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 /**
  53  * The DCT matrix stored as hex representation of floats. Equal to the following equation:
  54  * for (i = 0; i < 8; ++i)
  55  *    for (j = 0; j < 8; ++j)
  56  *       if (i == 0) const_matrix[i][j] = 1.0f / sqrtf(8.0f);
  57  *       else const_matrix[i][j] = sqrtf(2.0f / 8.0f) * cosf((2 * j + 1) * i * M_PI / (2.0f * 8.0f));
  58  */
  59 static const uint32_t const_matrix[8][8] = {
  60    { 0x3eb504f3, 0x3eb504f3, 0x3eb504f3, 0x3eb504f3, 0x3eb504f3, 0x3eb504f3, 0x3eb504f3, 0x3eb504f3 },
  61    { 0x3efb14be, 0x3ed4db31, 0x3e8e39da, 0x3dc7c5c4, 0xbdc7c5c2, 0xbe8e39d9, 0xbed4db32, 0xbefb14bf },
  62    { 0x3eec835f, 0x3e43ef15, 0xbe43ef14, 0xbeec835e, 0xbeec835f, 0xbe43ef1a, 0x3e43ef1b, 0x3eec835f },
  63    { 0x3ed4db31, 0xbdc7c5c2, 0xbefb14bf, 0xbe8e39dd, 0x3e8e39d7, 0x3efb14bf, 0x3dc7c5d0, 0xbed4db34 },
  64    { 0x3eb504f3, 0xbeb504f3, 0xbeb504f4, 0x3eb504f1, 0x3eb504f3, 0xbeb504f0, 0xbeb504ef, 0x3eb504f4 },
  65    { 0x3e8e39da, 0xbefb14bf, 0x3dc7c5c8, 0x3ed4db32, 0xbed4db34, 0xbdc7c5bb, 0x3efb14bf, 0xbe8e39d7 },
  66    { 0x3e43ef15, 0xbeec835f, 0x3eec835f, 0xbe43ef07, 0xbe43ef23, 0x3eec8361, 0xbeec835c, 0x3e43ef25 },
  67    { 0x3dc7c5c4, 0xbe8e39dd, 0x3ed4db32, 0xbefb14c0, 0x3efb14be, 0xbed4db31, 0x3e8e39ce, 0xbdc7c596 },
  68 };
  69
  70 static void
  71 calc_addr(struct ureg_program *shader, struct ureg_dst addr[2],
  72           struct ureg_src tc, struct ureg_src start, bool right_side,
  73           bool transposed, float size)
  74 {
  75    unsigned wm_start = (right_side == transposed) ? TGSI_WRITEMASK_X : TGSI_WRITEMASK_Y;
  76    unsigned sw_start = right_side ? TGSI_SWIZZLE_Y : TGSI_SWIZZLE_X;
  77
  78    unsigned wm_tc = (right_side == transposed) ? TGSI_WRITEMASK_Y : TGSI_WRITEMASK_X;
  79    unsigned sw_tc = right_side ? TGSI_SWIZZLE_X : TGSI_SWIZZLE_Y;
  80
  81    /*
  82     * addr[0..1].(start) = right_side ? start.x : tc.x
  83     * addr[0..1].(tc) = right_side ? tc.y : start.y
  84     * addr[0..1].z = tc.z
  85     * addr[1].(start) += 1.0f / scale
  86     */
  87    ureg_MOV(shader, ureg_writemask(addr[0], wm_start), ureg_scalar(start, sw_start));
  88    ureg_MOV(shader, ureg_writemask(addr[0], wm_tc), ureg_scalar(tc, sw_tc));
  89    ureg_MOV(shader, ureg_writemask(addr[0], TGSI_WRITEMASK_Z), tc);
  90
  91    ureg_ADD(shader, ureg_writemask(addr[1], wm_start), ureg_scalar(start, sw_start), ureg_imm1f(shader, 1.0f / size));
  92    ureg_MOV(shader, ureg_writemask(addr[1], wm_tc), ureg_scalar(tc, sw_tc));
  93    ureg_MOV(shader, ureg_writemask(addr[1], TGSI_WRITEMASK_Z), tc);
  94 }
  95
  96 static void
  97 increment_addr(struct ureg_program *shader, struct ureg_dst daddr[2],
  98                struct ureg_src saddr[2], bool right_side, bool transposed,
  99                int pos, float size)
 100 {
 101    unsigned wm_start = (right_side == transposed) ? TGSI_WRITEMASK_X : TGSI_WRITEMASK_Y;
