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 brw_shader *vs = NULL;
 102    bool start_busy = false;
 103    double start_time = 0;
 104
 105    if (prog)
 106       vs = (struct brw_shader *) prog->_LinkedShaders[MESA_SHADER_VERTEX];
 107
 108    memset(&prog_data, 0, sizeof(prog_data));
 109
 110    /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
 111    if (!prog)
 112       stage_prog_data->use_alt_mode = true;
 113
 114    mem_ctx = ralloc_context(NULL);
 115
 116    /* Allocate the references to the uniforms that will end up in the
 117     * prog_data associated with the compiled program, and which will be freed
 118     * by the state cache.
 119     */
 120    int param_count;
 121    if (vs) {
 122       /* We add padding around uniform values below vec4 size, with the worst
 123        * case being a float value that gets blown up to a vec4, so be
 124        * conservative here.
 125        */
 126       param_count = vs->base.num_uniform_components * 4 +
 127                     vs->base.NumImages * BRW_IMAGE_PARAM_SIZE;
 128       stage_prog_data->nr_image_params = vs->base.NumImages;
 129    } else {
 130       param_count = vp->program.Base.Parameters->NumParameters * 4;
 131    }
 132    /* vec4_visitor::setup_uniform_clipplane_values() also uploads user clip
 133     * planes as uniforms.
 134     */
 135    param_count += key->nr_userclip_plane_consts * 4;
 136
 137    stage_prog_data->param =
 138       rzalloc_array(NULL, const gl_constant_value *, param_count);
 139    stage_prog_data->pull_param =
 140       rzalloc_array(NULL, const gl_constant_value *, param_count);
 141    stage_prog_data->image_param =
 142       rzalloc_array(NULL, struct brw_image_param,
 143                     stage_prog_data->nr_image_params);
 144    stage_prog_data->nr_params = param_count;
 145
 146    GLbitfield64 outputs_written = vp->program.Base.OutputsWritten;
 147    prog_data.inputs_read = vp->program.Base.InputsRead;
 148
 149    if (key->copy_edgeflag) {
 150       outputs_written |= BITFIELD64_BIT(VARYING_SLOT_EDGE);
 151       prog_data.inputs_read |= VERT_BIT_EDGEFLAG;
 152    }
 153
 154    if (brw->gen < 6) {
 155       /* Put dummy slots into the VUE for the SF to put the replaced
 156        * point sprite coords in.  We shouldn't need these dummy slots,
 157        * which take up precious URB space, but it would mean that the SF
 158        * doesn't get nice aligned pairs of input coords into output
 159        * coords, which would be a pain to handle.
 160        */
 161       for (i = 0; i < 8; i++) {
 162          if (key->point_coord_replace & (1 << i))
 163             outputs_written |= BITFIELD64_BIT(VARYING_SLOT_TEX0 + i);
 164       }
 165
 166       /* if back colors are written, allocate slots for front colors too */
 167       if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC0))
 168          outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL0);
 169       if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC1))
 170          outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL1);
 171    }
 172
 173    /* In order for legacy clipping to work, we need to populate the clip
 174     * distance varying slots whenever clipping is enabled, even if the vertex
 175     * shader doesn't write to gl_ClipDistance.
 176     */
 177    if (key->nr_userclip_plane_consts > 0) {
 178       outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0);
 179       outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1);
 180    }
 181
 182    brw_compute_vue_map(brw->intelScreen->devinfo,
 183                        &prog_data.base.vue_map, outputs_written);
 184
 185    if (0) {
 186       _mesa_fprint_program_opt(stderr, &vp->program.Base, PROG_PRINT_DEBUG,
 187                                true);
 188    }
 189
 190    if (unlikely(brw->perf_debug)) {
 191       start_busy = (brw->batch.last_bo &&
 192                     drm_intel_bo_busy(brw->batch.last_bo));
 193       start_time = get_time();
 194    }
 195
 196    /* Emit GEN4 code.
