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 "vl_vertex_buffers.h"
  30 #include "util/u_draw.h"
  31 #include <assert.h>
  32 #include <pipe/p_context.h>
  33 #include <pipe/p_screen.h>
  34 #include <util/u_inlines.h>
  35 #include <util/u_sampler.h>
  36 #include <util/u_format.h>
  37 #include <tgsi/tgsi_ureg.h>
  38 #include "vl_types.h"
  39
  40 #define BLOCK_WIDTH 8
  41 #define BLOCK_HEIGHT 8
  42
  43 #define SCALE_FACTOR_16_TO_9 (32768.0f / 256.0f)
  44
  45 #define STAGE1_SCALE 4.0f
  46 #define STAGE2_SCALE (SCALE_FACTOR_16_TO_9 / STAGE1_SCALE)
  47
  48 #define NR_RENDER_TARGETS 1
  49
  50 struct vertex_shader_consts
  51 {
  52    struct vertex4f norm;
  53 };
  54
  55 enum VS_INPUT
  56 {
  57    VS_I_RECT,
  58    VS_I_VPOS,
  59
  60    NUM_VS_INPUTS
  61 };
  62
  63 enum VS_OUTPUT
  64 {
  65    VS_O_VPOS,
  66    VS_O_BLOCK,
  67    VS_O_TEX,
  68    VS_O_START
  69 };
  70
  71 static const float const_matrix[8][8] = {
  72    {  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.353553f,  0.3535530f },
  73    {  0.4903930f,  0.4157350f,  0.2777850f,  0.0975451f, -0.0975452f, -0.2777850f, -0.415735f, -0.4903930f },
  74    {  0.4619400f,  0.1913420f, -0.1913420f, -0.4619400f, -0.4619400f, -0.1913420f,  0.191342f,  0.4619400f },
  75    {  0.4157350f, -0.0975452f, -0.4903930f, -0.2777850f,  0.2777850f,  0.4903930f,  0.097545f, -0.4157350f },
  76    {  0.3535530f, -0.3535530f, -0.3535530f,  0.3535540f,  0.3535530f, -0.3535540f, -0.353553f,  0.3535530f },
  77    {  0.2777850f, -0.4903930f,  0.0975452f,  0.4157350f, -0.4157350f, -0.0975451f,  0.490393f, -0.2777850f },
  78    {  0.1913420f, -0.4619400f,  0.4619400f, -0.1913420f, -0.1913410f,  0.4619400f, -0.461940f,  0.1913420f },
  79    {  0.0975451f, -0.2777850f,  0.4157350f, -0.4903930f,  0.4903930f, -0.4157350f,  0.277786f, -0.0975458f }
  80 };
  81
  82 static void *
  83 create_vert_shader(struct vl_idct *idct, bool calc_src_cords)
  84 {
  85    struct ureg_program *shader;
  86    struct ureg_src scale;
  87    struct ureg_src vrect, vpos;
  88    struct ureg_dst t_vpos;
  89    struct ureg_dst o_vpos, o_block, o_tex, o_start;
  90
  91    shader = ureg_create(TGSI_PROCESSOR_VERTEX);
  92    if (!shader)
  93       return NULL;
  94
  95    t_vpos = ureg_DECL_temporary(shader);
  96
  97    vrect = ureg_DECL_vs_input(shader, VS_I_RECT);
  98    vpos = ureg_DECL_vs_input(shader, VS_I_VPOS);
  99
 100    o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
 101
 102    /*
 103     * scale = (BLOCK_WIDTH, BLOCK_HEIGHT) / (dst.width, dst.height)
 104     *
 105     * t_vpos = vpos + vrect
 106     * o_vpos.xy = t_vpos * scale
 107     * o_vpos.zw = vpos
 108     *
 109     * o_block = vrect
 110     * o_tex = t_pos
 111     * o_start = vpos * scale
 112     *
 113     */
 114    scale = ureg_imm2f(shader,
 115       (float)BLOCK_WIDTH / idct->destination->width0,
 116       (float)BLOCK_HEIGHT / idct->destination->height0);
 117
 118    ureg_ADD(shader, ureg_writemask(t_vpos, TGSI_WRITEMASK_XY), vpos, vrect);
 119    ureg_MUL(shader, ureg_writemask(t_vpos, TGSI_WRITEMASK_XY), ureg_src(t_vpos), scale);
 120    ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_XY), ureg_src(t_vpos));
 121    ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_ZW), vpos);
 122
 123    if(calc_src_cords) {
 124       o_block = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK);