 102    unsigned wm_tc = (right_side == transposed) ? TGSI_WRITEMASK_Y : TGSI_WRITEMASK_X;
 103
 104    /*
 105     * daddr[0..1].(start) = saddr[0..1].(start)
 106     * daddr[0..1].(tc) = saddr[0..1].(tc)
 107     */
 108
 109    ureg_MOV(shader, ureg_writemask(daddr[0], wm_start), saddr[0]);
 110    ureg_ADD(shader, ureg_writemask(daddr[0], wm_tc), saddr[0], ureg_imm1f(shader, pos / size));
 111    ureg_MOV(shader, ureg_writemask(daddr[1], wm_start), saddr[1]);
 112    ureg_ADD(shader, ureg_writemask(daddr[1], wm_tc), saddr[1], ureg_imm1f(shader, pos / size));
 113 }
 114
 115 static void
 116 fetch_four(struct ureg_program *shader, struct ureg_dst m[2], struct ureg_src addr[2], struct ureg_src sampler)
 117 {
 118    ureg_TEX(shader, m[0], TGSI_TEXTURE_3D, addr[0], sampler);
 119    ureg_TEX(shader, m[1], TGSI_TEXTURE_3D, addr[1], sampler);
 120 }
 121
 122 static void
 123 matrix_mul(struct ureg_program *shader, struct ureg_dst dst, struct ureg_dst l[2], struct ureg_dst r[2])
 124 {
 125    struct ureg_dst tmp;
 126
 127    tmp = ureg_DECL_temporary(shader);
 128
 129    /*
 130     * tmp.xy = dot4(m[0][0..1], m[1][0..1])
 131     * dst = tmp.x + tmp.y
 132     */
 133    ureg_DP4(shader, ureg_writemask(tmp, TGSI_WRITEMASK_X), ureg_src(l[0]), ureg_src(r[0]));
 134    ureg_DP4(shader, ureg_writemask(tmp, TGSI_WRITEMASK_Y), ureg_src(l[1]), ureg_src(r[1]));
 135    ureg_ADD(shader, dst,
 136       ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_X),
 137       ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_Y));
 138
 139    ureg_release_temporary(shader, tmp);
 140 }
 141
 142 static void *
 143 create_stage1_vert_shader(struct vl_idct *idct)
 144 {
 145    struct ureg_program *shader;
 146    struct ureg_src vrect, vpos;
 147    struct ureg_src scale;
 148    struct ureg_dst t_tex, t_start;
 149    struct ureg_dst o_vpos, o_l_addr[2], o_r_addr[2];
 150
 151    shader = ureg_create(TGSI_PROCESSOR_VERTEX);
 152    if (!shader)
 153       return NULL;
 154
 155    vrect = ureg_DECL_vs_input(shader, VS_I_RECT);
 156    vpos = ureg_DECL_vs_input(shader, VS_I_VPOS);
 157
 158    t_tex = ureg_DECL_temporary(shader);
 159    t_start = ureg_DECL_temporary(shader);
 160
 161    o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
 162
 163    o_l_addr[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR0);
 164    o_l_addr[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR1);
 165
 166    o_r_addr[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR0);
 167    o_r_addr[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR1);
 168
 169    /*
 170     * scale = (BLOCK_WIDTH, BLOCK_HEIGHT) / (dst.width, dst.height)
 171     *
 172     * t_vpos = vpos + vrect
 173     * o_vpos.xy = t_vpos * scale
 174     * o_vpos.zw = vpos
 175     *
 176     * o_l_addr = calc_addr(...)
 177     * o_r_addr = calc_addr(...)
