src/gallium/auxiliary/vl/vl_mc.c

   1 /**************************************************************************
   2  *
   3  * Copyright 2009 Younes Manton.
   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
  32 #include <util/u_sampler.h>
  33 #include <util/u_draw.h>
  34
  35 #include <tgsi/tgsi_ureg.h>
  36
  37 #include "vl_defines.h"
  38 #include "vl_vertex_buffers.h"
  39 #include "vl_mc.h"
  40 #include "vl_idct.h"
  41
  42 enum VS_OUTPUT
  43 {
  44    VS_O_VPOS,
  45    VS_O_VTOP,
  46    VS_O_VBOTTOM,
  47
  48    VS_O_FLAGS = VS_O_VTOP,
  49    VS_O_VTEX = VS_O_VBOTTOM
  50 };
  51
  52 static struct ureg_dst
  53 calc_position(struct vl_mc *r, struct ureg_program *shader, struct ureg_src block_scale)
  54 {
  55    struct ureg_src vrect, vpos;
  56    struct ureg_dst t_vpos;
  57    struct ureg_dst o_vpos;
  58
  59    vrect = ureg_DECL_vs_input(shader, VS_I_RECT);
  60    vpos = ureg_DECL_vs_input(shader, VS_I_VPOS);
  61
  62    t_vpos = ureg_DECL_temporary(shader);
  63
  64    o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
  65
  66    /*
  67     * block_scale = (MACROBLOCK_WIDTH, MACROBLOCK_HEIGHT) / (dst.width, dst.height)
  68     *
  69     * t_vpos = (vpos + vrect) * block_scale
  70     * o_vpos.xy = t_vpos
  71     * o_vpos.zw = vpos
  72     */
  73    ureg_ADD(shader, ureg_writemask(t_vpos, TGSI_WRITEMASK_XY), vpos, vrect);
  74    ureg_MUL(shader, ureg_writemask(t_vpos, TGSI_WRITEMASK_XY), ureg_src(t_vpos), block_scale);
  75    ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_XY), ureg_src(t_vpos));
  76    ureg_MOV(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_ZW), ureg_imm1f(shader, 1.0f));
  77
  78    return t_vpos;
  79 }
  80
  81 static struct ureg_dst
  82 calc_line(struct ureg_program *shader)
  83 {
  84    struct ureg_dst tmp;
  85    struct ureg_src pos;
  86
  87    tmp = ureg_DECL_temporary(shader);
  88
  89    pos = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS, TGSI_INTERPOLATE_LINEAR);
  90
  91    /*
  92     * tmp.y = fraction(pos.y / 2) >= 0.5 ? 1 : 0
  93     */
  94    ureg_MUL(shader, ureg_writemask(tmp, TGSI_WRITEMASK_Y), pos, ureg_imm1f(shader, 0.5f));
  95    ureg_FRC(shader, ureg_writemask(tmp, TGSI_WRITEMASK_Y), ureg_src(tmp));
  96    ureg_SGE(shader, ureg_writemask(tmp, TGSI_WRITEMASK_Y), ureg_src(tmp), ureg_imm1f(shader, 0.5f));
  97
  98    return tmp;
  99 }
 100
 101 static void *
 102 create_ref_vert_shader(struct vl_mc *r)
 103 {
 104    struct ureg_program *shader;
 105    struct ureg_src mv_scale;
 106    struct ureg_src vrect, vmv[2];
 107    struct ureg_dst t_vpos;
 108    struct ureg_dst o_vpos, o_vmv[2];
 109    unsigned i;
 110
 111    shader = ureg_create(TGSI_PROCESSOR_VERTEX);
 112    if (!shader)
 113       return NULL;
 114
 115    vrect = ureg_DECL_vs_input(shader, VS_I_RECT);
 116    vmv[0] = ureg_DECL_vs_input(shader, VS_I_MV_TOP);
 117    vmv[1] = ureg_DECL_vs_input(shader, VS_I_MV_BOTTOM);
 118
 119    t_vpos = calc_position(r, shader, ureg_imm2f(shader,
 120       (float)MACROBLOCK_WIDTH / r->buffer_width,
 121       (float)MACROBLOCK_HEIGHT / r->buffer_height)
 122    );
 123
 124    o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
 125    o_vmv[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_VTOP);
