t1 = dc[i*4+1] + dc[i*4+2];
t2 = dc[i*4+1] - dc[i*4+2];
t3 = dc[i*4+0] - dc[i*4+3] + 3; // rounding
-
- *block[i][0] = (t0 + t1) >> 3;
- *block[i][1] = (t3 + t2) >> 3;
- *block[i][2] = (t0 - t1) >> 3;
- *block[i][3] = (t3 - t2) >> 3;
+ dc[i*4+0] = 0;
+ dc[i*4+1] = 0;
+ dc[i*4+2] = 0;
+ dc[i*4+3] = 0;
+
+ block[i][0][0] = (t0 + t1) >> 3;
+ block[i][1][0] = (t3 + t2) >> 3;
+ block[i][2][0] = (t0 - t1) >> 3;
+ block[i][3][0] = (t3 - t2) >> 3;
}
}
+static void vp8_luma_dc_wht_dc_c(DCTELEM block[4][4][16], DCTELEM dc[16])
+{
+ int i, val = (dc[0] + 3) >> 3;
+ dc[0] = 0;
+
+ for (i = 0; i < 4; i++) {
+ block[i][0][0] = val;
+ block[i][1][0] = val;
+ block[i][2][0] = val;
+ block[i][3][0] = val;
+ }
+}
#define MUL_20091(a) ((((a)*20091) >> 16) + (a))
#define MUL_35468(a) (((a)*35468) >> 16)
static void vp8_idct_add_c(uint8_t *dst, DCTELEM block[16], int stride)
{
int i, t0, t1, t2, t3;
+ uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
DCTELEM tmp[16];
for (i = 0; i < 4; i++) {
t1 = block[0*4+i] - block[2*4+i];
t2 = MUL_35468(block[1*4+i]) - MUL_20091(block[3*4+i]);
t3 = MUL_20091(block[1*4+i]) + MUL_35468(block[3*4+i]);
+ block[0*4+i] = 0;
+ block[1*4+i] = 0;
+ block[2*4+i] = 0;
+ block[3*4+i] = 0;
tmp[i*4+0] = t0 + t3;
tmp[i*4+1] = t1 + t2;
t2 = MUL_35468(tmp[1*4+i]) - MUL_20091(tmp[3*4+i]);
t3 = MUL_20091(tmp[1*4+i]) + MUL_35468(tmp[3*4+i]);
- dst[0] = av_clip_uint8(dst[0] + ((t0 + t3 + 4) >> 3));
- dst[1] = av_clip_uint8(dst[1] + ((t1 + t2 + 4) >> 3));
- dst[2] = av_clip_uint8(dst[2] + ((t1 - t2 + 4) >> 3));
- dst[3] = av_clip_uint8(dst[3] + ((t0 - t3 + 4) >> 3));
+ dst[0] = cm[dst[0] + ((t0 + t3 + 4) >> 3)];
+ dst[1] = cm[dst[1] + ((t1 + t2 + 4) >> 3)];
+ dst[2] = cm[dst[2] + ((t1 - t2 + 4) >> 3)];
+ dst[3] = cm[dst[3] + ((t0 - t3 + 4) >> 3)];
dst += stride;
}
}
static void vp8_idct_dc_add_c(uint8_t *dst, DCTELEM block[16], int stride)
{
int i, dc = (block[0] + 4) >> 3;
+ uint8_t *cm = ff_cropTbl + MAX_NEG_CROP + dc;
+ block[0] = 0;
for (i = 0; i < 4; i++) {
- dst[0] = av_clip_uint8(dst[0] + dc);
- dst[1] = av_clip_uint8(dst[1] + dc);
- dst[2] = av_clip_uint8(dst[2] + dc);
- dst[3] = av_clip_uint8(dst[3] + dc);
+ dst[0] = cm[dst[0]];
+ dst[1] = cm[dst[1]];
+ dst[2] = cm[dst[2]];
+ dst[3] = cm[dst[3]];
dst += stride;
}
}
+static void vp8_idct_dc_add4uv_c(uint8_t *dst, DCTELEM block[4][16], int stride)
+{
+ vp8_idct_dc_add_c(dst+stride*0+0, block[0], stride);
+ vp8_idct_dc_add_c(dst+stride*0+4, block[1], stride);
+ vp8_idct_dc_add_c(dst+stride*4+0, block[2], stride);
+ vp8_idct_dc_add_c(dst+stride*4+4, block[3], stride);
+}
+
+static void vp8_idct_dc_add4y_c(uint8_t *dst, DCTELEM block[4][16], int stride)
+{
+ vp8_idct_dc_add_c(dst+ 0, block[0], stride);
+ vp8_idct_dc_add_c(dst+ 4, block[1], stride);
+ vp8_idct_dc_add_c(dst+ 8, block[2], stride);
+ vp8_idct_dc_add_c(dst+12, block[3], stride);
+}
// because I like only having two parameters to pass functions...
