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