From ea4d5f48373ab856e01a3cf05426db6e823e01d1 Mon Sep 17 00:00:00 2001
From: Janne Grunau <janne-libav@jannau.net>
Date: Wed, 21 Dec 2011 01:18:01 +0100
Subject: [PATCH] linux: use number of CPUs as automatic thread count

Use sched_getaffinity to determine the number of logical CPUs.
Limits the number of threads to 16 since slice threading of H.264
seems to be buggy with more than 16 threads.
---
 configure            |  2 ++
 libavcodec/pthread.c | 43 +++++++++++++++++++++++++++++++++++++++++++
 2 files changed, 45 insertions(+)

diff --git a/configure b/configure
index 8e2e08b95f..b5ee3b5bb2 100755
--- a/configure
+++ b/configure
@@ -1117,6 +1117,7 @@ HAVE_LIST="
     posix_memalign
     round
     roundf
+    sched_getaffinity
     sdl
     sdl_video_size
     setmode
@@ -2853,6 +2854,7 @@ check_func  setrlimit
 check_func  strerror_r
 check_func  strptime
 check_func  strtok_r
+check_func  sched_getaffinity
 check_func_headers io.h setmode
 check_func_headers lzo/lzo1x.h lzo1x_999_compress
 check_lib2 "windows.h psapi.h" GetProcessMemoryInfo -lpsapi
diff --git a/libavcodec/pthread.c b/libavcodec/pthread.c
index ac15fefd68..1800f291d7 100644
--- a/libavcodec/pthread.c
+++ b/libavcodec/pthread.c
@@ -30,6 +30,12 @@
  */
 
 #include "config.h"
+
+#if HAVE_SCHED_GETAFFINITY
+#define _GNU_SOURCE
+#include <sched.h>
+#endif
+
 #include "avcodec.h"
 #include "internal.h"
 #include "thread.h"
@@ -133,6 +139,29 @@ typedef struct FrameThreadContext {
     int die;                       ///< Set when threads should exit.
 } FrameThreadContext;
 
+
+/* H264 slice threading seems to be buggy with more than 16 threads,
+ * limit the number of threads to 16 for automatic detection */
+#define MAX_AUTO_THREADS 16
+
+static int get_logical_cpus(AVCodecContext *avctx)
+{
+    int ret, nb_cpus = 1;
+#if HAVE_SCHED_GETAFFINITY
+    cpu_set_t cpuset;
+
+    CPU_ZERO(&cpuset);
+
+    ret = sched_getaffinity(0, sizeof(cpuset), &cpuset);
+    if (!ret) {
+        nb_cpus = CPU_COUNT(&cpuset);
+    }
+#endif
+    av_log(avctx, AV_LOG_DEBUG, "detected %d logical cores\n", nb_cpus);
+    return FFMIN(nb_cpus, MAX_AUTO_THREADS);
+}
+
+
 static void* attribute_align_arg worker(void *v)
 {
     AVCodecContext *avctx = v;
@@ -237,6 +266,13 @@ static int thread_init(AVCodecContext *avctx)
     ThreadContext *c;
     int thread_count = avctx->thread_count;
 
+    if (!thread_count) {
+        int nb_cpus = get_logical_cpus(avctx);
+        // use number of cores + 1 as thread count if there is motre than one
+        if (nb_cpus > 1)
+            thread_count = avctx->thread_count = nb_cpus + 1;
+    }
+
     if (thread_count <= 1) {
         avctx->active_thread_type = 0;
         return 0;
@@ -697,6 +733,13 @@ static int frame_thread_init(AVCodecContext *avctx)
     FrameThreadContext *fctx;
     int i, err = 0;
 
+    if (!thread_count) {
+        int nb_cpus = get_logical_cpus(avctx);
+        // use number of cores + 1 as thread count if there is motre than one
+        if (nb_cpus > 1)
+            thread_count = avctx->thread_count = nb_cpus + 1;
+    }
+
     if (thread_count <= 1) {
         avctx->active_thread_type = 0;
         return 0;