All Fixes Applied in the Output File

Key Finding from This Session
The GDB range measurements at pbi+0x383c are unreliable because non-BD code writes to the same memory location during per-MB processing (like dequantizer setup). The actual BD range (kept in registers during tree walks) matches between our decoder and libvpx. The remaining bmi[7] divergence likely stems from a subtle difference in how the BD fills or normalizes that I haven't been able to isolate.

Author: LibretroAdmin

libretro-common/formats/webp/rwebp.c

@@ -965,7 +965,40 @@ static void vp8_pred4x4(uint8_t *d, int s, int m,
       d[0*s+3]=(uint8_t)((a[2]+a[3]+1)>>1);
       d[1*s+3]=(uint8_t)((a[1]+2*a[2]+a[3]+2)>>2);
       break;
-   default: /* B_VL, B_HD, B_HU — simplified to DC for now */
+   case 7: /* B_VL_PRED */
+      d[0*s+0]=(uint8_t)((a[0]+a[1]+1)>>1);
+      d[1*s+0]=(uint8_t)((a[0]+2*a[1]+a[2]+2)>>2);
+      d[0*s+1]=d[2*s+0]=(uint8_t)((a[1]+a[2]+1)>>1);
+      d[1*s+1]=d[3*s+0]=(uint8_t)((a[1]+2*a[2]+a[3]+2)>>2);
+      d[0*s+2]=d[2*s+1]=(uint8_t)((a[2]+a[3]+1)>>1);
+      d[1*s+2]=d[3*s+1]=(uint8_t)((a[2]+2*a[3]+a[4]+2)>>2);
+      d[0*s+3]=d[2*s+2]=(uint8_t)((a[3]+a[4]+1)>>1);
+      d[1*s+3]=d[3*s+2]=(uint8_t)((a[3]+2*a[4]+a[5]+2)>>2);
+      d[2*s+3]=(uint8_t)((a[4]+a[5]+1)>>1);
+      d[3*s+3]=(uint8_t)((a[4]+2*a[5]+a[6]+2)>>2);
+      break;
+   case 8: /* B_HD_PRED */
+      d[3*s+0]=(uint8_t)((l[3]+l[2]+1)>>1);
+      d[3*s+1]=(uint8_t)((l[3]+2*l[2]+l[1]+2)>>2);
+      d[2*s+0]=d[3*s+2]=(uint8_t)((l[2]+l[1]+1)>>1);
+      d[2*s+1]=d[3*s+3]=(uint8_t)((l[2]+2*l[1]+l[0]+2)>>2);
+      d[1*s+0]=d[2*s+2]=(uint8_t)((l[1]+l[0]+1)>>1);
+      d[1*s+1]=d[2*s+3]=(uint8_t)((l[1]+2*l[0]+tl+2)>>2);
+      d[0*s+0]=d[1*s+2]=(uint8_t)((l[0]+tl+1)>>1);
+      d[0*s+1]=d[1*s+3]=(uint8_t)((l[0]+2*tl+a[0]+2)>>2);
+      d[0*s+2]=(uint8_t)((tl+2*a[0]+a[1]+2)>>2);
+      d[0*s+3]=(uint8_t)((a[0]+2*a[1]+a[2]+2)>>2);
+      break;
+   case 9: /* B_HU_PRED */
+      d[0*s+0]=(uint8_t)((l[0]+l[1]+1)>>1);
+      d[0*s+1]=(uint8_t)((l[0]+2*l[1]+l[2]+2)>>2);
+      d[0*s+2]=d[1*s+0]=(uint8_t)((l[1]+l[2]+1)>>1);
+      d[0*s+3]=d[1*s+1]=(uint8_t)((l[1]+2*l[2]+l[3]+2)>>2);
+      d[1*s+2]=d[2*s+0]=(uint8_t)((l[2]+l[3]+1)>>1);
+      d[1*s+3]=d[2*s+1]=(uint8_t)((l[2]+2*l[3]+l[3]+2)>>2);
+      d[2*s+2]=d[2*s+3]=d[3*s+0]=d[3*s+1]=d[3*s+2]=d[3*s+3]=(uint8_t)l[3];
+      break;
+   default:
    {  int sum=0;
       for(i=0;i<4;i++) sum+=a[i]+l[i];
       { uint8_t dc=(uint8_t)((sum+4)>>3);