src/gallium/auxiliary/vl/vl_median_filter.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2012 Christian König.
   4  * All Rights Reserved.
   5  *
   6  * Permission is hereby granted, free of charge, to any person obtaining a
   7  * copy of this software and associated documentation files (the
   8  * "Software"), to deal in the Software without restriction, including
   9  * without limitation the rights to use, copy, modify, merge, publish,
  10  * distribute, sub license, and/or sell copies of the Software, and to
  11  * permit persons to whom the Software is furnished to do so, subject to
  12  * the following conditions:
  13  *
  14  * The above copyright notice and this permission notice (including the
  15  * next paragraph) shall be included in all copies or substantial portions
  16  * of the Software.
  17  *
  18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
  19  * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
  21  * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
  22  * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
  23  * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
  24  * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25  *
  26  **************************************************************************/
  27
  28 #include "pipe/p_context.h"
  29
  30 #include "tgsi/tgsi_ureg.h"
  31
  32 #include "util/u_draw.h"
  33 #include "util/u_memory.h"
  34 #include "util/u_math.h"
  35
  36 #include "vl_types.h"
  37 #include "vl_vertex_buffers.h"
  38 #include "vl_median_filter.h"
  39
  40 enum VS_OUTPUT
  41 {
  42    VS_O_VPOS = 0,
  43    VS_O_VTEX = 0
  44 };
  45
  46 static void *
  47 create_vert_shader(struct vl_median_filter *filter)
  48 {
  49    struct ureg_program *shader;
  50    struct ureg_src i_vpos;
  51    struct ureg_dst o_vpos, o_vtex;
  52
  53    shader = ureg_create(PIPE_SHADER_VERTEX);
  54    if (!shader)
  55       return NULL;
  56
  57    i_vpos = ureg_DECL_vs_input(shader, 0);
  58    o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
  59    o_vtex = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_VTEX);
  60
  61    ureg_MOV(shader, o_vpos, i_vpos);
  62    ureg_MOV(shader, o_vtex, i_vpos);
  63
  64    ureg_END(shader);
  65
  66    return ureg_create_shader_and_destroy(shader, filter->pipe);
  67 }
  68
  69 static inline bool
  70 is_vec_zero(struct vertex2f v)
  71 {
  72    return v.x == 0.0f && v.y == 0.0f;
  73 }
  74
  75 static void *
  76 create_frag_shader(struct vl_median_filter *filter,
  77                    struct vertex2f *offsets,
  78                    unsigned num_offsets)
  79 {
  80    struct pipe_screen *screen = filter->pipe->screen;
  81    struct ureg_program *shader;
  82    struct ureg_src i_vtex;
  83    struct ureg_src sampler;
  84    struct ureg_dst *t_array = MALLOC(sizeof(struct ureg_dst) * num_offsets);
  85    struct ureg_dst o_fragment;
  86    const unsigned median = num_offsets >> 1;
  87    unsigned i, j;
  88
  89    assert(num_offsets & 1); /* we need an odd number of offsets */
  90    if (!(num_offsets & 1)) { /* yeah, we REALLY need an odd number of offsets!!! */
  91       FREE(t_array);
  92       return NULL;
  93    }
  94
  95    if (num_offsets > screen->get_shader_param(
  96       screen, PIPE_SHADER_FRAGMENT, PIPE_SHADER_CAP_MAX_TEMPS)) {
  97
  98       FREE(t_array);
  99       return NULL;
 100    }
 101
 102    shader = ureg_create(PIPE_SHADER_FRAGMENT);
 103    if (!shader) {
 104       FREE(t_array);
 105       return NULL;
 106    }
 107
 108    i_vtex = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_VTEX, TGSI_INTERPOLATE_LINEAR);
 109    sampler = ureg_DECL_sampler(shader, 0);
 110
 111    for (i = 0; i < num_offsets; ++i)
 112       t_array[i] = ureg_DECL_temporary(shader);
 113    o_fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 114
 115    /*
 116     * t_array[0..*] = vtex + offset[0..*]
 117     * t_array[0..*] = tex(t_array[0..*], sampler)
 118     * result = partial_bubblesort(t_array)[mid]
 119     */
 120
 121    for (i = 0; i < num_offsets; ++i) {
 122       if (!is_vec_zero(offsets[i])) {
 123          ureg_ADD(shader, ureg_writemask(t_array[i], TGSI_WRITEMASK_XY),
 124                   i_vtex, ureg_imm2f(shader, offsets[i].x, offsets[i].y));
 125          ureg_MOV(shader, ureg_writemask(t_array[i], TGSI_WRITEMASK_ZW),
 126                   ureg_imm1f(shader, 0.0f));
 127       }
 128    }
 129
 130    for (i = 0; i < num_offsets; ++i) {
 131       struct ureg_src src = is_vec_zero(offsets[i]) ? i_vtex : ureg_src(t_array[i]);
 132       ureg_TEX(shader, t_array[i], TGSI_TEXTURE_2D, src, sampler);
 133    }
 134
 135    // TODO: Couldn't this be improved even more?
