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 #define STAGE1_SCALE 4.0f
  45 #define STAGE2_SCALE (SCALE_FACTOR_16_TO_9 / STAGE1_SCALE)
  46
  47 struct vertex_shader_consts
  48 {
  49    struct vertex4f norm;
  50 };
  51
  52 enum VS_INPUT
  53 {
  54    VS_I_RECT,
  55    VS_I_VPOS,
  56
  57    NUM_VS_INPUTS
  58 };
  59
  60 enum VS_OUTPUT
  61 {
  62    VS_O_VPOS,
  63    VS_O_BLOCK,
  64    VS_O_TEX,
  65    VS_O_START,
  66    VS_O_STEP
  67 };
  68
  69 static const float const_matrix[8][8] = {
  70    {  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.353553f,  0.3535530f },
  71    {  0.4903930f,  0.4157350f,  0.2777850f,  0.0975451f, -0.0975452f, -0.2777850f, -0.415735f, -0.4903930f },
  72    {  0.4619400f,  0.1913420f, -0.1913420f, -0.4619400f, -0.4619400f, -0.1913420f,  0.191342f,  0.4619400f },
  73    {  0.4157350f, -0.0975452f, -0.4903930f, -0.2777850f,  0.2777850f,  0.4903930f,  0.097545f, -0.4157350f },
  74    {  0.3535530f, -0.3535530f, -0.3535530f,  0.3535540f,  0.3535530f, -0.3535540f, -0.353553f,  0.3535530f },
  75    {  0.2777850f, -0.4903930f,  0.0975452f,  0.4157350f, -0.4157350f, -0.0975451f,  0.490393f, -0.2777850f },
  76    {  0.1913420f, -0.4619400f,  0.4619400f, -0.1913420f, -0.1913410f,  0.4619400f, -0.461940f,  0.1913420f },
  77    {  0.0975451f, -0.2777850f,  0.4157350f, -0.4903930f,  0.4903930f, -0.4157350f,  0.277786f, -0.0975458f }
  78 };
  79
  80 /* vertices for a quad covering a block */
  81 static const struct vertex2f const_quad[4] = {
  82    {0.0f, 0.0f}, {1.0f, 0.0f}, {1.0f, 1.0f}, {0.0f, 1.0f}
  83 };
  84
  85 static void *
  86 create_vert_shader(struct vl_idct *idct)
  87 {
  88    struct ureg_program *shader;
  89    struct ureg_src norm, bs;
  90    struct ureg_src vrect, vpos;
  91    struct ureg_dst scale, t_vpos;
  92    struct ureg_dst o_vpos, o_block, o_tex, o_start, o_step;
  93
  94    shader = ureg_create(TGSI_PROCESSOR_VERTEX);
  95    if (!shader)
  96       return NULL;
  97
  98    norm = ureg_DECL_constant(shader, 0);
  99    bs = ureg_imm2f(shader, BLOCK_WIDTH, BLOCK_HEIGHT);
 100
 101    scale = ureg_DECL_temporary(shader);
 102    t_vpos = ureg_DECL_temporary(shader);
 103
 104    vrect = ureg_DECL_vs_input(shader, VS_I_RECT);
 105    vpos = ureg_DECL_vs_input(shader, VS_I_VPOS);
 106
 107    o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
 108    o_block = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK);
 109    o_tex = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX);
 110    o_start = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_START);
 111    o_step = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_STEP);
 112
 113    /*
 114     * scale = norm * mbs;
 115     *
 116     * t_vpos = vpos + vrect
 117     * o_vpos.xy = t_vpos * scale
 118     * o_vpos.zw = vpos
 119     *
 120     * o_block = vrect
 121     * o_tex = t_pos
 122     * o_start = vpos * scale
 123     * o_step = norm
 124     *
 125     */
 126    ureg_MUL(shader, ureg_writemask(scale, TGSI_WRITEMASK_XY), norm, bs);
 127
 128    ureg_ADD(shader, ureg_writemask(t_vpos, TGSI_WRITEMASK_XY), vpos, vrect);
 129    ureg_MUL(shader, ureg_writemask(t_vpos, TGSI_WRITEMASK_XY), ureg_src(t_vpos), ureg_src(scale));
 130    ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_XY), ureg_src(t_vpos));
 131    ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_ZW), vpos);
 132
 133    ureg_MOV(shader, ureg_writemask(o_tex, TGSI_WRITEMASK_XY), ureg_src(t_vpos));
 134    ureg_MOV(shader, ureg_writemask(o_block, TGSI_WRITEMASK_XY), vrect);
 135    ureg_MUL(shader, ureg_writemask(o_start, TGSI_WRITEMASK_XY), vpos, ureg_src(scale));
 136    ureg_MOV(shader, ureg_writemask(o_step, TGSI_WRITEMASK_XY), norm);
 137
