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 <assert.h>
  29
  30 #include <pipe/p_context.h>
  31 #include <pipe/p_screen.h>
  32
  33 #include <util/u_draw.h>
  34 #include <util/u_sampler.h>
  35
  36 #include <tgsi/tgsi_ureg.h>
  37
  38 #include "vl_defines.h"
  39 #include "vl_types.h"
  40 #include "vl_vertex_buffers.h"
  41 #include "vl_idct.h"
  42
  43 enum VS_OUTPUT
  44 {
  45    VS_O_VPOS,
  46    VS_O_L_ADDR0,
  47    VS_O_L_ADDR1,
  48    VS_O_R_ADDR0,
  49    VS_O_R_ADDR1
  50 };
  51
  52 static const float const_matrix[8][8] = {
  53    {  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.353553f,  0.3535530f },
  54    {  0.4903930f,  0.4157350f,  0.2777850f,  0.0975451f, -0.0975452f, -0.2777850f, -0.415735f, -0.4903930f },
  55    {  0.4619400f,  0.1913420f, -0.1913420f, -0.4619400f, -0.4619400f, -0.1913420f,  0.191342f,  0.4619400f },
  56    {  0.4157350f, -0.0975452f, -0.4903930f, -0.2777850f,  0.2777850f,  0.4903930f,  0.097545f, -0.4157350f },
  57    {  0.3535530f, -0.3535530f, -0.3535530f,  0.3535540f,  0.3535530f, -0.3535540f, -0.353553f,  0.3535530f },
  58    {  0.2777850f, -0.4903930f,  0.0975452f,  0.4157350f, -0.4157350f, -0.0975451f,  0.490393f, -0.2777850f },
  59    {  0.1913420f, -0.4619400f,  0.4619400f, -0.1913420f, -0.1913410f,  0.4619400f, -0.461940f,  0.1913420f },
  60    {  0.0975451f, -0.2777850f,  0.4157350f, -0.4903930f,  0.4903930f, -0.4157350f,  0.277786f, -0.0975458f }
  61 };
  62
  63 static void
  64 calc_addr(struct ureg_program *shader, struct ureg_dst addr[2],
  65           struct ureg_src tc, struct ureg_src start, bool right_side,
  66           bool transposed, float size)
  67 {
  68    unsigned wm_start = (right_side == transposed) ? TGSI_WRITEMASK_X : TGSI_WRITEMASK_Y;
  69    unsigned sw_start = right_side ? TGSI_SWIZZLE_Y : TGSI_SWIZZLE_X;
  70
  71    unsigned wm_tc = (right_side == transposed) ? TGSI_WRITEMASK_Y : TGSI_WRITEMASK_X;
  72    unsigned sw_tc = right_side ? TGSI_SWIZZLE_X : TGSI_SWIZZLE_Y;
  73
  74    /*
  75     * addr[0..1].(start) = right_side ? start.x : tc.x
  76     * addr[0..1].(tc) = right_side ? tc.y : start.y
  77     * addr[0..1].z = tc.z
  78     * addr[1].(start) += 1.0f / scale
  79     */
  80    ureg_MOV(shader, ureg_writemask(addr[0], wm_start), ureg_scalar(start, sw_start));
  81    ureg_MOV(shader, ureg_writemask(addr[0], wm_tc), ureg_scalar(tc, sw_tc));
  82    ureg_MOV(shader, ureg_writemask(addr[0], TGSI_WRITEMASK_Z), tc);
  83
  84    ureg_ADD(shader, ureg_writemask(addr[1], wm_start), ureg_scalar(start, sw_start), ureg_imm1f(shader, 1.0f / size));
  85    ureg_MOV(shader, ureg_writemask(addr[1], wm_tc), ureg_scalar(tc, sw_tc));
  86    ureg_MOV(shader, ureg_writemask(addr[1], TGSI_WRITEMASK_Z), tc);
  87 }
  88
  89 static void *
  90 create_vert_shader(struct vl_idct *idct, bool matrix_stage)
  91 {
  92    struct ureg_program *shader;
  93    struct ureg_src vrect, vpos;
  94    struct ureg_src scale;
  95    struct ureg_dst t_tex, t_start;
  96    struct ureg_dst o_vpos, o_l_addr[2], o_r_addr[2];
  97
  98    shader = ureg_create(TGSI_PROCESSOR_VERTEX);
  99    if (!shader)
 100       return NULL;
 101
 102    t_tex = ureg_DECL_temporary(shader);
 103    t_start = ureg_DECL_temporary(shader);
 104
 105    vrect = ureg_DECL_vs_input(shader, VS_I_RECT);
 106    vpos = ureg_DECL_vs_input(shader, VS_I_VPOS);
 107
 108    o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
 109
 110    o_l_addr[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR0);
 111    o_l_addr[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR1);
 112
 113    o_r_addr[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR0);
 114    o_r_addr[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR1);
 115
 116    /*
 117     * scale = (BLOCK_WIDTH, BLOCK_HEIGHT) / (dst.width, dst.height)
 118     *
 119     * t_vpos = vpos + vrect
 120     * o_vpos.xy = t_vpos * scale
 121     * o_vpos.zw = vpos
 122     *
 123     * o_l_addr = calc_addr(...)
