shithub: opus

Info  •  Files  •  Log  •  Branches

The encoder would crash in the PVQ search if fed NaNs via the float interface. This patch protects against it in two sufficient ways: Making the PVQ search robust against NaNs and by squashing NaNs to zero on input.

Thanks to David Richards for reporting this failure mode.

--- a/libcelt/celt.c

+++ b/libcelt/celt.c

@@ -1086,6 +1086,8 @@

             x = SCALEIN(*pcmp);

 #ifndef FIXED_POINT

+            if (!(x==x))

+               x = 0;

             if (st->clip)

                x = MAX32(-65536.f, MIN32(65536.f,x));

 #endif

--- a/libcelt/vq.c

+++ b/libcelt/vq.c

@@ -223,7 +223,9 @@

 #ifdef FIXED_POINT

       if (sum <= K)

 #else

-      if (sum <= EPSILON)

+      /* Prevents infinities and NaNs from causing too many pulses

+         to be allocated. 64 is an approximation of infinity here. */

+      if (!(sum > EPSILON && sum < 64))

 #endif

       {

          X[0] = QCONST16(1.f,14);