src/gallium/auxiliary/vl/vl_idct.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2010 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 "vl_idct.h"
  29 #include "util/u_draw.h"
  30 #include <assert.h>
  31 #include <pipe/p_context.h>
  32 #include <pipe/p_screen.h>
  33 #include <util/u_inlines.h>
  34 #include <util/u_sampler.h>
  35 #include <util/u_format.h>
  36 #include <tgsi/tgsi_ureg.h>
  37 #include "vl_types.h"
  38
  39 #define BLOCK_WIDTH 8
  40 #define BLOCK_HEIGHT 8
  41
  42 #define SCALE_FACTOR_16_TO_9 (32768.0f / 256.0f)
  43
  44 struct vertex_shader_consts
  45 {
  46    struct vertex4f norm;
  47 };
  48
  49 enum VS_INPUT
  50 {
  51    VS_I_RECT,
  52    VS_I_VPOS,
  53
  54    NUM_VS_INPUTS
  55 };
  56
  57 enum VS_OUTPUT
  58 {
  59    VS_O_VPOS,
  60    VS_O_BLOCK,
  61    VS_O_TEX,
  62    VS_O_START,
  63    VS_O_STEP
  64 };
  65
  66 static const float const_matrix[8][8] = {
  67    {  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.353553f,  0.3535530f },
  68    {  0.4903930f,  0.4157350f,  0.2777850f,  0.0975451f, -0.0975452f, -0.2777850f, -0.415735f, -0.4903930f },
  69    {  0.4619400f,  0.1913420f, -0.1913420f, -0.4619400f, -0.4619400f, -0.1913420f,  0.191342f,  0.4619400f },
  70    {  0.4157350f, -0.0975452f, -0.4903930f, -0.2777850f,  0.2777850f,  0.4903930f,  0.097545f, -0.4157350f },
  71    {  0.3535530f, -0.3535530f, -0.3535530f,  0.3535540f,  0.3535530f, -0.3535540f, -0.353553f,  0.3535530f },
  72    {  0.2777850f, -0.4903930f,  0.0975452f,  0.4157350f, -0.4157350f, -0.0975451f,  0.490393f, -0.2777850f },
  73    {  0.1913420f, -0.4619400f,  0.4619400f, -0.1913420f, -0.1913410f,  0.4619400f, -0.461940f,  0.1913420f },
  74    {  0.0975451f, -0.2777850f,  0.4157350f, -0.4903930f,  0.4903930f, -0.4157350f,  0.277786f, -0.0975458f }
  75 };
  76
  77 /* vertices for a quad covering a block */
  78 static const struct vertex2f const_quad[4] = {
  79    {0.0f, 0.0f}, {1.0f, 0.0f}, {1.0f, 1.0f}, {0.0f, 1.0f}
  80 };
  81
  82 static void *
  83 create_vert_shader(struct vl_idct *idct)
  84 {
  85    struct ureg_program *shader;
  86    struct ureg_src norm, bs;
  87    struct ureg_src vrect, vpos;
  88    struct ureg_dst scale, t_vpos;
  89    struct ureg_dst o_vpos, o_block, o_tex, o_start, o_step;
  90
  91    shader = ureg_create(TGSI_PROCESSOR_VERTEX);
  92    if (!shader)
  93       return NULL;
  94
  95    norm = ureg_DECL_constant(shader, 0);
  96    bs = ureg_imm2f(shader, BLOCK_WIDTH, BLOCK_HEIGHT);
  97
  98    scale = ureg_DECL_temporary(shader);
  99    t_vpos = ureg_DECL_temporary(shader);
 100
 101    vrect = ureg_DECL_vs_input(shader, VS_I_RECT);
 102    vpos = ureg_DECL_vs_input(shader, VS_I_VPOS);
 103
 104    o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
 105    o_block = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK);
 106    o_tex = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX);
 107    o_start = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_START);
 108    o_step = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_STEP);
 109
 110    /*
 111     * scale = norm * mbs;
 112     *
 113     * t_vpos = vpos + vrect
 114     * o_vpos.xy = t_vpos * scale
 115     * o_vpos.zw = vpos
 116     *
 117     * o_block = vrect
 118     * o_tex = t_pos
 119     * o_start = vpos * scale
 120     * o_step = norm
 121     *
