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 NR_RENDER_TARGETS 4
  46
  47 enum VS_INPUT
  48 {
  49    VS_I_RECT,
  50    VS_I_VPOS,
  51
  52    NUM_VS_INPUTS
  53 };
  54
  55 enum VS_OUTPUT
  56 {
  57    VS_O_VPOS,
  58    VS_O_L_ADDR0,
  59    VS_O_L_ADDR1,
  60    VS_O_R_ADDR0,
  61    VS_O_R_ADDR1
  62 };
  63
  64 static const float const_matrix[8][8] = {
  65    {  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.353553f,  0.3535530f },
  66    {  0.4903930f,  0.4157350f,  0.2777850f,  0.0975451f, -0.0975452f, -0.2777850f, -0.415735f, -0.4903930f },
  67    {  0.4619400f,  0.1913420f, -0.1913420f, -0.4619400f, -0.4619400f, -0.1913420f,  0.191342f,  0.4619400f },
  68    {  0.4157350f, -0.0975452f, -0.4903930f, -0.2777850f,  0.2777850f,  0.4903930f,  0.097545f, -0.4157350f },
  69    {  0.3535530f, -0.3535530f, -0.3535530f,  0.3535540f,  0.3535530f, -0.3535540f, -0.353553f,  0.3535530f },
  70    {  0.2777850f, -0.4903930f,  0.0975452f,  0.4157350f, -0.4157350f, -0.0975451f,  0.490393f, -0.2777850f },
  71    {  0.1913420f, -0.4619400f,  0.4619400f, -0.1913420f, -0.1913410f,  0.4619400f, -0.461940f,  0.1913420f },
  72    {  0.0975451f, -0.2777850f,  0.4157350f, -0.4903930f,  0.4903930f, -0.4157350f,  0.277786f, -0.0975458f }
  73 };
  74
  75 static void
  76 calc_addr(struct ureg_program *shader, struct ureg_dst addr[2],
  77           struct ureg_src tc, struct ureg_src start, bool right_side,
  78           bool transposed, float size)
  79 {
  80    unsigned wm_start = (right_side == transposed) ? TGSI_WRITEMASK_X : TGSI_WRITEMASK_Y;
  81    unsigned sw_start = right_side ? TGSI_SWIZZLE_Y : TGSI_SWIZZLE_X;
  82
  83    unsigned wm_tc = (right_side == transposed) ? TGSI_WRITEMASK_Y : TGSI_WRITEMASK_X;
  84    unsigned sw_tc = right_side ? TGSI_SWIZZLE_X : TGSI_SWIZZLE_Y;
  85
  86    /*
  87     * addr[0..1].(start) = right_side ? start.x : tc.x
  88     * addr[0..1].(tc) = right_side ? tc.y : start.y
  89     * addr[0..1].z = tc.z
  90     * addr[1].(start) += 1.0f / scale
  91     */
  92    ureg_MOV(shader, ureg_writemask(addr[0], wm_start), ureg_scalar(start, sw_start));
  93    ureg_MOV(shader, ureg_writemask(addr[0], wm_tc), ureg_scalar(tc, sw_tc));
  94    ureg_MOV(shader, ureg_writemask(addr[0], TGSI_WRITEMASK_Z), tc);
  95
  96    ureg_ADD(shader, ureg_writemask(addr[1], wm_start), ureg_scalar(start, sw_start), ureg_imm1f(shader, 1.0f / size));
  97    ureg_MOV(shader, ureg_writemask(addr[1], wm_tc), ureg_scalar(tc, sw_tc));
  98    ureg_MOV(shader, ureg_writemask(addr[1], TGSI_WRITEMASK_Z), tc);
  99 }
 100
 101 static void *
 102 create_vert_shader(struct vl_idct *idct, bool matrix_stage)
 103 {
 104    struct ureg_program *shader;
 105    struct ureg_src scale;
 106    struct ureg_src vrect, vpos;
 107    struct ureg_dst t_tex, t_start;
 108    struct ureg_dst o_vpos, o_l_addr[2], o_r_addr[2];
 109
 110    shader = ureg_create(TGSI_PROCESSOR_VERTEX);
 111    if (!shader)
 112       return NULL;
 113
 114    t_tex = ureg_DECL_temporary(shader);
 115    t_start = ureg_DECL_temporary(shader);
 116
 117    vrect = ureg_DECL_vs_input(shader, VS_I_RECT);
 118    vpos = ureg_DECL_vs_input(shader, VS_I_VPOS);
 119
 120    o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
 121
 122    o_l_addr[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR0);
 123    o_l_addr[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR1);
 124
 125    o_r_addr[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR0);
 126    o_r_addr[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR1);
 127
 128    /*
 129     * scale = (BLOCK_WIDTH, BLOCK_HEIGHT) / (dst.width, dst.height)
 130     *
 131     * t_vpos = vpos + vrect
 132     * o_vpos.xy = t_vpos * scale
 133     * o_vpos.zw = vpos
 134     *
 135     * o_l_addr = calc_addr(...)