 138    ureg_release_temporary(shader, t_vpos);
 139    ureg_release_temporary(shader, scale);
 140
 141    ureg_END(shader);
 142
 143    return ureg_create_shader_and_destroy(shader, idct->pipe);
 144 }
 145
 146 static void
 147 matrix_mul(struct ureg_program *shader, struct ureg_dst dst,
 148            struct ureg_src tc[2], struct ureg_src sampler[2],
 149            struct ureg_src start[2], struct ureg_src step[2],
 150            bool fetch4[2], float scale)
 151 {
 152    struct ureg_dst t_tc[2], m[2][2], tmp[2];
 153    unsigned side, i, j;
 154
 155    for(i = 0; i < 2; ++i) {
 156       t_tc[i] = ureg_DECL_temporary(shader);
 157       for(j = 0; j < 2; ++j)
 158          m[i][j] = ureg_DECL_temporary(shader);
 159       tmp[i] = ureg_DECL_temporary(shader);
 160    }
 161
 162    /*
 163     * m[0..1][0] = ?
 164     * tmp[0..1] = dot4(m[0..1][0], m[0..1][1])
 165     * fragment = tmp[0] + tmp[1]
 166     */
 167    ureg_MOV(shader, ureg_writemask(t_tc[0], TGSI_WRITEMASK_X), start[0]);
 168    ureg_MOV(shader, ureg_writemask(t_tc[0], TGSI_WRITEMASK_Y), tc[0]);
 169
 170    if(fetch4[1]) {
 171       ureg_MOV(shader, ureg_writemask(t_tc[1], TGSI_WRITEMASK_X), ureg_scalar(start[1], TGSI_SWIZZLE_Y));
 172       ureg_MOV(shader, ureg_writemask(t_tc[1], TGSI_WRITEMASK_Y), ureg_scalar(tc[1], TGSI_SWIZZLE_X));
 173    } else {
 174       ureg_MOV(shader, ureg_writemask(t_tc[1], TGSI_WRITEMASK_X), tc[1]);
 175       ureg_MOV(shader, ureg_writemask(t_tc[1], TGSI_WRITEMASK_Y), start[1]);
 176    }
 177
 178    for(side = 0; side < 2; ++side) {
 179       for(i = 0; i < 2; ++i) {
 180          if(fetch4[side]) {
 181             ureg_TEX(shader, m[i][side], TGSI_TEXTURE_2D, ureg_src(t_tc[side]), sampler[side]);
 182             ureg_MOV(shader, ureg_writemask(t_tc[side], TGSI_WRITEMASK_X), step[side]);
 183
 184          } else for(j = 0; j < 4; ++j) {
 185             /* Nouveau and r600g can't writemask tex dst regs (yet?), do in two steps */
 186             ureg_TEX(shader, tmp[side], TGSI_TEXTURE_2D, ureg_src(t_tc[side]), sampler[side]);
 187             ureg_MOV(shader, ureg_writemask(m[i][side], TGSI_WRITEMASK_X << j), ureg_scalar(ureg_src(tmp[side]), TGSI_SWIZZLE_X));
 188
 189             ureg_ADD(shader, ureg_writemask(t_tc[side], TGSI_WRITEMASK_X << side), ureg_src(t_tc[side]), step[side]);
 190          }
 191       }
 192    }
 193
 194    ureg_DP4(shader, ureg_writemask(tmp[0], TGSI_WRITEMASK_X), ureg_src(m[0][0]), ureg_src(m[0][1]));
 195    ureg_DP4(shader, ureg_writemask(tmp[1], TGSI_WRITEMASK_X), ureg_src(m[1][0]), ureg_src(m[1][1]));
 196    ureg_ADD(shader, ureg_writemask(tmp[0], TGSI_WRITEMASK_X), ureg_src(tmp[0]), ureg_src(tmp[1]));
 197    ureg_MUL(shader, dst, ureg_src(tmp[0]), ureg_imm1f(shader, scale));
 198
 199    for(i = 0; i < 2; ++i) {
 200       ureg_release_temporary(shader, t_tc[i]);
 201       for(j = 0; j < 2; ++j)
 202          ureg_release_temporary(shader, m[i][j]);
 203       ureg_release_temporary(shader, tmp[i]);
 204    }
 205 }
 206
 207 static void *
 208 create_transpose_frag_shader(struct vl_idct *idct)
 209 {
 210    struct ureg_program *shader;
 211    struct ureg_src tc[2], sampler[2];
 212    struct ureg_src start[2], step[2];
 213    struct ureg_dst fragment;
 214    bool fetch4[2];
 215
 216    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 217    if (!shader)
 218       return NULL;
 219
 220    tc[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK, TGSI_INTERPOLATE_LINEAR);
 221    tc[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX, TGSI_INTERPOLATE_LINEAR);
 222
 223    start[0] = ureg_imm1f(shader, 0.0f);
 224    start[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_START, TGSI_INTERPOLATE_CONSTANT);
 225
