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 #define NR_RENDER_TARGETS 1
  48
  49 struct vertex_shader_consts
  50 {
  51    struct vertex4f norm;
  52 };
  53
  54 enum VS_INPUT
  55 {
  56    VS_I_RECT,
  57    VS_I_VPOS,
  58
  59    NUM_VS_INPUTS
  60 };
  61
  62 enum VS_OUTPUT
  63 {
  64    VS_O_VPOS,
  65    VS_O_BLOCK,
  66    VS_O_TEX,
  67    VS_O_START
  68 };
  69
  70 static const float const_matrix[8][8] = {
  71    {  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.353553f,  0.3535530f },
  72    {  0.4903930f,  0.4157350f,  0.2777850f,  0.0975451f, -0.0975452f, -0.2777850f, -0.415735f, -0.4903930f },
  73    {  0.4619400f,  0.1913420f, -0.1913420f, -0.4619400f, -0.4619400f, -0.1913420f,  0.191342f,  0.4619400f },
  74    {  0.4157350f, -0.0975452f, -0.4903930f, -0.2777850f,  0.2777850f,  0.4903930f,  0.097545f, -0.4157350f },
  75    {  0.3535530f, -0.3535530f, -0.3535530f,  0.3535540f,  0.3535530f, -0.3535540f, -0.353553f,  0.3535530f },
  76    {  0.2777850f, -0.4903930f,  0.0975452f,  0.4157350f, -0.4157350f, -0.0975451f,  0.490393f, -0.2777850f },
  77    {  0.1913420f, -0.4619400f,  0.4619400f, -0.1913420f, -0.1913410f,  0.4619400f, -0.461940f,  0.1913420f },
  78    {  0.0975451f, -0.2777850f,  0.4157350f, -0.4903930f,  0.4903930f, -0.4157350f,  0.277786f, -0.0975458f }
  79 };
  80
  81 /* vertices for a quad covering a block */
  82 static const struct vertex2f const_quad[4] = {
  83    {0.0f, 0.0f}, {1.0f, 0.0f}, {1.0f, 1.0f}, {0.0f, 1.0f}
  84 };
  85
  86 static void *
  87 create_vert_shader(struct vl_idct *idct, bool calc_src_cords)
  88 {
  89    struct ureg_program *shader;
  90    struct ureg_src scale;
  91    struct ureg_src vrect, vpos;
  92    struct ureg_dst t_vpos;
  93    struct ureg_dst o_vpos, o_block, o_tex, o_start;
  94
  95    shader = ureg_create(TGSI_PROCESSOR_VERTEX);
  96    if (!shader)
  97       return NULL;
  98
  99    t_vpos = ureg_DECL_temporary(shader);
 100
 101    vrect = ureg_DECL_vs_input(shader, VS_I_RECT);
 102    vpos = ureg_DECL_vs_input(shader, VS_I_VPOS);
 103
 104    o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
 105
 106    /*
 107     * scale = (BLOCK_WIDTH, BLOCK_HEIGHT) / (dst.width, dst.height)
 108     *
 109     * t_vpos = vpos + vrect
 110     * o_vpos.xy = t_vpos * scale
 111     * o_vpos.zw = vpos
 112     *
 113     * o_block = vrect
 114     * o_tex = t_pos
 115     * o_start = vpos * scale
 116     *
 117     */
 118    scale = ureg_imm2f(shader,
 119       (float)BLOCK_WIDTH / idct->destination->width0,
 120       (float)BLOCK_HEIGHT / idct->destination->height0);
 121
 122    ureg_ADD(shader, ureg_writemask(t_vpos, TGSI_WRITEMASK_XY), vpos, vrect);
 123    ureg_MUL(shader, ureg_writemask(t_vpos, TGSI_WRITEMASK_XY), ureg_src(t_vpos), scale);
 124    ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_XY), ureg_src(t_vpos));
 125    ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_ZW), vpos);
 126
 127    if(calc_src_cords) {
 128       o_block = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK);
 129       o_tex = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX);
 130       o_start = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_START);
 131
 132       ureg_MOV(shader, ureg_writemask(o_block, TGSI_WRITEMASK_XY), vrect);
 133       ureg_MOV(shader, ureg_writemask(o_tex, TGSI_WRITEMASK_XY), ureg_src(t_vpos));
 134       ureg_MUL(shader, ureg_writemask(o_start, TGSI_WRITEMASK_XY), vpos, scale);
 135    }
 136
 137    ureg_release_temporary(shader, t_vpos);
 138
 139    ureg_END(shader);
 140
 141    return ureg_create_shader_and_destroy(shader, idct->pipe);
 142 }
 143
