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 "vl_defines.h"
  31 #include "util/u_draw.h"
  32 #include <assert.h>
  33 #include <pipe/p_context.h>
  34 #include <pipe/p_screen.h>
  35 #include <util/u_inlines.h>
  36 #include <util/u_sampler.h>
  37 #include <util/u_format.h>
  38 #include <tgsi/tgsi_ureg.h>
  39 #include "vl_types.h"
  40
  41 enum VS_OUTPUT
  42 {
  43    VS_O_VPOS,
  44    VS_O_L_ADDR0,
  45    VS_O_L_ADDR1,
  46    VS_O_R_ADDR0,
  47    VS_O_R_ADDR1
  48 };
  49
  50 static const float const_matrix[8][8] = {
  51    {  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.3535530f,  0.353553f,  0.3535530f },
  52    {  0.4903930f,  0.4157350f,  0.2777850f,  0.0975451f, -0.0975452f, -0.2777850f, -0.415735f, -0.4903930f },
  53    {  0.4619400f,  0.1913420f, -0.1913420f, -0.4619400f, -0.4619400f, -0.1913420f,  0.191342f,  0.4619400f },
  54    {  0.4157350f, -0.0975452f, -0.4903930f, -0.2777850f,  0.2777850f,  0.4903930f,  0.097545f, -0.4157350f },
  55    {  0.3535530f, -0.3535530f, -0.3535530f,  0.3535540f,  0.3535530f, -0.3535540f, -0.353553f,  0.3535530f },
  56    {  0.2777850f, -0.4903930f,  0.0975452f,  0.4157350f, -0.4157350f, -0.0975451f,  0.490393f, -0.2777850f },
  57    {  0.1913420f, -0.4619400f,  0.4619400f, -0.1913420f, -0.1913410f,  0.4619400f, -0.461940f,  0.1913420f },
  58    {  0.0975451f, -0.2777850f,  0.4157350f, -0.4903930f,  0.4903930f, -0.4157350f,  0.277786f, -0.0975458f }
  59 };
  60
  61 static void
  62 calc_addr(struct ureg_program *shader, struct ureg_dst addr[2],
  63           struct ureg_src tc, struct ureg_src start, bool right_side,
  64           bool transposed, float size)
  65 {
  66    unsigned wm_start = (right_side == transposed) ? TGSI_WRITEMASK_X : TGSI_WRITEMASK_Y;
  67    unsigned sw_start = right_side ? TGSI_SWIZZLE_Y : TGSI_SWIZZLE_X;
  68
  69    unsigned wm_tc = (right_side == transposed) ? TGSI_WRITEMASK_Y : TGSI_WRITEMASK_X;
  70    unsigned sw_tc = right_side ? TGSI_SWIZZLE_X : TGSI_SWIZZLE_Y;
  71
  72    /*
  73     * addr[0..1].(start) = right_side ? start.x : tc.x
  74     * addr[0..1].(tc) = right_side ? tc.y : start.y
  75     * addr[0..1].z = tc.z
  76     * addr[1].(start) += 1.0f / scale
  77     */
  78    ureg_MOV(shader, ureg_writemask(addr[0], wm_start), ureg_scalar(start, sw_start));
  79    ureg_MOV(shader, ureg_writemask(addr[0], wm_tc), ureg_scalar(tc, sw_tc));
  80    ureg_MOV(shader, ureg_writemask(addr[0], TGSI_WRITEMASK_Z), tc);
  81
  82    ureg_ADD(shader, ureg_writemask(addr[1], wm_start), ureg_scalar(start, sw_start), ureg_imm1f(shader, 1.0f / size));
  83    ureg_MOV(shader, ureg_writemask(addr[1], wm_tc), ureg_scalar(tc, sw_tc));
  84    ureg_MOV(shader, ureg_writemask(addr[1], TGSI_WRITEMASK_Z), tc);
  85 }
  86
  87 static void *
  88 create_vert_shader(struct vl_idct *idct, bool matrix_stage)
  89 {
  90    struct ureg_program *shader;
  91    struct ureg_src vrect, vpos, vblock, eb;
  92    struct ureg_src scale, blocks_xy;
  93    struct ureg_dst t_tex, t_start;
  94    struct ureg_dst o_vpos, o_l_addr[2], o_r_addr[2];
  95    unsigned label;
  96
  97    shader = ureg_create(TGSI_PROCESSOR_VERTEX);
  98    if (!shader)
  99       return NULL;
 100
 101    t_tex = ureg_DECL_temporary(shader);
 102    t_start = ureg_DECL_temporary(shader);
 103
 104    vrect = ureg_DECL_vs_input(shader, VS_I_RECT);
 105    vpos = ureg_DECL_vs_input(shader, VS_I_VPOS);
 106    vblock = ureg_swizzle(vrect, TGSI_SWIZZLE_Z, TGSI_SWIZZLE_W, TGSI_SWIZZLE_X, TGSI_SWIZZLE_X);
 107
 108    o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
