src/mesa/drivers/dri/i965/brw_vs.c

   1 /*
   2  Copyright (C) Intel Corp.  2006.  All Rights Reserved.
   3  Intel funded Tungsten Graphics to
   4  develop this 3D driver.
   5
   6  Permission is hereby granted, free of charge, to any person obtaining
   7  a copy of this software and associated documentation files (the
   8  "Software"), to deal in the Software without restriction, including
   9  without limitation the rights to use, copy, modify, merge, publish,
  10  distribute, sublicense, and/or sell copies of the Software, and to
  11  permit persons to whom the Software is furnished to do so, subject to
  12  the following conditions:
  13
  14  The above copyright notice and this permission notice (including the
  15  next paragraph) shall be included in all copies or substantial
  16  portions of the Software.
  17
  18  THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  19  EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  20  MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
  21  IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
  22  LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
  23  OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
  24  WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
  25
  26  **********************************************************************/
  27  /*
  28   * Authors:
  29   *   Keith Whitwell <keithw@vmware.com>
  30   */
  31
  32
  33 #include "main/compiler.h"
  34 #include "brw_context.h"
  35 #include "brw_vs.h"
  36 #include "brw_util.h"
  37 #include "brw_state.h"
  38 #include "program/prog_print.h"
  39 #include "program/prog_parameter.h"
  40
  41 #include "util/ralloc.h"
  42
  43 /**
  44  * Decide which set of clip planes should be used when clipping via
  45  * gl_Position or gl_ClipVertex.
  46  */
  47 gl_clip_plane *brw_select_clip_planes(struct gl_context *ctx)
  48 {
  49    if (ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX]) {
  50       /* There is currently a GLSL vertex shader, so clip according to GLSL
  51        * rules, which means compare gl_ClipVertex (or gl_Position, if
  52        * gl_ClipVertex wasn't assigned) against the eye-coordinate clip planes
  53        * that were stored in EyeUserPlane at the time the clip planes were
  54        * specified.
  55        */
  56       return ctx->Transform.EyeUserPlane;
  57    } else {
  58       /* Either we are using fixed function or an ARB vertex program.  In
  59        * either case the clip planes are going to be compared against
  60        * gl_Position (which is in clip coordinates) so we have to clip using
  61        * _ClipUserPlane, which was transformed into clip coordinates by Mesa
  62        * core.
  63        */
  64       return ctx->Transform._ClipUserPlane;
  65    }
  66 }
  67
  68
  69 bool
  70 brw_vs_prog_data_compare(const void *in_a, const void *in_b)
  71 {
  72    const struct brw_vs_prog_data *a = in_a;
  73    const struct brw_vs_prog_data *b = in_b;
  74
  75    /* Compare the base structure. */
  76    if (!brw_stage_prog_data_compare(&a->base.base, &b->base.base))
  77       return false;
  78
  79    /* Compare the rest of the struct. */
  80    const unsigned offset = sizeof(struct brw_stage_prog_data);
  81    if (memcmp(((char *) a) + offset, ((char *) b) + offset,
  82               sizeof(struct brw_vs_prog_data) - offset)) {
  83       return false;
  84    }
  85
  86    return true;
  87 }
  88
  89 bool
  90 brw_codegen_vs_prog(struct brw_context *brw,
  91                     struct gl_shader_program *prog,
  92                     struct brw_vertex_program *vp,
  93                     struct brw_vs_prog_key *key)
  94 {
  95    GLuint program_size;
  96    const GLuint *program;
  97    struct brw_vs_prog_data prog_data;
  98    struct brw_stage_prog_data *stage_prog_data = &prog_data.base.base;
  99    void *mem_ctx;
 100    int i;
 101    struct gl_shader *vs = NULL;
 102
 103    if (prog)
 104       vs = prog->_LinkedShaders[MESA_SHADER_VERTEX];
 105
 106    memset(&prog_data, 0, sizeof(prog_data));
 107
 108    /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
 109    if (!prog)
 110       stage_prog_data->use_alt_mode = true;
 111
 112    mem_ctx = ralloc_context(NULL);
 113
 114    /* Allocate the references to the uniforms that will end up in the
 115     * prog_data associated with the compiled program, and which will be freed
 116     * by the state cache.