 144 static void
 145 fetch_one(struct ureg_program *shader, struct ureg_dst m[2],
 146           struct ureg_src tc, struct ureg_src sampler,
 147           struct ureg_src start, struct ureg_src block, float height)
 148 {
 149    struct ureg_dst t_tc, tmp;
 150    unsigned i, j;
 151
 152    t_tc = ureg_DECL_temporary(shader);
 153    tmp = ureg_DECL_temporary(shader);
 154
 155    m[0] = ureg_DECL_temporary(shader);
 156    m[1] = ureg_DECL_temporary(shader);
 157
 158    /*
 159     * t_tc.x = right_side ? start.x : tc.x
 160     * t_tc.y = right_side ? tc.y : start.y
 161     * m[0..1].xyzw = tex(t_tc++, sampler)
 162     */
 163    ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_X), ureg_scalar(tc, TGSI_SWIZZLE_X));
 164    ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Y), ureg_scalar(start, TGSI_SWIZZLE_Y));
 165
 166 #if NR_RENDER_TARGETS == 8
 167    ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Z), ureg_scalar(block, TGSI_SWIZZLE_X));
 168 #else
 169    ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Z), ureg_imm1f(shader, 0.0f));
 170 #endif
 171
 172    for(i = 0; i < 2; ++i) {
 173       for(j = 0; j < 4; ++j) {
 174          /* Nouveau and r600g can't writemask tex dst regs (yet?), do in two steps */
 175          ureg_TEX(shader, tmp, TGSI_TEXTURE_3D, ureg_src(t_tc), sampler);
 176          ureg_MOV(shader, ureg_writemask(m[i], TGSI_WRITEMASK_X << j), ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_X));
 177
 178          if(i != 1 || j != 3) /* skip the last add */
 179             ureg_ADD(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Y),
 180                ureg_src(t_tc), ureg_imm1f(shader, 1.0f / height));
 181       }
 182    }
 183
 184    ureg_release_temporary(shader, t_tc);
 185    ureg_release_temporary(shader, tmp);
 186 }
 187
 188 static void
 189 fetch_four(struct ureg_program *shader, struct ureg_dst m[2],
 190            struct ureg_src tc, struct ureg_src sampler,
 191            struct ureg_src start, bool right_side, float width)
 192 {
 193    struct ureg_dst t_tc;
 194
 195    t_tc = ureg_DECL_temporary(shader);
 196    m[0] = ureg_DECL_temporary(shader);
 197    m[1] = ureg_DECL_temporary(shader);
 198
 199    /*
 200     * t_tc.x = right_side ? start.x : tc.x
 201     * t_tc.y = right_side ? tc.y : start.y
 202     * m[0..1] = tex(t_tc++, sampler)
 203     */
 204    if(right_side) {
 205       ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_X), ureg_scalar(start, TGSI_SWIZZLE_Y));
 206       ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Y), ureg_scalar(tc, TGSI_SWIZZLE_X));
 207    } else {
 208       ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_X), ureg_scalar(start, TGSI_SWIZZLE_X));
 209       ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Y), ureg_scalar(tc, TGSI_SWIZZLE_Y));
 210    }
 211
 212    ureg_TEX(shader, m[0], TGSI_TEXTURE_2D, ureg_src(t_tc), sampler);
 213    ureg_ADD(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_X), ureg_src(t_tc), ureg_imm1f(shader, 1.0f / width));
 214    ureg_TEX(shader, m[1], TGSI_TEXTURE_2D, ureg_src(t_tc), sampler);
 215
 216    ureg_release_temporary(shader, t_tc);
 217 }
 218
 219 static void
 220 matrix_mul(struct ureg_program *shader, struct ureg_dst dst, struct ureg_dst l[2], struct ureg_dst r[2])
 221 {
 222    struct ureg_dst tmp[2];
 223    unsigned i;
 224
 225    for(i = 0; i < 2; ++i) {
 226       tmp[i] = ureg_DECL_temporary(shader);
 227    }
 228
 229    /*
 230     * tmp[0..1] = dot4(m[0][0..1], m[1][0..1])
 231     * dst = tmp[0] + tmp[1]
 232     */
 233    ureg_DP4(shader, ureg_writemask(tmp[0], TGSI_WRITEMASK_X), ureg_src(l[0]), ureg_src(r[0]));
