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 / 4;
  84
  85    if (vs)
  86       prog_data.base.base.nr_image_params = vs->base.NumImages;
  87
  88    /* vec4_visitor::setup_uniform_clipplane_values() also uploads user clip
  89     * planes as uniforms.
  90     */
  91    param_count += key->nr_userclip_plane_consts * 4;
  92
  93    stage_prog_data->param =
  94       rzalloc_array(NULL, const gl_constant_value *, param_count);
  95    stage_prog_data->pull_param =
  96       rzalloc_array(NULL, const gl_constant_value *, param_count);
  97    stage_prog_data->image_param =
  98       rzalloc_array(NULL, struct brw_image_param,
  99                     stage_prog_data->nr_image_params);
 100    stage_prog_data->nr_params = param_count;
 101
 102    if (prog) {
 103       brw_nir_setup_glsl_uniforms(vp->program.Base.nir, prog, &vp->program.Base,
 104                                   &prog_data.base.base,
 105                                   compiler->scalar_stage[MESA_SHADER_VERTEX]);
 106    } else {
 107       brw_nir_setup_arb_uniforms(vp->program.Base.nir, &vp->program.Base,
 108                                  &prog_data.base.base);
 109    }
 110
 111    GLbitfield64 outputs_written = vp->program.Base.OutputsWritten;
 112    prog_data.inputs_read = vp->program.Base.InputsRead;
 113
 114    if (key->copy_edgeflag) {
 115       outputs_written |= BITFIELD64_BIT(VARYING_SLOT_EDGE);
 116       prog_data.inputs_read |= VERT_BIT_EDGEFLAG;
 117    }
 118
 119    prog_data.base.cull_distance_mask =
 120       ((1 << vp->program.Base.CullDistanceArraySize) - 1) <<
 121       vp->program.Base.ClipDistanceArraySize;
 122
 123    if (brw->gen < 6) {
 124       /* Put dummy slots into the VUE for the SF to put the replaced
 125        * point sprite coords in.  We shouldn't need these dummy slots,
 126        * which take up precious URB space, but it would mean that the SF
 127        * doesn't get nice aligned pairs of input coords into output
 128        * coords, which would be a pain to handle.
 129        */
 130       for (i = 0; i < 8; i++) {
 131          if (key->point_coord_replace & (1 << i))
 132             outputs_written |= BITFIELD64_BIT(VARYING_SLOT_TEX0 + i);
 133       }
 134
 135       /* if back colors are written, allocate slots for front colors too */
 136       if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC0))
 137          outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL0);
 138       if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC1))
 139          outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL1);
 140    }
 141
 142    /* In order for legacy clipping to work, we need to populate the clip
 143     * distance varying slots whenever clipping is enabled, even if the vertex
 144     * shader doesn't write to gl_ClipDistance.
 145     */
 146    if (key->nr_userclip_plane_consts > 0) {
 147       outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0);
 148       outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1);
 149    }
 150
 151    brw_compute_vue_map(brw->intelScreen->devinfo,
 152                        &prog_data.base.vue_map, outputs_written,
 153                        prog ? prog->SeparateShader ||
 154                               prog->_LinkedShaders[MESA_SHADER_TESS_EVAL]
 155                             : false);
 156
 157    if (0) {
 158       _mesa_fprint_program_opt(stderr, &vp->program.Base, PROG_PRINT_DEBUG,
 159                                true);
 160    }
 161
 162    if (unlikely(brw->perf_debug)) {
 163       start_busy = (brw->batch.last_bo &&
 164                     drm_intel_bo_busy(brw->batch.last_bo));
 165       start_time = get_time();
 166    }
 167
 168    if (unlikely(INTEL_DEBUG & DEBUG_VS)) {
 169       brw_dump_ir("vertex", prog, vs ? &vs->base : NULL, &vp->program.Base);
 170
 171       fprintf(stderr, "VS Output ");
 172       brw_print_vue_map(stderr, &prog_data.base.vue_map);
 173    }
 174
 175    int st_index = -1;
 176    if (INTEL_DEBUG & DEBUG_SHADER_TIME)
 177       st_index = brw_get_shader_time_index(brw, prog, &vp->program.Base, ST_VS);
 178
 179    /* Emit GEN4 code.
