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_L_ADDR0,
  62    VS_O_L_ADDR1,
  63    VS_O_R_ADDR0,
  64    VS_O_R_ADDR1
  65 };
  66
  67 static const float const_matrix[8][8] = {
  68    {  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.353553f,  0.3535530f },
  69    {  0.4903930f,  0.4157350f,  0.2777850f,  0.0975451f, -0.0975452f, -0.2777850f, -0.415735f, -0.4903930f },
  70    {  0.4619400f,  0.1913420f, -0.1913420f, -0.4619400f, -0.4619400f, -0.1913420f,  0.191342f,  0.4619400f },
  71    {  0.4157350f, -0.0975452f, -0.4903930f, -0.2777850f,  0.2777850f,  0.4903930f,  0.097545f, -0.4157350f },
  72    {  0.3535530f, -0.3535530f, -0.3535530f,  0.3535540f,  0.3535530f, -0.3535540f, -0.353553f,  0.3535530f },
  73    {  0.2777850f, -0.4903930f,  0.0975452f,  0.4157350f, -0.4157350f, -0.0975451f,  0.490393f, -0.2777850f },
  74    {  0.1913420f, -0.4619400f,  0.4619400f, -0.1913420f, -0.1913410f,  0.4619400f, -0.461940f,  0.1913420f },
  75    {  0.0975451f, -0.2777850f,  0.4157350f, -0.4903930f,  0.4903930f, -0.4157350f,  0.277786f, -0.0975458f }
  76 };
  77
  78 static void
  79 calc_addr(struct ureg_program *shader, struct ureg_dst addr[2],
  80           struct ureg_src tc, struct ureg_src start, bool right_side,
  81           bool transposed, float size)
  82 {
  83    unsigned wm_start = (right_side == transposed) ? TGSI_WRITEMASK_X : TGSI_WRITEMASK_Y;
  84    unsigned sw_start = right_side ? TGSI_SWIZZLE_Y : TGSI_SWIZZLE_X;
  85
  86    unsigned wm_tc = (right_side == transposed) ? TGSI_WRITEMASK_Y : TGSI_WRITEMASK_X;
  87    unsigned sw_tc = right_side ? TGSI_SWIZZLE_X : TGSI_SWIZZLE_Y;
  88
  89    /*
  90     * addr[0..1].(start) = right_side ? start.x : tc.x
  91     * addr[0..1].(tc) = right_side ? tc.y : start.y
  92     * addr[0..1].z = tc.z
  93     * addr[1].(start) += 1.0f / scale
  94     */
  95    ureg_MOV(shader, ureg_writemask(addr[0], wm_start), ureg_scalar(start, sw_start));
  96    ureg_MOV(shader, ureg_writemask(addr[0], wm_tc), ureg_scalar(tc, sw_tc));
  97    ureg_MOV(shader, ureg_writemask(addr[0], TGSI_WRITEMASK_Z), tc);
  98
  99    ureg_ADD(shader, ureg_writemask(addr[1], wm_start), ureg_scalar(start, sw_start), ureg_imm1f(shader, 1.0f / size));
 100    ureg_MOV(shader, ureg_writemask(addr[1], wm_tc), ureg_scalar(tc, sw_tc));
 101    ureg_MOV(shader, ureg_writemask(addr[1], TGSI_WRITEMASK_Z), tc);
 102 }
 103
 104 static void *
 105 create_vert_shader(struct vl_idct *idct, bool matrix_stage)
 106 {
 107    struct ureg_program *shader;
 108    struct ureg_src scale;
 109    struct ureg_src vrect, vpos;
 110    struct ureg_dst t_tex, t_start;
 111    struct ureg_dst o_vpos, o_l_addr[2], o_r_addr[2];
 112
 113    shader = ureg_create(TGSI_PROCESSOR_VERTEX);
 114    if (!shader)
 115       return NULL;
 116
 117    t_tex = ureg_DECL_temporary(shader);
 118    t_start = ureg_DECL_temporary(shader);
 119
 120    vrect = ureg_DECL_vs_input(shader, VS_I_RECT);
 121    vpos = ureg_DECL_vs_input(shader, VS_I_VPOS);
 122
 123    o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
 124
 125    o_l_addr[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR0);
 126    o_l_addr[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR1);
 127
 128    o_r_addr[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR0);
 129    o_r_addr[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR1);
 130
 131    /*
 132     * scale = (BLOCK_WIDTH, BLOCK_HEIGHT) / (dst.width, dst.height)
 133     *
 134     * t_vpos = vpos + vrect
 135     * o_vpos.xy = t_vpos * scale
 136     * o_vpos.zw = vpos
 137     *
 138     * o_l_addr = calc_addr(...)