 109
 110    eb = ureg_DECL_vs_input(shader, VS_I_EB);
 111
 112    o_l_addr[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR0);
 113    o_l_addr[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR1);
 114
 115    o_r_addr[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR0);
 116    o_r_addr[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR1);
 117
 118    /*
 119     * scale = (BLOCK_WIDTH, BLOCK_HEIGHT) / (dst.width, dst.height)
 120     * blocks_xy = (blocks_x, blocks_y)
 121     *
 122     * if eb.(vblock.y, vblock.x)
 123     *    o_vpos.xy = -1
 124     * else
 125     *    t_tex = vpos * blocks_xy + vblock
 126     *    t_start = t_tex * scale
 127     *    t_tex = t_tex + vrect
 128     *    o_vpos.xy = t_tex * scale
 129     *
 130     *    o_l_addr = calc_addr(...)
 131     *    o_r_addr = calc_addr(...)
 132     * endif
 133     * o_vpos.zw = vpos
 134     *
 135     */
 136
 137    scale = ureg_imm2f(shader,
 138       (float)BLOCK_WIDTH / idct->buffer_width,
 139       (float)BLOCK_HEIGHT / idct->buffer_height);
 140
 141    blocks_xy = ureg_imm2f(shader, idct->blocks_x, idct->blocks_y);
 142
 143    if (idct->blocks_x > 1 || idct->blocks_y > 1) {
 144       ureg_CMP(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_XY),
 145          ureg_negate(ureg_scalar(vblock, TGSI_SWIZZLE_Y)),
 146          ureg_swizzle(eb, TGSI_SWIZZLE_Z, TGSI_SWIZZLE_W, TGSI_SWIZZLE_Z, TGSI_SWIZZLE_W),
 147          ureg_swizzle(eb, TGSI_SWIZZLE_X, TGSI_SWIZZLE_Y, TGSI_SWIZZLE_X, TGSI_SWIZZLE_Y));
 148
 149       ureg_CMP(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_X),
 150          ureg_negate(ureg_scalar(vblock, TGSI_SWIZZLE_X)),
 151          ureg_scalar(ureg_src(t_tex), TGSI_SWIZZLE_Y),
 152          ureg_scalar(ureg_src(t_tex), TGSI_SWIZZLE_X));
 153
 154       eb = ureg_src(t_tex);
 155    }
 156
 157    ureg_IF(shader, ureg_scalar(eb, TGSI_SWIZZLE_X), &label);
 158
 159       ureg_MOV(shader, o_vpos, ureg_imm1f(shader, -1.0f));
 160
 161    ureg_fixup_label(shader, label, ureg_get_instruction_number(shader));
 162    ureg_ELSE(shader, &label);
 163
 164       ureg_MAD(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_XY), vpos, blocks_xy, vblock);
 165       ureg_MUL(shader, ureg_writemask(t_start, TGSI_WRITEMASK_XY), ureg_src(t_tex), scale);
 166
 167       ureg_ADD(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_XY), ureg_src(t_tex), vrect);
 168
 169       ureg_MUL(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_XY), ureg_src(t_tex), scale);
 170       ureg_MUL(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_Z),
 171          ureg_scalar(vrect, TGSI_SWIZZLE_X),
 172          ureg_imm1f(shader, BLOCK_WIDTH / idct->nr_of_render_targets));
 173
 174       ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_XY), ureg_src(t_tex));
 175
 176       if(matrix_stage) {
 177          calc_addr(shader, o_l_addr, ureg_src(t_tex), ureg_src(t_start), false, false, idct->buffer_width / 4);
 178          calc_addr(shader, o_r_addr, vrect, ureg_imm1f(shader, 0.0f), true, true, BLOCK_WIDTH / 4);
 179       } else {
 180          calc_addr(shader, o_l_addr, vrect, ureg_imm1f(shader, 0.0f), false, false, BLOCK_WIDTH / 4);
 181          calc_addr(shader, o_r_addr, ureg_src(t_tex), ureg_src(t_start), true, false, idct->buffer_height / 4);
 182       }
 183
 184    ureg_fixup_label(shader, label, ureg_get_instruction_number(shader));
 185    ureg_ENDIF(shader);
 186
 187    ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_ZW), vpos);
 188