 180     */
 181    char *error_str;
 182    program = brw_compile_vs(compiler, brw, mem_ctx, key,
 183                             &prog_data, vp->program.Base.nir,
 184                             brw_select_clip_planes(&brw->ctx),
 185                             !_mesa_is_gles3(&brw->ctx),
 186                             st_index, &program_size, &error_str);
 187    if (program == NULL) {
 188       if (prog) {
 189          prog->LinkStatus = false;
 190          ralloc_strcat(&prog->InfoLog, error_str);
 191       }
 192
 193       _mesa_problem(NULL, "Failed to compile vertex shader: %s\n", error_str);
 194
 195       ralloc_free(mem_ctx);
 196       return false;
 197    }
 198
 199    if (unlikely(brw->perf_debug) && vs) {
 200       if (vs->compiled_once) {
 201          brw_vs_debug_recompile(brw, prog, key);
 202       }
 203       if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
 204          perf_debug("VS compile took %.03f ms and stalled the GPU\n",
 205                     (get_time() - start_time) * 1000);
 206       }
 207       vs->compiled_once = true;
 208    }
 209
 210    /* Scratch space is used for register spilling */
 211    brw_alloc_stage_scratch(brw, &brw->vs.base,
 212                            prog_data.base.base.total_scratch,
 213                            brw->max_vs_threads);
 214
 215    brw_upload_cache(&brw->cache, BRW_CACHE_VS_PROG,
 216                     key, sizeof(struct brw_vs_prog_key),
 217                     program, program_size,
 218                     &prog_data, sizeof(prog_data),
 219                     &brw->vs.base.prog_offset, &brw->vs.prog_data);
 220    ralloc_free(mem_ctx);
 221
 222    return true;
 223 }
 224
 225 void
 226 brw_vs_debug_recompile(struct brw_context *brw,
 227                        struct gl_shader_program *prog,
 228                        const struct brw_vs_prog_key *key)
 229 {
 230    struct brw_cache_item *c = NULL;
 231    const struct brw_vs_prog_key *old_key = NULL;
 232    bool found = false;
 233
 234    perf_debug("Recompiling vertex shader for program %d\n", prog->Name);
 235
 236    for (unsigned int i = 0; i < brw->cache.size; i++) {
 237       for (c = brw->cache.items[i]; c; c = c->next) {
 238          if (c->cache_id == BRW_CACHE_VS_PROG) {
 239             old_key = c->key;
 240
 241             if (old_key->program_string_id == key->program_string_id)
 242                break;
 243          }
 244       }
 245       if (c)
 246          break;
 247    }
 248
 249    if (!c) {
 250       perf_debug("  Didn't find previous compile in the shader cache for "
 251                  "debug\n");
 252       return;
 253    }
 254
 255    for (unsigned int i = 0; i < VERT_ATTRIB_MAX; i++) {
 256       found |= key_debug(brw, "Vertex attrib w/a flags",
 257                          old_key->gl_attrib_wa_flags[i],
 258                          key->gl_attrib_wa_flags[i]);
 259    }
 260
 261    found |= key_debug(brw, "legacy user clipping",
 262                       old_key->nr_userclip_plane_consts,
 263                       key->nr_userclip_plane_consts);
 264
 265    found |= key_debug(brw, "copy edgeflag",
 266                       old_key->copy_edgeflag, key->copy_edgeflag);
 267    found |= key_debug(brw, "PointCoord replace",
 268                       old_key->point_coord_replace, key->point_coord_replace);
 269    found |= key_debug(brw, "vertex color clamping",
 270                       old_key->clamp_vertex_color, key->clamp_vertex_color);
 271
 272    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
 273
 274    if (!found) {
 275       perf_debug("  Something else\n");
 276    }
 277 }
 278
 279 static bool
 280 brw_vs_state_dirty(const struct brw_context *brw)
 281 {
 282    return brw_state_dirty(brw,
 283                           _NEW_BUFFERS |
 284                           _NEW_LIGHT |
 285                           _NEW_POINT |
 286                           _NEW_POLYGON |
 287                           _NEW_TEXTURE |
 288                           _NEW_TRANSFORM,
 289                           BRW_NEW_VERTEX_PROGRAM |
 290                           BRW_NEW_VS_ATTRIB_WORKAROUNDS);
 291 }
 292
 293 static void
 294 brw_vs_populate_key(struct brw_context *brw,
 295                     struct brw_vs_prog_key *key)
 296 {
 297    struct gl_context *ctx = &brw->ctx;
 298    /* BRW_NEW_VERTEX_PROGRAM */
 299    struct brw_vertex_program *vp =
 300       (struct brw_vertex_program *)brw->vertex_program;
 301    struct gl_program *prog = (struct gl_program *) brw->vertex_program;
 302
 303    memset(key, 0, sizeof(*key));
 304
 305    /* Just upload the program verbatim for now.  Always send it all
 306     * the inputs it asks for, whether they are varying or not.