 139     * o_r_addr = calc_addr(...)
 140     *
 141     */
 142    scale = ureg_imm2f(shader,
 143       (float)BLOCK_WIDTH / idct->buffer_width,
 144       (float)BLOCK_HEIGHT / idct->buffer_height);
 145
 146    ureg_ADD(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_XY), vpos, vrect);
 147    ureg_MUL(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_XY), ureg_src(t_tex), scale);
 148    ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_XY), ureg_src(t_tex));
 149    ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_ZW), vpos);
 150
 151    ureg_MUL(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_Z),
 152       ureg_scalar(vrect, TGSI_SWIZZLE_X),
 153       ureg_imm1f(shader, BLOCK_WIDTH / NR_RENDER_TARGETS));
 154
 155    ureg_MUL(shader, ureg_writemask(t_start, TGSI_WRITEMASK_XY), vpos, scale);
 156
 157    if(matrix_stage) {
 158       calc_addr(shader, o_l_addr, ureg_src(t_tex), ureg_src(t_start), false, false, idct->buffer_width / 4);
 159       calc_addr(shader, o_r_addr, vrect, ureg_imm1f(shader, 0.0f), true, true, BLOCK_WIDTH / 4);
 160    } else {
 161       calc_addr(shader, o_l_addr, vrect, ureg_imm1f(shader, 0.0f), false, false, BLOCK_WIDTH / 4);
 162       calc_addr(shader, o_r_addr, ureg_src(t_tex), ureg_src(t_start), true, false, idct->buffer_height / 4);
 163    }
 164
 165    ureg_release_temporary(shader, t_tex);
 166    ureg_release_temporary(shader, t_start);
 167
 168    ureg_END(shader);
 169
 170    return ureg_create_shader_and_destroy(shader, idct->pipe);
 171 }
 172
 173 static void
 174 increment_addr(struct ureg_program *shader, struct ureg_dst addr[2],
 175                bool right_side, bool transposed, float size)
 176 {
 177    unsigned wm_tc = (right_side == transposed) ? TGSI_WRITEMASK_Y : TGSI_WRITEMASK_X;
 178
 179    /* addr[0..1]++ */
 180    ureg_ADD(shader, ureg_writemask(addr[0], wm_tc),
 181       ureg_src(addr[0]), ureg_imm1f(shader, 1.0f / size));
 182    ureg_ADD(shader, ureg_writemask(addr[1], wm_tc),
 183       ureg_src(addr[1]), ureg_imm1f(shader, 1.0f / size));
 184 }
 185
 186 static void
 187 fetch_four(struct ureg_program *shader, struct ureg_dst m[2], struct ureg_src addr[2], struct ureg_src sampler)
 188 {
 189    m[0] = ureg_DECL_temporary(shader);
 190    m[1] = ureg_DECL_temporary(shader);
 191
 192    ureg_TEX(shader, m[0], TGSI_TEXTURE_3D, addr[0], sampler);
 193    ureg_TEX(shader, m[1], TGSI_TEXTURE_3D, addr[1], sampler);
 194 }
 195
 196 static void
 197 matrix_mul(struct ureg_program *shader, struct ureg_dst dst, struct ureg_dst l[2], struct ureg_dst r[2])
 198 {
 199    struct ureg_dst tmp;
 200
 201    tmp = ureg_DECL_temporary(shader);
 202
 203    /*
 204     * tmp.xy = dot4(m[0][0..1], m[1][0..1])
 205     * dst = tmp.x + tmp.y
 206     */
 207    ureg_DP4(shader, ureg_writemask(tmp, TGSI_WRITEMASK_X), ureg_src(l[0]), ureg_src(r[0]));
 208    ureg_DP4(shader, ureg_writemask(tmp, TGSI_WRITEMASK_Y), ureg_src(l[1]), ureg_src(r[1]));
 209    ureg_ADD(shader, dst,
 210       ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_X),
 211       ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_Y));
 212
 213    ureg_release_temporary(shader, tmp);
 214 }
 215
 216 static void *