 124     * o_r_addr = calc_addr(...)
 125     *
 126     */
 127
 128    scale = ureg_imm2f(shader,
 129       (float)BLOCK_WIDTH / idct->buffer_width,
 130       (float)BLOCK_HEIGHT / idct->buffer_height);
 131
 132    ureg_ADD(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_XY), vpos, vrect);
 133    ureg_MUL(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_XY), ureg_src(t_tex), scale);
 134    ureg_MUL(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_Z),
 135       ureg_scalar(vrect, TGSI_SWIZZLE_X),
 136       ureg_imm1f(shader, BLOCK_WIDTH / idct->nr_of_render_targets));
 137
 138    ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_XY), ureg_src(t_tex));
 139    ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_ZW), ureg_imm1f(shader, 1.0f));
 140
 141    ureg_MUL(shader, ureg_writemask(t_start, TGSI_WRITEMASK_XY), vpos, scale);
 142
 143    if(matrix_stage) {
 144       calc_addr(shader, o_l_addr, ureg_src(t_tex), ureg_src(t_start), false, false, idct->buffer_width / 4);
 145       calc_addr(shader, o_r_addr, vrect, ureg_imm1f(shader, 0.0f), true, true, BLOCK_WIDTH / 4);
 146    } else {
 147       calc_addr(shader, o_l_addr, vrect, ureg_imm1f(shader, 0.0f), false, false, BLOCK_WIDTH / 4);
 148       calc_addr(shader, o_r_addr, ureg_src(t_tex), ureg_src(t_start), true, false, idct->buffer_height / 4);
 149    }
 150
 151    ureg_release_temporary(shader, t_tex);
 152    ureg_release_temporary(shader, t_start);
 153
 154    ureg_END(shader);
 155
 156    return ureg_create_shader_and_destroy(shader, idct->pipe);
 157 }
 158
 159 static void
 160 increment_addr(struct ureg_program *shader, struct ureg_dst daddr[2],
 161                struct ureg_src saddr[2], bool right_side, bool transposed,
 162                int pos, float size)
 163 {
 164    unsigned wm_start = (right_side == transposed) ? TGSI_WRITEMASK_X : TGSI_WRITEMASK_Y;
 165    unsigned wm_tc = (right_side == transposed) ? TGSI_WRITEMASK_Y : TGSI_WRITEMASK_X;
 166
 167    /*
 168     * daddr[0..1].(start) = saddr[0..1].(start)
 169     * daddr[0..1].(tc) = saddr[0..1].(tc)
 170     */
 171
 172    ureg_MOV(shader, ureg_writemask(daddr[0], wm_start), saddr[0]);
 173    ureg_ADD(shader, ureg_writemask(daddr[0], wm_tc), saddr[0], ureg_imm1f(shader, pos / size));
 174    ureg_MOV(shader, ureg_writemask(daddr[1], wm_start), saddr[1]);
 175    ureg_ADD(shader, ureg_writemask(daddr[1], wm_tc), saddr[1], ureg_imm1f(shader, pos / size));
 176 }
 177
 178 static void
 179 fetch_four(struct ureg_program *shader, struct ureg_dst m[2], struct ureg_src addr[2], struct ureg_src sampler)
 180 {
 181    ureg_TEX(shader, m[0], TGSI_TEXTURE_3D, addr[0], sampler);
 182    ureg_TEX(shader, m[1], TGSI_TEXTURE_3D, addr[1], sampler);
 183 }
 184
 185 static void
 186 matrix_mul(struct ureg_program *shader, struct ureg_dst dst, struct ureg_dst l[2], struct ureg_dst r[2])
 187 {
 188    struct ureg_dst tmp;
 189
 190    tmp = ureg_DECL_temporary(shader);
 191
 192    /*
 193     * tmp.xy = dot4(m[0][0..1], m[1][0..1])
 194     * dst = tmp.x + tmp.y
 195     */
 196    ureg_DP4(shader, ureg_writemask(tmp, TGSI_WRITEMASK_X), ureg_src(l[0]), ureg_src(r[0]));
 197    ureg_DP4(shader, ureg_writemask(tmp, TGSI_WRITEMASK_Y), ureg_src(l[1]), ureg_src(r[1]));
 198    ureg_ADD(shader, dst,
 199       ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_X),