 126    o_vmv[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_VBOTTOM);
 127
 128    /*
 129     * mv_scale.xy = 0.5 / (dst.width, dst.height);
 130     * mv_scale.z = 1.0f / 4.0f
 131     * mv_scale.w = 1.0f / 255.0f
 132     *
 133     * // Apply motion vectors
 134     * o_vmv[0..1].xy = vmv[0..1] * mv_scale + t_vpos
 135     * o_vmv[0..1].zw = vmv[0..1] * mv_scale
 136     *
 137     */
 138
 139    mv_scale = ureg_imm4f(shader,
 140       0.5f / r->buffer_width,
 141       0.5f / r->buffer_height,
 142       1.0f / 4.0f,
 143       1.0f / PIPE_VIDEO_MV_WEIGHT_MAX);
 144
 145    for (i = 0; i < 2; ++i) {
 146       ureg_MAD(shader, ureg_writemask(o_vmv[i], TGSI_WRITEMASK_XY), mv_scale, vmv[i], ureg_src(t_vpos));
 147       ureg_MUL(shader, ureg_writemask(o_vmv[i], TGSI_WRITEMASK_ZW), mv_scale, vmv[i]);
 148    }
 149
 150    ureg_release_temporary(shader, t_vpos);
 151
 152    ureg_END(shader);
 153
 154    return ureg_create_shader_and_destroy(shader, r->pipe);
 155 }
 156
 157 static void *
 158 create_ref_frag_shader(struct vl_mc *r)
 159 {
 160    const float y_scale =
 161       r->buffer_height / 2 *
 162       r->macroblock_size / MACROBLOCK_HEIGHT;
 163
 164    struct ureg_program *shader;
 165    struct ureg_src tc[2], sampler;
 166    struct ureg_dst ref, field;
 167    struct ureg_dst fragment;
 168    unsigned label;
 169
 170    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 171    if (!shader)
 172       return NULL;
 173
 174    tc[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_VTOP, TGSI_INTERPOLATE_LINEAR);
 175    tc[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_VBOTTOM, TGSI_INTERPOLATE_LINEAR);
 176
 177    sampler = ureg_DECL_sampler(shader, 0);
 178    ref = ureg_DECL_temporary(shader);
 179
 180    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 181
 182    field = calc_line(shader);
 183
 184    /*
 185     * ref = field.z ? tc[1] : tc[0]
 186     *
 187     * // Adjust tc acording to top/bottom field selection
 188     * if (|ref.z|) {
 189     *    ref.y *= y_scale
 190     *    ref.y = floor(ref.y)
 191     *    ref.y += ref.z
 192     *    ref.y /= y_scale
 193     * }
 194     * fragment.xyz = tex(ref, sampler[0])
 195     */
 196    ureg_CMP(shader, ureg_writemask(ref, TGSI_WRITEMASK_XYZ),
 197             ureg_negate(ureg_scalar(ureg_src(field), TGSI_SWIZZLE_Y)),
 198             tc[1], tc[0]);
 199    ureg_CMP(shader, ureg_writemask(fragment, TGSI_WRITEMASK_W),
 200             ureg_negate(ureg_scalar(ureg_src(field), TGSI_SWIZZLE_Y)),
 201             tc[1], tc[0]);
 202
 203    ureg_IF(shader, ureg_scalar(ureg_src(ref), TGSI_SWIZZLE_Z), &label);
 204
 205       ureg_MUL(shader, ureg_writemask(ref, TGSI_WRITEMASK_Y),
 206                ureg_src(ref), ureg_imm1f(shader, y_scale));
 207       ureg_FLR(shader, ureg_writemask(ref, TGSI_WRITEMASK_Y), ureg_src(ref));
 208       ureg_ADD(shader, ureg_writemask(ref, TGSI_WRITEMASK_Y),
 209                ureg_src(ref), ureg_scalar(ureg_src(ref), TGSI_SWIZZLE_Z));
 210       ureg_MUL(shader, ureg_writemask(ref, TGSI_WRITEMASK_Y),
 211                ureg_src(ref), ureg_imm1f(shader, 1.0f / y_scale));
 212
 213    ureg_fixup_label(shader, label, ureg_get_instruction_number(shader));
