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    ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Y), tex);
 288    for (i = 0; i < 4; ++i) {
 289       fetch_four(shader, l[i], ureg_src(t_tc), sampler[0], start[0], block, false, false, source->width0);
 290       ureg_MUL(shader, l[i][0], ureg_src(l[i][0]), ureg_imm1f(shader, STAGE1_SCALE));
 291       ureg_MUL(shader, l[i][1], ureg_src(l[i][1]), ureg_imm1f(shader, STAGE1_SCALE));
 292       if(i != 3)
 293          ureg_ADD(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_Y),
 294             ureg_src(t_tc), ureg_imm1f(shader, 1.0f / source->height0));
 295    }
 296
 297    for (i = 0; i < NR_RENDER_TARGETS; ++i) {
 298
 299 #if NR_RENDER_TARGETS == 8
 300       ureg_MOV(shader, ureg_writemask(t_tc, TGSI_WRITEMASK_X), ureg_imm1f(shader, 1.0f / BLOCK_WIDTH * i));
 301       fetch_four(shader, r, ureg_src(t_tc), sampler[1], start[1], block, true, true, matrix->width0);
 302 #elif NR_RENDER_TARGETS == 1
 303       fetch_four(shader, r, block, sampler[1], start[1], block, true, true, matrix->width0);
 304 #else
 305 #error invalid number of render targets
 306 #endif
 307
 308       for (j = 0; j < 4; ++j) {
 309          matrix_mul(shader, ureg_writemask(fragment[i], TGSI_WRITEMASK_X << j), l[j], r);
 310       }
 311       ureg_release_temporary(shader, r[0]);
 312       ureg_release_temporary(shader, r[1]);
 313    }
 314
 315    ureg_release_temporary(shader, t_tc);
 316    ureg_release_temporary(shader, tmp);
 317
 318    for (i = 0; i < 4; ++i) {
 319       ureg_release_temporary(shader, l[i][0]);
 320       ureg_release_temporary(shader, l[i][1]);
 321    }
 322
 323    ureg_END(shader);
 324
 325    return ureg_create_shader_and_destroy(shader, idct->pipe);
 326 }
 327
 328 static void *
 329 create_empty_block_frag_shader(struct vl_idct *idct)
 330 {
 331    struct ureg_program *shader;
 332    struct ureg_dst fragment;
 333
 334    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 335    if (!shader)
 336       return NULL;
 337
 338    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 339
 340    ureg_MOV(shader, fragment, ureg_imm1f(shader, 0.0f));
 341
 342    ureg_END(shader);
 343
 344    return ureg_create_shader_and_destroy(shader, idct->pipe);
 345 }
 346
 347 static bool
 348 init_shaders(struct vl_idct *idct)
 349 {
 350    idct->matrix_vs = create_vert_shader(idct, true);
 351    idct->matrix_fs = create_matrix_frag_shader(idct);
 352
 353    idct->transpose_vs = create_vert_shader(idct, true);
 354    idct->transpose_fs = create_transpose_frag_shader(idct);
 355
 356    idct->eb_vs = create_vert_shader(idct, false);
 357    idct->eb_fs = create_empty_block_frag_shader(idct);
 358
 359    return
 360       idct->transpose_vs != NULL && idct->transpose_fs != NULL &&
 361       idct->matrix_vs != NULL && idct->matrix_fs != NULL &&
 362       idct->eb_vs != NULL && idct->eb_fs != NULL;
 363 }
 364
 365 static void
 366 cleanup_shaders(struct vl_idct *idct)
 367 {
 368    idct->pipe->delete_vs_state(idct->pipe, idct->transpose_vs);
 369    idct->pipe->delete_fs_state(idct->pipe, idct->transpose_fs);
 370
 371    idct->pipe->delete_vs_state(idct->pipe, idct->matrix_vs);
 372    idct->pipe->delete_fs_state(idct->pipe, idct->matrix_fs);
 373
 374    idct->pipe->delete_vs_state(idct->pipe, idct->eb_vs);
 375    idct->pipe->delete_fs_state(idct->pipe, idct->eb_fs);
 376 }
 377
 378 static bool
 379 init_buffers(struct vl_idct *idct)
 380 {
 381    struct pipe_resource template;
 382    struct pipe_sampler_view sampler_view;
 383    struct pipe_vertex_element vertex_elems[2];
 384    unsigned i;
 385
 386    memset(&template, 0, sizeof(struct pipe_resource));
 387    template.last_level = 0;
 388    template.depth0 = 1;
 389    template.bind = PIPE_BIND_SAMPLER_VIEW;
 390    template.flags = 0;
 391
 392    template.target = PIPE_TEXTURE_2D;
 393    template.format = PIPE_FORMAT_R16G16B16A16_SNORM;
 394    template.width0 = idct->destination->width0 / 4;
 395    template.height0 = idct->destination->height0;
 396    template.depth0 = 1;
 397    template.usage = PIPE_USAGE_STREAM;
