vp9: fix overflow in 8x8 topleft 32x32 idct ssse3 version.

Also disable the mmx/iwht optimization when the bitexact flag is set.
With synthetically coded coefficients (i.e. these that lead to a
residual well outside the [-255,255] range), our optimizations will
overflow. It doesn't make sense to fix the overflows, since they can
only occur on synthetic input, not on real fwht-generated input. Thus,
add a bitexact flag that disables this optimization.

5 files changed, 16 insertions, 10 deletions

diff --git a/libavcodec/vp9.c b/libavcodec/vp9.c
index 98fd8a3046..fcdd6e128e 100644
--- a/libavcodec/vp9.c
+++ b/libavcodec/vp9.c
@@ -360,7 +360,7 @@ static int update_size(AVCodecContext *ctx, int w, int h, enum AVPixelFormat fmt
     av_freep(&s->block_base);
 
     if (s->bpp != s->last_bpp) {
-        ff_vp9dsp_init(&s->dsp, s->bpp);
+        ff_vp9dsp_init(&s->dsp, s->bpp, ctx->flags & AV_CODEC_FLAG_BITEXACT);
         ff_videodsp_init(&s->vdsp, s->bpp);
         s->last_bpp = s->bpp;
     }
diff --git a/libavcodec/vp9dsp.c b/libavcodec/vp9dsp.c
index 018cb61edf..54e77e267b 100644
--- a/libavcodec/vp9dsp.c
+++ b/libavcodec/vp9dsp.c
@@ -25,7 +25,7 @@
 #include "libavutil/common.h"
 #include "vp9dsp.h"
 
-av_cold void ff_vp9dsp_init(VP9DSPContext *dsp, int bpp)
+av_cold void ff_vp9dsp_init(VP9DSPContext *dsp, int bpp, int bitexact)
 {
     if (bpp == 8) {
         ff_vp9dsp_init_8(dsp);
@@ -36,6 +36,6 @@ av_cold void ff_vp9dsp_init(VP9DSPContext *dsp, int bpp)
         ff_vp9dsp_init_12(dsp);
     }
 
-    if (ARCH_X86) ff_vp9dsp_init_x86(dsp, bpp);
+    if (ARCH_X86) ff_vp9dsp_init_x86(dsp, bpp, bitexact);
     if (ARCH_MIPS) ff_vp9dsp_init_mips(dsp, bpp);
 }
diff --git a/libavcodec/vp9dsp.h b/libavcodec/vp9dsp.h
index 28903b3a70..016a9bb231 100644
--- a/libavcodec/vp9dsp.h
+++ b/libavcodec/vp9dsp.h
@@ -120,13 +120,13 @@ typedef struct VP9DSPContext {
     vp9_scaled_mc_func smc[5][4][2];
 } VP9DSPContext;
 
-void ff_vp9dsp_init(VP9DSPContext *dsp, int bpp);
+void ff_vp9dsp_init(VP9DSPContext *dsp, int bpp, int bitexact);
 
 void ff_vp9dsp_init_8(VP9DSPContext *dsp);
 void ff_vp9dsp_init_10(VP9DSPContext *dsp);
 void ff_vp9dsp_init_12(VP9DSPContext *dsp);
 
-void ff_vp9dsp_init_x86(VP9DSPContext *dsp, int bpp);
+void ff_vp9dsp_init_x86(VP9DSPContext *dsp, int bpp, int bitexact);
 void ff_vp9dsp_init_mips(VP9DSPContext *dsp, int bpp);
 
 #endif /* AVCODEC_VP9DSP_H */
diff --git a/libavcodec/x86/vp9dsp_init.c b/libavcodec/x86/vp9dsp_init.c
index 00e7125a0c..f24cb674fa 100644
--- a/libavcodec/x86/vp9dsp_init.c
+++ b/libavcodec/x86/vp9dsp_init.c
@@ -307,7 +307,7 @@ ipred_func(32, tm, avx2);
 
 #endif /* HAVE_YASM */
 
-av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp, int bpp)
+av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp, int bpp, int bitexact)
 {
 #if HAVE_YASM
     int cpu_flags;
@@ -388,10 +388,12 @@ av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp, int bpp)
     if (EXTERNAL_MMX(cpu_flags)) {
         init_fpel(4, 0,  4, put, mmx);
         init_fpel(3, 0,  8, put, mmx);
-        dsp->itxfm_add[4 /* lossless */][DCT_DCT] =
-        dsp->itxfm_add[4 /* lossless */][ADST_DCT] =
-        dsp->itxfm_add[4 /* lossless */][DCT_ADST] =
-        dsp->itxfm_add[4 /* lossless */][ADST_ADST] = ff_vp9_iwht_iwht_4x4_add_mmx;
+        if (!bitexact) {
+            dsp->itxfm_add[4 /* lossless */][DCT_DCT] =
+            dsp->itxfm_add[4 /* lossless */][ADST_DCT] =
+            dsp->itxfm_add[4 /* lossless */][DCT_ADST] =
+            dsp->itxfm_add[4 /* lossless */][ADST_ADST] = ff_vp9_iwht_iwht_4x4_add_mmx;
+        }
         init_ipred(8, mmx, v, VERT);
     }
 
diff --git a/libavcodec/x86/vp9itxfm.asm b/libavcodec/x86/vp9itxfm.asm
index 02d4d0a4e2..54462a6220 100644
--- a/libavcodec/x86/vp9itxfm.asm
+++ b/libavcodec/x86/vp9itxfm.asm
@@ -1127,10 +1127,14 @@ IADST8_FN iadst, IADST8, iadst, IADST8, avx, 16
     pmulhrsw            m7, m4, [pw_16069x2]        ; t6-7
     pmulhrsw            m4, [pw_3196x2]             ; t4-5
 
+%if 0 ; overflows :(
     paddw               m6, m7, m4
     psubw               m5, m7, m4
     pmulhrsw            m5, [pw_11585x2]            ; t5
     pmulhrsw            m6, [pw_11585x2]            ; t6
+%else
+    VP9_UNPACK_MULSUB_2W_4X  5, 6, 7, 4, 11585, 11585, [pd_8192], 0, 1 ; t5,  t6
+%endif
 
     psubw               m0, m3, m7
     paddw               m7, m3