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, bool calc_src_cords)
  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    t_vpos = ureg_DECL_temporary(shader);
  98
  99    vrect = ureg_DECL_vs_input(shader, VS_I_RECT);
 100    vpos = ureg_DECL_vs_input(shader, VS_I_VPOS);
 101
 102    o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
 103
 104    /*
 105     * scale = (BLOCK_WIDTH, BLOCK_HEIGHT) / (dst.width, dst.height)
 106     *
 107     * t_vpos = vpos + vrect
 108     * o_vpos.xy = t_vpos * scale
 109     * o_vpos.zw = vpos
 110     *
 111     * o_block = vrect
 112     * o_tex = t_pos
 113     * o_start = vpos * scale
 114     *
 115     */
 116    scale = ureg_imm2f(shader,
 117       (float)BLOCK_WIDTH / idct->destination->width0,
 118       (float)BLOCK_HEIGHT / idct->destination->height0);
 119
 120    ureg_ADD(shader, ureg_writemask(t_vpos, TGSI_WRITEMASK_XY), vpos, vrect);
 121    ureg_MUL(shader, ureg_writemask(t_vpos, TGSI_WRITEMASK_XY), ureg_src(t_vpos), scale);
 122    ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_XY), ureg_src(t_vpos));
 123    ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_ZW), vpos);
 124
 125    if(calc_src_cords) {
 126       o_block = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK);
 127       o_tex = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX);
 128       o_start = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_START);
 129
 130       ureg_MOV(shader, ureg_writemask(o_block, TGSI_WRITEMASK_XY), vrect);
 131       ureg_MOV(shader, ureg_writemask(o_tex, TGSI_WRITEMASK_XY), ureg_src(t_vpos));
 132       ureg_MUL(shader, ureg_writemask(o_start, TGSI_WRITEMASK_XY), vpos, scale);
 133    }
 134
 135    ureg_release_temporary(shader, t_vpos);
 136
 137    ureg_END(shader);
 138
 139    return ureg_create_shader_and_destroy(shader, idct->pipe);
 140 }
 141
 142 static void
 143 fetch_one(struct ureg_program *shader, struct ureg_dst m[2],
 144           struct ureg_src tc, struct ureg_src sampler,
 145           struct ureg_src start, struct ureg_src block, float height)
 146 {
 147    struct ureg_dst t_tc, tmp;
 148    unsigned i, j;
 149
 150    t_tc = ureg_DECL_temporary(shader);
 151    tmp = ureg_DECL_temporary(shader);
 152
 153    m[0] = ureg_DECL_temporary(shader);
 154    m[1] = ureg_DECL_temporary(shader);
 155
 156    /*
 157     * t_tc.x = right_side ? start.x : tc.x
 158     * t_tc.y = right_side ? tc.y : start.y
 159     * m[0..1].xyzw = tex(t_tc++, sampler)
 160     */
 161    ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_X), ureg_scalar(tc, TGSI_SWIZZLE_X));
 162    ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Y), ureg_scalar(start, TGSI_SWIZZLE_Y));
 163    ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Z), ureg_scalar(block, TGSI_SWIZZLE_X));
 164
 165    for(i = 0; i < 2; ++i) {
 166       for(j = 0; j < 4; ++j) {
 167          /* Nouveau and r600g can't writemask tex dst regs (yet?), do in two steps */
 168          ureg_TEX(shader, tmp, TGSI_TEXTURE_3D, ureg_src(t_tc), sampler);
 169          ureg_MOV(shader, ureg_writemask(m[i], TGSI_WRITEMASK_X << j), ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_X));
 170
 171          if(i != 1 || j != 3) /* skip the last add */
 172             ureg_ADD(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Y),
 173                ureg_src(t_tc), ureg_imm1f(shader, 1.0f / height));
 174       }
 175    }
 176
 177    ureg_release_temporary(shader, t_tc);
 178    ureg_release_temporary(shader, tmp);
 179 }
 180
 181 static void
 182 fetch_four(struct ureg_program *shader, struct ureg_dst m[2],
 183            struct ureg_src tc, struct ureg_src sampler,
 184            struct ureg_src start, bool right_side, float width)
 185 {
 186    struct ureg_dst t_tc;
 187
 188    t_tc = ureg_DECL_temporary(shader);
 189    m[0] = ureg_DECL_temporary(shader);