 125       o_tex = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX);
 126       o_start = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_START);
 127
 128       ureg_MOV(shader, ureg_writemask(o_block, TGSI_WRITEMASK_XY), vrect);
 129       ureg_MOV(shader, ureg_writemask(o_tex, TGSI_WRITEMASK_XY), ureg_src(t_vpos));
 130       ureg_MUL(shader, ureg_writemask(o_start, TGSI_WRITEMASK_XY), vpos, scale);
 131    }
 132
 133    ureg_release_temporary(shader, t_vpos);
 134
 135    ureg_END(shader);
 136
 137    return ureg_create_shader_and_destroy(shader, idct->pipe);
 138 }
 139
 140 static void
 141 fetch_four(struct ureg_program *shader, struct ureg_dst m[2],
 142            struct ureg_src tc, struct ureg_src sampler,
 143            struct ureg_src start, struct ureg_src block,
 144            bool right_side, bool transposed, float size)
 145 {
 146    struct ureg_dst t_tc;
 147    unsigned wm_start = (right_side == transposed) ? TGSI_WRITEMASK_X : TGSI_WRITEMASK_Y;
 148    unsigned wm_tc = (right_side == transposed) ? TGSI_WRITEMASK_Y : TGSI_WRITEMASK_X;
 149
 150    t_tc = ureg_DECL_temporary(shader);
 151    m[0] = ureg_DECL_temporary(shader);
 152    m[1] = ureg_DECL_temporary(shader);
 153
 154    /*
 155     * t_tc.x = right_side ? start.x : tc.x
 156     * t_tc.y = right_side ? tc.y : start.y
 157     * m[0..1] = tex(t_tc++, sampler)
 158     */
 159    if(!right_side) {
 160       ureg_MOV(shader, ureg_writemask(t_tc, wm_start), ureg_scalar(start, TGSI_SWIZZLE_X));
 161       ureg_MOV(shader, ureg_writemask(t_tc, wm_tc), ureg_scalar(tc, TGSI_SWIZZLE_Y));
 162    } else {
 163       ureg_MOV(shader, ureg_writemask(t_tc, wm_start), ureg_scalar(start, TGSI_SWIZZLE_Y));
 164       ureg_MOV(shader, ureg_writemask(t_tc, wm_tc), ureg_scalar(tc, TGSI_SWIZZLE_X));
 165    }
 166
 167 #if NR_RENDER_TARGETS == 8
 168    ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Z), ureg_scalar(block, TGSI_SWIZZLE_X));
 169 #else
 170    ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Z), ureg_imm1f(shader, 0.0f));
 171 #endif
 172
 173    ureg_TEX(shader, m[0], TGSI_TEXTURE_3D, ureg_src(t_tc), sampler);
 174    ureg_ADD(shader, ureg_writemask(t_tc, wm_start), ureg_src(t_tc), ureg_imm1f(shader, 1.0f / size));
 175    ureg_TEX(shader, m[1], TGSI_TEXTURE_3D, ureg_src(t_tc), sampler);
 176
 177    ureg_release_temporary(shader, t_tc);
 178 }
 179
 180 static void
 181 matrix_mul(struct ureg_program *shader, struct ureg_dst dst, struct ureg_dst l[2], struct ureg_dst r[2])
 182 {
 183    struct ureg_dst tmp[2];
 184    unsigned i;
 185
 186    for(i = 0; i < 2; ++i) {
 187       tmp[i] = ureg_DECL_temporary(shader);
 188    }
 189
 190    /*
 191     * tmp[0..1] = dot4(m[0][0..1], m[1][0..1])
 192     * dst = tmp[0] + tmp[1]
 193     */
 194    ureg_DP4(shader, ureg_writemask(tmp[0], TGSI_WRITEMASK_X), ureg_src(l[0]), ureg_src(r[0]));
 195    ureg_DP4(shader, ureg_writemask(tmp[1], TGSI_WRITEMASK_X), ureg_src(l[1]), ureg_src(r[1]));
 196    ureg_ADD(shader, dst,
 197       ureg_scalar(ureg_src(tmp[0]), TGSI_SWIZZLE_X),
 198       ureg_scalar(ureg_src(tmp[1]), TGSI_SWIZZLE_X));
 199
 200    for(i = 0; i < 2; ++i) {
 201       ureg_release_temporary(shader, tmp[i]);
 202    }
 203 }
 204
 205 static void *
 206 create_transpose_frag_shader(struct vl_idct *idct)
 207 {
 208    struct pipe_resource *transpose = idct->textures.individual.transpose;
