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 };
  67
  68 static const float const_matrix[8][8] = {
  69    {  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.353553f,  0.3535530f },
  70    {  0.4903930f,  0.4157350f,  0.2777850f,  0.0975451f, -0.0975452f, -0.2777850f, -0.415735f, -0.4903930f },
  71    {  0.4619400f,  0.1913420f, -0.1913420f, -0.4619400f, -0.4619400f, -0.1913420f,  0.191342f,  0.4619400f },
  72    {  0.4157350f, -0.0975452f, -0.4903930f, -0.2777850f,  0.2777850f,  0.4903930f,  0.097545f, -0.4157350f },
  73    {  0.3535530f, -0.3535530f, -0.3535530f,  0.3535540f,  0.3535530f, -0.3535540f, -0.353553f,  0.3535530f },
  74    {  0.2777850f, -0.4903930f,  0.0975452f,  0.4157350f, -0.4157350f, -0.0975451f,  0.490393f, -0.2777850f },
  75    {  0.1913420f, -0.4619400f,  0.4619400f, -0.1913420f, -0.1913410f,  0.4619400f, -0.461940f,  0.1913420f },
  76    {  0.0975451f, -0.2777850f,  0.4157350f, -0.4903930f,  0.4903930f, -0.4157350f,  0.277786f, -0.0975458f }
  77 };
  78
  79 /* vertices for a quad covering a block */
  80 static const struct vertex2f const_quad[4] = {
  81    {0.0f, 0.0f}, {1.0f, 0.0f}, {1.0f, 1.0f}, {0.0f, 1.0f}
  82 };
  83
  84 static void *
  85 create_vert_shader(struct vl_idct *idct)
  86 {
  87    struct ureg_program *shader;
  88    struct ureg_src scale;
  89    struct ureg_src vrect, vpos;
  90    struct ureg_dst t_vpos;
  91    struct ureg_dst o_vpos, o_block, o_tex, o_start;
  92
  93    shader = ureg_create(TGSI_PROCESSOR_VERTEX);
  94    if (!shader)
  95       return NULL;
  96
  97    scale = ureg_imm2f(shader,
  98       (float)BLOCK_WIDTH / idct->destination->width0,
  99       (float)BLOCK_HEIGHT / idct->destination->height0);
 100
 101    t_vpos = ureg_DECL_temporary(shader);
 102
 103    vrect = ureg_DECL_vs_input(shader, VS_I_RECT);
 104    vpos = ureg_DECL_vs_input(shader, VS_I_VPOS);
 105
 106    o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
 107    o_block = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK);
 108    o_tex = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX);
 109    o_start = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_START);
 110
 111    /*
 112     * t_vpos = vpos + vrect
 113     * o_vpos.xy = t_vpos * scale
 114     * o_vpos.zw = vpos
 115     *
 116     * o_block = vrect
 117     * o_tex = t_pos
 118     * o_start = vpos * scale
 119     *
 120     */
 121    ureg_ADD(shader, ureg_writemask(t_vpos, TGSI_WRITEMASK_XY), vpos, vrect);
 122    ureg_MUL(shader, ureg_writemask(t_vpos, TGSI_WRITEMASK_XY), ureg_src(t_vpos), scale);
 123    ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_XY), ureg_src(t_vpos));
 124    ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_ZW), vpos);
 125
 126    ureg_MOV(shader, ureg_writemask(o_block, TGSI_WRITEMASK_XY), vrect);
 127    ureg_MOV(shader, ureg_writemask(o_tex, TGSI_WRITEMASK_XY), ureg_src(t_vpos));
 128    ureg_MUL(shader, ureg_writemask(o_start, TGSI_WRITEMASK_XY), vpos, scale);
 129
 130    ureg_release_temporary(shader, t_vpos);
 131
 132    ureg_END(shader);
 133
 134    return ureg_create_shader_and_destroy(shader, idct->pipe);
 135 }
 136
 137 static void
 138 fetch_one(struct ureg_program *shader, struct ureg_dst m[2],
 139           struct ureg_src tc, struct ureg_src sampler,
 140           struct ureg_src start, bool right_side, float size)
 141 {
 142    struct ureg_dst t_tc, tmp;
 143    unsigned i, j;
 144
 145    t_tc = ureg_DECL_temporary(shader);
 146    tmp = ureg_DECL_temporary(shader);
 147
 148    m[0] = ureg_DECL_temporary(shader);
 149    m[1] = ureg_DECL_temporary(shader);
 150
 151    /*