 117     */
 118    int param_count;
 119    if (vs) {
 120       /* We add padding around uniform values below vec4 size, with the worst
 121        * case being a float value that gets blown up to a vec4, so be
 122        * conservative here.
 123        */
 124       param_count = vs->num_uniform_components * 4;
 125
 126    } else {
 127       param_count = vp->program.Base.Parameters->NumParameters * 4;
 128    }
 129    /* vec4_visitor::setup_uniform_clipplane_values() also uploads user clip
 130     * planes as uniforms.
 131     */
 132    param_count += key->base.nr_userclip_plane_consts * 4;
 133
 134    stage_prog_data->param =
 135       rzalloc_array(NULL, const gl_constant_value *, param_count);
 136    stage_prog_data->pull_param =
 137       rzalloc_array(NULL, const gl_constant_value *, param_count);
 138    stage_prog_data->nr_params = param_count;
 139
 140    GLbitfield64 outputs_written = vp->program.Base.OutputsWritten;
 141    prog_data.inputs_read = vp->program.Base.InputsRead;
 142
 143    if (key->copy_edgeflag) {
 144       outputs_written |= BITFIELD64_BIT(VARYING_SLOT_EDGE);
 145       prog_data.inputs_read |= VERT_BIT_EDGEFLAG;
 146    }
 147
 148    if (brw->gen < 6) {
 149       /* Put dummy slots into the VUE for the SF to put the replaced
 150        * point sprite coords in.  We shouldn't need these dummy slots,
 151        * which take up precious URB space, but it would mean that the SF
 152        * doesn't get nice aligned pairs of input coords into output
 153        * coords, which would be a pain to handle.
 154        */
 155       for (i = 0; i < 8; i++) {
 156          if (key->point_coord_replace & (1 << i))
 157             outputs_written |= BITFIELD64_BIT(VARYING_SLOT_TEX0 + i);
 158       }
 159
 160       /* if back colors are written, allocate slots for front colors too */
 161       if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC0))
 162          outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL0);
 163       if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC1))
 164          outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL1);
 165    }
 166
 167    /* In order for legacy clipping to work, we need to populate the clip
 168     * distance varying slots whenever clipping is enabled, even if the vertex
 169     * shader doesn't write to gl_ClipDistance.
 170     */
 171    if (key->base.userclip_active) {
 172       outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0);
 173       outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1);
 174    }
 175
 176    brw_compute_vue_map(brw->intelScreen->devinfo,
 177                        &prog_data.base.vue_map, outputs_written);
 178
 179    if (0) {
 180       _mesa_fprint_program_opt(stderr, &vp->program.Base, PROG_PRINT_DEBUG,
 181                                true);
 182    }
 183
 184    /* Emit GEN4 code.
 185     */
 186    program = brw_vs_emit(brw, mem_ctx, key, &prog_data,
 187                          &vp->program, prog, &program_size);
 188    if (program == NULL) {
 189       ralloc_free(mem_ctx);
 190       return false;
 191    }
 192
 193    /* Scratch space is used for register spilling */
 194    if (prog_data.base.base.total_scratch) {
 195       brw_get_scratch_bo(brw, &brw->vs.base.scratch_bo,
 196                          prog_data.base.base.total_scratch *
 197                          brw->max_vs_threads);
 198    }
 199
 200    brw_upload_cache(&brw->cache, BRW_CACHE_VS_PROG,
 201                     key, sizeof(struct brw_vs_prog_key),
 202                     program, program_size,
 203                     &prog_data, sizeof(prog_data),
 204                     &brw->vs.base.prog_offset, &brw->vs.prog_data);
 205    ralloc_free(mem_ctx);
 206
 207    return true;
 208 }
 209
 210 static bool
 211 key_debug(struct brw_context *brw, const char *name, int a, int b)
 212 {
 213    if (a != b) {
 214       perf_debug("  %s %d->%d\n", name, a, b);