 234    ureg_DP4(shader, ureg_writemask(tmp[1], TGSI_WRITEMASK_X), ureg_src(l[1]), ureg_src(r[1]));
 235    ureg_ADD(shader, dst, ureg_src(tmp[0]), ureg_src(tmp[1]));
 236
 237    for(i = 0; i < 2; ++i) {
 238       ureg_release_temporary(shader, tmp[i]);
 239    }
 240 }
 241
 242 static void *
 243 create_transpose_frag_shader(struct vl_idct *idct)
 244 {
 245    struct pipe_resource *transpose = idct->textures.individual.transpose;
 246    struct pipe_resource *intermediate = idct->textures.individual.intermediate;
 247
 248    struct ureg_program *shader;
 249
 250    struct ureg_src block, tex, sampler[2];
 251    struct ureg_src start[2];
 252
 253    struct ureg_dst m[2][2];
 254    struct ureg_dst tmp, fragment;
 255
 256    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 257    if (!shader)
 258       return NULL;
 259
 260    block = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK, TGSI_INTERPOLATE_LINEAR);
 261    tex = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX, TGSI_INTERPOLATE_CONSTANT);
 262
 263    sampler[0] = ureg_DECL_sampler(shader, 0);
 264    sampler[1] = ureg_DECL_sampler(shader, 1);
 265
 266    start[0] = ureg_imm1f(shader, 0.0f);
 267    start[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_START, TGSI_INTERPOLATE_CONSTANT);
 268
 269    fetch_four(shader, m[0], block, sampler[0], start[0], false, transpose->width0);
 270    fetch_one(shader, m[1], tex, sampler[1], start[1], block, intermediate->height0);
 271
 272    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 273
 274    tmp = ureg_DECL_temporary(shader);
 275    matrix_mul(shader, ureg_writemask(tmp, TGSI_WRITEMASK_X), m[0], m[1]);
 276    ureg_MUL(shader, fragment, ureg_src(tmp), ureg_imm1f(shader, STAGE2_SCALE));
 277
 278    ureg_release_temporary(shader, tmp);
 279    ureg_release_temporary(shader, m[0][0]);
 280    ureg_release_temporary(shader, m[0][1]);
 281    ureg_release_temporary(shader, m[1][0]);
 282    ureg_release_temporary(shader, m[1][1]);
 283
 284    ureg_END(shader);
 285
 286    return ureg_create_shader_and_destroy(shader, idct->pipe);
 287 }
 288
 289 static void *
 290 create_matrix_frag_shader(struct vl_idct *idct)
 291 {
 292    struct pipe_resource *matrix = idct->textures.individual.matrix;
 293    struct pipe_resource *source = idct->textures.individual.source;
 294
 295    struct ureg_program *shader;
 296
 297    struct ureg_src tc[2], sampler[2];
 298    struct ureg_src start[2];
 299
 300    struct ureg_dst l[2], r[2];
 301    struct ureg_dst t_tc, tmp, fragment[NR_RENDER_TARGETS];
 302
 303    unsigned i;
 304
 305    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 306    if (!shader)
 307       return NULL;
 308
 309    t_tc = ureg_DECL_temporary(shader);
 310    tmp = ureg_DECL_temporary(shader);
 311
 312    tc[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_TEX, TGSI_INTERPOLATE_LINEAR);
 313    tc[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_BLOCK, TGSI_INTERPOLATE_LINEAR);
 314
 315    sampler[0] = ureg_DECL_sampler(shader, 1);
 316    sampler[1] = ureg_DECL_sampler(shader, 0);
 317
 318    start[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_START, TGSI_INTERPOLATE_CONSTANT);
 319    start[1] = ureg_imm1f(shader, 0.0f);
 320
 321    for (i = 0; i < NR_RENDER_TARGETS; ++i)
 322        fragment[i] = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, i);
 323
 324    fetch_four(shader, l, tc[0], sampler[0], start[0], false, source->width0);
 325    ureg_MUL(shader, l[0], ureg_src(l[0]), ureg_scalar(ureg_imm1f(shader, STAGE1_SCALE), TGSI_SWIZZLE_X));
