---
libavcodec/dirac_dwt.c | 571 ++++++++++++++++++++++++++++++++++++++++++++++++
libavcodec/dirac_dwt.h | 74 +++++++
libavcodec/dwt.h | 7 +-
3 files changed, 650 insertions(+), 2 deletions(-)
create mode 100644 libavcodec/dirac_dwt.c
create mode 100644 libavcodec/dirac_dwt.h
diff --git a/libavcodec/dirac_dwt.c b/libavcodec/dirac_dwt.c
new file mode 100644
index 0000000..4b3908a
--- /dev/null
+++ b/libavcodec/dirac_dwt.c
@@ -0,0 +1,571 @@
+#include "dirac_dwt.h"
+
+static av_always_inline void interleave(IDWTELEM *dst, IDWTELEM *src0,
+ IDWTELEM *src1, int w2, int add,
+ int shift)
+{
+ int i;
+ for (i = 0; i < w2; i++) {
+ dst[2 * i] = src0[i] + add >> shift;
+ dst[2 * i + 1] = src1[i] + add >> shift;
+ }
+}
+
+static void horizontal_compose_dirac53i(IDWTELEM *b, IDWTELEM *temp, int w)
+{
+ const int w2 = w >> 1;
+ int x;
+
+ temp[0] = COMPOSE_53iL0(b[w2], b[0], b[w2]);
+ for (x = 1; x < w2; x++) {
+ temp[x] = COMPOSE_53iL0(b[x + w2 - 1], b[x], b[x + w2]);
+ temp[x + w2 - 1] = COMPOSE_DIRAC53iH0(temp[x - 1], b[x + w2 - 1],
+ temp[x]);
+ }
+ temp[w - 1] = COMPOSE_DIRAC53iH0(temp[w2 - 1], b[w - 1], temp[w2 - 1]);
+
+ interleave(b, temp, temp + w2, w2, 1, 1);
+}
+
+static void horizontal_compose_dd97i(IDWTELEM *b, IDWTELEM *tmp, int w)
+{
+ const int w2 = w >> 1;
+ int x;
+
+ tmp[0] = COMPOSE_53iL0(b[w2], b[0], b[w2]);
+ for (x = 1; x < w2; x++)
+ tmp[x] = COMPOSE_53iL0(b[x + w2 - 1], b[x], b[x + w2]);
+
+ // extend the edges
+ tmp[-1] = tmp[0];
+ tmp[w2 + 1] = tmp[w2] = tmp[w2 - 1];
+
+ for (x = 0; x < w2; x++) {
+ b[2 * x] = tmp[x] + 1 >> 1;
+ b[2 * x + 1] = COMPOSE_DD97iH0(tmp[x - 1], tmp[x], b[x + w2],
+ tmp[x + 1], tmp[x + 2]) + 1 >> 1;
+ }
+}
+
+static void horizontal_compose_dd137i(IDWTELEM *b, IDWTELEM *tmp, int w)
+{
+ const int w2 = w >> 1;
+ int x;
+
+ tmp[0] = COMPOSE_DD137iL0(b[w2], b[w2], b[0], b[w2], b[w2 + 1]);
+ tmp[1] = COMPOSE_DD137iL0(b[w2], b[w2], b[1], b[w2 + 1], b[w2 + 2]);
+ for (x = 2; x < w2 - 1; x++)
+ tmp[x] = COMPOSE_DD137iL0(b[x + w2 - 2], b[x + w2 - 1], b[x], b[x +
w2],
+ b[x + w2 + 1]);
+ tmp[w2 - 1] = COMPOSE_DD137iL0(b[w - 3], b[w - 2], b[w2 - 1], b[w - 1],
+ b[w - 1]);
+
+ // extend the edges
+ tmp[-1] = tmp[0];
+ tmp[w2 + 1] = tmp[w2] = tmp[w2 - 1];
+
+ for (x = 0; x < w2; x++) {
+ b[2 * x] = tmp[x] + 1 >> 1;
+ b[2 * x + 1] = COMPOSE_DD97iH0(tmp[x - 1], tmp[x], b[x + w2],
+ tmp[x + 1], tmp[x + 2]) + 1 >> 1;
+ }
+}
+
+static av_always_inline void horizontal_compose_haari(IDWTELEM *b,
+ IDWTELEM *temp, int w,