 136    for (i = 0; i <= median; ++i) {
 137       for (j = 1; j < (num_offsets - i - 1); ++j) {
 138          struct ureg_dst tmp = ureg_DECL_temporary(shader);
 139          ureg_MOV(shader, tmp, ureg_src(t_array[j]));
 140          ureg_MAX(shader, t_array[j], ureg_src(t_array[j]), ureg_src(t_array[j - 1]));
 141          ureg_MIN(shader, t_array[j - 1], ureg_src(tmp), ureg_src(t_array[j - 1]));
 142          ureg_release_temporary(shader, tmp);
 143       }
 144       if (i == median)
 145          ureg_MAX(shader, t_array[j], ureg_src(t_array[j]), ureg_src(t_array[j - 1]));
 146       else
 147          ureg_MIN(shader, t_array[j - 1], ureg_src(t_array[j]), ureg_src(t_array[j - 1]));
 148    }
 149    ureg_MOV(shader, o_fragment, ureg_src(t_array[median]));
 150
 151    ureg_END(shader);
 152
 153    FREE(t_array);
 154    return ureg_create_shader_and_destroy(shader, filter->pipe);
 155 }
 156
 157 static void
 158 generate_offsets(enum vl_median_filter_shape shape, unsigned size,
 159                  struct vertex2f **offsets, unsigned *num_offsets)
 160 {
 161    unsigned i = 0;
 162    int half_size;
 163    struct vertex2f v;
 164
 165    assert(offsets && num_offsets);
 166
 167    /* size needs to be odd */
 168    size = align(size + 1, 2) - 1;
 169    half_size = size >> 1;
 170
 171    switch(shape) {
 172    case VL_MEDIAN_FILTER_BOX:
 173       *num_offsets = size*size;
 174       break;
 175
 176    case VL_MEDIAN_FILTER_CROSS:
 177    case VL_MEDIAN_FILTER_X:
 178       *num_offsets = size + size - 1;
 179       break;
 180
 181    case VL_MEDIAN_FILTER_HORIZONTAL:
 182    case VL_MEDIAN_FILTER_VERTICAL:
 183       *num_offsets = size;
 184       break;
 185
 186    default:
 187       *num_offsets = 0;
 188       return;
 189    }
 190
 191    *offsets = MALLOC(sizeof(struct vertex2f) * *num_offsets);
 192    if (!*offsets)
 193       return;
 194
 195    switch(shape) {
 196    case VL_MEDIAN_FILTER_BOX:
 197       for (v.x = -half_size; v.x <= half_size; ++v.x)
 198          for (v.y = -half_size; v.y <= half_size; ++v.y)
 199             (*offsets)[i++] = v;
 200       break;
 201
 202    case VL_MEDIAN_FILTER_CROSS:
 203       v.y = 0.0f;
 204       for (v.x = -half_size; v.x <= half_size; ++v.x)
 205          (*offsets)[i++] = v;
 206
 207       v.x = 0.0f;
 208       for (v.y = -half_size; v.y <= half_size; ++v.y)
 209          if (v.y != 0.0f)
 210             (*offsets)[i++] = v;
 211       break;
 212
 213    case VL_MEDIAN_FILTER_X:
 214       for (v.x = v.y = -half_size; v.x <= half_size; ++v.x, ++v.y)
 215          (*offsets)[i++] = v;
 216
 217       for (v.x = -half_size, v.y = half_size; v.x <= half_size; ++v.x, --v.y)
 218          if (v.y != 0.0f)
 219             (*offsets)[i++] = v;
 220       break;
 221
 222    case VL_MEDIAN_FILTER_HORIZONTAL:
 223       v.y = 0.0f;
 224       for (v.x = -half_size; v.x <= half_size; ++v.x)
 225          (*offsets)[i++] = v;
 226       break;
 227
 228    case VL_MEDIAN_FILTER_VERTICAL:
 229       v.x = 0.0f;
 230       for (v.y = -half_size; v.y <= half_size; ++v.y)
