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 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    vmv[0] = ureg_DECL_vs_input(shader, VS_I_MV_TOP);
 116    vmv[1] = ureg_DECL_vs_input(shader, VS_I_MV_BOTTOM);
 117
 118    t_vpos = calc_position(r, shader, ureg_imm2f(shader,
 119       (float)MACROBLOCK_WIDTH / r->buffer_width,
 120       (float)MACROBLOCK_HEIGHT / r->buffer_height)
 121    );
 122
 123    /* XXX The position is not written, which may lead to undefined rendering.
 124     * XXX This is a serious bug. */
 125    o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
 126    o_vmv[0] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_VTOP);
 127    o_vmv[1] = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_VBOTTOM);
 128
 129    /*
 130     * mv_scale.xy = 0.5 / (dst.width, dst.height);
 131     * mv_scale.z = 1.0f / 4.0f
 132     * mv_scale.w = 1.0f / 255.0f
 133     *
 134     * // Apply motion vectors
 135     * o_vmv[0..1].xy = vmv[0..1] * mv_scale + t_vpos
 136     * o_vmv[0..1].zw = vmv[0..1] * mv_scale
 137     *
 138     */
 139
 140    mv_scale = ureg_imm4f(shader,
 141       0.5f / r->buffer_width,
 142       0.5f / r->buffer_height,
 143       1.0f / 4.0f,
 144       1.0f / PIPE_VIDEO_MV_WEIGHT_MAX);
 145
 146    for (i = 0; i < 2; ++i) {
 147       ureg_MAD(shader, ureg_writemask(o_vmv[i], TGSI_WRITEMASK_XY), mv_scale, vmv[i], ureg_src(t_vpos));
 148       ureg_MUL(shader, ureg_writemask(o_vmv[i], TGSI_WRITEMASK_ZW), mv_scale, vmv[i]);
 149    }
 150
 151    ureg_release_temporary(shader, t_vpos);
 152
 153    ureg_END(shader);
 154
 155    return ureg_create_shader_and_destroy(shader, r->pipe);
 156 }
 157
 158 static void *
 159 create_ref_frag_shader(struct vl_mc *r)
 160 {
 161    const float y_scale =
 162       r->buffer_height / 2 *
 163       r->macroblock_size / MACROBLOCK_HEIGHT;
 164
 165    struct ureg_program *shader;
 166    struct ureg_src tc[2], sampler;
 167    struct ureg_dst ref, field;
 168    struct ureg_dst fragment;
 169    unsigned label;
 170
 171    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 172    if (!shader)
 173       return NULL;
 174
 175    tc[0] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_VTOP, TGSI_INTERPOLATE_LINEAR);
 176    tc[1] = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_VBOTTOM, TGSI_INTERPOLATE_LINEAR);
 177
 178    sampler = ureg_DECL_sampler(shader, 0);
 179    ref = ureg_DECL_temporary(shader);
 180
 181    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 182
 183    field = calc_line(shader);
 184
 185    /*
 186     * ref = field.z ? tc[1] : tc[0]
 187     *
 188     * // Adjust tc acording to top/bottom field selection
 189     * if (|ref.z|) {
 190     *    ref.y *= y_scale
 191     *    ref.y = floor(ref.y)
 192     *    ref.y += ref.z
 193     *    ref.y /= y_scale
 194     * }
 195     * fragment.xyz = tex(ref, sampler[0])
 196     */
 197    ureg_CMP(shader, ureg_writemask(ref, TGSI_WRITEMASK_XYZ),
 198             ureg_negate(ureg_scalar(ureg_src(field), TGSI_SWIZZLE_Y)),
 199             tc[1], tc[0]);
 200    ureg_CMP(shader, ureg_writemask(fragment, TGSI_WRITEMASK_W),
 201             ureg_negate(ureg_scalar(ureg_src(field), TGSI_SWIZZLE_Y)),
 202             tc[1], tc[0]);
 203
 204    ureg_IF(shader, ureg_scalar(ureg_src(ref), TGSI_SWIZZLE_Z), &label);
 205