 217 create_matrix_frag_shader(struct vl_idct *idct)
 218 {
 219    struct ureg_program *shader;
 220
 221    struct ureg_src l_addr[2], r_addr[2], saddr[2];
 222
 223    struct ureg_dst addr[2], l[4][2], r[2];
 224    struct ureg_dst fragment[NR_RENDER_TARGETS];
 225
 226    unsigned i, j;
 227
 228    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 229    if (!shader)
 230       return NULL;
 231
 232    addr[0] = ureg_DECL_temporary(shader);
 233    addr[1] = ureg_DECL_temporary(shader);
 234
 235    saddr[0] = ureg_src(addr[0]);
 236    saddr[1] = ureg_src(addr[1]);
 237
 238    l_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR0, TGSI_INTERPOLATE_LINEAR);
 239    l_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR1, TGSI_INTERPOLATE_LINEAR);
 240
 241    r_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR0, TGSI_INTERPOLATE_LINEAR);
 242    r_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR1, TGSI_INTERPOLATE_LINEAR);
 243
 244    for (i = 0; i < NR_RENDER_TARGETS; ++i)
 245        fragment[i] = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, i);
 246
 247    for (i = 0; i < 4; ++i) {
 248       if(i == 0) {
 249          ureg_MOV(shader, addr[0], l_addr[0]);
 250          ureg_MOV(shader, addr[1], l_addr[1]);
 251       } else
 252          increment_addr(shader, addr, false, false, idct->buffer_height);
 253
 254       fetch_four(shader, l[i], saddr, ureg_DECL_sampler(shader, 1));
 255    }
 256
 257    for (i = 0; i < NR_RENDER_TARGETS; ++i) {
 258       if(i == 0) {
 259          ureg_MOV(shader, addr[0], r_addr[0]);
 260          ureg_MOV(shader, addr[1], r_addr[1]);
 261       } else
 262          increment_addr(shader, addr, true, true, BLOCK_HEIGHT);
 263
 264       fetch_four(shader, r, saddr, ureg_DECL_sampler(shader, 0));
 265
 266       for (j = 0; j < 4; ++j) {
 267          matrix_mul(shader, ureg_writemask(fragment[i], TGSI_WRITEMASK_X << j), l[j], r);
 268       }
 269       ureg_release_temporary(shader, r[0]);
 270       ureg_release_temporary(shader, r[1]);
 271    }
 272
 273    for (i = 0; i < 4; ++i) {
 274       ureg_release_temporary(shader, l[i][0]);
 275       ureg_release_temporary(shader, l[i][1]);
 276    }
 277    ureg_release_temporary(shader, addr[0]);
 278    ureg_release_temporary(shader, addr[1]);
 279
 280    ureg_END(shader);
 281
 282    return ureg_create_shader_and_destroy(shader, idct->pipe);
 283 }
 284
 285 static void *
 286 create_transpose_frag_shader(struct vl_idct *idct)
 287 {
 288    struct ureg_program *shader;
 289
 290    struct ureg_src l_addr[2], r_addr[2];
 291
 292    struct ureg_dst l[2], r[2];
 293    struct ureg_dst tmp, fragment;
 294
 295    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 296    if (!shader)
 297       return NULL;
 298
 299    l_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR0, TGSI_INTERPOLATE_LINEAR);
 300    l_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR1, TGSI_INTERPOLATE_LINEAR);
 301
 302    r_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR0, TGSI_INTERPOLATE_LINEAR);
 303    r_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR1, TGSI_INTERPOLATE_LINEAR);
 304
 305    fetch_four(shader, l, l_addr, ureg_DECL_sampler(shader, 0));
