8 files changed, 100 insertions, 10 deletions
diff --git a/plugins/lb-plugin/lb/lbhash.h b/plugins/lb-plugin/lb/lbhash.h
index d47b49828fa..ca3cc143dc2 100644
--- a/plugins/lb-plugin/lb/lbhash.h
+++ b/plugins/lb-plugin/lb/lbhash.h
@@ -31,7 +31,9 @@
 
 #include <vnet/vnet.h>
 
+#if defined (__SSE4_2__)
 #include <immintrin.h>
+#endif
 
 /*
  * @brief Number of entries per bucket.
diff --git a/vlib/vlib/buffer.h b/vlib/vlib/buffer.h
index 4ede684aa4b..5f1e62f08c9 100644
--- a/vlib/vlib/buffer.h
+++ b/vlib/vlib/buffer.h
@@ -56,7 +56,7 @@
 #define VLIB_BUFFER_PRE_DATA_SIZE	__PRE_DATA_SIZE
 #endif
 
-#ifdef CLIB_HAVE_VEC128
+#if defined (CLIB_HAVE_VEC128) || defined (__aarch64__)
 typedef u8x16 vlib_copy_unit_t;
 #else
 typedef u64 vlib_copy_unit_t;
diff --git a/vlib/vlib/node.c b/vlib/vlib/node.c
index 23f7ea0239e..c419a13a487 100644
--- a/vlib/vlib/node.c
+++ b/vlib/vlib/node.c
@@ -433,9 +433,9 @@ register_node (vlib_main_t * vm, vlib_node_registration_t * r)
     for (i = 0; i < vec_len (rt->errors); i++)
       rt->errors[i] = vlib_error_set (n->index, i);
 
-    STATIC_ASSERT (sizeof (vlib_node_runtime_t) == 2 * CLIB_CACHE_LINE_BYTES,
-		   "Size of vlib_node_runtime_t must be equal to 2 cachelines");
-    ASSERT (vec_len (n->runtime_data) <= sizeof (vlib_node_runtime_t) -
+    STATIC_ASSERT_SIZEOF (vlib_node_runtime_t, 128);
+    ASSERT (vec_len (n->runtime_data) <=
+	    sizeof (vlib_node_runtime_t) -
 	    STRUCT_OFFSET_OF (vlib_node_runtime_t, runtime_data));
 
     if (vec_len (n->runtime_data) > 0)
diff --git a/vppinfra/Makefile.am b/vppinfra/Makefile.am
index d0a023e25d3..6183cd7a74e 100644
--- a/vppinfra/Makefile.am
+++ b/vppinfra/Makefile.am
@@ -214,6 +214,7 @@ nobase_include_HEADERS = \
   vppinfra/vector_altivec.h \
   vppinfra/vector_funcs.h \
   vppinfra/vector_iwmmxt.h \
+  vppinfra/vector_neon.h \
   vppinfra/vector_sse2.h \
   vppinfra/valgrind.h \
   vppinfra/vm_unix.h \
diff --git a/vppinfra/vppinfra/cpu.c b/vppinfra/vppinfra/cpu.c
index 70b39214817..a26d5c9ae10 100644
--- a/vppinfra/vppinfra/cpu.c
+++ b/vppinfra/vppinfra/cpu.c
@@ -101,6 +101,8 @@ format_cpu_model_name (u8 * s, va_list * args)
   vec_free (name);
   return s;
 
+#elif defined(__aarch64__)
+  return format (s, "armv8");
 #else /* ! __x86_64__ */
   return format (s, "unknown");
 #endif
@@ -109,7 +111,7 @@ format_cpu_model_name (u8 * s, va_list * args)
 u8 *
 format_cpu_flags (u8 * s, va_list * args)
 {
-#if __x86_64__
+#if defined(__x86_64__)
 #define _(flag, func, reg, bit) \
   if (clib_cpu_supports_ ## flag()) \
     s = format (s, #flag " ");
diff --git a/vppinfra/vppinfra/cpu.h b/vppinfra/vppinfra/cpu.h
index 7d63f85b506..9c149f3fa2a 100644
--- a/vppinfra/vppinfra/cpu.h
+++ b/vppinfra/vppinfra/cpu.h
@@ -51,8 +51,6 @@
   return & fn;                                                         \
 }
 
-#if __x86_64__
-#include "cpuid.h"
 
 #define foreach_x86_64_flags \
 _ (sse3,     1, ecx, 0)   \
@@ -66,6 +64,9 @@ _ (aes,      1, ecx, 25)  \
 _ (sha,      7, ebx, 29)  \
 _ (invariant_tsc, 0x80000007, edx, 8)
 
+#if defined(__x86_64__)
+#include "cpuid.h"
+
 static inline int
 clib_get_cpuid (const u32 lev, u32 * eax, u32 * ebx, u32 * ecx, u32 * edx)
 {
@@ -90,13 +91,18 @@ clib_cpu_supports_ ## flag()						\
 }
 foreach_x86_64_flags
 #undef _
+#else
+
+#define _(flag, func, reg, bit) \
+static inline int clib_cpu_supports_ ## flag() { return 0; }
+foreach_x86_64_flags
+#undef _
+#endif
 #endif
   format_function_t format_cpu_uarch;
 format_function_t format_cpu_model_name;
 format_function_t format_cpu_flags;
 