 206       ureg_MUL(shader, ureg_writemask(ref, TGSI_WRITEMASK_Y),
 207                ureg_src(ref), ureg_imm1f(shader, y_scale));
 208       ureg_FLR(shader, ureg_writemask(ref, TGSI_WRITEMASK_Y), ureg_src(ref));
 209       ureg_ADD(shader, ureg_writemask(ref, TGSI_WRITEMASK_Y),
 210                ureg_src(ref), ureg_scalar(ureg_src(ref), TGSI_SWIZZLE_Z));
 211       ureg_MUL(shader, ureg_writemask(ref, TGSI_WRITEMASK_Y),
 212                ureg_src(ref), ureg_imm1f(shader, 1.0f / y_scale));
 213
 214    ureg_fixup_label(shader, label, ureg_get_instruction_number(shader));
 215    ureg_ENDIF(shader);
 216
 217    ureg_TEX(shader, ureg_writemask(fragment, TGSI_WRITEMASK_XYZ), TGSI_TEXTURE_2D, ureg_src(ref), sampler);
 218
 219    ureg_release_temporary(shader, ref);
 220
 221    ureg_release_temporary(shader, field);
 222    ureg_END(shader);
 223
 224    return ureg_create_shader_and_destroy(shader, r->pipe);
 225 }
 226
 227 static void *
 228 create_ycbcr_vert_shader(struct vl_mc *r, vl_mc_ycbcr_vert_shader vs_callback, void *callback_priv)
 229 {
 230    struct ureg_program *shader;
 231
 232    struct ureg_src vrect, vpos;
 233    struct ureg_dst t_vpos, t_vtex;
 234    struct ureg_dst o_vpos, o_flags;
 235
 236    struct vertex2f scale = {
 237       (float)BLOCK_WIDTH / r->buffer_width * MACROBLOCK_WIDTH / r->macroblock_size,
 238       (float)BLOCK_HEIGHT / r->buffer_height * MACROBLOCK_HEIGHT / r->macroblock_size
 239    };
 240
 241    unsigned label;
 242
 243    shader = ureg_create(TGSI_PROCESSOR_VERTEX);
 244    if (!shader)
 245       return NULL;
 246
 247    vrect = ureg_DECL_vs_input(shader, VS_I_RECT);
 248    vpos = ureg_DECL_vs_input(shader, VS_I_VPOS);
 249
 250    t_vpos = calc_position(r, shader, ureg_imm2f(shader, scale.x, scale.y));
 251    t_vtex = ureg_DECL_temporary(shader);
 252
 253    o_vpos = ureg_DECL_output(shader, TGSI_SEMANTIC_POSITION, VS_O_VPOS);
 254    o_flags = ureg_DECL_output(shader, TGSI_SEMANTIC_GENERIC, VS_O_FLAGS);
 255
 256    /*
 257     * o_vtex.xy = t_vpos
 258     * o_flags.z = intra * 0.5
 259     *
 260     * if(interlaced) {
 261     *    t_vtex.xy = vrect.y ? { 0, scale.y } : { -scale.y : 0 }
 262     *    t_vtex.z = vpos.y % 2
 263     *    t_vtex.y = t_vtex.z ? t_vtex.x : t_vtex.y
 264     *    o_vpos.y = t_vtex.y + t_vpos.y
 265     *
 266     *    o_flags.w = t_vtex.z ? 0 : 1
 267     * }
 268     *
 269     */
 270
 271    vs_callback(callback_priv, r, shader, VS_O_VTEX, t_vpos);
 272
 273    ureg_MUL(shader, ureg_writemask(o_flags, TGSI_WRITEMASK_Z),
 274             ureg_scalar(vpos, TGSI_SWIZZLE_Z), ureg_imm1f(shader, 0.5f));
 275    ureg_MOV(shader, ureg_writemask(o_flags, TGSI_WRITEMASK_W), ureg_imm1f(shader, -1.0f));
 276
 277    if (r->macroblock_size == MACROBLOCK_HEIGHT) { //TODO
 278       ureg_IF(shader, ureg_scalar(vpos, TGSI_SWIZZLE_W), &label);
 279
 280          ureg_CMP(shader, ureg_writemask(t_vtex, TGSI_WRITEMASK_XY),
 281                   ureg_negate(ureg_scalar(vrect, TGSI_SWIZZLE_Y)),
 282                   ureg_imm2f(shader, 0.0f, scale.y),
 283                   ureg_imm2f(shader, -scale.y, 0.0f));
 284          ureg_MUL(shader, ureg_writemask(t_vtex, TGSI_WRITEMASK_Z),
 285                   ureg_scalar(vpos, TGSI_SWIZZLE_Y), ureg_imm1f(shader, 0.5f));
 286
 287          ureg_FRC(shader, ureg_writemask(t_vtex, TGSI_WRITEMASK_Z), ureg_src(t_vtex));