 122     */
 123    ureg_MUL(shader, ureg_writemask(scale, TGSI_WRITEMASK_XY), norm, bs);
 124
 125    ureg_ADD(shader, ureg_writemask(t_vpos, TGSI_WRITEMASK_XY), vpos, vrect);
 126    ureg_MUL(shader, ureg_writemask(t_vpos, TGSI_WRITEMASK_XY), ureg_src(t_vpos), ureg_src(scale));
 127    ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_XY), ureg_src(t_vpos));
 128    ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_ZW), vpos);
 129
 130    ureg_MOV(shader, ureg_writemask(o_tex, TGSI_WRITEMASK_XY), ureg_src(t_vpos));
 131    ureg_MOV(shader, ureg_writemask(o_block, TGSI_WRITEMASK_XY), vrect);
 132    ureg_MUL(shader, ureg_writemask(o_start, TGSI_WRITEMASK_XY), vpos, ureg_src(scale));
 133    ureg_MOV(shader, ureg_writemask(o_step, TGSI_WRITEMASK_XY), norm);
 134
 135    ureg_release_temporary(shader, t_vpos);
 136    ureg_release_temporary(shader, scale);
 137
 138    ureg_END(shader);
 139
 140    return ureg_create_shader_and_destroy(shader, idct->pipe);
 141 }
 142
 143 static void
 144 matrix_mul(struct ureg_program *shader, struct ureg_dst dst,
 145            struct ureg_src tc[2], struct ureg_src sampler[2],
 146            struct ureg_src start[2], struct ureg_src step[2],
 147            float scale[2])
 148 {
 149    struct ureg_dst t_tc[2], m[2][2], tmp[2];
 150    unsigned i, j;
 151
 152    for(i = 0; i < 2; ++i) {
 153       t_tc[i] = ureg_DECL_temporary(shader);
 154       for(j = 0; j < 2; ++j)
 155          m[i][j] = ureg_DECL_temporary(shader);
 156       tmp[i] = ureg_DECL_temporary(shader);
 157    }
 158
 159    /*
 160     * m[0..1][0] = ?
 161     * tmp[0..1] = dot4(m[0..1][0], m[0..1][1])
 162     * fragment = tmp[0] + tmp[1]
 163     */
 164    ureg_MOV(shader, ureg_writemask(t_tc[0], TGSI_WRITEMASK_X), start[0]);
 165    ureg_MOV(shader, ureg_writemask(t_tc[0], TGSI_WRITEMASK_Y), tc[0]);
 166
 167    ureg_MOV(shader, ureg_writemask(t_tc[1], TGSI_WRITEMASK_X), tc[1]);
 168    ureg_MOV(shader, ureg_writemask(t_tc[1], TGSI_WRITEMASK_Y), start[1]);
 169
 170    for(i = 0; i < 2; ++i) {
 171       for(j = 0; j < 4; ++j) {
 172          /* Nouveau and r600g can't writemask tex dst regs (yet?), do in two steps */
 173          ureg_TEX(shader, tmp[0], TGSI_TEXTURE_2D, ureg_src(t_tc[0]), sampler[0]);
 174          ureg_MOV(shader, ureg_writemask(m[i][0], TGSI_WRITEMASK_X << j), ureg_scalar(ureg_src(tmp[0]), TGSI_SWIZZLE_X));
 175
 176          ureg_TEX(shader, tmp[1], TGSI_TEXTURE_2D, ureg_src(t_tc[1]), sampler[1]);
 177          ureg_MOV(shader, ureg_writemask(m[i][1], TGSI_WRITEMASK_X << j), ureg_scalar(ureg_src(tmp[1]), TGSI_SWIZZLE_X));
 178
 179          ureg_ADD(shader, ureg_writemask(t_tc[0], TGSI_WRITEMASK_X), ureg_src(t_tc[0]), step[0]);
 180          ureg_ADD(shader, ureg_writemask(t_tc[1], TGSI_WRITEMASK_Y), ureg_src(t_tc[1]), step[1]);
 181       }
 182
 183       if(scale[0] != 1.0f)
 184          ureg_MUL(shader, m[i][0], ureg_src(m[i][0]), ureg_scalar(ureg_imm1f(shader, scale[0]), TGSI_SWIZZLE_X));
 185
 186       if(scale[1] != 1.0f)
 187          ureg_MUL(shader, m[i][1], ureg_src(m[i][1]), ureg_scalar(ureg_imm1f(shader, scale[1]), TGSI_SWIZZLE_X));
 188    }
 189
 190    ureg_DP4(shader, ureg_writemask(tmp[0], TGSI_WRITEMASK_X), ureg_src(m[0][0]), ureg_src(m[0][1]));
 191    ureg_DP4(shader, ureg_writemask(tmp[1], TGSI_WRITEMASK_X), ureg_src(m[1][0]), ureg_src(m[1][1]));
