Retune CDEF dering threshold adjustment

Change the adjustment range from [50% ... 300%] to [25% ... 100%].
Adjustments above 100% were very rare, and capping the adjustment at
100% adds SIMD optimisation opportunities.  And lowering the other end
to 25% seems to help compresson slightly.

Low latency, used-cpu=0:
   PSNR | PSNR Cb | PSNR Cr | PSNR HVS |    SSIM | MS SSIM | CIEDE 2000
-0.0453 | -0.1118 | -0.1127 |  -0.0689 | -0.0429 | -0.0814 |    -0.0762

High latency, used-cpu=0:
   PSNR | PSNR Cb | PSNR Cr | PSNR HVS |   SSIM | MS SSIM | CIEDE 2000
-0.0303 |  0.0583 |  0.1740 |  -0.0440 | 0.0033 | -0.0042 |     0.0040

Change-Id: Id999158330a53e8c3383cd0e53a91c7f59fe062a

av1/common/od_dering.c

@@ -176,14 +176,6 @@ void od_filter_dering_direction_4x4_c(uint16_t *y, int ystride,
   }
 }
 
-/* This table approximates x^0.16 with the index being log2(x). It is clamped
-   to [-.5, 3]. The table is computed as:
-   round(256*min(3, max(.5, 1.08*(sqrt(2)*2.^([0:17]+8)/256/256).^.16))) */
-static const int16_t OD_THRESH_TABLE_Q8[18] = {
-  128, 134, 150, 168, 188, 210, 234, 262, 292,
-  327, 365, 408, 455, 509, 569, 635, 710, 768,
-};
-
 /* Compute deringing filter threshold for an 8x8 block based on the
    directional variance difference. A high variance difference means that we
    have a highly directional pattern (e.g. a high contrast edge), so we can
@@ -191,10 +183,9 @@ static const int16_t OD_THRESH_TABLE_Q8[18] = {
    contrast edge, or a non-directional texture, so we want to be careful not
    to blur. */
 static INLINE int od_adjust_thresh(int threshold, int32_t var) {
-  int v1;
+  const int i = var >> 6 ? AOMMIN(get_msb(var >> 6), 12) : 0;
   /* We use the variance of 8x8 blocks to adjust the threshold. */
-  v1 = OD_MINI(32767, var >> 6);
+  return var ? (threshold * (4 + i) + 8) >> 4 : 0;
-  return (threshold * OD_THRESH_TABLE_Q8[OD_ILOG(v1)] + 128) >> 8;
 }
 
 void copy_8x8_16bit_to_16bit_c(uint16_t *dst, int dstride, const uint16_t *src,