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            float scale)
 151 {
 152    struct ureg_dst t_tc[2], m[2][2], tmp[2];
 153    unsigned 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    ureg_MOV(shader, ureg_writemask(t_tc[1], TGSI_WRITEMASK_X), tc[1]);
 171    ureg_MOV(shader, ureg_writemask(t_tc[1], TGSI_WRITEMASK_Y), start[1]);
 172
 173    for(i = 0; i < 2; ++i) {
 174       for(j = 0; j < 4; ++j) {
 175          /* Nouveau and r600g can't writemask tex dst regs (yet?), do in two steps */
 176          ureg_TEX(shader, tmp[0], TGSI_TEXTURE_2D, ureg_src(t_tc[0]), sampler[0]);
 177          ureg_MOV(shader, ureg_writemask(m[i][0], TGSI_WRITEMASK_X << j), ureg_scalar(ureg_src(tmp[0]), TGSI_SWIZZLE_X));
 178
 179          ureg_TEX(shader, tmp[1], TGSI_TEXTURE_2D, ureg_src(t_tc[1]), sampler[1]);
 180          ureg_MOV(shader, ureg_writemask(m[i][1], TGSI_WRITEMASK_X << j), ureg_scalar(ureg_src(tmp[1]), TGSI_SWIZZLE_X));
 181
 182          ureg_ADD(shader, ureg_writemask(t_tc[0], TGSI_WRITEMASK_X), ureg_src(t_tc[0]), step[0]);
 183          ureg_ADD(shader, ureg_writemask(t_tc[1], TGSI_WRITEMASK_Y), ureg_src(t_tc[1]), step[1]);
 184       }
 185    }
 186
 187    ureg_DP4(shader, ureg_writemask(tmp[0], TGSI_WRITEMASK_X), ureg_src(m[0][0]), ureg_src(m[0][1]));
 188    ureg_DP4(shader, ureg_writemask(tmp[1], TGSI_WRITEMASK_X), ureg_src(m[1][0]), ureg_src(m[1][1]));
 189    ureg_ADD(shader, ureg_writemask(tmp[0], TGSI_WRITEMASK_X), ureg_src(tmp[0]), ureg_src(tmp[1]));
 190    ureg_MUL(shader, dst, ureg_src(tmp[0]), ureg_imm1f(shader, scale));
 191
 192    for(i = 0; i < 2; ++i) {
 193       ureg_release_temporary(shader, t_tc[i]);
 194       for(j = 0; j < 2; ++j)
 195          ureg_release_temporary(shader, m[i][j]);
 196       ureg_release_temporary(shader, tmp[i]);
 197    }
 198 }
 199
 200 static void *
 201 create_transpose_frag_shader(struct vl_idct *idct)
 202 {
 203    struct ureg_program *shader;
 204    struct ureg_src tc[2], sampler[2];
 205    struct ureg_src start[2], step[2];
 206    struct ureg_dst fragment;
 207
 208    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 209    if (!shader)
 210       return NULL;
 211
 212    tc[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK, TGSI_INTERPOLATE_LINEAR);
 213    tc[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX, TGSI_INTERPOLATE_LINEAR);
 214
 215    start[0] = ureg_imm1f(shader, 0.0f);
 216    start[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_START, TGSI_INTERPOLATE_CONSTANT);
 217
 218    step[0] = ureg_imm1f(shader, 1.0f / BLOCK_HEIGHT);
 219    step[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_STEP, TGSI_INTERPOLATE_CONSTANT);
 220
 221    sampler[0] = ureg_DECL_sampler(shader, 0);
 222    sampler[1] = ureg_DECL_sampler(shader, 1);
 223
 224    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 225
 226    matrix_mul(shader, fragment, tc, sampler, start, step, STAGE1_SCALE);
 227
 228    ureg_END(shader);
 229
 230    return ureg_create_shader_and_destroy(shader, idct->pipe);
 231 }
 232
 233 static void *
 234 create_matrix_frag_shader(struct vl_idct *idct)
 235 {
 236    struct ureg_program *shader;
 237    struct ureg_src tc[2], sampler[2];
 238    struct ureg_src start[2], step[2];
 239    struct ureg_dst fragment;
 240
 241    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 242    if (!shader)
 243       return NULL;
 244
 245    tc[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX, TGSI_INTERPOLATE_LINEAR);