 200       ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_Y));
 201
 202    ureg_release_temporary(shader, tmp);
 203 }
 204
 205 static void *
 206 create_matrix_frag_shader(struct vl_idct *idct)
 207 {
 208    struct ureg_program *shader;
 209
 210    struct ureg_src l_addr[2], r_addr[2];
 211
 212    struct ureg_dst l[4][2], r[2];
 213    struct ureg_dst fragment[idct->nr_of_render_targets];
 214
 215    unsigned i, j;
 216
 217    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 218    if (!shader)
 219       return NULL;
 220
 221    l_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR0, TGSI_INTERPOLATE_LINEAR);
 222    l_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR1, TGSI_INTERPOLATE_LINEAR);
 223
 224    r_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR0, TGSI_INTERPOLATE_LINEAR);
 225    r_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR1, TGSI_INTERPOLATE_LINEAR);
 226
 227    for (i = 0; i < idct->nr_of_render_targets; ++i)
 228        fragment[i] = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, i);
 229
 230    for (i = 0; i < 4; ++i) {
 231       l[i][0] = ureg_DECL_temporary(shader);
 232       l[i][1] = ureg_DECL_temporary(shader);
 233    }
 234
 235    r[0] = ureg_DECL_temporary(shader);
 236    r[1] = ureg_DECL_temporary(shader);
 237
 238    for (i = 1; i < 4; ++i) {
 239       increment_addr(shader, l[i], l_addr, false, false, i, idct->buffer_height);
 240    }
 241
 242    for (i = 0; i < 4; ++i) {
 243       struct ureg_src s_addr[2];
 244       s_addr[0] = i == 0 ? l_addr[0] : ureg_src(l[i][0]);
 245       s_addr[1] = i == 0 ? l_addr[1] : ureg_src(l[i][1]);
 246       fetch_four(shader, l[i], s_addr, ureg_DECL_sampler(shader, 1));
 247    }
 248
 249    for (i = 0; i < idct->nr_of_render_targets; ++i) {
 250       if(i > 0)
 251          increment_addr(shader, r, r_addr, true, true, i, BLOCK_HEIGHT);
 252
 253       struct ureg_src s_addr[2] = { ureg_src(r[0]), ureg_src(r[1]) };
 254       s_addr[0] = i == 0 ? r_addr[0] : ureg_src(r[0]);
 255       s_addr[1] = i == 0 ? r_addr[1] : ureg_src(r[1]);
 256       fetch_four(shader, r, s_addr, ureg_DECL_sampler(shader, 0));
 257
 258       for (j = 0; j < 4; ++j) {
 259          matrix_mul(shader, ureg_writemask(fragment[i], TGSI_WRITEMASK_X << j), l[j], r);
 260       }
 261    }
 262
 263    for (i = 0; i < 4; ++i) {
 264       ureg_release_temporary(shader, l[i][0]);
 265       ureg_release_temporary(shader, l[i][1]);
 266    }
 267    ureg_release_temporary(shader, r[0]);
 268    ureg_release_temporary(shader, r[1]);
 269
 270    ureg_END(shader);
 271
 272    return ureg_create_shader_and_destroy(shader, idct->pipe);
 273 }
 274
 275 static void *
 276 create_transpose_frag_shader(struct vl_idct *idct)
 277 {
 278    struct ureg_program *shader;
 279
 280    struct ureg_src l_addr[2], r_addr[2];
 281
 282    struct ureg_dst l[2], r[2];
 283    struct ureg_dst fragment;
 284
 285    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 286    if (!shader)
 287       return NULL;
 288
 289    l_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR0, TGSI_INTERPOLATE_LINEAR);
 290    l_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR1, TGSI_INTERPOLATE_LINEAR);
 291
 292    r_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR0, TGSI_INTERPOLATE_LINEAR);
 293    r_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR1, TGSI_INTERPOLATE_LINEAR);
 294