-#endif
-
 /*
  * fd.io coding-style-patch-verification: ON
  *
diff --git a/vppinfra/vppinfra/vector.h b/vppinfra/vppinfra/vector.h
index 58fc808abcc..491e7cfe547 100644
--- a/vppinfra/vppinfra/vector.h
+++ b/vppinfra/vppinfra/vector.h
@@ -42,7 +42,7 @@
 
 /* Vector types. */
 
-#if defined (__MMX__) || defined (__IWMMXT__)
+#if defined (__MMX__) || defined (__IWMMXT__) || defined (__aarch64__)
 #define CLIB_HAVE_VEC64
 #endif
 
@@ -63,6 +63,10 @@
 
 #if defined (__aarch64__) || defined (__arm__)
 typedef unsigned int u32x4 _vector_size (16);
+typedef u8 u8x16 _vector_size (16);
+typedef u16 u16x8 _vector_size (16);
+typedef u32 u32x4 _vector_size (16);
+typedef u64 u64x2 _vector_size (16);
 #endif
 
 #ifdef CLIB_HAVE_VEC64
@@ -245,6 +249,10 @@ _(i64, 2);
 #include <vppinfra/vector_iwmmxt.h>
 #endif
 
+#if defined (__aarch64__)
+#include <vppinfra/vector_neon.h>
+#endif
+
 #if (defined(CLIB_HAVE_VEC128) || defined(CLIB_HAVE_VEC64))
 #include <vppinfra/vector_funcs.h>
 #endif
diff --git a/vppinfra/vppinfra/vector_neon.h b/vppinfra/vppinfra/vector_neon.h
new file mode 100644
index 00000000000..cea5275949f
--- /dev/null
+++ b/vppinfra/vppinfra/vector_neon.h
@@ -0,0 +1,71 @@
+/*
+ * Copyright (c) 2015 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.
+ */
+
+#ifndef included_vector_neon_h
+#define included_vector_neon_h
+#include <arm_neon.h>
+
+/* Splats. */
+
+#define u8x16_splat(i) vdupq_n_u8(i)
+#define u16x8_splat(i) vdupq_n_u16(i)
+#define i16x8_splat(i) vdupq_n_s16(i)
+#define u32x4_splat(i) vdupq_n_u32(i)
+#define i32x4_splat(i) vdupq_n_s32(i)
+
+/* Arithmetic */
+#define u16x8_add(a,b) vaddq_u16(a,b)
+#define i16x8_add(a,b) vaddq_s16(a,b)
+#define u16x8_sub_saturate(a,b) vsubq_u16(a,b)
+#define i16x8_sub_saturate(a,b) vsubq_s16(a,b)
+
+#define u16x8_is_equal(a,b) vceqq_u16(a,b)
+#define i16x8_is_equal(a,b) vceqq_i16(a,b)
+
+always_inline u32
+u16x8_zero_byte_mask (u16x8 input)
+{
+  u8x16 vall_one = vdupq_n_u8 (0x0);
+  u8x16 res_values = { 0x01, 0x02, 0x04, 0x08,
+    0x10, 0x20, 0x40, 0x80,
+    0x01, 0x02, 0x04, 0x08,
+    0x10, 0x20, 0x40, 0x80
+  };
+
+  /* input --> [0x80, 0x40, 0x01, 0xf0, ... ] */
+  u8x16 test_result =
+    vreinterpretq_u8_u16 (vceqq_u16 (input, vreinterpretq_u16_u8 (vall_one)));
+  u8x16 before_merge = vminq_u8 (test_result, res_values);
+  /*before_merge--> [0x80, 0x00, 0x00, 0x10, ... ] */
+  /* u8x16 --> [a,b,c,d, e,f,g,h, i,j,k,l, m,n,o,p] */
+  /* pair add until we have 2 uint64_t  */
+  u16x8 merge1 = vpaddlq_u8 (before_merge);
+  /* u16x8-->  [a+b,c+d, e+f,g+h, i+j,k+l, m+n,o+p] */
+  u32x4 merge2 = vpaddlq_u16 (merge1);
+  /* u32x4-->  [a+b+c+d, e+f+g+h, i+j+k+l, m+n+o+p] */
+  u64x2 merge3 = vpaddlq_u32 (merge2);
+  /* u64x2-->  [a+b+c+d+e+f+g+h,  i+j+k+l+m+n+o+p]  */
+  return (u32) (vgetq_lane_u64 (merge3, 1) << 8) + vgetq_lane_u64 (merge3, 0);
+}
+
+#endif /* included_vector_neon_h */
+
+/*
+ * fd.io coding-style-patch-verification: ON
+ *
+ * Local Variables:
+ * eval: (c-set-style "gnu")
+ * End:
+ */