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 #include "brw_nir.h"
  41
  42 #include "util/ralloc.h"
  43
  44 /**
  45  * Decide which set of clip planes should be used when clipping via
  46  * gl_Position or gl_ClipVertex.
  47  */
  48 gl_clip_plane *brw_select_clip_planes(struct gl_context *ctx)
  49 {
  50    if (ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX]) {
  51       /* There is currently a GLSL vertex shader, so clip according to GLSL
  52        * rules, which means compare gl_ClipVertex (or gl_Position, if
  53        * gl_ClipVertex wasn't assigned) against the eye-coordinate clip planes
  54        * that were stored in EyeUserPlane at the time the clip planes were
  55        * specified.
  56        */
  57       return ctx->Transform.EyeUserPlane;
  58    } else {
  59       /* Either we are using fixed function or an ARB vertex program.  In
  60        * either case the clip planes are going to be compared against
  61        * gl_Position (which is in clip coordinates) so we have to clip using
  62        * _ClipUserPlane, which was transformed into clip coordinates by Mesa
  63        * core.
  64        */
  65       return ctx->Transform._ClipUserPlane;
  66    }
  67 }
  68
  69 bool
  70 brw_codegen_vs_prog(struct brw_context *brw,
  71                     struct gl_shader_program *prog,
  72                     struct brw_vertex_program *vp,
  73                     struct brw_vs_prog_key *key)
  74 {
  75    GLuint program_size;
  76    const GLuint *program;
  77    struct brw_vs_prog_data prog_data;
  78    struct brw_stage_prog_data *stage_prog_data = &prog_data.base.base;
  79    void *mem_ctx;
  80    int i;
  81    struct brw_shader *vs = NULL;
  82    bool start_busy = false;
  83    double start_time = 0;
  84
  85    if (!vp->program.Base.nir) {
  86       /* Normally we generate NIR in LinkShader() or
  87        * ProgramStringNotify(), but Mesa's fixed-function vertex program
  88        * handling doesn't notify the driver at all.  Just do it here, at
  89        * the last minute, even though it's lame.
  90        */
  91       assert(vp->program.Base.Id == 0 && prog == NULL);
  92       vp->program.Base.nir =
  93          brw_create_nir(brw, NULL, &vp->program.Base, MESA_SHADER_VERTEX,
  94                         brw->intelScreen->compiler->scalar_vs);
  95    }
  96
  97    if (prog)
  98       vs = (struct brw_shader *) prog->_LinkedShaders[MESA_SHADER_VERTEX];
  99
 100    memset(&prog_data, 0, sizeof(prog_data));
 101
 102    /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
 103    if (!prog)
 104       stage_prog_data->use_alt_mode = true;
 105
 106    mem_ctx = ralloc_context(NULL);
 107
 108    brw_assign_common_binding_table_offsets(MESA_SHADER_VERTEX,
 109                                            brw->intelScreen->devinfo,
 110                                            prog, &vp->program.Base,
 111                                            &prog_data.base.base, 0);
 112
 113    /* Allocate the references to the uniforms that will end up in the
 114     * prog_data associated with the compiled program, and which will be freed
 115     * by the state cache.
 116     */
 117    int param_count = vp->program.Base.nir->num_uniforms;
 118    if (!brw->intelScreen->compiler->scalar_vs)
 119       param_count *= 4;
 120
 121    if (vs)
 122       prog_data.base.base.nr_image_params = vs->base.NumImages;
 123
 124    /* vec4_visitor::setup_uniform_clipplane_values() also uploads user clip
 125     * planes as uniforms.
