00001
00002
00003
00004
00005
00006
00007
00008
00009
00010
00011
00012
00013
00014
00015
00016
00017
00018
00019
00020
00021 #include "dsputil.h"
00022 #include "diracdsp.h"
00023 #include "libavcodec/x86/diracdsp_mmx.h"
00024
00025 #define FILTER(src, stride) \
00026 ((21*((src)[ 0*stride] + (src)[1*stride]) \
00027 -7*((src)[-1*stride] + (src)[2*stride]) \
00028 +3*((src)[-2*stride] + (src)[3*stride]) \
00029 -1*((src)[-3*stride] + (src)[4*stride]) + 16) >> 5)
00030
00031 static void dirac_hpel_filter(uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, const uint8_t *src,
00032 int stride, int width, int height)
00033 {
00034 int x, y;
00035
00036 for (y = 0; y < height; y++) {
00037 for (x = -3; x < width+5; x++)
00038 dstv[x] = av_clip_uint8(FILTER(src+x, stride));
00039
00040 for (x = 0; x < width; x++)
00041 dstc[x] = av_clip_uint8(FILTER(dstv+x, 1));
00042
00043 for (x = 0; x < width; x++)
00044 dsth[x] = av_clip_uint8(FILTER(src+x, 1));
00045
00046 src += stride;
00047 dsth += stride;
00048 dstv += stride;
00049 dstc += stride;
00050 }
00051 }
00052
00053 #define PIXOP_BILINEAR(PFX, OP, WIDTH) \
00054 static void ff_ ## PFX ## _dirac_pixels ## WIDTH ## _bilinear_c(uint8_t *dst, const uint8_t *src[5], int stride, int h) \
00055 { \
00056 int x; \
00057 const uint8_t *s0 = src[0]; \
00058 const uint8_t *s1 = src[1]; \
00059 const uint8_t *s2 = src[2]; \
00060 const uint8_t *s3 = src[3]; \
00061 const uint8_t *w = src[4]; \
00062 \
00063 while (h--) { \
00064 for (x = 0; x < WIDTH; x++) { \
00065 OP(dst[x], (s0[x]*w[0] + s1[x]*w[1] + s2[x]*w[2] + s3[x]*w[3] + 8) >> 4); \
00066 } \
00067 \
00068 dst += stride; \
00069 s0 += stride; \
00070 s1 += stride; \
00071 s2 += stride; \
00072 s3 += stride; \
00073 } \
00074 }
00075
00076 #define OP_PUT(dst, val) (dst) = (val)
00077 #define OP_AVG(dst, val) (dst) = (((dst) + (val) + 1)>>1)
00078
00079 PIXOP_BILINEAR(put, OP_PUT, 8)
00080 PIXOP_BILINEAR(put, OP_PUT, 16)
00081 PIXOP_BILINEAR(put, OP_PUT, 32)
00082 PIXOP_BILINEAR(avg, OP_AVG, 8)
00083 PIXOP_BILINEAR(avg, OP_AVG, 16)
00084 PIXOP_BILINEAR(avg, OP_AVG, 32)
00085
00086 #define op_scale1(x) block[x] = av_clip_uint8( (block[x]*weight + (1<<(log2_denom-1))) >> log2_denom)
00087 #define op_scale2(x) dst[x] = av_clip_uint8( (src[x]*weights + dst[x]*weightd + (1<<(log2_denom-1))) >> log2_denom)
00088
00089 #define DIRAC_WEIGHT(W) \
00090 static void weight_dirac_pixels ## W ## _c(uint8_t *block, int stride, int log2_denom, \
00091 int weight, int h) { \
00092 int x; \
00093 while (h--) { \
00094 for (x = 0; x < W; x++) { \
00095 op_scale1(x); \
00096 op_scale1(x+1); \
00097 } \
00098 block += stride; \
00099 } \
00100 } \
00101 static void biweight_dirac_pixels ## W ## _c(uint8_t *dst, const uint8_t *src, int stride, int log2_denom, \
00102 int weightd, int weights, int h) { \
00103 int x; \
00104 while (h--) { \
00105 for (x = 0; x < W; x++) { \
00106 op_scale2(x); \
00107 op_scale2(x+1); \
00108 } \
00109 dst += stride; \
00110 src += stride; \