 189    ureg_release_temporary(shader, t_tex);
 190    ureg_release_temporary(shader, t_start);
 191
 192    ureg_END(shader);
 193
 194    return ureg_create_shader_and_destroy(shader, idct->pipe);
 195 }
 196
 197 static void
 198 increment_addr(struct ureg_program *shader, struct ureg_dst daddr[2],
 199                struct ureg_src saddr[2], bool right_side, bool transposed,
 200                int pos, float size)
 201 {
 202    unsigned wm_start = (right_side == transposed) ? TGSI_WRITEMASK_X : TGSI_WRITEMASK_Y;
 203    unsigned wm_tc = (right_side == transposed) ? TGSI_WRITEMASK_Y : TGSI_WRITEMASK_X;
 204
 205    /*
 206     * daddr[0..1].(start) = saddr[0..1].(start)
 207     * daddr[0..1].(tc) = saddr[0..1].(tc)
 208     */
 209
 210    ureg_MOV(shader, ureg_writemask(daddr[0], wm_start), saddr[0]);
 211    ureg_ADD(shader, ureg_writemask(daddr[0], wm_tc), saddr[0], ureg_imm1f(shader, pos / size));
 212    ureg_MOV(shader, ureg_writemask(daddr[1], wm_start), saddr[1]);
 213    ureg_ADD(shader, ureg_writemask(daddr[1], wm_tc), saddr[1], ureg_imm1f(shader, pos / size));
 214 }
 215
 216 static void
 217 fetch_four(struct ureg_program *shader, struct ureg_dst m[2], struct ureg_src addr[2], struct ureg_src sampler)
 218 {
 219    ureg_TEX(shader, m[0], TGSI_TEXTURE_3D, addr[0], sampler);
 220    ureg_TEX(shader, m[1], TGSI_TEXTURE_3D, addr[1], sampler);
 221 }
 222
 223 static void
 224 matrix_mul(struct ureg_program *shader, struct ureg_dst dst, struct ureg_dst l[2], struct ureg_dst r[2])
 225 {
 226    struct ureg_dst tmp;
 227
 228    tmp = ureg_DECL_temporary(shader);
 229
 230    /*
 231     * tmp.xy = dot4(m[0][0..1], m[1][0..1])
 232     * dst = tmp.x + tmp.y
 233     */
 234    ureg_DP4(shader, ureg_writemask(tmp, TGSI_WRITEMASK_X), ureg_src(l[0]), ureg_src(r[0]));
 235    ureg_DP4(shader, ureg_writemask(tmp, TGSI_WRITEMASK_Y), ureg_src(l[1]), ureg_src(r[1]));
 236    ureg_ADD(shader, dst,
 237       ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_X),
 238       ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_Y));
 239
 240    ureg_release_temporary(shader, tmp);
 241 }
 242
 243 static void *
 244 create_matrix_frag_shader(struct vl_idct *idct)
 245 {
 246    struct ureg_program *shader;
 247
 248    struct ureg_src l_addr[2], r_addr[2];
 249
 250    struct ureg_dst l[4][2], r[2];
 251    struct ureg_dst fragment[idct->nr_of_render_targets];
 252
 253    unsigned i, j;
 254
 255    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 256    if (!shader)
 257       return NULL;
 258
 259    l_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR0, TGSI_INTERPOLATE_LINEAR);
 260    l_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR1, TGSI_INTERPOLATE_LINEAR);
 261
 262    r_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR0, TGSI_INTERPOLATE_LINEAR);
 263    r_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR1, TGSI_INTERPOLATE_LINEAR);
 264
 265    for (i = 0; i < idct->nr_of_render_targets; ++i)
 266        fragment[i] = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, i);
 267
 268    for (i = 0; i < 4; ++i) {
 269       l[i][0] = ureg_DECL_temporary(shader);
 270       l[i][1] = ureg_DECL_temporary(shader);
 271    }
 272
 273    r[0] = ureg_DECL_temporary(shader);
 274    r[1] = ureg_DECL_temporary(shader);
 275
 276    for (i = 1; i < 4; ++i) {
 277       increment_addr(shader, l[i], l_addr, false, false, i, idct->buffer_height);
 278    }
 279
 280    for (i = 0; i < 4; ++i) {
 281       struct ureg_src s_addr[2];
 282       s_addr[0] = i == 0 ? l_addr[0] : ureg_src(l[i][0]);