 136     * o_r_addr = calc_addr(...)
 137     *
 138     */
 139    scale = ureg_imm2f(shader,
 140       (float)BLOCK_WIDTH / idct->buffer_width,
 141       (float)BLOCK_HEIGHT / idct->buffer_height);
 142
 143    ureg_ADD(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_XY), vpos, vrect);
 144    ureg_MUL(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_XY), ureg_src(t_tex), scale);
 145    ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_XY), ureg_src(t_tex));
 146    ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_ZW), vpos);
 147
 148    ureg_MUL(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_Z),
 149       ureg_scalar(vrect, TGSI_SWIZZLE_X),
 150       ureg_imm1f(shader, BLOCK_WIDTH / NR_RENDER_TARGETS));
 151
 152    ureg_MUL(shader, ureg_writemask(t_start, TGSI_WRITEMASK_XY), vpos, scale);
 153
 154    if(matrix_stage) {
 155       calc_addr(shader, o_l_addr, ureg_src(t_tex), ureg_src(t_start), false, false, idct->buffer_width / 4);
 156       calc_addr(shader, o_r_addr, vrect, ureg_imm1f(shader, 0.0f), true, true, BLOCK_WIDTH / 4);
 157    } else {
 158       calc_addr(shader, o_l_addr, vrect, ureg_imm1f(shader, 0.0f), false, false, BLOCK_WIDTH / 4);
 159       calc_addr(shader, o_r_addr, ureg_src(t_tex), ureg_src(t_start), true, false, idct->buffer_height / 4);
 160    }
 161
 162    ureg_release_temporary(shader, t_tex);
 163    ureg_release_temporary(shader, t_start);
 164
 165    ureg_END(shader);
 166
 167    return ureg_create_shader_and_destroy(shader, idct->pipe);
 168 }
 169
 170 static void
 171 increment_addr(struct ureg_program *shader, struct ureg_dst daddr[2],
 172                struct ureg_src saddr[2], bool right_side, bool transposed,
 173                int pos, float size)
 174 {
 175    unsigned wm_start = (right_side == transposed) ? TGSI_WRITEMASK_X : TGSI_WRITEMASK_Y;
 176    unsigned wm_tc = (right_side == transposed) ? TGSI_WRITEMASK_Y : TGSI_WRITEMASK_X;
 177
 178    /*
 179     * daddr[0..1].(start) = saddr[0..1].(start)
 180     * daddr[0..1].(tc) = saddr[0..1].(tc)
 181     */
 182
 183    ureg_MOV(shader, ureg_writemask(daddr[0], wm_start), saddr[0]);
 184    ureg_ADD(shader, ureg_writemask(daddr[0], wm_tc), saddr[0], ureg_imm1f(shader, pos / size));
 185    ureg_MOV(shader, ureg_writemask(daddr[1], wm_start), saddr[1]);
 186    ureg_ADD(shader, ureg_writemask(daddr[1], wm_tc), saddr[1], ureg_imm1f(shader, pos / size));
 187 }
 188
 189 static void
 190 fetch_four(struct ureg_program *shader, struct ureg_dst m[2], struct ureg_src addr[2], struct ureg_src sampler)
 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];
 222
 223    struct ureg_dst 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    l_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR0, TGSI_INTERPOLATE_LINEAR);
 233    l_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR1, TGSI_INTERPOLATE_LINEAR);
 234
 235    r_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR0, TGSI_INTERPOLATE_LINEAR);
 236    r_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR1, TGSI_INTERPOLATE_LINEAR);
 237
 238    for (i = 0; i < NR_RENDER_TARGETS; ++i)
 239        fragment[i] = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, i);
 240
 241    for (i = 0; i < 4; ++i) {
 242       l[i][0] = ureg_DECL_temporary(shader);
 243       l[i][1] = ureg_DECL_temporary(shader);
 244    }
 245
 246    r[0] = ureg_DECL_temporary(shader);
 247    r[1] = ureg_DECL_temporary(shader);
 248
 249    for (i = 1; i < 4; ++i) {
 250       increment_addr(shader, l[i], l_addr, false, false, i, idct->buffer_height);
 251    }
 252
 253    for (i = 0; i < 4; ++i) {