 326    ureg_MUL(shader, l[1], ureg_src(l[1]), ureg_scalar(ureg_imm1f(shader, STAGE1_SCALE), TGSI_SWIZZLE_X));
 327
 328    for (i = 0; i < NR_RENDER_TARGETS; ++i) {
 329
 330 #if NR_RENDER_TARGETS == 8
 331       ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_X), ureg_imm1f(shader, 1.0f / BLOCK_WIDTH * i));
 332       fetch_four(shader, r, ureg_src(t_tc), sampler[1], start[1], true, matrix->width0);
 333 #elif NR_RENDER_TARGETS == 1
 334       fetch_four(shader, r, tc[1], sampler[1], start[1], true, matrix->width0);
 335 #else
 336 #error invalid number of render targets
 337 #endif
 338
 339       matrix_mul(shader, fragment[i], l, r);
 340       ureg_release_temporary(shader, r[0]);
 341       ureg_release_temporary(shader, r[1]);
 342    }
 343
 344    ureg_release_temporary(shader, t_tc);
 345    ureg_release_temporary(shader, tmp);
 346    ureg_release_temporary(shader, l[0]);
 347    ureg_release_temporary(shader, l[1]);
 348
 349    ureg_END(shader);
 350
 351    return ureg_create_shader_and_destroy(shader, idct->pipe);
 352 }
 353
 354 static void *
 355 create_empty_block_frag_shader(struct vl_idct *idct)
 356 {
 357    struct ureg_program *shader;
 358    struct ureg_dst fragment;
 359
 360    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 361    if (!shader)
 362       return NULL;
 363
 364    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 365
 366    ureg_MOV(shader, fragment, ureg_imm1f(shader, 0.0f));
 367
 368    ureg_END(shader);
 369
 370    return ureg_create_shader_and_destroy(shader, idct->pipe);
 371 }
 372
 373 static bool
 374 init_shaders(struct vl_idct *idct)
 375 {
 376    idct->matrix_vs = create_vert_shader(idct, true);
 377    idct->matrix_fs = create_matrix_frag_shader(idct);
 378
 379    idct->transpose_vs = create_vert_shader(idct, true);
 380    idct->transpose_fs = create_transpose_frag_shader(idct);
 381
 382    idct->eb_vs = create_vert_shader(idct, false);
 383    idct->eb_fs = create_empty_block_frag_shader(idct);
 384
 385    return
 386       idct->transpose_vs != NULL && idct->transpose_fs != NULL &&
 387       idct->matrix_vs != NULL && idct->matrix_fs != NULL &&
 388       idct->eb_vs != NULL && idct->eb_fs != NULL;
 389 }
 390
 391 static void
 392 cleanup_shaders(struct vl_idct *idct)
 393 {
 394    idct->pipe->delete_vs_state(idct->pipe, idct->transpose_vs);
 395    idct->pipe->delete_fs_state(idct->pipe, idct->transpose_fs);
 396
 397    idct->pipe->delete_vs_state(idct->pipe, idct->matrix_vs);
 398    idct->pipe->delete_fs_state(idct->pipe, idct->matrix_fs);
 399
 400    idct->pipe->delete_vs_state(idct->pipe, idct->eb_vs);
 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.last_level = 0;
 419    template.depth0 = 1;
 420    template.bind = PIPE_BIND_SAMPLER_VIEW;
 421    template.flags = 0;
 422
 423    template.target = PIPE_TEXTURE_2D;
 424    template.format = PIPE_FORMAT_R16G16B16A16_SNORM;
 425    template.width0 = idct->destination->width0 / 4;
 426    template.height0 = idct->destination->height0;
 427    template.depth0 = 1;
 428    template.usage = PIPE_USAGE_STREAM;
 429    idct->textures.individual.source = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 430
 431    template.target = PIPE_TEXTURE_3D;
 432    template.format = PIPE_FORMAT_R16_SNORM;
 433    template.width0 = idct->destination->width0 / NR_RENDER_TARGETS;
 434    template.depth0 = NR_RENDER_TARGETS;
 435    template.usage = PIPE_USAGE_STATIC;
 436    idct->textures.individual.intermediate = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 437
 438    for (i = 0; i < 4; ++i) {
 439       if(idct->textures.all[i] == NULL)
 440          return false; /* a texture failed to allocate */