 214    ureg_ENDIF(shader);
 215
 216    ureg_TEX(shader, ureg_writemask(fragment, TGSI_WRITEMASK_XYZ), TGSI_TEXTURE_2D, ureg_src(ref), sampler);
 217
 218    ureg_release_temporary(shader, ref);
 219
 220    ureg_release_temporary(shader, field);
 221    ureg_END(shader);
 222
 223    return ureg_create_shader_and_destroy(shader, r->pipe);
 224 }
 225
 226 static void *
 227 create_ycbcr_vert_shader(struct vl_mc *r, vl_mc_ycbcr_vert_shader vs_callback, void *callback_priv)
 228 {
 229    struct ureg_program *shader;
 230
 231    struct ureg_src vrect, vpos;
 232    struct ureg_dst t_vpos, t_vtex;
 233    struct ureg_dst o_vpos, o_flags;
 234
 235    struct vertex2f scale = {
 236       (float)BLOCK_WIDTH / r->buffer_width * MACROBLOCK_WIDTH / r->macroblock_size,
 237       (float)BLOCK_HEIGHT / r->buffer_height * MACROBLOCK_HEIGHT / r->macroblock_size
 238    };
 239
 240    unsigned label;
 241
 242    shader = ureg_create(TGSI_PROCESSOR_VERTEX);
 243    if (!shader)
 244       return NULL;
 245
 246    vrect = ureg_DECL_vs_input(shader, VS_I_RECT);
 247    vpos = ureg_DECL_vs_input(shader, VS_I_VPOS);
 248
 249    t_vpos = calc_position(r, shader, ureg_imm2f(shader, scale.x, scale.y));
 250    t_vtex = ureg_DECL_temporary(shader);
 251
 252    o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
 253    o_flags = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_FLAGS);
 254
 255    /*
 256     * o_vtex.xy = t_vpos
 257     * o_flags.z = intra * 0.5
 258     *
 259     * if(interlaced) {
 260     *    t_vtex.xy = vrect.y ? { 0, scale.y } : { -scale.y : 0 }
 261     *    t_vtex.z = vpos.y % 2
 262     *    t_vtex.y = t_vtex.z ? t_vtex.x : t_vtex.y
 263     *    o_vpos.y = t_vtex.y + t_vpos.y
 264     *
 265     *    o_flags.w = t_vtex.z ? 0 : 1
 266     * }
 267     *
 268     */
 269
 270    vs_callback(callback_priv, r, shader, VS_O_VTEX, t_vpos);
 271
 272    ureg_MUL(shader, ureg_writemask(o_flags, TGSI_WRITEMASK_Z),
 273             ureg_scalar(vpos, TGSI_SWIZZLE_Z), ureg_imm1f(shader, 0.5f));
 274    ureg_MOV(shader, ureg_writemask(o_flags, TGSI_WRITEMASK_W), ureg_imm1f(shader, -1.0f));
 275
 276    if (r->macroblock_size == MACROBLOCK_HEIGHT) { //TODO
 277       ureg_IF(shader, ureg_scalar(vpos, TGSI_SWIZZLE_W), &label);
 278
 279          ureg_CMP(shader, ureg_writemask(t_vtex, TGSI_WRITEMASK_XY),
 280                   ureg_negate(ureg_scalar(vrect, TGSI_SWIZZLE_Y)),
 281                   ureg_imm2f(shader, 0.0f, scale.y),
 282                   ureg_imm2f(shader, -scale.y, 0.0f));
 283          ureg_MUL(shader, ureg_writemask(t_vtex, TGSI_WRITEMASK_Z),
 284                   ureg_scalar(vpos, TGSI_SWIZZLE_Y), ureg_imm1f(shader, 0.5f));
 285
 286          ureg_FRC(shader, ureg_writemask(t_vtex, TGSI_WRITEMASK_Z), ureg_src(t_vtex));
 287
 288          ureg_CMP(shader, ureg_writemask(t_vtex, TGSI_WRITEMASK_Y),
 289                   ureg_negate(ureg_scalar(ureg_src(t_vtex), TGSI_SWIZZLE_Z)),
 290                   ureg_scalar(ureg_src(t_vtex), TGSI_SWIZZLE_X),
 291                   ureg_scalar(ureg_src(t_vtex), TGSI_SWIZZLE_Y));
 292          ureg_ADD(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_Y),
 293                   ureg_src(t_vpos), ureg_src(t_vtex));