 192    ureg_ADD(shader, ureg_writemask(dst, TGSI_WRITEMASK_X), ureg_src(tmp[0]), ureg_src(tmp[1]));
 193
 194    for(i = 0; i < 2; ++i) {
 195       ureg_release_temporary(shader, t_tc[i]);
 196       for(j = 0; j < 2; ++j)
 197          ureg_release_temporary(shader, m[i][j]);
 198       ureg_release_temporary(shader, tmp[i]);
 199    }
 200 }
 201
 202 static void *
 203 create_transpose_frag_shader(struct vl_idct *idct)
 204 {
 205    struct ureg_program *shader;
 206    struct ureg_src tc[2], sampler[2];
 207    struct ureg_src start[2], step[2];
 208    struct ureg_dst fragment;
 209    float scale[2];
 210
 211    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 212    if (!shader)
 213       return NULL;
 214
 215    tc[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK, TGSI_INTERPOLATE_LINEAR);
 216    tc[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX, TGSI_INTERPOLATE_LINEAR);
 217
 218    start[0] = ureg_imm1f(shader, 0.0f);
 219    start[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_START, TGSI_INTERPOLATE_CONSTANT);
 220
 221    step[0] = ureg_imm1f(shader, 1.0f / BLOCK_HEIGHT);
 222    step[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_STEP, TGSI_INTERPOLATE_CONSTANT);
 223
 224    sampler[0] = ureg_DECL_sampler(shader, 0);
 225    sampler[1] = ureg_DECL_sampler(shader, 2);
 226
 227    scale[0] = 1.0f;
 228    scale[1] = SCALE_FACTOR_16_TO_9;
 229
 230    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 231
 232    matrix_mul(shader, fragment, tc, sampler, start, step, scale);
 233
 234    ureg_END(shader);
 235
 236    return ureg_create_shader_and_destroy(shader, idct->pipe);
 237 }
 238
 239 static void *
 240 create_matrix_frag_shader(struct vl_idct *idct)
 241 {
 242    struct ureg_program *shader;
 243    struct ureg_src tc[2], sampler[2];
 244    struct ureg_src start[2], step[2];
 245    struct ureg_dst fragment;
 246    float scale[2];
 247
 248    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 249    if (!shader)
 250       return NULL;
 251
 252    tc[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX, TGSI_INTERPOLATE_LINEAR);
 253    tc[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK, TGSI_INTERPOLATE_LINEAR);
 254
 255    start[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_START, TGSI_INTERPOLATE_CONSTANT);
 256    start[1] = ureg_imm1f(shader, 0.0f);
 257
 258    step[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_STEP, TGSI_INTERPOLATE_CONSTANT);
 259    step[1] = ureg_imm1f(shader, 1.0f / BLOCK_WIDTH);
 260
 261    sampler[0] = ureg_DECL_sampler(shader, 3);
 262    sampler[1] = ureg_DECL_sampler(shader, 1);
 263
 264    scale[0] = 1.0f;
 265    scale[1] = 1.0f;
 266
 267    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 268
 269    matrix_mul(shader, fragment, tc, sampler, start, step, scale);
 270
 271    ureg_END(shader);
 272
 273    return ureg_create_shader_and_destroy(shader, idct->pipe);
 274 }
 275
 276 static void *
 277 create_empty_block_frag_shader(struct vl_idct *idct)
 278 {
 279    struct ureg_program *shader;
 280    struct ureg_dst fragment;
 281
 282    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 283    if (!shader)
 284       return NULL;
 285
 286    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 287
 288    ureg_MOV(shader, fragment, ureg_imm1f(shader, 0.0f));
