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 "main/context.h"
  35 #include "brw_context.h"
  36 #include "brw_vs.h"
  37 #include "brw_util.h"
  38 #include "brw_state.h"
  39 #include "program/prog_print.h"
  40 #include "program/prog_parameter.h"
  41 #include "brw_nir.h"
  42 #include "brw_program.h"
  43
  44 #include "util/ralloc.h"
  45
  46 bool
  47 brw_codegen_vs_prog(struct brw_context *brw,
  48                     struct gl_shader_program *prog,
  49                     struct brw_vertex_program *vp,
  50                     struct brw_vs_prog_key *key)
  51 {
  52    const struct brw_compiler *compiler = brw->intelScreen->compiler;
  53    GLuint program_size;
  54    const GLuint *program;
  55    struct brw_vs_prog_data prog_data;
  56    struct brw_stage_prog_data *stage_prog_data = &prog_data.base.base;
  57    void *mem_ctx;
  58    int i;
  59    struct brw_shader *vs = NULL;
  60    bool start_busy = false;
  61    double start_time = 0;
  62
  63    if (prog)
  64       vs = (struct brw_shader *) prog->_LinkedShaders[MESA_SHADER_VERTEX];
  65
  66    memset(&prog_data, 0, sizeof(prog_data));
  67
  68    /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
  69    if (!prog)
  70       stage_prog_data->use_alt_mode = true;
  71
  72    mem_ctx = ralloc_context(NULL);
  73
  74    brw_assign_common_binding_table_offsets(MESA_SHADER_VERTEX,
  75                                            brw->intelScreen->devinfo,
  76                                            prog, &vp->program.Base,
  77                                            &prog_data.base.base, 0);
  78
  79    /* Allocate the references to the uniforms that will end up in the
  80     * prog_data associated with the compiled program, and which will be freed
  81     * by the state cache.
  82     */
  83    int param_count = vp->program.Base.nir->num_uniforms;
  84    if (!compiler->scalar_stage[MESA_SHADER_VERTEX])
  85       param_count *= 4;
  86
  87    if (vs)
  88       prog_data.base.base.nr_image_params = vs->base.NumImages;
  89
  90    /* vec4_visitor::setup_uniform_clipplane_values() also uploads user clip
  91     * planes as uniforms.
  92     */
  93    param_count += key->nr_userclip_plane_consts * 4;
  94
  95    stage_prog_data->param =
  96       rzalloc_array(NULL, const gl_constant_value *, param_count);
  97    stage_prog_data->pull_param =
  98       rzalloc_array(NULL, const gl_constant_value *, param_count);
  99    stage_prog_data->image_param =
 100       rzalloc_array(NULL, struct brw_image_param,
 101                     stage_prog_data->nr_image_params);
 102    stage_prog_data->nr_params = param_count;
 103
 104    if (prog) {
 105       brw_nir_setup_glsl_uniforms(vp->program.Base.nir, prog, &vp->program.Base,
 106                                   &prog_data.base.base,
 107                                   compiler->scalar_stage[MESA_SHADER_VERTEX]);
 108    } else {
 109       brw_nir_setup_arb_uniforms(vp->program.Base.nir, &vp->program.Base,
 110                                  &prog_data.base.base);
 111    }
 112
 113    GLbitfield64 outputs_written = vp->program.Base.OutputsWritten;
 114    prog_data.inputs_read = vp->program.Base.InputsRead;
 115
 116    if (key->copy_edgeflag) {
 117       outputs_written |= BITFIELD64_BIT(VARYING_SLOT_EDGE);
 118       prog_data.inputs_read |= VERT_BIT_EDGEFLAG;
 119    }
 120
 121    if (brw->gen < 6) {
 122       /* Put dummy slots into the VUE for the SF to put the replaced
 123        * point sprite coords in.  We shouldn't need these dummy slots,
 124        * which take up precious URB space, but it would mean that the SF
 125        * doesn't get nice aligned pairs of input coords into output
 126        * coords, which would be a pain to handle.
 127        */
 128       for (i = 0; i < 8; i++) {
 129          if (key->point_coord_replace & (1 << i))
 130             outputs_written |= BITFIELD64_BIT(VARYING_SLOT_TEX0 + i);
 131       }
 132
 133       /* if back colors are written, allocate slots for front colors too */
 134       if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC0))
 135          outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL0);
 136       if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC1))
 137          outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL1);
 138    }
 139
 140    /* In order for legacy clipping to work, we need to populate the clip
 141     * distance varying slots whenever clipping is enabled, even if the vertex
 142     * shader doesn't write to gl_ClipDistance.
