src/gallium/drivers/svga/svga_state_fs.c

   1 /**********************************************************
   2  * Copyright 2008-2009 VMware, Inc.  All rights reserved.
   3  *
   4  * Permission is hereby granted, free of charge, to any person
   5  * obtaining a copy of this software and associated documentation
   6  * files (the "Software"), to deal in the Software without
   7  * restriction, including without limitation the rights to use, copy,
   8  * modify, merge, publish, distribute, sublicense, and/or sell copies
   9  * of the Software, and to permit persons to whom the Software is
  10  * furnished to do so, subject to the following conditions:
  11  *
  12  * The above copyright notice and this permission notice shall be
  13  * included in all copies or substantial portions of the Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  16  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  17  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  18  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  19  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  20  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  21  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  22  * SOFTWARE.
  23  *
  24  **********************************************************/
  25
  26 #include "util/u_inlines.h"
  27 #include "pipe/p_defines.h"
  28 #include "util/u_format.h"
  29 #include "util/u_math.h"
  30 #include "util/u_memory.h"
  31 #include "util/u_bitmask.h"
  32 #include "tgsi/tgsi_ureg.h"
  33
  34 #include "svga_context.h"
  35 #include "svga_state.h"
  36 #include "svga_cmd.h"
  37 #include "svga_shader.h"
  38 #include "svga_resource_texture.h"
  39 #include "svga_tgsi.h"
  40 #include "svga_format.h"
  41
  42 #include "svga_hw_reg.h"
  43
  44
  45
  46 /**
  47  * If we fail to compile a fragment shader (because it uses too many
  48  * registers, for example) we'll use a dummy/fallback shader that
  49  * simply emits a constant color (red for debug, black for release).
  50  * We hit this with the Unigine/Heaven demo when Shaders = High.
  51  * With black, the demo still looks good.
  52  */
  53 static const struct tgsi_token *
  54 get_dummy_fragment_shader(void)
  55 {
  56 #ifdef DEBUG
  57    static const float color[4] = { 1.0, 0.0, 0.0, 0.0 }; /* red */
  58 #else
  59    static const float color[4] = { 0.0, 0.0, 0.0, 0.0 }; /* black */
  60 #endif
  61    struct ureg_program *ureg;
  62    const struct tgsi_token *tokens;
  63    struct ureg_src src;
  64    struct ureg_dst dst;
  65
  66    ureg = ureg_create(PIPE_SHADER_FRAGMENT);
  67    if (!ureg)
  68       return NULL;
  69
  70    dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_COLOR, 0);
  71    src = ureg_DECL_immediate(ureg, color, 4);
  72    ureg_MOV(ureg, dst, src);
  73    ureg_END(ureg);
  74
  75    tokens = ureg_get_tokens(ureg, NULL);
  76
  77    ureg_destroy(ureg);
  78
  79    return tokens;
  80 }
  81
  82
  83 static struct svga_shader_variant *
  84 translate_fragment_program(struct svga_context *svga,
  85                            const struct svga_fragment_shader *fs,
  86                            const struct svga_compile_key *key)
  87 {
  88    if (svga_have_vgpu10(svga)) {
  89       return svga_tgsi_vgpu10_translate(svga, &fs->base, key,
  90                                         PIPE_SHADER_FRAGMENT);
  91    }
  92    else {
  93       return svga_tgsi_vgpu9_translate(svga, &fs->base, key,
  94                                        PIPE_SHADER_FRAGMENT);
  95    }
  96 }
  97
  98
  99 /**
 100  * Replace the given shader's instruction with a simple constant-color
 101  * shader.  We use this when normal shader translation fails.
 102  */
 103 static struct svga_shader_variant *
 104 get_compiled_dummy_shader(struct svga_context *svga,
 105                           struct svga_fragment_shader *fs,
 106                           const struct svga_compile_key *key)
 107 {
 108    const struct tgsi_token *dummy = get_dummy_fragment_shader();
 109    struct svga_shader_variant *variant;
 110
 111    if (!dummy) {
 112       return NULL;
 113    }
 114
 115    FREE((void *) fs->base.tokens);
 116    fs->base.tokens = dummy;
 117
 118    variant = translate_fragment_program(svga, fs, key);
 119    return variant;
 120 }
 121
 122
 123 /**
 124  * Translate TGSI shader into an svga shader variant.