 126     */
 127    param_count += key->nr_userclip_plane_consts * 4;
 128
 129    stage_prog_data->param =
 130       rzalloc_array(NULL, const gl_constant_value *, param_count);
 131    stage_prog_data->pull_param =
 132       rzalloc_array(NULL, const gl_constant_value *, param_count);
 133    stage_prog_data->image_param =
 134       rzalloc_array(NULL, struct brw_image_param,
 135                     stage_prog_data->nr_image_params);
 136    stage_prog_data->nr_params = param_count;
 137
 138    if (prog) {
 139       brw_nir_setup_glsl_uniforms(vp->program.Base.nir, prog, &vp->program.Base,
 140                                   &prog_data.base.base,
 141                                   brw->intelScreen->compiler->scalar_vs);
 142    } else {
 143       brw_nir_setup_arb_uniforms(vp->program.Base.nir, &vp->program.Base,
 144                                  &prog_data.base.base);
 145    }
 146
 147    GLbitfield64 outputs_written = vp->program.Base.OutputsWritten;
 148    prog_data.inputs_read = vp->program.Base.InputsRead;
 149
 150    if (key->copy_edgeflag) {
 151       outputs_written |= BITFIELD64_BIT(VARYING_SLOT_EDGE);
 152       prog_data.inputs_read |= VERT_BIT_EDGEFLAG;
 153    }
 154
 155    if (brw->gen < 6) {
 156       /* Put dummy slots into the VUE for the SF to put the replaced
 157        * point sprite coords in.  We shouldn't need these dummy slots,
 158        * which take up precious URB space, but it would mean that the SF
 159        * doesn't get nice aligned pairs of input coords into output
 160        * coords, which would be a pain to handle.
 161        */
 162       for (i = 0; i < 8; i++) {
 163          if (key->point_coord_replace & (1 << i))
 164             outputs_written |= BITFIELD64_BIT(VARYING_SLOT_TEX0 + i);
 165       }
 166
 167       /* if back colors are written, allocate slots for front colors too */
 168       if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC0))
 169          outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL0);
 170       if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC1))
 171          outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL1);
 172    }
 173
 174    /* In order for legacy clipping to work, we need to populate the clip
 175     * distance varying slots whenever clipping is enabled, even if the vertex
 176     * shader doesn't write to gl_ClipDistance.
 177     */
 178    if (key->nr_userclip_plane_consts > 0) {
 179       outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0);
 180       outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1);
 181    }
 182
 183    brw_compute_vue_map(brw->intelScreen->devinfo,
 184                        &prog_data.base.vue_map, outputs_written,
 185                        prog ? prog->SeparateShader : false);
 186
 187    if (0) {
 188       _mesa_fprint_program_opt(stderr, &vp->program.Base, PROG_PRINT_DEBUG,
 189                                true);
 190    }
 191
 192    if (unlikely(brw->perf_debug)) {
 193       start_busy = (brw->batch.last_bo &&
 194                     drm_intel_bo_busy(brw->batch.last_bo));
 195       start_time = get_time();
 196    }
 197
 198    if (unlikely(INTEL_DEBUG & DEBUG_VS))
 199       brw_dump_ir("vertex", prog, &vs->base, &vp->program.Base);
 200
 201    /* Emit GEN4 code.