 152     * t_tc.x = right_side ? start.x : tc.x
 153     * t_tc.y = right_side ? tc.y : start.y
 154     * m[0..1].xyzw = tex(t_tc++, sampler)
 155     */
 156    if(right_side) {
 157       ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_X), ureg_scalar(tc, TGSI_SWIZZLE_X));
 158       ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Y), ureg_scalar(start, TGSI_SWIZZLE_Y));
 159    } else {
 160       ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_X), ureg_scalar(start, TGSI_SWIZZLE_X));
 161       ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Y), ureg_scalar(tc, TGSI_SWIZZLE_Y));
 162    }
 163    for(i = 0; i < 2; ++i) {
 164       for(j = 0; j < 4; ++j) {
 165          /* Nouveau and r600g can't writemask tex dst regs (yet?), do in two steps */
 166          ureg_TEX(shader, tmp, TGSI_TEXTURE_2D, ureg_src(t_tc), sampler);
 167          ureg_MOV(shader, ureg_writemask(m[i], TGSI_WRITEMASK_X << j), ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_X));
 168
 169          if(i != 1 || j != 3) /* skip the last add */
 170             ureg_ADD(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_X << right_side),
 171                ureg_src(t_tc), ureg_imm1f(shader, 1.0f / size));
 172       }
 173    }
 174
 175    ureg_release_temporary(shader, t_tc);
 176    ureg_release_temporary(shader, tmp);
 177 }
 178
 179 static void
 180 fetch_four(struct ureg_program *shader, struct ureg_dst m[2],
 181            struct ureg_src tc, struct ureg_src sampler,
 182            struct ureg_src start, bool right_side, float size)
 183 {
 184    struct ureg_dst t_tc;
 185
 186    t_tc = ureg_DECL_temporary(shader);
 187    m[0] = ureg_DECL_temporary(shader);
 188    m[1] = ureg_DECL_temporary(shader);
 189
 190    /*
 191     * t_tc.x = right_side ? start.x : tc.x
 192     * t_tc.y = right_side ? tc.y : start.y
 193     * m[0..1] = tex(t_tc++, sampler)
 194     */
 195    if(right_side) {
 196       ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_X), ureg_scalar(start, TGSI_SWIZZLE_Y));
 197       ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Y), ureg_scalar(tc, TGSI_SWIZZLE_X));
 198    } else {
 199       ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_X), ureg_scalar(start, TGSI_SWIZZLE_X));
 200       ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Y), ureg_scalar(tc, TGSI_SWIZZLE_Y));
 201    }
 202
 203    ureg_TEX(shader, m[0], TGSI_TEXTURE_2D, ureg_src(t_tc), sampler);
 204    ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_X), ureg_imm1f(shader, 4.0f / size));
 205    ureg_TEX(shader, m[1], TGSI_TEXTURE_2D, ureg_src(t_tc), sampler);
 206
 207    ureg_release_temporary(shader, t_tc);
 208 }
 209
 210 static struct ureg_dst
 211 matrix_mul(struct ureg_program *shader, struct ureg_dst m[2][2])
 212 {
 213    struct ureg_dst dst, tmp[2];
 214    unsigned i;
 215
 216    dst = ureg_DECL_temporary(shader);
 217    for(i = 0; i < 2; ++i) {
 218       tmp[i] = ureg_DECL_temporary(shader);
 219    }
 220
 221    /*
 222     * tmp[0..1] = dot4(m[0][0..1], m[1][0..1])
 223     * dst = tmp[0] + tmp[1]
 224     */
 225    ureg_DP4(shader, ureg_writemask(tmp[0], TGSI_WRITEMASK_X), ureg_src(m[0][0]), ureg_src(m[1][0]));
 226    ureg_DP4(shader, ureg_writemask(tmp[1], TGSI_WRITEMASK_X), ureg_src(m[0][1]), ureg_src(m[1][1]));
 227    ureg_ADD(shader, ureg_writemask(dst, TGSI_WRITEMASK_X), ureg_src(tmp[0]), ureg_src(tmp[1]));
 228
 229    for(i = 0; i < 2; ++i) {
 230       ureg_release_temporary(shader, tmp[i]);
 231    }
 232
 233    return dst;
 234 }
 235
 236 static void *
 237 create_transpose_frag_shader(struct vl_idct *idct)
 238 {
 239    struct ureg_program *shader;