 215       return true;
 216    }
 217    return false;
 218 }
 219
 220 void
 221 brw_vs_debug_recompile(struct brw_context *brw,
 222                        struct gl_shader_program *prog,
 223                        const struct brw_vs_prog_key *key)
 224 {
 225    struct brw_cache_item *c = NULL;
 226    const struct brw_vs_prog_key *old_key = NULL;
 227    bool found = false;
 228
 229    perf_debug("Recompiling vertex shader for program %d\n", prog->Name);
 230
 231    for (unsigned int i = 0; i < brw->cache.size; i++) {
 232       for (c = brw->cache.items[i]; c; c = c->next) {
 233          if (c->cache_id == BRW_CACHE_VS_PROG) {
 234             old_key = c->key;
 235
 236             if (old_key->base.program_string_id == key->base.program_string_id)
 237                break;
 238          }
 239       }
 240       if (c)
 241          break;
 242    }
 243
 244    if (!c) {
 245       perf_debug("  Didn't find previous compile in the shader cache for "
 246                  "debug\n");
 247       return;
 248    }
 249
 250    for (unsigned int i = 0; i < VERT_ATTRIB_MAX; i++) {
 251       found |= key_debug(brw, "Vertex attrib w/a flags",
 252                          old_key->gl_attrib_wa_flags[i],
 253                          key->gl_attrib_wa_flags[i]);
 254    }
 255
 256    found |= key_debug(brw, "user clip flags",
 257                       old_key->base.userclip_active, key->base.userclip_active);
 258
 259    found |= key_debug(brw, "user clipping planes as push constants",
 260                       old_key->base.nr_userclip_plane_consts,
 261                       key->base.nr_userclip_plane_consts);
 262
 263    found |= key_debug(brw, "copy edgeflag",
 264                       old_key->copy_edgeflag, key->copy_edgeflag);
 265    found |= key_debug(brw, "PointCoord replace",
 266                       old_key->point_coord_replace, key->point_coord_replace);
 267    found |= key_debug(brw, "vertex color clamping",
 268                       old_key->clamp_vertex_color, key->clamp_vertex_color);
 269
 270    found |= brw_debug_recompile_sampler_key(brw, &old_key->base.tex,
 271                                             &key->base.tex);
 272
 273    if (!found) {
 274       perf_debug("  Something else\n");
 275    }
 276 }
 277
 278
 279 void
 280 brw_setup_vue_key_clip_info(struct brw_context *brw,
 281                             struct brw_vue_prog_key *key,
 282                             bool program_uses_clip_distance)
 283 {
 284    struct gl_context *ctx = &brw->ctx;
 285
 286    key->userclip_active = (ctx->Transform.ClipPlanesEnabled != 0);
 287    if (key->userclip_active && !program_uses_clip_distance) {
 288       key->nr_userclip_plane_consts
 289          = _mesa_logbase2(ctx->Transform.ClipPlanesEnabled) + 1;
 290    }
 291 }
 292
 293 static bool
 294 brw_vs_state_dirty(struct brw_context *brw)
 295 {
 296    return brw_state_dirty(brw,
 297                           _NEW_BUFFERS |
 298                           _NEW_LIGHT |
 299                           _NEW_POINT |
 300                           _NEW_POLYGON |
 301                           _NEW_TEXTURE |
 302                           _NEW_TRANSFORM,
 303                           BRW_NEW_VERTEX_PROGRAM |
 304                           BRW_NEW_VS_ATTRIB_WORKAROUNDS);
 305 }
 306
 307 static void
 308 brw_vs_populate_key(struct brw_context *brw,
 309                     struct brw_vs_prog_key *key)
 310 {
 311    struct gl_context *ctx = &brw->ctx;
 312    /* BRW_NEW_VERTEX_PROGRAM */
 313    struct brw_vertex_program *vp =
 314       (struct brw_vertex_program *)brw->vertex_program;
 315    struct gl_program *prog = (struct gl_program *) brw->vertex_program;
 316    int i;
 317
 318    memset(key, 0, sizeof(*key));
 319
 320    /* Just upload the program verbatim for now.  Always send it all
 321     * the inputs it asks for, whether they are varying or not.