 294
 295          ureg_CMP(shader, ureg_writemask(o_flags, TGSI_WRITEMASK_W),
 296                   ureg_negate(ureg_scalar(ureg_src(t_vtex), TGSI_SWIZZLE_Z)),
 297                   ureg_imm1f(shader, 0.0f), ureg_imm1f(shader, 1.0f));
 298
 299       ureg_fixup_label(shader, label, ureg_get_instruction_number(shader));
 300       ureg_ENDIF(shader);
 301    }
 302
 303    ureg_release_temporary(shader, t_vtex);
 304    ureg_release_temporary(shader, t_vpos);
 305
 306    ureg_END(shader);
 307
 308    return ureg_create_shader_and_destroy(shader, r->pipe);
 309 }
 310
 311 static void *
 312 create_ycbcr_frag_shader(struct vl_mc *r, float scale, bool invert,
 313                          vl_mc_ycbcr_frag_shader fs_callback, void *callback_priv)
 314 {
 315    struct ureg_program *shader;
 316    struct ureg_src flags;
 317    struct ureg_dst tmp;
 318    struct ureg_dst fragment;
 319    unsigned label;
 320
 321    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 322    if (!shader)
 323       return NULL;
 324
 325    flags = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_FLAGS, TGSI_INTERPOLATE_LINEAR);
 326
 327    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 328
 329    tmp = calc_line(shader);
 330
 331    /*
 332     * if (field == tc.w)
 333     *    kill();
 334     * else {
 335     *    fragment.xyz  = tex(tc, sampler) * scale + tc.z
 336     *    fragment.w = 1.0f
 337     * }
 338     */
 339
 340    ureg_SEQ(shader, ureg_writemask(tmp, TGSI_WRITEMASK_Y),
 341             ureg_scalar(flags, TGSI_SWIZZLE_W), ureg_src(tmp));
 342
 343    ureg_IF(shader, ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_Y), &label);
 344
 345       ureg_KILP(shader);
 346
 347    ureg_fixup_label(shader, label, ureg_get_instruction_number(shader));
 348    ureg_ELSE(shader, &label);
 349
 350       fs_callback(callback_priv, r, shader, VS_O_VTEX, tmp);
 351
 352       if (scale != 1.0f)
 353          ureg_MAD(shader, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ),
 354                   ureg_src(tmp), ureg_imm1f(shader, scale),
 355                   ureg_scalar(flags, TGSI_SWIZZLE_Z));
 356       else
 357          ureg_ADD(shader, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ),
 358                   ureg_src(tmp), ureg_scalar(flags, TGSI_SWIZZLE_Z));
 359
 360       ureg_MUL(shader, ureg_writemask(fragment, TGSI_WRITEMASK_XYZ), ureg_src(tmp), ureg_imm1f(shader, invert ? -1.0f : 1.0f));
 361       ureg_MOV(shader, ureg_writemask(fragment, TGSI_WRITEMASK_W), ureg_imm1f(shader, 1.0f));
 362
 363    ureg_fixup_label(shader, label, ureg_get_instruction_number(shader));
 364    ureg_ENDIF(shader);
 365
 366    ureg_release_temporary(shader, tmp);
 367
 368    ureg_END(shader);
 369
 370    return ureg_create_shader_and_destroy(shader, r->pipe);
 371 }
 372
 373 static bool
 374 init_pipe_state(struct vl_mc *r)
 375 {
 376    struct pipe_sampler_state sampler;
 377    struct pipe_blend_state blend;
 378    struct pipe_rasterizer_state rs_state;
 379    unsigned i;
 380
 381    assert(r);
 382
 383    memset(&sampler, 0, sizeof(sampler));
 384    sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 385    sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 386    sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_BORDER;