#define LOAD_PIXELS\
int av_unused q2 = p[ 2*stride];\
int av_unused q3 = p[ 3*stride];
+#define clip_int8(n) (cm[n+0x80]-0x80)
+
static av_always_inline void filter_common(uint8_t *p, int stride, int is4tap)
{
LOAD_PIXELS
int a, f1, f2;
+ uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
a = 3*(q0 - p0);
if (is4tap)
- a += av_clip_int8(p1 - q1);
+ a += clip_int8(p1 - q1);
- a = av_clip_int8(a);
+ a = clip_int8(a);
// We deviate from the spec here with c(a+3) >> 3
// since that's what libvpx does.
// Despite what the spec says, we do need to clamp here to
// be bitexact with libvpx.
- p[-1*stride] = av_clip_uint8(p0 + f2);
- p[ 0*stride] = av_clip_uint8(q0 - f1);
+ p[-1*stride] = cm[p0 + f2];
+ p[ 0*stride] = cm[q0 - f1];
// only used for _inner on blocks without high edge variance
if (!is4tap) {
a = (f1+1)>>1;
- p[-2*stride] = av_clip_uint8(p1 + a);
- p[ 1*stride] = av_clip_uint8(q1 - a);
+ p[-2*stride] = cm[p1 + a];
+ p[ 1*stride] = cm[q1 - a];
}
}
static av_always_inline int normal_limit(uint8_t *p, int stride, int E, int I)
{
LOAD_PIXELS
- return simple_limit(p, stride, 2*E+I)
+ return simple_limit(p, stride, E)
&& FFABS(p3-p2) <= I && FFABS(p2-p1) <= I && FFABS(p1-p0) <= I
&& FFABS(q3-q2) <= I && FFABS(q2-q1) <= I && FFABS(q1-q0) <= I;
}
static av_always_inline void filter_mbedge(uint8_t *p, int stride)
{
int a0, a1, a2, w;
+ uint8_t *cm = ff_cropTbl + MAX_NEG_CROP;
LOAD_PIXELS
- w = av_clip_int8(p1-q1);
- w = av_clip_int8(w + 3*(q0-p0));
+ w = clip_int8(p1-q1);
+ w = clip_int8(w + 3*(q0-p0));
a0 = (27*w + 63) >> 7;
a1 = (18*w + 63) >> 7;
a2 = ( 9*w + 63) >> 7;
- p[-3*stride] = av_clip_uint8(p2 + a2);
- p[-2*stride] = av_clip_uint8(p1 + a1);
- p[-1*stride] = av_clip_uint8(p0 + a0);
- p[ 0*stride] = av_clip_uint8(q0 - a0);
- p[ 1*stride] = av_clip_uint8(q1 - a1);
- p[ 2*stride] = av_clip_uint8(q2 - a2);
+ p[-3*stride] = cm[p2 + a2];
+ p[-2*stride] = cm[p1 + a1];
+ p[-1*stride] = cm[p0 + a0];
+ p[ 0*stride] = cm[q0 - a0];
+ p[ 1*stride] = cm[q1 - a1];
+ p[ 2*stride] = cm[q2 - a2];
}
-#define LOOP_FILTER(dir, size, stridea, strideb) \
-static void vp8_ ## dir ## _loop_filter ## size ## _c(uint8_t *dst, int stride,\
+#define LOOP_FILTER(dir, size, stridea, strideb, maybe_inline) \
+static maybe_inline void vp8_ ## dir ## _loop_filter ## size ## _c(uint8_t *dst, int stride,\
int flim_E, int flim_I, int hev_thresh)\
{\
int i;\
}\
}\
\
-static void vp8_ ## dir ## _loop_filter ## size ## _inner_c(uint8_t *dst, int stride,\
