shithub: libvpx

Info  •  Files  •  Log  •  Branches

Earmark extra space for VSX.

Backend specific optimization for PPC VSX reads 16 bytes, whereas arm neon /
sse2 only reads <= 8 bytes. Although the extra bytes read are actually never
used, this is not a warrant for groping around.  Fixed by allocating more when
building for VSX. This is reported by asan.

Also note - PPC does have assembly that loads 64-bit content from memory - lxsdx
loads one 64-bit doubleword (whereas lxvd2x loads two 64-bit doubleword) from
memory. However, we only have "vec_vsx_ld" builtins that mapped to lxvd2x, no
builtins to lxsdx. The only way to access lxsdx is through inline assembly,
which does not fit well in the origin paradigm.

Refer:
  vsx:
    vpx_tm_predictor_4x4_vsx @ third_party/libvpx/git_root/vpx_dsp/ppc/intrapred_vsx.c
  neon:
    vpx_tm_predictor_4x4_neon @ third_party/libvpx/git_root/vpx_dsp/arm/intrapred_neon_asm.asm
  sse2:
    tm_predictor_4x4 @ third_party/libvpx/git_root/vpx_dsp/x86/intrapred_sse2.asm

BUG=b/63112600

Tested:
  asan tests passed.

Change-Id: I5f74b56e35c05b67851de8b5530aece213f2ce9d

--- a/vp8/common/reconintra.c

+++ b/vp8/common/reconintra.c

@@ -71,8 +71,16 @@

     unsigned char *uleft, unsigned char *vleft, int left_stride,

     unsigned char *upred_ptr, unsigned char *vpred_ptr, int pred_stride) {

   MB_PREDICTION_MODE uvmode = x->mode_info_context->mbmi.uv_mode;

+#if HAVE_VSX

+  /* Power PC implementation uses "vec_vsx_ld" to read 16 bytes from

+     uleft_col and vleft_col. Play it safe by reserving enough stack

+     space here. */

+  unsigned char uleft_col[16];

+  unsigned char vleft_col[16];

+#else

   unsigned char uleft_col[8];

   unsigned char vleft_col[8];

+#endif

   int i;

   intra_pred_fn fn;

--- a/vp8/common/reconintra4x4.c

+++ b/vp8/common/reconintra4x4.c

@@ -40,7 +40,15 @@

                           int left_stride, B_PREDICTION_MODE b_mode,

                           unsigned char *dst, int dst_stride,

                           unsigned char top_left) {

-  unsigned char Aboveb[12], *Above = Aboveb + 4;

+/* Power PC implementation uses "vec_vsx_ld" to read 16 bytes from

+   Above (aka, Aboveb + 4). Play it safe by reserving enough stack

+   space here. Similary for "Left". */

+#if HAVE_VSX

+  unsigned char Aboveb[20];

+#else

+  unsigned char Aboveb[12];

+#endif

+  unsigned char *Above = Aboveb + 4;

 #if HAVE_NEON

   // Neon intrinsics are unable to load 32 bits, or 4 8 bit values. Instead, it

   // over reads but does not use the extra 4 values.

@@ -50,6 +58,8 @@

   // indeed read, they are not used.

   vp8_zero_array(Left, 8);

 #endif  // VPX_WITH_ASAN

+#elif HAVE_VSX

+  unsigned char Left[16];

 #else

   unsigned char Left[4];

 #endif  // HAVE_NEON