 190    m[1] = ureg_DECL_temporary(shader);
 191
 192    /*
 193     * t_tc.x = right_side ? start.x : tc.x
 194     * t_tc.y = right_side ? tc.y : start.y
 195     * m[0..1] = tex(t_tc++, sampler)
 196     */
 197    if(right_side) {
 198       ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_X), ureg_scalar(start, TGSI_SWIZZLE_Y));
 199       ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Y), ureg_scalar(tc, TGSI_SWIZZLE_X));
 200    } else {
 201       ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_X), ureg_scalar(start, TGSI_SWIZZLE_X));
 202       ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Y), ureg_scalar(tc, TGSI_SWIZZLE_Y));
 203    }
 204
 205    ureg_TEX(shader, m[0], TGSI_TEXTURE_2D, ureg_src(t_tc), sampler);
 206    ureg_ADD(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_X), ureg_src(t_tc), ureg_imm1f(shader, 1.0f / width));
 207    ureg_TEX(shader, m[1], TGSI_TEXTURE_2D, ureg_src(t_tc), sampler);
 208
 209    ureg_release_temporary(shader, t_tc);
 210 }
 211
 212 static void
 213 matrix_mul(struct ureg_program *shader, struct ureg_dst dst, struct ureg_dst l[2], struct ureg_dst r[2])
 214 {
 215    struct ureg_dst tmp[2];
 216    unsigned i;
 217
 218    for(i = 0; i < 2; ++i) {
 219       tmp[i] = ureg_DECL_temporary(shader);
 220    }
 221
 222    /*
 223     * tmp[0..1] = dot4(m[0][0..1], m[1][0..1])
 224     * dst = tmp[0] + tmp[1]
 225     */
 226    ureg_DP4(shader, ureg_writemask(tmp[0], TGSI_WRITEMASK_X), ureg_src(l[0]), ureg_src(r[0]));
 227    ureg_DP4(shader, ureg_writemask(tmp[1], TGSI_WRITEMASK_X), ureg_src(l[1]), ureg_src(r[1]));
 228    ureg_ADD(shader, dst, ureg_src(tmp[0]), ureg_src(tmp[1]));
 229
 230    for(i = 0; i < 2; ++i) {
 231       ureg_release_temporary(shader, tmp[i]);
 232    }
 233 }
 234
 235 static void *
 236 create_transpose_frag_shader(struct vl_idct *idct)
 237 {
 238    struct pipe_resource *transpose = idct->textures.individual.transpose;
 239    struct pipe_resource *intermediate = idct->textures.individual.intermediate;
 240
 241    struct ureg_program *shader;
 242
 243    struct ureg_src block, tex, sampler[2];
 244    struct ureg_src start[2];
 245
 246    struct ureg_dst m[2][2];
 247    struct ureg_dst tmp, fragment;
 248
 249    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 250    if (!shader)
 251       return NULL;
 252
 253    block = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK, TGSI_INTERPOLATE_LINEAR);
 254    tex = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX, TGSI_INTERPOLATE_CONSTANT);
 255
 256    sampler[0] = ureg_DECL_sampler(shader, 0);
 257    sampler[1] = ureg_DECL_sampler(shader, 1);
 258
 259    start[0] = ureg_imm1f(shader, 0.0f);
 260    start[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_START, TGSI_INTERPOLATE_CONSTANT);
 261
 262    fetch_four(shader, m[0], block, sampler[0], start[0], false, transpose->width0);
 263    fetch_one(shader, m[1], tex, sampler[1], start[1], block, intermediate->height0);
 264
 265    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 266
 267    tmp = ureg_DECL_temporary(shader);
 268    matrix_mul(shader, ureg_writemask(tmp, TGSI_WRITEMASK_X), m[0], m[1]);
 269    ureg_MUL(shader, fragment, ureg_src(tmp), ureg_imm1f(shader, STAGE2_SCALE));
 270
 271    ureg_release_temporary(shader, tmp);
 272    ureg_release_temporary(shader, m[0][0]);
 273    ureg_release_temporary(shader, m[0][1]);
 274    ureg_release_temporary(shader, m[1][0]);
 275    ureg_release_temporary(shader, m[1][1]);
 276
 277    ureg_END(shader);
 278
 279    return ureg_create_shader_and_destroy(shader, idct->pipe);
 280 }
 281
 282 static void *
 283 create_matrix_frag_shader(struct vl_idct *idct)
 284 {
 285    struct pipe_resource *matrix = idct->textures.individual.matrix;
 286    struct pipe_resource *source = idct->textures.individual.source;
 287
 288    struct ureg_program *shader;
 289