+static maybe_inline void vp8_ ## dir ## _loop_filter ## size ## _inner_c(uint8_t *dst, int stride,\
int flim_E, int flim_I, int hev_thresh)\
{\
- int i, hv;\
+ int i;\
\
for (i = 0; i < size; i++)\
if (normal_limit(dst+i*stridea, strideb, flim_E, flim_I)) {\
- hv = hev(dst+i*stridea, strideb, hev_thresh);\
- filter_common(dst+i*stridea, strideb, hv);\
+ int hv = hev(dst+i*stridea, strideb, hev_thresh);\
+ if (hv) \
+ filter_common(dst+i*stridea, strideb, 1);\
+ else \
+ filter_common(dst+i*stridea, strideb, 0);\
}\
}
-LOOP_FILTER(v, 16, 1, stride)
-LOOP_FILTER(h, 16, stride, 1)
-LOOP_FILTER(v, 8, 1, stride)
-LOOP_FILTER(h, 8, stride, 1)
+LOOP_FILTER(v, 16, 1, stride,)
+LOOP_FILTER(h, 16, stride, 1,)
+
+#define UV_LOOP_FILTER(dir, stridea, strideb) \
+LOOP_FILTER(dir, 8, stridea, strideb, av_always_inline) \
+static void vp8_ ## dir ## _loop_filter8uv_c(uint8_t *dstU, uint8_t *dstV, int stride,\
+ int fE, int fI, int hev_thresh)\
+{\
+ vp8_ ## dir ## _loop_filter8_c(dstU, stride, fE, fI, hev_thresh);\
+ vp8_ ## dir ## _loop_filter8_c(dstV, stride, fE, fI, hev_thresh);\
+}\
+static void vp8_ ## dir ## _loop_filter8uv_inner_c(uint8_t *dstU, uint8_t *dstV, int stride,\
+ int fE, int fI, int hev_thresh)\
+{\
+ vp8_ ## dir ## _loop_filter8_inner_c(dstU, stride, fE, fI, hev_thresh);\
+ vp8_ ## dir ## _loop_filter8_inner_c(dstV, stride, fE, fI, hev_thresh);\
+}
+
+UV_LOOP_FILTER(v, 1, stride)
+UV_LOOP_FILTER(h, stride, 1)
static void vp8_v_loop_filter_simple_c(uint8_t *dst, int stride, int flim)
{
PUT_PIXELS(4)
#define FILTER_6TAP(src, F, stride) \
- av_clip_uint8((F[2]*src[x+0*stride] - F[1]*src[x-1*stride] + F[0]*src[x-2*stride] + \
- F[3]*src[x+1*stride] - F[4]*src[x+2*stride] + F[5]*src[x+3*stride] + 64) >> 7)
+ cm[(F[2]*src[x+0*stride] - F[1]*src[x-1*stride] + F[0]*src[x-2*stride] + \
+ F[3]*src[x+1*stride] - F[4]*src[x+2*stride] + F[5]*src[x+3*stride] + 64) >> 7]
#define FILTER_4TAP(src, F, stride) \
- av_clip_uint8((F[2]*src[x+0*stride] - F[1]*src[x-1*stride] + \
- F[3]*src[x+1*stride] - F[4]*src[x+2*stride] + 64) >> 7)
+ cm[(F[2]*src[x+0*stride] - F[1]*src[x-1*stride] + \
+ F[3]*src[x+1*stride] - F[4]*src[x+2*stride] + 64) >> 7]
#define VP8_EPEL_H(SIZE, FILTER, FILTERNAME) \
static void put_vp8_epel ## SIZE ## _ ## FILTERNAME ## _c(uint8_t *dst, int dststride, uint8_t *src, int srcstride, int h, int mx, int my) \
{ \
const uint8_t *filter = subpel_filters[mx-1]; \
+ uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; \
int x, y; \
\
for (y = 0; y < h; y++) { \
static void put_vp8_epel ## SIZE ## _ ## FILTERNAME ## _c(uint8_t *dst, int dststride, uint8_t *src, int srcstride, int h, int mx, int my) \
{ \
const uint8_t *filter = subpel_filters[my-1]; \
+ uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; \
int x, y; \
\
for (y = 0; y < h; y++) { \
static void put_vp8_epel ## SIZE ## _ ## FILTERNAME ## _c(uint8_t *dst, int dststride, uint8_t *src, int srcstride, int h, int mx, int my) \
{ \
const uint8_t *filter = subpel_filters[mx-1]; \
+ uint8_t *cm = ff_cropTbl + MAX_NEG_CROP; \
int x, y; \
uint8_t tmp_array[(2*SIZE+5)*SIZE]; \
uint8_t *tmp = tmp_array; \
av_cold void ff_vp8dsp_init(VP8DSPContext *dsp)
{
- dsp->vp8_luma_dc_wht = vp8_luma_dc_wht_c;
- dsp->vp8_idct_add = vp8_idct_add_c;
- dsp->vp8_idct_dc_add = vp8_idct_dc_add_c;
-
- dsp->vp8_v_loop_filter16 = vp8_v_loop_filter16_c;
- dsp->vp8_h_loop_filter16 = vp8_h_loop_filter16_c;
- dsp->vp8_v_loop_filter8 = vp8_v_loop_filter8_c;
- dsp->vp8_h_loop_filter8 = vp8_h_loop_filter8_c;
-
- dsp->vp8_v_loop_filter16_inner = vp8_v_loop_filter16_inner_c;
- dsp->vp8_h_loop_filter16_inner = vp8_h_loop_filter16_inner_c;
- dsp->vp8_v_loop_filter8_inner = vp8_v_loop_filter8_inner_c;
- dsp->vp8_h_loop_filter8_inner = vp8_h_loop_filter8_inner_c;
+ dsp->vp8_luma_dc_wht = vp8_luma_dc_wht_c;
+ dsp->vp8_luma_dc_wht_dc = vp8_luma_dc_wht_dc_c;
+ dsp->vp8_idct_add = vp8_idct_add_c;
+ dsp->vp8_idct_dc_add = vp8_idct_dc_add_c;
+ dsp->vp8_idct_dc_add4y = vp8_idct_dc_add4y_c;
+ dsp->vp8_idct_dc_add4uv = vp8_idct_dc_add4uv_c;
+
+ dsp->vp8_v_loop_filter16y = vp8_v_loop_filter16_c;
+ dsp->vp8_h_loop_filter16y = vp8_h_loop_filter16_c;
+ dsp->vp8_v_loop_filter8uv = vp8_v_loop_filter8uv_c;
+ dsp->vp8_h_loop_filter8uv = vp8_h_loop_filter8uv_c;
+
+ dsp->vp8_v_loop_filter16y_inner = vp8_v_loop_filter16_inner_c;
+ dsp->vp8_h_loop_filter16y_inner = vp8_h_loop_filter16_inner_c;
+ dsp->vp8_v_loop_filter8uv_inner = vp8_v_loop_filter8uv_inner_c;
+ dsp->vp8_h_loop_filter8uv_inner = vp8_h_loop_filter8uv_inner_c;
dsp->vp8_v_loop_filter_simple = vp8_v_loop_filter_simple_c;
dsp->vp8_h_loop_filter_simple = vp8_h_loop_filter_simple_c;
VP8_BILINEAR_MC_FUNC(0, 16);
VP8_BILINEAR_MC_FUNC(1, 8);
VP8_BILINEAR_MC_FUNC(2, 4);
+
+ if (HAVE_MMX)
+ ff_vp8dsp_init_x86(dsp);
+ if (HAVE_ALTIVEC)
+ ff_vp8dsp_init_altivec(dsp);
}
projects / ffmpeg / blobdiff