 240
 241    struct ureg_src tc[2], sampler[2];
 242    struct ureg_src start[2];
 243
 244    struct ureg_dst m[2][2];
 245    struct ureg_dst tmp, fragment;
 246
 247    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 248    if (!shader)
 249       return NULL;
 250
 251    tc[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK, TGSI_INTERPOLATE_LINEAR);
 252    tc[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX, TGSI_INTERPOLATE_LINEAR);
 253
 254    sampler[0] = ureg_DECL_sampler(shader, 0);
 255    sampler[1] = ureg_DECL_sampler(shader, 1);
 256
 257    start[0] = ureg_imm1f(shader, 0.0f);
 258    start[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_START, TGSI_INTERPOLATE_CONSTANT);
 259
 260    fetch_four(shader, m[0], tc[0], sampler[0], start[0], false, BLOCK_WIDTH);
 261    fetch_one(shader, m[1], tc[1], sampler[1], start[1], true, idct->destination->height0);
 262
 263    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 264
 265    tmp = matrix_mul(shader, m);
 266    ureg_MUL(shader, fragment, ureg_src(tmp), ureg_imm1f(shader, STAGE2_SCALE));
 267
 268    ureg_release_temporary(shader, tmp);
 269    ureg_release_temporary(shader, m[0][0]);
 270    ureg_release_temporary(shader, m[0][1]);
 271    ureg_release_temporary(shader, m[1][0]);
 272    ureg_release_temporary(shader, m[1][1]);
 273
 274    ureg_END(shader);
 275
 276    return ureg_create_shader_and_destroy(shader, idct->pipe);
 277 }
 278
 279 static void *
 280 create_matrix_frag_shader(struct vl_idct *idct)
 281 {
 282    struct ureg_program *shader;
 283
 284    struct ureg_src tc[2], sampler[2];
 285    struct ureg_src start[2];
 286
 287    struct ureg_dst m[2][2];
 288    struct ureg_dst tmp, fragment;
 289
 290    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 291    if (!shader)
 292       return NULL;
 293
 294    tc[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX, TGSI_INTERPOLATE_LINEAR);
 295    tc[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK, TGSI_INTERPOLATE_LINEAR);
 296
 297    sampler[0] = ureg_DECL_sampler(shader, 1);
 298    sampler[1] = ureg_DECL_sampler(shader, 0);
 299
 300    start[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_START, TGSI_INTERPOLATE_CONSTANT);
 301    start[1] = ureg_imm1f(shader, 0.0f);
 302
 303    fetch_four(shader, m[0], tc[0], sampler[0], start[0], false, idct->destination->width0);
 304    fetch_four(shader, m[1], tc[1], sampler[1], start[1], true, BLOCK_HEIGHT);
 305
 306    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 307
 308    tmp = matrix_mul(shader, m);
 309    ureg_MUL(shader, fragment, ureg_src(tmp), ureg_imm1f(shader, STAGE1_SCALE));
 310
 311    ureg_release_temporary(shader, tmp);
 312    ureg_release_temporary(shader, m[0][0]);
 313    ureg_release_temporary(shader, m[0][1]);
 314    ureg_release_temporary(shader, m[1][0]);
 315    ureg_release_temporary(shader, m[1][1]);
 316
 317    ureg_END(shader);
 318
 319    return ureg_create_shader_and_destroy(shader, idct->pipe);
 320 }
 321
 322 static void *
 323 create_empty_block_frag_shader(struct vl_idct *idct)
 324 {
 325    struct ureg_program *shader;
 326    struct ureg_dst fragment;
 327
 328    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 329    if (!shader)
 330       return NULL;
 331
 332    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 333
 334    ureg_MOV(shader, fragment, ureg_imm1f(shader, 0.0f));
 335
 336    ureg_END(shader);
 337
 338    return ureg_create_shader_and_destroy(shader, idct->pipe);
 339 }
 340
 341 static void
 342 xfer_buffers_map(struct vl_idct *idct)
 343 {
 344    struct pipe_box rect =
