vp8_sixtap_predict16x16_neon: fix overread

Shift the final read from the source by 3 to avoid breaking the
assumption that the 6-tap filter needs only 5 pixels outside of the
macroblock; this matches the sse2 and ssse3 implementations.

It's possible this restriction could be removed if the source buffers
are assumed to be padded.

Bug: webm:1795
Change-Id: I4c791e3a214898a503c78f4cedca154c75cdbaef
Fixed: webm:1795

diff --git a/test/predict_test.cc b/test/predict_test.cc
index e49d982..7472970 100644 (file)
--- a/test/predict_test.cc
+++ b/test/predict_test.cc
@@ -307,9 +307,7 @@ INSTANTIATE_TEST_SUITE_P(
 #if HAVE_NEON
 INSTANTIATE_TEST_SUITE_P(
     NEON, SixtapPredictTest,
-    ::testing::Values(/*TODO(https://crbug.com/webm/1795): enable this after
-                        buffer overflows are fixed.
-                      make_tuple(16, 16, &vp8_sixtap_predict16x16_neon),*/
+    ::testing::Values(make_tuple(16, 16, &vp8_sixtap_predict16x16_neon),
                       make_tuple(8, 8, &vp8_sixtap_predict8x8_neon),
                       make_tuple(8, 4, &vp8_sixtap_predict8x4_neon),
                       make_tuple(4, 4, &vp8_sixtap_predict4x4_neon)));

diff --git a/vp8/common/arm/neon/sixtappredict_neon.c b/vp8/common/arm/neon/sixtappredict_neon.c
index 4960d16..b15cfb4 100644 (file)
--- a/vp8/common/arm/neon/sixtappredict_neon.c
+++ b/vp8/common/arm/neon/sixtappredict_neon.c
@@ -1253,9 +1253,6 @@ void vp8_sixtap_predict8x8_neon(unsigned char *src_ptr, int src_pixels_per_line,
   return;
 }
 
-// TODO(https://crbug.com/webm/1795): enable this after buffer overflows are
-// fixed.
-#if 0
 void vp8_sixtap_predict16x16_neon(unsigned char *src_ptr,
                                   int src_pixels_per_line, int xoffset,
                                   int yoffset, unsigned char *dst_ptr,
@@ -1507,7 +1504,9 @@ void vp8_sixtap_predict16x16_neon(unsigned char *src_ptr,
     src += src_pixels_per_line;
     d12u8 = vld1_u8(src);
     d13u8 = vld1_u8(src + 8);
-    d14u8 = vld1_u8(src + 16);
+    // Only 5 pixels are needed, avoid a potential out of bounds read.
+    d14u8 = vld1_u8(src + 13);
+    d14u8 = vext_u8(d14u8, d14u8, 3);
     src += src_pixels_per_line;
 
     __builtin_prefetch(src);
@@ -1731,4 +1730,3 @@ void vp8_sixtap_predict16x16_neon(unsigned char *src_ptr,
   }
   return;
 }
-#endif  // 0

diff --git a/vp8/common/rtcd_defs.pl b/vp8/common/rtcd_defs.pl
index 05e67ce..739a612 100644 (file)
--- a/vp8/common/rtcd_defs.pl
+++ b/vp8/common/rtcd_defs.pl
@@ -146,9 +146,7 @@ if (vpx_config("CONFIG_POSTPROC") eq "yes") {
 # Subpixel
 #
 add_proto qw/void vp8_sixtap_predict16x16/, "unsigned char *src_ptr, int src_pixels_per_line, int xoffset, int yoffset, unsigned char *dst_ptr, int dst_pitch";
-# TODO(https://crbug.com/webm/1795): enable neon after buffer overflows are
-# fixed.
-specialize qw/vp8_sixtap_predict16x16 sse2 ssse3 dspr2 msa mmi lsx/;
+specialize qw/vp8_sixtap_predict16x16 sse2 ssse3 neon dspr2 msa mmi lsx/;
 
 add_proto qw/void vp8_sixtap_predict8x8/, "unsigned char *src_ptr, int src_pixels_per_line, int xoffset, int yoffset, unsigned char *dst_ptr, int dst_pitch";
 specialize qw/vp8_sixtap_predict8x8 sse2 ssse3 neon dspr2 msa mmi lsx/;