 398    idct->textures.individual.source = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 399
 400    template.target = PIPE_TEXTURE_3D;
 401    template.format = PIPE_FORMAT_R16G16B16A16_SNORM;
 402    template.width0 = idct->destination->width0 / NR_RENDER_TARGETS;
 403    template.height0 = idct->destination->height0 / 4;
 404    template.depth0 = NR_RENDER_TARGETS;
 405    template.usage = PIPE_USAGE_STATIC;
 406    idct->textures.individual.intermediate = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 407
 408    for (i = 0; i < 4; ++i) {
 409       if(idct->textures.all[i] == NULL)
 410          return false; /* a texture failed to allocate */
 411
 412       u_sampler_view_default_template(&sampler_view, idct->textures.all[i], idct->textures.all[i]->format);
 413       idct->sampler_views.all[i] = idct->pipe->create_sampler_view(idct->pipe, idct->textures.all[i], &sampler_view);
 414    }
 415
 416    idct->vertex_bufs.individual.quad = vl_vb_upload_quads(idct->pipe, idct->max_blocks);
 417
 418    if(idct->vertex_bufs.individual.quad.buffer == NULL)
 419       return false;
 420
 421    idct->vertex_bufs.individual.pos.stride = sizeof(struct vertex2f);
 422    idct->vertex_bufs.individual.pos.max_index = 4 * idct->max_blocks - 1;
 423    idct->vertex_bufs.individual.pos.buffer_offset = 0;
 424    idct->vertex_bufs.individual.pos.buffer = pipe_buffer_create
 425    (
 426       idct->pipe->screen,
 427       PIPE_BIND_VERTEX_BUFFER,
 428       sizeof(struct vertex2f) * 4 * idct->max_blocks
 429    );
 430
 431    if(idct->vertex_bufs.individual.pos.buffer == NULL)
 432       return false;
 433
 434    /* Rect element */
 435    vertex_elems[0].src_offset = 0;
 436    vertex_elems[0].instance_divisor = 0;
 437    vertex_elems[0].vertex_buffer_index = 0;
 438    vertex_elems[0].src_format = PIPE_FORMAT_R32G32_FLOAT;
 439
 440    /* Pos element */
 441    vertex_elems[1].src_offset = 0;
 442    vertex_elems[1].instance_divisor = 0;
 443    vertex_elems[1].vertex_buffer_index = 1;
 444    vertex_elems[1].src_format = PIPE_FORMAT_R32G32_FLOAT;
 445
 446    idct->vertex_elems_state = idct->pipe->create_vertex_elements_state(idct->pipe, 2, vertex_elems);
 447
 448    return true;
 449 }
 450
 451 static void
 452 cleanup_buffers(struct vl_idct *idct)
 453 {
 454    unsigned i;
 455
 456    assert(idct);
 457
 458    for (i = 0; i < 4; ++i) {
 459       pipe_sampler_view_reference(&idct->sampler_views.all[i], NULL);
 460       pipe_resource_reference(&idct->textures.all[i], NULL);
 461    }
 462
 463    idct->pipe->delete_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 464    pipe_resource_reference(&idct->vertex_bufs.individual.quad.buffer, NULL);
 465    pipe_resource_reference(&idct->vertex_bufs.individual.pos.buffer, NULL);
 466 }
 467
 468 static void
 469 init_state(struct vl_idct *idct)
 470 {
 471    struct pipe_sampler_state sampler;
 472    struct pipe_rasterizer_state rs_state;
 473    unsigned i;
 474
 475    idct->viewport[0].scale[0] = idct->textures.individual.intermediate->width0;
 476    idct->viewport[0].scale[1] = idct->textures.individual.intermediate->height0;
 477
 478    idct->viewport[1].scale[0] = idct->destination->width0;
 479    idct->viewport[1].scale[1] = idct->destination->height0;
 480
 481    idct->fb_state[0].width = idct->textures.individual.intermediate->width0;
 482    idct->fb_state[0].height = idct->textures.individual.intermediate->height0;
 483
 484    idct->fb_state[0].nr_cbufs = NR_RENDER_TARGETS;
 485    for(i = 0; i < NR_RENDER_TARGETS; ++i) {
 486       idct->fb_state[0].cbufs[i] = idct->pipe->screen->get_tex_surface(
 487          idct->pipe->screen, idct->textures.individual.intermediate, 0, 0, i,
 488          PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET);
 489    }
 490
 491    idct->fb_state[1].width = idct->destination->width0;
 492    idct->fb_state[1].height = idct->destination->height0;
 493
 494    idct->fb_state[1].nr_cbufs = 1;
 495    idct->fb_state[1].cbufs[0] = idct->pipe->screen->get_tex_surface(
 496       idct->pipe->screen, idct->destination, 0, 0, 0,
 497       PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET);
 498
 499    for(i = 0; i < 2; ++i) {
