Remove calls to crc_u32 and add clib_crc32c for armv8+crc

crc_u32 was not defined for non x86_64 with SSE4.2 processors.

Calls to "crc_u32" are removed and replaced by either a call to
clib_crc32c or a call to clib_xxhash, as the result is not used
as a check value but as a hash.

Change-Id: I3af4d68e2e5ebd0c9b0a6090f848d043cb0f20a2
Signed-off-by: Christophe Fontaine <christophe.fontaine@enea.com>

6 files changed, 61 insertions, 92 deletions

diff --git a/src/plugins/flowprobe/node.c b/src/plugins/flowprobe/node.c
index 6a539db9..80bfa9b5 100644
--- a/src/plugins/flowprobe/node.c
+++ b/src/plugins/flowprobe/node.c
@@ -17,6 +17,7 @@
 #include <vlib/vlib.h>
 #include <vnet/vnet.h>
 #include <vnet/pg/pg.h>
+#include <vppinfra/crc32.h>
 #include <vppinfra/error.h>
 #include <flowprobe/flowprobe.h>
 #include <vnet/ip/ip6_packet.h>
@@ -258,10 +259,16 @@ static inline u32
 flowprobe_hash (flowprobe_key_t * k)
 {
   flowprobe_main_t *fm = &flowprobe_main;
-  int i;
   u32 h = 0;
-  for (i = 0; i < sizeof (k->as_u32) / sizeof (u32); i++)
-    h = crc_u32 (k->as_u32[i], h);
+
+#ifdef clib_crc32c_uses_intrinsics
+  h = clib_crc32c ((u8 *) k->as_u32, FLOWPROBE_KEY_IN_U32);
+#else
+  u64 tmp =
+    k->as_u32[0] ^ k->as_u32[1] ^ k->as_u32[2] ^ k->as_u32[3] ^ k->as_u32[4];
+  h = clib_xxhash (tmp);
+#endif
+
   return h >> (32 - fm->ht_log2len);
 }
 
diff --git a/src/vnet/map/map.c b/src/vnet/map/map.c
index 8579cdf6..d246f6c8 100644
--- a/src/vnet/map/map.c
+++ b/src/vnet/map/map.c
@@ -19,29 +19,10 @@
 #include <vnet/fib/ip6_fib.h>
 #include <vnet/adj/adj.h>
 #include <vnet/map/map_dpo.h>
+#include <vppinfra/crc32.h>
 
 #include "map.h"
 
-#ifdef __SSE4_2__
-static inline u32
-crc_u32 (u32 data, u32 value)
-{
-  __asm__ volatile ("crc32l %[data], %[value];":[value] "+r" (value):[data]
-		    "rm" (data));
-  return value;
-}
-#else
-#include <vppinfra/xxhash.h>
-
-static inline u32
-crc_u32 (u32 data, u32 value)
-{
-  u64 tmp = ((u64) data << 32) | (u64) value;
-  return (u32) clib_xxhash (tmp);
-}
-#endif
-
-
 /*
  * This code supports the following MAP modes:
  *
@@ -1488,10 +1469,12 @@ map_ip4_reass_get (u32 src, u32 dst, u16 fragment_id,
   };
 
   u32 h = 0;
-  h = crc_u32 (k.as_u32[0], h);
-  h = crc_u32 (k.as_u32[1], h);
-  h = crc_u32 (k.as_u32[2], h);
-  h = crc_u32 (k.as_u32[3], h);
+#ifdef clib_crc32c_uses_intrinsics
+  h = clib_crc32c ((u8 *) k.as_u32, 16);
+#else
+  u64 tmp = k.as_u32[0] ^ k.as_u32[1] ^ k.as_u32[2] ^ k.as_u32[3];
+  h = clib_xxhash (tmp);
+#endif
   h = h >> (32 - mm->ip4_reass_ht_log2len);
 