 289
 290    ureg_END(shader);
 291
 292    return ureg_create_shader_and_destroy(shader, idct->pipe);
 293 }
 294
 295 static void
 296 xfer_buffers_map(struct vl_idct *idct)
 297 {
 298    struct pipe_box rect =
 299    {
 300       0, 0, 0,
 301       idct->destination->width0,
 302       idct->destination->height0,
 303       1
 304    };
 305
 306    idct->tex_transfer = idct->pipe->get_transfer
 307    (
 308 #if 1
 309       idct->pipe, idct->textures.individual.source,
 310 #else
 311       idct->pipe, idct->destination,
 312 #endif
 313       u_subresource(0, 0),
 314       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 315       &rect
 316    );
 317
 318    idct->texels = idct->pipe->transfer_map(idct->pipe, idct->tex_transfer);
 319
 320    idct->vectors = pipe_buffer_map
 321    (
 322       idct->pipe,
 323       idct->vertex_bufs.individual.pos.buffer,
 324       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 325       &idct->vec_transfer
 326    );
 327 }
 328
 329 static void
 330 xfer_buffers_unmap(struct vl_idct *idct)
 331 {
 332    pipe_buffer_unmap(idct->pipe, idct->vertex_bufs.individual.pos.buffer, idct->vec_transfer);
 333
 334    idct->pipe->transfer_unmap(idct->pipe, idct->tex_transfer);
 335    idct->pipe->transfer_destroy(idct->pipe, idct->tex_transfer);
 336 }
 337
 338 static bool
 339 init_shaders(struct vl_idct *idct)
 340 {
 341    assert(idct->vs = create_vert_shader(idct));
 342    assert(idct->transpose_fs = create_transpose_frag_shader(idct));
 343    assert(idct->matrix_fs = create_matrix_frag_shader(idct));
 344    assert(idct->eb_fs = create_empty_block_frag_shader(idct));
 345
 346    return
 347       idct->vs != NULL &&
 348       idct->transpose_fs != NULL &&
 349       idct->matrix_fs != NULL &&
 350       idct->eb_fs != NULL;
 351 }
 352
 353 static void
 354 cleanup_shaders(struct vl_idct *idct)
 355 {
 356    idct->pipe->delete_vs_state(idct->pipe, idct->vs);
 357    idct->pipe->delete_fs_state(idct->pipe, idct->transpose_fs);
 358    idct->pipe->delete_fs_state(idct->pipe, idct->matrix_fs);
 359    idct->pipe->delete_fs_state(idct->pipe, idct->eb_fs);
 360 }
 361
 362 static bool
 363 init_buffers(struct vl_idct *idct)
 364 {
 365    struct pipe_resource template;
 366    struct pipe_sampler_view sampler_view;
 367    struct pipe_vertex_element vertex_elems[2];
 368    unsigned i;
 369
 370    idct->max_blocks =
 371       align(idct->destination->width0, BLOCK_WIDTH) / BLOCK_WIDTH *
 372       align(idct->destination->height0, BLOCK_HEIGHT) / BLOCK_HEIGHT *
 373       idct->destination->depth0;
 374
 375    memset(&template, 0, sizeof(struct pipe_resource));
 376    template.target = PIPE_TEXTURE_2D;
 377    template.format = PIPE_FORMAT_R32_FLOAT;
 378    template.last_level = 0;
 379    template.width0 = 8;
 380    template.height0 = 8;
 381    template.depth0 = 1;
 382    template.usage = PIPE_USAGE_IMMUTABLE;
 383    template.bind = PIPE_BIND_SAMPLER_VIEW;
 384    template.flags = 0;
 385
 386    idct->textures.individual.matrix = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 387    idct->textures.individual.transpose = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 388
 389    template.format = idct->destination->format;
 390    template.width0 = idct->destination->width0;
 391    template.height0 = idct->destination->height0;
