src/gallium/drivers/svga/svga_state_vs.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_math.h"
  29 #include "util/u_memory.h"
  30 #include "util/u_bitmask.h"
  31 #include "translate/translate.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_tgsi.h"
  39
  40 #include "svga_hw_reg.h"
  41
  42
  43 /**
  44  * If we fail to compile a vertex shader we'll use a dummy/fallback shader
  45  * that simply emits a (0,0,0,1) vertex position.
  46  */
  47 static const struct tgsi_token *
  48 get_dummy_vertex_shader(void)
  49 {
  50    static const float zero[4] = { 0.0, 0.0, 0.0, 1.0 };
  51    struct ureg_program *ureg;
  52    const struct tgsi_token *tokens;
  53    struct ureg_src src;
  54    struct ureg_dst dst;
  55
  56    ureg = ureg_create(PIPE_SHADER_VERTEX);
  57    if (!ureg)
  58       return NULL;
  59
  60    dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0);
  61    src = ureg_DECL_immediate(ureg, zero, 4);
  62    ureg_MOV(ureg, dst, src);
  63    ureg_END(ureg);
  64
  65    tokens = ureg_get_tokens(ureg, NULL);
  66
  67    ureg_destroy(ureg);
  68
  69    return tokens;
  70 }
  71
  72
  73 static struct svga_shader_variant *
  74 translate_vertex_program(struct svga_context *svga,
  75                          const struct svga_vertex_shader *vs,
  76                          const struct svga_compile_key *key)
  77 {
  78    if (svga_have_vgpu10(svga)) {
  79       return svga_tgsi_vgpu10_translate(svga, &vs->base, key,
  80                                         PIPE_SHADER_VERTEX);
  81    }
  82    else {
  83       return svga_tgsi_vgpu9_translate(svga, &vs->base, key,
  84                                        PIPE_SHADER_VERTEX);
  85    }
  86 }
  87
  88
  89 /**
  90  * Replace the given shader's instruction with a simple / dummy shader.
  91  * We use this when normal shader translation fails.
  92  */
  93 static struct svga_shader_variant *
  94 get_compiled_dummy_vertex_shader(struct svga_context *svga,
  95                                  struct svga_vertex_shader *vs,
  96                                  const struct svga_compile_key *key)
  97 {
  98    const struct tgsi_token *dummy = get_dummy_vertex_shader();
  99    struct svga_shader_variant *variant;
 100
 101    if (!dummy) {
 102       return NULL;
 103    }
 104
 105    FREE((void *) vs->base.tokens);
 106    vs->base.tokens = dummy;
 107
 108    variant = translate_vertex_program(svga, vs, key);
 109    return variant;
 110 }
 111
 112
 113 /**
 114  * Translate TGSI shader into an svga shader variant.