 125  */
 126 static enum pipe_error
 127 compile_fs(struct svga_context *svga,
 128            struct svga_fragment_shader *fs,
 129            const struct svga_compile_key *key,
 130            struct svga_shader_variant **out_variant)
 131 {
 132    struct svga_shader_variant *variant;
 133    enum pipe_error ret = PIPE_ERROR;
 134
 135    variant = translate_fragment_program(svga, fs, key);
 136    if (variant == NULL) {
 137       debug_printf("Failed to compile fragment shader,"
 138                    " using dummy shader instead.\n");
 139       variant = get_compiled_dummy_shader(svga, fs, key);
 140    }
 141    else if (svga_shader_too_large(svga, variant)) {
 142       /* too big, use dummy shader */
 143       debug_printf("Shader too large (%u bytes),"
 144                    " using dummy shader instead.\n",
 145                    (unsigned) (variant->nr_tokens
 146                                * sizeof(variant->tokens[0])));
 147       /* Free the too-large variant */
 148       svga_destroy_shader_variant(svga, SVGA3D_SHADERTYPE_PS, variant);
 149       /* Use simple pass-through shader instead */
 150       variant = get_compiled_dummy_shader(svga, fs, key);
 151    }
 152
 153    if (!variant) {
 154       return PIPE_ERROR;
 155    }
 156
 157    ret = svga_define_shader(svga, SVGA3D_SHADERTYPE_PS, variant);
 158    if (ret != PIPE_OK) {
 159       svga_destroy_shader_variant(svga, SVGA3D_SHADERTYPE_PS, variant);
 160       return ret;
 161    }
 162
 163    *out_variant = variant;
 164
 165    /* insert variant at head of linked list */
 166    variant->next = fs->base.variants;
 167    fs->base.variants = variant;
 168
 169    return PIPE_OK;
 170 }
 171
 172
 173 /* SVGA_NEW_TEXTURE_BINDING
 174  * SVGA_NEW_RAST
 175  * SVGA_NEW_NEED_SWTNL
 176  * SVGA_NEW_SAMPLER
 177  */
 178 static enum pipe_error
 179 make_fs_key(const struct svga_context *svga,
 180             struct svga_fragment_shader *fs,
 181             struct svga_compile_key *key)
 182 {
 183    const enum pipe_shader_type shader = PIPE_SHADER_FRAGMENT;
 184    unsigned i;
 185
 186    memset(key, 0, sizeof *key);
 187
 188    memcpy(key->generic_remap_table, fs->generic_remap_table,
 189           sizeof(fs->generic_remap_table));
 190
 191    /* SVGA_NEW_GS, SVGA_NEW_VS
 192     */
 193    if (svga->curr.gs) {
 194       key->fs.gs_generic_outputs = svga->curr.gs->generic_outputs;
 195    } else {
 196       key->fs.vs_generic_outputs = svga->curr.vs->generic_outputs;
 197    }
 198
 199    /* Only need fragment shader fixup for twoside lighting if doing
 200     * hwtnl.  Otherwise the draw module does the whole job for us.
 201     *
 202     * SVGA_NEW_SWTNL
 203     */
 204    if (!svga->state.sw.need_swtnl) {
 205       /* SVGA_NEW_RAST, SVGA_NEW_REDUCED_PRIMITIVE
 206        */
 207       key->fs.light_twoside = svga->curr.rast->templ.light_twoside;
 208       key->fs.front_ccw = svga->curr.rast->templ.front_ccw;
 209       key->fs.pstipple = (svga->curr.rast->templ.poly_stipple_enable &&
 210                           svga->curr.reduced_prim == PIPE_PRIM_TRIANGLES);
 211       key->fs.aa_point = (svga->curr.rast->templ.point_smooth &&
 212                           svga->curr.reduced_prim == PIPE_PRIM_POINTS &&
 213                           (svga->curr.rast->pointsize > 1.0 ||
 214                            svga->curr.vs->base.info.writes_psize));
 215       if (key->fs.aa_point) {
 216          assert(svga->curr.gs != NULL);
 217          assert(svga->curr.gs->aa_point_coord_index != -1);
 218          key->fs.aa_point_coord_index = svga->curr.gs->aa_point_coord_index;
 219       }
 220    }
 221
 222    /* The blend workaround for simulating logicop xor behaviour
 223     * requires that the incoming fragment color be white.  This change
 224     * achieves that by creating a variant of the current fragment
 225     * shader that overrides all output colors with 1,1,1,1
 226     *
 227     * This will work for most shaders, including those containing
 228     * TEXKIL and/or depth-write.  However, it will break on the
 229     * combination of xor-logicop plus alphatest.