 345    {
 346       0, 0, 0,
 347       idct->destination->width0,
 348       idct->destination->height0,
 349       1
 350    };
 351
 352    idct->tex_transfer = idct->pipe->get_transfer
 353    (
 354       idct->pipe, idct->textures.individual.source,
 355       u_subresource(0, 0),
 356       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 357       &rect
 358    );
 359
 360    idct->texels = idct->pipe->transfer_map(idct->pipe, idct->tex_transfer);
 361
 362    idct->vectors = pipe_buffer_map
 363    (
 364       idct->pipe,
 365       idct->vertex_bufs.individual.pos.buffer,
 366       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 367       &idct->vec_transfer
 368    );
 369 }
 370
 371 static void
 372 xfer_buffers_unmap(struct vl_idct *idct)
 373 {
 374    pipe_buffer_unmap(idct->pipe, idct->vertex_bufs.individual.pos.buffer, idct->vec_transfer);
 375
 376    idct->pipe->transfer_unmap(idct->pipe, idct->tex_transfer);
 377    idct->pipe->transfer_destroy(idct->pipe, idct->tex_transfer);
 378 }
 379
 380 static bool
 381 init_shaders(struct vl_idct *idct)
 382 {
 383    idct->vs = create_vert_shader(idct);
 384    idct->transpose_fs = create_transpose_frag_shader(idct);
 385    idct->matrix_fs = create_matrix_frag_shader(idct);
 386    idct->eb_fs = create_empty_block_frag_shader(idct);
 387
 388    return
 389       idct->vs != NULL &&
 390       idct->transpose_fs != NULL &&
 391       idct->matrix_fs != NULL &&
 392       idct->eb_fs != NULL;
 393 }
 394
 395 static void
 396 cleanup_shaders(struct vl_idct *idct)
 397 {
 398    idct->pipe->delete_vs_state(idct->pipe, idct->vs);
 399    idct->pipe->delete_fs_state(idct->pipe, idct->transpose_fs);
 400    idct->pipe->delete_fs_state(idct->pipe, idct->matrix_fs);
 401    idct->pipe->delete_fs_state(idct->pipe, idct->eb_fs);
 402 }
 403
 404 static bool
 405 init_buffers(struct vl_idct *idct)
 406 {
 407    struct pipe_resource template;
 408    struct pipe_sampler_view sampler_view;
 409    struct pipe_vertex_element vertex_elems[2];
 410    unsigned i;
 411
 412    idct->max_blocks =
 413       align(idct->destination->width0, BLOCK_WIDTH) / BLOCK_WIDTH *
 414       align(idct->destination->height0, BLOCK_HEIGHT) / BLOCK_HEIGHT *
 415       idct->destination->depth0;
 416
 417    memset(&template, 0, sizeof(struct pipe_resource));
 418    template.target = PIPE_TEXTURE_2D;
 419    template.format = PIPE_FORMAT_R32G32B32A32_FLOAT;
 420    template.last_level = 0;
 421    template.width0 = 2;
 422    template.height0 = 8;
 423    template.depth0 = 1;
 424    template.usage = PIPE_USAGE_IMMUTABLE;
 425    template.bind = PIPE_BIND_SAMPLER_VIEW;
 426    template.flags = 0;
 427
 428    template.format = PIPE_FORMAT_R16G16B16A16_SNORM;
 429    template.width0 = idct->destination->width0;
 430    template.height0 = idct->destination->height0;
 431    template.depth0 = idct->destination->depth0;
 432    template.usage = PIPE_USAGE_STREAM;
 433    idct->textures.individual.source = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 434
 435    template.format = idct->destination->format;
 436    template.usage = PIPE_USAGE_STATIC;
 437    idct->textures.individual.intermediate = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 438
 439    for (i = 0; i < 4; ++i) {
 440       if(idct->textures.all[i] == NULL)
 441          return false; /* a texture failed to allocate */
 442
 443       u_sampler_view_default_template(&sampler_view, idct->textures.all[i], idct->textures.all[i]->format);
 444       idct->sampler_views.all[i] = idct->pipe->create_sampler_view(idct->pipe, idct->textures.all[i], &sampler_view);