 226    step[0] = ureg_imm1f(shader, 4.0f / BLOCK_HEIGHT);
 227    step[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_STEP, TGSI_INTERPOLATE_CONSTANT);
 228
 229    sampler[0] = ureg_DECL_sampler(shader, 0);
 230    sampler[1] = ureg_DECL_sampler(shader, 1);
 231
 232    fetch4[0] = true;
 233    fetch4[1] = false;
 234
 235    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 236
 237    matrix_mul(shader, fragment, tc, sampler, start, step, fetch4, STAGE1_SCALE);
 238
 239    ureg_END(shader);
 240
 241    return ureg_create_shader_and_destroy(shader, idct->pipe);
 242 }
 243
 244 static void *
 245 create_matrix_frag_shader(struct vl_idct *idct)
 246 {
 247    struct ureg_program *shader;
 248    struct ureg_src tc[2], sampler[2];
 249    struct ureg_src start[2], step[2];
 250    struct ureg_dst fragment;
 251    bool fetch4[2];
 252
 253    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 254    if (!shader)
 255       return NULL;
 256
 257    tc[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX, TGSI_INTERPOLATE_LINEAR);
 258    tc[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK, TGSI_INTERPOLATE_LINEAR);
 259
 260    start[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_START, TGSI_INTERPOLATE_CONSTANT);
 261    start[1] = ureg_imm1f(shader, 0.0f);
 262
 263    step[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_STEP, TGSI_INTERPOLATE_CONSTANT);
 264    step[1] = ureg_imm1f(shader, 4.0f / BLOCK_WIDTH);
 265
 266    sampler[0] = ureg_DECL_sampler(shader, 1);
 267    sampler[1] = ureg_DECL_sampler(shader, 0);
 268
 269    fetch4[0] = false;
 270    fetch4[1] = true;
 271
 272    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 273
 274    matrix_mul(shader, fragment, tc, sampler, start, step, fetch4, STAGE2_SCALE);
 275
 276    ureg_END(shader);
 277
 278    return ureg_create_shader_and_destroy(shader, idct->pipe);
 279 }
 280
 281 static void *
 282 create_empty_block_frag_shader(struct vl_idct *idct)
 283 {
 284    struct ureg_program *shader;
 285    struct ureg_dst fragment;
 286
 287    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 288    if (!shader)
 289       return NULL;
 290
 291    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 292
 293    ureg_MOV(shader, fragment, ureg_imm1f(shader, 0.0f));
 294
 295    ureg_END(shader);
 296
 297    return ureg_create_shader_and_destroy(shader, idct->pipe);
 298 }
 299
 300 static void
 301 xfer_buffers_map(struct vl_idct *idct)
 302 {
 303    struct pipe_box rect =
 304    {
 305       0, 0, 0,
 306       idct->destination->width0,
 307       idct->destination->height0,
 308       1
 309    };
 310
 311    idct->tex_transfer = idct->pipe->get_transfer
 312    (
 313       idct->pipe, idct->textures.individual.source,
 314       u_subresource(0, 0),
 315       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 316       &rect
 317    );
 318
 319    idct->texels = idct->pipe->transfer_map(idct->pipe, idct->tex_transfer);
 320
 321    idct->vectors = pipe_buffer_map
 322    (
 323       idct->pipe,
 324       idct->vertex_bufs.individual.pos.buffer,
 325       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 326       &idct->vec_transfer
 327    );
 328 }
 329
 330 static void
 331 xfer_buffers_unmap(struct vl_idct *idct)
 332 {
 333    pipe_buffer_unmap(idct->pipe, idct->vertex_bufs.individual.pos.buffer, idct->vec_transfer);
 334
 335    idct->pipe->transfer_unmap(idct->pipe, idct->tex_transfer);
 336    idct->pipe->transfer_destroy(idct->pipe, idct->tex_transfer);
 337 }
 338
 339 static bool
 340 init_shaders(struct vl_idct *idct)
 341 {
 342    idct->vs = create_vert_shader(idct);
 343    idct->transpose_fs = create_transpose_frag_shader(idct);
 344    idct->matrix_fs = create_matrix_frag_shader(idct);