 209    struct pipe_resource *intermediate = idct->textures.individual.intermediate;
 210
 211    struct ureg_program *shader;
 212
 213    struct ureg_src block, tex, sampler[2];
 214    struct ureg_src start[2];
 215
 216    struct ureg_dst m[2][2];
 217    struct ureg_dst tmp, fragment;
 218
 219    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 220    if (!shader)
 221       return NULL;
 222
 223    block = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK, TGSI_INTERPOLATE_LINEAR);
 224    tex = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX, TGSI_INTERPOLATE_CONSTANT);
 225
 226    sampler[0] = ureg_DECL_sampler(shader, 0);
 227    sampler[1] = ureg_DECL_sampler(shader, 1);
 228
 229    start[0] = ureg_imm1f(shader, 0.0f);
 230    start[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_START, TGSI_INTERPOLATE_CONSTANT);
 231
 232    fetch_four(shader, m[0], block, sampler[0], start[0], block, false, false, transpose->width0);
 233    fetch_four(shader, m[1], tex, sampler[1], start[1], block, true, false, intermediate->height0);
 234
 235    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 236
 237    tmp = ureg_DECL_temporary(shader);
 238    matrix_mul(shader, ureg_writemask(tmp, TGSI_WRITEMASK_X), m[0], m[1]);
 239    ureg_MUL(shader, fragment, ureg_src(tmp), ureg_imm1f(shader, STAGE2_SCALE));
 240
 241    ureg_release_temporary(shader, tmp);
 242    ureg_release_temporary(shader, m[0][0]);
 243    ureg_release_temporary(shader, m[0][1]);
 244    ureg_release_temporary(shader, m[1][0]);
 245    ureg_release_temporary(shader, m[1][1]);
 246
 247    ureg_END(shader);
 248
 249    return ureg_create_shader_and_destroy(shader, idct->pipe);
 250 }
 251
 252 static void *
 253 create_matrix_frag_shader(struct vl_idct *idct)
 254 {
 255    struct pipe_resource *matrix = idct->textures.individual.matrix;
 256    struct pipe_resource *source = idct->textures.individual.source;
 257
 258    struct ureg_program *shader;
 259
 260    struct ureg_src tex, block, sampler[2];
 261    struct ureg_src start[2];
 262
 263    struct ureg_dst l[4][2], r[2];
 264    struct ureg_dst t_tc, tmp, fragment[NR_RENDER_TARGETS];
 265
 266    unsigned i, j;
 267
 268    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 269    if (!shader)
 270       return NULL;
 271
 272    t_tc = ureg_DECL_temporary(shader);
 273    tmp = ureg_DECL_temporary(shader);
 274
 275    tex = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX, TGSI_INTERPOLATE_LINEAR);
 276    block = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK, TGSI_INTERPOLATE_LINEAR);
 277
 278    sampler[0] = ureg_DECL_sampler(shader, 1);
 279    sampler[1] = ureg_DECL_sampler(shader, 0);
 280
 281    start[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_START, TGSI_INTERPOLATE_CONSTANT);
 282    start[1] = ureg_imm1f(shader, 0.0f);
 283
 284    for (i = 0; i < NR_RENDER_TARGETS; ++i)
 285        fragment[i] = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, i);
 286
 287    /* pixel center is at 0.5 not 0.0 !!! */
 288    ureg_ADD(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Y),
 289       tex, ureg_imm1f(shader, -2.0f / source->height0));
 290
 291    for (i = 0; i < 4; ++i) {
 292       fetch_four(shader, l[i], ureg_src(t_tc), sampler[0], start[0], block, false, false, source->width0);
 293       ureg_MUL(shader, l[i][0], ureg_src(l[i][0]), ureg_imm1f(shader, STAGE1_SCALE));