 387    sampler.min_img_filter = PIPE_TEX_FILTER_LINEAR;
 388    sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
 389    sampler.mag_img_filter = PIPE_TEX_FILTER_LINEAR;
 390    sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
 391    sampler.compare_func = PIPE_FUNC_ALWAYS;
 392    sampler.normalized_coords = 1;
 393    r->sampler_ref = r->pipe->create_sampler_state(r->pipe, &sampler);
 394    if (!r->sampler_ref)
 395       goto error_sampler_ref;
 396
 397    for (i = 0; i < VL_MC_NUM_BLENDERS; ++i) {
 398       memset(&blend, 0, sizeof blend);
 399       blend.independent_blend_enable = 0;
 400       blend.rt[0].blend_enable = 1;
 401       blend.rt[0].rgb_func = PIPE_BLEND_ADD;
 402       blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_SRC_ALPHA;
 403       blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
 404       blend.rt[0].alpha_func = PIPE_BLEND_ADD;
 405       blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_SRC_ALPHA;
 406       blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
 407       blend.logicop_enable = 0;
 408       blend.logicop_func = PIPE_LOGICOP_CLEAR;
 409       blend.rt[0].colormask = i;
 410       blend.dither = 0;
 411       r->blend_clear[i] = r->pipe->create_blend_state(r->pipe, &blend);
 412       if (!r->blend_clear[i])
 413          goto error_blend;
 414
 415       blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ONE;
 416       blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ONE;
 417       r->blend_add[i] = r->pipe->create_blend_state(r->pipe, &blend);
 418       if (!r->blend_add[i])
 419          goto error_blend;
 420
 421       blend.rt[0].rgb_func = PIPE_BLEND_REVERSE_SUBTRACT;
 422       blend.rt[0].alpha_dst_factor = PIPE_BLEND_REVERSE_SUBTRACT;
 423       r->blend_sub[i] = r->pipe->create_blend_state(r->pipe, &blend);
 424       if (!r->blend_sub[i])
 425          goto error_blend;
 426    }
 427
 428    memset(&rs_state, 0, sizeof(rs_state));
 429    /*rs_state.sprite_coord_enable */
 430    rs_state.sprite_coord_mode = PIPE_SPRITE_COORD_UPPER_LEFT;
 431    rs_state.point_quad_rasterization = true;
 432    rs_state.point_size = BLOCK_WIDTH;
 433    rs_state.gl_rasterization_rules = true;
 434    r->rs_state = r->pipe->create_rasterizer_state(r->pipe, &rs_state);
 435    if (!r->rs_state)
 436       goto error_rs_state;
 437
 438    return true;
 439
 440 error_rs_state:
 441 error_blend:
 442    for (i = 0; i < VL_MC_NUM_BLENDERS; ++i) {
 443       if (r->blend_sub[i])
 444          r->pipe->delete_blend_state(r->pipe, r->blend_sub[i]);
 445
 446       if (r->blend_add[i])
 447          r->pipe->delete_blend_state(r->pipe, r->blend_add[i]);
 448
 449       if (r->blend_clear[i])
 450          r->pipe->delete_blend_state(r->pipe, r->blend_clear[i]);
 451    }
 452
 453    r->pipe->delete_sampler_state(r->pipe, r->sampler_ref);
 454
 455 error_sampler_ref:
 456    return false;
 457 }
 458
 459 static void
 460 cleanup_pipe_state(struct vl_mc *r)
 461 {
 462    unsigned i;
 463
 464    assert(r);
 465
 466    r->pipe->delete_sampler_state(r->pipe, r->sampler_ref);
 467    for (i = 0; i < VL_MC_NUM_BLENDERS; ++i) {
