Disable Libav's AVX and FMA4 optimizations under Windows x86_64.

They cause crashes in avresample.

SSE* optimizations are still enabled, so the performance hit should be pretty minimal.

git-svn-id: svn://svn.handbrake.fr/HandBrake/trunk@5409 b64f7644-9d1e-0410-96f1-a4d463321fa5

diff --git a/libhb/hb.c b/libhb/hb.c
index a7df7a220f162a40e90952e261f51dced6a77044..35e341b8a375e0164733fbb7e2b58dd4786f4f8e 100644 (file)
--- a/libhb/hb.c
+++ b/libhb/hb.c
@@ -102,8 +102,16 @@ static int ff_lockmgr_cb(void **mutex, enum AVLockOp op)
 
 void hb_avcodec_init()
 {
-    av_lockmgr_register( ff_lockmgr_cb );
+    av_lockmgr_register(ff_lockmgr_cb);
     av_register_all();
+#ifdef _WIN64
+    // avresample's assembly optimizations can cause crashes under Win x86_64
+    // (see http://bugzilla.libav.org/show_bug.cgi?id=496)
+    // disable AVX and FMA4 as a workaround
+    hb_deep_log(2, "hb_avcodec_init: Windows x86_64, disabling AVX and FMA4");
+    int cpu_flags = av_get_cpu_flags() & ~AV_CPU_FLAG_AVX & ~AV_CPU_FLAG_FMA4;
+    av_set_cpu_flags_mask(cpu_flags);
+#endif
 }
 
 int hb_avcodec_open(AVCodecContext *avctx, AVCodec *codec,