 295    l[0] = ureg_DECL_temporary(shader);
 296    l[1] = ureg_DECL_temporary(shader);
 297    r[0] = ureg_DECL_temporary(shader);
 298    r[1] = ureg_DECL_temporary(shader);
 299
 300    fetch_four(shader, l, l_addr, ureg_DECL_sampler(shader, 0));
 301    fetch_four(shader, r, r_addr, ureg_DECL_sampler(shader, 1));
 302
 303    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 304
 305    matrix_mul(shader, ureg_writemask(fragment, TGSI_WRITEMASK_X), l, r);
 306
 307    ureg_release_temporary(shader, l[0]);
 308    ureg_release_temporary(shader, l[1]);
 309    ureg_release_temporary(shader, r[0]);
 310    ureg_release_temporary(shader, r[1]);
 311
 312    ureg_END(shader);
 313
 314    return ureg_create_shader_and_destroy(shader, idct->pipe);
 315 }
 316
 317 static bool
 318 init_shaders(struct vl_idct *idct)
 319 {
 320    idct->matrix_vs = create_vert_shader(idct, true);
 321    if (!idct->matrix_vs)
 322       goto error_matrix_vs;
 323
 324    idct->matrix_fs = create_matrix_frag_shader(idct);
 325    if (!idct->matrix_fs)
 326       goto error_matrix_fs;
 327
 328    idct->transpose_vs = create_vert_shader(idct, false);
 329    if (!idct->transpose_vs)
 330       goto error_transpose_vs;
 331
 332    idct->transpose_fs = create_transpose_frag_shader(idct);
 333    if (!idct->transpose_fs)
 334       goto error_transpose_fs;
 335
 336    return true;
 337
 338 error_transpose_fs:
 339    idct->pipe->delete_vs_state(idct->pipe, idct->transpose_vs);
 340
 341 error_transpose_vs:
 342    idct->pipe->delete_fs_state(idct->pipe, idct->matrix_fs);
 343
 344 error_matrix_fs:
 345    idct->pipe->delete_vs_state(idct->pipe, idct->matrix_vs);
 346
 347 error_matrix_vs:
 348    return false;
 349 }
 350
 351 static void
 352 cleanup_shaders(struct vl_idct *idct)
 353 {
 354    idct->pipe->delete_vs_state(idct->pipe, idct->matrix_vs);
 355    idct->pipe->delete_fs_state(idct->pipe, idct->matrix_fs);
 356    idct->pipe->delete_vs_state(idct->pipe, idct->transpose_vs);
 357    idct->pipe->delete_fs_state(idct->pipe, idct->transpose_fs);
 358 }
 359
 360 static bool
 361 init_state(struct vl_idct *idct)
 362 {
 363    struct pipe_blend_state blend;
 364    struct pipe_rasterizer_state rs_state;
 365    struct pipe_sampler_state sampler;
 366    unsigned i;
 367
 368    assert(idct);
 369
 370    memset(&rs_state, 0, sizeof(rs_state));
 371    rs_state.gl_rasterization_rules = false;
 372    idct->rs_state = idct->pipe->create_rasterizer_state(idct->pipe, &rs_state);
 373    if (!idct->rs_state)
 374       goto error_rs_state;
 375
 376    memset(&blend, 0, sizeof blend);
 377
 378    blend.independent_blend_enable = 0;
 379    blend.rt[0].blend_enable = 0;
 380    blend.rt[0].rgb_func = PIPE_BLEND_ADD;
 381    blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE;
 382    blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ONE;
 383    blend.rt[0].alpha_func = PIPE_BLEND_ADD;
 384    blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
 385    blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ONE;
 386    blend.logicop_enable = 0;
 387    blend.logicop_func = PIPE_LOGICOP_CLEAR;
 388    /* Needed to allow color writes to FB, even if blending disabled */
 389    blend.rt[0].colormask = PIPE_MASK_RGBA;
 390    blend.dither = 0;
 391    idct->blend = idct->pipe->create_blend_state(idct->pipe, &blend);
 392    if (!idct->blend)