 283       s_addr[1] = i == 0 ? l_addr[1] : ureg_src(l[i][1]);
 284       fetch_four(shader, l[i], s_addr, ureg_DECL_sampler(shader, 1));
 285    }
 286
 287    for (i = 0; i < idct->nr_of_render_targets; ++i) {
 288       if(i > 0)
 289          increment_addr(shader, r, r_addr, true, true, i, BLOCK_HEIGHT);
 290
 291       struct ureg_src s_addr[2] = { ureg_src(r[0]), ureg_src(r[1]) };
 292       s_addr[0] = i == 0 ? r_addr[0] : ureg_src(r[0]);
 293       s_addr[1] = i == 0 ? r_addr[1] : ureg_src(r[1]);
 294       fetch_four(shader, r, s_addr, ureg_DECL_sampler(shader, 0));
 295
 296       for (j = 0; j < 4; ++j) {
 297          matrix_mul(shader, ureg_writemask(fragment[i], TGSI_WRITEMASK_X << j), l[j], r);
 298       }
 299    }
 300
 301    for (i = 0; i < 4; ++i) {
 302       ureg_release_temporary(shader, l[i][0]);
 303       ureg_release_temporary(shader, l[i][1]);
 304    }
 305    ureg_release_temporary(shader, r[0]);
 306    ureg_release_temporary(shader, r[1]);
 307
 308    ureg_END(shader);
 309
 310    return ureg_create_shader_and_destroy(shader, idct->pipe);
 311 }
 312
 313 static void *
 314 create_transpose_frag_shader(struct vl_idct *idct)
 315 {
 316    struct ureg_program *shader;
 317
 318    struct ureg_src l_addr[2], r_addr[2];
 319
 320    struct ureg_dst l[2], r[2];
 321    struct ureg_dst fragment;
 322
 323    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 324    if (!shader)
 325       return NULL;
 326
 327    l_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR0, TGSI_INTERPOLATE_LINEAR);
 328    l_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR1, TGSI_INTERPOLATE_LINEAR);
 329
 330    r_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR0, TGSI_INTERPOLATE_LINEAR);
 331    r_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR1, TGSI_INTERPOLATE_LINEAR);
 332
 333    l[0] = ureg_DECL_temporary(shader);
 334    l[1] = ureg_DECL_temporary(shader);
 335    r[0] = ureg_DECL_temporary(shader);
 336    r[1] = ureg_DECL_temporary(shader);
 337
 338    fetch_four(shader, l, l_addr, ureg_DECL_sampler(shader, 0));
 339    fetch_four(shader, r, r_addr, ureg_DECL_sampler(shader, 1));
 340
 341    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 342
 343    matrix_mul(shader, ureg_writemask(fragment, TGSI_WRITEMASK_X), l, r);
 344
 345    ureg_release_temporary(shader, l[0]);
 346    ureg_release_temporary(shader, l[1]);
 347    ureg_release_temporary(shader, r[0]);
 348    ureg_release_temporary(shader, r[1]);
 349
 350    ureg_END(shader);
 351
 352    return ureg_create_shader_and_destroy(shader, idct->pipe);
 353 }
 354
 355 static bool
 356 init_shaders(struct vl_idct *idct)
 357 {
 358    idct->matrix_vs = create_vert_shader(idct, true);
 359    if (!idct->matrix_vs)
 360       goto error_matrix_vs;
 361
 362    idct->matrix_fs = create_matrix_frag_shader(idct);
 363    if (!idct->matrix_fs)
 364       goto error_matrix_fs;
 365
 366    idct->transpose_vs = create_vert_shader(idct, false);
 367    if (!idct->transpose_vs)
 368       goto error_transpose_vs;
 369
 370    idct->transpose_fs = create_transpose_frag_shader(idct);
 371    if (!idct->transpose_fs)
 372       goto error_transpose_fs;
 373
 374    return true;
 375
 376 error_transpose_fs:
 377    idct->pipe->delete_vs_state(idct->pipe, idct->transpose_vs);
 378
 379 error_transpose_vs:
 380    idct->pipe->delete_fs_state(idct->pipe, idct->matrix_fs);
 381
 382 error_matrix_fs:
 383    idct->pipe->delete_vs_state(idct->pipe, idct->matrix_vs);
 384
 385 error_matrix_vs:
 386    return false;
 387 }
 388
 389 static void
 390 cleanup_shaders(struct vl_idct *idct)
 391 {
 392    idct->pipe->delete_vs_state(idct->pipe, idct->matrix_vs);
 393    idct->pipe->delete_fs_state(idct->pipe, idct->matrix_fs);
 394    idct->pipe->delete_vs_state(idct->pipe, idct->transpose_vs);
 395    idct->pipe->delete_fs_state(idct->pipe, idct->transpose_fs);
 396 }
 397
 398 static bool
 399 init_state(struct vl_idct *idct)
 400 {
 401    struct pipe_sampler_state sampler;
 402    struct pipe_rasterizer_state rs_state;
 403    unsigned i;
 404
 405    assert(idct);
 406
 407    memset(&rs_state, 0, sizeof(rs_state));
 408    rs_state.gl_rasterization_rules = false;
 409    idct->rs_state = idct->pipe->create_rasterizer_state(idct->pipe, &rs_state);
 410    if (!idct->rs_state)
 411       goto error_rs_state;
 412
 413    for (i = 0; i < 2; ++i) {
 414       memset(&sampler, 0, sizeof(sampler));
 415       sampler.wrap_s = PIPE_TEX_WRAP_REPEAT;
 416       sampler.wrap_t = PIPE_TEX_WRAP_REPEAT;
 417       sampler.wrap_r = PIPE_TEX_WRAP_REPEAT;
 418       sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
 419       sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
 420       sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
 421       sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
 422       sampler.compare_func = PIPE_FUNC_ALWAYS;
 423       sampler.normalized_coords = 1;
 424       idct->samplers[i] = idct->pipe->create_sampler_state(idct->pipe, &sampler);
 425       if (!idct->samplers[i])
 426          goto error_samplers;
 427    }
 428
 429    return true;
 430
 431 error_samplers:
 432    for (i = 0; i < 2; ++i)
 433       if (idct->samplers[i])
 434          idct->pipe->delete_sampler_state(idct->pipe, idct->samplers[i]);
 435
 436    idct->pipe->delete_rasterizer_state(idct->pipe, idct->rs_state);
 437
 438 error_rs_state:
 439    return false;
 440 }
 441
 442 static void
 443 cleanup_state(struct vl_idct *idct)
 444 {
 445    unsigned i;
 446
 447    for (i = 0; i < 2; ++i)
 448       idct->pipe->delete_sampler_state(idct->pipe, idct->samplers[i]);
 449
 450    idct->pipe->delete_rasterizer_state(idct->pipe, idct->rs_state);
 451 }
 452
 453 static bool
 454 init_intermediate(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 455 {
 456    struct pipe_resource *tex;
 457    struct pipe_surface surf_templ;
 458    unsigned i;
 459
 460    assert(idct && buffer);
 461
 462    tex = buffer->sampler_views.individual.intermediate->texture;
 463
 464    buffer->fb_state[0].width = tex->width0;
 465    buffer->fb_state[0].height = tex->height0;
 466    buffer->fb_state[0].nr_cbufs = idct->nr_of_render_targets;
 467    for(i = 0; i < idct->nr_of_render_targets; ++i) {
 468       memset(&surf_templ, 0, sizeof(surf_templ));
 469       surf_templ.format = tex->format;
 470       surf_templ.u.tex.first_layer = i;
 471       surf_templ.u.tex.last_layer = i;
 472       surf_templ.usage = PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET;
 473       buffer->fb_state[0].cbufs[i] = idct->pipe->create_surface(
 474          idct->pipe, tex, &surf_templ);
 475
 476       if (!buffer->fb_state[0].cbufs[i])
 477          goto error_surfaces;
 478    }
 479
 480    buffer->viewport[0].scale[0] = tex->width0;
 481    buffer->viewport[0].scale[1] = tex->height0;
 482
 483    return true;
 484
 485 error_surfaces:
 486    for(i = 0; i < idct->nr_of_render_targets; ++i)
 487       pipe_surface_reference(&buffer->fb_state[0].cbufs[i], NULL);
 488
 489    return false;
 490 }
 491
 492 static void
