vp9: drop support for real (non-emulated) edges

They are not measurably faster on x86, they might be somewhat faster on
other platforms due to missing emu edge SIMD, but the gain is not large
enough to justify the added complexity.

libavcodec/vp9block.c

@@ -1583,9 +1583,9 @@ int ff_vp9_decode_block(AVCodecContext *avctx, int row, int col,
      * This allows to support emu-edge and so on even if we have large
      * block overhangs. */
     emu[0] = (col + w4) * 8 > s->cur_frame->linesize[0] ||
-             (row + h4) > s->rows + 2 * !(avctx->flags & CODEC_FLAG_EMU_EDGE);
+             (row + h4) > s->rows;
     emu[1] = (col + w4) * 4 > s->cur_frame->linesize[1] ||
-             (row + h4) > s->rows + 2 * !(avctx->flags & CODEC_FLAG_EMU_EDGE);
+             (row + h4) > s->rows;
     if (emu[0]) {
         b->dst[0]   = s->tmp_y;
         b->y_stride = 64;

tests/fate/vpx.mak

@@ -78,7 +78,6 @@ $(eval $(call FATE_VP9_SUITE,tiling-pedestrian,$(1),$(2)))
 endef
 
 $(eval $(call FATE_VP9_FULL))
-$(eval $(call FATE_VP9_FULL,-emu-edge,-flags +emu_edge))
 
 FATE_SAMPLES_AVCONV-$(CONFIG_VP9_DECODER) += $(FATE_VP9-yes)
 fate-vp9: $(FATE_VP9-yes)