 307     */
 308    key->program_string_id = vp->id;
 309
 310    if (ctx->Transform.ClipPlanesEnabled != 0 &&
 311        (ctx->API == API_OPENGL_COMPAT ||
 312         ctx->API == API_OPENGLES) &&
 313        vp->program.Base.ClipDistanceArraySize == 0) {
 314       key->nr_userclip_plane_consts =
 315          _mesa_logbase2(ctx->Transform.ClipPlanesEnabled) + 1;
 316    }
 317
 318    /* _NEW_POLYGON */
 319    if (brw->gen < 6) {
 320       key->copy_edgeflag = (ctx->Polygon.FrontMode != GL_FILL ||
 321                             ctx->Polygon.BackMode != GL_FILL);
 322    }
 323
 324    if (prog->OutputsWritten & (VARYING_BIT_COL0 | VARYING_BIT_COL1 |
 325                                VARYING_BIT_BFC0 | VARYING_BIT_BFC1)) {
 326       /* _NEW_LIGHT | _NEW_BUFFERS */
 327       key->clamp_vertex_color = ctx->Light._ClampVertexColor;
 328    }
 329
 330    /* _NEW_POINT */
 331    if (brw->gen < 6 && ctx->Point.PointSprite) {
 332       key->point_coord_replace = ctx->Point.CoordReplaceBits & 0xff;
 333    }
 334
 335    /* _NEW_TEXTURE */
 336    brw_populate_sampler_prog_key_data(ctx, prog, brw->vs.base.sampler_count,
 337                                       &key->tex);
 338
 339    /* BRW_NEW_VS_ATTRIB_WORKAROUNDS */
 340    memcpy(key->gl_attrib_wa_flags, brw->vb.attrib_wa_flags,
 341           sizeof(brw->vb.attrib_wa_flags));
 342 }
 343
 344 void
 345 brw_upload_vs_prog(struct brw_context *brw)
 346 {
 347    struct gl_context *ctx = &brw->ctx;
 348    struct gl_shader_program **current = ctx->_Shader->CurrentProgram;
 349    struct brw_vs_prog_key key;
 350    /* BRW_NEW_VERTEX_PROGRAM */
 351    struct brw_vertex_program *vp =
 352       (struct brw_vertex_program *)brw->vertex_program;
 353
 354    if (!brw_vs_state_dirty(brw))
 355       return;
 356
 357    brw_vs_populate_key(brw, &key);
 358
 359    if (!brw_search_cache(&brw->cache, BRW_CACHE_VS_PROG,
 360                          &key, sizeof(key),
 361                          &brw->vs.base.prog_offset, &brw->vs.prog_data)) {
 362       bool success = brw_codegen_vs_prog(brw, current[MESA_SHADER_VERTEX],
 363                                          vp, &key);
 364       (void) success;
 365       assert(success);
 366    }
 367    brw->vs.base.prog_data = &brw->vs.prog_data->base.base;
 368 }
 369
 370 bool
 371 brw_vs_precompile(struct gl_context *ctx,
 372                   struct gl_shader_program *shader_prog,
 373                   struct gl_program *prog)
 374 {
 375    struct brw_context *brw = brw_context(ctx);
 376    struct brw_vs_prog_key key;
 377    uint32_t old_prog_offset = brw->vs.base.prog_offset;
 378    struct brw_vs_prog_data *old_prog_data = brw->vs.prog_data;
 379    bool success;
 380
 381    struct gl_vertex_program *vp = (struct gl_vertex_program *) prog;
 382    struct brw_vertex_program *bvp = brw_vertex_program(vp);
 383
 384    memset(&key, 0, sizeof(key));
 385
 386    brw_setup_tex_for_precompile(brw, &key.tex, prog);
 387    key.program_string_id = bvp->id;
 388    key.clamp_vertex_color =
 389       (prog->OutputsWritten & (VARYING_BIT_COL0 | VARYING_BIT_COL1 |
 390                                VARYING_BIT_BFC0 | VARYING_BIT_BFC1));
 391
 392    success = brw_codegen_vs_prog(brw, shader_prog, bvp, &key);
 393
 394    brw->vs.base.prog_offset = old_prog_offset;
 395    brw->vs.prog_data = old_prog_data;
 396
 397    return success;
 398 }