 254       struct ureg_src s_addr[2];
 255       s_addr[0] = i == 0 ? l_addr[0] : ureg_src(l[i][0]);
 256       s_addr[1] = i == 0 ? l_addr[1] : ureg_src(l[i][1]);
 257       fetch_four(shader, l[i], s_addr, ureg_DECL_sampler(shader, 1));
 258    }
 259
 260    for (i = 0; i < NR_RENDER_TARGETS; ++i) {
 261       if(i > 0)
 262          increment_addr(shader, r, r_addr, true, true, i, BLOCK_HEIGHT);
 263
 264       struct ureg_src s_addr[2] = { ureg_src(r[0]), ureg_src(r[1]) };
 265       s_addr[0] = i == 0 ? r_addr[0] : ureg_src(r[0]);
 266       s_addr[1] = i == 0 ? r_addr[1] : ureg_src(r[1]);
 267       fetch_four(shader, r, s_addr, ureg_DECL_sampler(shader, 0));
 268
 269       for (j = 0; j < 4; ++j) {
 270          matrix_mul(shader, ureg_writemask(fragment[i], TGSI_WRITEMASK_X << j), l[j], r);
 271       }
 272    }
 273
 274    for (i = 0; i < 4; ++i) {
 275       ureg_release_temporary(shader, l[i][0]);
 276       ureg_release_temporary(shader, l[i][1]);
 277    }
 278    ureg_release_temporary(shader, r[0]);
 279    ureg_release_temporary(shader, r[1]);
 280
 281    ureg_END(shader);
 282
 283    return ureg_create_shader_and_destroy(shader, idct->pipe);
 284 }
 285
 286 static void *
 287 create_transpose_frag_shader(struct vl_idct *idct)
 288 {
 289    struct ureg_program *shader;
 290
 291    struct ureg_src l_addr[2], r_addr[2];
 292
 293    struct ureg_dst l[2], r[2];
 294    struct ureg_dst fragment;
 295
 296    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 297    if (!shader)
 298       return NULL;
 299
 300    l_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR0, TGSI_INTERPOLATE_LINEAR);
 301    l_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR1, TGSI_INTERPOLATE_LINEAR);
 302
 303    r_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR0, TGSI_INTERPOLATE_LINEAR);
 304    r_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR1, TGSI_INTERPOLATE_LINEAR);
 305
 306    l[0] = ureg_DECL_temporary(shader);
 307    l[1] = ureg_DECL_temporary(shader);
 308    r[0] = ureg_DECL_temporary(shader);
 309    r[1] = ureg_DECL_temporary(shader);
 310
 311    fetch_four(shader, l, l_addr, ureg_DECL_sampler(shader, 0));
 312    fetch_four(shader, r, r_addr, ureg_DECL_sampler(shader, 1));
 313
 314    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 315
 316    matrix_mul(shader, ureg_writemask(fragment, TGSI_WRITEMASK_X), l, r);
 317
 318    ureg_release_temporary(shader, l[0]);
 319    ureg_release_temporary(shader, l[1]);
 320    ureg_release_temporary(shader, r[0]);
 321    ureg_release_temporary(shader, r[1]);
 322
 323    ureg_END(shader);
 324
 325    return ureg_create_shader_and_destroy(shader, idct->pipe);
 326 }
 327
 328 static bool
 329 init_shaders(struct vl_idct *idct)
 330 {
 331    idct->matrix_vs = create_vert_shader(idct, true);
 332    idct->matrix_fs = create_matrix_frag_shader(idct);
 333
 334    idct->transpose_vs = create_vert_shader(idct, false);
 335    idct->transpose_fs = create_transpose_frag_shader(idct);
 336
 337    return
 338       idct->matrix_vs != NULL &&
 339       idct->matrix_fs != NULL &&
 340       idct->transpose_vs != NULL &&
 341       idct->transpose_fs != NULL;
 342 }
 343
 344 static void
 345 cleanup_shaders(struct vl_idct *idct)
 346 {
 347    idct->pipe->delete_vs_state(idct->pipe, idct->matrix_vs);
 348    idct->pipe->delete_fs_state(idct->pipe, idct->matrix_fs);
 349    idct->pipe->delete_vs_state(idct->pipe, idct->transpose_vs);
 350    idct->pipe->delete_fs_state(idct->pipe, idct->transpose_fs);
 351 }
 352
 353 static bool
 354 init_state(struct vl_idct *idct)
 355 {