 178     *
 179     */
 180
 181    scale = ureg_imm2f(shader,
 182       (float)BLOCK_WIDTH / idct->buffer_width,
 183       (float)BLOCK_HEIGHT / idct->buffer_height);
 184
 185    ureg_ADD(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_XY), vpos, vrect);
 186    ureg_MUL(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_XY), ureg_src(t_tex), scale);
 187
 188    ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_XY), ureg_src(t_tex));
 189    ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_ZW), ureg_imm1f(shader, 1.0f));
 190
 191    ureg_MUL(shader, ureg_writemask(t_tex, TGSI_WRITEMASK_Z),
 192       ureg_scalar(vrect, TGSI_SWIZZLE_X),
 193       ureg_imm1f(shader, BLOCK_WIDTH / idct->nr_of_render_targets));
 194    ureg_MUL(shader, ureg_writemask(t_start, TGSI_WRITEMASK_XY), vpos, scale);
 195
 196    calc_addr(shader, o_l_addr, ureg_src(t_tex), ureg_src(t_start), false, false, idct->buffer_width / 4);
 197    calc_addr(shader, o_r_addr, vrect, ureg_imm1f(shader, 0.0f), true, true, BLOCK_WIDTH / 4);
 198
 199    ureg_release_temporary(shader, t_tex);
 200    ureg_release_temporary(shader, t_start);
 201
 202    ureg_END(shader);
 203
 204    return ureg_create_shader_and_destroy(shader, idct->pipe);
 205 }
 206
 207 static void *
 208 create_stage1_frag_shader(struct vl_idct *idct)
 209 {
 210    struct ureg_program *shader;
 211
 212    struct ureg_src l_addr[2], r_addr[2];
 213
 214    struct ureg_dst l[4][2], r[2];
 215    struct ureg_dst fragment[idct->nr_of_render_targets];
 216
 217    int i, j;
 218
 219    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 220    if (!shader)
 221       return NULL;
 222
 223    l_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR0, TGSI_INTERPOLATE_LINEAR);
 224    l_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_L_ADDR1, TGSI_INTERPOLATE_LINEAR);
 225
 226    r_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR0, TGSI_INTERPOLATE_LINEAR);
 227    r_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_R_ADDR1, TGSI_INTERPOLATE_LINEAR);
 228
 229    for (i = 0; i < idct->nr_of_render_targets; ++i)
 230        fragment[i] = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, i);
 231
 232    for (i = 0; i < 4; ++i) {
 233       l[i][0] = ureg_DECL_temporary(shader);
 234       l[i][1] = ureg_DECL_temporary(shader);
 235    }
 236
 237    r[0] = ureg_DECL_temporary(shader);
 238    r[1] = ureg_DECL_temporary(shader);
 239
 240    for (i = 0; i < 4; ++i) {
 241       increment_addr(shader, l[i], l_addr, false, false, i - 2, idct->buffer_height);
 242    }
 243
 244    for (i = 0; i < 4; ++i) {
 245       struct ureg_src s_addr[2] = { ureg_src(l[i][0]), 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       increment_addr(shader, r, r_addr, true, true, i - (signed)idct->nr_of_render_targets / 2, BLOCK_HEIGHT);
 251
 252       struct ureg_src s_addr[2] = { ureg_src(r[0]), ureg_src(r[1]) };
 253       fetch_four(shader, r, s_addr, ureg_DECL_sampler(shader, 0));
 254
 255       for (j = 0; j < 4; ++j) {
 256          matrix_mul(shader, ureg_writemask(fragment[i], TGSI_WRITEMASK_X << j), l[j], r);
 257       }
 258    }
 259
 260    for (i = 0; i < 4; ++i) {
 261       ureg_release_temporary(shader, l[i][0]);
 262       ureg_release_temporary(shader, l[i][1]);
 263    }
 264    ureg_release_temporary(shader, r[0]);
 265    ureg_release_temporary(shader, r[1]);
 266
 267    ureg_END(shader);
 268
 269    return ureg_create_shader_and_destroy(shader, idct->pipe);
 270 }
 271
 272 void