 115  */
 116 static enum pipe_error
 117 compile_vs(struct svga_context *svga,
 118            struct svga_vertex_shader *vs,
 119            const struct svga_compile_key *key,
 120            struct svga_shader_variant **out_variant)
 121 {
 122    struct svga_shader_variant *variant;
 123    enum pipe_error ret = PIPE_ERROR;
 124
 125    variant = translate_vertex_program(svga, vs, key);
 126    if (variant == NULL) {
 127       debug_printf("Failed to compile vertex shader,"
 128                    " using dummy shader instead.\n");
 129       variant = get_compiled_dummy_vertex_shader(svga, vs, key);
 130    }
 131    else if (svga_shader_too_large(svga, variant)) {
 132       /* too big, use dummy shader */
 133       debug_printf("Shader too large (%u bytes),"
 134                    " using dummy shader instead.\n",
 135                    (unsigned) (variant->nr_tokens
 136                                * sizeof(variant->tokens[0])));
 137       /* Free the too-large variant */
 138       svga_destroy_shader_variant(svga, SVGA3D_SHADERTYPE_VS, variant);
 139       /* Use simple pass-through shader instead */
 140       variant = get_compiled_dummy_vertex_shader(svga, vs, key);
 141    }
 142
 143    if (!variant) {
 144       return PIPE_ERROR;
 145    }
 146
 147    ret = svga_define_shader(svga, SVGA3D_SHADERTYPE_VS, variant);
 148    if (ret != PIPE_OK) {
 149       svga_destroy_shader_variant(svga, SVGA3D_SHADERTYPE_VS, variant);
 150       return ret;
 151    }
 152
 153    *out_variant = variant;
 154
 155    return PIPE_OK;
 156 }
 157
 158
 159 /* SVGA_NEW_PRESCALE, SVGA_NEW_RAST, SVGA_NEW_FS
 160  */
 161 static void
 162 make_vs_key(struct svga_context *svga, struct svga_compile_key *key)
 163 {
 164    const enum pipe_shader_type shader = PIPE_SHADER_VERTEX;
 165
 166    memset(key, 0, sizeof *key);
 167
 168    if (svga->state.sw.need_swtnl && svga_have_vgpu10(svga)) {
 169       /* Set both of these flags, to match compile_passthrough_vs() */
 170       key->vs.passthrough = 1;
 171       key->vs.undo_viewport = 1;
 172       return;
 173    }
 174
 175    /* SVGA_NEW_PRESCALE */
 176    key->vs.need_prescale = svga->state.hw_clear.prescale.enabled &&
 177                            (svga->curr.gs == NULL);
 178
 179    /* SVGA_NEW_RAST */
 180    key->vs.allow_psiz = svga->curr.rast->templ.point_size_per_vertex;
 181
 182    /* SVGA_NEW_FS */
 183    key->vs.fs_generic_inputs = svga->curr.fs->generic_inputs;
 184
 185    svga_remap_generics(key->vs.fs_generic_inputs, key->generic_remap_table);
 186
 187    /* SVGA_NEW_VELEMENT */
 188    key->vs.adjust_attrib_range = svga->curr.velems->adjust_attrib_range;
 189    key->vs.adjust_attrib_w_1 = svga->curr.velems->adjust_attrib_w_1;
 190    key->vs.attrib_is_pure_int = svga->curr.velems->attrib_is_pure_int;
 191    key->vs.adjust_attrib_itof = svga->curr.velems->adjust_attrib_itof;
 192    key->vs.adjust_attrib_utof = svga->curr.velems->adjust_attrib_utof;
 193    key->vs.attrib_is_bgra = svga->curr.velems->attrib_is_bgra;
 194    key->vs.attrib_puint_to_snorm = svga->curr.velems->attrib_puint_to_snorm;
 195    key->vs.attrib_puint_to_uscaled = svga->curr.velems->attrib_puint_to_uscaled;
 196    key->vs.attrib_puint_to_sscaled = svga->curr.velems->attrib_puint_to_sscaled;
 197
 198    /* SVGA_NEW_TEXTURE_BINDING | SVGA_NEW_SAMPLER */
 199    svga_init_shader_key_common(svga, shader, key);
 200
 201    /* SVGA_NEW_RAST */
 202    key->clip_plane_enable = svga->curr.rast->templ.clip_plane_enable;
 203 }
 204
 205
 206 /**
 207  * svga_reemit_vs_bindings - Reemit the vertex shader bindings
 208  */
 209 enum pipe_error
 210 svga_reemit_vs_bindings(struct svga_context *svga)
 211 {
 212    enum pipe_error ret;
 213    struct svga_winsys_gb_shader *gbshader = NULL;
 214    SVGA3dShaderId shaderId = SVGA3D_INVALID_ID;
 215
 216    assert(svga->rebind.flags.vs);
 217    assert(svga_have_gb_objects(svga));
 218
 219    if (svga->state.hw_draw.vs) {
 220       gbshader = svga->state.hw_draw.vs->gb_shader;
 221       shaderId = svga->state.hw_draw.vs->id;
 222    }
 223
 224    if (!svga_need_to_rebind_resources(svga)) {
 225       ret =  svga->swc->resource_rebind(svga->swc, NULL, gbshader,
 226                                         SVGA_RELOC_READ);
 227       goto out;
 228    }
 229
 230    if (svga_have_vgpu10(svga))
 231       ret = SVGA3D_vgpu10_SetShader(svga->swc, SVGA3D_SHADERTYPE_VS,
 232                                     gbshader, shaderId);
 233    else
 234       ret = SVGA3D_SetGBShader(svga->swc, SVGA3D_SHADERTYPE_VS, gbshader);
 235
 236  out:
 237    if (ret != PIPE_OK)
 238       return ret;
 239
 240    svga->rebind.flags.vs = FALSE;
 241    return PIPE_OK;
 242 }
 243
 244
 245 /**
 246  * The current vertex shader is already executed by the 'draw'
 247  * module, so we just need to generate a simple vertex shader
 248  * to pass through all those VS outputs that will
 249  * be consumed by the fragment shader.