 230     *
 231     * Ultimately, we could implement alphatest in the shader using
 232     * texkil prior to overriding the outgoing fragment color.
 233     *
 234     * SVGA_NEW_BLEND
 235     */
 236    key->fs.white_fragments = svga->curr.blend->need_white_fragments;
 237
 238    key->fs.alpha_to_one = svga->curr.blend->alpha_to_one;
 239
 240 #ifdef DEBUG
 241    /*
 242     * We expect a consistent set of samplers and sampler views.
 243     * Do some debug checks/warnings here.
 244     */
 245    {
 246       static boolean warned = FALSE;
 247       unsigned i, n = MAX2(svga->curr.num_sampler_views[shader],
 248                            svga->curr.num_samplers[shader]);
 249       /* Only warn once to prevent too much debug output */
 250       if (!warned) {
 251          if (svga->curr.num_sampler_views[shader] !=
 252              svga->curr.num_samplers[shader]) {
 253             debug_printf("svga: mismatched number of sampler views (%u) "
 254                          "vs. samplers (%u)\n",
 255                          svga->curr.num_sampler_views[shader],
 256                          svga->curr.num_samplers[shader]);
 257          }
 258          for (i = 0; i < n; i++) {
 259             if ((svga->curr.sampler_views[shader][i] == NULL) !=
 260                 (svga->curr.sampler[shader][i] == NULL))
 261                debug_printf("sampler_view[%u] = %p but sampler[%u] = %p\n",
 262                             i, svga->curr.sampler_views[shader][i],
 263                             i, svga->curr.sampler[shader][i]);
 264          }
 265          warned = TRUE;
 266       }
 267    }
 268 #endif
 269
 270    /* XXX: want to limit this to the textures that the shader actually
 271     * refers to.
 272     *
 273     * SVGA_NEW_TEXTURE_BINDING | SVGA_NEW_SAMPLER
 274     */
 275    svga_init_shader_key_common(svga, shader, key);
 276
 277    for (i = 0; i < svga->curr.num_samplers[shader]; ++i) {
 278       struct pipe_sampler_view *view = svga->curr.sampler_views[shader][i];
 279       const struct svga_sampler_state *sampler = svga->curr.sampler[shader][i];
 280       if (view) {
 281          struct pipe_resource *tex = view->texture;
 282          if (tex->target != PIPE_BUFFER) {
 283             struct svga_texture *stex = svga_texture(tex);
 284             SVGA3dSurfaceFormat format = stex->key.format;
 285
 286             if (!svga_have_vgpu10(svga) &&
 287                 (format == SVGA3D_Z_D16 ||
 288                  format == SVGA3D_Z_D24X8 ||
 289                  format == SVGA3D_Z_D24S8)) {
 290                /* If we're sampling from a SVGA3D_Z_D16, SVGA3D_Z_D24X8,
 291                 * or SVGA3D_Z_D24S8 surface, we'll automatically get
 292                 * shadow comparison.  But we only get LEQUAL mode.
 293                 * Set TEX_COMPARE_NONE here so we don't emit the extra FS
 294                 * code for shadow comparison.
 295                 */
 296                key->tex[i].compare_mode = PIPE_TEX_COMPARE_NONE;
 297                key->tex[i].compare_func = PIPE_FUNC_NEVER;
 298                /* These depth formats _only_ support comparison mode and
 299                 * not ordinary sampling so warn if the later is expected.