 246    tc[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK, TGSI_INTERPOLATE_LINEAR);
 247
 248    start[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_START, TGSI_INTERPOLATE_CONSTANT);
 249    start[1] = ureg_imm1f(shader, 0.0f);
 250
 251    step[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_STEP, TGSI_INTERPOLATE_CONSTANT);
 252    step[1] = ureg_imm1f(shader, 1.0f / BLOCK_WIDTH);
 253
 254    sampler[0] = ureg_DECL_sampler(shader, 1);
 255    sampler[1] = ureg_DECL_sampler(shader, 0);
 256
 257    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 258
 259    matrix_mul(shader, fragment, tc, sampler, start, step, STAGE2_SCALE);
 260
 261    ureg_END(shader);
 262
 263    return ureg_create_shader_and_destroy(shader, idct->pipe);
 264 }
 265
 266 static void *
 267 create_empty_block_frag_shader(struct vl_idct *idct)
 268 {
 269    struct ureg_program *shader;
 270    struct ureg_dst fragment;
 271
 272    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 273    if (!shader)
 274       return NULL;
 275
 276    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 277
 278    ureg_MOV(shader, fragment, ureg_imm1f(shader, 0.0f));
 279
 280    ureg_END(shader);
 281
 282    return ureg_create_shader_and_destroy(shader, idct->pipe);
 283 }
 284
 285 static void
 286 xfer_buffers_map(struct vl_idct *idct)
 287 {
 288    struct pipe_box rect =
 289    {
 290       0, 0, 0,
 291       idct->destination->width0,
 292       idct->destination->height0,
 293       1
 294    };
 295
 296    idct->tex_transfer = idct->pipe->get_transfer
 297    (
 298       idct->pipe, idct->textures.individual.source,
 299       u_subresource(0, 0),
 300       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 301       &rect
 302    );
 303
 304    idct->texels = idct->pipe->transfer_map(idct->pipe, idct->tex_transfer);
 305
 306    idct->vectors = pipe_buffer_map
 307    (
 308       idct->pipe,
 309       idct->vertex_bufs.individual.pos.buffer,
 310       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 311       &idct->vec_transfer
 312    );
 313 }
 314
 315 static void
 316 xfer_buffers_unmap(struct vl_idct *idct)
 317 {
 318    pipe_buffer_unmap(idct->pipe, idct->vertex_bufs.individual.pos.buffer, idct->vec_transfer);
 319
 320    idct->pipe->transfer_unmap(idct->pipe, idct->tex_transfer);
 321    idct->pipe->transfer_destroy(idct->pipe, idct->tex_transfer);
 322 }
 323
 324 static bool
 325 init_shaders(struct vl_idct *idct)
 326 {
 327    idct->vs = create_vert_shader(idct);
 328    idct->transpose_fs = create_transpose_frag_shader(idct);
 329    idct->matrix_fs = create_matrix_frag_shader(idct);
 330    idct->eb_fs = create_empty_block_frag_shader(idct);
 331
 332    return
 333       idct->vs != NULL &&
 334       idct->transpose_fs != NULL &&
 335       idct->matrix_fs != NULL &&
 336       idct->eb_fs != NULL;
 337 }
 338
 339 static void
 340 cleanup_shaders(struct vl_idct *idct)
 341 {
 342    idct->pipe->delete_vs_state(idct->pipe, idct->vs);
 343    idct->pipe->delete_fs_state(idct->pipe, idct->transpose_fs);
 344    idct->pipe->delete_fs_state(idct->pipe, idct->matrix_fs);
 345    idct->pipe->delete_fs_state(idct->pipe, idct->eb_fs);
 346 }
 347
 348 static bool
 349 init_buffers(struct vl_idct *idct)
 350 {
 351    struct pipe_resource template;
 352    struct pipe_sampler_view sampler_view;
 353    struct pipe_vertex_element vertex_elems[2];
 354    unsigned i;
 355
 356    idct->max_blocks =