 273 vl_idct_stage2_vert_shader(struct vl_idct *idct, struct ureg_program *shader,
 274                            unsigned first_output, struct ureg_dst tex)
 275 {
 276    struct ureg_src vrect, vpos;
 277    struct ureg_src scale;
 278    struct ureg_dst t_start;
 279    struct ureg_dst o_l_addr[2], o_r_addr[2];
 280
 281    vrect = ureg_DECL_vs_input(shader, VS_I_RECT);
 282    vpos = ureg_DECL_vs_input(shader, VS_I_VPOS);
 283
 284    t_start = ureg_DECL_temporary(shader);
 285
 286    --first_output;
 287
 288    o_l_addr[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, first_output + VS_O_L_ADDR0);
 289    o_l_addr[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, first_output + VS_O_L_ADDR1);
 290
 291    o_r_addr[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, first_output + VS_O_R_ADDR0);
 292    o_r_addr[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, first_output + VS_O_R_ADDR1);
 293
 294    scale = ureg_imm2f(shader,
 295       (float)BLOCK_WIDTH / idct->buffer_width,
 296       (float)BLOCK_HEIGHT / idct->buffer_height);
 297
 298    ureg_MUL(shader, ureg_writemask(tex, TGSI_WRITEMASK_Z),
 299       ureg_scalar(vrect, TGSI_SWIZZLE_X),
 300       ureg_imm1f(shader, BLOCK_WIDTH / idct->nr_of_render_targets));
 301    ureg_MUL(shader, ureg_writemask(t_start, TGSI_WRITEMASK_XY), vpos, scale);
 302
 303    calc_addr(shader, o_l_addr, vrect, ureg_imm1f(shader, 0.0f), false, false, BLOCK_WIDTH / 4);
 304    calc_addr(shader, o_r_addr, ureg_src(tex), ureg_src(t_start), true, false, idct->buffer_height / 4);
 305 }
 306
 307 void
 308 vl_idct_stage2_frag_shader(struct vl_idct *idct, struct ureg_program *shader,
 309                            unsigned first_input, struct ureg_dst fragment)
 310 {
 311    struct ureg_src l_addr[2], r_addr[2];
 312
 313    struct ureg_dst l[2], r[2];
 314
 315    --first_input;
 316
 317    l_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, first_input + VS_O_L_ADDR0, TGSI_INTERPOLATE_LINEAR);
 318    l_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, first_input + VS_O_L_ADDR1, TGSI_INTERPOLATE_LINEAR);
 319
 320    r_addr[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, first_input + VS_O_R_ADDR0, TGSI_INTERPOLATE_LINEAR);
 321    r_addr[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, first_input + VS_O_R_ADDR1, TGSI_INTERPOLATE_LINEAR);
 322
 323    l[0] = ureg_DECL_temporary(shader);
 324    l[1] = ureg_DECL_temporary(shader);
 325    r[0] = ureg_DECL_temporary(shader);
 326    r[1] = ureg_DECL_temporary(shader);
 327
 328    fetch_four(shader, l, l_addr, ureg_DECL_sampler(shader, 0));
 329    fetch_four(shader, r, r_addr, ureg_DECL_sampler(shader, 1));
 330
 331    matrix_mul(shader, fragment, l, r);
 332
 333    ureg_release_temporary(shader, l[0]);
 334    ureg_release_temporary(shader, l[1]);
 335    ureg_release_temporary(shader, r[0]);
 336    ureg_release_temporary(shader, r[1]);
 337 }
 338
 339 static bool
 340 init_shaders(struct vl_idct *idct)
 341 {
 342    idct->vs = create_stage1_vert_shader(idct);
 343    if (!idct->vs)
 344       goto error_vs;
 345
 346    idct->fs = create_stage1_frag_shader(idct);
 347    if (!idct->fs)
 348       goto error_fs;
 349
 350    return true;
 351
 352 error_fs:
 353    idct->pipe->delete_vs_state(idct->pipe, idct->vs);
 354
 355 error_vs:
 356    return false;
 357 }
 358
 359 static void
 360 cleanup_shaders(struct vl_idct *idct)
 361 {
 362    idct->pipe->delete_vs_state(idct->pipe, idct->vs);
 363    idct->pipe->delete_fs_state(idct->pipe, idct->fs);
 364 }
 365
 366 static bool
 367 init_state(struct vl_idct *idct)