 392    template.depth0 = idct->destination->depth0;
 393    template.usage = PIPE_USAGE_DYNAMIC;
 394    idct->textures.individual.source = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 395
 396    template.usage = PIPE_USAGE_STATIC;
 397    idct->textures.individual.intermediate = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 398
 399    for (i = 0; i < 4; ++i) {
 400       if(idct->textures.all[i] == NULL)
 401          return false; /* a texture failed to allocate */
 402
 403       u_sampler_view_default_template(&sampler_view, idct->textures.all[i], idct->textures.all[i]->format);
 404       idct->sampler_views.all[i] = idct->pipe->create_sampler_view(idct->pipe, idct->textures.all[i], &sampler_view);
 405    }
 406
 407    idct->vertex_bufs.individual.quad.stride = sizeof(struct vertex2f);
 408    idct->vertex_bufs.individual.quad.max_index = 4 * idct->max_blocks - 1;
 409    idct->vertex_bufs.individual.quad.buffer_offset = 0;
 410    idct->vertex_bufs.individual.quad.buffer = pipe_buffer_create
 411    (
 412       idct->pipe->screen,
 413       PIPE_BIND_VERTEX_BUFFER,
 414       sizeof(struct vertex2f) * 4 * idct->max_blocks
 415    );
 416
 417    if(idct->vertex_bufs.individual.quad.buffer == NULL)
 418       return false;
 419
 420    idct->vertex_bufs.individual.pos.stride = sizeof(struct vertex2f);
 421    idct->vertex_bufs.individual.pos.max_index = 4 * idct->max_blocks - 1;
 422    idct->vertex_bufs.individual.pos.buffer_offset = 0;
 423    idct->vertex_bufs.individual.pos.buffer = pipe_buffer_create
 424    (
 425       idct->pipe->screen,
 426       PIPE_BIND_VERTEX_BUFFER,
 427       sizeof(struct vertex2f) * 4 * idct->max_blocks
 428    );
 429
 430    if(idct->vertex_bufs.individual.pos.buffer == NULL)
 431       return false;
 432
 433    /* Rect element */
 434    vertex_elems[0].src_offset = 0;
 435    vertex_elems[0].instance_divisor = 0;
 436    vertex_elems[0].vertex_buffer_index = 0;
 437    vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
 438
 439    /* Pos element */
 440    vertex_elems[1].src_offset = 0;
 441    vertex_elems[1].instance_divisor = 0;
 442    vertex_elems[1].vertex_buffer_index = 1;
 443    vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
 444
 445    idct->vertex_elems_state = idct->pipe->create_vertex_elements_state(idct->pipe, 2, vertex_elems);
 446
 447    idct->vs_const_buf = pipe_buffer_create
 448    (
 449       idct->pipe->screen,
 450       PIPE_BIND_CONSTANT_BUFFER,
 451       sizeof(struct vertex_shader_consts)
 452    );
 453
 454    if(idct->vs_const_buf == NULL)
 455       return false;
 456
 457    return true;
 458 }
 459
 460 static void
 461 cleanup_buffers(struct vl_idct *idct)
 462 {
 463    unsigned i;
 464
 465    assert(idct);
 466
 467    pipe_resource_reference(&idct->vs_const_buf, NULL);
 468
 469    for (i = 0; i < 4; ++i) {
 470       pipe_sampler_view_reference(&idct->sampler_views.all[i], NULL);
 471       pipe_resource_reference(&idct->textures.all[i], NULL);
 472    }
 473
 474    idct->pipe->delete_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 475    pipe_resource_reference(&idct->vertex_bufs.individual.quad.buffer, NULL);
 476    pipe_resource_reference(&idct->vertex_bufs.individual.pos.buffer, NULL);
 477 }
 478
 479 static void
 480 init_constants(struct vl_idct *idct)
 481 {
 482    struct pipe_transfer *buf_transfer;