 290    struct ureg_src tc[2], sampler[2];
 291    struct ureg_src start[2];
 292
 293    struct ureg_dst l[2], r[2];
 294    struct ureg_dst t_tc, tmp, fragment[BLOCK_WIDTH];
 295
 296    unsigned i;
 297
 298    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 299    if (!shader)
 300       return NULL;
 301
 302    t_tc = ureg_DECL_temporary(shader);
 303    tmp = ureg_DECL_temporary(shader);
 304
 305    tc[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX, TGSI_INTERPOLATE_LINEAR);
 306    tc[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK, TGSI_INTERPOLATE_LINEAR);
 307
 308    sampler[0] = ureg_DECL_sampler(shader, 1);
 309    sampler[1] = ureg_DECL_sampler(shader, 0);
 310
 311    start[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_START, TGSI_INTERPOLATE_CONSTANT);
 312    start[1] = ureg_imm1f(shader, 0.0f);
 313
 314    for (i = 0; i < BLOCK_WIDTH; ++i)
 315        fragment[i] = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, i);
 316
 317    fetch_four(shader, l, tc[0], sampler[0], start[0], false, source->width0);
 318    ureg_MUL(shader, l[0], ureg_src(l[0]), ureg_scalar(ureg_imm1f(shader, STAGE1_SCALE), TGSI_SWIZZLE_X));
 319    ureg_MUL(shader, l[1], ureg_src(l[1]), ureg_scalar(ureg_imm1f(shader, STAGE1_SCALE), TGSI_SWIZZLE_X));
 320
 321    for (i = 0; i < BLOCK_WIDTH; ++i) {
 322       ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_X), ureg_imm1f(shader, 1.0f / BLOCK_WIDTH * i));
 323       fetch_four(shader, r, ureg_src(t_tc), sampler[1], start[1], true, matrix->width0);
 324       matrix_mul(shader, fragment[i], l, r);
 325       ureg_release_temporary(shader, r[0]);
 326       ureg_release_temporary(shader, r[1]);
 327    }
 328
 329    ureg_release_temporary(shader, t_tc);
 330    ureg_release_temporary(shader, tmp);
 331    ureg_release_temporary(shader, l[0]);
 332    ureg_release_temporary(shader, l[1]);
 333
 334    ureg_END(shader);
 335
 336    return ureg_create_shader_and_destroy(shader, idct->pipe);
 337 }
 338
 339 static void *
 340 create_empty_block_frag_shader(struct vl_idct *idct)
 341 {
 342    struct ureg_program *shader;
 343    struct ureg_dst fragment;
 344
 345    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 346    if (!shader)
 347       return NULL;
 348
 349    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 350
 351    ureg_MOV(shader, fragment, ureg_imm1f(shader, 0.0f));
 352
 353    ureg_END(shader);
 354
 355    return ureg_create_shader_and_destroy(shader, idct->pipe);
 356 }
 357
 358 static bool
 359 init_shaders(struct vl_idct *idct)
 360 {
 361    idct->matrix_vs = create_vert_shader(idct, true);
 362    idct->matrix_fs = create_matrix_frag_shader(idct);
 363
 364    idct->transpose_vs = create_vert_shader(idct, true);
 365    idct->transpose_fs = create_transpose_frag_shader(idct);
 366
 367    idct->eb_vs = create_vert_shader(idct, false);
 368    idct->eb_fs = create_empty_block_frag_shader(idct);
 369
 370    return
 371       idct->transpose_vs != NULL && idct->transpose_fs != NULL &&
 372       idct->matrix_vs != NULL && idct->matrix_fs != NULL &&
 373       idct->eb_vs != NULL && idct->eb_fs != NULL;
 374 }
 375
 376 static void
 377 cleanup_shaders(struct vl_idct *idct)
 378 {
 379    idct->pipe->delete_vs_state(idct->pipe, idct->transpose_vs);
 380    idct->pipe->delete_fs_state(idct->pipe, idct->transpose_fs);
 381
 382    idct->pipe->delete_vs_state(idct->pipe, idct->matrix_vs);
 383    idct->pipe->delete_fs_state(idct->pipe, idct->matrix_fs);
 384
 385    idct->pipe->delete_vs_state(idct->pipe, idct->eb_vs);
 386    idct->pipe->delete_fs_state(idct->pipe, idct->eb_fs);
 387 }
 388
 389 static bool
 390 init_buffers(struct vl_idct *idct)
 391 {
 392    struct pipe_resource template;
 393    struct pipe_sampler_view sampler_view;
 394    struct pipe_vertex_element vertex_elems[2];
 395    unsigned i;
 396
 397    idct->max_blocks =