 357       align(idct->destination->width0, BLOCK_WIDTH) / BLOCK_WIDTH *
 358       align(idct->destination->height0, BLOCK_HEIGHT) / BLOCK_HEIGHT *
 359       idct->destination->depth0;
 360
 361    memset(&template, 0, sizeof(struct pipe_resource));
 362    template.target = PIPE_TEXTURE_2D;
 363    template.format = PIPE_FORMAT_R32_FLOAT;
 364    template.last_level = 0;
 365    template.width0 = 8;
 366    template.height0 = 8;
 367    template.depth0 = 1;
 368    template.usage = PIPE_USAGE_IMMUTABLE;
 369    template.bind = PIPE_BIND_SAMPLER_VIEW;
 370    template.flags = 0;
 371
 372    idct->textures.individual.matrix = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 373    idct->textures.individual.transpose = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 374
 375    template.format = idct->destination->format;
 376    template.width0 = idct->destination->width0;
 377    template.height0 = idct->destination->height0;
 378    template.depth0 = idct->destination->depth0;
 379    template.usage = PIPE_USAGE_DYNAMIC;
 380    idct->textures.individual.source = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 381
 382    template.usage = PIPE_USAGE_STATIC;
 383    idct->textures.individual.intermediate = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 384
 385    for (i = 0; i < 4; ++i) {
 386       if(idct->textures.all[i] == NULL)
 387          return false; /* a texture failed to allocate */
 388
 389       u_sampler_view_default_template(&sampler_view, idct->textures.all[i], idct->textures.all[i]->format);
 390       idct->sampler_views.all[i] = idct->pipe->create_sampler_view(idct->pipe, idct->textures.all[i], &sampler_view);
 391    }
 392
 393    idct->vertex_bufs.individual.quad.stride = sizeof(struct vertex2f);
 394    idct->vertex_bufs.individual.quad.max_index = 4 * idct->max_blocks - 1;
 395    idct->vertex_bufs.individual.quad.buffer_offset = 0;
 396    idct->vertex_bufs.individual.quad.buffer = pipe_buffer_create
 397    (
 398       idct->pipe->screen,
 399       PIPE_BIND_VERTEX_BUFFER,
 400       sizeof(struct vertex2f) * 4 * idct->max_blocks
 401    );
 402
 403    if(idct->vertex_bufs.individual.quad.buffer == NULL)
 404       return false;
 405
 406    idct->vertex_bufs.individual.pos.stride = sizeof(struct vertex2f);
 407    idct->vertex_bufs.individual.pos.max_index = 4 * idct->max_blocks - 1;
 408    idct->vertex_bufs.individual.pos.buffer_offset = 0;
 409    idct->vertex_bufs.individual.pos.buffer = pipe_buffer_create
 410    (
 411       idct->pipe->screen,
 412       PIPE_BIND_VERTEX_BUFFER,
 413       sizeof(struct vertex2f) * 4 * idct->max_blocks
 414    );
 415
 416    if(idct->vertex_bufs.individual.pos.buffer == NULL)
 417       return false;
 418
 419    /* Rect element */
 420    vertex_elems[0].src_offset = 0;
 421    vertex_elems[0].instance_divisor = 0;
 422    vertex_elems[0].vertex_buffer_index = 0;
 423    vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
 424
 425    /* Pos element */
 426    vertex_elems[1].src_offset = 0;
 427    vertex_elems[1].instance_divisor = 0;
 428    vertex_elems[1].vertex_buffer_index = 1;
 429    vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
 430
 431    idct->vertex_elems_state = idct->pipe->create_vertex_elements_state(idct->pipe, 2, vertex_elems);
 432
 433    idct->vs_const_buf = pipe_buffer_create
 434    (
 435       idct->pipe->screen,
 436       PIPE_BIND_CONSTANT_BUFFER,
 437       sizeof(struct vertex_shader_consts)