 306    fetch_four(shader, r, r_addr, ureg_DECL_sampler(shader, 1));
 307
 308    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 309
 310    tmp = ureg_DECL_temporary(shader);
 311    matrix_mul(shader, ureg_writemask(tmp, TGSI_WRITEMASK_X), l, r);
 312    ureg_MUL(shader, fragment, ureg_src(tmp), ureg_imm1f(shader, STAGE2_SCALE));
 313
 314    ureg_release_temporary(shader, tmp);
 315    ureg_release_temporary(shader, l[0]);
 316    ureg_release_temporary(shader, l[1]);
 317    ureg_release_temporary(shader, r[0]);
 318    ureg_release_temporary(shader, r[1]);
 319
 320    ureg_END(shader);
 321
 322    return ureg_create_shader_and_destroy(shader, idct->pipe);
 323 }
 324
 325 static bool
 326 init_shaders(struct vl_idct *idct)
 327 {
 328    idct->matrix_vs = create_vert_shader(idct, true);
 329    idct->matrix_fs = create_matrix_frag_shader(idct);
 330
 331    idct->transpose_vs = create_vert_shader(idct, false);
 332    idct->transpose_fs = create_transpose_frag_shader(idct);
 333
 334    return
 335       idct->matrix_vs != NULL &&
 336       idct->matrix_fs != NULL &&
 337       idct->transpose_vs != NULL &&
 338       idct->transpose_fs != NULL;
 339 }
 340
 341 static void
 342 cleanup_shaders(struct vl_idct *idct)
 343 {
 344    idct->pipe->delete_vs_state(idct->pipe, idct->matrix_vs);
 345    idct->pipe->delete_fs_state(idct->pipe, idct->matrix_fs);
 346    idct->pipe->delete_vs_state(idct->pipe, idct->transpose_vs);
 347    idct->pipe->delete_fs_state(idct->pipe, idct->transpose_fs);
 348 }
 349
 350 static bool
 351 init_state(struct vl_idct *idct)
 352 {
 353    struct pipe_vertex_element vertex_elems[NUM_VS_INPUTS];
 354    struct pipe_sampler_state sampler;
 355    struct pipe_rasterizer_state rs_state;
 356    unsigned i;
 357
 358    assert(idct);
 359
 360    idct->quad = vl_vb_upload_quads(idct->pipe, idct->max_blocks);
 361
 362    if(idct->quad.buffer == NULL)
 363       return false;
 364
 365    for (i = 0; i < 4; ++i) {
 366       memset(&sampler, 0, sizeof(sampler));
 367       sampler.wrap_s = PIPE_TEX_WRAP_REPEAT;
 368       sampler.wrap_t = PIPE_TEX_WRAP_REPEAT;
 369       sampler.wrap_r = PIPE_TEX_WRAP_REPEAT;
 370       sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
 371       sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
 372       sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
 373       sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
 374       sampler.compare_func = PIPE_FUNC_ALWAYS;
 375       sampler.normalized_coords = 1;
 376       /*sampler.shadow_ambient = ; */
 377       /*sampler.lod_bias = ; */
 378       sampler.min_lod = 0;
 379       /*sampler.max_lod = ; */
 380       /*sampler.border_color[0] = ; */
 381       /*sampler.max_anisotropy = ; */
 382       idct->samplers.all[i] = idct->pipe->create_sampler_state(idct->pipe, &sampler);
 383    }
 384
 385    memset(&rs_state, 0, sizeof(rs_state));
 386    /*rs_state.sprite_coord_enable */
 387    rs_state.sprite_coord_mode = PIPE_SPRITE_COORD_UPPER_LEFT;
 388    rs_state.point_quad_rasterization = true;
 389    rs_state.point_size = BLOCK_WIDTH;
 390    rs_state.gl_rasterization_rules = false;