 356    struct pipe_vertex_element vertex_elems[NUM_VS_INPUTS];
 357    struct pipe_sampler_state sampler;
 358    struct pipe_rasterizer_state rs_state;
 359    unsigned i;
 360
 361    assert(idct);
 362
 363    idct->quad = vl_vb_upload_quads(idct->pipe, idct->max_blocks);
 364
 365    if(idct->quad.buffer == NULL)
 366       return false;
 367
 368    for (i = 0; i < 4; ++i) {
 369       memset(&sampler, 0, sizeof(sampler));
 370       sampler.wrap_s = PIPE_TEX_WRAP_REPEAT;
 371       sampler.wrap_t = PIPE_TEX_WRAP_REPEAT;
 372       sampler.wrap_r = PIPE_TEX_WRAP_REPEAT;
 373       sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
 374       sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
 375       sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
 376       sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
 377       sampler.compare_func = PIPE_FUNC_ALWAYS;
 378       sampler.normalized_coords = 1;
 379       /*sampler.shadow_ambient = ; */
 380       /*sampler.lod_bias = ; */
 381       sampler.min_lod = 0;
 382       /*sampler.max_lod = ; */
 383       /*sampler.border_color[0] = ; */
 384       /*sampler.max_anisotropy = ; */
 385       idct->samplers.all[i] = idct->pipe->create_sampler_state(idct->pipe, &sampler);
 386    }
 387
 388    memset(&rs_state, 0, sizeof(rs_state));
 389    /*rs_state.sprite_coord_enable */
 390    rs_state.sprite_coord_mode = PIPE_SPRITE_COORD_UPPER_LEFT;
 391    rs_state.point_quad_rasterization = true;
 392    rs_state.point_size = BLOCK_WIDTH;
 393    rs_state.gl_rasterization_rules = false;
 394    idct->rs_state = idct->pipe->create_rasterizer_state(idct->pipe, &rs_state);
 395
 396    vertex_elems[VS_I_RECT] = vl_vb_get_quad_vertex_element();
 397
 398    /* Pos element */
 399    vertex_elems[VS_I_VPOS].src_format = PIPE_FORMAT_R32G32_FLOAT;
 400
 401    idct->vertex_buffer_stride = vl_vb_element_helper(&vertex_elems[VS_I_VPOS], 1, 1);
 402    idct->vertex_elems_state = idct->pipe->create_vertex_elements_state(idct->pipe, 2, vertex_elems);
 403
 404    return true;
 405 }
 406
 407 static void
 408 cleanup_state(struct vl_idct *idct)
 409 {
 410    unsigned i;
 411
 412    for (i = 0; i < 4; ++i)
 413       idct->pipe->delete_sampler_state(idct->pipe, idct->samplers.all[i]);
 414
 415    idct->pipe->delete_rasterizer_state(idct->pipe, idct->rs_state);
 416    idct->pipe->delete_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 417 }
 418
 419 static bool
 420 init_textures(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 421 {
 422    struct pipe_resource template;
 423    struct pipe_sampler_view sampler_view;
 424    unsigned i;
 425
 426    assert(idct && buffer);
 427
 428    /* create textures */
 429    memset(&template, 0, sizeof(struct pipe_resource));
 430    template.last_level = 0;
 431    template.depth0 = 1;
 432    template.bind = PIPE_BIND_SAMPLER_VIEW;
 433    template.flags = 0;
 434
 435    template.target = PIPE_TEXTURE_2D;
 436    template.format = PIPE_FORMAT_R16G16B16A16_SNORM;
 437    template.width0 = idct->buffer_width / 4;
 438    template.height0 = idct->buffer_height;
 439    template.depth0 = 1;
 440    template.usage = PIPE_USAGE_STREAM;
 441    buffer->textures.individual.source = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 442
 443    template.target = PIPE_TEXTURE_3D;
 444    template.format = PIPE_FORMAT_R16G16B16A16_SNORM;
 445    template.width0 = idct->buffer_width / NR_RENDER_TARGETS;
 446    template.height0 = idct->buffer_height / 4;
 447    template.depth0 = NR_RENDER_TARGETS;
 448    template.usage = PIPE_USAGE_STATIC;