 445    }
 446
 447    idct->vertex_bufs.individual.quad.stride = sizeof(struct vertex2f);
 448    idct->vertex_bufs.individual.quad.max_index = 4 * idct->max_blocks - 1;
 449    idct->vertex_bufs.individual.quad.buffer_offset = 0;
 450    idct->vertex_bufs.individual.quad.buffer = pipe_buffer_create
 451    (
 452       idct->pipe->screen,
 453       PIPE_BIND_VERTEX_BUFFER,
 454       sizeof(struct vertex2f) * 4 * idct->max_blocks
 455    );
 456
 457    if(idct->vertex_bufs.individual.quad.buffer == NULL)
 458       return false;
 459
 460    idct->vertex_bufs.individual.pos.stride = sizeof(struct vertex2f);
 461    idct->vertex_bufs.individual.pos.max_index = 4 * idct->max_blocks - 1;
 462    idct->vertex_bufs.individual.pos.buffer_offset = 0;
 463    idct->vertex_bufs.individual.pos.buffer = pipe_buffer_create
 464    (
 465       idct->pipe->screen,
 466       PIPE_BIND_VERTEX_BUFFER,
 467       sizeof(struct vertex2f) * 4 * idct->max_blocks
 468    );
 469
 470    if(idct->vertex_bufs.individual.pos.buffer == NULL)
 471       return false;
 472
 473    /* Rect element */
 474    vertex_elems[0].src_offset = 0;
 475    vertex_elems[0].instance_divisor = 0;
 476    vertex_elems[0].vertex_buffer_index = 0;
 477    vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
 478
 479    /* Pos element */
 480    vertex_elems[1].src_offset = 0;
 481    vertex_elems[1].instance_divisor = 0;
 482    vertex_elems[1].vertex_buffer_index = 1;
 483    vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
 484
 485    idct->vertex_elems_state = idct->pipe->create_vertex_elements_state(idct->pipe, 2, vertex_elems);
 486
 487    return true;
 488 }
 489
 490 static void
 491 cleanup_buffers(struct vl_idct *idct)
 492 {
 493    unsigned i;
 494
 495    assert(idct);
 496
 497    for (i = 0; i < 4; ++i) {
 498       pipe_sampler_view_reference(&idct->sampler_views.all[i], NULL);
 499       pipe_resource_reference(&idct->textures.all[i], NULL);
 500    }
 501
 502    idct->pipe->delete_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 503    pipe_resource_reference(&idct->vertex_bufs.individual.quad.buffer, NULL);
 504    pipe_resource_reference(&idct->vertex_bufs.individual.pos.buffer, NULL);
 505 }
 506
 507 static void
 508 init_constants(struct vl_idct *idct)
 509 {
 510    struct pipe_transfer *buf_transfer;
 511    struct vertex2f *v;
 512
 513    unsigned i;
 514
 515    /* quad vectors */
 516    v = pipe_buffer_map
 517    (
 518       idct->pipe,
 519       idct->vertex_bufs.individual.quad.buffer,
 520       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 521       &buf_transfer
 522    );
 523    for ( i = 0; i < idct->max_blocks; ++i)
 524      memcpy(v + i * 4, &const_quad, sizeof(const_quad));
 525    pipe_buffer_unmap(idct->pipe, idct->vertex_bufs.individual.quad.buffer, buf_transfer);
 526 }
 527
 528 static void
 529 init_state(struct vl_idct *idct)
 530 {
 531    struct pipe_sampler_state sampler;
 532    unsigned i;
 533
 534    idct->num_blocks = 0;
 535    idct->num_empty_blocks = 0;
 536
 537    idct->viewport.scale[0] = idct->destination->width0;
 538    idct->viewport.scale[1] = idct->destination->height0;
 539    idct->viewport.scale[2] = 1;
 540    idct->viewport.scale[3] = 1;
 541    idct->viewport.translate[0] = 0;
 542    idct->viewport.translate[1] = 0;
 543    idct->viewport.translate[2] = 0;
 544    idct->viewport.translate[3] = 0;
 545
 546    idct->fb_state.width = idct->destination->width0;
 547    idct->fb_state.height = idct->destination->height0;
 548    idct->fb_state.nr_cbufs = 1;
 549    idct->fb_state.zsbuf = NULL;
 550
 551    for (i = 0; i < 4; ++i) {