 368 {
 369    struct pipe_blend_state blend;
 370    struct pipe_rasterizer_state rs_state;
 371    struct pipe_sampler_state sampler;
 372    unsigned i;
 373
 374    assert(idct);
 375
 376    memset(&rs_state, 0, sizeof(rs_state));
 377    rs_state.gl_rasterization_rules = true;
 378    idct->rs_state = idct->pipe->create_rasterizer_state(idct->pipe, &rs_state);
 379    if (!idct->rs_state)
 380       goto error_rs_state;
 381
 382    memset(&blend, 0, sizeof blend);
 383
 384    blend.independent_blend_enable = 0;
 385    blend.rt[0].blend_enable = 0;
 386    blend.rt[0].rgb_func = PIPE_BLEND_ADD;
 387    blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_ONE;
 388    blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ONE;
 389    blend.rt[0].alpha_func = PIPE_BLEND_ADD;
 390    blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_ONE;
 391    blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ONE;
 392    blend.logicop_enable = 0;
 393    blend.logicop_func = PIPE_LOGICOP_CLEAR;
 394    /* Needed to allow color writes to FB, even if blending disabled */
 395    blend.rt[0].colormask = PIPE_MASK_RGBA;
 396    blend.dither = 0;
 397    idct->blend = idct->pipe->create_blend_state(idct->pipe, &blend);
 398    if (!idct->blend)
 399       goto error_blend;
 400
 401    for (i = 0; i < 2; ++i) {
 402       memset(&sampler, 0, sizeof(sampler));
 403       sampler.wrap_s = PIPE_TEX_WRAP_REPEAT;
 404       sampler.wrap_t = PIPE_TEX_WRAP_REPEAT;
 405       sampler.wrap_r = PIPE_TEX_WRAP_REPEAT;
 406       sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
 407       sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
 408       sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
 409       sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
 410       sampler.compare_func = PIPE_FUNC_ALWAYS;
 411       sampler.normalized_coords = 1;
 412       idct->samplers[i] = idct->pipe->create_sampler_state(idct->pipe, &sampler);
 413       if (!idct->samplers[i])
 414          goto error_samplers;
 415    }
 416
 417    return true;
 418
 419 error_samplers:
 420    for (i = 0; i < 2; ++i)
 421       if (idct->samplers[i])
 422          idct->pipe->delete_sampler_state(idct->pipe, idct->samplers[i]);
 423
 424    idct->pipe->delete_rasterizer_state(idct->pipe, idct->rs_state);
 425
 426 error_blend:
 427    idct->pipe->delete_blend_state(idct->pipe, idct->blend);
 428
 429 error_rs_state:
 430    return false;
 431 }
 432
 433 static void
 434 cleanup_state(struct vl_idct *idct)
 435 {
 436    unsigned i;
 437
 438    for (i = 0; i < 2; ++i)
 439       idct->pipe->delete_sampler_state(idct->pipe, idct->samplers[i]);
 440
 441    idct->pipe->delete_rasterizer_state(idct->pipe, idct->rs_state);
 442    idct->pipe->delete_blend_state(idct->pipe, idct->blend);
 443 }
 444
 445 static bool
 446 init_intermediate(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 447 {
 448    struct pipe_resource *tex;
 449    struct pipe_surface surf_templ;
 450    unsigned i;
 451
 452    assert(idct && buffer);
 453
 454    tex = buffer->sampler_views.individual.intermediate->texture;
 455
 456    buffer->fb_state.width = tex->width0;
 457    buffer->fb_state.height = tex->height0;
 458    buffer->fb_state.nr_cbufs = idct->nr_of_render_targets;
 459    for(i = 0; i < idct->nr_of_render_targets; ++i) {
 460       memset(&surf_templ, 0, sizeof(surf_templ));
 461       surf_templ.format = tex->format;
 462       surf_templ.u.tex.first_layer = i;
 463       surf_templ.u.tex.last_layer = i;