 288
 289          ureg_CMP(shader, ureg_writemask(t_vtex, TGSI_WRITEMASK_Y),
 290                   ureg_negate(ureg_scalar(ureg_src(t_vtex), TGSI_SWIZZLE_Z)),
 291                   ureg_scalar(ureg_src(t_vtex), TGSI_SWIZZLE_X),
 292                   ureg_scalar(ureg_src(t_vtex), TGSI_SWIZZLE_Y));
 293          ureg_ADD(shader, ureg_writemask(o_vpos, TGSI_WRITEMASK_Y),
 294                   ureg_src(t_vpos), ureg_src(t_vtex));
 295
 296          ureg_CMP(shader, ureg_writemask(o_flags, TGSI_WRITEMASK_W),
 297                   ureg_negate(ureg_scalar(ureg_src(t_vtex), TGSI_SWIZZLE_Z)),
 298                   ureg_imm1f(shader, 0.0f), ureg_imm1f(shader, 1.0f));
 299
 300       ureg_fixup_label(shader, label, ureg_get_instruction_number(shader));
 301       ureg_ENDIF(shader);
 302    }
 303
 304    ureg_release_temporary(shader, t_vtex);
 305    ureg_release_temporary(shader, t_vpos);
 306
 307    ureg_END(shader);
 308
 309    return ureg_create_shader_and_destroy(shader, r->pipe);
 310 }
 311
 312 static void *
 313 create_ycbcr_frag_shader(struct vl_mc *r, float scale, bool invert,
 314                          vl_mc_ycbcr_frag_shader fs_callback, void *callback_priv)
 315 {
 316    struct ureg_program *shader;
 317    struct ureg_src flags;
 318    struct ureg_dst tmp;
 319    struct ureg_dst fragment;
 320    unsigned label;
 321
 322    shader = ureg_create(TGSI_PROCESSOR_FRAGMENT);
 323    if (!shader)
 324       return NULL;
 325
 326    flags = ureg_DECL_fs_input(shader, TGSI_SEMANTIC_GENERIC, VS_O_FLAGS, TGSI_INTERPOLATE_LINEAR);
 327
 328    fragment = ureg_DECL_output(shader, TGSI_SEMANTIC_COLOR, 0);
 329
 330    tmp = calc_line(shader);
 331
 332    /*
 333     * if (field == tc.w)
 334     *    kill();
 335     * else {
 336     *    fragment.xyz  = tex(tc, sampler) * scale + tc.z
 337     *    fragment.w = 1.0f
 338     * }
 339     */
 340
 341    ureg_SEQ(shader, ureg_writemask(tmp, TGSI_WRITEMASK_Y),
 342             ureg_scalar(flags, TGSI_SWIZZLE_W), ureg_src(tmp));
 343
 344    ureg_IF(shader, ureg_scalar(ureg_src(tmp), TGSI_SWIZZLE_Y), &label);
 345
 346       ureg_KILP(shader);
 347
 348    ureg_fixup_label(shader, label, ureg_get_instruction_number(shader));
 349    ureg_ELSE(shader, &label);
 350
 351       fs_callback(callback_priv, r, shader, VS_O_VTEX, tmp);
 352
 353       if (scale != 1.0f)
 354          ureg_MAD(shader, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ),
 355                   ureg_src(tmp), ureg_imm1f(shader, scale),
 356                   ureg_scalar(flags, TGSI_SWIZZLE_Z));
 357       else
 358          ureg_ADD(shader, ureg_writemask(tmp, TGSI_WRITEMASK_XYZ),
 359                   ureg_src(tmp), ureg_scalar(flags, TGSI_SWIZZLE_Z));
 360
 361       ureg_MUL(shader, ureg_writemask(fragment, TGSI_WRITEMASK_XYZ), ureg_src(tmp), ureg_imm1f(shader, invert ? -1.0f : 1.0f));
 362       ureg_MOV(shader, ureg_writemask(fragment, TGSI_WRITEMASK_W), ureg_imm1f(shader, 1.0f));
 363
 364    ureg_fixup_label(shader, label, ureg_get_instruction_number(shader));
 365    ureg_ENDIF(shader);
 366
 367    ureg_release_temporary(shader, tmp);
 368
 369    ureg_END(shader);
 370
 371    return ureg_create_shader_and_destroy(shader, r->pipe);
 372 }
 373
 374 static bool
 375 init_pipe_state(struct vl_mc *r)
 376 {
 377    struct pipe_sampler_state sampler;