 143     */
 144    if (key->nr_userclip_plane_consts > 0) {
 145       outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0);
 146       outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1);
 147    }
 148
 149    brw_compute_vue_map(brw->intelScreen->devinfo,
 150                        &prog_data.base.vue_map, outputs_written,
 151                        prog ? prog->SeparateShader : false);
 152
 153    if (0) {
 154       _mesa_fprint_program_opt(stderr, &vp->program.Base, PROG_PRINT_DEBUG,
 155                                true);
 156    }
 157
 158    if (unlikely(brw->perf_debug)) {
 159       start_busy = (brw->batch.last_bo &&
 160                     drm_intel_bo_busy(brw->batch.last_bo));
 161       start_time = get_time();
 162    }
 163
 164    if (unlikely(INTEL_DEBUG & DEBUG_VS)) {
 165       brw_dump_ir("vertex", prog, vs ? &vs->base : NULL, &vp->program.Base);
 166
 167       fprintf(stderr, "VS Output ");
 168       brw_print_vue_map(stderr, &prog_data.base.vue_map);
 169    }
 170
 171    int st_index = -1;
 172    if (INTEL_DEBUG & DEBUG_SHADER_TIME)
 173       st_index = brw_get_shader_time_index(brw, prog, &vp->program.Base, ST_VS);
 174
 175    /* Emit GEN4 code.
 176     */
 177    char *error_str;
 178    program = brw_compile_vs(compiler, brw, mem_ctx, key,
 179                             &prog_data, vp->program.Base.nir,
 180                             brw_select_clip_planes(&brw->ctx),
 181                             !_mesa_is_gles3(&brw->ctx),
 182                             st_index, &program_size, &error_str);
 183    if (program == NULL) {
 184       if (prog) {
 185          prog->LinkStatus = false;
 186          ralloc_strcat(&prog->InfoLog, error_str);
 187       }
 188
 189       _mesa_problem(NULL, "Failed to compile vertex shader: %s\n", error_str);
 190
 191       ralloc_free(mem_ctx);
 192       return false;
 193    }
 194
 195    if (unlikely(brw->perf_debug) && vs) {
 196       if (vs->compiled_once) {
 197          brw_vs_debug_recompile(brw, prog, key);
 198       }
 199       if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
 200          perf_debug("VS compile took %.03f ms and stalled the GPU\n",
 201                     (get_time() - start_time) * 1000);
 202       }
 203       vs->compiled_once = true;
 204    }
 205
 206    /* Scratch space is used for register spilling */
 207    if (prog_data.base.base.total_scratch) {
 208       brw_get_scratch_bo(brw, &brw->vs.base.scratch_bo,
 209                          prog_data.base.base.total_scratch *
 210                          brw->max_vs_threads);
 211    }
 212
 213    brw_upload_cache(&brw->cache, BRW_CACHE_VS_PROG,
 214                     key, sizeof(struct brw_vs_prog_key),
 215                     program, program_size,
 216                     &prog_data, sizeof(prog_data),
 217                     &brw->vs.base.prog_offset, &brw->vs.prog_data);
 218    ralloc_free(mem_ctx);
 219
 220    return true;
 221 }
 222
 223 void
 224 brw_vs_debug_recompile(struct brw_context *brw,
 225                        struct gl_shader_program *prog,
 226                        const struct brw_vs_prog_key *key)
 227 {
 228    struct brw_cache_item *c = NULL;
 229    const struct brw_vs_prog_key *old_key = NULL;
 230    bool found = false;
 231
 232    perf_debug("Recompiling vertex shader for program %d\n", prog->Name);
 233
 234    for (unsigned int i = 0; i < brw->cache.size; i++) {
 235       for (c = brw->cache.items[i]; c; c = c->next) {
 236          if (c->cache_id == BRW_CACHE_VS_PROG) {
 237             old_key = c->key;
 238
 239             if (old_key->program_string_id == key->program_string_id)
 240                break;
 241          }
 242       }
 243       if (c)
 244          break;
 245    }
 246
 247    if (!c) {
 248       perf_debug("  Didn't find previous compile in the shader cache for "
 249                  "debug\n");
 250       return;
 251    }
 252
 253    for (unsigned int i = 0; i < VERT_ATTRIB_MAX; i++) {
 254       found |= key_debug(brw, "Vertex attrib w/a flags",
 255                          old_key->gl_attrib_wa_flags[i],
 256                          key->gl_attrib_wa_flags[i]);
 257    }
 258
 259    found |= key_debug(brw, "legacy user clipping",
 260                       old_key->nr_userclip_plane_consts,
 261                       key->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->tex, &key->tex);
 271
 272    if (!found) {
 273       perf_debug("  Something else\n");
 274    }
 275 }
 276
 277 static bool
 278 brw_vs_state_dirty(struct brw_context *brw)
 279 {
 280    return brw_state_dirty(brw,
 281                           _NEW_BUFFERS |
 282                           _NEW_LIGHT |
 283                           _NEW_POINT |
 284                           _NEW_POLYGON |
 285                           _NEW_TEXTURE |
 286                           _NEW_TRANSFORM,
 287                           BRW_NEW_VERTEX_PROGRAM |
 288                           BRW_NEW_VS_ATTRIB_WORKAROUNDS);
 289 }
 290
 291 static void
 292 brw_vs_populate_key(struct brw_context *brw,
 293                     struct brw_vs_prog_key *key)
 294 {
 295    struct gl_context *ctx = &brw->ctx;
 296    /* BRW_NEW_VERTEX_PROGRAM */
 297    struct brw_vertex_program *vp =
 298       (struct brw_vertex_program *)brw->vertex_program;
 299    struct gl_program *prog = (struct gl_program *) brw->vertex_program;
 300    int i;
 301
 302    memset(key, 0, sizeof(*key));
 303
 304    /* Just upload the program verbatim for now.  Always send it all
 305     * the inputs it asks for, whether they are varying or not.