 294       ureg_MUL(shader, l[i][1], ureg_src(l[i][1]), ureg_imm1f(shader, STAGE1_SCALE));
 295       if(i != 3)
 296          ureg_ADD(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Y),
 297             ureg_src(t_tc), ureg_imm1f(shader, 1.0f / source->height0));
 298    }
 299
 300    for (i = 0; i < NR_RENDER_TARGETS; ++i) {
 301
 302 #if NR_RENDER_TARGETS == 8
 303       ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_X), ureg_imm1f(shader, 1.0f / BLOCK_WIDTH * i));
 304       fetch_four(shader, r, ureg_src(t_tc), sampler[1], start[1], block, true, true, matrix->width0);
 305 #elif NR_RENDER_TARGETS == 1
 306       fetch_four(shader, r, block, sampler[1], start[1], block, true, true, matrix->width0);
 307 #else
 308 #error invalid number of render targets
 309 #endif
 310
 311       for (j = 0; j < 4; ++j) {
 312          matrix_mul(shader, ureg_writemask(fragment[i], TGSI_WRITEMASK_X << j), l[j], r);
 313       }
 314       ureg_release_temporary(shader, r[0]);
 315       ureg_release_temporary(shader, r[1]);
 316    }
 317
 318    ureg_release_temporary(shader, t_tc);
 319    ureg_release_temporary(shader, tmp);
 320
 321    for (i = 0; i < 4; ++i) {
 322       ureg_release_temporary(shader, l[i][0]);
 323       ureg_release_temporary(shader, l[i][1]);
 324    }
 325
 326    ureg_END(shader);
 327
 328    return ureg_create_shader_and_destroy(shader, idct->pipe);
 329 }
 330
 331 static void *
 332 create_empty_block_frag_shader(struct vl_idct *idct)
 333 {
 334    struct ureg_program *shader;
 335    struct ureg_dst fragment;
 336
 337    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 338    if (!shader)
 339       return NULL;
 340
 341    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 342
 343    ureg_MOV(shader, fragment, ureg_imm1f(shader, 0.0f));
 344
 345    ureg_END(shader);
 346
 347    return ureg_create_shader_and_destroy(shader, idct->pipe);
 348 }
 349
 350 static bool
 351 init_shaders(struct vl_idct *idct)
 352 {
 353    idct->matrix_vs = create_vert_shader(idct, true);
 354    idct->matrix_fs = create_matrix_frag_shader(idct);
 355
 356    idct->transpose_vs = create_vert_shader(idct, true);
 357    idct->transpose_fs = create_transpose_frag_shader(idct);
 358
 359    idct->eb_vs = create_vert_shader(idct, false);
 360    idct->eb_fs = create_empty_block_frag_shader(idct);
 361
 362    return
 363       idct->transpose_vs != NULL && idct->transpose_fs != NULL &&
 364       idct->matrix_vs != NULL && idct->matrix_fs != NULL &&
 365       idct->eb_vs != NULL && idct->eb_fs != NULL;
 366 }
 367
 368 static void
 369 cleanup_shaders(struct vl_idct *idct)
 370 {
 371    idct->pipe->delete_vs_state(idct->pipe, idct->transpose_vs);
 372    idct->pipe->delete_fs_state(idct->pipe, idct->transpose_fs);
 373
 374    idct->pipe->delete_vs_state(idct->pipe, idct->matrix_vs);
 375    idct->pipe->delete_fs_state(idct->pipe, idct->matrix_fs);
 376
 377    idct->pipe->delete_vs_state(idct->pipe, idct->eb_vs);
 378    idct->pipe->delete_fs_state(idct->pipe, idct->eb_fs);
 379 }
 380
 381 static bool
 382 init_buffers(struct vl_idct *idct)
 383 {
 384    struct pipe_resource template;
 385    struct pipe_sampler_view sampler_view;
 386    struct pipe_vertex_element vertex_elems[2];
 387    unsigned i;
 388
 389    memset(&template, 0, sizeof(struct pipe_resource));
 390    template.last_level = 0;
 391    template.depth0 = 1;
 392    template.bind = PIPE_BIND_SAMPLER_VIEW;
 393    template.flags = 0;