 438    );
 439
 440    if(idct->vs_const_buf == NULL)
 441       return false;
 442
 443    return true;
 444 }
 445
 446 static void
 447 cleanup_buffers(struct vl_idct *idct)
 448 {
 449    unsigned i;
 450
 451    assert(idct);
 452
 453    pipe_resource_reference(&idct->vs_const_buf, NULL);
 454
 455    for (i = 0; i < 4; ++i) {
 456       pipe_sampler_view_reference(&idct->sampler_views.all[i], NULL);
 457       pipe_resource_reference(&idct->textures.all[i], NULL);
 458    }
 459
 460    idct->pipe->delete_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 461    pipe_resource_reference(&idct->vertex_bufs.individual.quad.buffer, NULL);
 462    pipe_resource_reference(&idct->vertex_bufs.individual.pos.buffer, NULL);
 463 }
 464
 465 static void
 466 init_constants(struct vl_idct *idct)
 467 {
 468    struct pipe_transfer *buf_transfer;
 469    struct vertex_shader_consts *vs_consts;
 470    struct vertex2f *v;
 471    float *f;
 472
 473    struct pipe_box rect =
 474    {
 475       0, 0, 0,
 476       BLOCK_WIDTH,
 477       BLOCK_HEIGHT,
 478       1
 479    };
 480
 481    unsigned i, j, pitch;
 482
 483    /* quad vectors */
 484    v = pipe_buffer_map
 485    (
 486       idct->pipe,
 487       idct->vertex_bufs.individual.quad.buffer,
 488       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 489       &buf_transfer
 490    );
 491    for ( i = 0; i < idct->max_blocks; ++i)
 492      memcpy(v + i * 4, &const_quad, sizeof(const_quad));
 493    pipe_buffer_unmap(idct->pipe, idct->vertex_bufs.individual.quad.buffer, buf_transfer);
 494
 495    /* transposed matrix */
 496    buf_transfer = idct->pipe->get_transfer
 497    (
 498       idct->pipe, idct->textures.individual.transpose,
 499       u_subresource(0, 0),
 500       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 501       &rect
 502    );
 503    pitch = buf_transfer->stride / util_format_get_blocksize(buf_transfer->resource->format);
 504
 505    f = idct->pipe->transfer_map(idct->pipe, buf_transfer);
 506    for(i = 0; i < BLOCK_HEIGHT; ++i)
 507       for(j = 0; j < BLOCK_WIDTH; ++j)
 508          f[i * pitch + j] = const_matrix[j][i]; // transpose
 509
 510    idct->pipe->transfer_unmap(idct->pipe, buf_transfer);
 511    idct->pipe->transfer_destroy(idct->pipe, buf_transfer);
 512
 513    /* matrix */
 514    buf_transfer = idct->pipe->get_transfer
 515    (
 516       idct->pipe, idct->textures.individual.matrix,
 517       u_subresource(0, 0),
 518       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 519       &rect
 520    );
 521    pitch = buf_transfer->stride / util_format_get_blocksize(buf_transfer->resource->format);
 522
 523    f = idct->pipe->transfer_map(idct->pipe, buf_transfer);
 524    for(i = 0; i < BLOCK_HEIGHT; ++i)
 525       for(j = 0; j < BLOCK_WIDTH; ++j)
 526          f[i * pitch + j] = const_matrix[i][j];
 527
 528    idct->pipe->transfer_unmap(idct->pipe, buf_transfer);
 529    idct->pipe->transfer_destroy(idct->pipe, buf_transfer);
 530
 531    /* normalisation constants */
 532    vs_consts = pipe_buffer_map
 533    (
 534       idct->pipe, idct->vs_const_buf,
 535       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 536       &buf_transfer
 537    );
 538
 539    vs_consts->norm.x = 1.0f / idct->destination->width0;
 540    vs_consts->norm.y = 1.0f / idct->destination->height0;
 541
 542    pipe_buffer_unmap(idct->pipe, idct->vs_const_buf, buf_transfer);
 543 }
 544
 545 static void