 393       goto error_blend;
 394
 395    for (i = 0; i < 2; ++i) {
 396       memset(&sampler, 0, sizeof(sampler));
 397       sampler.wrap_s = PIPE_TEX_WRAP_REPEAT;
 398       sampler.wrap_t = PIPE_TEX_WRAP_REPEAT;
 399       sampler.wrap_r = PIPE_TEX_WRAP_REPEAT;
 400       sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
 401       sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
 402       sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
 403       sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
 404       sampler.compare_func = PIPE_FUNC_ALWAYS;
 405       sampler.normalized_coords = 1;
 406       idct->samplers[i] = idct->pipe->create_sampler_state(idct->pipe, &sampler);
 407       if (!idct->samplers[i])
 408          goto error_samplers;
 409    }
 410
 411    return true;
 412
 413 error_samplers:
 414    for (i = 0; i < 2; ++i)
 415       if (idct->samplers[i])
 416          idct->pipe->delete_sampler_state(idct->pipe, idct->samplers[i]);
 417
 418    idct->pipe->delete_rasterizer_state(idct->pipe, idct->rs_state);
 419
 420 error_blend:
 421    idct->pipe->delete_blend_state(idct->pipe, idct->blend);
 422
 423 error_rs_state:
 424    return false;
 425 }
 426
 427 static void
 428 cleanup_state(struct vl_idct *idct)
 429 {
 430    unsigned i;
 431
 432    for (i = 0; i < 2; ++i)
 433       idct->pipe->delete_sampler_state(idct->pipe, idct->samplers[i]);
 434
 435    idct->pipe->delete_rasterizer_state(idct->pipe, idct->rs_state);
 436    idct->pipe->delete_blend_state(idct->pipe, idct->blend);
 437 }
 438
 439 static bool
 440 init_intermediate(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 441 {
 442    struct pipe_resource *tex;
 443    struct pipe_surface surf_templ;
 444    unsigned i;
 445
 446    assert(idct && buffer);
 447
 448    tex = buffer->sampler_views.individual.intermediate->texture;
 449
 450    buffer->fb_state[0].width = tex->width0;
 451    buffer->fb_state[0].height = tex->height0;
 452    buffer->fb_state[0].nr_cbufs = idct->nr_of_render_targets;
 453    for(i = 0; i < idct->nr_of_render_targets; ++i) {
 454       memset(&surf_templ, 0, sizeof(surf_templ));
 455       surf_templ.format = tex->format;
 456       surf_templ.u.tex.first_layer = i;
 457       surf_templ.u.tex.last_layer = i;
 458       surf_templ.usage = PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET;
 459       buffer->fb_state[0].cbufs[i] = idct->pipe->create_surface(
 460          idct->pipe, tex, &surf_templ);
 461
 462       if (!buffer->fb_state[0].cbufs[i])
 463          goto error_surfaces;
 464    }
 465
 466    buffer->viewport[0].scale[0] = tex->width0;
 467    buffer->viewport[0].scale[1] = tex->height0;
 468
 469    return true;
 470
 471 error_surfaces:
 472    for(i = 0; i < idct->nr_of_render_targets; ++i)
 473       pipe_surface_reference(&buffer->fb_state[0].cbufs[i], NULL);
 474
 475    return false;
 476 }
 477
 478 static void
 479 cleanup_intermediate(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 480 {
 481    unsigned i;
 482
 483    assert(idct && buffer);
 484
 485    for(i = 0; i < idct->nr_of_render_targets; ++i)
 486       pipe_surface_reference(&buffer->fb_state[0].cbufs[i], NULL);
 487
 488    pipe_sampler_view_reference(&buffer->sampler_views.individual.intermediate, NULL);
 489 }
 490
 491 struct pipe_sampler_view *
 492 vl_idct_upload_matrix(struct pipe_context *pipe, float scale)