 394
 395    template.target = PIPE_TEXTURE_2D;
 396    template.format = PIPE_FORMAT_R16G16B16A16_SNORM;
 397    template.width0 = idct->destination->width0 / 4;
 398    template.height0 = idct->destination->height0;
 399    template.depth0 = 1;
 400    template.usage = PIPE_USAGE_STREAM;
 401    idct->textures.individual.source = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 402
 403    template.target = PIPE_TEXTURE_3D;
 404    template.format = PIPE_FORMAT_R16G16B16A16_SNORM;
 405    template.width0 = idct->destination->width0 / NR_RENDER_TARGETS;
 406    template.height0 = idct->destination->height0 / 4;
 407    template.depth0 = NR_RENDER_TARGETS;
 408    template.usage = PIPE_USAGE_STATIC;
 409    idct->textures.individual.intermediate = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 410
 411    for (i = 0; i < 4; ++i) {
 412       if(idct->textures.all[i] == NULL)
 413          return false; /* a texture failed to allocate */
 414
 415       u_sampler_view_default_template(&sampler_view, idct->textures.all[i], idct->textures.all[i]->format);
 416       idct->sampler_views.all[i] = idct->pipe->create_sampler_view(idct->pipe, idct->textures.all[i], &sampler_view);
 417    }
 418
 419    idct->vertex_bufs.individual.quad = vl_vb_upload_quads(idct->pipe, idct->max_blocks);
 420
 421    if(idct->vertex_bufs.individual.quad.buffer == NULL)
 422       return false;
 423
 424    idct->vertex_bufs.individual.pos.stride = sizeof(struct vertex2f);
 425    idct->vertex_bufs.individual.pos.max_index = 4 * idct->max_blocks - 1;
 426    idct->vertex_bufs.individual.pos.buffer_offset = 0;
 427    idct->vertex_bufs.individual.pos.buffer = pipe_buffer_create
 428    (
 429       idct->pipe->screen,
 430       PIPE_BIND_VERTEX_BUFFER,
 431       sizeof(struct vertex2f) * 4 * idct->max_blocks
 432    );
 433
 434    if(idct->vertex_bufs.individual.pos.buffer == NULL)
 435       return false;
 436
 437    /* Rect element */
 438    vertex_elems[0].src_offset = 0;
 439    vertex_elems[0].instance_divisor = 0;
 440    vertex_elems[0].vertex_buffer_index = 0;
 441    vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
 442
 443    /* Pos element */
 444    vertex_elems[1].src_offset = 0;
 445    vertex_elems[1].instance_divisor = 0;
 446    vertex_elems[1].vertex_buffer_index = 1;
 447    vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
 448
 449    idct->vertex_elems_state = idct->pipe->create_vertex_elements_state(idct->pipe, 2, vertex_elems);
 450
 451    return true;
 452 }
 453
 454 static void
 455 cleanup_buffers(struct vl_idct *idct)
 456 {
 457    unsigned i;
 458
 459    assert(idct);
 460
 461    for (i = 0; i < 4; ++i) {
 462       pipe_sampler_view_reference(&idct->sampler_views.all[i], NULL);
 463       pipe_resource_reference(&idct->textures.all[i], NULL);
 464    }
 465
 466    idct->pipe->delete_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 467    pipe_resource_reference(&idct->vertex_bufs.individual.quad.buffer, NULL);
 468    pipe_resource_reference(&idct->vertex_bufs.individual.pos.buffer, NULL);
 469 }
 470
 471 static void
 472 init_state(struct vl_idct *idct)
 473 {
 474    struct pipe_sampler_state sampler;
 475    unsigned i;
 476
 477    idct->viewport[0].scale[0] = idct->textures.individual.intermediate->width0;
 478    idct->viewport[0].scale[1] = idct->textures.individual.intermediate->height0;
 479