 398       align(idct->destination->width0, BLOCK_WIDTH) / BLOCK_WIDTH *
 399       align(idct->destination->height0, BLOCK_HEIGHT) / BLOCK_HEIGHT *
 400       idct->destination->depth0;
 401
 402    memset(&template, 0, sizeof(struct pipe_resource));
 403    template.last_level = 0;
 404    template.depth0 = 1;
 405    template.bind = PIPE_BIND_SAMPLER_VIEW;
 406    template.flags = 0;
 407
 408    template.target = PIPE_TEXTURE_2D;
 409    template.format = PIPE_FORMAT_R16G16B16A16_SNORM;
 410    template.width0 = idct->destination->width0 / 4;
 411    template.height0 = idct->destination->height0;
 412    template.depth0 = 1;
 413    template.usage = PIPE_USAGE_STREAM;
 414    idct->textures.individual.source = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 415
 416    template.target = PIPE_TEXTURE_3D;
 417    template.format = PIPE_FORMAT_R16_SNORM;
 418    template.width0 = idct->destination->width0 / 8;
 419    template.depth0 = 8;
 420    template.usage = PIPE_USAGE_STATIC;
 421    idct->textures.individual.intermediate = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 422
 423    for (i = 0; i < 4; ++i) {
 424       if(idct->textures.all[i] == NULL)
 425          return false; /* a texture failed to allocate */
 426
 427       u_sampler_view_default_template(&sampler_view, idct->textures.all[i], idct->textures.all[i]->format);
 428       idct->sampler_views.all[i] = idct->pipe->create_sampler_view(idct->pipe, idct->textures.all[i], &sampler_view);
 429    }
 430
 431    idct->vertex_bufs.individual.quad.stride = sizeof(struct vertex2f);
 432    idct->vertex_bufs.individual.quad.max_index = 4 * idct->max_blocks - 1;
 433    idct->vertex_bufs.individual.quad.buffer_offset = 0;
 434    idct->vertex_bufs.individual.quad.buffer = pipe_buffer_create
 435    (
 436       idct->pipe->screen,
 437       PIPE_BIND_VERTEX_BUFFER,
 438       sizeof(struct vertex2f) * 4 * idct->max_blocks
 439    );
 440
 441    if(idct->vertex_bufs.individual.quad.buffer == NULL)
 442       return false;
 443
 444    idct->vertex_bufs.individual.pos.stride = sizeof(struct vertex2f);
 445    idct->vertex_bufs.individual.pos.max_index = 4 * idct->max_blocks - 1;
 446    idct->vertex_bufs.individual.pos.buffer_offset = 0;
 447    idct->vertex_bufs.individual.pos.buffer = pipe_buffer_create
 448    (
 449       idct->pipe->screen,
 450       PIPE_BIND_VERTEX_BUFFER,
 451       sizeof(struct vertex2f) * 4 * idct->max_blocks
 452    );
 453
 454    if(idct->vertex_bufs.individual.pos.buffer == NULL)
 455       return false;
 456
 457    /* Rect element */
 458    vertex_elems[0].src_offset = 0;
 459    vertex_elems[0].instance_divisor = 0;
 460    vertex_elems[0].vertex_buffer_index = 0;
 461    vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
 462
 463    /* Pos element */
 464    vertex_elems[1].src_offset = 0;
 465    vertex_elems[1].instance_divisor = 0;
 466    vertex_elems[1].vertex_buffer_index = 1;
 467    vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
 468
 469    idct->vertex_elems_state = idct->pipe->create_vertex_elements_state(idct->pipe, 2, vertex_elems);
 470
 471    return true;
 472 }
 473
 474 static void
 475 cleanup_buffers(struct vl_idct *idct)
 476 {
 477    unsigned i;
 478
 479    assert(idct);
 480
 481    for (i = 0; i < 4; ++i) {
 482       pipe_sampler_view_reference(&idct->sampler_views.all[i], NULL);
 483       pipe_resource_reference(&idct->textures.all[i], NULL);
 484    }
 485
 486    idct->pipe->delete_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 487    pipe_resource_reference(&idct->vertex_bufs.individual.quad.buffer, NULL);
 488    pipe_resource_reference(&idct->vertex_bufs.individual.pos.buffer, NULL);
 489 }
 490
 491 static void
 492 init_constants(struct vl_idct *idct)
 493 {
 494    struct pipe_transfer *buf_transfer;
 495    struct vertex2f *v;
 496
 497    unsigned i;
 498
 499    /* quad vectors */