 300                 */
 301                if (sampler->compare_mode != PIPE_TEX_COMPARE_R_TO_TEXTURE) {
 302                   debug_warn_once("Unsupported shadow compare mode");
 303                }
 304                /* The shader translation code can emit code to
 305                 * handle ALWAYS and NEVER compare functions
 306                 */
 307                else if (sampler->compare_func == PIPE_FUNC_ALWAYS ||
 308                         sampler->compare_func == PIPE_FUNC_NEVER) {
 309                   key->tex[i].compare_mode = sampler->compare_mode;
 310                   key->tex[i].compare_func = sampler->compare_func;
 311                }
 312                else if (sampler->compare_func != PIPE_FUNC_LEQUAL) {
 313                   debug_warn_once("Unsupported shadow compare function");
 314                }
 315             }
 316             else {
 317                /* For other texture formats, just use the compare func/mode
 318                 * as-is.  Should be no-ops for color textures.  For depth
 319                 * textures, we do not get automatic depth compare.  We have
 320                 * to do it ourselves in the shader.  And we don't get PCF.
 321                 */
 322                key->tex[i].compare_mode = sampler->compare_mode;
 323                key->tex[i].compare_func = sampler->compare_func;
 324             }
 325          }
 326       }
 327    }
 328
 329    /* sprite coord gen state */
 330    for (i = 0; i < svga->curr.num_samplers[shader]; ++i) {
 331       key->tex[i].sprite_texgen =
 332          svga->curr.rast->templ.sprite_coord_enable & (1 << i);
 333    }
 334
 335    key->sprite_origin_lower_left = (svga->curr.rast->templ.sprite_coord_mode
 336                                     == PIPE_SPRITE_COORD_LOWER_LEFT);
 337
 338    key->fs.flatshade = svga->curr.rast->templ.flatshade;
 339
 340    /* SVGA_NEW_DEPTH_STENCIL_ALPHA */
 341    if (svga_have_vgpu10(svga)) {
 342       /* Alpha testing is not supported in integer-valued render targets. */
 343       if (svga_has_any_integer_cbufs(svga)) {
 344          key->fs.alpha_func = SVGA3D_CMP_ALWAYS;
 345          key->fs.alpha_ref = 0;
 346       }
 347       else {
 348          key->fs.alpha_func = svga->curr.depth->alphafunc;
 349          key->fs.alpha_ref = svga->curr.depth->alpharef;
 350       }
 351    }
 352
 353    /* SVGA_NEW_FRAME_BUFFER | SVGA_NEW_BLEND */
 354    if (fs->base.info.properties[TGSI_PROPERTY_FS_COLOR0_WRITES_ALL_CBUFS] ||
 355        svga->curr.blend->need_white_fragments) {
 356       /* Replicate color0 output (or white) to N colorbuffers */
 357       key->fs.write_color0_to_n_cbufs = svga->curr.framebuffer.nr_cbufs;
 358    }
 359
 360    return PIPE_OK;
 361 }
 362
 363
 364 /**
 365  * svga_reemit_fs_bindings - Reemit the fragment shader bindings
 366  */
 367 enum pipe_error
 368 svga_reemit_fs_bindings(struct svga_context *svga)
 369 {
 370    enum pipe_error ret;
 371
 372    assert(svga->rebind.flags.fs);
 373    assert(svga_have_gb_objects(svga));
 374
 375    if (!svga->state.hw_draw.fs)
 376       return PIPE_OK;
 377
 378    if (!svga_need_to_rebind_resources(svga)) {
 379       ret =  svga->swc->resource_rebind(svga->swc, NULL,
 380                                         svga->state.hw_draw.fs->gb_shader,
 381                                         SVGA_RELOC_READ);
 382       goto out;
 383    }
 384
 385    if (svga_have_vgpu10(svga))
 386       ret = SVGA3D_vgpu10_SetShader(svga->swc, SVGA3D_SHADERTYPE_PS,
 387                                     svga->state.hw_draw.fs->gb_shader,
 388                                     svga->state.hw_draw.fs->id);
 389    else
 390       ret = SVGA3D_SetGBShader(svga->swc, SVGA3D_SHADERTYPE_PS,
 391                                svga->state.hw_draw.fs->gb_shader);
 392
 393  out:
 394    if (ret != PIPE_OK)
 395       return ret;
 396
 397    svga->rebind.flags.fs = FALSE;
 398    return PIPE_OK;
 399 }
 400
 401
 402
 403 static enum pipe_error
 404 emit_hw_fs(struct svga_context *svga, unsigned dirty)
 405 {
 406    struct svga_shader_variant *variant = NULL;
 407    enum pipe_error ret = PIPE_OK;
 408    struct svga_fragment_shader *fs = svga->curr.fs;
 409    struct svga_compile_key key;
 410
 411    SVGA_STATS_TIME_PUSH(svga_sws(svga), SVGA_STATS_TIME_EMITFS);
 412
 413    /* Disable rasterization if rasterizer_discard flag is set or
 414     * vs/gs does not output position.
