Merge remote-tracking branch 'rbultje/vp9-simd'

* rbultje/vp9-simd:
  vp9: make mv bounds 32bit.
  vp9: reset contextual caches on frame size change with mt enabled.
  vp9/x86: idct_32x32_add_ssse3 sub-8x8-idct.
  vp9/x86: idct_32x32_add_ssse3 sub-16x16-idct.
  vp9/x86: idct_32x32_add_ssse3.
Merged-by: Michael Niedermayer <michaelni@gmx.at>

libavcodec/vp9.c

@@ -232,7 +232,7 @@ typedef struct VP9Context {
     // block reconstruction intermediates
     int16_t *block_base, *block, *uvblock_base[2], *uvblock[2];
     uint8_t *eob_base, *uveob_base[2], *eob, *uveob[2];
-    VP56mv min_mv, max_mv;
+    struct { int x, y; } min_mv, max_mv;
     DECLARE_ALIGNED(32, uint8_t, tmp_y)[64*64];
     DECLARE_ALIGNED(32, uint8_t, tmp_uv)[2][32*32];
 } VP9Context;
@@ -3450,6 +3450,13 @@ static void adapt_probs(VP9Context *s)
     }
 }
 
+static void free_buffers(VP9Context *s)
+{
+    av_freep(&s->above_partition_ctx);
+    av_freep(&s->b_base);
+    av_freep(&s->block_base);
+}
+
 static av_cold int vp9_decode_free(AVCodecContext *ctx)
 {
     VP9Context *s = ctx->priv_data;
@@ -3468,11 +3475,9 @@ static av_cold int vp9_decode_free(AVCodecContext *ctx)
             ff_thread_release_buffer(ctx, &s->next_refs[i]);
         av_frame_free(&s->next_refs[i].f);
     }
-    av_freep(&s->above_partition_ctx);
+    free_buffers(s);
     av_freep(&s->c_b);
     s->c_b_size = 0;
-    av_freep(&s->b_base);
-    av_freep(&s->block_base);
 
     return 0;
 }
@@ -3762,7 +3767,10 @@ static int vp9_decode_update_thread_context(AVCodecContext *dst, const AVCodecCo
     int i, res;
     VP9Context *s = dst->priv_data, *ssrc = src->priv_data;
 
-    // FIXME scalability, size, etc.
+    // detect size changes in other threads
+    if (s->above_partition_ctx && (s->cols != ssrc->cols || s->rows != ssrc->rows)) {
+        free_buffers(s);
+    }
 
     for (i = 0; i < 2; i++) {
         if (s->frames[i].tf.f->data[0])

libavcodec/x86/vp9dsp_init.c

@@ -159,6 +159,7 @@ filters_8tap_1d_fn3(avg)
 void ff_vp9_idct_idct_4x4_add_ssse3(uint8_t *dst, ptrdiff_t stride, int16_t *block, int eob);
 void ff_vp9_idct_idct_8x8_add_ssse3(uint8_t *dst, ptrdiff_t stride, int16_t *block, int eob);
 void ff_vp9_idct_idct_16x16_add_ssse3(uint8_t *dst, ptrdiff_t stride, int16_t *block, int eob);
+void ff_vp9_idct_idct_32x32_add_ssse3(uint8_t *dst, ptrdiff_t stride, int16_t *block, int eob);
 
 #endif /* HAVE_YASM */
 
@@ -217,6 +218,10 @@ av_cold void ff_vp9dsp_init_x86(VP9DSPContext *dsp)
         if (ARCH_X86_64) {
             dsp->itxfm_add[TX_8X8][DCT_DCT] = ff_vp9_idct_idct_8x8_add_ssse3;
             dsp->itxfm_add[TX_16X16][DCT_DCT] = ff_vp9_idct_idct_16x16_add_ssse3;
+            dsp->itxfm_add[TX_32X32][ADST_ADST] =
+            dsp->itxfm_add[TX_32X32][ADST_DCT] =
+            dsp->itxfm_add[TX_32X32][DCT_ADST] =
+            dsp->itxfm_add[TX_32X32][DCT_DCT] = ff_vp9_idct_idct_32x32_add_ssse3;
         }
     }