 441
 442       u_sampler_view_default_template(&sampler_view, idct->textures.all[i], idct->textures.all[i]->format);
 443       idct->sampler_views.all[i] = idct->pipe->create_sampler_view(idct->pipe, idct->textures.all[i], &sampler_view);
 444    }
 445
 446    idct->vertex_bufs.individual.quad.stride = sizeof(struct vertex2f);
 447    idct->vertex_bufs.individual.quad.max_index = 4 * idct->max_blocks - 1;
 448    idct->vertex_bufs.individual.quad.buffer_offset = 0;
 449    idct->vertex_bufs.individual.quad.buffer = pipe_buffer_create
 450    (
 451       idct->pipe->screen,
 452       PIPE_BIND_VERTEX_BUFFER,
 453       sizeof(struct vertex2f) * 4 * idct->max_blocks
 454    );
 455
 456    if(idct->vertex_bufs.individual.quad.buffer == NULL)
 457       return false;
 458
 459    idct->vertex_bufs.individual.pos.stride = sizeof(struct vertex2f);
 460    idct->vertex_bufs.individual.pos.max_index = 4 * idct->max_blocks - 1;
 461    idct->vertex_bufs.individual.pos.buffer_offset = 0;
 462    idct->vertex_bufs.individual.pos.buffer = pipe_buffer_create
 463    (
 464       idct->pipe->screen,
 465       PIPE_BIND_VERTEX_BUFFER,
 466       sizeof(struct vertex2f) * 4 * idct->max_blocks
 467    );
 468
 469    if(idct->vertex_bufs.individual.pos.buffer == NULL)
 470       return false;
 471
 472    /* Rect element */
 473    vertex_elems[0].src_offset = 0;
 474    vertex_elems[0].instance_divisor = 0;
 475    vertex_elems[0].vertex_buffer_index = 0;
 476    vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
 477
 478    /* Pos element */
 479    vertex_elems[1].src_offset = 0;
 480    vertex_elems[1].instance_divisor = 0;
 481    vertex_elems[1].vertex_buffer_index = 1;
 482    vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
 483
 484    idct->vertex_elems_state = idct->pipe->create_vertex_elements_state(idct->pipe, 2, vertex_elems);
 485
 486    return true;
 487 }
 488
 489 static void
 490 cleanup_buffers(struct vl_idct *idct)
 491 {
 492    unsigned i;
 493
 494    assert(idct);
 495
 496    for (i = 0; i < 4; ++i) {
 497       pipe_sampler_view_reference(&idct->sampler_views.all[i], NULL);
 498       pipe_resource_reference(&idct->textures.all[i], NULL);
 499    }
 500
 501    idct->pipe->delete_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 502    pipe_resource_reference(&idct->vertex_bufs.individual.quad.buffer, NULL);
 503    pipe_resource_reference(&idct->vertex_bufs.individual.pos.buffer, NULL);
 504 }
 505
 506 static void
 507 init_constants(struct vl_idct *idct)
 508 {
 509    struct pipe_transfer *buf_transfer;
 510    struct vertex2f *v;
 511
 512    unsigned i;
 513
 514    /* quad vectors */
 515    v = pipe_buffer_map
 516    (
 517       idct->pipe,
 518       idct->vertex_bufs.individual.quad.buffer,
 519       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 520       &buf_transfer
 521    );
 522    for ( i = 0; i < idct->max_blocks; ++i)
 523      memcpy(v + i * 4, &const_quad, sizeof(const_quad));
 524    pipe_buffer_unmap(idct->pipe, idct->vertex_bufs.individual.quad.buffer, buf_transfer);
 525 }
 526
 527 static void
 528 init_state(struct vl_idct *idct)
 529 {
 530    struct pipe_sampler_state sampler;
 531    unsigned i;
 532
 533    idct->num_blocks = 0;
 534    idct->num_empty_blocks = 0;
 535
 536    idct->viewport[0].scale[0] = idct->textures.individual.intermediate->width0;
 537    idct->viewport[0].scale[1] = idct->textures.individual.intermediate->height0;
 538
 539    idct->viewport[1].scale[0] = idct->destination->width0;
 540    idct->viewport[1].scale[1] = idct->destination->height0;
 541
 542    idct->fb_state[0].width = idct->textures.individual.intermediate->width0;
