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 bool
  45 brw_codegen_vs_prog(struct brw_context *brw,
  46                     struct gl_shader_program *prog,
  47                     struct brw_vertex_program *vp,
  48                     struct brw_vs_prog_key *key)
  49 {
  50    GLuint program_size;
  51    const GLuint *program;
  52    struct brw_vs_prog_data prog_data;
  53    struct brw_stage_prog_data *stage_prog_data = &prog_data.base.base;
  54    void *mem_ctx;
  55    int i;
  56    struct brw_shader *vs = NULL;
  57    bool start_busy = false;
  58    double start_time = 0;
  59
  60    if (!vp->program.Base.nir) {
  61       /* Normally we generate NIR in LinkShader() or
  62        * ProgramStringNotify(), but Mesa's fixed-function vertex program
  63        * handling doesn't notify the driver at all.  Just do it here, at
  64        * the last minute, even though it's lame.
  65        */
  66       assert(vp->program.Base.Id == 0 && prog == NULL);
  67       vp->program.Base.nir =
  68          brw_create_nir(brw, NULL, &vp->program.Base, MESA_SHADER_VERTEX,
  69                         brw->intelScreen->compiler->scalar_vs);
  70    }
  71
  72    if (prog)
  73       vs = (struct brw_shader *) prog->_LinkedShaders[MESA_SHADER_VERTEX];
  74
  75    memset(&prog_data, 0, sizeof(prog_data));
  76
  77    /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */
  78    if (!prog)
  79       stage_prog_data->use_alt_mode = true;
  80
  81    mem_ctx = ralloc_context(NULL);
  82
  83    brw_assign_common_binding_table_offsets(MESA_SHADER_VERTEX,
  84                                            brw->intelScreen->devinfo,
  85                                            prog, &vp->program.Base,
  86                                            &prog_data.base.base, 0);
  87
  88    /* Allocate the references to the uniforms that will end up in the
  89     * prog_data associated with the compiled program, and which will be freed
  90     * by the state cache.
  91     */
  92    int param_count = vp->program.Base.nir->num_uniforms;
  93    if (!brw->intelScreen->compiler->scalar_vs)
  94       param_count *= 4;
  95
  96    if (vs)
  97       prog_data.base.base.nr_image_params = vs->base.NumImages;
  98
  99    /* vec4_visitor::setup_uniform_clipplane_values() also uploads user clip
 100     * planes as uniforms.
 101     */
 102    param_count += key->nr_userclip_plane_consts * 4;
 103
 104    stage_prog_data->param =
 105       rzalloc_array(NULL, const gl_constant_value *, param_count);
 106    stage_prog_data->pull_param =
 107       rzalloc_array(NULL, const gl_constant_value *, param_count);
 108    stage_prog_data->image_param =
 109       rzalloc_array(NULL, struct brw_image_param,
 110                     stage_prog_data->nr_image_params);
 111    stage_prog_data->nr_params = param_count;
 112
 113    if (prog) {
 114       brw_nir_setup_glsl_uniforms(vp->program.Base.nir, prog, &vp->program.Base,
 115                                   &prog_data.base.base,
 116                                   brw->intelScreen->compiler->scalar_vs);
 117    } else {
 118       brw_nir_setup_arb_uniforms(vp->program.Base.nir, &vp->program.Base,
 119                                  &prog_data.base.base);
 120    }
 121
 122    GLbitfield64 outputs_written = vp->program.Base.OutputsWritten;
 123    prog_data.inputs_read = vp->program.Base.InputsRead;
 124
 125    if (key->copy_edgeflag) {
 126       outputs_written |= BITFIELD64_BIT(VARYING_SLOT_EDGE);
 127       prog_data.inputs_read |= VERT_BIT_EDGEFLAG;
 128    }
 129
 130    if (brw->gen < 6) {
 131       /* Put dummy slots into the VUE for the SF to put the replaced
 132        * point sprite coords in.  We shouldn't need these dummy slots,
 133        * which take up precious URB space, but it would mean that the SF
 134        * doesn't get nice aligned pairs of input coords into output
 135        * coords, which would be a pain to handle.