 391    idct->rs_state = idct->pipe->create_rasterizer_state(idct->pipe, &rs_state);
 392
 393    vertex_elems[VS_I_RECT] = vl_vb_get_quad_vertex_element();
 394
 395    /* Pos element */
 396    vertex_elems[VS_I_VPOS].src_format = PIPE_FORMAT_R32G32_FLOAT;
 397
 398    idct->vertex_buffer_stride = vl_vb_element_helper(&vertex_elems[VS_I_VPOS], 1, 1);
 399    idct->vertex_elems_state = idct->pipe->create_vertex_elements_state(idct->pipe, 2, vertex_elems);
 400
 401    return true;
 402 }
 403
 404 static void
 405 cleanup_state(struct vl_idct *idct)
 406 {
 407    unsigned i;
 408
 409    for (i = 0; i < 4; ++i)
 410       idct->pipe->delete_sampler_state(idct->pipe, idct->samplers.all[i]);
 411
 412    idct->pipe->delete_rasterizer_state(idct->pipe, idct->rs_state);
 413    idct->pipe->delete_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 414 }
 415
 416 static bool
 417 init_textures(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 418 {
 419    struct pipe_resource template;
 420    struct pipe_sampler_view sampler_view;
 421    unsigned i;
 422
 423    assert(idct && buffer);
 424
 425    /* create textures */
 426    memset(&template, 0, sizeof(struct pipe_resource));
 427    template.last_level = 0;
 428    template.depth0 = 1;
 429    template.bind = PIPE_BIND_SAMPLER_VIEW;
 430    template.flags = 0;
 431
 432    template.target = PIPE_TEXTURE_2D;
 433    template.format = PIPE_FORMAT_R16G16B16A16_SNORM;
 434    template.width0 = idct->buffer_width / 4;
 435    template.height0 = idct->buffer_height;
 436    template.depth0 = 1;
 437    template.usage = PIPE_USAGE_STREAM;
 438    buffer->textures.individual.source = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 439
 440    template.target = PIPE_TEXTURE_3D;
 441    template.format = PIPE_FORMAT_R16G16B16A16_SNORM;
 442    template.width0 = idct->buffer_width / NR_RENDER_TARGETS;
 443    template.height0 = idct->buffer_height / 4;
 444    template.depth0 = NR_RENDER_TARGETS;
 445    template.usage = PIPE_USAGE_STATIC;
 446    buffer->textures.individual.intermediate = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 447
 448    for (i = 0; i < 4; ++i) {
 449       if(buffer->textures.all[i] == NULL)
 450          return false; /* a texture failed to allocate */
 451
 452       u_sampler_view_default_template(&sampler_view, buffer->textures.all[i], buffer->textures.all[i]->format);
 453       buffer->sampler_views.all[i] = idct->pipe->create_sampler_view(idct->pipe, buffer->textures.all[i], &sampler_view);
 454    }
 455
 456    return true;
 457 }
 458
 459 static void
 460 cleanup_textures(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 461 {
 462    unsigned i;
 463
 464    assert(idct && buffer);
 465
 466    for (i = 0; i < 4; ++i) {
 467       pipe_sampler_view_reference(&buffer->sampler_views.all[i], NULL);
 468       pipe_resource_reference(&buffer->textures.all[i], NULL);
 469    }
 470 }
 471
 472 static bool
 473 init_vertex_buffers(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 474 {
 475    assert(idct && buffer);
 476
 477    buffer->vertex_bufs.individual.quad.stride = idct->quad.stride;
 478    buffer->vertex_bufs.individual.quad.max_index = idct->quad.max_index;
