From: Nathan Moinvaziri <nathan@nathanm.com>
Date: Sat, 13 Dec 2025 01:50:15 +0000 (-0800)
Subject: Use different bit accumulator type for x86 compiler optimization
X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=HEAD;p=thirdparty%2Fzlib-ng.git

Use different bit accumulator type for x86 compiler optimization
---

diff --git a/infback.c b/infback.c
index dca46dab..b3b6d86d 100644
--- a/infback.c
+++ b/infback.c
@@ -154,7 +154,7 @@ int32_t Z_EXPORT PREFIX(inflateBack)(PREFIX3(stream) *strm, in_func in, void *in
     unsigned char *put;          /* next output */
     unsigned have, left;         /* available input and output */
     uint64_t hold;               /* bit buffer */
-    uint8_t bits;                /* bits in bit buffer */
+    bits_t bits;                 /* bits in bit buffer */
     unsigned copy;               /* number of stored or match bytes to copy */
     unsigned char *from;         /* where to copy match bytes from */
     code here;                   /* current decoding table entry */
diff --git a/inffast_tpl.h b/inffast_tpl.h
index 316bd3a3..4798d81e 100644
--- a/inffast_tpl.h
+++ b/inffast_tpl.h
@@ -101,7 +101,7 @@ void Z_INTERNAL INFLATE_FAST(PREFIX3(stream) *strm, uint32_t start) {
        with (1<<bits)-1 to drop those excess bits so that, on function exit, we
        keep the invariant that (state->hold >> state->bits) == 0.
     */
-    uint8_t bits;               /* local strm->bits */
+    bits_t bits;                /* local strm->bits */
     uint64_t hold;              /* local strm->hold */
     unsigned lmask;             /* mask for first level of length codes */
     unsigned dmask;             /* mask for first level of distance codes */
@@ -128,7 +128,7 @@ void Z_INTERNAL INFLATE_FAST(PREFIX3(stream) *strm, uint32_t start) {
     wnext = state->wnext;
     window = state->window;
     hold = state->hold;
-    bits = (uint8_t)state->bits;
+    bits = (bits_t)state->bits;
     lcode = state->lencode;
     dcode = state->distcode;
     lmask = (1U << state->lenbits) - 1;
@@ -305,7 +305,7 @@ void Z_INTERNAL INFLATE_FAST(PREFIX3(stream) *strm, uint32_t start) {
     /* return unused bytes (on entry, bits < 8, so in won't go too far back) */
     len = bits >> 3;
     in -= len;
-    bits -= (uint8_t)(len << 3);
+    bits -= (bits_t)(len << 3);
     hold &= (UINT64_C(1) << bits) - 1;
 
     /* update state and return */
diff --git a/inflate.c b/inflate.c
index edba74cb..4cac4817 100644
--- a/inflate.c
+++ b/inflate.c
@@ -479,7 +479,7 @@ int32_t Z_EXPORT PREFIX(inflate)(PREFIX3(stream) *strm, int32_t flush) {
     unsigned char *from;        /* where to copy match bytes from */
     unsigned have, left;        /* available input and output */
     uint64_t hold;              /* bit buffer */
-    unsigned bits;              /* bits in bit buffer */
+    bits_t bits;                /* bits in bit buffer */
     uint32_t in, out;           /* save starting available input and output */
     unsigned copy;              /* number of stored or match bytes to copy */
     code here;                  /* current decoding table entry */
diff --git a/inflate_p.h b/inflate_p.h
index 425709ae..fad4b53f 100644
--- a/inflate_p.h
+++ b/inflate_p.h
@@ -72,6 +72,13 @@
     } while (0)
 #endif
 
+/* Compiler optimization for bit accumulator on x86 architectures */
+#if defined(__x86_64__) || defined(__i386__)
+typedef uint8_t bits_t;
+#else
+typedef unsigned bits_t;
+#endif
+
 /* Load registers with state in inflate() for speed */
 #define LOAD() \
     do { \
@@ -80,7 +87,7 @@
         next = strm->next_in; \
         have = strm->avail_in; \
         hold = state->hold; \
-        bits = (uint8_t)state->bits; \
+        bits = (bits_t)state->bits; \
     } while (0)
 
 /* Restore state from registers in inflate() */
@@ -105,7 +112,7 @@
    not enough available input to do that, then return from inflate()/inflateBack(). */
 #define NEEDBITS(n) \
     do { \
-        while (bits < (uint8_t)(n)) \
+        while (bits < (bits_t)(n)) \
             PULLBYTE(); \
     } while (0)
 
@@ -117,7 +124,7 @@
 #define DROPBITS(n) \
     do { \
         hold >>= (n); \
-        bits -= (uint8_t)(n); \
+        bits -= (bits_t)(n); \
     } while (0)
 
 /* Remove zero to seven bits as needed to go to a byte boundary */