 250  * Used when we employ the 'draw' module.
 251  */
 252 static enum pipe_error
 253 compile_passthrough_vs(struct svga_context *svga,
 254                        struct svga_vertex_shader *vs,
 255                        struct svga_fragment_shader *fs,
 256                        struct svga_shader_variant **out_variant)
 257 {
 258    struct svga_shader_variant *variant = NULL;
 259    unsigned num_inputs;
 260    unsigned i;
 261    unsigned num_elements;
 262    struct svga_vertex_shader new_vs;
 263    struct ureg_src src[PIPE_MAX_SHADER_INPUTS];
 264    struct ureg_dst dst[PIPE_MAX_SHADER_OUTPUTS];
 265    struct ureg_program *ureg;
 266    struct svga_compile_key key;
 267    enum pipe_error ret;
 268
 269    assert(svga_have_vgpu10(svga));
 270    assert(fs);
 271
 272    num_inputs = fs->base.info.num_inputs;
 273
 274    ureg = ureg_create(PIPE_SHADER_VERTEX);
 275    if (!ureg)
 276       return PIPE_ERROR_OUT_OF_MEMORY;
 277
 278    /* draw will always add position */
 279    dst[0] = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0);
 280    src[0] = ureg_DECL_vs_input(ureg, 0);
 281    num_elements = 1;
 282
 283    /**
 284     * swtnl backend redefines the input layout based on the
 285     * fragment shader's inputs. So we only need to passthrough
 286     * those inputs that will be consumed by the fragment shader.
 287     * Note: DX10 requires the number of vertex elements
 288     * specified in the input layout to be no less than the
 289     * number of inputs to the vertex shader.