 468       r->pipe->delete_blend_state(r->pipe, r->blend_clear[i]);
 469       r->pipe->delete_blend_state(r->pipe, r->blend_add[i]);
 470       r->pipe->delete_blend_state(r->pipe, r->blend_sub[i]);
 471    }
 472    r->pipe->delete_rasterizer_state(r->pipe, r->rs_state);
 473 }
 474
 475 bool
 476 vl_mc_init(struct vl_mc *renderer, struct pipe_context *pipe,
 477            unsigned buffer_width, unsigned buffer_height,
 478            unsigned macroblock_size, float scale,
 479            vl_mc_ycbcr_vert_shader vs_callback,
 480            vl_mc_ycbcr_frag_shader fs_callback,
 481            void *callback_priv)
 482 {
 483    assert(renderer);
 484    assert(pipe);
 485
 486    memset(renderer, 0, sizeof(struct vl_mc));
 487
 488    renderer->pipe = pipe;
 489    renderer->buffer_width = buffer_width;
 490    renderer->buffer_height = buffer_height;
 491    renderer->macroblock_size = macroblock_size;
 492
 493    if (!init_pipe_state(renderer))
 494       goto error_pipe_state;
 495
 496    renderer->vs_ref = create_ref_vert_shader(renderer);
 497    if (!renderer->vs_ref)
 498       goto error_vs_ref;
 499
 500    renderer->vs_ycbcr = create_ycbcr_vert_shader(renderer, vs_callback, callback_priv);
 501    if (!renderer->vs_ycbcr)
 502       goto error_vs_ycbcr;
 503
 504    renderer->fs_ref = create_ref_frag_shader(renderer);
 505    if (!renderer->fs_ref)
 506       goto error_fs_ref;
 507
 508    renderer->fs_ycbcr = create_ycbcr_frag_shader(renderer, scale, false, fs_callback, callback_priv);
 509    if (!renderer->fs_ycbcr)
 510       goto error_fs_ycbcr;
 511
 512    renderer->fs_ycbcr_sub = create_ycbcr_frag_shader(renderer, scale, true, fs_callback, callback_priv);
 513    if (!renderer->fs_ycbcr_sub)
 514       goto error_fs_ycbcr_sub;
 515
 516    return true;
 517
 518 error_fs_ycbcr_sub:
 519    renderer->pipe->delete_fs_state(renderer->pipe, renderer->fs_ycbcr);
 520
 521 error_fs_ycbcr:
 522    renderer->pipe->delete_fs_state(renderer->pipe, renderer->fs_ref);
 523
 524 error_fs_ref:
 525    renderer->pipe->delete_vs_state(renderer->pipe, renderer->vs_ycbcr);
 526
 527 error_vs_ycbcr:
 528    renderer->pipe->delete_vs_state(renderer->pipe, renderer->vs_ref);
 529
 530 error_vs_ref:
 531    cleanup_pipe_state(renderer);
 532
 533 error_pipe_state:
 534    return false;
 535 }
 536
 537 void
 538 vl_mc_cleanup(struct vl_mc *renderer)
 539 {
 540    assert(renderer);
 541
 542    cleanup_pipe_state(renderer);
 543
 544    renderer->pipe->delete_vs_state(renderer->pipe, renderer->vs_ref);
 545    renderer->pipe->delete_vs_state(renderer->pipe, renderer->vs_ycbcr);
 546    renderer->pipe->delete_fs_state(renderer->pipe, renderer->fs_ref);
 547    renderer->pipe->delete_fs_state(renderer->pipe, renderer->fs_ycbcr);
 548    renderer->pipe->delete_fs_state(renderer->pipe, renderer->fs_ycbcr_sub);
 549 }
 550
 551 bool
 552 vl_mc_init_buffer(struct vl_mc *renderer, struct vl_mc_buffer *buffer)
 553 {
 554    assert(renderer && buffer);
 555
 556    buffer->renderer = renderer;
 557
 558    buffer->viewport.scale[2] = 1;
 559    buffer->viewport.scale[3] = 1;
 560    buffer->viewport.translate[0] = 0;
 561    buffer->viewport.translate[1] = 0;
 562    buffer->viewport.translate[2] = 0;
 563    buffer->viewport.translate[3] = 0;