 500       idct->viewport[i].scale[2] = 1;
 501       idct->viewport[i].scale[3] = 1;
 502       idct->viewport[i].translate[0] = 0;
 503       idct->viewport[i].translate[1] = 0;
 504       idct->viewport[i].translate[2] = 0;
 505       idct->viewport[i].translate[3] = 0;
 506
 507       idct->fb_state[i].zsbuf = NULL;
 508    }
 509
 510    for (i = 0; i < 4; ++i) {
 511       memset(&sampler, 0, sizeof(sampler));
 512       sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 513       sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 514       sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 515       sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
 516       sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
 517       sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
 518       sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
 519       sampler.compare_func = PIPE_FUNC_ALWAYS;
 520       sampler.normalized_coords = 1;
 521       /*sampler.shadow_ambient = ; */
 522       /*sampler.lod_bias = ; */
 523       sampler.min_lod = 0;
 524       /*sampler.max_lod = ; */
 525       /*sampler.border_color[0] = ; */
 526       /*sampler.max_anisotropy = ; */
 527       idct->samplers.all[i] = idct->pipe->create_sampler_state(idct->pipe, &sampler);
 528    }
 529
 530    memset(&rs_state, 0, sizeof(rs_state));
 531    /*rs_state.sprite_coord_enable */
 532    rs_state.sprite_coord_mode = PIPE_SPRITE_COORD_UPPER_LEFT;
 533    rs_state.point_quad_rasterization = true;
 534    rs_state.point_size = BLOCK_WIDTH;
 535    rs_state.gl_rasterization_rules = false;
 536    idct->rs_state = idct->pipe->create_rasterizer_state(idct->pipe, &rs_state);
 537 }
 538
 539 static void
 540 cleanup_state(struct vl_idct *idct)
 541 {
 542    unsigned i;
 543
 544    for(i = 0; i < NR_RENDER_TARGETS; ++i) {
 545       idct->pipe->screen->tex_surface_destroy(idct->fb_state[0].cbufs[i]);
 546    }
 547
 548    idct->pipe->screen->tex_surface_destroy(idct->fb_state[1].cbufs[0]);
 549
 550    for (i = 0; i < 4; ++i)
 551       idct->pipe->delete_sampler_state(idct->pipe, idct->samplers.all[i]);
 552
 553    idct->pipe->delete_rasterizer_state(idct->pipe, idct->rs_state);
 554 }
 555
 556 struct pipe_resource *
 557 vl_idct_upload_matrix(struct pipe_context *pipe)
 558 {
 559    struct pipe_resource template, *matrix;
 560    struct pipe_transfer *buf_transfer;
 561    unsigned i, j, pitch;
 562    float *f;
 563
 564    struct pipe_box rect =
 565    {
 566       0, 0, 0,
 567       BLOCK_WIDTH,
 568       BLOCK_HEIGHT,
 569       1
 570    };
 571
 572    memset(&template, 0, sizeof(struct pipe_resource));
 573    template.target = PIPE_TEXTURE_2D;
 574    template.format = PIPE_FORMAT_R32G32B32A32_FLOAT;
 575    template.last_level = 0;
 576    template.width0 = 2;
 577    template.height0 = 8;
 578    template.depth0 = 1;
 579    template.usage = PIPE_USAGE_IMMUTABLE;
 580    template.bind = PIPE_BIND_SAMPLER_VIEW;
 581    template.flags = 0;
 582
 583    matrix = pipe->screen->resource_create(pipe->screen, &template);
 584
 585    /* matrix */
 586    buf_transfer = pipe->get_transfer
 587    (
 588       pipe, matrix,
 589       u_subresource(0, 0),
 590       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 591       &rect
 592    );
 593    pitch = buf_transfer->stride / sizeof(float);
 594
 595    f = pipe->transfer_map(pipe, buf_transfer);
 596    for(i = 0; i < BLOCK_HEIGHT; ++i)
 597       for(j = 0; j < BLOCK_WIDTH; ++j)
 598          f[i * pitch + j] = const_matrix[j][i]; // transpose
 599
 600    pipe->transfer_unmap(pipe, buf_transfer);
 601    pipe->transfer_destroy(pipe, buf_transfer);
 602
 603    return matrix;
 604 }
 605
 606 static void
 607 xfer_buffers_map(struct vl_idct *idct)
 608 {
 609    struct pipe_box rect =
 610    {
 611       0, 0, 0,
 612       idct->textures.individual.source->width0,
 613       idct->textures.individual.source->height0,
 614       1
 615    };
 616
 617    idct->tex_transfer = idct->pipe->get_transfer
 618    (
 619       idct->pipe, idct->textures.individual.source,
 620       u_subresource(0, 0),