 231          (*offsets)[i++] = v;
 232       break;
 233    }
 234
 235    assert(i == *num_offsets);
 236 }
 237
 238 bool
 239 vl_median_filter_init(struct vl_median_filter *filter, struct pipe_context *pipe,
 240                       unsigned width, unsigned height, unsigned size,
 241                       enum vl_median_filter_shape shape)
 242 {
 243    struct pipe_rasterizer_state rs_state;
 244    struct pipe_blend_state blend;
 245    struct pipe_sampler_state sampler;
 246    struct vertex2f *offsets = NULL;
 247    struct pipe_vertex_element ve;
 248    unsigned i, num_offsets = 0;
 249
 250    assert(filter && pipe);
 251    assert(width && height);
 252    assert(size > 1 && size < 20);
 253
 254    memset(filter, 0, sizeof(*filter));
 255    filter->pipe = pipe;
 256
 257    memset(&rs_state, 0, sizeof(rs_state));
 258    rs_state.half_pixel_center = true;
 259    rs_state.bottom_edge_rule = true;
 260    rs_state.depth_clip = 1;
 261    filter->rs_state = pipe->create_rasterizer_state(pipe, &rs_state);
 262    if (!filter->rs_state)
 263       goto error_rs_state;
 264
 265    memset(&blend, 0, sizeof blend);
 266    blend.rt[0].rgb_func = PIPE_BLEND_ADD;
 267    blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE;
 268    blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ONE;
 269    blend.rt[0].alpha_func = PIPE_BLEND_ADD;
 270    blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
 271    blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ONE;
 272    blend.logicop_func = PIPE_LOGICOP_CLEAR;
 273    blend.rt[0].colormask = PIPE_MASK_RGBA;
 274    filter->blend = pipe->create_blend_state(pipe, &blend);
 275    if (!filter->blend)
 276       goto error_blend;
 277
 278    memset(&sampler, 0, sizeof(sampler));
 279    sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 280    sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 281    sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 282    sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
 283    sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
 284    sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
 285    sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
 286    sampler.compare_func = PIPE_FUNC_ALWAYS;
 287    sampler.normalized_coords = 1;
 288    filter->sampler = pipe->create_sampler_state(pipe, &sampler);
 289    if (!filter->sampler)
 290       goto error_sampler;
 291
 292    filter->quad = vl_vb_upload_quads(pipe);
 293    if(!filter->quad.buffer)
 294       goto error_quad;
 295
 296    memset(&ve, 0, sizeof(ve));
 297    ve.src_offset = 0;
 298    ve.instance_divisor = 0;
 299    ve.vertex_buffer_index = 0;
 300    ve.src_format = PIPE_FORMAT_R32G32_FLOAT;
 301    filter->ves = pipe->create_vertex_elements_state(pipe, 1, &ve);
 302    if (!filter->ves)
 303       goto error_ves;
 304
 305    generate_offsets(shape, size, &offsets, &num_offsets);
 306    if (!offsets)
 307       goto error_offsets;
 308
 309    for (i = 0; i < num_offsets; ++i) {
 310       offsets[i].x /= width;
 311       offsets[i].y /= height;
 312    }