+ int shift)
+{
+ const int w2 = w >> 1;
+ int x;
+
+ for (x = 0; x < w2; x++) {
+ temp[x] = COMPOSE_HAARiL0(b[x], b[x + w2]);
+ temp[x + w2] = COMPOSE_HAARiH0(b[x + w2], temp[x]);
+ }
+
+ interleave(b, temp, temp + w2, w2, shift, shift);
+}
+
+static void horizontal_compose_haar0i(IDWTELEM *b, IDWTELEM *temp, int w)
+{
+ horizontal_compose_haari(b, temp, w, 0);
+}
+
+static void horizontal_compose_haar1i(IDWTELEM *b, IDWTELEM *temp, int w)
+{
+ horizontal_compose_haari(b, temp, w, 1);
+}
+
+static void horizontal_compose_fidelityi(IDWTELEM *b, IDWTELEM *tmp, int w)
+{
+ const int w2 = w >> 1;
+ int i, x;
+ IDWTELEM v[8];
+
+ for (x = 0; x < w2; x++) {
+ for (i = 0; i < 8; i++)
+ v[i] = b[av_clip(x - 3 + i, 0, w2 - 1)];
+ tmp[x] = COMPOSE_FIDELITYiH0(v[0], v[1], v[2], v[3], b[x + w2], v[4],
+ v[5], v[6], v[7]);
+ }
+
+ for (x = 0; x < w2; x++) {
+ for (i = 0; i < 8; i++)
+ v[i] = tmp[av_clip(x - 4 + i, 0, w2 - 1)];
+ tmp[x + w2] = COMPOSE_FIDELITYiL0(v[0], v[1], v[2], v[3], b[x], v[4],
+ v[5], v[6], v[7]);
+ }
+
+ interleave(b, tmp + w2, tmp, w2, 0, 0);
+}
+
+static void horizontal_compose_daub97i(IDWTELEM *b, IDWTELEM *temp, int w)
+{
+ const int w2 = w >> 1;
+ int x, b0, b1, b2;
+
+ temp[0] = COMPOSE_DAUB97iL1(b[w2], b[0], b[w2]);
+ for (x = 1; x < w2; x++) {
+ temp[x] = COMPOSE_DAUB97iL1(b[x + w2 - 1], b[x], b[x + w2]);
+ temp[x + w2 - 1] = COMPOSE_DAUB97iH1(temp[x - 1], b[x + w2 - 1],
+ temp[x]);
+ }
+ temp[w - 1] = COMPOSE_DAUB97iH1(temp[w2 - 1], b[w - 1], temp[w2 - 1]);
+
+ // second stage combined with interleave and shift
+ b0 = b2 = COMPOSE_DAUB97iL0(temp[w2], temp[0], temp[w2]);
+ b[0] = b0 + 1 >> 1;
+ for (x = 1; x < w2; x++) {
+ b2 = COMPOSE_DAUB97iL0(temp[x + w2 - 1], temp[x], temp[x +
w2]);
+ b1 = COMPOSE_DAUB97iH0(b0, temp[x + w2 - 1], b2);
+ b[2 * x - 1] = b1 + 1 >> 1;
+ b[2 * x] = b2 + 1 >> 1;
+ b0 = b2;
+ }
+ b[w - 1] = (COMPOSE_DAUB97iH0(b2, temp[w - 1], b2) + 1) >> 1;
+}
+
+static void vertical_compose_dirac53iH0(IDWTELEM *b0, IDWTELEM *b1,
+ IDWTELEM *b2, int width)
+{
+ int i;
+
+ for (i = 0; i < width; i++)
+ b1[i] = COMPOSE_DIRAC53iH0(b0[i], b1[i], b2[i]);
+}
+
+static void vertical_compose_dd97iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
+ IDWTELEM *b3, IDWTELEM *b4, int width)
+{
+ int i;
+
+ for (i = 0; i < width; i++)
+ b2[i] = COMPOSE_DD97iH0(b0[i], b1[i], b2[i], b3[i], b4[i]);
+}
+
+static void vertical_compose_dd137iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