 543    idct->fb_state[0].height = idct->textures.individual.intermediate->height0;
 544
 545    idct->fb_state[0].nr_cbufs = NR_RENDER_TARGETS;
 546    for(i = 0; i < NR_RENDER_TARGETS; ++i) {
 547       idct->fb_state[0].cbufs[i] = idct->pipe->screen->get_tex_surface(
 548          idct->pipe->screen, idct->textures.individual.intermediate, 0, 0, i,
 549          PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET);
 550    }
 551
 552    idct->fb_state[1].width = idct->destination->width0;
 553    idct->fb_state[1].height = idct->destination->height0;
 554
 555    idct->fb_state[1].nr_cbufs = 1;
 556    idct->fb_state[1].cbufs[0] = idct->pipe->screen->get_tex_surface(
 557       idct->pipe->screen, idct->destination, 0, 0, 0,
 558       PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET);
 559
 560    for(i = 0; i < 2; ++i) {
 561       idct->viewport[i].scale[2] = 1;
 562       idct->viewport[i].scale[3] = 1;
 563       idct->viewport[i].translate[0] = 0;
 564       idct->viewport[i].translate[1] = 0;
 565       idct->viewport[i].translate[2] = 0;
 566       idct->viewport[i].translate[3] = 0;
 567
 568       idct->fb_state[i].zsbuf = NULL;
 569    }
 570
 571    for (i = 0; i < 4; ++i) {
 572       memset(&sampler, 0, sizeof(sampler));
 573       sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 574       sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 575       sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 576       sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
 577       sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
 578       sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
 579       sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
 580       sampler.compare_func = PIPE_FUNC_ALWAYS;
 581       sampler.normalized_coords = 1;
 582       /*sampler.shadow_ambient = ; */
 583       /*sampler.lod_bias = ; */
 584       sampler.min_lod = 0;
 585       /*sampler.max_lod = ; */
 586       /*sampler.border_color[0] = ; */
 587       /*sampler.max_anisotropy = ; */
 588       idct->samplers.all[i] = idct->pipe->create_sampler_state(idct->pipe, &sampler);
 589    }
 590 }
 591
 592 static void
 593 cleanup_state(struct vl_idct *idct)
 594 {
 595    unsigned i;
 596
 597    for(i = 0; i < NR_RENDER_TARGETS; ++i) {
 598       idct->pipe->screen->tex_surface_destroy(idct->fb_state[0].cbufs[i]);
 599    }
 600
 601    idct->pipe->screen->tex_surface_destroy(idct->fb_state[1].cbufs[0]);
 602
 603    for (i = 0; i < 4; ++i)
 604       idct->pipe->delete_sampler_state(idct->pipe, idct->samplers.all[i]);
 605 }
 606
 607 struct pipe_resource *
 608 vl_idct_upload_matrix(struct pipe_context *pipe)
 609 {
 610    struct pipe_resource template, *matrix;
 611    struct pipe_transfer *buf_transfer;
 612    unsigned i, j, pitch;
 613    float *f;
 614
 615    struct pipe_box rect =
 616    {
 617       0, 0, 0,
 618       BLOCK_WIDTH,
 619       BLOCK_HEIGHT,
 620       1
 621    };
 622
 623    memset(&template, 0, sizeof(struct pipe_resource));
 624    template.target = PIPE_TEXTURE_2D;
 625    template.format = PIPE_FORMAT_R32G32B32A32_FLOAT;
 626    template.last_level = 0;
 627    template.width0 = 2;
 628    template.height0 = 8;
 629    template.depth0 = 1;
 630    template.usage = PIPE_USAGE_IMMUTABLE;
 631    template.bind = PIPE_BIND_SAMPLER_VIEW;
 632    template.flags = 0;
 633
 634    matrix = pipe->screen->resource_create(pipe->screen, &template);
 635
 636    /* matrix */
 637    buf_transfer = pipe->get_transfer
 638    (
 639       pipe, matrix,
 640       u_subresource(0, 0),