 449    buffer->textures.individual.intermediate = idct->pipe->screen->resource_create(idct->pipe->screen, &template);
 450
 451    for (i = 0; i < 4; ++i) {
 452       if(buffer->textures.all[i] == NULL)
 453          return false; /* a texture failed to allocate */
 454
 455       u_sampler_view_default_template(&sampler_view, buffer->textures.all[i], buffer->textures.all[i]->format);
 456       buffer->sampler_views.all[i] = idct->pipe->create_sampler_view(idct->pipe, buffer->textures.all[i], &sampler_view);
 457    }
 458
 459    return true;
 460 }
 461
 462 static void
 463 cleanup_textures(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 464 {
 465    unsigned i;
 466
 467    assert(idct && buffer);
 468
 469    for (i = 0; i < 4; ++i) {
 470       pipe_sampler_view_reference(&buffer->sampler_views.all[i], NULL);
 471       pipe_resource_reference(&buffer->textures.all[i], NULL);
 472    }
 473 }
 474
 475 static bool
 476 init_vertex_buffers(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 477 {
 478    assert(idct && buffer);
 479
 480    buffer->vertex_bufs.individual.quad.stride = idct->quad.stride;
 481    buffer->vertex_bufs.individual.quad.max_index = idct->quad.max_index;
 482    buffer->vertex_bufs.individual.quad.buffer_offset = idct->quad.buffer_offset;
 483    pipe_resource_reference(&buffer->vertex_bufs.individual.quad.buffer, idct->quad.buffer);
 484
 485    buffer->vertex_bufs.individual.pos = vl_vb_init(
 486       &buffer->blocks, idct->pipe, idct->max_blocks, 2,
 487       idct->vertex_buffer_stride);
 488
 489    if(buffer->vertex_bufs.individual.pos.buffer == NULL)
 490       return false;
 491
 492    return true;
 493 }
 494
 495 static void
 496 cleanup_vertex_buffers(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 497 {
 498    assert(idct && buffer);
 499
 500    pipe_resource_reference(&buffer->vertex_bufs.individual.quad.buffer, NULL);
 501    pipe_resource_reference(&buffer->vertex_bufs.individual.pos.buffer, NULL);
 502
 503    vl_vb_cleanup(&buffer->blocks);
 504 }
 505
 506 struct pipe_resource *
 507 vl_idct_upload_matrix(struct pipe_context *pipe)
 508 {
 509    struct pipe_resource template, *matrix;
 510    struct pipe_transfer *buf_transfer;
 511    unsigned i, j, pitch;
 512    float *f;
 513
 514    struct pipe_box rect =
 515    {
 516       0, 0, 0,
 517       BLOCK_WIDTH / 4,
 518       BLOCK_HEIGHT,
 519       1
 520    };
 521
 522    memset(&template, 0, sizeof(struct pipe_resource));
 523    template.target = PIPE_TEXTURE_2D;
 524    template.format = PIPE_FORMAT_R32G32B32A32_FLOAT;
 525    template.last_level = 0;
 526    template.width0 = 2;
 527    template.height0 = 8;
 528    template.depth0 = 1;
 529    template.usage = PIPE_USAGE_IMMUTABLE;
 530    template.bind = PIPE_BIND_SAMPLER_VIEW;
 531    template.flags = 0;
 532
 533    matrix = pipe->screen->resource_create(pipe->screen, &template);
 534
 535    /* matrix */
 536    buf_transfer = pipe->get_transfer
 537    (
 538       pipe, matrix,
 539       0, PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 540       &rect
 541    );
 542    pitch = buf_transfer->stride / sizeof(float);
 543
 544    f = pipe->transfer_map(pipe, buf_transfer);
 545    for(i = 0; i < BLOCK_HEIGHT; ++i)
 546       for(j = 0; j < BLOCK_WIDTH; ++j)
 547          // transpose and scale
 548          f[i * pitch + j] = const_matrix[j][i] * sqrtf(SCALE_FACTOR_16_TO_9);
 549
 550    pipe->transfer_unmap(pipe, buf_transfer);
 551    pipe->transfer_destroy(pipe, buf_transfer);
 552
 553    return matrix;
 554 }
 555
 556 bool vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe,
 557                   unsigned buffer_width, unsigned buffer_height,