+ IDWTELEM *b3, IDWTELEM *b4, int width)
+{
+ int i;
+
+ for (i = 0; i < width; i++)
+ b2[i] = COMPOSE_DD137iL0(b0[i], b1[i], b2[i], b3[i], b4[i]);
+}
+
+static void vertical_compose_haar(IDWTELEM *b0, IDWTELEM *b1, int width)
+{
+ int i;
+
+ for (i = 0; i < width; i++) {
+ b0[i] = COMPOSE_HAARiL0(b0[i], b1[i]);
+ b1[i] = COMPOSE_HAARiH0(b1[i], b0[i]);
+ }
+}
+
+static void vertical_compose_fidelityiH0(IDWTELEM *dst, IDWTELEM *b[8],
+ int width)
+{
+ int i;
+
+ for (i = 0; i < width; i++)
+ dst[i] = COMPOSE_FIDELITYiH0(b[0][i], b[1][i], b[2][i], b[3][i],
dst[i],
+ b[4][i], b[5][i], b[6][i], b[7][i]);
+}
+
+static void vertical_compose_fidelityiL0(IDWTELEM *dst, IDWTELEM *b[8],
+ int width)
+{
+ int i;
+
+ for (i = 0; i < width; i++)
+ dst[i] = COMPOSE_FIDELITYiL0(b[0][i], b[1][i], b[2][i], b[3][i],
dst[i],
+ b[4][i], b[5][i], b[6][i], b[7][i]);
+}
+
+static void vertical_compose_daub97iH0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM
*b2,
+ int width)
+{
+ int i;
+
+ for (i = 0; i < width; i++)
+ b1[i] = COMPOSE_DAUB97iH0(b0[i], b1[i], b2[i]);
+}
+
+static void vertical_compose_daub97iH1(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM
*b2,
+ int width)
+{
+ int i;
+
+ for (i = 0; i < width; i++)
+ b1[i] = COMPOSE_DAUB97iH1(b0[i], b1[i], b2[i]);
+}
+
+static void vertical_compose_daub97iL0(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM
*b2,
+ int width)
+{
+ int i;
+
+ for (i = 0; i < width; i++)
+ b1[i] = COMPOSE_DAUB97iL0(b0[i], b1[i], b2[i]);
+}
+
+static void vertical_compose_daub97iL1(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM
*b2,
+ int width)
+{
+ int i;
+
+ for (i = 0; i < width; i++)
+ b1[i] = COMPOSE_DAUB97iL1(b0[i], b1[i], b2[i]);
+}
+
+static void spatial_compose_dd97i_dy(DWTContext *d, int level, int width,
+ int height, int stride)
+{
+ vertical_compose_3tap vertical_compose_l0 = d->vertical_compose_l0;
+ vertical_compose_5tap vertical_compose_h0 = d->vertical_compose_h0;
+ DWTCompose *cs = d->cs + level;
+
+ int i, y = cs->y;
+ IDWTELEM *b[8];
+ for (i = 0; i < 6; i++)
+ b[i] = cs->b[i];
+ b[6] = d->buffer + av_clip(y + 5, 0, height - 2) * stride;
+ b[7] = d->buffer + av_clip(y + 6, 1, height - 1) * stride;
+
+ if (y + 5 < (unsigned)height)
+ vertical_compose_l0(b[5], b[6], b[7], width);
+ if (y + 1 < (unsigned)height)
+ vertical_compose_h0(b[0], b[2], b[3], b[4], b[6], width);
+ if (y - 1 < (unsigned)height)
+ d->horizontal_compose(b[0], d->temp, width);
+ if (y + 0 < (unsigned)height)