 345    idct->eb_fs = create_empty_block_frag_shader(idct);
 346
 347    return
 348       idct->vs != NULL &&
 349       idct->transpose_fs != NULL &&
 350       idct->matrix_fs != NULL &&
 351       idct->eb_fs != NULL;
 352 }
 353
 354 static void
 355 cleanup_shaders(struct vl_idct *idct)
 356 {
 357    idct->pipe->delete_vs_state(idct->pipe, idct->vs);
 358    idct->pipe->delete_fs_state(idct->pipe, idct->transpose_fs);
 359    idct->pipe->delete_fs_state(idct->pipe, idct->matrix_fs);
 360    idct->pipe->delete_fs_state(idct->pipe, idct->eb_fs);
 361 }
 362
 363 static bool
 364 init_buffers(struct vl_idct *idct)
 365 {
 366    struct pipe_resource template;
 367    struct pipe_sampler_view sampler_view;
 368    struct pipe_vertex_element vertex_elems[2];
 369    unsigned i;
 370
 371    idct->max_blocks =
 372       align(idct->destination->width0, BLOCK_WIDTH) / BLOCK_WIDTH *
 373       align(idct->destination->height0, BLOCK_HEIGHT) / BLOCK_HEIGHT *
 374       idct->destination->depth0;
 375
 376    memset(&template, 0, sizeof(struct pipe_resource));
 377    template.target = PIPE_TEXTURE_2D;
 378    template.format = PIPE_FORMAT_R32G32B32A32_FLOAT;
 379    template.last_level = 0;
 380    template.width0 = 2;
 381    template.height0 = 8;
 382    template.depth0 = 1;
 383    template.usage = PIPE_USAGE_IMMUTABLE;
 384    template.bind = PIPE_BIND_SAMPLER_VIEW;
 385    template.flags = 0;
 386
 387    template.format = idct->destination->format;
 388    template.width0 = idct->destination->width0;
 389    template.height0 = idct->destination->height0;
 390    template.depth0 = idct->destination->depth0;
 391    template.usage = PIPE_USAGE_DYNAMIC;
 392    idct->textures.individual.source = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 393
 394    template.usage = PIPE_USAGE_STATIC;
 395    idct->textures.individual.intermediate = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 396
 397    for (i = 0; i < 4; ++i) {
 398       if(idct->textures.all[i] == NULL)
 399          return false; /* a texture failed to allocate */
 400
 401       u_sampler_view_default_template(&sampler_view, idct->textures.all[i], idct->textures.all[i]->format);
 402       idct->sampler_views.all[i] = idct->pipe->create_sampler_view(idct->pipe, idct->textures.all[i], &sampler_view);
 403    }
 404
 405    idct->vertex_bufs.individual.quad.stride = sizeof(struct vertex2f);
 406    idct->vertex_bufs.individual.quad.max_index = 4 * idct->max_blocks - 1;
 407    idct->vertex_bufs.individual.quad.buffer_offset = 0;
 408    idct->vertex_bufs.individual.quad.buffer = pipe_buffer_create
 409    (
 410       idct->pipe->screen,
 411       PIPE_BIND_VERTEX_BUFFER,
 412       sizeof(struct vertex2f) * 4 * idct->max_blocks
 413    );
 414
 415    if(idct->vertex_bufs.individual.quad.buffer == NULL)
 416       return false;
 417
 418    idct->vertex_bufs.individual.pos.stride = sizeof(struct vertex2f);
 419    idct->vertex_bufs.individual.pos.max_index = 4 * idct->max_blocks - 1;
 420    idct->vertex_bufs.individual.pos.buffer_offset = 0;
 421    idct->vertex_bufs.individual.pos.buffer = pipe_buffer_create
 422    (
 423       idct->pipe->screen,
 424       PIPE_BIND_VERTEX_BUFFER,
 425       sizeof(struct vertex2f) * 4 * idct->max_blocks
 426    );
 427
 428    if(idct->vertex_bufs.individual.pos.buffer == NULL)
 429       return false;
 430
 431    /* Rect element */
 432    vertex_elems[0].src_offset = 0;
 433    vertex_elems[0].instance_divisor = 0;
 434    vertex_elems[0].vertex_buffer_index = 0;
 435    vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
 436
 437    /* Pos element */
 438    vertex_elems[1].src_offset = 0;
 439    vertex_elems[1].instance_divisor = 0;