 202     */
 203    program = brw_vs_emit(brw, mem_ctx, key, &prog_data,
 204                          &vp->program, prog, &program_size);
 205    if (program == NULL) {
 206       ralloc_free(mem_ctx);
 207       return false;
 208    }
 209
 210    if (unlikely(brw->perf_debug) && vs) {
 211       if (vs->compiled_once) {
 212          brw_vs_debug_recompile(brw, prog, key);
 213       }
 214       if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
 215          perf_debug("VS compile took %.03f ms and stalled the GPU\n",
 216                     (get_time() - start_time) * 1000);
 217       }
 218       vs->compiled_once = true;
 219    }
 220
 221    /* Scratch space is used for register spilling */
 222    if (prog_data.base.base.total_scratch) {
 223       brw_get_scratch_bo(brw, &brw->vs.base.scratch_bo,
 224                          prog_data.base.base.total_scratch *
 225                          brw->max_vs_threads);
 226    }
 227
 228    brw_upload_cache(&brw->cache, BRW_CACHE_VS_PROG,
 229                     key, sizeof(struct brw_vs_prog_key),
 230                     program, program_size,
 231                     &prog_data, sizeof(prog_data),
 232                     &brw->vs.base.prog_offset, &brw->vs.prog_data);
 233    ralloc_free(mem_ctx);
 234
 235    return true;
 236 }
 237
 238 static bool
 239 key_debug(struct brw_context *brw, const char *name, int a, int b)
 240 {
 241    if (a != b) {
 242       perf_debug("  %s %d->%d\n", name, a, b);
 243       return true;
 244    }
 245    return false;
 246 }
 247
 248 void
 249 brw_vs_debug_recompile(struct brw_context *brw,
 250                        struct gl_shader_program *prog,
 251                        const struct brw_vs_prog_key *key)
 252 {
 253    struct brw_cache_item *c = NULL;
 254    const struct brw_vs_prog_key *old_key = NULL;
 255    bool found = false;
 256
 257    perf_debug("Recompiling vertex shader for program %d\n", prog->Name);
 258
 259    for (unsigned int i = 0; i < brw->cache.size; i++) {
 260       for (c = brw->cache.items[i]; c; c = c->next) {
 261          if (c->cache_id == BRW_CACHE_VS_PROG) {
 262             old_key = c->key;
 263
 264             if (old_key->program_string_id == key->program_string_id)
 265                break;
 266          }
 267       }
 268       if (c)
 269          break;
 270    }
 271
 272    if (!c) {
 273       perf_debug("  Didn't find previous compile in the shader cache for "
 274                  "debug\n");
 275       return;
 276    }
 277
 278    for (unsigned int i = 0; i < VERT_ATTRIB_MAX; i++) {
 279       found |= key_debug(brw, "Vertex attrib w/a flags",
 280                          old_key->gl_attrib_wa_flags[i],
 281                          key->gl_attrib_wa_flags[i]);
 282    }
 283
 284    found |= key_debug(brw, "legacy user clipping",
 285                       old_key->nr_userclip_plane_consts,
 286                       key->nr_userclip_plane_consts);
 287
 288    found |= key_debug(brw, "copy edgeflag",
 289                       old_key->copy_edgeflag, key->copy_edgeflag);
 290    found |= key_debug(brw, "PointCoord replace",
 291                       old_key->point_coord_replace, key->point_coord_replace);
 292    found |= key_debug(brw, "vertex color clamping",
 293                       old_key->clamp_vertex_color, key->clamp_vertex_color);
 294
 295    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
 296
 297    if (!found) {
 298       perf_debug("  Something else\n");
 299    }
 300 }
 301
 302 static bool
 303 brw_vs_state_dirty(struct brw_context *brw)
 304 {
 305    return brw_state_dirty(brw,
 306                           _NEW_BUFFERS |
 307                           _NEW_LIGHT |
 308                           _NEW_POINT |
 309                           _NEW_POLYGON |
 310                           _NEW_TEXTURE |
 311                           _NEW_TRANSFORM,
 312                           BRW_NEW_VERTEX_PROGRAM |
 313                           BRW_NEW_VS_ATTRIB_WORKAROUNDS);
 314 }
 315
 316 static void
 317 brw_vs_populate_key(struct brw_context *brw,
 318                     struct brw_vs_prog_key *key)
 319 {
 320    struct gl_context *ctx = &brw->ctx;
 321    /* BRW_NEW_VERTEX_PROGRAM */
 322    struct brw_vertex_program *vp =
 323       (struct brw_vertex_program *)brw->vertex_program;
 324    struct gl_program *prog = (struct gl_program *) brw->vertex_program;
 325    int i;
 326
 327    memset(key, 0, sizeof(*key));
 328
 329    /* Just upload the program verbatim for now.  Always send it all
 330     * the inputs it asks for, whether they are varying or not.