 546 init_state(struct vl_idct *idct)
 547 {
 548    struct pipe_sampler_state sampler;
 549    unsigned i;
 550
 551    idct->num_blocks = 0;
 552    idct->num_empty_blocks = 0;
 553
 554    idct->viewport.scale[0] = idct->destination->width0;
 555    idct->viewport.scale[1] = idct->destination->height0;
 556    idct->viewport.scale[2] = 1;
 557    idct->viewport.scale[3] = 1;
 558    idct->viewport.translate[0] = 0;
 559    idct->viewport.translate[1] = 0;
 560    idct->viewport.translate[2] = 0;
 561    idct->viewport.translate[3] = 0;
 562
 563    idct->fb_state.width = idct->destination->width0;
 564    idct->fb_state.height = idct->destination->height0;
 565    idct->fb_state.nr_cbufs = 1;
 566    idct->fb_state.zsbuf = NULL;
 567
 568    for (i = 0; i < 4; ++i) {
 569       memset(&sampler, 0, sizeof(sampler));
 570       sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 571       sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 572       sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 573       sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
 574       sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
 575       sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
 576       sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
 577       sampler.compare_func = PIPE_FUNC_ALWAYS;
 578       sampler.normalized_coords = 1;
 579       /*sampler.shadow_ambient = ; */
 580       /*sampler.lod_bias = ; */
 581       sampler.min_lod = 0;
 582       /*sampler.max_lod = ; */
 583       /*sampler.border_color[0] = ; */
 584       /*sampler.max_anisotropy = ; */
 585       idct->samplers.all[i] = idct->pipe->create_sampler_state(idct->pipe, &sampler);
 586    }
 587 }
 588
 589 static void
 590 cleanup_state(struct vl_idct *idct)
 591 {
 592    unsigned i;
 593
 594    for (i = 0; i < 4; ++i)
 595       idct->pipe->delete_sampler_state(idct->pipe, idct->samplers.all[i]);
 596 }
 597
 598 bool
 599 vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe, struct pipe_resource *dst)
 600 {
 601    assert(idct && pipe && dst);
 602
 603    idct->pipe = pipe;
 604    pipe_resource_reference(&idct->destination, dst);
 605
 606    init_state(idct);
 607
 608    if(!init_shaders(idct))
 609       return false;
 610
 611    if(!init_buffers(idct)) {
 612       cleanup_shaders(idct);
 613       return false;
 614    }
 615
 616    idct->surfaces.intermediate = idct->pipe->screen->get_tex_surface(
 617       idct->pipe->screen, idct->textures.individual.intermediate, 0, 0, 0,
 618       PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET);
 619
 620    idct->surfaces.destination = idct->pipe->screen->get_tex_surface(
 621       idct->pipe->screen, idct->destination, 0, 0, 0,
 622       PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET);
 623
 624    init_constants(idct);
 625    xfer_buffers_map(idct);
 626
 627    return true;
 628 }
 629
 630 void
 631 vl_idct_cleanup(struct vl_idct *idct)
 632 {
 633    idct->pipe->screen->tex_surface_destroy(idct->surfaces.destination);
 634    idct->pipe->screen->tex_surface_destroy(idct->surfaces.intermediate);
 635
 636    cleanup_shaders(idct);
 637    cleanup_buffers(idct);
 638
 639    cleanup_state(idct);
 640
 641    pipe_resource_reference(&idct->destination, NULL);
 642 }
 643
 644 void
 645 vl_idct_add_block(struct vl_idct *idct, unsigned x, unsigned y, short *block)
 646 {
 647    struct vertex2f v, *v_dst;
 648
 649    unsigned tex_pitch;
 650    short *texels;
 651
