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    unsigned num_tokens;
  56
  57    ureg = ureg_create(TGSI_PROCESSOR_VERTEX);
  58    if (!ureg)
  59       return NULL;
  60
  61    dst = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0);
  62    src = ureg_DECL_immediate(ureg, zero, 4);
  63    ureg_MOV(ureg, dst, src);
  64    ureg_END(ureg);
  65
  66    tokens = ureg_get_tokens(ureg, &num_tokens);
  67
  68    ureg_destroy(ureg);
  69
  70    return tokens;
  71 }
  72
  73
  74 static struct svga_shader_variant *
  75 translate_vertex_program(struct svga_context *svga,
  76                          const struct svga_vertex_shader *vs,
  77                          const struct svga_compile_key *key)
  78 {
  79    if (svga_have_vgpu10(svga)) {
  80       return svga_tgsi_vgpu10_translate(svga, &vs->base, key,
  81                                         PIPE_SHADER_VERTEX);
  82    }
  83    else {
  84       return svga_tgsi_vgpu9_translate(&vs->base, key, 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 unsigned 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    key->vs.need_prescale = svga->state.hw_clear.prescale.enabled &&
 176                            (svga->curr.gs == NULL);
 177    key->vs.allow_psiz = svga->curr.rast->templ.point_size_per_vertex;
 178
 179    /* SVGA_NEW_FS */
 180    key->vs.fs_generic_inputs = svga->curr.fs->generic_inputs;
 181
 182    svga_remap_generics(key->vs.fs_generic_inputs, key->generic_remap_table);
 183
 184    /* SVGA_NEW_VELEMENT */
 185    key->vs.adjust_attrib_range = svga->curr.velems->adjust_attrib_range;
 186    key->vs.adjust_attrib_w_1 = svga->curr.velems->adjust_attrib_w_1;
 187    key->vs.attrib_is_pure_int = svga->curr.velems->attrib_is_pure_int;
 188    key->vs.adjust_attrib_itof = svga->curr.velems->adjust_attrib_itof;
 189    key->vs.adjust_attrib_utof = svga->curr.velems->adjust_attrib_utof;
 190    key->vs.attrib_is_bgra = svga->curr.velems->attrib_is_bgra;
 191    key->vs.attrib_puint_to_snorm = svga->curr.velems->attrib_puint_to_snorm;
 192    key->vs.attrib_puint_to_uscaled = svga->curr.velems->attrib_puint_to_uscaled;
 193    key->vs.attrib_puint_to_sscaled = svga->curr.velems->attrib_puint_to_sscaled;
 194
 195    /* SVGA_NEW_TEXTURE_BINDING | SVGA_NEW_SAMPLER */
 196    svga_init_shader_key_common(svga, shader, key);
 197
 198    /* SVGA_NEW_RAST */
 199    key->clip_plane_enable = svga->curr.rast->templ.clip_plane_enable;
 200 }
 201
 202
 203 /**
 204  * svga_reemit_vs_bindings - Reemit the vertex shader bindings
 205  */
 206 enum pipe_error
 207 svga_reemit_vs_bindings(struct svga_context *svga)
 208 {
 209    enum pipe_error ret;
 210    struct svga_winsys_gb_shader *gbshader = NULL;
 211    SVGA3dShaderId shaderId = SVGA3D_INVALID_ID;
 212
 213    assert(svga->rebind.flags.vs);
 214    assert(svga_have_gb_objects(svga));
 215
 216    if (svga->state.hw_draw.vs) {
 217       gbshader = svga->state.hw_draw.vs->gb_shader;
 218       shaderId = svga->state.hw_draw.vs->id;
 219    }
 220
 221    if (!svga_need_to_rebind_resources(svga)) {
 222       ret =  svga->swc->resource_rebind(svga->swc, NULL, gbshader,
 223                                         SVGA_RELOC_READ);
 224       goto out;
 225    }
 226
 227    if (svga_have_vgpu10(svga))
 228       ret = SVGA3D_vgpu10_SetShader(svga->swc, SVGA3D_SHADERTYPE_VS,
 229                                     gbshader, shaderId);
 230    else
 231       ret = SVGA3D_SetGBShader(svga->swc, SVGA3D_SHADERTYPE_VS, gbshader);
 232
 233  out:
 234    if (ret != PIPE_OK)
 235       return ret;
 236
 237    svga->rebind.flags.vs = FALSE;
 238    return PIPE_OK;
 239 }
 240
 241
 242 /**
 243  * The current vertex shader is already executed by the 'draw'
 244  * module, so we just need to generate a simple vertex shader
 245  * to pass through all those VS outputs that will
 246  * be consumed by the fragment shader.
 247  * Used when we employ the 'draw' module.
 248  */
 249 static enum pipe_error
 250 compile_passthrough_vs(struct svga_context *svga,
 251                        struct svga_vertex_shader *vs,
 252                        struct svga_fragment_shader *fs,
 253                        struct svga_shader_variant **out_variant)
 254 {
 255    struct svga_shader_variant *variant = NULL;
 256    unsigned num_inputs;
 257    unsigned i;
 258    unsigned num_elements;
 259    struct svga_vertex_shader new_vs;
 260    struct ureg_src src[PIPE_MAX_SHADER_INPUTS];
 261    struct ureg_dst dst[PIPE_MAX_SHADER_OUTPUTS];
 262    struct ureg_program *ureg;
 263    unsigned num_tokens;
 264    struct svga_compile_key key;
 265    enum pipe_error ret;
 266
 267    assert(svga_have_vgpu10(svga));
 268    assert(fs);
 269
 270    num_inputs = fs->base.info.num_inputs;
 271
 272    ureg = ureg_create(TGSI_PROCESSOR_VERTEX);
 273    if (!ureg)
 274       return PIPE_ERROR_OUT_OF_MEMORY;
 275
 276    /* draw will always add position */
 277    dst[0] = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0);
 278    src[0] = ureg_DECL_vs_input(ureg, 0);
 279    num_elements = 1;
 280
 281    /**
 282     * swtnl backend redefines the input layout based on the
 283     * fragment shader's inputs. So we only need to passthrough
 284     * those inputs that will be consumed by the fragment shader.
