MVUNALIGN_LEFT47 (dst, src, n, srcofs);
 
-	/**
-	 * Copy whatever left
-	 */
+  /**
+   * Copy whatever left
+   */
   goto COPY_BLOCK_64_BACK15;
 }
 
diff --git a/src/vppinfra/string.h b/src/vppinfra/string.h
index 0e0627d088d..7873e4647d9 100644
--- a/src/vppinfra/string.h
+++ b/src/vppinfra/string.h
@@ -65,8 +65,10 @@ void clib_memswap (void *_a, void *_b, uword bytes);
  * so don't let it anywhere near them.
  */
 #ifndef __COVERITY__
-#if __AVX__
-#include <vppinfra/memcpy_avx.h>
+#if __AVX512F__
+#include <vppinfra/memcpy_avx512.h>
+#elif __AVX2__
+#include <vppinfra/memcpy_avx2.h>
 #elif __SSSE3__
 #include <vppinfra/memcpy_sse3.h>
 #else