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