 500    v = pipe_buffer_map
 501    (
 502       idct->pipe,
 503       idct->vertex_bufs.individual.quad.buffer,
 504       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 505       &buf_transfer
 506    );
 507    for ( i = 0; i < idct->max_blocks; ++i)
 508      memcpy(v + i * 4, &const_quad, sizeof(const_quad));
 509    pipe_buffer_unmap(idct->pipe, idct->vertex_bufs.individual.quad.buffer, 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[0].scale[0] = idct->textures.individual.intermediate->width0;
 522    idct->viewport[0].scale[1] = idct->textures.individual.intermediate->height0;
 523
 524    idct->viewport[1].scale[0] = idct->destination->width0;
 525    idct->viewport[1].scale[1] = idct->destination->height0;
 526
 527    idct->fb_state[0].width = idct->textures.individual.intermediate->width0;
 528    idct->fb_state[0].height = idct->textures.individual.intermediate->height0;
 529
 530    idct->fb_state[0].nr_cbufs = 8;
 531    for(i = 0; i < 8; ++i) {
 532       idct->fb_state[0].cbufs[i] = idct->pipe->screen->get_tex_surface(
 533          idct->pipe->screen, idct->textures.individual.intermediate, 0, 0, i,
 534          PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET);
 535    }
 536
 537    idct->fb_state[1].width = idct->destination->width0;
 538    idct->fb_state[1].height = idct->destination->height0;
 539
 540    idct->fb_state[1].nr_cbufs = 1;
 541    idct->fb_state[1].cbufs[0] = idct->pipe->screen->get_tex_surface(
 542       idct->pipe->screen, idct->destination, 0, 0, 0,
 543       PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET);
 544
 545    for(i = 0; i < 2; ++i) {
 546       idct->viewport[i].scale[2] = 1;
 547       idct->viewport[i].scale[3] = 1;
 548       idct->viewport[i].translate[0] = 0;
 549       idct->viewport[i].translate[1] = 0;
 550       idct->viewport[i].translate[2] = 0;
 551       idct->viewport[i].translate[3] = 0;
 552
 553       idct->fb_state[i].zsbuf = NULL;
 554    }
 555
 556    for (i = 0; i < 4; ++i) {
 557       memset(&sampler, 0, sizeof(sampler));
 558       sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 559       sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 560       sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 561       sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
 562       sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
 563       sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
 564       sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
 565       sampler.compare_func = PIPE_FUNC_ALWAYS;
 566       sampler.normalized_coords = 1;
 567       /*sampler.shadow_ambient = ; */
 568       /*sampler.lod_bias = ; */
 569       sampler.min_lod = 0;
 570       /*sampler.max_lod = ; */
 571       /*sampler.border_color[0] = ; */
 572       /*sampler.max_anisotropy = ; */
 573       idct->samplers.all[i] = idct->pipe->create_sampler_state(idct->pipe, &sampler);
 574    }
 575 }
 576
 577 static void
 578 cleanup_state(struct vl_idct *idct)
 579 {
 580    unsigned i;
 581
 582    for(i = 0; i < 8; ++i) {
 583       idct->pipe->screen->tex_surface_destroy(idct->fb_state[0].cbufs[i]);
 584    }
 585
 586    idct->pipe->screen->tex_surface_destroy(idct->fb_state[1].cbufs[0]);
 587
 588    for (i = 0; i < 4; ++i)
 589       idct->pipe->delete_sampler_state(idct->pipe, idct->samplers.all[i]);
 590 }
 591
 592 struct pipe_resource *
 593 vl_idct_upload_matrix(struct pipe_context *pipe)
 594 {
 595    struct pipe_resource template, *matrix;
 596    struct pipe_transfer *buf_transfer;
 597    unsigned i, j, pitch;
 598    float *f;
 599
 600    struct pipe_box rect =
 601    {
 602       0, 0, 0,
 603       BLOCK_WIDTH,
 604       BLOCK_HEIGHT,
 605       1
 606    };
 607
 608    memset(&template, 0, sizeof(struct pipe_resource));
 609    template.target = PIPE_TEXTURE_2D;
 610    template.format = PIPE_FORMAT_R32G32B32A32_FLOAT;