 479    buffer->vertex_bufs.individual.quad.buffer_offset = idct->quad.buffer_offset;
 480    pipe_resource_reference(&buffer->vertex_bufs.individual.quad.buffer, idct->quad.buffer);
 481
 482    buffer->vertex_bufs.individual.pos = vl_vb_init(
 483       &buffer->blocks, idct->pipe, idct->max_blocks, 2,
 484       idct->vertex_buffer_stride);
 485
 486    if(buffer->vertex_bufs.individual.pos.buffer == NULL)
 487       return false;
 488
 489    return true;
 490 }
 491
 492 static void
 493 cleanup_vertex_buffers(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 494 {
 495    assert(idct && buffer);
 496
 497    pipe_resource_reference(&buffer->vertex_bufs.individual.quad.buffer, NULL);
 498    pipe_resource_reference(&buffer->vertex_bufs.individual.pos.buffer, NULL);
 499
 500    vl_vb_cleanup(&buffer->blocks);
 501 }
 502
 503 struct pipe_resource *
 504 vl_idct_upload_matrix(struct pipe_context *pipe)
 505 {
 506    struct pipe_resource template, *matrix;
 507    struct pipe_transfer *buf_transfer;
 508    unsigned i, j, pitch;
 509    float *f;
 510
 511    struct pipe_box rect =
 512    {
 513       0, 0, 0,
 514       BLOCK_WIDTH / 4,
 515       BLOCK_HEIGHT,
 516       1
 517    };
 518
 519    memset(&template, 0, sizeof(struct pipe_resource));
 520    template.target = PIPE_TEXTURE_2D;
 521    template.format = PIPE_FORMAT_R32G32B32A32_FLOAT;
 522    template.last_level = 0;
 523    template.width0 = 2;
 524    template.height0 = 8;
 525    template.depth0 = 1;
 526    template.usage = PIPE_USAGE_IMMUTABLE;
 527    template.bind = PIPE_BIND_SAMPLER_VIEW;
 528    template.flags = 0;
 529
 530    matrix = pipe->screen->resource_create(pipe->screen, &template);
 531
 532    /* matrix */
 533    buf_transfer = pipe->get_transfer
 534    (
 535       pipe, matrix,
 536       0, PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 537       &rect
 538    );
 539    pitch = buf_transfer->stride / sizeof(float);
 540
 541    f = pipe->transfer_map(pipe, buf_transfer);
 542    for(i = 0; i < BLOCK_HEIGHT; ++i)
 543       for(j = 0; j < BLOCK_WIDTH; ++j)
 544          // transpose and scale
 545          f[i * pitch + j] = const_matrix[j][i] * STAGE1_SCALE;
 546
 547    pipe->transfer_unmap(pipe, buf_transfer);
 548    pipe->transfer_destroy(pipe, buf_transfer);
 549
 550    return matrix;
 551 }
 552
 553 bool vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe,
 554                   unsigned buffer_width, unsigned buffer_height,
 555                   struct pipe_resource *matrix)
 556 {
 557    assert(idct && pipe && matrix);
 558
 559    idct->pipe = pipe;
 560    idct->buffer_width = buffer_width;
 561    idct->buffer_height = buffer_height;
 562    pipe_resource_reference(&idct->matrix, matrix);
 563
 564    idct->max_blocks =
 565       align(buffer_width, BLOCK_WIDTH) / BLOCK_WIDTH *
 566       align(buffer_height, BLOCK_HEIGHT) / BLOCK_HEIGHT;
 567
 568    if(!init_shaders(idct))
 569       return false;
 570
 571    if(!init_state(idct)) {
 572       cleanup_shaders(idct);
 573       return false;
 574    }
 575
 576    return true;
 577 }
 578
 579 void
 580 vl_idct_cleanup(struct vl_idct *idct)
 581 {
 582    cleanup_shaders(idct);
 583    cleanup_state(idct);
 584
