This is also faster.
Unified · Split
| @@ -104,56 +104,74 @@ typedef struct ThreadData { · 103 unmodified lines | ||
| 104 | 104 | int omin[4]; |
| 105 | 105 | } ThreadData; |
| 106 | 106 | |
| 107 | −#define LOAD_COMMON \ | |
| 108 | − ColorLevelsContext *s = ctx->priv; \ | |
| 109 | − const ThreadData *td = arg; \ | |
| 110 | − const int process_h = td->h; \ | |
| 111 | − const int slice_start = (process_h * jobnr ) / nb_jobs; \ | |
| 112 | − const int slice_end = (process_h * (jobnr+1)) / nb_jobs; \ | |
| 113 | − const uint8_t *srcrow = td->srcrow; \ | |
| 114 | − uint8_t *dstrow = td->dstrow; \ | |
| 115 | − const int step = s->step; | |
| 107 | +#define DO_COMMON(type, clip) \ | |
| 108 | + ColorLevelsContext *s = ctx->priv; \ | |
| 109 | + const ThreadData *td = arg; \ | |
| 110 | + const int linesize = s->linesize; \ | |
| 111 | + const int step = s->step; \ | |
| 112 | + const int process_h = td->h; \ | |
| 113 | + const int slice_start = (process_h * jobnr ) / nb_jobs; \ | |
| 114 | + const int slice_end = (process_h * (jobnr+1)) / nb_jobs; \ | |
| 115 | + const int src_linesize = td->src_linesize / sizeof(type); \ | |
| 116 | + const int dst_linesize = td->dst_linesize / sizeof(type); \ | |
| 117 | + const type *srcrow = (const type *)td->srcrow + src_linesize * slice_start; \ | |
| 118 | + type *dstrow = (type *)td->dstrow + dst_linesize * slice_start; \ | |
| 119 | + const uint8_t offset_r = s->rgba_map[R]; \ | |
| 120 | + const uint8_t offset_g = s->rgba_map[G]; \ | |
| 121 | + const uint8_t offset_b = s->rgba_map[B]; \ | |
| 122 | + const uint8_t offset_a = s->rgba_map[A]; \ | |
| 123 | + const int imin_r = td->imin[R]; \ | |
| 124 | + const int imin_g = td->imin[G]; \ | |
| 125 | + const int imin_b = td->imin[B]; \ | |
| 126 | + const int imin_a = td->imin[A]; \ | |
| 127 | + const int omin_r = td->omin[R]; \ | |
| 128 | + const int omin_g = td->omin[G]; \ | |
| 129 | + const int omin_b = td->omin[B]; \ | |
| 130 | + const int omin_a = td->omin[A]; \ | |
| 131 | + const float coeff_r = td->coeff[R]; \ | |
| 132 | + const float coeff_g = td->coeff[G]; \ | |
| 133 | + const float coeff_b = td->coeff[B]; \ | |
| 134 | + const float coeff_a = td->coeff[A]; \ | |
| 135 | + const type *src_r = srcrow + offset_r; \ | |
| 136 | + const type *src_g = srcrow + offset_g; \ | |
| 137 | + const type *src_b = srcrow + offset_b; \ | |
| 138 | + const type *src_a = srcrow + offset_a; \ | |
| 139 | + type *dst_r = dstrow + offset_r; \ | |
| 140 | + type *dst_g = dstrow + offset_g; \ | |
| 141 | + type *dst_b = dstrow + offset_b; \ | |
| 142 | + type *dst_a = dstrow + offset_a; \ | |
| 143 | + \ | |
| 144 | + for (int y = slice_start; y < slice_end; y++) { \ | |
| 145 | + for (int x = 0; x < linesize; x += step) { \ | |
| 146 | + dst_r[x] = clip((src_r[x] - imin_r) * coeff_r + omin_r); \ | |
| 147 | + dst_g[x] = clip((src_g[x] - imin_g) * coeff_g + omin_g); \ | |
| 148 | + dst_b[x] = clip((src_b[x] - imin_b) * coeff_b + omin_b); \ | |
| 149 | + } \ | |
| 150 | + \ | |
| 151 | + for (int x = 0; x < linesize && s->nb_comp == 4; x += step) \ | |
| 152 | + dst_a[x] = clip((src_a[x] - imin_a) * coeff_a + omin_a); \ | |
| 153 | + \ | |
| 154 | + src_r += src_linesize; \ | |
| 155 | + src_g += src_linesize; \ | |
| 156 | + src_b += src_linesize; \ | |
| 157 | + src_a += src_linesize; \ | |
| 158 | + \ | |
| 159 | + dst_r += dst_linesize; \ | |
| 160 | + dst_g += dst_linesize; \ | |
| 161 | + dst_b += dst_linesize; \ | |
| 162 | + dst_a += dst_linesize; \ | |
| 163 | + } | |
| 116 | 164 | |
| 117 | 165 | static int colorlevels_slice_8(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) |
| 118 | 166 | { |
| 119 | − LOAD_COMMON | |
| 120 | − | |
| 121 | − for (int comp = 0; comp < s->nb_comp; comp++) { | |
| 122 | − const uint8_t offset = s->rgba_map[comp]; | |
| 123 | − const int imin = td->imin[comp]; | |
| 124 | − const int omin = td->omin[comp]; | |
| 125 | − const float coeff = td->coeff[comp]; | |
| 126 | − | |
| 127 | − for (int y = slice_start; y < slice_end; y++) { | |
| 128 | − const uint8_t *src = srcrow + y * td->src_linesize; | |
| 129 | − uint8_t *dst = dstrow + y * td->dst_linesize; | |
| 130 | − | |
| 131 | − for (int x = 0; x < s->linesize; x += step) | |
| 132 | − dst[x + offset] = av_clip_uint8((src[x + offset] - imin) * coeff + omin); | |
| 133 | − } | |
| 134 | − } | |
| 167 | + DO_COMMON(uint8_t, av_clip_uint8) | |
| 135 | 168 | |
| 136 | 169 | return 0; |
| 137 | 170 | } |
| 138 | 171 | |
| 139 | 172 | static int colorlevels_slice_16(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) |
| 140 | 173 | { |
| 141 | − LOAD_COMMON | |
| 142 | − | |
| 143 | − for (int comp = 0; comp < s->nb_comp; comp++) { | |
| 144 | − const uint8_t offset = s->rgba_map[comp]; | |
| 145 | − const int imin = td->imin[comp]; | |
| 146 | − const int omin = td->omin[comp]; | |
| 147 | − const float coeff = td->coeff[comp]; | |
| 148 | − | |
| 149 | − for (int y = slice_start; y < slice_end; y++) { | |
| 150 | − const uint16_t *src = (const uint16_t *)(srcrow + y * td->src_linesize); | |
| 151 | − uint16_t *dst = (uint16_t *)(dstrow + y * td->dst_linesize); | |
| 152 | − | |
| 153 | − for (int x = 0; x < s->linesize; x += step) | |
| 154 | − dst[x + offset] = av_clip_uint16((src[x + offset] - imin) * coeff + omin); | |
| 155 | − } | |
| 156 | − } | |
| 174 | + DO_COMMON(uint16_t, av_clip_uint16) | |
| 157 | 175 | |
| 158 | 176 | return 0; |
| 159 | 177 | } |