 483    struct vertex_shader_consts *vs_consts;
 484    struct vertex2f *v;
 485    float *f;
 486
 487    struct pipe_box rect =
 488    {
 489       0, 0, 0,
 490       BLOCK_WIDTH,
 491       BLOCK_HEIGHT,
 492       1
 493    };
 494
 495    unsigned i, j, pitch;
 496
 497    /* quad vectors */
 498    v = pipe_buffer_map
 499    (
 500       idct->pipe,
 501       idct->vertex_bufs.individual.quad.buffer,
 502       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 503       &buf_transfer
 504    );
 505    for ( i = 0; i < idct->max_blocks; ++i)
 506      memcpy(v + i * 4, &const_quad, sizeof(const_quad));
 507    pipe_buffer_unmap(idct->pipe, idct->vertex_bufs.individual.quad.buffer, buf_transfer);
 508
 509    /* transposed matrix */
 510    buf_transfer = idct->pipe->get_transfer
 511    (
 512       idct->pipe, idct->textures.individual.transpose,
 513       u_subresource(0, 0),
 514       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 515       &rect
 516    );
 517    pitch = buf_transfer->stride / util_format_get_blocksize(buf_transfer->resource->format);
 518
 519    f = idct->pipe->transfer_map(idct->pipe, buf_transfer);
 520    for(i = 0; i < BLOCK_HEIGHT; ++i)
 521       for(j = 0; j < BLOCK_WIDTH; ++j)
 522          f[i * pitch + j] = const_matrix[j][i]; // transpose
 523
 524    idct->pipe->transfer_unmap(idct->pipe, buf_transfer);
 525    idct->pipe->transfer_destroy(idct->pipe, buf_transfer);
 526
 527    /* matrix */
 528    buf_transfer = idct->pipe->get_transfer
 529    (
 530       idct->pipe, idct->textures.individual.matrix,
 531       u_subresource(0, 0),
 532       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 533       &rect
 534    );
 535    pitch = buf_transfer->stride / util_format_get_blocksize(buf_transfer->resource->format);
 536
 537    f = idct->pipe->transfer_map(idct->pipe, buf_transfer);
 538    for(i = 0; i < BLOCK_HEIGHT; ++i)
 539       for(j = 0; j < BLOCK_WIDTH; ++j)
 540          f[i * pitch + j] = const_matrix[i][j];
 541
 542    idct->pipe->transfer_unmap(idct->pipe, buf_transfer);
 543    idct->pipe->transfer_destroy(idct->pipe, buf_transfer);
 544
 545    /* normalisation constants */
 546    vs_consts = pipe_buffer_map
 547    (
 548       idct->pipe, idct->vs_const_buf,
 549       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 550       &buf_transfer
 551    );
 552
 553    vs_consts->norm.x = 1.0f / idct->destination->width0;
 554    vs_consts->norm.y = 1.0f / idct->destination->height0;
 555
 556    pipe_buffer_unmap(idct->pipe, idct->vs_const_buf, buf_transfer);
 557 }
 558
 559 static void
 560 init_state(struct vl_idct *idct)
 561 {
 562    struct pipe_sampler_state sampler;
 563    unsigned i;
 564
 565    idct->num_blocks = 0;
 566    idct->num_empty_blocks = 0;
 567
 568    idct->viewport.scale[0] = idct->destination->width0;
 569    idct->viewport.scale[1] = idct->destination->height0;
 570    idct->viewport.scale[2] = 1;
 571    idct->viewport.scale[3] = 1;
 572    idct->viewport.translate[0] = 0;
 573    idct->viewport.translate[1] = 0;
 574    idct->viewport.translate[2] = 0;
 575    idct->viewport.translate[3] = 0;
 576
 577    idct->fb_state.width = idct->destination->width0;
 578    idct->fb_state.height = idct->destination->height0;
 579    idct->fb_state.nr_cbufs = 1;
 580    idct->fb_state.zsbuf = NULL;