 641       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 642       &rect
 643    );
 644    pitch = buf_transfer->stride / sizeof(float);
 645
 646    f = pipe->transfer_map(pipe, buf_transfer);
 647    for(i = 0; i < BLOCK_HEIGHT; ++i)
 648       for(j = 0; j < BLOCK_WIDTH; ++j)
 649          f[i * pitch + j] = const_matrix[j][i]; // transpose
 650
 651    pipe->transfer_unmap(pipe, buf_transfer);
 652    pipe->transfer_destroy(pipe, buf_transfer);
 653
 654    return matrix;
 655 }
 656
 657 static void
 658 xfer_buffers_map(struct vl_idct *idct)
 659 {
 660    struct pipe_box rect =
 661    {
 662       0, 0, 0,
 663       idct->textures.individual.source->width0,
 664       idct->textures.individual.source->height0,
 665       1
 666    };
 667
 668    idct->tex_transfer = idct->pipe->get_transfer
 669    (
 670       idct->pipe, idct->textures.individual.source,
 671       u_subresource(0, 0),
 672       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 673       &rect
 674    );
 675
 676    idct->texels = idct->pipe->transfer_map(idct->pipe, idct->tex_transfer);
 677
 678    idct->vectors = pipe_buffer_map
 679    (
 680       idct->pipe,
 681       idct->vertex_bufs.individual.pos.buffer,
 682       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 683       &idct->vec_transfer
 684    );
 685
 686    idct->next_empty_block.l_x = ~1;
 687    idct->next_empty_block.l_y = ~1;
 688    idct->next_empty_block.r_x = ~1;
 689    idct->next_empty_block.r_y = ~1;
 690 }
 691
 692 static void
 693 xfer_buffers_unmap(struct vl_idct *idct)
 694 {
 695    pipe_buffer_unmap(idct->pipe, idct->vertex_bufs.individual.pos.buffer, idct->vec_transfer);
 696
 697    idct->pipe->transfer_unmap(idct->pipe, idct->tex_transfer);
 698    idct->pipe->transfer_destroy(idct->pipe, idct->tex_transfer);
 699 }
 700
 701 bool
 702 vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe, struct pipe_resource *dst, struct pipe_resource *matrix)
 703 {
 704    assert(idct && pipe && dst);
 705
 706    idct->pipe = pipe;
 707    pipe_resource_reference(&idct->textures.individual.matrix, matrix);
 708    pipe_resource_reference(&idct->textures.individual.transpose, matrix);
 709    pipe_resource_reference(&idct->destination, dst);
 710
 711    if(!init_buffers(idct))
 712       return false;
 713
 714    if(!init_shaders(idct)) {
 715       cleanup_buffers(idct);
 716       return false;
 717    }
 718
 719    init_state(idct);
 720
 721    init_constants(idct);
 722    xfer_buffers_map(idct);
 723
 724    return true;
 725 }
 726
 727 void
 728 vl_idct_cleanup(struct vl_idct *idct)
 729 {
 730    cleanup_shaders(idct);
 731    cleanup_buffers(idct);
 732
 733    cleanup_state(idct);
 734
 735    pipe_resource_reference(&idct->destination, NULL);
 736 }
 737
 738 static void
 739 flush_empty_block(struct vl_idct *idct, unsigned new_x, unsigned new_y)
 740 {
 741    if (idct->next_empty_block.l_x == ~1 ||
 742        idct->next_empty_block.l_y == ~1) {
 743
 744       idct->next_empty_block.l_x = new_x;
 745       idct->next_empty_block.l_y = new_y;
 746
 747    } else if (idct->next_empty_block.r_x != (new_x - 1) ||
 748               idct->next_empty_block.r_y != new_y) {
 749
 750       struct vertex2f l, r, *v_dst;
 751
 752       v_dst = idct->vectors + (idct->max_blocks - idct->num_empty_blocks) * 4 - 4;
 753
 754       l.x = idct->next_empty_block.l_x;
 755       l.y = idct->next_empty_block.l_y;
 756       r.x = idct->next_empty_block.r_x;
 757       r.y = idct->next_empty_block.r_y;
 758       v_dst[0] = v_dst[3] = l;
 759       v_dst[1] = v_dst[2] = r;
 760
 761       idct->next_empty_block.l_x = new_x;
 762       idct->next_empty_block.l_y = new_y;