 480    idct->viewport[1].scale[0] = idct->destination->width0;
 481    idct->viewport[1].scale[1] = idct->destination->height0;
 482
 483    idct->fb_state[0].width = idct->textures.individual.intermediate->width0;
 484    idct->fb_state[0].height = idct->textures.individual.intermediate->height0;
 485
 486    idct->fb_state[0].nr_cbufs = NR_RENDER_TARGETS;
 487    for(i = 0; i < NR_RENDER_TARGETS; ++i) {
 488       idct->fb_state[0].cbufs[i] = idct->pipe->screen->get_tex_surface(
 489          idct->pipe->screen, idct->textures.individual.intermediate, 0, 0, i,
 490          PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET);
 491    }
 492
 493    idct->fb_state[1].width = idct->destination->width0;
 494    idct->fb_state[1].height = idct->destination->height0;
 495
 496    idct->fb_state[1].nr_cbufs = 1;
 497    idct->fb_state[1].cbufs[0] = idct->pipe->screen->get_tex_surface(
 498       idct->pipe->screen, idct->destination, 0, 0, 0,
 499       PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET);
 500
 501    for(i = 0; i < 2; ++i) {
 502       idct->viewport[i].scale[2] = 1;
 503       idct->viewport[i].scale[3] = 1;
 504       idct->viewport[i].translate[0] = 0;
 505       idct->viewport[i].translate[1] = 0;
 506       idct->viewport[i].translate[2] = 0;
 507       idct->viewport[i].translate[3] = 0;
 508
 509       idct->fb_state[i].zsbuf = NULL;
 510    }
 511
 512    for (i = 0; i < 4; ++i) {
 513       memset(&sampler, 0, sizeof(sampler));
 514       sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 515       sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 516       sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 517       sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
 518       sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
 519       sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
 520       sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
 521       sampler.compare_func = PIPE_FUNC_ALWAYS;
 522       sampler.normalized_coords = 1;
 523       /*sampler.shadow_ambient = ; */
 524       /*sampler.lod_bias = ; */
 525       sampler.min_lod = 0;
 526       /*sampler.max_lod = ; */
 527       /*sampler.border_color[0] = ; */
 528       /*sampler.max_anisotropy = ; */
 529       idct->samplers.all[i] = idct->pipe->create_sampler_state(idct->pipe, &sampler);
 530    }
 531 }
 532
 533 static void
 534 cleanup_state(struct vl_idct *idct)
 535 {
 536    unsigned i;
 537
 538    for(i = 0; i < NR_RENDER_TARGETS; ++i) {
 539       idct->pipe->screen->tex_surface_destroy(idct->fb_state[0].cbufs[i]);
 540    }
 541
 542    idct->pipe->screen->tex_surface_destroy(idct->fb_state[1].cbufs[0]);
 543
 544    for (i = 0; i < 4; ++i)
 545       idct->pipe->delete_sampler_state(idct->pipe, idct->samplers.all[i]);
 546 }
 547
 548 struct pipe_resource *
 549 vl_idct_upload_matrix(struct pipe_context *pipe)
 550 {
 551    struct pipe_resource template, *matrix;
 552    struct pipe_transfer *buf_transfer;
 553    unsigned i, j, pitch;
 554    float *f;
 555
 556    struct pipe_box rect =
 557    {
 558       0, 0, 0,
 559       BLOCK_WIDTH,
 560       BLOCK_HEIGHT,
 561       1
 562    };
 563
 564    memset(&template, 0, sizeof(struct pipe_resource));
 565    template.target = PIPE_TEXTURE_2D;
 566    template.format = PIPE_FORMAT_R32G32B32A32_FLOAT;
 567    template.last_level = 0;
 568    template.width0 = 2;
 569    template.height0 = 8;