 621       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 622       &rect
 623    );
 624
 625    idct->texels = idct->pipe->transfer_map(idct->pipe, idct->tex_transfer);
 626 }
 627
 628 static void
 629 xfer_buffers_unmap(struct vl_idct *idct)
 630 {
 631    idct->pipe->transfer_unmap(idct->pipe, idct->tex_transfer);
 632    idct->pipe->transfer_destroy(idct->pipe, idct->tex_transfer);
 633 }
 634
 635 bool
 636 vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe, struct pipe_resource *dst, struct pipe_resource *matrix)
 637 {
 638    assert(idct && pipe && dst);
 639
 640    idct->pipe = pipe;
 641    pipe_resource_reference(&idct->textures.individual.matrix, matrix);
 642    pipe_resource_reference(&idct->textures.individual.transpose, matrix);
 643    pipe_resource_reference(&idct->destination, dst);
 644
 645    idct->max_blocks =
 646       align(idct->destination->width0, BLOCK_WIDTH) / BLOCK_WIDTH *
 647       align(idct->destination->height0, BLOCK_HEIGHT) / BLOCK_HEIGHT *
 648       idct->destination->depth0;
 649
 650    if(!init_buffers(idct))
 651       return false;
 652
 653    if(!init_shaders(idct)) {
 654       cleanup_buffers(idct);
 655       return false;
 656    }
 657
 658    if(!vl_vb_init(&idct->blocks, idct->max_blocks)) {
 659       cleanup_shaders(idct);
 660       cleanup_buffers(idct);
 661       return false;
 662    }
 663
 664    init_state(idct);
 665
 666    xfer_buffers_map(idct);
 667
 668    return true;
 669 }
 670
 671 void
 672 vl_idct_cleanup(struct vl_idct *idct)
 673 {
 674    vl_vb_cleanup(&idct->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    tex_pitch = idct->tex_transfer->stride / sizeof(short);
 694    texels = idct->texels + y * tex_pitch * BLOCK_HEIGHT + x * BLOCK_WIDTH;
 695
 696    for (i = 0; i < BLOCK_HEIGHT; ++i)
 697       memcpy(texels + i * tex_pitch, block + i * BLOCK_WIDTH, BLOCK_WIDTH * sizeof(short));
 698
 699    vl_vb_add_block(&idct->blocks, x, y);
 700 }
 701
 702 void
 703 vl_idct_flush(struct vl_idct *idct)
 704 {
 705    struct pipe_transfer *vec_transfer;
 706    struct quadf *vectors;
 707    unsigned num_blocks;
 708
 709    assert(idct);
 710
 711    vectors = pipe_buffer_map
 712    (
 713       idct->pipe,
 714       idct->vertex_bufs.individual.pos.buffer,
 715       PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 716       &vec_transfer
 717    );
 718
 719    num_blocks = vl_vb_upload(&idct->blocks, vectors);
 720
 721    pipe_buffer_unmap(idct->pipe, idct->vertex_bufs.individual.pos.buffer, vec_transfer);
 722
 723    xfer_buffers_unmap(idct);
 724
 725    if(num_blocks > 0) {
 726
 727       idct->pipe->bind_rasterizer_state(idct->pipe, idct->rs_state);
 728
 729       /* first stage */
 730       idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state[0]);
 731       idct->pipe->set_viewport_state(idct->pipe, &idct->viewport[0]);
 732
 733       idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all);
 734       idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 735       idct->pipe->set_fragment_sampler_views(idct->pipe, 2, idct->sampler_views.stage[0]);
 736       idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[0]);
 737       idct->pipe->bind_vs_state(idct->pipe, idct->matrix_vs);
 738       idct->pipe->bind_fs_state(idct->pipe, idct->matrix_fs);
 739
 740       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, num_blocks * 4);
 741
 742       /* second stage */
 743       idct->pipe->set_framebuffer_state(idct->pipe, &idct->fb_state[1]);
 744       idct->pipe->set_viewport_state(idct->pipe, &idct->viewport[1]);
 745
 746       idct->pipe->set_vertex_buffers(idct->pipe, 2, idct->vertex_bufs.all);
 747       idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 748       idct->pipe->set_fragment_sampler_views(idct->pipe, 2, idct->sampler_views.stage[1]);
 749       idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[1]);
 750       idct->pipe->bind_vs_state(idct->pipe, idct->transpose_vs);
 751       idct->pipe->bind_fs_state(idct->pipe, idct->transpose_fs);
 752
 753       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, num_blocks * 4);
 754    }
 755
 756    xfer_buffers_map(idct);
 757 }