From 8935175266e343ac1d52106e2e790810b54f26c1 Mon Sep 17 00:00:00 2001
From: liqiang64 <liqiang64@huawei.com>
Date: Tue, 3 Dec 2019 03:22:00 +0000
Subject: [PATCH] zlib: Optimize CRC32

This patch uses the NEON instruction set to optimize the CRC32
algorithm.

On the ARM architecture, we can optimize the efficiency of
crc32 through the interface provided by the neon instruction
set.
Modify by Li Qiang.
---
 crc32.c | 50 ++++++++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 50 insertions(+)

diff --git a/crc32.c b/crc32.c
index f8357b0..5c53068 100644
--- a/crc32.c
+++ b/crc32.c
@@ -28,6 +28,9 @@
 #endif /* MAKECRCH */
 
 #include "zutil.h"      /* for Z_U4, Z_U8, z_crc_t, and FAR definitions */
+#ifdef __aarch64__
+#include "arm_acle.h"
+#endif
 
  /*
   A CRC of a message is computed on N braids of words in the message, where
@@ -553,6 +556,50 @@ const z_crc_t FAR * ZEXPORT get_crc_table(void) {
     return (const z_crc_t FAR *)crc_table;
 }

+#ifdef __aarch64__
+ulg crc32_neon(crc, buf, len)
+    unsigned long crc;
+    const unsigned char FAR *buf;
+    z_size_t len;
+{
+    register uint32_t crc_result = 0xFFFFFFFFU;
+    register const uint8_t  *buf1;
+    register const uint16_t *buf2;
+    register const uint32_t *buf4;
+    register const uint64_t *buf8;
+    int64_t length = (int64_t)len;
+    buf8 = (const  uint64_t *)(const void *)buf;
+
+    if (buf == NULL) {
+        crc_result = 0xffffffffL;
+    } else {
+        crc_result = crc^0xffffffffUL;
+
+        while((length -= sizeof(uint64_t)) >= 0) {
+            crc_result = __crc32d((crc_result), *buf8++);
+        }
+
+        buf4 = (const uint32_t *)(const void *)buf8;
+        if (length & sizeof(uint32_t)) {
+            crc_result = __crc32w((crc_result), *buf4++);
+        }
+
+        buf2 = (const uint16_t *)(const void *)buf4;
+        if(length & sizeof(uint16_t)) {
+            crc_result = __crc32h((crc_result), *buf2++);
+        }
+
+        buf1 = (const uint8_t *)(const void *)buf2;
+        if (length & sizeof(uint8_t)) {
+            crc_result = __crc32b((crc_result), *buf1);
+        }
+    }
+
+    return (crc_result ^ 0xffffffffL);
+}
+#endif
+
+
 /* =========================================================================
  * Use ARM machine instructions if available. This will compute the CRC about
  * ten times faster than the braided calculation. This code does not check for
@@ -581,6 +628,10 @@ unsigned long ZEXPORT crc32_z(unsigned long crc, const unsigned char FAR *buf,
     z_size_t last, last2, i;
     z_size_t num;
 
+    #ifdef __aarch64__
+    return crc32_neon(crc, buf, len);
+    #endif
+
     /* Return initial CRC, if requested. */
     if (buf == Z_NULL) return 0;
 
-- 
2.27.0