 378    struct pipe_blend_state blend;
 379    struct pipe_rasterizer_state rs_state;
 380    unsigned i;
 381
 382    assert(r);
 383
 384    memset(&sampler, 0, sizeof(sampler));
 385    sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 386    sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
 387    sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_BORDER;
 388    sampler.min_img_filter = PIPE_TEX_FILTER_LINEAR;
 389    sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
 390    sampler.mag_img_filter = PIPE_TEX_FILTER_LINEAR;
 391    sampler.compare_mode = PIPE_TEX_COMPARE_NONE;
 392    sampler.compare_func = PIPE_FUNC_ALWAYS;
 393    sampler.normalized_coords = 1;
 394    r->sampler_ref = r->pipe->create_sampler_state(r->pipe, &sampler);
 395    if (!r->sampler_ref)
 396       goto error_sampler_ref;
 397
 398    for (i = 0; i < VL_MC_NUM_BLENDERS; ++i) {
 399       memset(&blend, 0, sizeof blend);
 400       blend.independent_blend_enable = 0;
 401       blend.rt[0].blend_enable = 1;
 402       blend.rt[0].rgb_func = PIPE_BLEND_ADD;
 403       blend.rt[0].rgb_src_factor = PIPE_BLENDFACTOR_SRC_ALPHA;
 404       blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
 405       blend.rt[0].alpha_func = PIPE_BLEND_ADD;
 406       blend.rt[0].alpha_src_factor = PIPE_BLENDFACTOR_SRC_ALPHA;
 407       blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
 408       blend.logicop_enable = 0;
 409       blend.logicop_func = PIPE_LOGICOP_CLEAR;
 410       blend.rt[0].colormask = i;
 411       blend.dither = 0;
 412       r->blend_clear[i] = r->pipe->create_blend_state(r->pipe, &blend);
 413       if (!r->blend_clear[i])
 414          goto error_blend;
 415
 416       blend.rt[0].rgb_dst_factor = PIPE_BLENDFACTOR_ONE;
 417       blend.rt[0].alpha_dst_factor = PIPE_BLENDFACTOR_ONE;
 418       r->blend_add[i] = r->pipe->create_blend_state(r->pipe, &blend);
 419       if (!r->blend_add[i])
 420          goto error_blend;
 421
 422       blend.rt[0].rgb_func = PIPE_BLEND_REVERSE_SUBTRACT;
 423       blend.rt[0].alpha_dst_factor = PIPE_BLEND_REVERSE_SUBTRACT;
 424       r->blend_sub[i] = r->pipe->create_blend_state(r->pipe, &blend);
 425       if (!r->blend_sub[i])
 426          goto error_blend;
 427    }
 428
 429    memset(&rs_state, 0, sizeof(rs_state));
 430    /*rs_state.sprite_coord_enable */
 431    rs_state.sprite_coord_mode = PIPE_SPRITE_COORD_UPPER_LEFT;
 432    rs_state.point_quad_rasterization = true;
 433    rs_state.point_size = BLOCK_WIDTH;
 434    rs_state.gl_rasterization_rules = true;
 435    r->rs_state = r->pipe->create_rasterizer_state(r->pipe, &rs_state);
 436    if (!r->rs_state)
 437       goto error_rs_state;
 438
 439    return true;
 440
 441 error_rs_state:
 442 error_blend:
 443    for (i = 0; i < VL_MC_NUM_BLENDERS; ++i) {
 444       if (r->blend_sub[i])
 445          r->pipe->delete_blend_state(r->pipe, r->blend_sub[i]);
 446
 447       if (r->blend_add[i])
 448          r->pipe->delete_blend_state(r->pipe, r->blend_add[i]);
 449
 450       if (r->blend_clear[i])
 451          r->pipe->delete_blend_state(r->pipe, r->blend_clear[i]);
 452    }
 453
 454    r->pipe->delete_sampler_state(r->pipe, r->sampler_ref);
 455
 456 error_sampler_ref:
 457    return false;
 458 }
 459
 460 static void
 461 cleanup_pipe_state(struct vl_mc *r)
 462 {
 463    unsigned i;
 464
