avcodec/x86/lossless_videoencdsp_init: Don't read from before the buffer

sub_median_pred_mmxext() calculates a predictor from the left, top
and topleft pixel values. The left value is simply read via
ptr[-1], although this is not guaranteed to be inside the buffer
in case of negative strides. This happens e.g. with

ffmpeg -i fate-suite/mpeg2/dvd_single_frame.vob -vf vflip \
       -c:v magicyuv -pred median -f null -

Fix this by reading the first value like the topleft value.
Also change the documentation of sub_median_pred to reflect this
change (and the one from 791b5954bc).

Reviewed-by: Lynne <dev@lynne.ee>
Signed-off-by: Andreas Rheinhardt <andreas.rheinhardt@outlook.com>

libavcodec/lossless_videoencdsp.h

@@ -29,7 +29,6 @@ typedef struct LLVidEncDSPContext {
                        intptr_t w);
     /**
      * Subtract HuffYUV's variant of median prediction.
-     * Note, this might read from src1[-1], src2[-1].
      */
     void (*sub_median_pred)(uint8_t *dst, const uint8_t *src1,
                             const uint8_t *src2, intptr_t w,

libavcodec/x86/lossless_videoencdsp_init.c

@@ -49,9 +49,13 @@ static void sub_median_pred_mmxext(uint8_t *dst, const uint8_t *src1,
     __asm__ volatile (
         "movq  (%1, %0), %%mm0          \n\t" // LT
         "psllq $8, %%mm0                \n\t"
+        "movq  (%2, %0), %%mm2          \n\t" // L
+        "psllq $8, %%mm2                \n\t"
+        "jmp 2f                         \n\t"
         "1:                             \n\t"
-        "movq  (%1, %0), %%mm1          \n\t" // T
         "movq  -1(%2, %0), %%mm2        \n\t" // L
+        "2:                             \n\t"
+        "movq  (%1, %0), %%mm1          \n\t" // T
         "movq  (%2, %0), %%mm3          \n\t" // X
         "movq %%mm2, %%mm4              \n\t" // L
         "psubb %%mm0, %%mm2             \n\t"