From 32462a072cb63fda5e84e9f372f7690a3eb691e4 Mon Sep 17 00:00:00 2001
From: Pascal Massimino <pascal.massimino@gmail.com>
Date: Tue, 23 Jun 2015 16:38:18 +0200
Subject: [PATCH] Speedup to HuffmanCostCombinedCount

~3% speedup for lossless encoding
Improves compression ratio by ~0.03%

Change-Id: Ic6d05fb0b1099b5ca56689b92b1c6515d54a5d6b
---
 src/dsp/lossless_enc.c        | 16 +++++-----------
 src/dsp/lossless_enc_mips32.c | 23 ++++++++---------------
 2 files changed, 13 insertions(+), 26 deletions(-)

diff --git a/src/dsp/lossless_enc.c b/src/dsp/lossless_enc.c
index d20ce1f8..54670fe5 100644
--- a/src/dsp/lossless_enc.c
+++ b/src/dsp/lossless_enc.c
@@ -1143,24 +1143,18 @@ static VP8LStreaks HuffmanCostCombinedCount(const uint32_t* X,
                                             const uint32_t* Y, int length) {
   int i;
   int streak = 0;
+  uint32_t xy_prev = 0xffffffff;
   VP8LStreaks stats;
   memset(&stats, 0, sizeof(stats));
-  for (i = 0; i < length - 1; ++i) {
-    const int xy = X[i] + Y[i];
-    const int xy_next = X[i + 1] + Y[i + 1];
+  for (i = 0; i < length; ++i) {
+    const uint32_t xy = X[i] + Y[i];
     ++streak;
-    if (xy == xy_next) {
-      continue;
-    }
+    if (xy != xy_prev) {
     stats.counts[xy != 0] += (streak > 3);
     stats.streaks[xy != 0][(streak > 3)] += streak;
     streak = 0;
+      xy_prev = xy;
   }
-  {
-    const int xy = X[i] + Y[i];
-    ++streak;
-    stats.counts[xy != 0] += (streak > 3);
-    stats.streaks[xy != 0][(streak > 3)] += streak;
   }
   return stats;
 }
diff --git a/src/dsp/lossless_enc_mips32.c b/src/dsp/lossless_enc_mips32.c
index 922d88e7..c4cdd908 100644
--- a/src/dsp/lossless_enc_mips32.c
+++ b/src/dsp/lossless_enc_mips32.c
@@ -245,29 +245,22 @@ static VP8LStreaks HuffmanCostCombinedCount(const uint32_t* X,
                                             const uint32_t* Y, int length) {
   int i;
   int streak = 0;
+  uint32_t xy_prev = 0xffffffff;
   VP8LStreaks stats;
   int* const pstreaks = &stats.streaks[0][0];
   int* const pcnts = &stats.counts[0];
   int temp0, temp1, temp2, temp3;
   memset(&stats, 0, sizeof(stats));
-  for (i = 0; i < length - 1; ++i) {
-    const uint32_t xy = X[i] + Y[i];
-    const uint32_t xy_next = X[i + 1] + Y[i + 1];
-    ++streak;
-    if (xy == xy_next) {
-      continue;
-    }
-    temp0 = (xy != 0);
-    HUFFMAN_COST_PASS
-    streak = 0;
-  }
-  {
+  for (i = 0; i < length; ++i) {
     const uint32_t xy = X[i] + Y[i];
     ++streak;
-    temp0 = (xy != 0);
-    HUFFMAN_COST_PASS
+    if (xy != xy_prev) {
+      temp0 = (xy != 0);
+      HUFFMAN_COST_PASS
+      streak = 0;
+      xy_prev = xy;
+  }
   }
-
   return stats;
 }