 465    assert(r);
 466
 467    r->pipe->delete_sampler_state(r->pipe, r->sampler_ref);
 468    for (i = 0; i < VL_MC_NUM_BLENDERS; ++i) {
 469       r->pipe->delete_blend_state(r->pipe, r->blend_clear[i]);
 470       r->pipe->delete_blend_state(r->pipe, r->blend_add[i]);
 471       r->pipe->delete_blend_state(r->pipe, r->blend_sub[i]);
 472    }
 473    r->pipe->delete_rasterizer_state(r->pipe, r->rs_state);
 474 }
 475
 476 bool
 477 vl_mc_init(struct vl_mc *renderer, struct pipe_context *pipe,
 478            unsigned buffer_width, unsigned buffer_height,
 479            unsigned macroblock_size, float scale,
 480            vl_mc_ycbcr_vert_shader vs_callback,
 481            vl_mc_ycbcr_frag_shader fs_callback,
 482            void *callback_priv)
 483 {
 484    assert(renderer);
 485    assert(pipe);
 486
 487    memset(renderer, 0, sizeof(struct vl_mc));
 488
 489    renderer->pipe = pipe;
 490    renderer->buffer_width = buffer_width;
 491    renderer->buffer_height = buffer_height;
 492    renderer->macroblock_size = macroblock_size;
 493
 494    if (!init_pipe_state(renderer))
 495       goto error_pipe_state;
 496
 497    renderer->vs_ref = create_ref_vert_shader(renderer);
 498    if (!renderer->vs_ref)
 499       goto error_vs_ref;
 500
 501    renderer->vs_ycbcr = create_ycbcr_vert_shader(renderer, vs_callback, callback_priv);
 502    if (!renderer->vs_ycbcr)
 503       goto error_vs_ycbcr;
 504
 505    renderer->fs_ref = create_ref_frag_shader(renderer);
 506    if (!renderer->fs_ref)
 507       goto error_fs_ref;
 508
 509    renderer->fs_ycbcr = create_ycbcr_frag_shader(renderer, scale, false, fs_callback, callback_priv);
 510    if (!renderer->fs_ycbcr)
 511       goto error_fs_ycbcr;
 512
 513    renderer->fs_ycbcr_sub = create_ycbcr_frag_shader(renderer, scale, true, fs_callback, callback_priv);
 514    if (!renderer->fs_ycbcr_sub)
 515       goto error_fs_ycbcr_sub;
 516
 517    return true;
 518
 519 error_fs_ycbcr_sub:
 520    renderer->pipe->delete_fs_state(renderer->pipe, renderer->fs_ycbcr);
 521
 522 error_fs_ycbcr:
 523    renderer->pipe->delete_fs_state(renderer->pipe, renderer->fs_ref);
 524
 525 error_fs_ref:
 526    renderer->pipe->delete_vs_state(renderer->pipe, renderer->vs_ycbcr);
 527
 528 error_vs_ycbcr:
 529    renderer->pipe->delete_vs_state(renderer->pipe, renderer->vs_ref);
 530
 531 error_vs_ref:
 532    cleanup_pipe_state(renderer);
 533
 534 error_pipe_state:
 535    return false;
 536 }
 537
 538 void
 539 vl_mc_cleanup(struct vl_mc *renderer)
 540 {
 541    assert(renderer);
 542
 543    cleanup_pipe_state(renderer);
 544
 545    renderer->pipe->delete_vs_state(renderer->pipe, renderer->vs_ref);
 546    renderer->pipe->delete_vs_state(renderer->pipe, renderer->vs_ycbcr);
 547    renderer->pipe->delete_fs_state(renderer->pipe, renderer->fs_ref);
 548    renderer->pipe->delete_fs_state(renderer->pipe, renderer->fs_ycbcr);
 549    renderer->pipe->delete_fs_state(renderer->pipe, renderer->fs_ycbcr_sub);
 550 }
 551
 552 bool
 553 vl_mc_init_buffer(struct vl_mc *renderer, struct vl_mc_buffer *buffer)
 554 {
 555    assert(renderer && buffer);
 556
 557    buffer->renderer = renderer;
 558
 559    buffer->viewport.scale[2] = 1;
 560    buffer->viewport.scale[3] = 1;
 561    buffer->viewport.translate[0] = 0;
 562    buffer->viewport.translate[1] = 0;