 652    unsigned i;
 653
 654    assert(idct);
 655
 656    if(block) {
 657       tex_pitch = idct->tex_transfer->stride / util_format_get_blocksize(idct->tex_transfer->resource->format);
 658       texels = idct->texels + y * tex_pitch * BLOCK_HEIGHT + x * BLOCK_WIDTH;
 659
 660       for (i = 0; i < BLOCK_HEIGHT; ++i)
 661          memcpy(texels + i * tex_pitch, block + i * BLOCK_WIDTH, BLOCK_WIDTH * 2);
 662
 663       /* non empty blocks fills the vector buffer from left to right */
 664       v_dst = idct->vectors + idct->num_blocks * 4;
 665
 666       idct->num_blocks++;
 667
 668    } else {
 669
 670       /* while empty blocks fills the vector buffer from right to left */
 671       v_dst = idct->vectors + (idct->max_blocks - idct->num_empty_blocks) * 4 - 4;
 672
 673       idct->num_empty_blocks++;
 674    }
 675
 676    v.x = x;
 677    v.y = y;
 678
 679    for (i = 0; i < 4; ++i) {
 680       v_dst[i] = v;
 681    }
 682 }
 683
 684 void
 685 vl_idct_flush(struct vl_idct *idct)
 686 {
 687    xfer_buffers_unmap(idct);
 688
 689    idct->pipe->set_constant_buffer(idct->pipe, PIPE_SHADER_VERTEX, 0, idct->vs_const_buf);
 690
 691    if(idct->num_blocks > 0) {
 692
 693       /* first stage */
 694       idct->fb_state.cbufs[0] = idct->surfaces.intermediate;
 695       idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state);
 696       idct->pipe->set_viewport_state(idct->pipe, &idct->viewport);
 697
 698       idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all);
 699       idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 700       idct->pipe->set_fragment_sampler_views(idct->pipe, 2, idct->sampler_views.stage[0]);
 701       idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[0]);
 702       idct->pipe->bind_vs_state(idct->pipe, idct->vs);
 703       idct->pipe->bind_fs_state(idct->pipe, idct->transpose_fs);
 704
 705       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, idct->num_blocks * 4);
 706
 707       /* second stage */
 708       idct->fb_state.cbufs[0] = idct->surfaces.destination;
 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, 2, idct->sampler_views.stage[1]);
 715       idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[1]);
 716       idct->pipe->bind_vs_state(idct->pipe, idct->vs);
 717       idct->pipe->bind_fs_state(idct->pipe, idct->matrix_fs);
 718
 719       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, idct->num_blocks * 4);
 720    }
 721
 722    if(idct->num_empty_blocks > 0) {
 723
 724       /* empty block handling */
 725       idct->fb_state.cbufs[0] = idct->surfaces.destination;
 726       idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state);
 727       idct->pipe->set_viewport_state(idct->pipe, &idct->viewport);
 728
 729       idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all);
 730       idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 731       idct->pipe->set_fragment_sampler_views(idct->pipe, 4, idct->sampler_views.all);
 732       idct->pipe->bind_fragment_sampler_states(idct->pipe, 4, idct->samplers.all);
 733       idct->pipe->bind_vs_state(idct->pipe, idct->vs);
 734       idct->pipe->bind_fs_state(idct->pipe, idct->eb_fs);
 735
 736       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS,
 737          (idct->max_blocks - idct->num_empty_blocks) * 4,
 738          idct->num_empty_blocks * 4);
 739    }
 740
 741    idct->num_blocks = 0;
 742    idct->num_empty_blocks = 0;
 743    xfer_buffers_map(idct);
 744 }