 440    vertex_elems[1].vertex_buffer_index = 1;
 441    vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
 442
 443    idct->vertex_elems_state = idct->pipe->create_vertex_elements_state(idct->pipe, 2, vertex_elems);
 444
 445    idct->vs_const_buf = pipe_buffer_create
 446    (
 447       idct->pipe->screen,
 448       PIPE_BIND_CONSTANT_BUFFER,
 449       sizeof(struct vertex_shader_consts)
 450    );
 451
 452    if(idct->vs_const_buf == NULL)
 453       return false;
 454
 455    return true;
 456 }
 457
 458 static void
 459 cleanup_buffers(struct vl_idct *idct)
 460 {
 461    unsigned i;
 462
 463    assert(idct);
 464
 465    pipe_resource_reference(&idct->vs_const_buf, NULL);
 466
 467    for (i = 0; i < 4; ++i) {
 468       pipe_sampler_view_reference(&idct->sampler_views.all[i], NULL);
 469       pipe_resource_reference(&idct->textures.all[i], NULL);
 470    }
 471
 472    idct->pipe->delete_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 473    pipe_resource_reference(&idct->vertex_bufs.individual.quad.buffer, NULL);
 474    pipe_resource_reference(&idct->vertex_bufs.individual.pos.buffer, NULL);
 475 }
 476
 477 static void
 478 init_constants(struct vl_idct *idct)
 479 {
 480    struct pipe_transfer *buf_transfer;
 481    struct vertex_shader_consts *vs_consts;
 482    struct vertex2f *v;
 483
 484    unsigned i;
 485
 486    /* quad vectors */
 487    v = pipe_buffer_map
 488    (
 489       idct->pipe,
 490       idct->vertex_bufs.individual.quad.buffer,
 491       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 492       &buf_transfer
 493    );
 494    for ( i = 0; i < idct->max_blocks; ++i)
 495      memcpy(v + i * 4, &const_quad, sizeof(const_quad));
 496    pipe_buffer_unmap(idct->pipe, idct->vertex_bufs.individual.quad.buffer, buf_transfer);
 497
 498    /* normalisation constants */
 499    vs_consts = pipe_buffer_map
 500    (
 501       idct->pipe, idct->vs_const_buf,
 502       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 503       &buf_transfer
 504    );
 505
 506    vs_consts->norm.x = 1.0f / idct->destination->width0;
 507    vs_consts->norm.y = 1.0f / idct->destination->height0;
 508
 509    pipe_buffer_unmap(idct->pipe, idct->vs_const_buf, buf_transfer);
 510 }
 511
 512 static void
 513 init_state(struct vl_idct *idct)
 514 {
 515    struct pipe_sampler_state sampler;
 516    unsigned i;
 517
 518    idct->num_blocks = 0;
 519    idct->num_empty_blocks = 0;
 520
 521    idct->viewport.scale[0] = idct->destination->width0;
 522    idct->viewport.scale[1] = idct->destination->height0;
 523    idct->viewport.scale[2] = 1;
 524    idct->viewport.scale[3] = 1;
 525    idct->viewport.translate[0] = 0;
 526    idct->viewport.translate[1] = 0;
 527    idct->viewport.translate[2] = 0;
 528    idct->viewport.translate[3] = 0;
 529
 530    idct->fb_state.width = idct->destination->width0;
 531    idct->fb_state.height = idct->destination->height0;
 532    idct->fb_state.nr_cbufs = 1;
 533    idct->fb_state.zsbuf = NULL;
 534
 535    for (i = 0; i < 4; ++i) {
 536       memset(&sampler, 0, sizeof(sampler));
 537       sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 538       sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 539       sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 540       sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
 541       sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
 542       sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
 543       sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
 544       sampler.compare_func = PIPE_FUNC_ALWAYS;
 545       sampler.normalized_coords = 1;
 546       /*sampler.shadow_ambient = ; */
 547       /*sampler.lod_bias = ; */
 548       sampler.min_lod = 0;