 563    buffer->viewport.translate[2] = 0;
 564    buffer->viewport.translate[3] = 0;
 565
 566    buffer->fb_state.nr_cbufs = 1;
 567    buffer->fb_state.zsbuf = NULL;
 568
 569    return true;
 570 }
 571
 572 void
 573 vl_mc_cleanup_buffer(struct vl_mc_buffer *buffer)
 574 {
 575    assert(buffer);
 576 }
 577
 578 void
 579 vl_mc_set_surface(struct vl_mc_buffer *buffer, struct pipe_surface *surface)
 580 {
 581    assert(buffer && surface);
 582
 583    buffer->surface_cleared = false;
 584
 585    buffer->viewport.scale[0] = surface->width;
 586    buffer->viewport.scale[1] = surface->height;
 587
 588    buffer->fb_state.width = surface->width;
 589    buffer->fb_state.height = surface->height;
 590    buffer->fb_state.cbufs[0] = surface;
 591 }
 592
 593 static void
 594 prepare_pipe_4_rendering(struct vl_mc_buffer *buffer, unsigned component, unsigned mask)
 595 {
 596    struct vl_mc *renderer;
 597
 598    assert(buffer);
 599
 600    renderer = buffer->renderer;
 601    renderer->pipe->bind_rasterizer_state(renderer->pipe, renderer->rs_state);
 602
 603    if (buffer->surface_cleared || component > 0)
 604       renderer->pipe->bind_blend_state(renderer->pipe, renderer->blend_add[mask]);
 605    else
 606       renderer->pipe->bind_blend_state(renderer->pipe, renderer->blend_clear[mask]);
 607
 608    renderer->pipe->set_framebuffer_state(renderer->pipe, &buffer->fb_state);
 609    renderer->pipe->set_viewport_state(renderer->pipe, &buffer->viewport);
 610 }
 611
 612 void
 613 vl_mc_render_ref(struct vl_mc_buffer *buffer, struct pipe_sampler_view *ref)
 614 {
 615    struct vl_mc *renderer;
 616
 617    assert(buffer && ref);
 618
 619    prepare_pipe_4_rendering(buffer, 0, PIPE_MASK_R | PIPE_MASK_G | PIPE_MASK_B);
 620
 621    renderer = buffer->renderer;
 622
 623    renderer->pipe->bind_vs_state(renderer->pipe, renderer->vs_ref);
 624    renderer->pipe->bind_fs_state(renderer->pipe, renderer->fs_ref);
 625
 626    renderer->pipe->set_fragment_sampler_views(renderer->pipe, 1, &ref);
 627    renderer->pipe->bind_fragment_sampler_states(renderer->pipe, 1, &renderer->sampler_ref);
 628
 629    util_draw_arrays_instanced(renderer->pipe, PIPE_PRIM_QUADS, 0, 4, 0,
 630                               renderer->buffer_width / MACROBLOCK_WIDTH *
 631                               renderer->buffer_height / MACROBLOCK_HEIGHT);
 632
 633    buffer->surface_cleared = true;
 634 }
 635
 636 void
 637 vl_mc_render_ycbcr(struct vl_mc_buffer *buffer, unsigned component, unsigned num_instances)
 638 {
 639    struct vl_mc *renderer;
 640    unsigned mask = 1 << component;
 641
 642    assert(buffer);
 643
 644    if (num_instances == 0)
 645       return;
 646
 647    prepare_pipe_4_rendering(buffer, component, mask);
 648
 649    renderer = buffer->renderer;
 650
 651    renderer->pipe->bind_vs_state(renderer->pipe, renderer->vs_ycbcr);
 652    renderer->pipe->bind_fs_state(renderer->pipe, renderer->fs_ycbcr);
 653
 654    util_draw_arrays_instanced(renderer->pipe, PIPE_PRIM_QUADS, 0, 4, 0, num_instances);
 655
 656    if (buffer->surface_cleared) {
 657       renderer->pipe->bind_blend_state(renderer->pipe, renderer->blend_sub[mask]);
 658       renderer->pipe->bind_fs_state(renderer->pipe, renderer->fs_ycbcr_sub);
 659       util_draw_arrays_instanced(renderer->pipe, PIPE_PRIM_QUADS, 0, 4, 0, num_instances);
 660    }
 661 }