 585    pipe_resource_reference(&idct->matrix, NULL);
 586 }
 587
 588 bool
 589 vl_idct_init_buffer(struct vl_idct *idct, struct vl_idct_buffer *buffer, struct pipe_resource *dst)
 590 {
 591    struct pipe_surface template;
 592
 593    unsigned i;
 594
 595    assert(buffer);
 596    assert(idct);
 597    assert(dst);
 598
 599    pipe_resource_reference(&buffer->textures.individual.matrix, idct->matrix);
 600    pipe_resource_reference(&buffer->textures.individual.transpose, idct->matrix);
 601    pipe_resource_reference(&buffer->destination, dst);
 602
 603    if (!init_textures(idct, buffer))
 604       return false;
 605
 606    if (!init_vertex_buffers(idct, buffer))
 607       return false;
 608
 609    /* init state */
 610    buffer->viewport[0].scale[0] = buffer->textures.individual.intermediate->width0;
 611    buffer->viewport[0].scale[1] = buffer->textures.individual.intermediate->height0;
 612
 613    buffer->viewport[1].scale[0] = buffer->destination->width0;
 614    buffer->viewport[1].scale[1] = buffer->destination->height0;
 615
 616    buffer->fb_state[0].width = buffer->textures.individual.intermediate->width0;
 617    buffer->fb_state[0].height = buffer->textures.individual.intermediate->height0;
 618
 619    buffer->fb_state[0].nr_cbufs = NR_RENDER_TARGETS;
 620    for(i = 0; i < NR_RENDER_TARGETS; ++i) {
 621       memset(&template, 0, sizeof(template));
 622       template.format = buffer->textures.individual.intermediate->format;
 623       template.u.tex.first_layer = i;
 624       template.u.tex.last_layer = i;
 625       template.usage = PIPE_BIND_RENDER_TARGET;
 626       buffer->fb_state[0].cbufs[i] = idct->pipe->create_surface(
 627          idct->pipe, buffer->textures.individual.intermediate,
 628          &template);
 629    }
 630
 631    buffer->fb_state[1].width = buffer->destination->width0;
 632    buffer->fb_state[1].height = buffer->destination->height0;
 633
 634    buffer->fb_state[1].nr_cbufs = 1;
 635
 636    memset(&template, 0, sizeof(template));
 637    template.format = buffer->destination->format;
 638    template.usage = PIPE_BIND_RENDER_TARGET;
 639    buffer->fb_state[1].cbufs[0] = idct->pipe->create_surface(
 640       idct->pipe, buffer->destination, &template);
 641
 642    for(i = 0; i < 2; ++i) {
 643       buffer->viewport[i].scale[2] = 1;
 644       buffer->viewport[i].scale[3] = 1;
 645       buffer->viewport[i].translate[0] = 0;
 646       buffer->viewport[i].translate[1] = 0;
 647       buffer->viewport[i].translate[2] = 0;
 648       buffer->viewport[i].translate[3] = 0;
 649
 650       buffer->fb_state[i].zsbuf = NULL;
 651    }
 652
 653    return true;
 654 }
 655
 656 void
 657 vl_idct_cleanup_buffer(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 658 {
 659    unsigned i;
 660
 661    assert(buffer);
 662
 663    for(i = 0; i < NR_RENDER_TARGETS; ++i) {
 664       idct->pipe->surface_destroy(idct->pipe, buffer->fb_state[0].cbufs[i]);
 665    }
 666
 667    idct->pipe->surface_destroy(idct->pipe, buffer->fb_state[1].cbufs[0]);
 668
 669    cleanup_textures(idct, buffer);
 670    cleanup_vertex_buffers(idct, buffer);
 671 }
 672
 673 void
 674 vl_idct_map_buffers(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 675 {