   f64 now = vlib_time_now (mm->vlib_main);
@@ -1660,8 +1643,15 @@ map_ip6_reass_get (ip6_address_t * src, ip6_address_t * dst, u32 fragment_id,
 
   u32 h = 0;
   int i;
-  for (i = 0; i < 10; i++)
-    h = crc_u32 (k.as_u32[i], h);
+
+#ifdef clib_crc32c_uses_intrinsics
+  h = clib_crc32c ((u8 *) k.as_u32, 40);
+#else
+  u64 tmp =
+    k.as_u64[0] ^ k.as_u64[1] ^ k.as_u64[2] ^ k.as_u64[3] ^ k.as_u64[4];
+  h = clib_xxhash (tmp);
+#endif
+
   h = h >> (32 - mm->ip6_reass_ht_log2len);
 
   f64 now = vlib_time_now (mm->vlib_main);
diff --git a/src/vppinfra/bihash_16_8.h b/src/vppinfra/bihash_16_8.h
index ce80f70e..6b1b563e 100644
--- a/src/vppinfra/bihash_16_8.h
+++ b/src/vppinfra/bihash_16_8.h
@@ -24,6 +24,7 @@
 #include <vppinfra/format.h>
 #include <vppinfra/pool.h>
 #include <vppinfra/xxhash.h>
+#include <vppinfra/crc32.h>
 
 typedef struct
 {
@@ -40,39 +41,16 @@ clib_bihash_is_free_16_8 (clib_bihash_kv_16_8_t * v)
   return 0;
 }
 
-#if __SSE4_2__
-#ifndef __defined_crc_u32__
-#define __defined_crc_u32__
-static inline u32
-crc_u32 (u32 data, u32 value)
-{
-  __asm__ volatile ("crc32l %[data], %[value];":[value] "+r" (value):[data]
-		    "rm" (data));
-  return value;
-}
-#endif /* __defined_crc_u32__ */
-
 static inline u64
 clib_bihash_hash_16_8 (clib_bihash_kv_16_8_t * v)
 {
-  u32 *dp = (u32 *) & v->key[0];
-  u32 value = 0;
-
-  value = crc_u32 (dp[0], value);
-  value = crc_u32 (dp[1], value);
-  value = crc_u32 (dp[2], value);
-  value = crc_u32 (dp[3], value);
-
-  return value;
-}
+#ifdef clib_crc32c_uses_intrinsics
+  return clib_crc32c ((u8 *) v->key, 16);
 #else
-static inline u64
-clib_bihash_hash_16_8 (clib_bihash_kv_16_8_t * v)
-{
   u64 tmp = v->key[0] ^ v->key[1];
   return clib_xxhash (tmp);
-}
 #endif
+}
 
 static inline u8 *
 format_bihash_kvp_16_8 (u8 * s, va_list * args)
diff --git a/src/vppinfra/bihash_24_8.h b/src/vppinfra/bihash_24_8.h
index 655dab80..db77daa4 100644
--- a/src/vppinfra/bihash_24_8.h
+++ b/src/vppinfra/bihash_24_8.h
@@ -20,11 +20,11 @@
 #ifndef __included_bihash_24_8_h__
 #define __included_bihash_24_8_h__
 
+#include <vppinfra/crc32.h>
 #include <vppinfra/heap.h>
 #include <vppinfra/format.h>
 #include <vppinfra/pool.h>
 #include <vppinfra/xxhash.h>
-#include <vppinfra/crc32.h>
 
 typedef struct
 {
@@ -44,7 +44,7 @@ clib_bihash_is_free_24_8 (const clib_bihash_kv_24_8_t * v)
 static inline u64
 clib_bihash_hash_24_8 (const clib_bihash_kv_24_8_t * v)
 {
-#if __SSE4_2__
+#ifdef clib_crc32c_uses_intrinsics
   return clib_crc32c ((u8 *) v->key, 24);
 #else
   u64 tmp = v->key[0] ^ v->key[1] ^ v->key[2];
diff --git a/src/vppinfra/bihash_48_8.h b/src/vppinfra/bihash_48_8.h
index 1a6e7691..48079e0a 100644
--- a/src/vppinfra/bihash_48_8.h
+++ b/src/vppinfra/bihash_48_8.h
@@ -21,6 +21,7 @@
 #ifndef __included_bihash_48_8_h__
 #define __included_bihash_48_8_h__
 