 558                   struct pipe_resource *matrix)
 559 {
 560    assert(idct && pipe && matrix);
 561
 562    idct->pipe = pipe;
 563    idct->buffer_width = buffer_width;
 564    idct->buffer_height = buffer_height;
 565    pipe_resource_reference(&idct->matrix, matrix);
 566
 567    idct->max_blocks =
 568       align(buffer_width, BLOCK_WIDTH) / BLOCK_WIDTH *
 569       align(buffer_height, BLOCK_HEIGHT) / BLOCK_HEIGHT;
 570
 571    if(!init_shaders(idct))
 572       return false;
 573
 574    if(!init_state(idct)) {
 575       cleanup_shaders(idct);
 576       return false;
 577    }
 578
 579    return true;
 580 }
 581
 582 void
 583 vl_idct_cleanup(struct vl_idct *idct)
 584 {
 585    cleanup_shaders(idct);
 586    cleanup_state(idct);
 587
 588    pipe_resource_reference(&idct->matrix, NULL);
 589 }
 590
 591 bool
 592 vl_idct_init_buffer(struct vl_idct *idct, struct vl_idct_buffer *buffer, struct pipe_resource *dst)
 593 {
 594    struct pipe_surface template;
 595
 596    unsigned i;
 597
 598    assert(buffer);
 599    assert(idct);
 600    assert(dst);
 601
 602    pipe_resource_reference(&buffer->textures.individual.matrix, idct->matrix);
 603    pipe_resource_reference(&buffer->textures.individual.transpose, idct->matrix);
 604    pipe_resource_reference(&buffer->destination, dst);
 605
 606    if (!init_textures(idct, buffer))
 607       return false;
 608
 609    if (!init_vertex_buffers(idct, buffer))
 610       return false;
 611
 612    /* init state */
 613    buffer->viewport[0].scale[0] = buffer->textures.individual.intermediate->width0;
 614    buffer->viewport[0].scale[1] = buffer->textures.individual.intermediate->height0;
 615
 616    buffer->viewport[1].scale[0] = buffer->destination->width0;
 617    buffer->viewport[1].scale[1] = buffer->destination->height0;
 618
 619    buffer->fb_state[0].width = buffer->textures.individual.intermediate->width0;
 620    buffer->fb_state[0].height = buffer->textures.individual.intermediate->height0;
 621
 622    buffer->fb_state[0].nr_cbufs = NR_RENDER_TARGETS;
 623    for(i = 0; i < NR_RENDER_TARGETS; ++i) {
 624       memset(&template, 0, sizeof(template));
 625       template.format = buffer->textures.individual.intermediate->format;
 626       template.u.tex.first_layer = i;
 627       template.u.tex.last_layer = i;
 628       template.usage = PIPE_BIND_RENDER_TARGET;
 629       buffer->fb_state[0].cbufs[i] = idct->pipe->create_surface(
 630          idct->pipe, buffer->textures.individual.intermediate,
 631          &template);
 632    }
 633
 634    buffer->fb_state[1].width = buffer->destination->width0;
 635    buffer->fb_state[1].height = buffer->destination->height0;
 636
 637    buffer->fb_state[1].nr_cbufs = 1;
 638
 639    memset(&template, 0, sizeof(template));
 640    template.format = buffer->destination->format;
 641    template.usage = PIPE_BIND_RENDER_TARGET;
 642    buffer->fb_state[1].cbufs[0] = idct->pipe->create_surface(
 643       idct->pipe, buffer->destination, &template);
 644
 645    for(i = 0; i < 2; ++i) {
 646       buffer->viewport[i].scale[2] = 1;
 647       buffer->viewport[i].scale[3] = 1;
 648       buffer->viewport[i].translate[0] = 0;
 649       buffer->viewport[i].translate[1] = 0;
 650       buffer->viewport[i].translate[2] = 0;
 651       buffer->viewport[i].translate[3] = 0;
 652
 653       buffer->fb_state[i].zsbuf = NULL;
 654    }
 655
 656    return true;
 657 }
 658
 659 void
 660 vl_idct_cleanup_buffer(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 661 {
 662    unsigned i;
 663
 664    assert(buffer);
 665