 464       surf_templ.usage = PIPE_BIND_SAMPLER_VIEW | PIPE_BIND_RENDER_TARGET;
 465       buffer->fb_state.cbufs[i] = idct->pipe->create_surface(
 466          idct->pipe, tex, &surf_templ);
 467
 468       if (!buffer->fb_state.cbufs[i])
 469          goto error_surfaces;
 470    }
 471
 472    buffer->viewport.scale[0] = tex->width0;
 473    buffer->viewport.scale[1] = tex->height0;
 474
 475    return true;
 476
 477 error_surfaces:
 478    for(i = 0; i < idct->nr_of_render_targets; ++i)
 479       pipe_surface_reference(&buffer->fb_state.cbufs[i], NULL);
 480
 481    return false;
 482 }
 483
 484 static void
 485 cleanup_intermediate(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 486 {
 487    unsigned i;
 488
 489    assert(idct && buffer);
 490
 491    for(i = 0; i < idct->nr_of_render_targets; ++i)
 492       pipe_surface_reference(&buffer->fb_state.cbufs[i], NULL);
 493
 494    pipe_sampler_view_reference(&buffer->sampler_views.individual.intermediate, NULL);
 495 }
 496
 497 struct pipe_sampler_view *
 498 vl_idct_upload_matrix(struct pipe_context *pipe, float scale)
 499 {
 500    struct pipe_resource tex_templ, *matrix;
 501    struct pipe_sampler_view sv_templ, *sv;
 502    struct pipe_transfer *buf_transfer;
 503    unsigned i, j, pitch;
 504    float *f;
 505
 506    struct pipe_box rect =
 507    {
 508       0, 0, 0,
 509       BLOCK_WIDTH / 4,
 510       BLOCK_HEIGHT,
 511       1
 512    };
 513
 514    assert(pipe);
 515
 516    memset(&tex_templ, 0, sizeof(tex_templ));
 517    tex_templ.target = PIPE_TEXTURE_2D;
 518    tex_templ.format = PIPE_FORMAT_R32G32B32A32_FLOAT;
 519    tex_templ.last_level = 0;
 520    tex_templ.width0 = 2;
 521    tex_templ.height0 = 8;
 522    tex_templ.depth0 = 1;
 523    tex_templ.array_size = 1;
 524    tex_templ.usage = PIPE_USAGE_IMMUTABLE;
 525    tex_templ.bind = PIPE_BIND_SAMPLER_VIEW;
 526    tex_templ.flags = 0;
 527
 528    matrix = pipe->screen->resource_create(pipe->screen, &tex_templ);
 529    if (!matrix)
 530       goto error_matrix;
 531
 532    buf_transfer = pipe->get_transfer
 533    (
 534       pipe, matrix,
 535       0, PIPE_TRANSFER_WRITE | PIPE_TRANSFER_DISCARD,
 536       &rect
 537    );
 538    if (!buf_transfer)
 539       goto error_transfer;
 540
 541    pitch = buf_transfer->stride / sizeof(float);
 542
 543    f = pipe->transfer_map(pipe, buf_transfer);
 544    if (!f)
 545       goto error_map;
 546
 547    for(i = 0; i < BLOCK_HEIGHT; ++i)
 548       for(j = 0; j < BLOCK_WIDTH; ++j)
 549          // transpose and scale
 550          f[i * pitch + j] = ((const float (*)[8])const_matrix)[j][i] * scale;
 551
 552    pipe->transfer_unmap(pipe, buf_transfer);
 553    pipe->transfer_destroy(pipe, buf_transfer);
 554
 555    memset(&sv_templ, 0, sizeof(sv_templ));
 556    u_sampler_view_default_template(&sv_templ, matrix, matrix->format);
 557    sv = pipe->create_sampler_view(pipe, matrix, &sv_templ);
 558    pipe_resource_reference(&matrix, NULL);
 559    if (!sv)
 560       goto error_map;
 561
 562    return sv;
 563
 564 error_map:
 565    pipe->transfer_destroy(pipe, buf_transfer);
 566
 567 error_transfer:
 568    pipe_resource_reference(&matrix, NULL);
 569
 570 error_matrix:
 571    return NULL;
 572 }
 573
 574 bool vl_idct_init(struct vl_idct *idct, struct pipe_context *pipe,
 575                   unsigned buffer_width, unsigned buffer_height,
 576                   unsigned nr_of_render_targets,
 577                   struct pipe_sampler_view *matrix,
 578                   struct pipe_sampler_view *transpose)
 579 {
