From: Danielle Rozenblit <drozenblit@fb.com>
Date: Mon, 17 Oct 2022 17:24:29 +0000 (-0700)
Subject: Rough draft speed optimization
X-Git-Tag: v1.5.4^2~66^2~9
X-Git-Url: http://git.ipfire.org/cgi-bin/gitweb.cgi?a=commitdiff_plain;h=a08fabd51a4191b9b1a2d580454ae6b7b805ee35;p=thirdparty%2Fzstd.git

Rough draft speed optimization
---

diff --git a/lib/common/huf.h b/lib/common/huf.h
index d79acd1d5..0db6fb369 100644
--- a/lib/common/huf.h
+++ b/lib/common/huf.h
@@ -173,7 +173,7 @@ size_t HUF_decompress4X2_DCtx_wksp(HUF_DTable* dctx, void* dst, size_t dstSize,
 /* ****************************************
  *  HUF detailed API
  * ****************************************/
-#define HUF_OPTIMAL_DEPTH_THRESHOLD ZSTD_btultra
+#define HUF_OPTIMAL_DEPTH_THRESHOLD 3
 typedef enum {
    HUF_depth_fast, /** Use heuristic to find the table depth**/
    HUF_depth_optimal /** Test possible table depths to find the one that produces the smallest header + encoded size**/
diff --git a/lib/compress/huf_compress.c b/lib/compress/huf_compress.c
index 54b58f191..d8bbaa6b4 100644
--- a/lib/compress/huf_compress.c
+++ b/lib/compress/huf_compress.c
@@ -1258,7 +1258,8 @@ unsigned HUF_minTableLog(size_t srcSize, unsigned symbolCardinality)
 
 unsigned HUF_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxSymbolValue, void* workSpace, size_t wkspSize, HUF_CElt* table, const unsigned* count, HUF_depth_mode depthMode)
 {
-    unsigned optLog = FSE_optimalTableLog_internal(maxTableLog, srcSize, maxSymbolValue, 1);
+    unsigned optLogGuess = FSE_optimalTableLog_internal(maxTableLog, srcSize, maxSymbolValue, 1);
+    unsigned optLog = optLogGuess;
 
     if (depthMode == HUF_depth_optimal) { /** Test valid depths and return optimal **/
         BYTE* dst = (BYTE*)workSpace + sizeof(HUF_WriteCTableWksp);
@@ -1267,10 +1268,11 @@ unsigned HUF_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxS
         unsigned huffLog;
         size_t maxBits, hSize, newSize;
         const unsigned symbolCardinality = HUF_cardinality(count, maxSymbolValue);
+        const unsigned minTableLog = HUF_minTableLog(srcSize, symbolCardinality);
 
         if (wkspSize < sizeof(HUF_buildCTable_wksp_tables)) return optLog;
 
-        for (huffLog = HUF_minTableLog(srcSize, symbolCardinality); huffLog <= maxTableLog; huffLog++) {
+        for (huffLog = optLogGuess; huffLog >= minTableLog; huffLog--) {
             maxBits = HUF_buildCTable_wksp(table, count,
                                             maxSymbolValue, huffLog,
                                             workSpace, wkspSize);
@@ -1282,7 +1284,29 @@ unsigned HUF_optimalTableLog(unsigned maxTableLog, size_t srcSize, unsigned maxS
 
             newSize = HUF_estimateCompressedSize(table, count, maxSymbolValue) + hSize;
 
-            if (newSize < optSize) {
+            if (newSize > optSize) {
+                break;
+            } else {
+                optSize = newSize;
+                optLog = huffLog;
+            }
+        }
+
+        for (huffLog = optLogGuess + 1; huffLog <= maxTableLog; huffLog++) {
+            maxBits = HUF_buildCTable_wksp(table, count,
+                                            maxSymbolValue, huffLog,
+                                            workSpace, wkspSize);
+            if (ERR_isError(maxBits)) continue;
+
+            hSize = HUF_writeCTable_wksp(dst, dstSize, table, maxSymbolValue, (U32)maxBits,
+                                              workSpace, wkspSize);
+            if (ERR_isError(hSize)) continue;
+
+            newSize = HUF_estimateCompressedSize(table, count, maxSymbolValue) + hSize;
+
+            if (newSize > optSize) {
+                break;
+            } else {
                 optSize = newSize;
                 optLog = huffLog;
             }