 611    template.last_level = 0;
 612    template.width0 = 2;
 613    template.height0 = 8;
 614    template.depth0 = 1;
 615    template.usage = PIPE_USAGE_IMMUTABLE;
 616    template.bind = PIPE_BIND_SAMPLER_VIEW;
 617    template.flags = 0;
 618
 619    matrix = pipe->screen->resource_create(pipe->screen, &template);
 620
 621    /* matrix */
 622    buf_transfer = pipe->get_transfer
 623    (
 624       pipe, matrix,
 625       u_subresource(0, 0),
 626       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 627       &rect
 628    );
 629    pitch = buf_transfer->stride / sizeof(float);
 630
 631    f = pipe->transfer_map(pipe, buf_transfer);
 632    for(i = 0; i < BLOCK_HEIGHT; ++i)
 633       for(j = 0; j < BLOCK_WIDTH; ++j)
 634          f[i * pitch + j] = const_matrix[j][i]; // transpose
 635
 636    pipe->transfer_unmap(pipe, buf_transfer);
 637    pipe->transfer_destroy(pipe, buf_transfer);
 638
 639    return matrix;
 640 }
 641
 642 static void
 643 xfer_buffers_map(struct vl_idct *idct)
 644 {
 645    struct pipe_box rect =
 646    {
 647       0, 0, 0,
 648       idct->textures.individual.source->width0,
 649       idct->textures.individual.source->height0,
 650       1
 651    };
 652
 653    idct->tex_transfer = idct->pipe->get_transfer
 654    (
 655       idct->pipe, idct->textures.individual.source,
 656       u_subresource(0, 0),
 657       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 658       &rect
 659    );
 660
 661    idct->texels = idct->pipe->transfer_map(idct->pipe, idct->tex_transfer);
 662
 663    idct->vectors = pipe_buffer_map
 664    (
 665       idct->pipe,
 666       idct->vertex_bufs.individual.pos.buffer,
 667       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 668       &idct->vec_transfer
 669    );
 670
 671    idct->next_empty_block.l_x = ~1;
 672    idct->next_empty_block.l_y = ~1;
 673    idct->next_empty_block.r_x = ~1;
 674    idct->next_empty_block.r_y = ~1;
 675 }
 676
 677 static void
 678 xfer_buffers_unmap(struct vl_idct *idct)
 679 {
 680    pipe_buffer_unmap(idct->pipe, idct->vertex_bufs.individual.pos.buffer, idct->vec_transfer);
 681
 682    idct->pipe->transfer_unmap(idct->pipe, idct->tex_transfer);
 683    idct->pipe->transfer_destroy(idct->pipe, idct->tex_transfer);
 684 }
 685
 686 bool
 687 vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe, struct pipe_resource *dst, struct pipe_resource *matrix)
 688 {
 689    assert(idct && pipe && dst);
 690
 691    idct->pipe = pipe;
 692    pipe_resource_reference(&idct->textures.individual.matrix, matrix);
 693    pipe_resource_reference(&idct->textures.individual.transpose, matrix);
 694    pipe_resource_reference(&idct->destination, dst);
 695
 696    if(!init_buffers(idct))
 697       return false;
 698
 699    if(!init_shaders(idct)) {
 700       cleanup_buffers(idct);
 701       return false;
 702    }
 703
 704    init_state(idct);
 705
 706    init_constants(idct);
 707    xfer_buffers_map(idct);
 708
 709    return true;
 710 }
 711
 712 void
 713 vl_idct_cleanup(struct vl_idct *idct)
 714 {
 715    cleanup_shaders(idct);
 716    cleanup_buffers(idct);
 717
 718    cleanup_state(idct);
 719
 720    pipe_resource_reference(&idct->destination, NULL);
 721 }
 722
 723 static void
 724 flush_empty_block(struct vl_idct *idct, unsigned new_x, unsigned new_y)
 725 {
 726    if (idct->next_empty_block.l_x == ~1 ||
 727        idct->next_empty_block.l_y == ~1) {
 728
 729       idct->next_empty_block.l_x = new_x;
 730       idct->next_empty_block.l_y = new_y;
 731
 732    } else if (idct->next_empty_block.r_x != (new_x - 1) ||
 733               idct->next_empty_block.r_y != new_y) {
 734
 735       struct vertex2f l, r, *v_dst;
 736
 737       v_dst = idct->vectors + (idct->max_blocks - idct->num_empty_blocks) * 4 - 4;
 738
 739       l.x = idct->next_empty_block.l_x;
 740       l.y = idct->next_empty_block.l_y;
