src/gallium/auxiliary/vl/vl_matrix_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 TUNGSTEN GRAPHICS 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 <stdio.h>
  29
  30 #include "pipe/p_context.h"
  31
  32 #include "tgsi/tgsi_ureg.h"
  33
  34 #include "util/u_draw.h"
  35 #include "util/u_memory.h"
  36 #include "util/u_math.h"
  37
  38 #include "vl_types.h"
  39 #include "vl_vertex_buffers.h"
  40 #include "vl_matrix_filter.h"
  41
  42 enum VS_OUTPUT
  43 {
  44    VS_O_VPOS = 0,
  45    VS_O_VTEX = 0
  46 };
  47
  48 static void *
  49 create_vert_shader(struct vl_matrix_filter *filter)
  50 {
  51    struct ureg_program *shader;
  52    struct ureg_src i_vpos;
  53    struct ureg_dst o_vpos, o_vtex;
  54
  55    shader = ureg_create(TGSI_PROCESSOR_VERTEX);
  56    if (!shader)
  57       return NULL;
  58
  59    i_vpos = ureg_DECL_vs_input(shader, 0);
  60    o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
  61    o_vtex = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_VTEX);
  62
  63    ureg_MOV(shader, o_vpos, i_vpos);
  64    ureg_MOV(shader, o_vtex, i_vpos);
  65
  66    ureg_END(shader);
  67
  68    return ureg_create_shader_and_destroy(shader, filter->pipe);
  69 }
  70
  71 static inline bool
  72 is_vec_zero(struct vertex2f v)
  73 {
  74    return v.x == 0.0f && v.y == 0.0f;
  75 }
  76
  77 static void *
  78 create_frag_shader(struct vl_matrix_filter *filter, unsigned num_offsets,
  79                    struct vertex2f *offsets, const float *matrix_values)
  80 {
  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 t_sum;
  86    struct ureg_dst o_fragment;
  87    bool first;
  88    int i;
  89
  90    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
  91    if (!shader) {
  92       FREE(t_array);
  93       return NULL;
  94    }
  95
  96    i_vtex = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_VTEX, TGSI_INTERPOLATE_LINEAR);
  97    sampler = ureg_DECL_sampler(shader, 0);
  98
  99    for (i = 0; i < num_offsets; ++i)
 100       if (matrix_values[i] != 0.0f)
 101          t_array[i] = ureg_DECL_temporary(shader);
 102
 103    o_fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 104
 105    /*
 106     * t_array[0..*] = vtex + offset[0..*]
 107     * t_array[0..*] = tex(t_array[0..*], sampler)
 108     * o_fragment = sum(t_array[0..*] * matrix_values[0..*])
 109     */
 110
 111    for (i = 0; i < num_offsets; ++i) {
 112       if (matrix_values[i] != 0.0f && !is_vec_zero(offsets[i])) {
 113          ureg_ADD(shader, ureg_writemask(t_array[i], TGSI_WRITEMASK_XY),
 114                   i_vtex, ureg_imm2f(shader, offsets[i].x, offsets[i].y));
 115          ureg_MOV(shader, ureg_writemask(t_array[i], TGSI_WRITEMASK_ZW),
 116                   ureg_imm1f(shader, 0.0f));
 117       }
 118    }
 119
 120    for (i = 0; i < num_offsets; ++i) {
 121       if (matrix_values[i] != 0.0f) {
 122          struct ureg_src src = is_vec_zero(offsets[i]) ? i_vtex : ureg_src(t_array[i]);
 123          ureg_TEX(shader, t_array[i], TGSI_TEXTURE_2D, src, sampler);
 124       }
 125    }
 126
 127    for (i = 0, first = true; i < num_offsets; ++i) {
 128       if (matrix_values[i] != 0.0f) {
 129          if (first) {
 130             t_sum = t_array[i];
 131             ureg_MUL(shader, t_sum, ureg_src(t_array[i]),
 132                      ureg_imm1f(shader, matrix_values[i]));
 133             first = false;
 134          } else