 290     */
 291    for (i = 0; i < num_inputs; i++) {
 292       switch (fs->base.info.input_semantic_name[i]) {
 293       case TGSI_SEMANTIC_COLOR:
 294       case TGSI_SEMANTIC_GENERIC:
 295       case TGSI_SEMANTIC_FOG:
 296          dst[num_elements] = ureg_DECL_output(ureg,
 297                                 fs->base.info.input_semantic_name[i],
 298                                 fs->base.info.input_semantic_index[i]);
 299          src[num_elements] = ureg_DECL_vs_input(ureg, num_elements);
 300          num_elements++;
 301          break;
 302       default:
 303          break;
 304       }
 305    }
 306
 307    for (i = 0; i < num_elements; i++) {
 308       ureg_MOV(ureg, dst[i], src[i]);
 309    }
 310
 311    ureg_END(ureg);
 312
 313    memset(&new_vs, 0, sizeof(new_vs));
 314    new_vs.base.tokens = ureg_get_tokens(ureg, NULL);
 315    tgsi_scan_shader(new_vs.base.tokens, &new_vs.base.info);
 316
 317    memset(&key, 0, sizeof(key));
 318    key.vs.undo_viewport = 1;
 319
 320    ret = compile_vs(svga, &new_vs, &key, &variant);
 321    if (ret != PIPE_OK)
 322       return ret;
 323
 324    ureg_free_tokens(new_vs.base.tokens);
 325    ureg_destroy(ureg);
 326
 327    /* Overwrite the variant key to indicate it's a pass-through VS */
 328    memset(&variant->key, 0, sizeof(variant->key));
 329    variant->key.vs.passthrough = 1;
 330    variant->key.vs.undo_viewport = 1;
 331
 332    *out_variant = variant;
 333
 334    return PIPE_OK;
 335 }
 336
 337
 338 static enum pipe_error
 339 emit_hw_vs(struct svga_context *svga, unsigned dirty)
 340 {
 341    struct svga_shader_variant *variant;
 342    struct svga_vertex_shader *vs = svga->curr.vs;
 343    struct svga_fragment_shader *fs = svga->curr.fs;
 344    enum pipe_error ret = PIPE_OK;
 345    struct svga_compile_key key;
 346
 347    SVGA_STATS_TIME_PUSH(svga_sws(svga), SVGA_STATS_TIME_EMITVS);
 348
 349    /* If there is an active geometry shader, and it has stream output
 350     * defined, then we will skip the stream output from the vertex shader
 351     */
 352    if (!svga_have_gs_streamout(svga)) {
 353       /* No GS stream out */
 354       if (svga_have_vs_streamout(svga)) {
 355          /* Set VS stream out */
 356          svga_set_stream_output(svga, vs->base.stream_output);
 357       }
 358       else {
 359          /* turn off stream out */
 360          svga_set_stream_output(svga, NULL);
 361       }
 362    }
 363
 364    /* SVGA_NEW_NEED_SWTNL */
 365    if (svga->state.sw.need_swtnl && !svga_have_vgpu10(svga)) {
 366       /* No vertex shader is needed */
 367       variant = NULL;
 368    }
 369    else {
 370       make_vs_key(svga, &key);
 371
 372       /* See if we already have a VS variant that matches the key */
 373       variant = svga_search_shader_key(&vs->base, &key);
 374
 375       if (!variant) {
 376          /* Create VS variant now */
 377          if (key.vs.passthrough) {
 378             ret = compile_passthrough_vs(svga, vs, fs, &variant);
 379          }
 380          else {
 381             ret = compile_vs(svga, vs, &key, &variant);
 382          }
 383          if (ret != PIPE_OK)
 384             goto done;
 385
 386          /* insert the new variant at head of linked list */
 387          assert(variant);
 388          variant->next = vs->base.variants;
 389          vs->base.variants = variant;
 390       }
 391    }
 392
 393    if (variant != svga->state.hw_draw.vs) {
 394       /* Bind the new variant */
 395       if (variant) {
 396          ret = svga_set_shader(svga, SVGA3D_SHADERTYPE_VS, variant);
 397          if (ret != PIPE_OK)
 398             goto done;
 399          svga->rebind.flags.vs = FALSE;
 400       }
 401
 402       svga->dirty |= SVGA_NEW_VS_VARIANT;
 403       svga->state.hw_draw.vs = variant;
 404    }
 405
 406 done:
 407    SVGA_STATS_TIME_POP(svga_sws(svga));
 408    return ret;
 409 }
 410
 411 struct svga_tracked_state svga_hw_vs =
 412 {
 413    "vertex shader (hwtnl)",
 414    (SVGA_NEW_VS |
 415     SVGA_NEW_FS |
 416     SVGA_NEW_TEXTURE_BINDING |
 417     SVGA_NEW_SAMPLER |
 418     SVGA_NEW_RAST |
 419     SVGA_NEW_PRESCALE |
 420     SVGA_NEW_VELEMENT |
 421     SVGA_NEW_NEED_SWTNL),
 422    emit_hw_vs
 423 };