 564
 565    buffer->fb_state.nr_cbufs = 1;
 566    buffer->fb_state.zsbuf = NULL;
 567
 568    return true;
 569 }
 570
 571 void
 572 vl_mc_cleanup_buffer(struct vl_mc_buffer *buffer)
 573 {
 574    assert(buffer);
 575 }
 576
 577 void
 578 vl_mc_set_surface(struct vl_mc_buffer *buffer, struct pipe_surface *surface)
 579 {
 580    assert(buffer && surface);
 581
 582    buffer->surface_cleared = false;
 583
 584    buffer->viewport.scale[0] = surface->width;
 585    buffer->viewport.scale[1] = surface->height;
 586
 587    buffer->fb_state.width = surface->width;
 588    buffer->fb_state.height = surface->height;
 589    buffer->fb_state.cbufs[0] = surface;
 590 }
 591
 592 static void
 593 prepare_pipe_4_rendering(struct vl_mc_buffer *buffer, unsigned mask)
 594 {
 595    struct vl_mc *renderer;
 596
 597    assert(buffer);
 598
 599    renderer = buffer->renderer;
 600    renderer->pipe->bind_rasterizer_state(renderer->pipe, renderer->rs_state);
 601
 602    if (buffer->surface_cleared)
 603       renderer->pipe->bind_blend_state(renderer->pipe, renderer->blend_add[mask]);
 604    else
 605       renderer->pipe->bind_blend_state(renderer->pipe, renderer->blend_clear[mask]);
 606
 607    renderer->pipe->set_framebuffer_state(renderer->pipe, &buffer->fb_state);
 608    renderer->pipe->set_viewport_state(renderer->pipe, &buffer->viewport);
 609 }
 610
 611 void
 612 vl_mc_render_ref(struct vl_mc_buffer *buffer, struct pipe_sampler_view *ref)
 613 {
 614    struct vl_mc *renderer;
 615
 616    assert(buffer && ref);
 617
 618    prepare_pipe_4_rendering(buffer, PIPE_MASK_R | PIPE_MASK_G | PIPE_MASK_B);
 619
 620    renderer = buffer->renderer;
 621
 622    renderer->pipe->bind_vs_state(renderer->pipe, renderer->vs_ref);
 623    renderer->pipe->bind_fs_state(renderer->pipe, renderer->fs_ref);
 624
 625    renderer->pipe->set_fragment_sampler_views(renderer->pipe, 1, &ref);
 626    renderer->pipe->bind_fragment_sampler_states(renderer->pipe, 1, &renderer->sampler_ref);
 627
 628    util_draw_arrays_instanced(renderer->pipe, PIPE_PRIM_QUADS, 0, 4, 0,
 629                               renderer->buffer_width / MACROBLOCK_WIDTH *
 630                               renderer->buffer_height / MACROBLOCK_HEIGHT);
 631
 632    buffer->surface_cleared = true;
 633 }
 634
 635 void
 636 vl_mc_render_ycbcr(struct vl_mc_buffer *buffer, unsigned component, unsigned num_instances)
 637 {
 638    struct vl_mc *renderer;
 639    unsigned mask = 1 << component;
 640
 641    assert(buffer);
 642
 643    if (num_instances == 0)
 644       return;
 645
 646    prepare_pipe_4_rendering(buffer, mask);
 647
 648    renderer = buffer->renderer;
 649
 650    renderer->pipe->bind_vs_state(renderer->pipe, renderer->vs_ycbcr);
 651    renderer->pipe->bind_fs_state(renderer->pipe, renderer->fs_ycbcr);
 652
 653    util_draw_arrays_instanced(renderer->pipe, PIPE_PRIM_QUADS, 0, 4, 0, num_instances);
 654
 655    if (buffer->surface_cleared) {
 656       renderer->pipe->bind_blend_state(renderer->pipe, renderer->blend_sub[mask]);
 657       renderer->pipe->bind_fs_state(renderer->pipe, renderer->fs_ycbcr_sub);
 658       util_draw_arrays_instanced(renderer->pipe, PIPE_PRIM_QUADS, 0, 4, 0, num_instances);
 659    }
 660 }