more generic skip formula

improving speed

2 files changed, 5 insertions, 14 deletions

diff --git a/lib/lz4hc.c b/lib/lz4hc.c
index 941cda0..2d8671d 100644
--- a/lib/lz4hc.c
+++ b/lib/lz4hc.c
@@ -369,7 +369,7 @@ LZ4_FORCE_INLINE int LZ4HC_encodeSequence (
     U32 const mlAdd = (matchLength>=19) ? ((matchLength-19) / 255) + 1 : 0;
     U32 const cost = 1 + llAdd + ll + 2 + mlAdd;
     if (start==NULL) start = *anchor;  /* only works for single segment */
-    //g_debuglog_enable = (pos >= 112705) & (pos <= 112760);
+    //g_debuglog_enable = (pos >= 2228) & (pos <= 2262);
     DEBUGLOG(2, "pos:%7u -- literals:%3u, match:%4i, offset:%5u, cost:%3u + %u",
                 pos,
                 (U32)(*ip - *anchor), matchLength, (U32)(*ip-match),
diff --git a/lib/lz4opt.h b/lib/lz4opt.h
index c9fab04..0463cb1 100644
--- a/lib/lz4opt.h
+++ b/lib/lz4opt.h
@@ -46,7 +46,6 @@ typedef struct {
     int off;
     int mlen;
     int litlen;
-    int toSearch;
 } LZ4HC_optimal_t;
 
 
@@ -244,6 +243,7 @@ static int LZ4HC_compress_optimal (
     const BYTE* const matchlimit = iend - LASTLITERALS;
     BYTE* op = (BYTE*) dst;
     BYTE* const oend = op + dstCapacity;
+    int const front = fullUpdate ? 2 : 1;
 
     /* init */
     DEBUGLOG(5, "LZ4HC_compress_optimal");
@@ -279,7 +279,6 @@ static int LZ4HC_compress_optimal (
                 opt[rPos].off = 0;
                 opt[rPos].litlen = llen + rPos;
                 opt[rPos].price = cost;
-                opt[rPos].toSearch = 1;
                 DEBUGLOG(7, "rPos:%3i => price:%3i (litlen=%i) -- initial setup",
                             rPos, cost, opt[rPos].litlen);
         }   }
@@ -296,21 +295,16 @@ static int LZ4HC_compress_optimal (
                     opt[mlen].off = offset;
                     opt[mlen].litlen = llen;
                     opt[mlen].price = cost;
-                    opt[mlen].toSearch = (((mlen - 18) % 255) == 0);
                     DEBUGLOG(7, "rPos:%3i => price:%3i (matchlen=%i) -- initial setup",
                                 mlen, cost, mlen);
         }   }   }
         last_match_pos = matches[nb_matches_initial-1].len;
-        if (fullUpdate) opt[last_match_pos-2].toSearch = 1;   /* 1 byte on calgary */
-        if (fullUpdate) opt[last_match_pos-1].toSearch = 1;   /* 1 byte on calgary */
-        opt[last_match_pos].toSearch = 1;
         {   int addLit;
             for (addLit = 1; addLit <= 3; addLit ++) {
                 opt[last_match_pos+addLit].mlen = 1; /* literal */
                 opt[last_match_pos+addLit].off = 0;
                 opt[last_match_pos+addLit].litlen = addLit;
                 opt[last_match_pos+addLit].price = opt[last_match_pos].price + LZ4HC_literalsPrice(addLit);
-                opt[last_match_pos+addLit].toSearch = 1;
                 DEBUGLOG(7, "rPos:%3i => price:%3i (litlen=%i) -- initial setup",
                             last_match_pos+addLit, opt[last_match_pos+addLit].price, addLit);
         }   }
@@ -321,7 +315,9 @@ static int LZ4HC_compress_optimal (
             int nb_matches;
 
             if (curPtr >= mflimit) break;
-            if (opt[cur].toSearch == 0) continue;
+            DEBUGLOG(7, "rPos:%u[%u] vs [%u]%u",
+                    cur, opt[cur].price, opt[cur+1].price, cur+1);
+            if (opt[cur+front].price <= opt[cur].price) continue;
 
             DEBUGLOG(7, "search at rPos:%u", cur);
             //nb_matches = LZ4HC_BinTree_GetAllMatches(ctx, curPtr, matchlimit, MINMATCH-1, matches, fullUpdate);
@@ -388,19 +384,14 @@ static int LZ4HC_compress_optimal (
                             opt[pos].off = offset;
                             opt[pos].litlen = ll;
                             opt[pos].price = price;
-                            opt[pos].toSearch = (((ml-18) % 255) == 0);
             }   }   }   }
             /* complete following positions with literals */
-            if (fullUpdate) opt[last_match_pos-2].toSearch = 1;   /* 2 bytes on enwik7 */
-            if (fullUpdate) opt[last_match_pos-1].toSearch = 1;   /* 53 bytes on enwik7, 13 bytes on calgary */
-            opt[last_match_pos].toSearch = 1;
             {   int addLit;
                 for (addLit = 1; addLit <= 3; addLit ++) {
                     opt[last_match_pos+addLit].mlen = 1; /* literal */
                     opt[last_match_pos+addLit].off = 0;
                     opt[last_match_pos+addLit].litlen = addLit;
                     opt[last_match_pos+addLit].price = opt[last_match_pos].price + LZ4HC_literalsPrice(addLit);
-                    opt[last_match_pos+addLit].toSearch = 1;
                     DEBUGLOG(7, "rPos:%3i => price:%3i (litlen=%i)", last_match_pos+addLit, opt[last_match_pos+addLit].price, addLit);
             }   }
         }  /* for (cur = 1; cur <= last_match_pos; cur++) */