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    tgsi_scan_shader(vs->base.tokens, &vs->base.info);
 109    vs->generic_outputs = svga_get_generic_outputs_mask(&vs->base.info);
 110
 111    variant = translate_vertex_program(svga, vs, key);
 112    return variant;
 113 }
 114
 115
 116 /**
 117  * Translate TGSI shader into an svga shader variant.
 118  */
 119 static enum pipe_error
 120 compile_vs(struct svga_context *svga,
 121            struct svga_vertex_shader *vs,
 122            const struct svga_compile_key *key,
 123            struct svga_shader_variant **out_variant)
 124 {
 125    struct svga_shader_variant *variant;
 126    enum pipe_error ret = PIPE_ERROR;
 127
 128    variant = translate_vertex_program(svga, vs, key);
 129    if (variant == NULL) {
 130       debug_printf("Failed to compile vertex shader,"
 131                    " using dummy shader instead.\n");
 132       variant = get_compiled_dummy_vertex_shader(svga, vs, key);
 133    }
 134    else if (svga_shader_too_large(svga, variant)) {
 135       /* too big, use dummy shader */
 136       debug_printf("Shader too large (%u bytes),"
 137                    " using dummy shader instead.\n",
 138                    (unsigned) (variant->nr_tokens
 139                                * sizeof(variant->tokens[0])));
 140       /* Free the too-large variant */
 141       svga_destroy_shader_variant(svga, variant);
 142       /* Use simple pass-through shader instead */
 143       variant = get_compiled_dummy_vertex_shader(svga, vs, key);
 144    }
 145
 146    if (!variant) {
 147       return PIPE_ERROR;
 148    }
 149
 150    ret = svga_define_shader(svga, variant);
 151    if (ret != PIPE_OK) {
 152       svga_destroy_shader_variant(svga, variant);
 153       return ret;
 154    }
 155
 156    *out_variant = variant;
 157
 158    return PIPE_OK;
 159 }
 160
 161
 162 /* SVGA_NEW_PRESCALE, SVGA_NEW_RAST, SVGA_NEW_FS
 163  */
 164 static void
 165 make_vs_key(struct svga_context *svga, struct svga_compile_key *key)
 166 {
 167    struct svga_vertex_shader *vs = svga->curr.vs;
 168
 169    memset(key, 0, sizeof *key);
 170
 171    if (svga->state.sw.need_swtnl && svga_have_vgpu10(svga)) {
 172       /* Set both of these flags, to match compile_passthrough_vs() */
 173       key->vs.passthrough = 1;
 174       key->vs.undo_viewport = 1;
 175       return;
 176    }
 177
 178    /* SVGA_NEW_PRESCALE */
 179    key->vs.need_prescale = svga->state.hw_clear.prescale[0].enabled &&
 180                            (svga->curr.tes == NULL) &&
 181                            (svga->curr.gs == NULL);
 182
 183    /* SVGA_NEW_RAST */
 184    key->vs.allow_psiz = svga->curr.rast->templ.point_size_per_vertex;
 185
 186    /* SVGA_NEW_FS */
 187    key->vs.fs_generic_inputs = svga->curr.fs->generic_inputs;
 188
 189    svga_remap_generics(key->vs.fs_generic_inputs, key->generic_remap_table);
 190
 191    /* SVGA_NEW_VELEMENT */
 192    key->vs.adjust_attrib_range = svga->curr.velems->adjust_attrib_range;
 193    key->vs.adjust_attrib_w_1 = svga->curr.velems->adjust_attrib_w_1;
 194    key->vs.attrib_is_pure_int = svga->curr.velems->attrib_is_pure_int;
 195    key->vs.adjust_attrib_itof = svga->curr.velems->adjust_attrib_itof;
 196    key->vs.adjust_attrib_utof = svga->curr.velems->adjust_attrib_utof;
 197    key->vs.attrib_is_bgra = svga->curr.velems->attrib_is_bgra;
 198    key->vs.attrib_puint_to_snorm = svga->curr.velems->attrib_puint_to_snorm;
 199    key->vs.attrib_puint_to_uscaled = svga->curr.velems->attrib_puint_to_uscaled;
 200    key->vs.attrib_puint_to_sscaled = svga->curr.velems->attrib_puint_to_sscaled;
 201
 202    /* SVGA_NEW_TEXTURE_BINDING | SVGA_NEW_SAMPLER */
 203    svga_init_shader_key_common(svga, PIPE_SHADER_VERTEX, &vs->base, key);
 204
 205    /* SVGA_NEW_RAST */
 206    key->clip_plane_enable = svga->curr.rast->templ.clip_plane_enable;
 207
 208    /* Determine if this shader is the last shader in the vertex
 209     * processing stage.