 493 cleanup_intermediate(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 494 {
 495    unsigned i;
 496
 497    assert(idct && buffer);
 498
 499    for(i = 0; i < idct->nr_of_render_targets; ++i)
 500       pipe_surface_reference(&buffer->fb_state[0].cbufs[i], NULL);
 501
 502    pipe_sampler_view_reference(&buffer->sampler_views.individual.intermediate, NULL);
 503 }
 504
 505 struct pipe_sampler_view *
 506 vl_idct_upload_matrix(struct pipe_context *pipe, float scale)
 507 {
 508    struct pipe_resource tex_templ, *matrix;
 509    struct pipe_sampler_view sv_templ, *sv;
 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    assert(pipe);
 523
 524    memset(&tex_templ, 0, sizeof(tex_templ));
 525    tex_templ.target = PIPE_TEXTURE_2D;
 526    tex_templ.format = PIPE_FORMAT_R32G32B32A32_FLOAT;
 527    tex_templ.last_level = 0;
 528    tex_templ.width0 = 2;
 529    tex_templ.height0 = 8;
 530    tex_templ.depth0 = 1;
 531    tex_templ.array_size = 1;
 532    tex_templ.usage = PIPE_USAGE_IMMUTABLE;
 533    tex_templ.bind = PIPE_BIND_SAMPLER_VIEW;
 534    tex_templ.flags = 0;
 535
 536    matrix = pipe->screen->resource_create(pipe->screen, &tex_templ);
 537    if (!matrix)
 538       goto error_matrix;
 539
 540    buf_transfer = pipe->get_transfer
 541    (
 542       pipe, matrix,
 543       0, PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 544       &rect
 545    );
 546    if (!buf_transfer)
 547       goto error_transfer;
 548
 549    pitch = buf_transfer->stride / sizeof(float);
 550
 551    f = pipe->transfer_map(pipe, buf_transfer);
 552    if (!f)
 553       goto error_map;
 554
 555    for(i = 0; i < BLOCK_HEIGHT; ++i)
 556       for(j = 0; j < BLOCK_WIDTH; ++j)
 557          // transpose and scale
 558          f[i * pitch + j] = const_matrix[j][i] * scale;
 559
 560    pipe->transfer_unmap(pipe, buf_transfer);
 561    pipe->transfer_destroy(pipe, buf_transfer);
 562
 563    memset(&sv_templ, 0, sizeof(sv_templ));
 564    u_sampler_view_default_template(&sv_templ, matrix, matrix->format);
 565    sv = pipe->create_sampler_view(pipe, matrix, &sv_templ);
 566    pipe_resource_reference(&matrix, NULL);
 567    if (!sv)
 568       goto error_map;
 569
 570    return sv;
 571
 572 error_map:
 573    pipe->transfer_destroy(pipe, buf_transfer);
 574
 575 error_transfer:
 576    pipe_resource_reference(&matrix, NULL);
 577
 578 error_matrix:
 579    return NULL;
 580 }
 581
 582 bool vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe,
 583                   unsigned buffer_width, unsigned buffer_height,
 584                   unsigned blocks_x, unsigned blocks_y,
 585                   unsigned nr_of_render_targets,
 586                   struct pipe_sampler_view *matrix)
 587 {
 588    assert(idct && pipe && matrix);
 589
 590    idct->pipe = pipe;
 591    idct->buffer_width = buffer_width;
 592    idct->buffer_height = buffer_height;
 593    idct->blocks_x = blocks_x;
 594    idct->blocks_y = blocks_y;
 595    idct->nr_of_render_targets = nr_of_render_targets;
 596    pipe_sampler_view_reference(&idct->matrix, matrix);
 597
 598    if(!init_shaders(idct))
 599       return false;
 600
 601    if(!init_state(idct)) {
 602       cleanup_shaders(idct);
 603       return false;
 604    }
 605
 606    return true;
 607 }
 608
 609 void
 610 vl_idct_cleanup(struct vl_idct *idct)
 611 {
 612    cleanup_shaders(idct);
 613    cleanup_state(idct);
 614
 615    pipe_sampler_view_reference(&idct->matrix, NULL);
 616 }
 617
 618 bool
 619 vl_idct_init_buffer(struct vl_idct *idct, struct vl_idct_buffer *buffer,