 570    template.depth0 = 1;
 571    template.usage = PIPE_USAGE_IMMUTABLE;
 572    template.bind = PIPE_BIND_SAMPLER_VIEW;
 573    template.flags = 0;
 574
 575    matrix = pipe->screen->resource_create(pipe->screen, &template);
 576
 577    /* matrix */
 578    buf_transfer = pipe->get_transfer
 579    (
 580       pipe, matrix,
 581       u_subresource(0, 0),
 582       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 583       &rect
 584    );
 585    pitch = buf_transfer->stride / sizeof(float);
 586
 587    f = pipe->transfer_map(pipe, buf_transfer);
 588    for(i = 0; i < BLOCK_HEIGHT; ++i)
 589       for(j = 0; j < BLOCK_WIDTH; ++j)
 590          f[i * pitch + j] = const_matrix[j][i]; // transpose
 591
 592    pipe->transfer_unmap(pipe, buf_transfer);
 593    pipe->transfer_destroy(pipe, buf_transfer);
 594
 595    return matrix;
 596 }
 597
 598 static void
 599 xfer_buffers_map(struct vl_idct *idct)
 600 {
 601    struct pipe_box rect =
 602    {
 603       0, 0, 0,
 604       idct->textures.individual.source->width0,
 605       idct->textures.individual.source->height0,
 606       1
 607    };
 608
 609    idct->tex_transfer = idct->pipe->get_transfer
 610    (
 611       idct->pipe, idct->textures.individual.source,
 612       u_subresource(0, 0),
 613       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 614       &rect
 615    );
 616
 617    idct->texels = idct->pipe->transfer_map(idct->pipe, idct->tex_transfer);
 618 }
 619
 620 static void
 621 xfer_buffers_unmap(struct vl_idct *idct)
 622 {
 623    idct->pipe->transfer_unmap(idct->pipe, idct->tex_transfer);
 624    idct->pipe->transfer_destroy(idct->pipe, idct->tex_transfer);
 625 }
 626
 627 bool
 628 vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe, struct pipe_resource *dst, struct pipe_resource *matrix)
 629 {
 630    assert(idct && pipe && dst);
 631
 632    idct->pipe = pipe;
 633    pipe_resource_reference(&idct->textures.individual.matrix, matrix);
 634    pipe_resource_reference(&idct->textures.individual.transpose, matrix);
 635    pipe_resource_reference(&idct->destination, dst);
 636
 637    idct->max_blocks =
 638       align(idct->destination->width0, BLOCK_WIDTH) / BLOCK_WIDTH *
 639       align(idct->destination->height0, BLOCK_HEIGHT) / BLOCK_HEIGHT *
 640       idct->destination->depth0;
 641
 642    if(!init_buffers(idct))
 643       return false;
 644
 645    if(!init_shaders(idct)) {
 646       cleanup_buffers(idct);
 647       return false;
 648    }
 649
 650    if(!vl_vb_init(&idct->blocks, idct->max_blocks)) {
 651       cleanup_shaders(idct);
 652       cleanup_buffers(idct);
 653       return false;
 654    }
 655
 656    if(!vl_vb_init(&idct->empty_blocks, idct->max_blocks)) {
 657       vl_vb_cleanup(&idct->blocks);
 658       cleanup_shaders(idct);
 659       cleanup_buffers(idct);
 660       return false;
 661    }
 662
 663    init_state(idct);
 664
 665    xfer_buffers_map(idct);
 666
 667    return true;
 668 }
 669
 670 void
 671 vl_idct_cleanup(struct vl_idct *idct)
 672 {
 673    vl_vb_cleanup(&idct->blocks);
 674    vl_vb_cleanup(&idct->empty_blocks);
 675    cleanup_shaders(idct);
 676    cleanup_buffers(idct);
 677
 678    cleanup_state(idct);
 679
 680    pipe_resource_reference(&idct->destination, NULL);
 681 }
 682
 683 void
 684 vl_idct_add_block(struct vl_idct *idct, unsigned x, unsigned y, short *block)
 685 {
 686    unsigned tex_pitch;