 210     */
 211    key->last_vertex_stage = !(svga->curr.gs ||
 212                               svga->curr.tcs || svga->curr.tes);
 213 }
 214
 215
 216 /**
 217  * svga_reemit_vs_bindings - Reemit the vertex shader bindings
 218  */
 219 enum pipe_error
 220 svga_reemit_vs_bindings(struct svga_context *svga)
 221 {
 222    enum pipe_error ret;
 223    struct svga_winsys_gb_shader *gbshader = NULL;
 224    SVGA3dShaderId shaderId = SVGA3D_INVALID_ID;
 225
 226    assert(svga->rebind.flags.vs);
 227    assert(svga_have_gb_objects(svga));
 228
 229    if (svga->state.hw_draw.vs) {
 230       gbshader = svga->state.hw_draw.vs->gb_shader;
 231       shaderId = svga->state.hw_draw.vs->id;
 232    }
 233
 234    if (!svga_need_to_rebind_resources(svga)) {
 235       ret =  svga->swc->resource_rebind(svga->swc, NULL, gbshader,
 236                                         SVGA_RELOC_READ);
 237    }
 238    else {
 239       if (svga_have_vgpu10(svga))
 240          ret = SVGA3D_vgpu10_SetShader(svga->swc, SVGA3D_SHADERTYPE_VS,
 241                                        gbshader, shaderId);
 242       else
 243          ret = SVGA3D_SetGBShader(svga->swc, SVGA3D_SHADERTYPE_VS, gbshader);
 244    }
 245
 246    if (ret != PIPE_OK)
 247       return ret;
 248
 249    svga->rebind.flags.vs = FALSE;
 250    return PIPE_OK;
 251 }
 252
 253
 254 /**
 255  * The current vertex shader is already executed by the 'draw'
 256  * module, so we just need to generate a simple vertex shader
 257  * to pass through all those VS outputs that will
 258  * be consumed by the fragment shader.
 259  * Used when we employ the 'draw' module.
 260  */
 261 static enum pipe_error
 262 compile_passthrough_vs(struct svga_context *svga,
 263                        struct svga_vertex_shader *vs,
 264                        struct svga_fragment_shader *fs,
 265                        struct svga_shader_variant **out_variant)
 266 {
 267    struct svga_shader_variant *variant = NULL;
 268    unsigned num_inputs;
 269    unsigned i;
 270    unsigned num_elements;
 271    struct svga_vertex_shader new_vs;
 272    struct ureg_src src[PIPE_MAX_SHADER_INPUTS];
 273    struct ureg_dst dst[PIPE_MAX_SHADER_OUTPUTS];
 274    struct ureg_program *ureg;
 275    struct svga_compile_key key;
 276    enum pipe_error ret;
 277
 278    assert(svga_have_vgpu10(svga));
 279    assert(fs);
 280
 281    num_inputs = fs->base.info.num_inputs;
 282
 283    ureg = ureg_create(PIPE_SHADER_VERTEX);
 284    if (!ureg)
 285       return PIPE_ERROR_OUT_OF_MEMORY;
 286
 287    /* draw will always add position */
 288    dst[0] = ureg_DECL_output(ureg, TGSI_SEMANTIC_POSITION, 0);
 289    src[0] = ureg_DECL_vs_input(ureg, 0);
 290    num_elements = 1;
 291
 292    /**
 293     * swtnl backend redefines the input layout based on the
 294     * fragment shader's inputs. So we only need to passthrough
 295     * those inputs that will be consumed by the fragment shader.
 296     * Note: DX10 requires the number of vertex elements
 297     * specified in the input layout to be no less than the
 298     * number of inputs to the vertex shader.