+ d->horizontal_compose(b[1], d->temp, width);
+
+ for (i = 0; i < 6; i++)
+ cs->b[i] = b[i + 2];
+ cs->y += 2;
+}
+
+static void spatial_compose_dirac53i_dy(DWTContext *d, int level, int width,
+ int height, int stride)
+{
+ vertical_compose_3tap vertical_compose_l0 = d->vertical_compose_l0;
+ vertical_compose_3tap vertical_compose_h0 = d->vertical_compose_h0;
+ DWTCompose *cs = d->cs + level;
+
+ int y = cs->y;
+ IDWTELEM *b[4] = { cs->b[0], cs->b[1] };
+ b[2] = d->buffer + mirror(y + 1, height - 1) * stride;
+ b[3] = d->buffer + mirror(y + 2, height - 1) * stride;
+
+ if (y + 1 < (unsigned)height)
+ vertical_compose_l0(b[1], b[2], b[3], width);
+ if (y + 0 < (unsigned)height)
+ vertical_compose_h0(b[0], b[1], b[2], width);
+ if (y - 1 < (unsigned)height)
+ d->horizontal_compose(b[0], d->temp, width);
+ if (y + 0 < (unsigned)height)
+ d->horizontal_compose(b[1], d->temp, width);
+
+ cs->b[0] = b[2];
+ cs->b[1] = b[3];
+ cs->y += 2;
+}
+
+static void spatial_compose_dd137i_dy(DWTContext *d, int level, int width,
+ int height, int stride)
+{
+ vertical_compose_5tap vertical_compose_l0 = d->vertical_compose_l0;
+ vertical_compose_5tap vertical_compose_h0 = d->vertical_compose_h0;
+ DWTCompose *cs = d->cs + level;
+
+ int i, y = cs->y;
+ IDWTELEM *b[10];
+ for (i = 0; i < 8; i++)
+ b[i] = cs->b[i];
+ b[8] = d->buffer + av_clip(y + 7, 0, height - 2) * stride;
+ b[9] = d->buffer + av_clip(y + 8, 1, height - 1) * stride;
+
+ if (y + 5 < (unsigned)height)
+ vertical_compose_l0(b[3], b[5], b[6], b[7], b[9], width);
+ if (y + 1 < (unsigned)height)
+ vertical_compose_h0(b[0], b[2], b[3], b[4], b[6], width);
+ if (y - 1 < (unsigned)height)
+ d->horizontal_compose(b[0], d->temp, width);
+ if (y + 0 < (unsigned)height)
+ d->horizontal_compose(b[1], d->temp, width);
+
+ for (i = 0; i < 8; i++)
+ cs->b[i] = b[i + 2];
+ cs->y += 2;
+}
+
+// haar makes the assumption that height is even (always true for dirac)
+static void spatial_compose_haari_dy(DWTContext *d, int level, int width,
+ int height, int stride)
+{
+ vertical_compose_2tap vertical_compose = d->vertical_compose;
+ int y = d->cs[level].y;
+ IDWTELEM *b0 = d->buffer + (y - 1) * stride;
+ IDWTELEM *b1 = d->buffer + y * stride;
+
+ vertical_compose(b0, b1, width);
+ d->horizontal_compose(b0, d->temp, width);
+ d->horizontal_compose(b1, d->temp, width);
+
+ d->cs[level].y += 2;
+}
+
+// Don't do sliced idwt for fidelity; the 9 tap filter makes it a bit annoying
+// Fortunately, this filter isn't used in practice.