 581
 582    for (i = 0; i < 4; ++i) {
 583       memset(&sampler, 0, sizeof(sampler));
 584       sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 585       sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 586       sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 587       sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
 588       sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
 589       sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
 590       sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
 591       sampler.compare_func = PIPE_FUNC_ALWAYS;
 592       sampler.normalized_coords = 1;
 593       /*sampler.shadow_ambient = ; */
 594       /*sampler.lod_bias = ; */
 595       sampler.min_lod = 0;
 596       /*sampler.max_lod = ; */
 597       /*sampler.border_color[0] = ; */
 598       /*sampler.max_anisotropy = ; */
 599       idct->samplers.all[i] = idct->pipe->create_sampler_state(idct->pipe, &sampler);
 600    }
 601 }
 602
 603 static void
 604 cleanup_state(struct vl_idct *idct)
 605 {
 606    unsigned i;
 607
 608    for (i = 0; i < 4; ++i)
 609       idct->pipe->delete_sampler_state(idct->pipe, idct->samplers.all[i]);
 610 }
 611
 612 bool
 613 vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe, struct pipe_resource *dst)
 614 {
 615    assert(idct && pipe && dst);
 616
 617    idct->pipe = pipe;
 618    pipe_resource_reference(&idct->destination, dst);
 619
 620    init_state(idct);
 621
 622    if(!init_shaders(idct))
 623       return false;
 624
 625    if(!init_buffers(idct)) {
 626       cleanup_shaders(idct);
 627       return false;
 628    }
 629
 630    idct->surfaces.intermediate = idct->pipe->screen->get_tex_surface(
 631       idct->pipe->screen, idct->textures.individual.intermediate, 0, 0, 0,
 632       PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET);
 633
 634    idct->surfaces.destination = idct->pipe->screen->get_tex_surface(
 635       idct->pipe->screen, idct->destination, 0, 0, 0,
 636       PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET);
 637
 638    init_constants(idct);
 639    xfer_buffers_map(idct);
 640
 641    return true;
 642 }
 643
 644 void
 645 vl_idct_cleanup(struct vl_idct *idct)
 646 {
 647    idct->pipe->screen->tex_surface_destroy(idct->surfaces.destination);
 648    idct->pipe->screen->tex_surface_destroy(idct->surfaces.intermediate);
 649
 650    cleanup_shaders(idct);
 651    cleanup_buffers(idct);
 652
 653    cleanup_state(idct);
 654
 655    pipe_resource_reference(&idct->destination, NULL);
 656 }
 657
 658 void
 659 vl_idct_add_block(struct vl_idct *idct, unsigned x, unsigned y, short *block)
 660 {
 661    struct vertex2f v, *v_dst;
 662
 663    unsigned tex_pitch;
 664    short *texels;
 665
 666    unsigned i;
 667
 668    assert(idct);
 669
 670    if(block) {
 671       tex_pitch = idct->tex_transfer->stride / util_format_get_blocksize(idct->tex_transfer->resource->format);
 672       texels = idct->texels + y * tex_pitch * BLOCK_HEIGHT + x * BLOCK_WIDTH;
 673
 674       for (i = 0; i < BLOCK_HEIGHT; ++i)
 675          memcpy(texels + i * tex_pitch, block + i * BLOCK_WIDTH, BLOCK_WIDTH * 2);
 676
 677       /* non empty blocks fills the vector buffer from left to right */
 678       v_dst = idct->vectors + idct->num_blocks * 4;
 679
 680       idct->num_blocks++;
 681
 682    } else {