+#include <vppinfra/crc32.h>
 #include <vppinfra/heap.h>
 #include <vppinfra/format.h>
 #include <vppinfra/pool.h>
@@ -41,48 +42,17 @@ clib_bihash_is_free_48_8 (const clib_bihash_kv_48_8_t * v)
   return 0;
 }
 
-#if __SSE4_2__
-#ifndef __defined_crc_u32__
-#define __defined_crc_u32__
-static inline u32
-crc_u32 (u32 data, u32 value)
-{
-  __asm__ volatile ("crc32l %[data], %[value];":[value] "+r" (value):[data]
-		    "rm" (data));
-  return value;
-}
-#endif /* __defined_crc_u32__ */
-
 static inline u64
 clib_bihash_hash_48_8 (const clib_bihash_kv_48_8_t * v)
 {
-  const u32 *dp = (const u32 *) &v->key[0];
-  u32 value = 0;
-
-  value = crc_u32 (dp[0], value);
-  value = crc_u32 (dp[1], value);
-  value = crc_u32 (dp[2], value);
-  value = crc_u32 (dp[3], value);
-  value = crc_u32 (dp[4], value);
-  value = crc_u32 (dp[5], value);
-  value = crc_u32 (dp[6], value);
-  value = crc_u32 (dp[7], value);
-  value = crc_u32 (dp[8], value);
-  value = crc_u32 (dp[9], value);
-  value = crc_u32 (dp[10], value);
-  value = crc_u32 (dp[11], value);
-
-  return value;
-}
+#ifdef clib_crc32c_uses_intrinsics
+  return clib_crc32c ((u8 *) v->key, 48);
 #else
-static inline u64
-clib_bihash_hash_48_8 (const clib_bihash_kv_48_8_t * v)
-{
   u64 tmp = v->key[0] ^ v->key[1] ^ v->key[2] ^ v->key[3] ^ v->key[4]
     ^ v->key[5];
   return clib_xxhash (tmp);
-}
 #endif
+}
 
 static inline u8 *
 format_bihash_kvp_48_8 (u8 * s, va_list * args)
diff --git a/src/vppinfra/crc32.h b/src/vppinfra/crc32.h
index abb2953f..5a47236a 100644
--- a/src/vppinfra/crc32.h
+++ b/src/vppinfra/crc32.h
@@ -17,6 +17,7 @@
 #define __included_crc32_h__
 
 #if __SSE4_2__
+#define clib_crc32c_uses_intrinsics
 #include <x86intrin.h>
 
 static_always_inline u32
@@ -45,8 +46,31 @@ clib_crc32c (u8 * s, int len)
   return v;
 }
 
-#endif
+#elif __ARM_FEATURE_CRC32
+#define clib_crc32c_with_intrinsics
+#include <arm_acle.h>
+
+static_always_inline u32
+clib_crc32c (u8 * s, int len)
+{
+  u32 v = 0;
+
+  for (; len >= 8; len -= 8, s += 8)
+    v = __crc32cd (v, *((u64 *) s));
+
+  for (; len >= 4; len -= 4, s += 4)
+    v = __crc32cw (v, *((u32 *) s));
+
+  for (; len >= 2; len -= 2, s += 2)
+    v = __crc32ch (v, *((u16 *) s));
+
+  for (; len >= 1; len -= 1, s += 1)
+    v = __crc32cb (v, *((u8 *) s));
 
+  return v;
+}
+
+#endif
 #endif /* __included_crc32_h__ */
 
 /*