 763       idct->num_empty_blocks++;
 764    }
 765
 766    idct->next_empty_block.r_x = new_x;
 767    idct->next_empty_block.r_y = new_y;
 768 }
 769
 770 void
 771 vl_idct_add_block(struct vl_idct *idct, unsigned x, unsigned y, short *block)
 772 {
 773    struct vertex2f v, *v_dst;
 774
 775    unsigned tex_pitch;
 776    short *texels;
 777
 778    unsigned i;
 779
 780    assert(idct);
 781
 782    if(block) {
 783       tex_pitch = idct->tex_transfer->stride / sizeof(short);
 784       texels = idct->texels + y * tex_pitch * BLOCK_HEIGHT + x * BLOCK_WIDTH;
 785
 786       for (i = 0; i < BLOCK_HEIGHT; ++i)
 787          memcpy(texels + i * tex_pitch, block + i * BLOCK_WIDTH, BLOCK_WIDTH * sizeof(short));
 788
 789       /* non empty blocks fills the vector buffer from left to right */
 790       v_dst = idct->vectors + idct->num_blocks * 4;
 791
 792       idct->num_blocks++;
 793
 794       v.x = x;
 795       v.y = y;
 796
 797       for (i = 0; i < 4; ++i) {
 798          v_dst[i] = v;
 799       }
 800
 801    } else {
 802
 803       /* while empty blocks fills the vector buffer from right to left */
 804       flush_empty_block(idct, x, y);
 805    }
 806 }
 807
 808 void
 809 vl_idct_flush(struct vl_idct *idct)
 810 {
 811    flush_empty_block(idct, ~1, ~1);
 812    xfer_buffers_unmap(idct);
 813
 814    if(idct->num_blocks > 0) {
 815
 816       /* first stage */
 817       idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state[0]);
 818       idct->pipe->set_viewport_state(idct->pipe, &idct->viewport[0]);
 819
 820       idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all);
 821       idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 822       idct->pipe->set_fragment_sampler_views(idct->pipe, 2, idct->sampler_views.stage[0]);
 823       idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[0]);
 824       idct->pipe->bind_vs_state(idct->pipe, idct->matrix_vs);
 825       idct->pipe->bind_fs_state(idct->pipe, idct->matrix_fs);
 826
 827       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, idct->num_blocks * 4);
 828
 829       /* second stage */
 830       idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state[1]);
 831       idct->pipe->set_viewport_state(idct->pipe, &idct->viewport[1]);
 832
 833       idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all);
 834       idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 835       idct->pipe->set_fragment_sampler_views(idct->pipe, 2, idct->sampler_views.stage[1]);
 836       idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[1]);
 837       idct->pipe->bind_vs_state(idct->pipe, idct->transpose_vs);
 838       idct->pipe->bind_fs_state(idct->pipe, idct->transpose_fs);
 839
 840       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, idct->num_blocks * 4);
 841    }
 842
 843    if(idct->num_empty_blocks > 0) {
 844
 845       /* empty block handling */
 846       idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state[1]);
 847       idct->pipe->set_viewport_state(idct->pipe, &idct->viewport[1]);
 848
 849       idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all);
 850       idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 851       idct->pipe->bind_vs_state(idct->pipe, idct->eb_vs);
 852       idct->pipe->bind_fs_state(idct->pipe, idct->eb_fs);
 853
 854       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS,
 855          (idct->max_blocks - idct->num_empty_blocks) * 4,
 856          idct->num_empty_blocks * 4);
 857    }
 858
 859    idct->num_blocks = 0;
 860    idct->num_empty_blocks = 0;
 861    xfer_buffers_map(idct);
 862 }