+static void spatial_compose_fidelity(DWTContext *d, int level, int width,
+ int height, int stride)
+{
+ vertical_compose_9tap vertical_compose_l0 = d->vertical_compose_l0;
+ vertical_compose_9tap vertical_compose_h0 = d->vertical_compose_h0;
+ int i, y;
+ IDWTELEM *b[8];
+
+ for (y = 1; y < height; y += 2) {
+ for (i = 0; i < 8; i++)
+ b[i] = d->buffer + av_clip((y - 7 + 2 * i), 0, height - 2) *
stride;
+ vertical_compose_h0(d->buffer + y * stride, b, width);
+ }
+
+ for (y = 0; y < height; y += 2) {
+ for (i = 0; i < 8; i++)
+ b[i] = d->buffer + av_clip((y - 7 + 2 * i), 1, height - 1) *
stride;
+ vertical_compose_l0(d->buffer + y * stride, b, width);
+ }
+
+ for (y = 0; y < height; y++)
+ d->horizontal_compose(d->buffer + y * stride, d->temp, width);
+
+ d->cs[level].y = height + 1;
+}
+
+static void spatial_compose_daub97i_dy(DWTContext *d, int level, int width,
+ int height, int stride)
+{
+ vertical_compose_3tap vertical_compose_l0 = d->vertical_compose_l0;
+ vertical_compose_3tap vertical_compose_h0 = d->vertical_compose_h0;
+ vertical_compose_3tap vertical_compose_l1 = d->vertical_compose_l1;
+ vertical_compose_3tap vertical_compose_h1 = d->vertical_compose_h1;
+ DWTCompose *cs = d->cs + level;
+
+ int i, y = cs->y;
+ IDWTELEM *b[6];
+ for (i = 0; i < 4; i++)
+ b[i] = cs->b[i];
+ b[4] = d->buffer + mirror(y + 3, height - 1) * stride;
+ b[5] = d->buffer + mirror(y + 4, height - 1) * stride;
+
+ if (y + 3 < (unsigned)height)
+ vertical_compose_l1(b[3], b[4], b[5], width);
+ if (y + 2 < (unsigned)height)
+ vertical_compose_h1(b[2], b[3], b[4], width);
+ if (y + 1 < (unsigned)height)
+ vertical_compose_l0(b[1], b[2], b[3], width);
+ if (y + 0 < (unsigned)height)
+ vertical_compose_h0(b[0], b[1], b[2], width);
+
+ if (y - 1 < (unsigned)height)
+ d->horizontal_compose(b[0], d->temp, width);
+ if (y + 0 < (unsigned)height)
+ d->horizontal_compose(b[1], d->temp, width);
+
+ for (i = 0; i < 4; i++)
+ cs->b[i] = b[i + 2];
+ cs->y += 2;
+}
+
+static void spatial_compose97i_init2(DWTCompose *cs, IDWTELEM *buffer,
+ int height, int stride)
+{
+ cs->b[0] = buffer + mirror(-3 - 1, height - 1) * stride;
+ cs->b[1] = buffer + mirror(-3, height - 1) * stride;
+ cs->b[2] = buffer + mirror(-3 + 1, height - 1) * stride;
+ cs->b[3] = buffer + mirror(-3 + 2, height - 1) * stride;
+ cs->y = -3;
+}
+
+static void spatial_compose53i_init2(DWTCompose *cs, IDWTELEM *buffer,
+ int height, int stride)
+{
+ cs->b[0] = buffer + mirror(-1 - 1, height - 1) * stride;
+ cs->b[1] = buffer + mirror(-1, height - 1) * stride;
+ cs->y = -1;
+}
+
+static void spatial_compose_dd97i_init(DWTCompose *cs, IDWTELEM *buffer,