 322     */
 323    key->base.program_string_id = vp->id;
 324    brw_setup_vue_key_clip_info(brw, &key->base,
 325                                vp->program.Base.UsesClipDistanceOut);
 326
 327    /* _NEW_POLYGON */
 328    if (brw->gen < 6) {
 329       key->copy_edgeflag = (ctx->Polygon.FrontMode != GL_FILL ||
 330                             ctx->Polygon.BackMode != GL_FILL);
 331    }
 332
 333    if (prog->OutputsWritten & (VARYING_BIT_COL0 | VARYING_BIT_COL1 |
 334                                VARYING_BIT_BFC0 | VARYING_BIT_BFC1)) {
 335       /* _NEW_LIGHT | _NEW_BUFFERS */
 336       key->clamp_vertex_color = ctx->Light._ClampVertexColor;
 337    }
 338
 339    /* _NEW_POINT */
 340    if (brw->gen < 6 && ctx->Point.PointSprite) {
 341       for (i = 0; i < 8; i++) {
 342          if (ctx->Point.CoordReplace[i])
 343             key->point_coord_replace |= (1 << i);
 344       }
 345    }
 346
 347    /* _NEW_TEXTURE */
 348    brw_populate_sampler_prog_key_data(ctx, prog, brw->vs.base.sampler_count,
 349                                       &key->base.tex);
 350
 351    /* BRW_NEW_VS_ATTRIB_WORKAROUNDS */
 352    memcpy(key->gl_attrib_wa_flags, brw->vb.attrib_wa_flags,
 353           sizeof(brw->vb.attrib_wa_flags));
 354 }
 355
 356 void
 357 brw_upload_vs_prog(struct brw_context *brw)
 358 {
 359    struct gl_context *ctx = &brw->ctx;
 360    struct gl_shader_program **current = ctx->_Shader->CurrentProgram;
 361    struct brw_vs_prog_key key;
 362    /* BRW_NEW_VERTEX_PROGRAM */
 363    struct brw_vertex_program *vp =
 364       (struct brw_vertex_program *)brw->vertex_program;
 365
 366    if (!brw_vs_state_dirty(brw))
 367       return;
 368
 369    brw_vs_populate_key(brw, &key);
 370
 371    if (!brw_search_cache(&brw->cache, BRW_CACHE_VS_PROG,
 372                          &key, sizeof(key),
 373                          &brw->vs.base.prog_offset, &brw->vs.prog_data)) {
 374       bool success = brw_codegen_vs_prog(brw, current[MESA_SHADER_VERTEX],
 375                                          vp, &key);
 376       (void) success;
 377       assert(success);
 378    }
 379    brw->vs.base.prog_data = &brw->vs.prog_data->base.base;
 380
 381    if (memcmp(&brw->vs.prog_data->base.vue_map, &brw->vue_map_geom_out,
 382               sizeof(brw->vue_map_geom_out)) != 0) {
 383       brw->vue_map_vs = brw->vs.prog_data->base.vue_map;
 384       brw->ctx.NewDriverState |= BRW_NEW_VUE_MAP_VS;
 385       if (brw->gen < 6) {
 386          /* No geometry shader support, so the VS VUE map is the VUE map for
 387           * the output of the "geometry" portion of the pipeline.
 388           */
 389          brw->vue_map_geom_out = brw->vue_map_vs;
 390          brw->ctx.NewDriverState |= BRW_NEW_VUE_MAP_GEOM_OUT;
 391       }
 392    }
 393 }
 394
 395 bool
 396 brw_vs_precompile(struct gl_context *ctx,
 397                   struct gl_shader_program *shader_prog,
 398                   struct gl_program *prog)
 399 {
 400    struct brw_context *brw = brw_context(ctx);
 401    struct brw_vs_prog_key key;
 402    uint32_t old_prog_offset = brw->vs.base.prog_offset;
 403    struct brw_vs_prog_data *old_prog_data = brw->vs.prog_data;
 404    bool success;
 405
 406    struct gl_vertex_program *vp = (struct gl_vertex_program *) prog;
 407    struct brw_vertex_program *bvp = brw_vertex_program(vp);
 408
 409    memset(&key, 0, sizeof(key));
 410
 411    brw_vue_setup_prog_key_for_precompile(ctx, &key.base, bvp->id, &vp->Base);
 412    key.clamp_vertex_color =
 413       (prog->OutputsWritten & (VARYING_BIT_COL0 | VARYING_BIT_COL1 |
 414                                VARYING_BIT_BFC0 | VARYING_BIT_BFC1));
 415
 416    success = brw_codegen_vs_prog(brw, shader_prog, bvp, &key);
 417
 418    brw->vs.base.prog_offset = old_prog_offset;
 419    brw->vs.prog_data = old_prog_data;
 420
 421    return success;
 422 }