 135             ureg_MAD(shader, t_sum, ureg_src(t_array[i]),
 136                      ureg_imm1f(shader, matrix_values[i]), ureg_src(t_sum));
 137       }
 138    }
 139    if (first)
 140       ureg_MOV(shader, o_fragment, ureg_imm1f(shader, 0.0f));
 141    else
 142       ureg_MOV(shader, o_fragment, ureg_src(t_sum));
 143
 144    ureg_END(shader);
 145
 146    FREE(t_array);
 147    return ureg_create_shader_and_destroy(shader, filter->pipe);
 148 }
 149
 150 bool
 151 vl_matrix_filter_init(struct vl_matrix_filter *filter, struct pipe_context *pipe,
 152                       unsigned video_width, unsigned video_height,
 153                       unsigned matrix_width, unsigned matrix_height,
 154                       const float *matrix_values)
 155 {
 156    struct pipe_rasterizer_state rs_state;
 157    struct pipe_blend_state blend;
 158    struct pipe_sampler_state sampler;
 159    struct pipe_vertex_element ve;
 160    struct vertex2f *offsets, v, sizes;
 161    unsigned i, num_offsets = matrix_width * matrix_height;
 162
 163    assert(filter && pipe);
 164    assert(video_width && video_height);
 165    assert(matrix_width && matrix_height);
 166
 167    memset(filter, 0, sizeof(*filter));
 168    filter->pipe = pipe;
 169
 170    memset(&rs_state, 0, sizeof(rs_state));
 171    rs_state.half_pixel_center = true;
 172    rs_state.bottom_edge_rule = true;
 173    rs_state.depth_clip = 1;
 174    filter->rs_state = pipe->create_rasterizer_state(pipe, &rs_state);
 175    if (!filter->rs_state)
 176       goto error_rs_state;
 177
 178    memset(&blend, 0, sizeof blend);
 179    blend.rt[0].rgb_func = PIPE_BLEND_ADD;
 180    blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE;
 181    blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ONE;
 182    blend.rt[0].alpha_func = PIPE_BLEND_ADD;
 183    blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
 184    blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ONE;
 185    blend.logicop_func = PIPE_LOGICOP_CLEAR;
 186    blend.rt[0].colormask = PIPE_MASK_RGBA;
 187    filter->blend = pipe->create_blend_state(pipe, &blend);
 188    if (!filter->blend)
 189       goto error_blend;
 190
 191    memset(&sampler, 0, sizeof(sampler));
 192    sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 193    sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 194    sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 195    sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
 196    sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
 197    sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
 198    sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
 199    sampler.compare_func = PIPE_FUNC_ALWAYS;
 200    sampler.normalized_coords = 1;
 201    filter->sampler = pipe->create_sampler_state(pipe, &sampler);
 202    if (!filter->sampler)
 203       goto error_sampler;
 204
 205    filter->quad = vl_vb_upload_quads(pipe);
 206    if(!filter->quad.buffer)
 207       goto error_quad;
 208
 209    memset(&ve, 0, sizeof(ve));
 210    ve.src_offset = 0;
 211    ve.instance_divisor = 0;
 212    ve.vertex_buffer_index = 0;
 213    ve.src_format = PIPE_FORMAT_R32G32_FLOAT;
 214    filter->ves = pipe->create_vertex_elements_state(pipe, 1, &ve);
 215    if (!filter->ves)
 216       goto error_ves;
 217
 218    offsets = MALLOC(sizeof(struct vertex2f) * num_offsets);
 219    if (!offsets)
 220       goto error_offsets;
 221
 222    sizes.x = (float)(matrix_width - 1) / 2.0f;
