From: Yann Collet <cyan@fb.com>
Date: Wed, 30 Nov 2016 01:42:46 +0000 (-0800)
Subject: minor decompression speed improvement
X-Git-Tag: v1.1.2~60
X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=ff504de39127f1d738bafb3de9d033193470de3a;p=thirdparty%2Fzstd.git

minor decompression speed improvement
---

diff --git a/lib/decompress/zstd_decompress.c b/lib/decompress/zstd_decompress.c
index 3e9be3f6a..3106fa014 100644
--- a/lib/decompress/zstd_decompress.c
+++ b/lib/decompress/zstd_decompress.c
@@ -873,17 +873,17 @@ static seq_t ZSTD_decodeSequence(seqState_t* seqState)
                             0x1003, 0x2003, 0x4003, 0x8003, 0x10003 };
 
     static const U32 OF_base[MaxOff+1] = {
-                 0,        1,       1,       5,     0xD,     0x1D,     0x3D,     0x7D,
-                 0xFD,   0x1FD,   0x3FD,   0x7FD,   0xFFD,   0x1FFD,   0x3FFD,   0x7FFD,
-                 0xFFFD, 0x1FFFD, 0x3FFFD, 0x7FFFD, 0xFFFFD, 0x1FFFFD, 0x3FFFFD, 0x7FFFFD,
-                 0xFFFFFD, 0x1FFFFFD, 0x3FFFFFD, 0x7FFFFFD, 0xFFFFFFD };
+                             0,        1,       1,       5,     0xD,     0x1D,     0x3D,     0x7D,
+                             0xFD,   0x1FD,   0x3FD,   0x7FD,   0xFFD,   0x1FFD,   0x3FFD,   0x7FFD,
+                             0xFFFD, 0x1FFFD, 0x3FFFD, 0x7FFFD, 0xFFFFD, 0x1FFFFD, 0x3FFFFD, 0x7FFFFD,
+                             0xFFFFFD, 0x1FFFFFD, 0x3FFFFFD, 0x7FFFFFD, 0xFFFFFFD };
 
     /* sequence */
     {   size_t offset;
         if (!ofCode)
             offset = 0;
         else {
-            offset = OF_base[ofCode] + BIT_readBits(&seqState->DStream, ofBits);   /* <=  (ZSTD_WINDOWLOG_MAX-1) bits */
+            offset = OF_base[ofCode] + BIT_readBitsFast(&seqState->DStream, ofBits);   /* <=  (ZSTD_WINDOWLOG_MAX-1) bits */
             if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);
         }
 
@@ -906,10 +906,10 @@ static seq_t ZSTD_decodeSequence(seqState_t* seqState)
         seq.offset = offset;
     }
 
-    seq.matchLength = ML_base[mlCode] + ((mlCode>31) ? BIT_readBits(&seqState->DStream, mlBits) : 0);  /* <=  16 bits */
+    seq.matchLength = ML_base[mlCode] + ((mlCode>31) ? BIT_readBitsFast(&seqState->DStream, mlBits) : 0);  /* <=  16 bits */
     if (MEM_32bits() && (mlBits+llBits>24)) BIT_reloadDStream(&seqState->DStream);
 
-    seq.litLength = LL_base[llCode] + ((llCode>15) ? BIT_readBits(&seqState->DStream, llBits) : 0);    /* <=  16 bits */
+    seq.litLength = LL_base[llCode] + ((llCode>15) ? BIT_readBitsFast(&seqState->DStream, llBits) : 0);    /* <=  16 bits */
     if (MEM_32bits() ||
        (totalBits > 64 - 7 - (LLFSELog+MLFSELog+OffFSELog)) ) BIT_reloadDStream(&seqState->DStream);
 
@@ -1086,17 +1086,17 @@ static seq_t ZSTD_decodeSequenceLong(seqState_t* seqState)
                             0x1003, 0x2003, 0x4003, 0x8003, 0x10003 };
 
     static const U32 OF_base[MaxOff+1] = {
-                 0,        1,       1,       5,     0xD,     0x1D,     0x3D,     0x7D,
-                 0xFD,   0x1FD,   0x3FD,   0x7FD,   0xFFD,   0x1FFD,   0x3FFD,   0x7FFD,
-                 0xFFFD, 0x1FFFD, 0x3FFFD, 0x7FFFD, 0xFFFFD, 0x1FFFFD, 0x3FFFFD, 0x7FFFFD,
-                 0xFFFFFD, 0x1FFFFFD, 0x3FFFFFD, 0x7FFFFFD, 0xFFFFFFD };
+                             0,        1,       1,       5,     0xD,     0x1D,     0x3D,     0x7D,
+                             0xFD,   0x1FD,   0x3FD,   0x7FD,   0xFFD,   0x1FFD,   0x3FFD,   0x7FFD,
+                             0xFFFD, 0x1FFFD, 0x3FFFD, 0x7FFFD, 0xFFFFD, 0x1FFFFD, 0x3FFFFD, 0x7FFFFD,
+                             0xFFFFFD, 0x1FFFFFD, 0x3FFFFFD, 0x7FFFFFD, 0xFFFFFFD };
 
     /* sequence */
     {   size_t offset;
         if (!ofCode)
             offset = 0;
         else {
-            offset = OF_base[ofCode] + BIT_readBits(&seqState->DStream, ofBits);   /* <=  (ZSTD_WINDOWLOG_MAX-1) bits */
+            offset = OF_base[ofCode] + BIT_readBitsFast(&seqState->DStream, ofBits);   /* <=  (ZSTD_WINDOWLOG_MAX-1) bits */
             if (MEM_32bits()) BIT_reloadDStream(&seqState->DStream);
         }
 
@@ -1119,10 +1119,10 @@ static seq_t ZSTD_decodeSequenceLong(seqState_t* seqState)
         seq.offset = offset;
     }
 
-    seq.matchLength = ML_base[mlCode] + ((mlCode>31) ? BIT_readBits(&seqState->DStream, mlBits) : 0);  /* <=  16 bits */
+    seq.matchLength = ML_base[mlCode] + ((mlCode>31) ? BIT_readBitsFast(&seqState->DStream, mlBits) : 0);  /* <=  16 bits */
     if (MEM_32bits() && (mlBits+llBits>24)) BIT_reloadDStream(&seqState->DStream);
 
-    seq.litLength = LL_base[llCode] + ((llCode>15) ? BIT_readBits(&seqState->DStream, llBits) : 0);    /* <=  16 bits */
+    seq.litLength = LL_base[llCode] + ((llCode>15) ? BIT_readBitsFast(&seqState->DStream, llBits) : 0);    /* <=  16 bits */
     if (MEM_32bits() ||
        (totalBits > 64 - 7 - (LLFSELog+MLFSELog+OffFSELog)) ) BIT_reloadDStream(&seqState->DStream);