From: Nathan Moinvaziri <nathan@nathanm.com>
Date: Sun, 16 Aug 2020 03:06:26 +0000 (-0700)
Subject: Remove NO_DIVIDE from adler32.
X-Git-Tag: 1.9.9-b1~105
X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=193d8fd7dfb7927facab7a3034daa27ad5b9df1c;p=thirdparty%2Fzlib-ng.git

Remove NO_DIVIDE from adler32.
---

diff --git a/adler32.c b/adler32.c
index 247c45be..989d7b4e 100644
--- a/adler32.c
+++ b/adler32.c
@@ -46,8 +46,8 @@ ZLIB_INTERNAL uint32_t adler32_c(uint32_t adler, const unsigned char *buf, size_
             buf += 8;
 #endif
         } while (--n);
-        MOD(adler);
-        MOD(sum2);
+        adler %= BASE;
+        sum2 %= BASE;
     }
 
     /* do remaining bytes (less than NMAX, still just one modulo) */
@@ -69,8 +69,8 @@ ZLIB_INTERNAL uint32_t adler32_c(uint32_t adler, const unsigned char *buf, size_
             adler += *buf++;
             sum2 += adler;
         }
-        MOD(adler);
-        MOD(sum2);
+        adler %= BASE;
+        sum2 %= BASE;
     }
 
     /* return recombined sums */
@@ -109,11 +109,11 @@ static uint32_t adler32_combine_(uint32_t adler1, uint32_t adler2, z_off64_t len
         return 0xffffffff;
 
     /* the derivation of this formula is left as an exercise for the reader */
-    MOD63(len2);                /* assumes len2 >= 0 */
+    len2 %= BASE;                 /* assumes len2 >= 0 */
     rem = (unsigned)len2;
     sum1 = adler1 & 0xffff;
     sum2 = rem * sum1;
-    MOD(sum2);
+    sum2 %= BASE;
     sum1 += (adler2 & 0xffff) + BASE - 1;
     sum2 += ((adler1 >> 16) & 0xffff) + ((adler2 >> 16) & 0xffff) + BASE - rem;
     if (sum1 >= BASE) sum1 -= BASE;
diff --git a/adler32_p.h b/adler32_p.h
index c1967320..7f75c71e 100644
--- a/adler32_p.h
+++ b/adler32_p.h
@@ -18,46 +18,6 @@
 #define DO8(sum1, sum2, buf, i)  {DO4(sum1, sum2, buf, i); DO4(sum1, sum2, buf, i+4);}
 #define DO16(sum1, sum2, buf)    {DO8(sum1, sum2, buf, 0); DO8(sum1, sum2, buf, 8);}
 
-/* use NO_DIVIDE if your processor does not do division in hardware --
-   try it both ways to see which is faster */
-#ifdef NO_DIVIDE
-/* note that this assumes BASE is 65521, where 65536 % 65521 == 15
-   (thank you to John Reiser for pointing this out) */
-#  define CHOP(a) \
-    do { \
-        uint32_t tmp = a >> 16; \
-        a &= 0xffff; \
-        a += (tmp << 4) - tmp; \
-    } while (0)
-#  define MOD28(a) \
-    do { \
-        CHOP(a); \
-        if (a >= BASE) a -= BASE; \
-    } while (0)
-#  define MOD(a) \
-    do { \
-        CHOP(a); \
-        MOD28(a); \
-    } while (0)
-#  define MOD63(a) \
-    do { /* this assumes a is not negative */ \
-        z_off64_t tmp = a >> 32; \
-        a &= 0xffffffffL; \
-        a += (tmp << 8) - (tmp << 5) + tmp; \
-        tmp = a >> 16; \
-        a &= 0xffffL; \
-        a += (tmp << 4) - tmp; \
-        tmp = a >> 16; \
-        a &= 0xffffL; \
-        a += (tmp << 4) - tmp; \
-        if (a >= BASE) a -= BASE; \
-    } while (0)
-#else
-#  define MOD(a) a %= BASE
-#  define MOD28(a) a %= BASE
-#  define MOD63(a) a %= BASE
-#endif
-
 static inline uint32_t adler32_len_1(uint32_t adler, const unsigned char *buf, uint32_t sum2) {
     adler += buf[0];
     if (adler >= BASE)
@@ -76,7 +36,7 @@ static inline uint32_t adler32_len_16(uint32_t adler, const unsigned char *buf,
     }
     if (adler >= BASE)
         adler -= BASE;
-    MOD28(sum2);            /* only added so many BASE's */
+    sum2 %= BASE;            /* only added so many BASE's */
     return adler | (sum2 << 16);
 }
 
diff --git a/arch/x86/adler32_avx.c b/arch/x86/adler32_avx.c
index cd5823f8..60d3da4b 100644
--- a/arch/x86/adler32_avx.c
+++ b/arch/x86/adler32_avx.c
@@ -93,12 +93,12 @@ ZLIB_INTERNAL uint32_t adler32_avx2(uint32_t adler, const unsigned char *buf, si
 
        adler = (s1_unpack[0] % BASE) + (s1_unpack[1] % BASE) + (s1_unpack[2] % BASE) + (s1_unpack[3] % BASE) +
                (s1_unpack[4] % BASE) + (s1_unpack[5] % BASE) + (s1_unpack[6] % BASE) + (s1_unpack[7] % BASE);
-       MOD(adler);
+       adler %= BASE;
        s1[7] = adler;
 
        sum2 = (s2_unpack[0] % BASE) + (s2_unpack[1] % BASE) + (s2_unpack[2] % BASE) + (s2_unpack[3] % BASE) +
               (s2_unpack[4] % BASE) + (s2_unpack[5] % BASE) + (s2_unpack[6] % BASE) + (s2_unpack[7] % BASE);
-       MOD(sum2);
+       sum2 %= BASE;
        s2[7] = sum2;
     }
 
@@ -107,8 +107,8 @@ ZLIB_INTERNAL uint32_t adler32_avx2(uint32_t adler, const unsigned char *buf, si
         adler += *buf++;
         sum2 += adler;
     }
-    MOD(adler);
-    MOD(sum2);
+    adler %= BASE;
+    sum2 %= BASE;
 
     /* return recombined sums */
     return adler | (sum2 << 16);
diff --git a/arch/x86/adler32_ssse3.c b/arch/x86/adler32_ssse3.c
index 931ea90c..c1a48e98 100644
--- a/arch/x86/adler32_ssse3.c
+++ b/arch/x86/adler32_ssse3.c
@@ -95,11 +95,11 @@ ZLIB_INTERNAL uint32_t adler32_ssse3(uint32_t adler, const unsigned char *buf, s
        _mm_store_si128((__m128i*)s2_unpack, vs2);
 
        adler = (s1_unpack[0] % BASE) + (s1_unpack[1] % BASE) + (s1_unpack[2] % BASE) + (s1_unpack[3] % BASE);
-       MOD(adler);
+       adler %= BASE;
        s1[3] = adler;
 
        sum2 = (s2_unpack[0] % BASE) + (s2_unpack[1] % BASE) + (s2_unpack[2] % BASE) + (s2_unpack[3] % BASE);
-       MOD(sum2);
+       sum2 %= BASE;
        s2[3] = sum2;
     }
 
@@ -108,8 +108,8 @@ ZLIB_INTERNAL uint32_t adler32_ssse3(uint32_t adler, const unsigned char *buf, s
         adler += *buf++;
         sum2 += adler;
     }
-    MOD(adler);
-    MOD(sum2);
+    adler %= BASE;
+    sum2 %= BASE;
 
     /* return recombined sums */
     return adler | (sum2 << 16);