 197     */
 198    program = brw_vs_emit(brw, mem_ctx, key, &prog_data,
 199                          &vp->program, prog, &program_size);
 200    if (program == NULL) {
 201       ralloc_free(mem_ctx);
 202       return false;
 203    }
 204
 205    if (unlikely(brw->perf_debug) && vs) {
 206       if (vs->compiled_once) {
 207          brw_vs_debug_recompile(brw, prog, key);
 208       }
 209       if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
 210          perf_debug("VS compile took %.03f ms and stalled the GPU\n",
 211                     (get_time() - start_time) * 1000);
 212       }
 213       vs->compiled_once = true;
 214    }
 215
 216    /* Scratch space is used for register spilling */
 217    if (prog_data.base.base.total_scratch) {
 218       brw_get_scratch_bo(brw, &brw->vs.base.scratch_bo,
 219                          prog_data.base.base.total_scratch *
 220                          brw->max_vs_threads);
 221    }
 222
 223    brw_upload_cache(&brw->cache, BRW_CACHE_VS_PROG,
 224                     key, sizeof(struct brw_vs_prog_key),
 225                     program, program_size,
 226                     &prog_data, sizeof(prog_data),
 227                     &brw->vs.base.prog_offset, &brw->vs.prog_data);
 228    ralloc_free(mem_ctx);
 229
 230    return true;
 231 }
 232
 233 static bool
 234 key_debug(struct brw_context *brw, const char *name, int a, int b)
 235 {
 236    if (a != b) {
 237       perf_debug("  %s %d->%d\n", name, a, b);
 238       return true;
 239    }
 240    return false;
 241 }
 242
 243 void
 244 brw_vs_debug_recompile(struct brw_context *brw,
 245                        struct gl_shader_program *prog,
 246                        const struct brw_vs_prog_key *key)
 247 {
 248    struct brw_cache_item *c = NULL;
 249    const struct brw_vs_prog_key *old_key = NULL;
 250    bool found = false;
 251
 252    perf_debug("Recompiling vertex shader for program %d\n", prog->Name);
 253
 254    for (unsigned int i = 0; i < brw->cache.size; i++) {
 255       for (c = brw->cache.items[i]; c; c = c->next) {
 256          if (c->cache_id == BRW_CACHE_VS_PROG) {
 257             old_key = c->key;
 258
 259             if (old_key->program_string_id == key->program_string_id)
 260                break;
 261          }
 262       }
 263       if (c)
 264          break;
 265    }
 266
 267    if (!c) {
 268       perf_debug("  Didn't find previous compile in the shader cache for "
 269                  "debug\n");
 270       return;
 271    }
 272
 273    for (unsigned int i = 0; i < VERT_ATTRIB_MAX; i++) {
 274       found |= key_debug(brw, "Vertex attrib w/a flags",
 275                          old_key->gl_attrib_wa_flags[i],
 276                          key->gl_attrib_wa_flags[i]);
 277    }
 278
 279    found |= key_debug(brw, "legacy user clipping",
 280                       old_key->nr_userclip_plane_consts,
 281                       key->nr_userclip_plane_consts);
 282
 283    found |= key_debug(brw, "copy edgeflag",
 284                       old_key->copy_edgeflag, key->copy_edgeflag);
 285    found |= key_debug(brw, "PointCoord replace",
 286                       old_key->point_coord_replace, key->point_coord_replace);
 287    found |= key_debug(brw, "vertex color clamping",
 288                       old_key->clamp_vertex_color, key->clamp_vertex_color);
 289
 290    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
 291
 292    if (!found) {
 293       perf_debug("  Something else\n");
 294    }
 295 }
 296
 297 static bool
 298 brw_vs_state_dirty(struct brw_context *brw)
 299 {
 300    return brw_state_dirty(brw,
 301                           _NEW_BUFFERS |
 302                           _NEW_LIGHT |
 303                           _NEW_POINT |
 304                           _NEW_POLYGON |
 305                           _NEW_TEXTURE |
 306                           _NEW_TRANSFORM,
 307                           BRW_NEW_VERTEX_PROGRAM |
 308                           BRW_NEW_VS_ATTRIB_WORKAROUNDS);
 309 }
 310
 311 static void
 312 brw_vs_populate_key(struct brw_context *brw,
 313                     struct brw_vs_prog_key *key)
 314 {
 315    struct gl_context *ctx = &brw->ctx;
 316    /* BRW_NEW_VERTEX_PROGRAM */
 317    struct brw_vertex_program *vp =
 318       (struct brw_vertex_program *)brw->vertex_program;
 319    struct gl_program *prog = (struct gl_program *) brw->vertex_program;
 320    int i;
 321
 322    memset(key, 0, sizeof(*key));
 323
 324    /* Just upload the program verbatim for now.  Always send it all
 325     * the inputs it asks for, whether they are varying or not.