 285     * Note: DX10 requires the number of vertex elements
 286     * specified in the input layout to be no less than the
 287     * number of inputs to the vertex shader.
 288     */
 289    for (i = 0; i < num_inputs; i++) {
 290       switch (fs->base.info.input_semantic_name[i]) {
 291       case TGSI_SEMANTIC_COLOR:
 292       case TGSI_SEMANTIC_GENERIC:
 293       case TGSI_SEMANTIC_FOG:
 294          dst[num_elements] = ureg_DECL_output(ureg,
 295                                 fs->base.info.input_semantic_name[i],
 296                                 fs->base.info.input_semantic_index[i]);
 297          src[num_elements] = ureg_DECL_vs_input(ureg, num_elements);
 298          num_elements++;
 299          break;
 300       default:
 301          break;
 302       }
 303    }
 304
 305    for (i = 0; i < num_elements; i++) {
 306       ureg_MOV(ureg, dst[i], src[i]);
 307    }
 308
 309    ureg_END(ureg);
 310
 311    memset(&new_vs, 0, sizeof(new_vs));
 312    new_vs.base.tokens = ureg_get_tokens(ureg, &num_tokens);
 313    tgsi_scan_shader(new_vs.base.tokens, &new_vs.base.info);
 314
 315    memset(&key, 0, sizeof(key));
 316    key.vs.undo_viewport = 1;
 317
 318    ret = compile_vs(svga, &new_vs, &key, &variant);
 319    if (ret != PIPE_OK)
 320       return ret;
 321
 322    ureg_free_tokens(new_vs.base.tokens);
 323    ureg_destroy(ureg);
 324
 325    /* Overwrite the variant key to indicate it's a pass-through VS */
 326    memset(&variant->key, 0, sizeof(variant->key));
 327    variant->key.vs.passthrough = 1;
 328    variant->key.vs.undo_viewport = 1;
 329
 330    *out_variant = variant;
 331
 332    return PIPE_OK;
 333 }
 334
 335
 336 static enum pipe_error
 337 emit_hw_vs(struct svga_context *svga, unsigned dirty)
 338 {
 339    struct svga_shader_variant *variant;
 340    struct svga_vertex_shader *vs = svga->curr.vs;
 341    struct svga_fragment_shader *fs = svga->curr.fs;
 342    enum pipe_error ret = PIPE_OK;
 343    struct svga_compile_key key;
 344
 345    /* If there is an active geometry shader, and it has stream output
 346     * defined, then we will skip the stream output from the vertex shader
 347     */
 348    if (!svga_have_gs_streamout(svga)) {
 349       /* No GS stream out */
 350       if (svga_have_vs_streamout(svga)) {
 351          /* Set VS stream out */
 352          svga_set_stream_output(svga, vs->base.stream_output);
 353       }
 354       else {
 355          /* turn off stream out */
 356          svga_set_stream_output(svga, NULL);
 357       }
 358    }
 359
 360    /* SVGA_NEW_NEED_SWTNL */
 361    if (svga->state.sw.need_swtnl && !svga_have_vgpu10(svga)) {
 362       /* No vertex shader is needed */
 363       variant = NULL;
 364    }
 365    else {
 366       make_vs_key(svga, &key);
 367
 368       /* See if we already have a VS variant that matches the key */
 369       variant = svga_search_shader_key(&vs->base, &key);
 370
 371       if (!variant) {
 372          /* Create VS variant now */
 373          if (key.vs.passthrough) {
 374             ret = compile_passthrough_vs(svga, vs, fs, &variant);
 375          }
 376          else {
 377             ret = compile_vs(svga, vs, &key, &variant);
 378          }
 379          if (ret != PIPE_OK)
 380             return ret;
 381
 382          /* insert the new variant at head of linked list */
 383          assert(variant);
 384          variant->next = vs->base.variants;
 385          vs->base.variants = variant;
 386       }
 387    }
 388
 389    if (variant != svga->state.hw_draw.vs) {
 390       /* Bind the new variant */
 391       if (variant) {
 392          ret = svga_set_shader(svga, SVGA3D_SHADERTYPE_VS, variant);
 393          if (ret != PIPE_OK)
 394             return ret;
 395          svga->rebind.flags.vs = FALSE;
 396       }
 397
 398       svga->dirty |= SVGA_NEW_VS_VARIANT;
 399       svga->state.hw_draw.vs = variant;
 400    }
 401
 402    return PIPE_OK;
 403 }
 404
 405 struct svga_tracked_state svga_hw_vs =
 406 {
 407    "vertex shader (hwtnl)",
 408    (SVGA_NEW_VS |
 409     SVGA_NEW_FS |
 410     SVGA_NEW_TEXTURE_BINDING |
 411     SVGA_NEW_SAMPLER |
 412     SVGA_NEW_RAST |
 413     SVGA_NEW_PRESCALE |
 414     SVGA_NEW_VELEMENT |
 415     SVGA_NEW_NEED_SWTNL),
 416    emit_hw_vs
 417 };