 552       memset(&sampler, 0, sizeof(sampler));
 553       sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 554       sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 555       sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 556       sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
 557       sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
 558       sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
 559       sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
 560       sampler.compare_func = PIPE_FUNC_ALWAYS;
 561       sampler.normalized_coords = 1;
 562       /*sampler.shadow_ambient = ; */
 563       /*sampler.lod_bias = ; */
 564       sampler.min_lod = 0;
 565       /*sampler.max_lod = ; */
 566       /*sampler.border_color[0] = ; */
 567       /*sampler.max_anisotropy = ; */
 568       idct->samplers.all[i] = idct->pipe->create_sampler_state(idct->pipe, &sampler);
 569    }
 570 }
 571
 572 static void
 573 cleanup_state(struct vl_idct *idct)
 574 {
 575    unsigned i;
 576
 577    for (i = 0; i < 4; ++i)
 578       idct->pipe->delete_sampler_state(idct->pipe, idct->samplers.all[i]);
 579 }
 580
 581 struct pipe_resource *
 582 vl_idct_upload_matrix(struct pipe_context *pipe)
 583 {
 584    struct pipe_resource template, *matrix;
 585    struct pipe_transfer *buf_transfer;
 586    unsigned i, j, pitch;
 587    float *f;
 588
 589    struct pipe_box rect =
 590    {
 591       0, 0, 0,
 592       BLOCK_WIDTH,
 593       BLOCK_HEIGHT,
 594       1
 595    };
 596
 597    memset(&template, 0, sizeof(struct pipe_resource));
 598    template.target = PIPE_TEXTURE_2D;
 599    template.format = PIPE_FORMAT_R32G32B32A32_FLOAT;
 600    template.last_level = 0;
 601    template.width0 = 2;
 602    template.height0 = 8;
 603    template.depth0 = 1;
 604    template.usage = PIPE_USAGE_IMMUTABLE;
 605    template.bind = PIPE_BIND_SAMPLER_VIEW;
 606    template.flags = 0;
 607
 608    matrix = pipe->screen->resource_create(pipe->screen, &template);
 609
 610    /* matrix */
 611    buf_transfer = pipe->get_transfer
 612    (
 613       pipe, matrix,
 614       u_subresource(0, 0),
 615       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 616       &rect
 617    );
 618    pitch = buf_transfer->stride / util_format_get_blocksize(buf_transfer->resource->format);
 619
 620    f = pipe->transfer_map(pipe, buf_transfer);
 621    for(i = 0; i < BLOCK_HEIGHT; ++i)
 622       for(j = 0; j < BLOCK_WIDTH; ++j)
 623          f[i * pitch * 4 + j] = const_matrix[j][i]; // transpose
 624
 625    pipe->transfer_unmap(pipe, buf_transfer);
 626    pipe->transfer_destroy(pipe, buf_transfer);
 627
 628    return matrix;
 629 }
 630
 631 bool
 632 vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe, struct pipe_resource *dst, struct pipe_resource *matrix)
 633 {
 634    assert(idct && pipe && dst);
 635
 636    idct->pipe = pipe;
 637    pipe_resource_reference(&idct->textures.individual.matrix, matrix);
 638    pipe_resource_reference(&idct->textures.individual.transpose, matrix);
 639    pipe_resource_reference(&idct->destination, dst);
 640
 641    init_state(idct);
 642
 643    if(!init_shaders(idct))
 644       return false;
 645
 646    if(!init_buffers(idct)) {
 647       cleanup_shaders(idct);
 648       return false;
 649    }
 650
 651    idct->surfaces.intermediate = idct->pipe->screen->get_tex_surface(
 652       idct->pipe->screen, idct->textures.individual.intermediate, 0, 0, 0,
 653       PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET);
 654
 655    idct->surfaces.destination = idct->pipe->screen->get_tex_surface(