 306     */
 307    key->program_string_id = vp->id;
 308
 309    if (ctx->Transform.ClipPlanesEnabled != 0 &&
 310        (ctx->API == API_OPENGL_COMPAT ||
 311         ctx->API == API_OPENGLES) &&
 312        vp->program.Base.ClipDistanceArraySize == 0) {
 313       key->nr_userclip_plane_consts =
 314          _mesa_logbase2(ctx->Transform.ClipPlanesEnabled) + 1;
 315    }
 316
 317    /* _NEW_POLYGON */
 318    if (brw->gen < 6) {
 319       key->copy_edgeflag = (ctx->Polygon.FrontMode != GL_FILL ||
 320                             ctx->Polygon.BackMode != GL_FILL);
 321    }
 322
 323    if (prog->OutputsWritten & (VARYING_BIT_COL0 | VARYING_BIT_COL1 |
 324                                VARYING_BIT_BFC0 | VARYING_BIT_BFC1)) {
 325       /* _NEW_LIGHT | _NEW_BUFFERS */
 326       key->clamp_vertex_color = ctx->Light._ClampVertexColor;
 327    }
 328
 329    /* _NEW_POINT */
 330    if (brw->gen < 6 && ctx->Point.PointSprite) {
 331       for (i = 0; i < 8; i++) {
 332          if (ctx->Point.CoordReplace[i])
 333             key->point_coord_replace |= (1 << i);
 334       }
 335    }
 336
 337    /* _NEW_TEXTURE */
 338    brw_populate_sampler_prog_key_data(ctx, prog, brw->vs.base.sampler_count,
 339                                       &key->tex);
 340
 341    /* BRW_NEW_VS_ATTRIB_WORKAROUNDS */
 342    memcpy(key->gl_attrib_wa_flags, brw->vb.attrib_wa_flags,
 343           sizeof(brw->vb.attrib_wa_flags));
 344 }
 345
 346 void
 347 brw_upload_vs_prog(struct brw_context *brw)
 348 {
 349    struct gl_context *ctx = &brw->ctx;
 350    struct gl_shader_program **current = ctx->_Shader->CurrentProgram;
 351    struct brw_vs_prog_key key;
 352    /* BRW_NEW_VERTEX_PROGRAM */
 353    struct brw_vertex_program *vp =
 354       (struct brw_vertex_program *)brw->vertex_program;
 355
 356    if (!brw_vs_state_dirty(brw))
 357       return;
 358
 359    brw_vs_populate_key(brw, &key);
 360
 361    if (!brw_search_cache(&brw->cache, BRW_CACHE_VS_PROG,
 362                          &key, sizeof(key),
 363                          &brw->vs.base.prog_offset, &brw->vs.prog_data)) {
 364       bool success = brw_codegen_vs_prog(brw, current[MESA_SHADER_VERTEX],
 365                                          vp, &key);
 366       (void) success;
 367       assert(success);
 368    }
 369    brw->vs.base.prog_data = &brw->vs.prog_data->base.base;
 370 }
 371
 372 bool
 373 brw_vs_precompile(struct gl_context *ctx,
 374                   struct gl_shader_program *shader_prog,
 375                   struct gl_program *prog)
 376 {
 377    struct brw_context *brw = brw_context(ctx);
 378    struct brw_vs_prog_key key;
 379    uint32_t old_prog_offset = brw->vs.base.prog_offset;
 380    struct brw_vs_prog_data *old_prog_data = brw->vs.prog_data;
 381    bool success;
 382
 383    struct gl_vertex_program *vp = (struct gl_vertex_program *) prog;
 384    struct brw_vertex_program *bvp = brw_vertex_program(vp);
 385
 386    memset(&key, 0, sizeof(key));
 387
 388    brw_setup_tex_for_precompile(brw, &key.tex, prog);
 389    key.program_string_id = bvp->id;
 390    key.clamp_vertex_color =
 391       (prog->OutputsWritten & (VARYING_BIT_COL0 | VARYING_BIT_COL1 |
 392                                VARYING_BIT_BFC0 | VARYING_BIT_BFC1));
 393
 394    success = brw_codegen_vs_prog(brw, shader_prog, bvp, &key);
 395
 396    brw->vs.base.prog_offset = old_prog_offset;
 397    brw->vs.prog_data = old_prog_data;
 398
 399    return success;
 400 }