+ int height, int stride)
+{
+ cs->b[0] = buffer + av_clip(-5 - 1, 0, height - 2) * stride;
+ cs->b[1] = buffer + av_clip(-5, 1, height - 1) * stride;
+ cs->b[2] = buffer + av_clip(-5 + 1, 0, height - 2) * stride;
+ cs->b[3] = buffer + av_clip(-5 + 2, 1, height - 1) * stride;
+ cs->b[4] = buffer + av_clip(-5 + 3, 0, height - 2) * stride;
+ cs->b[5] = buffer + av_clip(-5 + 4, 1, height - 1) * stride;
+ cs->y = -5;
+}
+
+static void spatial_compose_dd137i_init(DWTCompose *cs, IDWTELEM *buffer,
+ int height, int stride)
+{
+ cs->b[0] = buffer + av_clip(-5 - 1, 0, height - 2) * stride;
+ cs->b[1] = buffer + av_clip(-5, 1, height - 1) * stride;
+ cs->b[2] = buffer + av_clip(-5 + 1, 0, height - 2) * stride;
+ cs->b[3] = buffer + av_clip(-5 + 2, 1, height - 1) * stride;
+ cs->b[4] = buffer + av_clip(-5 + 3, 0, height - 2) * stride;
+ cs->b[5] = buffer + av_clip(-5 + 4, 1, height - 1) * stride;
+ cs->b[6] = buffer + av_clip(-5 + 5, 0, height - 2) * stride;
+ cs->b[7] = buffer + av_clip(-5 + 6, 1, height - 1) * stride;
+ cs->y = -5;
+}
+
+int ff_spatial_idwt_init2(DWTContext *d, IDWTELEM *buffer, int width,
+ int height, int stride, enum dwt_type type,
+ int decomposition_count, IDWTELEM *temp)
+{
+ int level;
+
+ d->buffer = buffer;
+ d->width = width;
+ d->height = height;
+ d->stride = stride;
+ d->decomposition_count = decomposition_count;
+ d->temp = temp + 8;
+
+ for (level = decomposition_count - 1; level >= 0; level--) {
+ int hl = height >> level;
+ int stride_l = stride << level;
+
+ switch (type) {
+ case DWT_DIRAC_DD9_7:
+ spatial_compose_dd97i_init(d->cs + level, buffer, hl, stride_l);
+ break;
+ case DWT_DIRAC_LEGALL5_3:
+ spatial_compose53i_init2(d->cs + level, buffer, hl, stride_l);
+ break;
+ case DWT_DIRAC_DD13_7:
+ spatial_compose_dd137i_init(d->cs + level, buffer, hl, stride_l);
+ break;
+ case DWT_DIRAC_HAAR0:
+ case DWT_DIRAC_HAAR1:
+ d->cs[level].y = 1;
+ break;
+ case DWT_DIRAC_DAUB9_7:
+ spatial_compose97i_init2(d->cs + level, buffer, hl, stride_l);
+ break;
+ default:
+ d->cs[level].y = 0;
+ break;
+ }
+ }
+
+ switch (type) {
+ case DWT_DIRAC_DD9_7:
+ d->spatial_compose = spatial_compose_dd97i_dy;
+ d->vertical_compose_l0 = vertical_compose53iL0;
+ d->vertical_compose_h0 = vertical_compose_dd97iH0;
+ d->horizontal_compose = horizontal_compose_dd97i;
+ d->support = 7;
+ break;
+ case DWT_DIRAC_LEGALL5_3:
+ d->spatial_compose = spatial_compose_dirac53i_dy;
+ d->vertical_compose_l0 = vertical_compose53iL0;
+ d->vertical_compose_h0 = vertical_compose_dirac53iH0;
+ d->horizontal_compose = horizontal_compose_dirac53i;
+ d->support = 3;