 326     */
 327    key->program_string_id = vp->id;
 328
 329    if (ctx->Transform.ClipPlanesEnabled != 0 &&
 330        ctx->API == API_OPENGL_COMPAT &&
 331        !vp->program.Base.UsesClipDistanceOut) {
 332       key->nr_userclip_plane_consts =
 333          _mesa_logbase2(ctx->Transform.ClipPlanesEnabled) + 1;
 334    }
 335
 336    /* _NEW_POLYGON */
 337    if (brw->gen < 6) {
 338       key->copy_edgeflag = (ctx->Polygon.FrontMode != GL_FILL ||
 339                             ctx->Polygon.BackMode != GL_FILL);
 340    }
 341
 342    if (prog->OutputsWritten & (VARYING_BIT_COL0 | VARYING_BIT_COL1 |
 343                                VARYING_BIT_BFC0 | VARYING_BIT_BFC1)) {
 344       /* _NEW_LIGHT | _NEW_BUFFERS */
 345       key->clamp_vertex_color = ctx->Light._ClampVertexColor;
 346    }
 347
 348    /* _NEW_POINT */
 349    if (brw->gen < 6 && ctx->Point.PointSprite) {
 350       for (i = 0; i < 8; i++) {
 351          if (ctx->Point.CoordReplace[i])
 352             key->point_coord_replace |= (1 << i);
 353       }
 354    }
 355
 356    /* _NEW_TEXTURE */
 357    brw_populate_sampler_prog_key_data(ctx, prog, brw->vs.base.sampler_count,
 358                                       &key->tex);
 359
 360    /* BRW_NEW_VS_ATTRIB_WORKAROUNDS */
 361    memcpy(key->gl_attrib_wa_flags, brw->vb.attrib_wa_flags,
 362           sizeof(brw->vb.attrib_wa_flags));
 363 }
 364
 365 void
 366 brw_upload_vs_prog(struct brw_context *brw)
 367 {
 368    struct gl_context *ctx = &brw->ctx;
 369    struct gl_shader_program **current = ctx->_Shader->CurrentProgram;
 370    struct brw_vs_prog_key key;
 371    /* BRW_NEW_VERTEX_PROGRAM */
 372    struct brw_vertex_program *vp =
 373       (struct brw_vertex_program *)brw->vertex_program;
 374
 375    if (!brw_vs_state_dirty(brw))
 376       return;
 377
 378    brw_vs_populate_key(brw, &key);
 379
 380    if (!brw_search_cache(&brw->cache, BRW_CACHE_VS_PROG,
 381                          &key, sizeof(key),
 382                          &brw->vs.base.prog_offset, &brw->vs.prog_data)) {
 383       bool success = brw_codegen_vs_prog(brw, current[MESA_SHADER_VERTEX],
 384                                          vp, &key);
 385       (void) success;
 386       assert(success);
 387    }
 388    brw->vs.base.prog_data = &brw->vs.prog_data->base.base;
 389
 390    if (brw->vs.prog_data->base.vue_map.slots_valid !=
 391        brw->vue_map_geom_out.slots_valid) {
 392       brw->vue_map_vs = brw->vs.prog_data->base.vue_map;
 393       brw->ctx.NewDriverState |= BRW_NEW_VUE_MAP_VS;
 394       if (brw->gen < 6) {
 395          /* No geometry shader support, so the VS VUE map is the VUE map for
 396           * the output of the "geometry" portion of the pipeline.
 397           */
 398          brw->vue_map_geom_out = brw->vue_map_vs;
 399          brw->ctx.NewDriverState |= BRW_NEW_VUE_MAP_GEOM_OUT;
 400       }
 401    }
 402 }
 403
 404 bool
 405 brw_vs_precompile(struct gl_context *ctx,
 406                   struct gl_shader_program *shader_prog,
 407                   struct gl_program *prog)
 408 {
 409    struct brw_context *brw = brw_context(ctx);
 410    struct brw_vs_prog_key key;
 411    uint32_t old_prog_offset = brw->vs.base.prog_offset;
 412    struct brw_vs_prog_data *old_prog_data = brw->vs.prog_data;
 413    bool success;
 414
 415    struct gl_vertex_program *vp = (struct gl_vertex_program *) prog;
 416    struct brw_vertex_program *bvp = brw_vertex_program(vp);
 417
 418    memset(&key, 0, sizeof(key));
 419
 420    brw_setup_tex_for_precompile(brw, &key.tex, prog);
 421    key.program_string_id = bvp->id;
 422    key.clamp_vertex_color =
 423       (prog->OutputsWritten & (VARYING_BIT_COL0 | VARYING_BIT_COL1 |
 424                                VARYING_BIT_BFC0 | VARYING_BIT_BFC1));
 425
 426    success = brw_codegen_vs_prog(brw, shader_prog, bvp, &key);
 427
 428    brw->vs.base.prog_offset = old_prog_offset;
 429    brw->vs.prog_data = old_prog_data;
 430
 431    return success;
 432 }