ref: 74c9802f42842d458165f5acedc5307fda383f7b
parent: b865e6c7141f3cb7d56f8868a307fe24905b17a4
author: David Michael Barr <b@rr-dav.id.au>
date: Sat Sep 29 18:35:58 EDT 2018
Remove DC row buffer for CfL prediction
--- a/src/ipred.c
+++ b/src/ipred.c
@@ -881,14 +881,13 @@
static NOINLINE void
cfl_pred_1_c(pixel *dst, const ptrdiff_t stride, const int16_t *ac,
- const pixel *const dc_pred, const int8_t alpha,
- const int width, const int height)
+ const int8_t alpha, const int width, const int height)
{+ const pixel dc = *dst;
for (int y = 0; y < height; y++) { for (int x = 0; x < width; x++) {const int diff = alpha * ac[x];
- dst[x] = iclip_pixel(dc_pred[0] + apply_sign((abs(diff) + 32) >> 6,
- diff));
+ dst[x] = iclip_pixel(dc + apply_sign((abs(diff) + 32) >> 6, diff));
}
ac += width;
dst += PXSTRIDE(stride);
@@ -899,11 +898,10 @@
static void cfl_pred_1_##width##xN_c(pixel *const dst, \
const ptrdiff_t stride, \
const int16_t *const ac, \
- const pixel *const dc_pred, \
const int8_t alpha, \
const int height) \
{ \- cfl_pred_1_c(dst, stride, ac, dc_pred, alpha, width, height); \
+ cfl_pred_1_c(dst, stride, ac, alpha, width, height); \
}
cfl_pred_1_fn( 4)
@@ -912,18 +910,16 @@
cfl_pred_1_fn(32)
static NOINLINE void
-cfl_pred_c(pixel *dstU, pixel *dstV, const ptrdiff_t stride,
- const int16_t *ac, const pixel *const dc_pred,
+cfl_pred_c(pixel *dstU, pixel *dstV, const ptrdiff_t stride, const int16_t *ac,
const int8_t *const alphas, const int width, const int height)
{+ const pixel dcU = *dstU, dcV = *dstV;
for (int y = 0; y < height; y++) { for (int x = 0; x < width; x++) {const int diff1 = alphas[0] * ac[x];
- dstU[x] = iclip_pixel(dc_pred[ 0] + apply_sign((abs(diff1) + 32) >> 6,
- diff1));
+ dstU[x] = iclip_pixel(dcU + apply_sign((abs(diff1) + 32) >> 6, diff1));
const int diff2 = alphas[1] * ac[x];
- dstV[x] = iclip_pixel(dc_pred[32] + apply_sign((abs(diff2) + 32) >> 6,
- diff2));
+ dstV[x] = iclip_pixel(dcV + apply_sign((abs(diff2) + 32) >> 6, diff2));
}
ac += width;
dstU += PXSTRIDE(stride);
@@ -936,11 +932,10 @@
pixel *const dstV, \
const ptrdiff_t stride, \
const int16_t *const ac, \
- const pixel *const dc_pred, \
const int8_t *const alphas, \
const int height) \
{ \- cfl_pred_c(dstU, dstV, stride, ac, dc_pred, alphas, width, height); \
+ cfl_pred_c(dstU, dstV, stride, ac, alphas, width, height); \
}
cfl_pred_fn( 4)
--- a/src/ipred.h
+++ b/src/ipred.h
@@ -55,26 +55,23 @@
typedef decl_cfl_ac_fn(*cfl_ac_fn);
/*
- * dst[x,y] = dc + alpha * ac[x,y]
+ * dst[x,y] += alpha * ac[x,y]
* - alpha contains a q3 scalar in [-16,16] range;
- * - dc_pred[] is the first line of the plane's DC prediction
*/
#define decl_cfl_pred_1_fn(name) \
void (name)(pixel *dst, ptrdiff_t stride, \
- const int16_t *ac, const pixel *dc_pred, \
- const int8_t alpha, const int height)
+ const int16_t *ac, const int8_t alpha, \
+ const int height)
typedef decl_cfl_pred_1_fn(*cfl_pred_1_fn);
/*
- * dst[plane][x,y] = dc[plane] + alpha[plane] * ac[x,y]
+ * dst[plane][x,y] += alpha[plane] * ac[x,y]
* - alphas contains two q3 scalars (one for each plane) in [-16,16] range;
- * - dc_pred[] is the first line of each plane's DC prediction, the second plane
- * starting at an offset of 16 * sizeof(pixel) bytes.
*/
#define decl_cfl_pred_fn(name) \
void (name)(pixel *u_dst, pixel *v_dst, ptrdiff_t stride, \
- const int16_t *ac, const pixel *dc_pred, \
- const int8_t *const alphas, const int height)
+ const int16_t *ac, const int8_t *const alphas, \
+ const int height)
typedef decl_cfl_pred_fn(*cfl_pred_fn);
/*
--- a/src/recon.c
+++ b/src/recon.c
@@ -836,7 +836,6 @@
assert(!init_x && !init_y);
int16_t *const ac = t->scratch.ac;
- ALIGN_STK_32(pixel, uv_pred, 2 * 32,);
pixel *y_src = ((pixel *) f->cur.p.data[0]) + 4 * (t->bx & ~ss_hor) +
4 * (t->by & ~ss_ver) * PXSTRIDE(f->cur.p.stride[0]);
const ptrdiff_t uv_off = 4 * ((t->bx >> ss_hor) +
@@ -870,13 +869,7 @@
top_sb_edge, DC_PRED, &angle,
cfl_uv_t_dim->w,
cfl_uv_t_dim->h, edge);
- if (b->cfl_alpha[pl]) {- dsp->ipred.intra_pred[cfl_uvtx][m](&uv_pred[32 * pl],
- 0, edge, 0);
- } else {- dsp->ipred.intra_pred[cfl_uvtx][m](uv_dst[pl], stride,
- edge, 0);
- }
+ dsp->ipred.intra_pred[cfl_uvtx][m](uv_dst[pl], stride, edge, 0);
}
const int furthest_r =
((cw4 << ss_hor) + t_dim->w - 1) & ~(t_dim->w - 1);
@@ -889,14 +882,12 @@
if (b->cfl_alpha[0] && b->cfl_alpha[1]) {dsp->ipred.cfl_pred[cfl_uv_t_dim->lw](uv_dst[0],
uv_dst[1], stride,
- ac, uv_pred,
- b->cfl_alpha,
+ ac, b->cfl_alpha,
cbh4 * 4);
} else {const int pl = !b->cfl_alpha[0];
dsp->ipred.cfl_pred_1[cfl_uv_t_dim->lw](uv_dst[pl],
stride, ac,
- &uv_pred[32 * pl],
b->cfl_alpha[pl],
cbh4 * 4);
}