+ break;
+ case DWT_DIRAC_DD13_7:
+ d->spatial_compose = spatial_compose_dd137i_dy;
+ d->vertical_compose_l0 = vertical_compose_dd137iL0;
+ d->vertical_compose_h0 = vertical_compose_dd97iH0;
+ d->horizontal_compose = horizontal_compose_dd137i;
+ d->support = 7;
+ break;
+ case DWT_DIRAC_HAAR0:
+ case DWT_DIRAC_HAAR1:
+ d->spatial_compose = spatial_compose_haari_dy;
+ d->vertical_compose = vertical_compose_haar;
+ if (type == DWT_DIRAC_HAAR0)
+ d->horizontal_compose = horizontal_compose_haar0i;
+ else
+ d->horizontal_compose = horizontal_compose_haar1i;
+ d->support = 1;
+ break;
+ case DWT_DIRAC_FIDELITY:
+ d->spatial_compose = spatial_compose_fidelity;
+ d->vertical_compose_l0 = vertical_compose_fidelityiL0;
+ d->vertical_compose_h0 = vertical_compose_fidelityiH0;
+ d->horizontal_compose = horizontal_compose_fidelityi;
+ break;
+ case DWT_DIRAC_DAUB9_7:
+ d->spatial_compose = spatial_compose_daub97i_dy;
+ d->vertical_compose_l0 = vertical_compose_daub97iL0;
+ d->vertical_compose_h0 = vertical_compose_daub97iH0;
+ d->vertical_compose_l1 = vertical_compose_daub97iL1;
+ d->vertical_compose_h1 = vertical_compose_daub97iH1;
+ d->horizontal_compose = horizontal_compose_daub97i;
+ d->support = 5;
+ break;
+ default:
+ av_log(NULL, AV_LOG_ERROR, "Unknown wavelet type %d\n", type);
+ return AVERROR_OPTION_NOT_FOUND;
+ }
+
+ return 0;
+}
+
+void ff_spatial_idwt_slice2(DWTContext *d, int y)
+{
+ int level, support = d->support;
+
+ for (level = d->decomposition_count - 1; level >= 0; level--) {
+ int wl = d->width >> level;
+ int hl = d->height >> level;
+ int stride_l = d->stride << level;
+
+ while (d->cs[level].y <= FFMIN((y >> level) + support, hl))
+ d->spatial_compose(d, level, wl, hl, stride_l);
+ }
+}
+
+int ff_spatial_idwt2(IDWTELEM *buffer, int width, int height, int stride,
+ enum dwt_type type, int decomposition_count,
+ IDWTELEM *temp)
+{
+ DWTContext d;
+ int y;
+ int ret;
+
+ if (ret = ff_spatial_idwt_init2(&d, buffer, width, height, stride, type,
+ decomposition_count, temp))
+ return ret;
+
+ for (y = 0; y < d.height; y += 4)
+ ff_spatial_idwt_slice2(&d, y);
+
+ return 0;
+}
diff --git a/libavcodec/dirac_dwt.h b/libavcodec/dirac_dwt.h
new file mode 100644
index 0000000..ff0e4ba
--- /dev/null
+++ b/libavcodec/dirac_dwt.h
@@ -0,0 +1,74 @@
+#include "dwt.h"
+
+enum dwt_type {
+ DWT_SNOW_DAUB9_7,
+ DWT_SNOW_LEGALL5_3,
+ DWT_DIRAC_DD9_7,
+ DWT_DIRAC_LEGALL5_3,
+ DWT_DIRAC_DD13_7,
+ DWT_DIRAC_HAAR0,
+ DWT_DIRAC_HAAR1,
+ DWT_DIRAC_FIDELITY,
+ DWT_DIRAC_DAUB9_7,
+ DWT_NUM_TYPES
+};
+
+
+// Possible prototypes for vertical_compose functions