 136        */
 137       for (i = 0; i < 8; i++) {
 138          if (key->point_coord_replace & (1 << i))
 139             outputs_written |= BITFIELD64_BIT(VARYING_SLOT_TEX0 + i);
 140       }
 141
 142       /* if back colors are written, allocate slots for front colors too */
 143       if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC0))
 144          outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL0);
 145       if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC1))
 146          outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL1);
 147    }
 148
 149    /* In order for legacy clipping to work, we need to populate the clip
 150     * distance varying slots whenever clipping is enabled, even if the vertex
 151     * shader doesn't write to gl_ClipDistance.
 152     */
 153    if (key->nr_userclip_plane_consts > 0) {
 154       outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0);
 155       outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1);
 156    }
 157
 158    brw_compute_vue_map(brw->intelScreen->devinfo,
 159                        &prog_data.base.vue_map, outputs_written,
 160                        prog ? prog->SeparateShader : false);
 161
 162    unsigned nr_attributes = _mesa_bitcount_64(prog_data.inputs_read);
 163
 164    /* gl_VertexID and gl_InstanceID are system values, but arrive via an
 165     * incoming vertex attribute.  So, add an extra slot.
 166     */
 167    if (vp->program.Base.SystemValuesRead &
 168        (BITFIELD64_BIT(SYSTEM_VALUE_VERTEX_ID_ZERO_BASE) |
 169         BITFIELD64_BIT(SYSTEM_VALUE_INSTANCE_ID))) {
 170       nr_attributes++;
 171    }
 172
 173    /* The 3DSTATE_VS documentation lists the lower bound on "Vertex URB Entry
 174     * Read Length" as 1 in vec4 mode, and 0 in SIMD8 mode.  Empirically, in
 175     * vec4 mode, the hardware appears to wedge unless we read something.
 176     */
 177    if (brw->intelScreen->compiler->scalar_vs)
 178       prog_data.base.urb_read_length = DIV_ROUND_UP(nr_attributes, 2);
 179    else
 180       prog_data.base.urb_read_length = DIV_ROUND_UP(MAX2(nr_attributes, 1), 2);
 181
 182    prog_data.nr_attributes = nr_attributes;
 183
 184    /* Since vertex shaders reuse the same VUE entry for inputs and outputs
 185     * (overwriting the original contents), we need to make sure the size is
 186     * the larger of the two.
 187     */
 188    const unsigned vue_entries =
 189       MAX2(nr_attributes, prog_data.base.vue_map.num_slots);
 190
 191    if (brw->gen == 6)
 192       prog_data.base.urb_entry_size = DIV_ROUND_UP(vue_entries, 8);
 193    else
 194       prog_data.base.urb_entry_size = DIV_ROUND_UP(vue_entries, 4);
 195
 196    if (0) {
 197       _mesa_fprint_program_opt(stderr, &vp->program.Base, PROG_PRINT_DEBUG,
 198                                true);
 199    }
 200
 201    if (unlikely(brw->perf_debug)) {
 202       start_busy = (brw->batch.last_bo &&
 203                     drm_intel_bo_busy(brw->batch.last_bo));
 204       start_time = get_time();
 205    }
 206
 207    if (unlikely(INTEL_DEBUG & DEBUG_VS))
 208       brw_dump_ir("vertex", prog, vs ? &vs->base : NULL, &vp->program.Base);
 209
 210    int st_index = -1;
 211    if (INTEL_DEBUG & DEBUG_SHADER_TIME)
 212       st_index = brw_get_shader_time_index(brw, prog, &vp->program.Base, ST_VS);
 213
 214    /* Emit GEN4 code.