 676    assert(idct);
 677
 678    struct pipe_box rect =
 679    {
 680       0, 0, 0,
 681       buffer->textures.individual.source->width0,
 682       buffer->textures.individual.source->height0,
 683       1
 684    };
 685
 686    buffer->tex_transfer = idct->pipe->get_transfer
 687    (
 688       idct->pipe, buffer->textures.individual.source,
 689       0, PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 690       &rect
 691    );
 692
 693    buffer->texels = idct->pipe->transfer_map(idct->pipe, buffer->tex_transfer);
 694
 695    vl_vb_map(&buffer->blocks, idct->pipe);
 696 }
 697
 698 void
 699 vl_idct_add_block(struct vl_idct_buffer *buffer, unsigned x, unsigned y, short *block)
 700 {
 701    struct vertex2f v;
 702    unsigned tex_pitch;
 703    short *texels;
 704
 705    unsigned i;
 706
 707    assert(buffer);
 708
 709    tex_pitch = buffer->tex_transfer->stride / sizeof(short);
 710    texels = buffer->texels + y * tex_pitch * BLOCK_HEIGHT + x * BLOCK_WIDTH;
 711
 712    for (i = 0; i < BLOCK_HEIGHT; ++i)
 713       memcpy(texels + i * tex_pitch, block + i * BLOCK_WIDTH, BLOCK_WIDTH * sizeof(short));
 714
 715    v.x = x;
 716    v.y = y;
 717    vl_vb_add_block(&buffer->blocks, (float*)&v);
 718 }
 719
 720 void
 721 vl_idct_unmap_buffers(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 722 {
 723    assert(idct && buffer);
 724
 725    idct->pipe->transfer_unmap(idct->pipe, buffer->tex_transfer);
 726    idct->pipe->transfer_destroy(idct->pipe, buffer->tex_transfer);
 727    vl_vb_unmap(&buffer->blocks, idct->pipe);
 728 }
 729
 730 void
 731 vl_idct_flush(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 732 {
 733    unsigned num_verts;
 734
 735    assert(idct);
 736
 737    num_verts = vl_vb_restart(&buffer->blocks);
 738
 739    if(num_verts > 0) {
 740
 741       idct->pipe->bind_rasterizer_state(idct->pipe, idct->rs_state);
 742       idct->pipe->set_vertex_buffers(idct->pipe, 2, buffer->vertex_bufs.all);
 743       idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 744
 745       /* first stage */
 746       idct->pipe->set_framebuffer_state(idct->pipe, &buffer->fb_state[0]);
 747       idct->pipe->set_viewport_state(idct->pipe, &buffer->viewport[0]);
 748       idct->pipe->set_fragment_sampler_views(idct->pipe, 2, buffer->sampler_views.stage[0]);
 749       idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[0]);
 750       idct->pipe->bind_vs_state(idct->pipe, idct->matrix_vs);
 751       idct->pipe->bind_fs_state(idct->pipe, idct->matrix_fs);
 752       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, num_verts);
 753
 754       /* second stage */
 755       idct->pipe->set_framebuffer_state(idct->pipe, &buffer->fb_state[1]);
 756       idct->pipe->set_viewport_state(idct->pipe, &buffer->viewport[1]);
 757       idct->pipe->set_fragment_sampler_views(idct->pipe, 2, buffer->sampler_views.stage[1]);
 758       idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[1]);
 759       idct->pipe->bind_vs_state(idct->pipe, idct->transpose_vs);
 760       idct->pipe->bind_fs_state(idct->pipe, idct->transpose_fs);
 761       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, num_verts);
 762    }
 763 }