+typedef void (*vertical_compose_2tap)(IDWTELEM *b0, IDWTELEM *b1, int width);
+typedef void (*vertical_compose_3tap)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
+ int width);
+typedef void (*vertical_compose_5tap)(IDWTELEM *b0, IDWTELEM *b1, IDWTELEM *b2,
+ IDWTELEM *b3, IDWTELEM *b4, int width);
+typedef void (*vertical_compose_9tap)(IDWTELEM *dst, IDWTELEM *b[8],
+ int width);
+
+// -1 if an error occurred, e.g. the dwt_type isn't recognized
+int ff_spatial_idwt_init2(DWTContext *d, IDWTELEM *buffer, int width,
+ int height, int stride, enum dwt_type type,
+ int decomposition_count, IDWTELEM *temp);
+
+int ff_spatial_idwt2(IDWTELEM *buffer, int width, int height, int stride,
+ enum dwt_type type, int decomposition_count,
+ IDWTELEM *temp);
+
+void ff_spatial_idwt_slice2(DWTContext *d, int y);
+
+// shared stuff for simd optimiztions
+#define COMPOSE_53iL0(b0, b1, b2) \
+ (b1 - ((b0 + b2 + 2) >> 2))
+
+#define COMPOSE_DIRAC53iH0(b0, b1, b2) \
+ (b1 + ((b0 + b2 + 1) >> 1))
+
+#define COMPOSE_DD97iH0(b0, b1, b2, b3, b4) \
+ (b2 + ((-b0 + 9 * b1 + 9 * b3 - b4 + 8) >> 4))
+
+#define COMPOSE_DD137iL0(b0, b1, b2, b3, b4) \
+ (b2 - ((-b0 + 9 * b1 + 9 * b3 - b4 + 16) >> 5))
+
+#define COMPOSE_HAARiL0(b0, b1) \
+ (b0 - ((b1 + 1) >> 1))
+
+#define COMPOSE_HAARiH0(b0, b1) \
+ (b0 + b1)
+
+#define COMPOSE_FIDELITYiL0(b0, b1, b2, b3, b4, b5, b6, b7, b8) \
+ (b4 - ((-8 * (b0 + b8) + 21 * (b1 + b7) - 46 * (b2 + b6) + \
+ 161 * (b3 + b5) + 128) >> 8))
+
+#define COMPOSE_FIDELITYiH0(b0, b1, b2, b3, b4, b5, b6, b7, b8) \
+ (b4 + ((-2 * (b0 + b8) + 10 * (b1 + b7) - 25 * (b2 + b6) + \
+ 81 * (b3 + b5) + 128) >> 8))
+
+#define COMPOSE_DAUB97iL1(b0, b1, b2) \
+ (b1 - ((1817 * (b0 + b2) + 2048) >> 12))
+
+#define COMPOSE_DAUB97iH1(b0, b1, b2) \
+ (b1 - ((113 * (b0 + b2) + 64) >> 7))
+
+#define COMPOSE_DAUB97iL0(b0, b1, b2) \
+ (b1 + ((217 * (b0 + b2) + 2048) >> 12))
+
+#define COMPOSE_DAUB97iH0(b0, b1, b2) \
+ (b1 + ((6497 * (b0 + b2) + 2048) >> 12))
diff --git a/libavcodec/dwt.h b/libavcodec/dwt.h
index b41e128..8cf51f0 100644
--- a/libavcodec/dwt.h
+++ b/libavcodec/dwt.h
@@ -26,7 +26,12 @@
typedef int DWTELEM;
typedef short IDWTELEM;
+#define MAX_DWT_SUPPORT 8
+#define MAX_DECOMPOSITIONS 8
+
typedef struct {
+ IDWTELEM *b[MAX_DWT_SUPPORT];
+
IDWTELEM *b0;
IDWTELEM *b1;
IDWTELEM *b2;
@@ -57,8 +62,6 @@ typedef struct DWTContext {
int add, uint8_t *dst8);
} DWTContext;
-#define MAX_DECOMPOSITIONS 8
-
#define DWT_97 0
#define DWT_53 1
--
1.7.10.4
_______________________________________________
libav-devel mailing list
libav-devel@libav.org
https://lists.libav.org/mailman/listinfo/libav-devel