 549       /*sampler.max_lod = ; */
 550       /*sampler.border_color[0] = ; */
 551       /*sampler.max_anisotropy = ; */
 552       idct->samplers.all[i] = idct->pipe->create_sampler_state(idct->pipe, &sampler);
 553    }
 554 }
 555
 556 static void
 557 cleanup_state(struct vl_idct *idct)
 558 {
 559    unsigned i;
 560
 561    for (i = 0; i < 4; ++i)
 562       idct->pipe->delete_sampler_state(idct->pipe, idct->samplers.all[i]);
 563 }
 564
 565 struct pipe_resource *
 566 vl_idct_upload_matrix(struct pipe_context *pipe)
 567 {
 568    struct pipe_resource template, *matrix;
 569    struct pipe_transfer *buf_transfer;
 570    unsigned i, j, pitch;
 571    float *f;
 572
 573    struct pipe_box rect =
 574    {
 575       0, 0, 0,
 576       BLOCK_WIDTH,
 577       BLOCK_HEIGHT,
 578       1
 579    };
 580
 581    memset(&template, 0, sizeof(struct pipe_resource));
 582    template.target = PIPE_TEXTURE_2D;
 583    template.format = PIPE_FORMAT_R32G32B32A32_FLOAT;
 584    template.last_level = 0;
 585    template.width0 = 2;
 586    template.height0 = 8;
 587    template.depth0 = 1;
 588    template.usage = PIPE_USAGE_IMMUTABLE;
 589    template.bind = PIPE_BIND_SAMPLER_VIEW;
 590    template.flags = 0;
 591
 592    matrix = pipe->screen->resource_create(pipe->screen, &template);
 593
 594    /* matrix */
 595    buf_transfer = pipe->get_transfer
 596    (
 597       pipe, matrix,
 598       u_subresource(0, 0),
 599       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 600       &rect
 601    );
 602    pitch = buf_transfer->stride / util_format_get_blocksize(buf_transfer->resource->format);
 603
 604    f = pipe->transfer_map(pipe, buf_transfer);
 605    for(i = 0; i < BLOCK_HEIGHT; ++i)
 606       for(j = 0; j < BLOCK_WIDTH; ++j)
 607          f[i * pitch * 4 + j] = const_matrix[j][i]; // transpose
 608
 609    pipe->transfer_unmap(pipe, buf_transfer);
 610    pipe->transfer_destroy(pipe, buf_transfer);
 611
 612    return matrix;
 613 }
 614
 615 bool
 616 vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe, struct pipe_resource *dst, struct pipe_resource *matrix)
 617 {
 618    assert(idct && pipe && dst);
 619
 620    idct->pipe = pipe;
 621    pipe_resource_reference(&idct->textures.individual.matrix, matrix);
 622    pipe_resource_reference(&idct->textures.individual.transpose, matrix);
 623    pipe_resource_reference(&idct->destination, dst);
 624
 625    init_state(idct);
 626
 627    if(!init_shaders(idct))
 628       return false;
 629
 630    if(!init_buffers(idct)) {
 631       cleanup_shaders(idct);
 632       return false;
 633    }
 634
 635    idct->surfaces.intermediate = idct->pipe->screen->get_tex_surface(
 636       idct->pipe->screen, idct->textures.individual.intermediate, 0, 0, 0,
 637       PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET);
 638
 639    idct->surfaces.destination = idct->pipe->screen->get_tex_surface(
 640       idct->pipe->screen, idct->destination, 0, 0, 0,
 641       PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET);
 642
 643    init_constants(idct);
 644    xfer_buffers_map(idct);
 645
 646    return true;
 647 }
 648
 649 void
 650 vl_idct_cleanup(struct vl_idct *idct)
 651 {
 652    idct->pipe->screen->tex_surface_destroy(idct->surfaces.destination);
 653    idct->pipe->screen->tex_surface_destroy(idct->surfaces.intermediate);
 654
 655    cleanup_shaders(idct);
 656    cleanup_buffers(idct);
 657
 658    cleanup_state(idct);
 659
 660    pipe_resource_reference(&idct->destination, NULL);
 661 }
 662
 663 void
 664 vl_idct_add_block(struct vl_idct *idct, unsigned x, unsigned y, short *block)
 665 {
 666    struct vertex2f v, *v_dst;
 667
 668    unsigned tex_pitch;
 669    short *texels;
 670
 671    unsigned i;
 672
 673    assert(idct);