 666    for(i = 0; i < NR_RENDER_TARGETS; ++i) {
 667       idct->pipe->surface_destroy(idct->pipe, buffer->fb_state[0].cbufs[i]);
 668    }
 669
 670    idct->pipe->surface_destroy(idct->pipe, buffer->fb_state[1].cbufs[0]);
 671
 672    cleanup_textures(idct, buffer);
 673    cleanup_vertex_buffers(idct, buffer);
 674 }
 675
 676 void
 677 vl_idct_map_buffers(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 678 {
 679    assert(idct);
 680
 681    struct pipe_box rect =
 682    {
 683       0, 0, 0,
 684       buffer->textures.individual.source->width0,
 685       buffer->textures.individual.source->height0,
 686       1
 687    };
 688
 689    buffer->tex_transfer = idct->pipe->get_transfer
 690    (
 691       idct->pipe, buffer->textures.individual.source,
 692       0, PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 693       &rect
 694    );
 695
 696    buffer->texels = idct->pipe->transfer_map(idct->pipe, buffer->tex_transfer);
 697
 698    vl_vb_map(&buffer->blocks, idct->pipe);
 699 }
 700
 701 void
 702 vl_idct_add_block(struct vl_idct_buffer *buffer, unsigned x, unsigned y, short *block)
 703 {
 704    struct vertex2f v;
 705    unsigned tex_pitch;
 706    short *texels;
 707
 708    unsigned i;
 709
 710    assert(buffer);
 711
 712    tex_pitch = buffer->tex_transfer->stride / sizeof(short);
 713    texels = buffer->texels + y * tex_pitch * BLOCK_HEIGHT + x * BLOCK_WIDTH;
 714
 715    for (i = 0; i < BLOCK_HEIGHT; ++i)
 716       memcpy(texels + i * tex_pitch, block + i * BLOCK_WIDTH, BLOCK_WIDTH * sizeof(short));
 717
 718    v.x = x;
 719    v.y = y;
 720    vl_vb_add_block(&buffer->blocks, (float*)&v);
 721 }
 722
 723 void
 724 vl_idct_unmap_buffers(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 725 {
 726    assert(idct && buffer);
 727
 728    idct->pipe->transfer_unmap(idct->pipe, buffer->tex_transfer);
 729    idct->pipe->transfer_destroy(idct->pipe, buffer->tex_transfer);
 730    vl_vb_unmap(&buffer->blocks, idct->pipe);
 731 }
 732
 733 void
 734 vl_idct_flush(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 735 {
 736    unsigned num_verts;
 737
 738    assert(idct);
 739
 740    num_verts = vl_vb_restart(&buffer->blocks);
 741
 742    if(num_verts > 0) {
 743
 744       idct->pipe->bind_rasterizer_state(idct->pipe, idct->rs_state);
 745       idct->pipe->set_vertex_buffers(idct->pipe, 2, buffer->vertex_bufs.all);
 746       idct->pipe->bind_vertex_elements_state(idct->pipe, idct->vertex_elems_state);
 747
 748       /* first stage */
 749       idct->pipe->set_framebuffer_state(idct->pipe, &buffer->fb_state[0]);
 750       idct->pipe->set_viewport_state(idct->pipe, &buffer->viewport[0]);
 751       idct->pipe->set_fragment_sampler_views(idct->pipe, 2, buffer->sampler_views.stage[0]);
 752       idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[0]);
 753       idct->pipe->bind_vs_state(idct->pipe, idct->matrix_vs);
 754       idct->pipe->bind_fs_state(idct->pipe, idct->matrix_fs);
 755       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, num_verts);
 756
 757       /* second stage */
 758       idct->pipe->set_framebuffer_state(idct->pipe, &buffer->fb_state[1]);
 759       idct->pipe->set_viewport_state(idct->pipe, &buffer->viewport[1]);
 760       idct->pipe->set_fragment_sampler_views(idct->pipe, 2, buffer->sampler_views.stage[1]);
 761       idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers.stage[1]);
 762       idct->pipe->bind_vs_state(idct->pipe, idct->transpose_vs);
 763       idct->pipe->bind_fs_state(idct->pipe, idct->transpose_fs);
 764       util_draw_arrays(idct->pipe, PIPE_PRIM_QUADS, 0, num_verts);
 765    }
 766 }