 493 {
 494    struct pipe_resource tex_templ, *matrix;
 495    struct pipe_sampler_view sv_templ, *sv;
 496    struct pipe_transfer *buf_transfer;
 497    unsigned i, j, pitch;
 498    float *f;
 499
 500    struct pipe_box rect =
 501    {
 502       0, 0, 0,
 503       BLOCK_WIDTH / 4,
 504       BLOCK_HEIGHT,
 505       1
 506    };
 507
 508    assert(pipe);
 509
 510    memset(&tex_templ, 0, sizeof(tex_templ));
 511    tex_templ.target = PIPE_TEXTURE_2D;
 512    tex_templ.format = PIPE_FORMAT_R32G32B32A32_FLOAT;
 513    tex_templ.last_level = 0;
 514    tex_templ.width0 = 2;
 515    tex_templ.height0 = 8;
 516    tex_templ.depth0 = 1;
 517    tex_templ.array_size = 1;
 518    tex_templ.usage = PIPE_USAGE_IMMUTABLE;
 519    tex_templ.bind = PIPE_BIND_SAMPLER_VIEW;
 520    tex_templ.flags = 0;
 521
 522    matrix = pipe->screen->resource_create(pipe->screen, &tex_templ);
 523    if (!matrix)
 524       goto error_matrix;
 525
 526    buf_transfer = pipe->get_transfer
 527    (
 528       pipe, matrix,
 529       0, PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 530       &rect
 531    );
 532    if (!buf_transfer)
 533       goto error_transfer;
 534
 535    pitch = buf_transfer->stride / sizeof(float);
 536
 537    f = pipe->transfer_map(pipe, buf_transfer);
 538    if (!f)
 539       goto error_map;
 540
 541    for(i = 0; i < BLOCK_HEIGHT; ++i)
 542       for(j = 0; j < BLOCK_WIDTH; ++j)
 543          // transpose and scale
 544          f[i * pitch + j] = const_matrix[j][i] * scale;
 545
 546    pipe->transfer_unmap(pipe, buf_transfer);
 547    pipe->transfer_destroy(pipe, buf_transfer);
 548
 549    memset(&sv_templ, 0, sizeof(sv_templ));
 550    u_sampler_view_default_template(&sv_templ, matrix, matrix->format);
 551    sv = pipe->create_sampler_view(pipe, matrix, &sv_templ);
 552    pipe_resource_reference(&matrix, NULL);
 553    if (!sv)
 554       goto error_map;
 555
 556    return sv;
 557
 558 error_map:
 559    pipe->transfer_destroy(pipe, buf_transfer);
 560
 561 error_transfer:
 562    pipe_resource_reference(&matrix, NULL);
 563
 564 error_matrix:
 565    return NULL;
 566 }
 567
 568 bool vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe,
 569                   unsigned buffer_width, unsigned buffer_height,
 570                   unsigned nr_of_render_targets,
 571                   struct pipe_sampler_view *matrix,
 572                   struct pipe_sampler_view *transpose)
 573 {
 574    assert(idct && pipe && matrix);
 575
 576    idct->pipe = pipe;
 577    idct->buffer_width = buffer_width;
 578    idct->buffer_height = buffer_height;
 579    idct->nr_of_render_targets = nr_of_render_targets;
 580
 581    pipe_sampler_view_reference(&idct->matrix, matrix);
 582    pipe_sampler_view_reference(&idct->transpose, transpose);
 583
 584    if(!init_shaders(idct))
 585       return false;
 586
 587    if(!init_state(idct)) {
 588       cleanup_shaders(idct);
 589       return false;
 590    }
 591
 592    return true;
 593 }
 594
 595 void
 596 vl_idct_cleanup(struct vl_idct *idct)
 597 {
 598    cleanup_shaders(idct);
 599    cleanup_state(idct);
 600
 601    pipe_sampler_view_reference(&idct->matrix, NULL);
 602 }
 603
 604 bool
 605 vl_idct_init_buffer(struct vl_idct *idct, struct vl_idct_buffer *buffer,
 606                     struct pipe_sampler_view *source,
 607                     struct pipe_sampler_view *intermediate,