 687    short *texels;
 688
 689    unsigned i;
 690
 691    assert(idct);
 692
 693    if(block) {
 694       tex_pitch = idct->tex_transfer->stride / sizeof(short);
 695       texels = idct->texels + y * tex_pitch * BLOCK_HEIGHT + x * BLOCK_WIDTH;
 696
 697       for (i = 0; i < BLOCK_HEIGHT; ++i)
 698          memcpy(texels + i * tex_pitch, block + i * BLOCK_WIDTH, BLOCK_WIDTH * sizeof(short));
 699
 700       vl_vb_add_block(&idct->blocks, false, x, y);
 701    } else {
 702
 703       vl_vb_add_block(&idct->empty_blocks, true, x, y);
 704    }
 705 }
 706
 707 void
 708 vl_idct_flush(struct vl_idct *idct)
 709 {
 710    struct pipe_transfer *vec_transfer;
 711    struct quadf *vectors;
 712    unsigned num_blocks, num_empty_blocks;
 713
 714    assert(idct);
 715
 716    vectors = pipe_buffer_map
 717    (
 718       idct->pipe,
 719       idct->vertex_bufs.individual.pos.buffer,
 720       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 721       &vec_transfer
 722    );
 723
 724    num_blocks = vl_vb_upload(&idct->blocks, vectors);
 725    num_empty_blocks = vl_vb_upload(&idct->empty_blocks, vectors + num_blocks);
 726
 727    pipe_buffer_unmap(idct->pipe, idct->vertex_bufs.individual.pos.buffer, vec_transfer);
 728
 729    xfer_buffers_unmap(idct);
 730
 731    if(num_blocks > 0) {
 732
 733       /* first stage */
 734       idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state[0]);
 735       idct->pipe->set_viewport_state(idct->pipe, &idct->viewport[0]);
 736
 737       idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all);
 738       idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 739       idct->pipe->set_fragment_sampler_views(idct->pipe, 2, idct->sampler_views.stage[0]);
 740       idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[0]);
 741       idct->pipe->bind_vs_state(idct->pipe, idct->matrix_vs);
 742       idct->pipe->bind_fs_state(idct->pipe, idct->matrix_fs);
 743
 744       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, num_blocks * 4);
 745
 746       /* second stage */
 747       idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state[1]);
 748       idct->pipe->set_viewport_state(idct->pipe, &idct->viewport[1]);
 749
 750       idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all);
 751       idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 752       idct->pipe->set_fragment_sampler_views(idct->pipe, 2, idct->sampler_views.stage[1]);
 753       idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[1]);
 754       idct->pipe->bind_vs_state(idct->pipe, idct->transpose_vs);
 755       idct->pipe->bind_fs_state(idct->pipe, idct->transpose_fs);
 756
 757       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, num_blocks * 4);
 758    }
 759
 760    if(num_empty_blocks > 0) {
 761
 762       /* empty block handling */
 763       idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state[1]);
 764       idct->pipe->set_viewport_state(idct->pipe, &idct->viewport[1]);
 765
 766       idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all);
 767       idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 768       idct->pipe->bind_vs_state(idct->pipe, idct->eb_vs);
 769       idct->pipe->bind_fs_state(idct->pipe, idct->eb_fs);
 770
 771       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, num_blocks * 4, num_empty_blocks * 4);
 772    }
 773
 774    xfer_buffers_map(idct);
 775 }