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