 215     */
 216    program = brw_vs_emit(brw, mem_ctx, key, &prog_data,
 217                          &vp->program, prog, st_index, &program_size);
 218    if (program == NULL) {
 219       ralloc_free(mem_ctx);
 220       return false;
 221    }
 222
 223    if (unlikely(brw->perf_debug) && vs) {
 224       if (vs->compiled_once) {
 225          brw_vs_debug_recompile(brw, prog, key);
 226       }
 227       if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
 228          perf_debug("VS compile took %.03f ms and stalled the GPU\n",
 229                     (get_time() - start_time) * 1000);
 230       }
 231       vs->compiled_once = true;
 232    }
 233
 234    /* Scratch space is used for register spilling */
 235    if (prog_data.base.base.total_scratch) {
 236       brw_get_scratch_bo(brw, &brw->vs.base.scratch_bo,
 237                          prog_data.base.base.total_scratch *
 238                          brw->max_vs_threads);
 239    }
 240
 241    brw_upload_cache(&brw->cache, BRW_CACHE_VS_PROG,
 242                     key, sizeof(struct brw_vs_prog_key),
 243                     program, program_size,
 244                     &prog_data, sizeof(prog_data),
 245                     &brw->vs.base.prog_offset, &brw->vs.prog_data);
 246    ralloc_free(mem_ctx);
 247
 248    return true;
 249 }
 250
 251 static bool
 252 key_debug(struct brw_context *brw, const char *name, int a, int b)
 253 {
 254    if (a != b) {
 255       perf_debug("  %s %d->%d\n", name, a, b);
 256       return true;
 257    }
 258    return false;
 259 }
 260
 261 void
 262 brw_vs_debug_recompile(struct brw_context *brw,
 263                        struct gl_shader_program *prog,
 264                        const struct brw_vs_prog_key *key)
 265 {
 266    struct brw_cache_item *c = NULL;
 267    const struct brw_vs_prog_key *old_key = NULL;
 268    bool found = false;
 269
 270    perf_debug("Recompiling vertex shader for program %d\n", prog->Name);
 271
 272    for (unsigned int i = 0; i < brw->cache.size; i++) {
 273       for (c = brw->cache.items[i]; c; c = c->next) {
 274          if (c->cache_id == BRW_CACHE_VS_PROG) {
 275             old_key = c->key;
 276
 277             if (old_key->program_string_id == key->program_string_id)
 278                break;
 279          }
 280       }
 281       if (c)
 282          break;
 283    }
 284
 285    if (!c) {
 286       perf_debug("  Didn't find previous compile in the shader cache for "
 287                  "debug\n");
 288       return;
 289    }
 290
 291    for (unsigned int i = 0; i < VERT_ATTRIB_MAX; i++) {
 292       found |= key_debug(brw, "Vertex attrib w/a flags",
 293                          old_key->gl_attrib_wa_flags[i],
 294                          key->gl_attrib_wa_flags[i]);
 295    }
 296
 297    found |= key_debug(brw, "legacy user clipping",
 298                       old_key->nr_userclip_plane_consts,
 299                       key->nr_userclip_plane_consts);
 300
 301    found |= key_debug(brw, "copy edgeflag",
 302                       old_key->copy_edgeflag, key->copy_edgeflag);
 303    found |= key_debug(brw, "PointCoord replace",
 304                       old_key->point_coord_replace, key->point_coord_replace);
 305    found |= key_debug(brw, "vertex color clamping",
 306                       old_key->clamp_vertex_color, key->clamp_vertex_color);
 307
 308    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
 309
 310    if (!found) {
 311       perf_debug("  Something else\n");
 312    }
 313 }
 314
 315 static bool
 316 brw_vs_state_dirty(struct brw_context *brw)
 317 {
 318    return brw_state_dirty(brw,
 319                           _NEW_BUFFERS |
 320                           _NEW_LIGHT |
 321                           _NEW_POINT |
 322                           _NEW_POLYGON |
 323                           _NEW_TEXTURE |
 324                           _NEW_TRANSFORM,
 325                           BRW_NEW_VERTEX_PROGRAM |
 326                           BRW_NEW_VS_ATTRIB_WORKAROUNDS);
 327 }
 328
 329 static void
 330 brw_vs_populate_key(struct brw_context *brw,
 331                     struct brw_vs_prog_key *key)
 332 {
 333    struct gl_context *ctx = &brw->ctx;
 334    /* BRW_NEW_VERTEX_PROGRAM */
 335    struct brw_vertex_program *vp =
 336       (struct brw_vertex_program *)brw->vertex_program;
 337    struct gl_program *prog = (struct gl_program *) brw->vertex_program;
 338    int i;
 339
 340    memset(key, 0, sizeof(*key));
 341
 342    /* Just upload the program verbatim for now.  Always send it all
 343     * the inputs it asks for, whether they are varying or not.