 223    sizes.y = (float)(matrix_height - 1) / 2.0f;
 224
 225    for (v.x = -sizes.x, i = 0; v.x <= sizes.x; v.x += 1.0f)
 226       for (v.y = -sizes.y; v.y <= sizes.y; v.y += 1.0f)
 227          offsets[i++] = v;
 228
 229    for (i = 0; i < num_offsets; ++i) {
 230       offsets[i].x /= video_width;
 231       offsets[i].y /= video_height;
 232    }
 233
 234    filter->vs = create_vert_shader(filter);
 235    if (!filter->vs)
 236       goto error_vs;
 237
 238    filter->fs = create_frag_shader(filter, num_offsets, offsets, matrix_values);
 239    if (!filter->fs)
 240       goto error_fs;
 241
 242    FREE(offsets);
 243    return true;
 244
 245 error_fs:
 246    pipe->delete_vs_state(pipe, filter->vs);
 247
 248 error_vs:
 249    FREE(offsets);
 250
 251 error_offsets:
 252    pipe->delete_vertex_elements_state(pipe, filter->ves);
 253
 254 error_ves:
 255    pipe_resource_reference(&filter->quad.buffer, NULL);
 256
 257 error_quad:
 258    pipe->delete_sampler_state(pipe, filter->sampler);
 259
 260 error_sampler:
 261    pipe->delete_blend_state(pipe, filter->blend);
 262
 263 error_blend:
 264    pipe->delete_rasterizer_state(pipe, filter->rs_state);
 265
 266 error_rs_state:
 267    return false;
 268 }
 269
 270 void
 271 vl_matrix_filter_cleanup(struct vl_matrix_filter *filter)
 272 {
 273    assert(filter);
 274
 275    filter->pipe->delete_sampler_state(filter->pipe, filter->sampler);
 276    filter->pipe->delete_blend_state(filter->pipe, filter->blend);
 277    filter->pipe->delete_rasterizer_state(filter->pipe, filter->rs_state);
 278    filter->pipe->delete_vertex_elements_state(filter->pipe, filter->ves);
 279    pipe_resource_reference(&filter->quad.buffer, NULL);
 280
 281    filter->pipe->delete_vs_state(filter->pipe, filter->vs);
 282    filter->pipe->delete_fs_state(filter->pipe, filter->fs);
 283 }
 284
 285 void
 286 vl_matrix_filter_render(struct vl_matrix_filter *filter,
 287                         struct pipe_sampler_view *src,
 288                         struct pipe_surface *dst)
 289 {
 290    struct pipe_viewport_state viewport;
 291    struct pipe_framebuffer_state fb_state;
 292
 293    assert(filter && src && dst);
 294
 295    memset(&viewport, 0, sizeof(viewport));
 296    viewport.scale[0] = dst->width;
 297    viewport.scale[1] = dst->height;
 298    viewport.scale[2] = 1;
 299    viewport.scale[3] = 1;
 300
 301    memset(&fb_state, 0, sizeof(fb_state));
 302    fb_state.width = dst->width;
 303    fb_state.height = dst->height;
 304    fb_state.nr_cbufs = 1;
 305    fb_state.cbufs[0] = dst;
 306
 307    filter->pipe->bind_rasterizer_state(filter->pipe, filter->rs_state);
 308    filter->pipe->bind_blend_state(filter->pipe, filter->blend);
 309    filter->pipe->bind_sampler_states(filter->pipe, PIPE_SHADER_FRAGMENT,
 310                                      0, 1, &filter->sampler);
 311    filter->pipe->set_fragment_sampler_views(filter->pipe, 1, &src);
 312    filter->pipe->bind_vs_state(filter->pipe, filter->vs);
 313    filter->pipe->bind_fs_state(filter->pipe, filter->fs);
 314    filter->pipe->set_framebuffer_state(filter->pipe, &fb_state);
 315    filter->pipe->set_viewport_states(filter->pipe, 0, 1, &viewport);
 316    filter->pipe->set_vertex_buffers(filter->pipe, 0, 1, &filter->quad);
 317    filter->pipe->bind_vertex_elements_state(filter->pipe, filter->ves);
 318
 319    util_draw_arrays(filter->pipe, PIPE_PRIM_QUADS, 0, 4);
 320 }