 299     */
 300    for (i = 0; i < num_inputs; i++) {
 301       switch (fs->base.info.input_semantic_name[i]) {
 302       case TGSI_SEMANTIC_COLOR:
 303       case TGSI_SEMANTIC_GENERIC:
 304       case TGSI_SEMANTIC_FOG:
 305          dst[num_elements] = ureg_DECL_output(ureg,
 306                                 fs->base.info.input_semantic_name[i],
 307                                 fs->base.info.input_semantic_index[i]);
 308          src[num_elements] = ureg_DECL_vs_input(ureg, num_elements);
 309          num_elements++;
 310          break;
 311       default:
 312          break;
 313       }
 314    }
 315
 316    for (i = 0; i < num_elements; i++) {
 317       ureg_MOV(ureg, dst[i], src[i]);
 318    }
 319
 320    ureg_END(ureg);
 321
 322    memset(&new_vs, 0, sizeof(new_vs));
 323    new_vs.base.tokens = ureg_get_tokens(ureg, NULL);
 324    tgsi_scan_shader(new_vs.base.tokens, &new_vs.base.info);
 325
 326    memset(&key, 0, sizeof(key));
 327    key.vs.undo_viewport = 1;
 328
 329    ret = compile_vs(svga, &new_vs, &key, &variant);
 330    if (ret != PIPE_OK)
 331       return ret;
 332
 333    ureg_free_tokens(new_vs.base.tokens);
 334    ureg_destroy(ureg);
 335
 336    /* Overwrite the variant key to indicate it's a pass-through VS */
 337    memset(&variant->key, 0, sizeof(variant->key));
 338    variant->key.vs.passthrough = 1;
 339    variant->key.vs.undo_viewport = 1;
 340
 341    *out_variant = variant;
 342
 343    return PIPE_OK;
 344 }
 345
 346
 347 static enum pipe_error
 348 emit_hw_vs(struct svga_context *svga, uint64_t dirty)
 349 {
 350    struct svga_shader_variant *variant;
 351    struct svga_vertex_shader *vs = svga->curr.vs;
 352    struct svga_fragment_shader *fs = svga->curr.fs;
 353    enum pipe_error ret = PIPE_OK;
 354    struct svga_compile_key key;
 355
 356    SVGA_STATS_TIME_PUSH(svga_sws(svga), SVGA_STATS_TIME_EMITVS);
 357
 358    /* If there is an active geometry shader, and it has stream output
 359     * defined, then we will skip the stream output from the vertex shader
 360     */
 361    if (!svga_have_gs_streamout(svga)) {
 362       /* No GS stream out */
 363       if (svga_have_vs_streamout(svga)) {
 364          /* Set VS stream out */
 365          ret = svga_set_stream_output(svga, vs->base.stream_output);
 366       }
 367       else {
 368          /* turn off stream out */
 369          ret = svga_set_stream_output(svga, NULL);
 370       }
 371       if (ret != PIPE_OK) {
 372          goto done;
 373       }
 374    }
 375
 376    /* SVGA_NEW_NEED_SWTNL */
 377    if (svga->state.sw.need_swtnl && !svga_have_vgpu10(svga)) {
 378       /* No vertex shader is needed */
 379       variant = NULL;
 380    }
 381    else {
 382       make_vs_key(svga, &key);
 383
 384       /* See if we already have a VS variant that matches the key */
 385       variant = svga_search_shader_key(&vs->base, &key);
 386
 387       if (!variant) {
 388          /* Create VS variant now */
 389          if (key.vs.passthrough) {
 390             ret = compile_passthrough_vs(svga, vs, fs, &variant);
 391          }
 392          else {
 393             ret = compile_vs(svga, vs, &key, &variant);
 394          }
 395          if (ret != PIPE_OK)
 396             goto done;
 397
 398          /* insert the new variant at head of linked list */
 399          assert(variant);
 400          variant->next = vs->base.variants;
 401          vs->base.variants = variant;
 402       }
 403    }
 404
 405    if (variant != svga->state.hw_draw.vs) {
 406       /* Bind the new variant */
 407       if (variant) {
 408          ret = svga_set_shader(svga, SVGA3D_SHADERTYPE_VS, variant);
 409          if (ret != PIPE_OK)
 410             goto done;
 411          svga->rebind.flags.vs = FALSE;
 412       }
 413
 414       svga->dirty |= SVGA_NEW_VS_VARIANT;
 415       svga->state.hw_draw.vs = variant;
 416    }
 417
 418 done:
 419    SVGA_STATS_TIME_POP(svga_sws(svga));
 420    return ret;
 421 }
 422
 423 struct svga_tracked_state svga_hw_vs =
 424 {
 425    "vertex shader (hwtnl)",
 426    (SVGA_NEW_VS |
 427     SVGA_NEW_FS |
 428     SVGA_NEW_TEXTURE_BINDING |
 429     SVGA_NEW_SAMPLER |
 430     SVGA_NEW_RAST |
 431     SVGA_NEW_PRESCALE |
 432     SVGA_NEW_VELEMENT |
 433     SVGA_NEW_NEED_SWTNL),
 434    emit_hw_vs
 435 };