 331     */
 332    key->program_string_id = vp->id;
 333
 334    if (ctx->Transform.ClipPlanesEnabled != 0 &&
 335        ctx->API == API_OPENGL_COMPAT &&
 336        !vp->program.Base.UsesClipDistanceOut) {
 337       key->nr_userclip_plane_consts =
 338          _mesa_logbase2(ctx->Transform.ClipPlanesEnabled) + 1;
 339    }
 340
 341    /* _NEW_POLYGON */
 342    if (brw->gen < 6) {
 343       key->copy_edgeflag = (ctx->Polygon.FrontMode != GL_FILL ||
 344                             ctx->Polygon.BackMode != GL_FILL);
 345    }
 346
 347    if (prog->OutputsWritten & (VARYING_BIT_COL0 | VARYING_BIT_COL1 |
 348                                VARYING_BIT_BFC0 | VARYING_BIT_BFC1)) {
 349       /* _NEW_LIGHT | _NEW_BUFFERS */
 350       key->clamp_vertex_color = ctx->Light._ClampVertexColor;
 351    }
 352
 353    /* _NEW_POINT */
 354    if (brw->gen < 6 && ctx->Point.PointSprite) {
 355       for (i = 0; i < 8; i++) {
 356          if (ctx->Point.CoordReplace[i])
 357             key->point_coord_replace |= (1 << i);
 358       }
 359    }
 360
 361    /* _NEW_TEXTURE */
 362    brw_populate_sampler_prog_key_data(ctx, prog, brw->vs.base.sampler_count,
 363                                       &key->tex);
 364
 365    /* BRW_NEW_VS_ATTRIB_WORKAROUNDS */
 366    memcpy(key->gl_attrib_wa_flags, brw->vb.attrib_wa_flags,
 367           sizeof(brw->vb.attrib_wa_flags));
 368 }
 369
 370 void
 371 brw_upload_vs_prog(struct brw_context *brw)
 372 {
 373    struct gl_context *ctx = &brw->ctx;
 374    struct gl_shader_program **current = ctx->_Shader->CurrentProgram;
 375    struct brw_vs_prog_key key;
 376    /* BRW_NEW_VERTEX_PROGRAM */
 377    struct brw_vertex_program *vp =
 378       (struct brw_vertex_program *)brw->vertex_program;
 379
 380    if (!brw_vs_state_dirty(brw))
 381       return;
 382
 383    brw_vs_populate_key(brw, &key);
 384
 385    if (!brw_search_cache(&brw->cache, BRW_CACHE_VS_PROG,
 386                          &key, sizeof(key),
 387                          &brw->vs.base.prog_offset, &brw->vs.prog_data)) {
 388       bool success = brw_codegen_vs_prog(brw, current[MESA_SHADER_VERTEX],
 389                                          vp, &key);
 390       (void) success;
 391       assert(success);
 392    }
 393    brw->vs.base.prog_data = &brw->vs.prog_data->base.base;
 394 }
 395
 396 bool
 397 brw_vs_precompile(struct gl_context *ctx,
 398                   struct gl_shader_program *shader_prog,
 399                   struct gl_program *prog)
 400 {
 401    struct brw_context *brw = brw_context(ctx);
 402    struct brw_vs_prog_key key;
 403    uint32_t old_prog_offset = brw->vs.base.prog_offset;
 404    struct brw_vs_prog_data *old_prog_data = brw->vs.prog_data;
 405    bool success;
 406
 407    struct gl_vertex_program *vp = (struct gl_vertex_program *) prog;
 408    struct brw_vertex_program *bvp = brw_vertex_program(vp);
 409
 410    memset(&key, 0, sizeof(key));
 411
 412    brw_setup_tex_for_precompile(brw, &key.tex, prog);
 413    key.program_string_id = bvp->id;
 414    key.clamp_vertex_color =
 415       (prog->OutputsWritten & (VARYING_BIT_COL0 | VARYING_BIT_COL1 |
 416                                VARYING_BIT_BFC0 | VARYING_BIT_BFC1));
 417
 418    success = brw_codegen_vs_prog(brw, shader_prog, bvp, &key);
 419
 420    brw->vs.base.prog_offset = old_prog_offset;
 421    brw->vs.prog_data = old_prog_data;
 422
 423    return success;
 424 }