 344     */
 345    key->program_string_id = vp->id;
 346
 347    if (ctx->Transform.ClipPlanesEnabled != 0 &&
 348        ctx->API == API_OPENGL_COMPAT &&
 349        !vp->program.Base.UsesClipDistanceOut) {
 350       key->nr_userclip_plane_consts =
 351          _mesa_logbase2(ctx->Transform.ClipPlanesEnabled) + 1;
 352    }
 353
 354    /* _NEW_POLYGON */
 355    if (brw->gen < 6) {
 356       key->copy_edgeflag = (ctx->Polygon.FrontMode != GL_FILL ||
 357                             ctx->Polygon.BackMode != GL_FILL);
 358    }
 359
 360    if (prog->OutputsWritten & (VARYING_BIT_COL0 | VARYING_BIT_COL1 |
 361                                VARYING_BIT_BFC0 | VARYING_BIT_BFC1)) {
 362       /* _NEW_LIGHT | _NEW_BUFFERS */
 363       key->clamp_vertex_color = ctx->Light._ClampVertexColor;
 364    }
 365
 366    /* _NEW_POINT */
 367    if (brw->gen < 6 && ctx->Point.PointSprite) {
 368       for (i = 0; i < 8; i++) {
 369          if (ctx->Point.CoordReplace[i])
 370             key->point_coord_replace |= (1 << i);
 371       }
 372    }
 373
 374    /* _NEW_TEXTURE */
 375    brw_populate_sampler_prog_key_data(ctx, prog, brw->vs.base.sampler_count,
 376                                       &key->tex);
 377
 378    /* BRW_NEW_VS_ATTRIB_WORKAROUNDS */
 379    memcpy(key->gl_attrib_wa_flags, brw->vb.attrib_wa_flags,
 380           sizeof(brw->vb.attrib_wa_flags));
 381 }
 382
 383 void
 384 brw_upload_vs_prog(struct brw_context *brw)
 385 {
 386    struct gl_context *ctx = &brw->ctx;
 387    struct gl_shader_program **current = ctx->_Shader->CurrentProgram;
 388    struct brw_vs_prog_key key;
 389    /* BRW_NEW_VERTEX_PROGRAM */
 390    struct brw_vertex_program *vp =
 391       (struct brw_vertex_program *)brw->vertex_program;
 392
 393    if (!brw_vs_state_dirty(brw))
 394       return;
 395
 396    brw_vs_populate_key(brw, &key);
 397
 398    if (!brw_search_cache(&brw->cache, BRW_CACHE_VS_PROG,
 399                          &key, sizeof(key),
 400                          &brw->vs.base.prog_offset, &brw->vs.prog_data)) {
 401       bool success = brw_codegen_vs_prog(brw, current[MESA_SHADER_VERTEX],
 402                                          vp, &key);
 403       (void) success;
 404       assert(success);
 405    }
 406    brw->vs.base.prog_data = &brw->vs.prog_data->base.base;
 407 }
 408
 409 bool
 410 brw_vs_precompile(struct gl_context *ctx,
 411                   struct gl_shader_program *shader_prog,
 412                   struct gl_program *prog)
 413 {
 414    struct brw_context *brw = brw_context(ctx);
 415    struct brw_vs_prog_key key;
 416    uint32_t old_prog_offset = brw->vs.base.prog_offset;
 417    struct brw_vs_prog_data *old_prog_data = brw->vs.prog_data;
 418    bool success;
 419
 420    struct gl_vertex_program *vp = (struct gl_vertex_program *) prog;
 421    struct brw_vertex_program *bvp = brw_vertex_program(vp);
 422
 423    memset(&key, 0, sizeof(key));
 424
 425    brw_setup_tex_for_precompile(brw, &key.tex, prog);
 426    key.program_string_id = bvp->id;
 427    key.clamp_vertex_color =
 428       (prog->OutputsWritten & (VARYING_BIT_COL0 | VARYING_BIT_COL1 |
 429                                VARYING_BIT_BFC0 | VARYING_BIT_BFC1));
 430
 431    success = brw_codegen_vs_prog(brw, shader_prog, bvp, &key);
 432
 433    brw->vs.base.prog_offset = old_prog_offset;
 434    brw->vs.prog_data = old_prog_data;
 435
 436    return success;
 437 }