 608                     struct pipe_surface *destination)
 609 {
 610    unsigned i;
 611
 612    assert(buffer);
 613    assert(idct);
 614    assert(source);
 615    assert(destination);
 616
 617    pipe_sampler_view_reference(&buffer->sampler_views.individual.matrix, idct->matrix);
 618    pipe_sampler_view_reference(&buffer->sampler_views.individual.source, source);
 619    pipe_sampler_view_reference(&buffer->sampler_views.individual.transpose, idct->transpose);
 620    pipe_sampler_view_reference(&buffer->sampler_views.individual.intermediate, intermediate);
 621
 622    if (!init_intermediate(idct, buffer))
 623       return false;
 624
 625    /* init state */
 626    buffer->fb_state[1].width = destination->texture->width0;
 627    buffer->fb_state[1].height = destination->texture->height0;
 628    buffer->fb_state[1].nr_cbufs = 1;
 629    pipe_surface_reference(&buffer->fb_state[1].cbufs[0], destination);
 630
 631    buffer->viewport[1].scale[0] = destination->texture->width0;
 632    buffer->viewport[1].scale[1] = destination->texture->height0;
 633
 634    for(i = 0; i < 2; ++i) {
 635       buffer->viewport[i].scale[2] = 1;
 636       buffer->viewport[i].scale[3] = 1;
 637       buffer->viewport[i].translate[0] = 0;
 638       buffer->viewport[i].translate[1] = 0;
 639       buffer->viewport[i].translate[2] = 0;
 640       buffer->viewport[i].translate[3] = 0;
 641
 642       buffer->fb_state[i].zsbuf = NULL;
 643    }
 644
 645    return true;
 646 }
 647
 648 void
 649 vl_idct_cleanup_buffer(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 650 {
 651    unsigned i;
 652
 653    assert(idct && buffer);
 654
 655    for(i = 0; i < idct->nr_of_render_targets; ++i)
 656       pipe_surface_reference(&buffer->fb_state[0].cbufs[i], NULL);
 657
 658    pipe_surface_reference(&buffer->fb_state[1].cbufs[0], NULL);
 659
 660    cleanup_intermediate(idct, buffer);
 661 }
 662
 663 void
 664 vl_idct_flush(struct vl_idct *idct, struct vl_idct_buffer *buffer, unsigned num_instances)
 665 {
 666    assert(idct);
 667    assert(buffer);
 668
 669    if(num_instances > 0) {
 670       idct->pipe->bind_rasterizer_state(idct->pipe, idct->rs_state);
 671       idct->pipe->bind_blend_state(idct->pipe, idct->blend);
 672       idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers);
 673
 674       /* first stage */
 675       idct->pipe->set_framebuffer_state(idct->pipe, &buffer->fb_state[0]);
 676       idct->pipe->set_viewport_state(idct->pipe, &buffer->viewport[0]);
 677       idct->pipe->set_fragment_sampler_views(idct->pipe, 2, buffer->sampler_views.stage[0]);
 678       idct->pipe->bind_vs_state(idct->pipe, idct->matrix_vs);
 679       idct->pipe->bind_fs_state(idct->pipe, idct->matrix_fs);
 680       util_draw_arrays_instanced(idct->pipe, PIPE_PRIM_QUADS, 0, 4, 0, num_instances);
 681
 682       /* second stage */
 683       idct->pipe->set_framebuffer_state(idct->pipe, &buffer->fb_state[1]);
 684       idct->pipe->set_viewport_state(idct->pipe, &buffer->viewport[1]);
 685       idct->pipe->set_fragment_sampler_views(idct->pipe, 2, buffer->sampler_views.stage[1]);
 686       idct->pipe->bind_vs_state(idct->pipe, idct->transpose_vs);
 687       idct->pipe->bind_fs_state(idct->pipe, idct->transpose_fs);
 688       util_draw_arrays_instanced(idct->pipe, PIPE_PRIM_QUADS, 0, 4, 0, num_instances);
 689    }
 690 }