 674
 675    if(block) {
 676       tex_pitch = idct->tex_transfer->stride / util_format_get_blocksize(idct->tex_transfer->resource->format);
 677       texels = idct->texels + y * tex_pitch * BLOCK_HEIGHT + x * BLOCK_WIDTH;
 678
 679       for (i = 0; i < BLOCK_HEIGHT; ++i)
 680          memcpy(texels + i * tex_pitch, block + i * BLOCK_WIDTH, BLOCK_WIDTH * 2);
 681
 682       /* non empty blocks fills the vector buffer from left to right */
 683       v_dst = idct->vectors + idct->num_blocks * 4;
 684
 685       idct->num_blocks++;
 686
 687    } else {
 688
 689       /* while empty blocks fills the vector buffer from right to left */
 690       v_dst = idct->vectors + (idct->max_blocks - idct->num_empty_blocks) * 4 - 4;
 691
 692       idct->num_empty_blocks++;
 693    }
 694
 695    v.x = x;
 696    v.y = y;
 697
 698    for (i = 0; i < 4; ++i) {
 699       v_dst[i] = v;
 700    }
 701 }
 702
 703 void
 704 vl_idct_flush(struct vl_idct *idct)
 705 {
 706    xfer_buffers_unmap(idct);
 707
 708    idct->pipe->set_constant_buffer(idct->pipe, PIPE_SHADER_VERTEX, 0, idct->vs_const_buf);
 709
 710    if(idct->num_blocks > 0) {
 711
 712       /* first stage */
 713       idct->fb_state.cbufs[0] = idct->surfaces.intermediate;
 714       idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state);
 715       idct->pipe->set_viewport_state(idct->pipe, &idct->viewport);
 716
 717       idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all);
 718       idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 719       idct->pipe->set_fragment_sampler_views(idct->pipe, 2, idct->sampler_views.stage[0]);
 720       idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[0]);
 721       idct->pipe->bind_vs_state(idct->pipe, idct->vs);
 722       idct->pipe->bind_fs_state(idct->pipe, idct->transpose_fs);
 723
 724       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, idct->num_blocks * 4);
 725
 726       /* second stage */
 727       idct->fb_state.cbufs[0] = idct->surfaces.destination;
 728       idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state);
 729       idct->pipe->set_viewport_state(idct->pipe, &idct->viewport);
 730
 731       idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all);
 732       idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 733       idct->pipe->set_fragment_sampler_views(idct->pipe, 2, idct->sampler_views.stage[1]);
 734       idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[1]);
 735       idct->pipe->bind_vs_state(idct->pipe, idct->vs);
 736       idct->pipe->bind_fs_state(idct->pipe, idct->matrix_fs);
 737
 738       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, idct->num_blocks * 4);
 739    }
 740
 741    if(idct->num_empty_blocks > 0) {
 742
 743       /* empty block handling */
 744       idct->fb_state.cbufs[0] = idct->surfaces.destination;
 745       idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state);
 746       idct->pipe->set_viewport_state(idct->pipe, &idct->viewport);
 747
 748       idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all);
 749       idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 750       idct->pipe->set_fragment_sampler_views(idct->pipe, 4, idct->sampler_views.all);
 751       idct->pipe->bind_fragment_sampler_states(idct->pipe, 4, idct->samplers.all);
 752       idct->pipe->bind_vs_state(idct->pipe, idct->vs);
 753       idct->pipe->bind_fs_state(idct->pipe, idct->eb_fs);
 754
 755       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS,
 756          (idct->max_blocks - idct->num_empty_blocks) * 4,
 757          idct->num_empty_blocks * 4);
 758    }
 759
 760    idct->num_blocks = 0;
 761    idct->num_empty_blocks = 0;
 762    xfer_buffers_map(idct);
 763 }