 656       idct->pipe->screen, idct->destination, 0, 0, 0,
 657       PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET);
 658
 659    init_constants(idct);
 660    xfer_buffers_map(idct);
 661
 662    return true;
 663 }
 664
 665 void
 666 vl_idct_cleanup(struct vl_idct *idct)
 667 {
 668    idct->pipe->screen->tex_surface_destroy(idct->surfaces.destination);
 669    idct->pipe->screen->tex_surface_destroy(idct->surfaces.intermediate);
 670
 671    cleanup_shaders(idct);
 672    cleanup_buffers(idct);
 673
 674    cleanup_state(idct);
 675
 676    pipe_resource_reference(&idct->destination, NULL);
 677 }
 678
 679 void
 680 vl_idct_add_block(struct vl_idct *idct, unsigned x, unsigned y, short *block)
 681 {
 682    struct vertex2f v, *v_dst;
 683
 684    unsigned tex_pitch;
 685    short *texels;
 686
 687    unsigned i;
 688
 689    assert(idct);
 690
 691    if(block) {
 692       tex_pitch = idct->tex_transfer->stride / util_format_get_blocksize(idct->tex_transfer->resource->format);
 693       texels = idct->texels + (y * tex_pitch * BLOCK_HEIGHT + x * BLOCK_WIDTH) * 4;
 694
 695       for (i = 0; i < BLOCK_HEIGHT; ++i)
 696          memcpy(texels + i * tex_pitch * 4, block + i * BLOCK_WIDTH, BLOCK_WIDTH * 2);
 697
 698       /* non empty blocks fills the vector buffer from left to right */
 699       v_dst = idct->vectors + idct->num_blocks * 4;
 700
 701       idct->num_blocks++;
 702
 703    } else {
 704
 705       /* while empty blocks fills the vector buffer from right to left */
 706       v_dst = idct->vectors + (idct->max_blocks - idct->num_empty_blocks) * 4 - 4;
 707
 708       idct->num_empty_blocks++;
 709    }
 710
 711    v.x = x;
 712    v.y = y;
 713
 714    for (i = 0; i < 4; ++i) {
 715       v_dst[i] = v;
 716    }
 717 }
 718
 719 void
 720 vl_idct_flush(struct vl_idct *idct)
 721 {
 722    xfer_buffers_unmap(idct);
 723
 724    if(idct->num_blocks > 0) {
 725
 726       /* first stage */
 727       idct->fb_state.cbufs[0] = idct->surfaces.intermediate;
 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[0]);
 734       idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[0]);
 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       /* second stage */
 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, 2, idct->sampler_views.stage[1]);
 748       idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[1]);
 749       idct->pipe->bind_vs_state(idct->pipe, idct->vs);
 750       idct->pipe->bind_fs_state(idct->pipe, idct->transpose_fs);
 751
 752       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, idct->num_blocks * 4);
 753    }
 754
 755    if(idct->num_empty_blocks > 0) {
 756
 757       /* empty block handling */
 758       idct->fb_state.cbufs[0] = idct->surfaces.destination;
 759       idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state);
 760       idct->pipe->set_viewport_state(idct->pipe, &idct->viewport);
 761
 762       idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all);
 763       idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 764       idct->pipe->set_fragment_sampler_views(idct->pipe, 4, idct->sampler_views.all);
 765       idct->pipe->bind_fragment_sampler_states(idct->pipe, 4, idct->samplers.all);
 766       idct->pipe->bind_vs_state(idct->pipe, idct->vs);
 767       idct->pipe->bind_fs_state(idct->pipe, idct->eb_fs);
 768
 769       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS,
 770          (idct->max_blocks - idct->num_empty_blocks) * 4,
 771          idct->num_empty_blocks * 4);
 772    }
 773
 774    idct->num_blocks = 0;
 775    idct->num_empty_blocks = 0;
 776    xfer_buffers_map(idct);
 777 }