 415     */
 416    svga->disable_rasterizer =
 417       svga->curr.rast->templ.rasterizer_discard ||
 418       (svga->curr.gs && !svga->curr.gs->base.info.writes_position) ||
 419       (!svga->curr.gs && !svga->curr.vs->base.info.writes_position);
 420
 421    /* Set FS to NULL when rasterization is to be disabled */
 422    if (svga->disable_rasterizer) {
 423       /* Set FS to NULL if it has not been done */
 424       if (svga->state.hw_draw.fs) {
 425          ret = svga_set_shader(svga, SVGA3D_SHADERTYPE_PS, NULL);
 426          if (ret != PIPE_OK)
 427             goto done;
 428       }
 429       svga->rebind.flags.fs = FALSE;
 430       svga->state.hw_draw.fs = NULL;
 431       goto done;
 432    }
 433
 434    /* SVGA_NEW_BLEND
 435     * SVGA_NEW_TEXTURE_BINDING
 436     * SVGA_NEW_RAST
 437     * SVGA_NEW_NEED_SWTNL
 438     * SVGA_NEW_SAMPLER
 439     * SVGA_NEW_FRAME_BUFFER
 440     * SVGA_NEW_DEPTH_STENCIL_ALPHA
 441     * SVGA_NEW_VS
 442     */
 443    ret = make_fs_key(svga, fs, &key);
 444    if (ret != PIPE_OK)
 445       goto done;
 446
 447    variant = svga_search_shader_key(&fs->base, &key);
 448    if (!variant) {
 449       ret = compile_fs(svga, fs, &key, &variant);
 450       if (ret != PIPE_OK)
 451          goto done;
 452    }
 453
 454    assert(variant);
 455
 456    if (variant != svga->state.hw_draw.fs) {
 457       ret = svga_set_shader(svga, SVGA3D_SHADERTYPE_PS, variant);
 458       if (ret != PIPE_OK)
 459          goto done;
 460
 461       svga->rebind.flags.fs = FALSE;
 462
 463       svga->dirty |= SVGA_NEW_FS_VARIANT;
 464       svga->state.hw_draw.fs = variant;
 465    }
 466
 467 done:
 468    SVGA_STATS_TIME_POP(svga_sws(svga));
 469    return ret;
 470 }
 471
 472 struct svga_tracked_state svga_hw_fs =
 473 {
 474    "fragment shader (hwtnl)",
 475    (SVGA_NEW_FS |
 476     SVGA_NEW_GS |
 477     SVGA_NEW_VS |
 478     SVGA_NEW_TEXTURE_BINDING |
 479     SVGA_NEW_NEED_SWTNL |
 480     SVGA_NEW_RAST |
 481     SVGA_NEW_STIPPLE |
 482     SVGA_NEW_REDUCED_PRIMITIVE |
 483     SVGA_NEW_SAMPLER |
 484     SVGA_NEW_FRAME_BUFFER |
 485     SVGA_NEW_DEPTH_STENCIL_ALPHA |
 486     SVGA_NEW_BLEND),
 487    emit_hw_fs
 488 };
 489
 490
 491