 620                     struct pipe_sampler_view *source,
 621                     struct pipe_sampler_view *intermediate,
 622                     struct pipe_surface *destination)
 623 {
 624    unsigned i;
 625
 626    assert(buffer);
 627    assert(idct);
 628    assert(source);
 629    assert(destination);
 630
 631    pipe_sampler_view_reference(&buffer->sampler_views.individual.matrix, idct->matrix);
 632    pipe_sampler_view_reference(&buffer->sampler_views.individual.source, source);
 633    pipe_sampler_view_reference(&buffer->sampler_views.individual.transpose, idct->matrix);
 634    pipe_sampler_view_reference(&buffer->sampler_views.individual.intermediate, intermediate);
 635
 636    if (!init_intermediate(idct, buffer))
 637       return false;
 638
 639    /* init state */
 640    buffer->fb_state[1].width = destination->texture->width0;
 641    buffer->fb_state[1].height = destination->texture->height0;
 642    buffer->fb_state[1].nr_cbufs = 1;
 643    pipe_surface_reference(&buffer->fb_state[1].cbufs[0], destination);
 644
 645    buffer->viewport[1].scale[0] = destination->texture->width0;
 646    buffer->viewport[1].scale[1] = destination->texture->height0;
 647
 648    for(i = 0; i < 2; ++i) {
 649       buffer->viewport[i].scale[2] = 1;
 650       buffer->viewport[i].scale[3] = 1;
 651       buffer->viewport[i].translate[0] = 0;
 652       buffer->viewport[i].translate[1] = 0;
 653       buffer->viewport[i].translate[2] = 0;
 654       buffer->viewport[i].translate[3] = 0;
 655
 656       buffer->fb_state[i].zsbuf = NULL;
 657    }
 658
 659    return true;
 660 }
 661
 662 void
 663 vl_idct_cleanup_buffer(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 664 {
 665    unsigned i;
 666
 667    assert(idct && buffer);
 668
 669    for(i = 0; i < idct->nr_of_render_targets; ++i)
 670       pipe_surface_reference(&buffer->fb_state[0].cbufs[i], NULL);
 671
 672    pipe_surface_reference(&buffer->fb_state[1].cbufs[0], NULL);
 673
 674    cleanup_intermediate(idct, buffer);
 675 }
 676
 677 void
 678 vl_idct_flush(struct vl_idct *idct, struct vl_idct_buffer *buffer, unsigned num_instances)
 679 {
 680    unsigned num_verts;
 681
 682    assert(idct);
 683    assert(buffer);
 684
 685    if(num_instances > 0) {
 686       num_verts = idct->blocks_x * idct->blocks_y * 4;
 687
 688       idct->pipe->bind_rasterizer_state(idct->pipe, idct->rs_state);
 689       idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers);
 690
 691       /* first stage */
 692       idct->pipe->set_framebuffer_state(idct->pipe, &buffer->fb_state[0]);
 693       idct->pipe->set_viewport_state(idct->pipe, &buffer->viewport[0]);
 694       idct->pipe->set_fragment_sampler_views(idct->pipe, 2, buffer->sampler_views.stage[0]);
 695       idct->pipe->bind_vs_state(idct->pipe, idct->matrix_vs);
 696       idct->pipe->bind_fs_state(idct->pipe, idct->matrix_fs);
 697       util_draw_arrays_instanced(idct->pipe, PIPE_PRIM_QUADS, 0, num_verts, 0, num_instances);
 698
 699       /* second stage */
 700       idct->pipe->set_framebuffer_state(idct->pipe, &buffer->fb_state[1]);
 701       idct->pipe->set_viewport_state(idct->pipe, &buffer->viewport[1]);
 702       idct->pipe->set_fragment_sampler_views(idct->pipe, 2, buffer->sampler_views.stage[1]);
 703       idct->pipe->bind_vs_state(idct->pipe, idct->transpose_vs);
 704       idct->pipe->bind_fs_state(idct->pipe, idct->transpose_fs);
 705       util_draw_arrays_instanced(idct->pipe, PIPE_PRIM_QUADS, 0, num_verts, 0, num_instances);
 706    }
 707 }