 313
 314    filter->vs = create_vert_shader(filter);
 315    if (!filter->vs)
 316       goto error_vs;
 317
 318    filter->fs = create_frag_shader(filter, offsets, num_offsets);
 319    if (!filter->fs)
 320       goto error_fs;
 321
 322    FREE(offsets);
 323    return true;
 324
 325 error_fs:
 326    pipe->delete_vs_state(pipe, filter->vs);
 327
 328 error_vs:
 329    FREE(offsets);
 330
 331 error_offsets:
 332    pipe->delete_vertex_elements_state(pipe, filter->ves);
 333
 334 error_ves:
 335    pipe_resource_reference(&filter->quad.buffer, NULL);
 336
 337 error_quad:
 338    pipe->delete_sampler_state(pipe, filter->sampler);
 339
 340 error_sampler:
 341    pipe->delete_blend_state(pipe, filter->blend);
 342
 343 error_blend:
 344    pipe->delete_rasterizer_state(pipe, filter->rs_state);
 345
 346 error_rs_state:
 347    return false;
 348 }
 349
 350 void
 351 vl_median_filter_cleanup(struct vl_median_filter *filter)
 352 {
 353    assert(filter);
 354
 355    filter->pipe->delete_sampler_state(filter->pipe, filter->sampler);
 356    filter->pipe->delete_blend_state(filter->pipe, filter->blend);
 357    filter->pipe->delete_rasterizer_state(filter->pipe, filter->rs_state);
 358    filter->pipe->delete_vertex_elements_state(filter->pipe, filter->ves);
 359    pipe_resource_reference(&filter->quad.buffer, NULL);
 360
 361    filter->pipe->delete_vs_state(filter->pipe, filter->vs);
 362    filter->pipe->delete_fs_state(filter->pipe, filter->fs);
 363 }
 364
 365 void
 366 vl_median_filter_render(struct vl_median_filter *filter,
 367                         struct pipe_sampler_view *src,
 368                         struct pipe_surface *dst)
 369 {
 370    struct pipe_viewport_state viewport;
 371    struct pipe_framebuffer_state fb_state;
 372
 373    assert(filter && src && dst);
 374
 375    memset(&viewport, 0, sizeof(viewport));
 376    viewport.scale[0] = dst->width;
 377    viewport.scale[1] = dst->height;
 378    viewport.scale[2] = 1;
 379
 380    memset(&fb_state, 0, sizeof(fb_state));
 381    fb_state.width = dst->width;
 382    fb_state.height = dst->height;
 383    fb_state.nr_cbufs = 1;
 384    fb_state.cbufs[0] = dst;
 385
 386    filter->pipe->bind_rasterizer_state(filter->pipe, filter->rs_state);
 387    filter->pipe->bind_blend_state(filter->pipe, filter->blend);
 388    filter->pipe->bind_sampler_states(filter->pipe, PIPE_SHADER_FRAGMENT,
 389                                      0, 1, &filter->sampler);
 390    filter->pipe->set_sampler_views(filter->pipe, PIPE_SHADER_FRAGMENT,
 391                                    0, 1, &src);
 392    filter->pipe->bind_vs_state(filter->pipe, filter->vs);
 393    filter->pipe->bind_fs_state(filter->pipe, filter->fs);
 394    filter->pipe->set_framebuffer_state(filter->pipe, &fb_state);
 395    filter->pipe->set_viewport_states(filter->pipe, 0, 1, &viewport);
 396    filter->pipe->set_vertex_buffers(filter->pipe, 0, 1, &filter->quad);
 397    filter->pipe->bind_vertex_elements_state(filter->pipe, filter->ves);
 398
 399    util_draw_arrays(filter->pipe, PIPE_PRIM_QUADS, 0, 4);
 400 }