 580    assert(idct && pipe && matrix);
 581
 582    idct->pipe = pipe;
 583    idct->buffer_width = buffer_width;
 584    idct->buffer_height = buffer_height;
 585    idct->nr_of_render_targets = nr_of_render_targets;
 586
 587    pipe_sampler_view_reference(&idct->matrix, matrix);
 588    pipe_sampler_view_reference(&idct->transpose, transpose);
 589
 590    if(!init_shaders(idct))
 591       return false;
 592
 593    if(!init_state(idct)) {
 594       cleanup_shaders(idct);
 595       return false;
 596    }
 597
 598    return true;
 599 }
 600
 601 void
 602 vl_idct_cleanup(struct vl_idct *idct)
 603 {
 604    cleanup_shaders(idct);
 605    cleanup_state(idct);
 606
 607    pipe_sampler_view_reference(&idct->matrix, NULL);
 608 }
 609
 610 bool
 611 vl_idct_init_buffer(struct vl_idct *idct, struct vl_idct_buffer *buffer,
 612                     struct pipe_sampler_view *source,
 613                     struct pipe_sampler_view *intermediate,
 614                     struct pipe_surface *destination)
 615 {
 616    assert(buffer);
 617    assert(idct);
 618    assert(source);
 619    assert(destination);
 620
 621    memset(buffer, 0, sizeof(struct vl_idct_buffer));
 622
 623    pipe_sampler_view_reference(&buffer->sampler_views.individual.matrix, idct->matrix);
 624    pipe_sampler_view_reference(&buffer->sampler_views.individual.source, source);
 625    pipe_sampler_view_reference(&buffer->sampler_views.individual.transpose, idct->transpose);
 626    pipe_sampler_view_reference(&buffer->sampler_views.individual.intermediate, intermediate);
 627
 628    if (!init_intermediate(idct, buffer))
 629       return false;
 630
 631    buffer->viewport.scale[2] = 1;
 632    buffer->viewport.scale[3] = 1;
 633    buffer->viewport.translate[0] = 0;
 634    buffer->viewport.translate[1] = 0;
 635    buffer->viewport.translate[2] = 0;
 636    buffer->viewport.translate[3] = 0;
 637
 638    return true;
 639 }
 640
 641 void
 642 vl_idct_cleanup_buffer(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 643 {
 644    unsigned i;
 645
 646    assert(idct && buffer);
 647
 648    for(i = 0; i < idct->nr_of_render_targets; ++i)
 649       pipe_surface_reference(&buffer->fb_state.cbufs[i], NULL);
 650
 651    cleanup_intermediate(idct, buffer);
 652 }
 653
 654 void
 655 vl_idct_flush(struct vl_idct *idct, struct vl_idct_buffer *buffer, unsigned num_instances)
 656 {
 657    assert(idct);
 658    assert(buffer);
 659
 660    idct->pipe->bind_rasterizer_state(idct->pipe, idct->rs_state);
 661    idct->pipe->bind_blend_state(idct->pipe, idct->blend);
 662    idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers);
 663
 664    /* first stage */
 665    idct->pipe->set_framebuffer_state(idct->pipe, &buffer->fb_state);
 666    idct->pipe->set_viewport_state(idct->pipe, &buffer->viewport);
 667    idct->pipe->set_fragment_sampler_views(idct->pipe, 2, buffer->sampler_views.stage[0]);
 668    idct->pipe->bind_vs_state(idct->pipe, idct->vs);
 669    idct->pipe->bind_fs_state(idct->pipe, idct->fs);
 670    util_draw_arrays_instanced(idct->pipe, PIPE_PRIM_QUADS, 0, 4, 0, num_instances);
 671 }
 672
 673 void
 674 vl_idct_prepare_stage2(struct vl_idct *idct, struct vl_idct_buffer *buffer)
 675 {
 676    assert(idct);
 677    assert(buffer);
 678
 679    /* second stage */
 680    idct->pipe->bind_rasterizer_state(idct->pipe, idct->rs_state);
 681    idct->pipe->bind_fragment_sampler_states(idct->pipe, 2, idct->samplers);
 682    idct->pipe->set_fragment_sampler_views(idct->pipe, 2, buffer->sampler_views.stage[1]);
 683 }
 684