 683
 684       /* while empty blocks fills the vector buffer from right to left */
 685       v_dst = idct->vectors + (idct->max_blocks - idct->num_empty_blocks) * 4 - 4;
 686
 687       idct->num_empty_blocks++;
 688    }
 689
 690    v.x = x;
 691    v.y = y;
 692
 693    for (i = 0; i < 4; ++i) {
 694       v_dst[i] = v;
 695    }
 696 }
 697
 698 void
 699 vl_idct_flush(struct vl_idct *idct)
 700 {
 701    xfer_buffers_unmap(idct);
 702
 703    idct->pipe->set_constant_buffer(idct->pipe, PIPE_SHADER_VERTEX, 0, idct->vs_const_buf);
 704
 705    if(idct->num_blocks > 0) {
 706
 707       /* first stage */
 708       idct->fb_state.cbufs[0] = idct->surfaces.intermediate;
 709       idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state);
 710       idct->pipe->set_viewport_state(idct->pipe, &idct->viewport);
 711
 712       idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all);
 713       idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 714       idct->pipe->set_fragment_sampler_views(idct->pipe, 4, idct->sampler_views.all);
 715       idct->pipe->bind_fragment_sampler_states(idct->pipe, 4, idct->samplers.all);
 716       idct->pipe->bind_vs_state(idct->pipe, idct->vs);
 717       idct->pipe->bind_fs_state(idct->pipe, idct->transpose_fs);
 718
 719       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, idct->num_blocks * 4);
 720
 721       idct->pipe->flush(idct->pipe, PIPE_FLUSH_RENDER_CACHE, NULL);
 722
 723       /* second stage */
 724       idct->fb_state.cbufs[0] = idct->surfaces.destination;
 725       idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state);
 726       idct->pipe->set_viewport_state(idct->pipe, &idct->viewport);
 727
 728       idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all);
 729       idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 730       idct->pipe->set_fragment_sampler_views(idct->pipe, 4, idct->sampler_views.all);
 731       idct->pipe->bind_fragment_sampler_states(idct->pipe, 4, idct->samplers.all);
 732       idct->pipe->bind_vs_state(idct->pipe, idct->vs);
 733       idct->pipe->bind_fs_state(idct->pipe, idct->matrix_fs);
 734
 735       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, idct->num_blocks * 4);
 736    }
 737
 738    if(idct->num_empty_blocks > 0) {
 739
 740       /* empty block handling */
 741       idct->fb_state.cbufs[0] = idct->surfaces.destination;
 742       idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state);
 743       idct->pipe->set_viewport_state(idct->pipe, &idct->viewport);
 744
 745       idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all);
 746       idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 747       idct->pipe->set_fragment_sampler_views(idct->pipe, 4, idct->sampler_views.all);
 748       idct->pipe->bind_fragment_sampler_states(idct->pipe, 4, idct->samplers.all);
 749       idct->pipe->bind_vs_state(idct->pipe, idct->vs);
 750       idct->pipe->bind_fs_state(idct->pipe, idct->eb_fs);
 751
 752       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS,
 753          (idct->max_blocks - idct->num_empty_blocks) * 4,
 754          idct->num_empty_blocks * 4);
 755    }
 756
 757    idct->pipe->flush(idct->pipe, PIPE_FLUSH_RENDER_CACHE, NULL);
 758
 759    idct->num_blocks = 0;
 760    idct->num_empty_blocks = 0;
 761    xfer_buffers_map(idct);
 762 }