00111 } \
00112 }
00113
00114 DIRAC_WEIGHT(8)
00115 DIRAC_WEIGHT(16)
00116 DIRAC_WEIGHT(32)
00117
00118 #define ADD_OBMC(xblen) \
00119 static void add_obmc ## xblen ## _c(uint16_t *dst, const uint8_t *src, int stride, \
00120 const uint8_t *obmc_weight, int yblen) \
00121 { \
00122 int x; \
00123 while (yblen--) { \
00124 for (x = 0; x < xblen; x += 2) { \
00125 dst[x ] += src[x ] * obmc_weight[x ]; \
00126 dst[x+1] += src[x+1] * obmc_weight[x+1]; \
00127 } \
00128 dst += stride; \
00129 src += stride; \
00130 obmc_weight += 32; \
00131 } \
00132 }
00133
00134 ADD_OBMC(8)
00135 ADD_OBMC(16)
00136 ADD_OBMC(32)
00137
00138 static void put_signed_rect_clamped_c(uint8_t *dst, int dst_stride, const int16_t *src, int src_stride, int width, int height)
00139 {
00140 int x, y;
00141 for (y = 0; y < height; y++) {
00142 for (x = 0; x < width; x+=4) {
00143 dst[x ] = av_clip_uint8(src[x ] + 128);
00144 dst[x+1] = av_clip_uint8(src[x+1] + 128);
00145 dst[x+2] = av_clip_uint8(src[x+2] + 128);
00146 dst[x+3] = av_clip_uint8(src[x+3] + 128);
00147 }
00148 dst += dst_stride;
00149 src += src_stride;
00150 }
00151 }
00152
00153 static void add_rect_clamped_c(uint8_t *dst, const uint16_t *src, int stride,
00154 const int16_t *idwt, int idwt_stride,
00155 int width, int height)
00156 {
00157 int x, y;
00158
00159 for (y = 0; y < height; y++) {
00160 for (x = 0; x < width; x+=2) {
00161 dst[x ] = av_clip_uint8(((src[x ]+32)>>6) + idwt[x ]);
00162 dst[x+1] = av_clip_uint8(((src[x+1]+32)>>6) + idwt[x+1]);
00163 }
00164 dst += stride;
00165 src += stride;
00166 idwt += idwt_stride;
00167 }
00168 }
00169
00170 #define PIXFUNC(PFX, WIDTH) \
00171 c->PFX ## _dirac_pixels_tab[WIDTH>>4][0] = ff_ ## PFX ## _dirac_pixels ## WIDTH ## _c; \
00172 c->PFX ## _dirac_pixels_tab[WIDTH>>4][1] = ff_ ## PFX ## _dirac_pixels ## WIDTH ## _l2_c; \
00173 c->PFX ## _dirac_pixels_tab[WIDTH>>4][2] = ff_ ## PFX ## _dirac_pixels ## WIDTH ## _l4_c; \
00174 c->PFX ## _dirac_pixels_tab[WIDTH>>4][3] = ff_ ## PFX ## _dirac_pixels ## WIDTH ## _bilinear_c
00175
00176 void ff_diracdsp_init(DiracDSPContext *c)
00177 {
00178 c->dirac_hpel_filter = dirac_hpel_filter;
00179 c->add_rect_clamped = add_rect_clamped_c;
00180 c->put_signed_rect_clamped = put_signed_rect_clamped_c;
00181
00182 c->add_dirac_obmc[0] = add_obmc8_c;
00183 c->add_dirac_obmc[1] = add_obmc16_c;
00184 c->add_dirac_obmc[2] = add_obmc32_c;
00185
00186 c->weight_dirac_pixels_tab[0] = weight_dirac_pixels8_c;
00187 c->weight_dirac_pixels_tab[1] = weight_dirac_pixels16_c;
00188 c->weight_dirac_pixels_tab[2] = weight_dirac_pixels32_c;
00189 c->biweight_dirac_pixels_tab[0] = biweight_dirac_pixels8_c;
00190 c->biweight_dirac_pixels_tab[1] = biweight_dirac_pixels16_c;
00191 c->biweight_dirac_pixels_tab[2] = biweight_dirac_pixels32_c;
00192
00193 PIXFUNC(put, 8);
00194 PIXFUNC(put, 16);
00195 PIXFUNC(put, 32);
00196 PIXFUNC(avg, 8);
00197 PIXFUNC(avg, 16);
00198 PIXFUNC(avg, 32);
00199
00200 if (HAVE_MMX && HAVE_YASM) ff_diracdsp_init_mmx(c);
00201 }