 741       r.x = idct->next_empty_block.r_x;
 742       r.y = idct->next_empty_block.r_y;
 743       v_dst[0] = v_dst[3] = l;
 744       v_dst[1] = v_dst[2] = r;
 745
 746       idct->next_empty_block.l_x = new_x;
 747       idct->next_empty_block.l_y = new_y;
 748       idct->num_empty_blocks++;
 749    }
 750
 751    idct->next_empty_block.r_x = new_x;
 752    idct->next_empty_block.r_y = new_y;
 753 }
 754
 755 void
 756 vl_idct_add_block(struct vl_idct *idct, unsigned x, unsigned y, short *block)
 757 {
 758    struct vertex2f v, *v_dst;
 759
 760    unsigned tex_pitch;
 761    short *texels;
 762
 763    unsigned i;
 764
 765    assert(idct);
 766
 767    if(block) {
 768       tex_pitch = idct->tex_transfer->stride / sizeof(short);
 769       texels = idct->texels + y * tex_pitch * BLOCK_HEIGHT + x * BLOCK_WIDTH;
 770
 771       for (i = 0; i < BLOCK_HEIGHT; ++i)
 772          memcpy(texels + i * tex_pitch, block + i * BLOCK_WIDTH, BLOCK_WIDTH * sizeof(short));
 773
 774       /* non empty blocks fills the vector buffer from left to right */
 775       v_dst = idct->vectors + idct->num_blocks * 4;
 776
 777       idct->num_blocks++;
 778
 779       v.x = x;
 780       v.y = y;
 781
 782       for (i = 0; i < 4; ++i) {
 783          v_dst[i] = v;
 784       }
 785
 786    } else {
 787
 788       /* while empty blocks fills the vector buffer from right to left */
 789       flush_empty_block(idct, x, y);
 790    }
 791 }
 792
 793 void
 794 vl_idct_flush(struct vl_idct *idct)
 795 {
 796    flush_empty_block(idct, ~1, ~1);
 797    xfer_buffers_unmap(idct);
 798
 799    if(idct->num_blocks > 0) {
 800
 801       /* first stage */
 802       idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state[0]);
 803       idct->pipe->set_viewport_state(idct->pipe, &idct->viewport[0]);
 804
 805       idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all);
 806       idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 807       idct->pipe->set_fragment_sampler_views(idct->pipe, 2, idct->sampler_views.stage[0]);
 808       idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[0]);
 809       idct->pipe->bind_vs_state(idct->pipe, idct->matrix_vs);
 810       idct->pipe->bind_fs_state(idct->pipe, idct->matrix_fs);
 811
 812       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, idct->num_blocks * 4);
 813
 814       /* second stage */
 815       idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state[1]);
 816       idct->pipe->set_viewport_state(idct->pipe, &idct->viewport[1]);
 817
 818       idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all);
 819       idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 820       idct->pipe->set_fragment_sampler_views(idct->pipe, 2, idct->sampler_views.stage[1]);
 821       idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[1]);
 822       idct->pipe->bind_vs_state(idct->pipe, idct->transpose_vs);
 823       idct->pipe->bind_fs_state(idct->pipe, idct->transpose_fs);
 824
 825       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, idct->num_blocks * 4);
 826    }
 827
 828    if(idct->num_empty_blocks > 0) {
 829
 830       /* empty block handling */
 831       idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state[1]);
 832       idct->pipe->set_viewport_state(idct->pipe, &idct->viewport[1]);
 833
 834       idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all);
 835       idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 836       idct->pipe->bind_vs_state(idct->pipe, idct->eb_vs);
 837       idct->pipe->bind_fs_state(idct->pipe, idct->eb_fs);
 838
 839       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS,
 840          (idct->max_blocks - idct->num_empty_blocks) * 4,
 841          idct->num_empty_blocks * 4);
 842    }
 843
 844    idct->num_blocks = 0;
 845    idct->num_empty_blocks = 0;
 846    xfer_buffers_map(idct);
 847 }