[x265] [PATCH] api: simplistic auto-determination of frame thread count

[Steve Borho]

# HG changeset patch
# User Steve Borho <steve at borho.org>
# Date 1383795498 21600
#      Wed Nov 06 21:38:18 2013 -0600
# Node ID 90d9c1067f50c750e5e0b23952ac0edd2595e575
# Parent  93cccbe49a93dd4c054ef06aca76974948793613
api: simplistic auto-determination of frame thread count

diff -r 93cccbe49a93 -r 90d9c1067f50 source/common/common.cpp
--- a/source/common/common.cpp	Wed Nov 06 19:49:38 2013 -0600
+++ b/source/common/common.cpp	Wed Nov 06 21:38:18 2013 -0600
@@ -141,7 +141,7 @@
     /* Applying non-zero default values to all elements in the param structure */
     param->logLevel = X265_LOG_INFO;
     param->bEnableWavefront = 1;
-    param->frameNumThreads = 1;
+    param->frameNumThreads = 0;
     param->inputBitDepth = 8;
     param->sourceCsp = X265_CSP_I420;
 
@@ -444,8 +444,8 @@
           "Search Range must be less than 32768");
     CHECK(param->keyframeMax < 0,
           "Keyframe interval must be 0 (auto) 1 (intra-only) or greater than 1");
-    CHECK(param->frameNumThreads <= 0,
-          "frameNumThreads (--frame-threads) must be 1 or higher");
+    CHECK(param->frameNumThreads < 0,
+          "frameNumThreads (--frame-threads) must be 0 or higher");
     CHECK(param->cbQpOffset < -12, "Min. Chroma Cb QP Offset is -12");
     CHECK(param->cbQpOffset >  12, "Max. Chroma Cb QP Offset is  12");
     CHECK(param->crQpOffset < -12, "Min. Chroma Cr QP Offset is -12");
diff -r 93cccbe49a93 -r 90d9c1067f50 source/encoder/encoder.cpp
--- a/source/encoder/encoder.cpp	Wed Nov 06 19:49:38 2013 -0600
+++ b/source/encoder/encoder.cpp	Wed Nov 06 21:38:18 2013 -0600
@@ -1024,11 +1024,26 @@
         _param->poolNumThreads = 1;
 
     setThreadPool(ThreadPool::allocThreadPool(_param->poolNumThreads));
-    int actual = ThreadPool::getThreadPool()->getThreadCount();
-    if (actual > 1)
+    int poolThreadCount = ThreadPool::getThreadPool()->getThreadCount();
+    int rows = (_param->sourceHeight + _param->maxCUSize - 1) / _param->maxCUSize;
+
+    if (_param->frameNumThreads == 0)
     {
-        x265_log(_param, X265_LOG_INFO, "WPP streams / pool / frames  : %d / %d / %d\n",
-                 (_param->sourceHeight + _param->maxCUSize - 1) / _param->maxCUSize, actual, _param->frameNumThreads);
+        // auto-detect frame threads
+        if (poolThreadCount > 32)
+            _param->frameNumThreads = 6;  // dual-socket 10-core IvyBridge or higher
+        else if (poolThreadCount >= 16)
+            _param->frameNumThreads = 5;  // 8 HT cores, or dual socket
+        else if (poolThreadCount >= 12)
+            _param->frameNumThreads = 3;  // 6 HT cores
+        else if (poolThreadCount >= 4)
+            _param->frameNumThreads = 2;  // Dual or Quad core
+        else
+            _param->frameNumThreads = 1;
+    }
+    if (poolThreadCount > 1)
+    {
+        x265_log(_param, X265_LOG_INFO, "WPP streams / pool / frames  : %d / %d / %d\n", rows, poolThreadCount, _param->frameNumThreads);
     }
     else if (_param->frameNumThreads > 1)
     {
@@ -1044,7 +1059,6 @@
     {
         x265_log(_param, X265_LOG_INFO, "Warning: picture-based SAO used with frame parallelism\n");
     }
-
     if (!_param->keyframeMin)
     {
         _param->keyframeMin = _param->keyframeMax;
diff -r 93cccbe49a93 -r 90d9c1067f50 source/x265.cpp
--- a/source/x265.cpp	Wed Nov 06 19:49:38 2013 -0600
+++ b/source/x265.cpp	Wed Nov 06 21:38:18 2013 -0600
@@ -266,7 +266,7 @@
     H0("   --threads                     Number of threads for thread pool (0: detect CPU core count, default)\n");
     H0("-p/--preset                      ultrafast, veryfast, faster, fast, medium, slow, slower, veryslow, or placebo\n");
     H0("-t/--tune                        Tune the settings for a particular type of source or situation\n");
-    H0("-F/--frame-threads               Number of concurrently encoded frames. Default %d\n", param->frameNumThreads);
+    H0("-F/--frame-threads               Number of concurrently encoded frames. 0: auto-determined by core count\n");
     H0("   --log                         Logging level 0:ERROR 1:WARNING 2:INFO 3:DEBUG -1:NONE. Default %d\n", param->logLevel);
     H0("   --csv                         Comma separated log file, log level >= 3 frame log, else one line per run\n");
     H0("   --y4m                         Parse input stream as YUV4MPEG2 regardless of file extension\n");
diff -r 93cccbe49a93 -r 90d9c1067f50 source/x265.h
--- a/source/x265.h	Wed Nov 06 19:49:38 2013 -0600
+++ b/source/x265.h	Wed Nov 06 21:38:18 2013 -0600
@@ -289,8 +289,8 @@
 {
     int       logLevel;
     int       bEnableWavefront;                ///< enable wavefront parallel processing
-    int       poolNumThreads;                  ///< number of threads to allocate for thread pool
-    int       frameNumThreads;                 ///< number of concurrently encoded frames
+    int       poolNumThreads;                  ///< number of threads to allocate for thread pool, 0 implies auto-detection (default)
+    int       frameNumThreads;                 ///< number of concurrently encoded frames, 0 implies auto-detection (default)
     const char *csvfn;                         ///< csv log filename. logLevel >= 3 is frame logging, else one line per run
 
     // source specification