src/mesa/drivers/dri/i965/brw_tes.c

   1 /*
   2  * Copyright © 2014 Intel Corporation
   3  *
   4  * Permission is hereby granted, free of charge, to any person obtaining a
   5  * copy of this software and associated documentation files (the "Software"),
   6  * to deal in the Software without restriction, including without limitation
   7  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
   8  * and/or sell copies of the Software, and to permit persons to whom the
   9  * Software is furnished to do so, subject to the following conditions:
  10  *
  11  * The above copyright notice and this permission notice (including the next
  12  * paragraph) shall be included in all copies or substantial portions of the
  13  * Software.
  14  *
  15  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
  16  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
  17  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
  18  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
  19  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
  20  * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
  21  * DEALINGS IN THE SOFTWARE.
  22  */
  23
  24 /**
  25  * \file brw_tes.c
  26  *
  27  * Tessellation evaluation shader state upload code.
  28  */
  29
  30 #include "brw_context.h"
  31 #include "brw_nir.h"
  32 #include "brw_program.h"
  33 #include "brw_shader.h"
  34 #include "brw_state.h"
  35 #include "program/prog_parameter.h"
  36
  37 static void
  38 brw_tes_debug_recompile(struct brw_context *brw,
  39                        struct gl_shader_program *shader_prog,
  40                        const struct brw_tes_prog_key *key)
  41 {
  42    struct brw_cache_item *c = NULL;
  43    const struct brw_tes_prog_key *old_key = NULL;
  44    bool found = false;
  45
  46    perf_debug("Recompiling tessellation evaluation shader for program %d\n",
  47               shader_prog->Name);
  48
  49    for (unsigned int i = 0; i < brw->cache.size; i++) {
  50       for (c = brw->cache.items[i]; c; c = c->next) {
  51          if (c->cache_id == BRW_CACHE_TES_PROG) {
  52             old_key = c->key;
  53
  54             if (old_key->program_string_id == key->program_string_id)
  55                break;
  56          }
  57       }
  58       if (c)
  59          break;
  60    }
  61
  62    if (!c) {
  63       perf_debug("  Didn't find previous compile in the shader cache for "
  64                  "debug\n");
  65       return;
  66    }
  67
  68    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
  69    found |= key_debug(brw, "inputs read", old_key->inputs_read,
  70                       key->inputs_read);
  71    found |= key_debug(brw, "patch inputs read", old_key->patch_inputs_read,
  72                       key->patch_inputs_read);
  73
  74    if (!found) {
  75       perf_debug("  Something else\n");
  76    }
  77 }
  78
  79 static bool
  80 brw_codegen_tes_prog(struct brw_context *brw,
  81                      struct gl_shader_program *shader_prog,
  82                      struct brw_tess_eval_program *tep,
  83                      struct brw_tes_prog_key *key)
  84 {
  85    const struct brw_compiler *compiler = brw->screen->compiler;
  86    const struct gen_device_info *devinfo = &brw->screen->devinfo;
  87    struct brw_stage_state *stage_state = &brw->tes.base;
  88    nir_shader *nir = tep->program.nir;
  89    struct brw_tes_prog_data prog_data;
  90    bool start_busy = false;
  91    double start_time = 0;
  92
  93    memset(&prog_data, 0, sizeof(prog_data));
  94
  95    brw_assign_common_binding_table_offsets(MESA_SHADER_TESS_EVAL, devinfo,
  96                                            shader_prog, &tep->program,
  97                                            &prog_data.base.base, 0);
  98
  99    switch (tep->program.info.tes.spacing) {
 100    case GL_EQUAL:
 101       prog_data.partitioning = BRW_TESS_PARTITIONING_INTEGER;
 102       break;
 103    case GL_FRACTIONAL_ODD:
 104       prog_data.partitioning = BRW_TESS_PARTITIONING_ODD_FRACTIONAL;
 105       break;
 106    case GL_FRACTIONAL_EVEN:
 107       prog_data.partitioning = BRW_TESS_PARTITIONING_EVEN_FRACTIONAL;
 108       break;
 109    default:
 110       unreachable("invalid domain shader spacing");
 111    }
 112
 113    switch (tep->program.info.tes.primitive_mode) {
 114    case GL_QUADS:
 115       prog_data.domain = BRW_TESS_DOMAIN_QUAD;
 116       break;
 117    case GL_TRIANGLES:
 118       prog_data.domain = BRW_TESS_DOMAIN_TRI;
 119       break;
 120    case GL_ISOLINES:
 121       prog_data.domain = BRW_TESS_DOMAIN_ISOLINE;
 122       break;
 123    default:
 124       unreachable("invalid domain shader primitive mode");
 125    }
 126
 127    if (tep->program.info.tes.point_mode) {
 128       prog_data.output_topology = BRW_TESS_OUTPUT_TOPOLOGY_POINT;
 129    } else if (tep->program.info.tes.primitive_mode == GL_ISOLINES) {
 130       prog_data.output_topology = BRW_TESS_OUTPUT_TOPOLOGY_LINE;
 131    } else {
 132       /* Hardware winding order is backwards from OpenGL */
 133       switch (tep->program.info.tes.vertex_order) {
 134       case GL_CCW:
 135          prog_data.output_topology = BRW_TESS_OUTPUT_TOPOLOGY_TRI_CW;
 136          break;
 137       case GL_CW:
 138          prog_data.output_topology = BRW_TESS_OUTPUT_TOPOLOGY_TRI_CCW;
 139          break;
 140       default:
 141          unreachable("invalid domain shader vertex order");
 142       }
 143    }
 144
 145    /* Allocate the references to the uniforms that will end up in the
 146     * prog_data associated with the compiled program, and which will be freed
 147     * by the state cache.
 148     *
 149     * Note: param_count needs to be num_uniform_components * 4, since we add
 150     * padding around uniform values below vec4 size, so the worst case is that
 151     * every uniform is a float which gets padded to the size of a vec4.
 152     */
 153    struct gl_linked_shader *tes =
 154       shader_prog->_LinkedShaders[MESA_SHADER_TESS_EVAL];
 155    int param_count = nir->num_uniforms / 4;
 156
 157    prog_data.base.base.param =
 158       rzalloc_array(NULL, const gl_constant_value *, param_count);
 159    prog_data.base.base.pull_param =
 160       rzalloc_array(NULL, const gl_constant_value *, param_count);
 161    prog_data.base.base.image_param =
 162       rzalloc_array(NULL, struct brw_image_param, tes->NumImages);
 163    prog_data.base.base.nr_params = param_count;
 164    prog_data.base.base.nr_image_params = tes->NumImages;
 165
 166    prog_data.base.cull_distance_mask =
 167       ((1 << tep->program.CullDistanceArraySize) - 1) <<
 168       tep->program.ClipDistanceArraySize;
 169
 170    brw_nir_setup_glsl_uniforms(nir, shader_prog, &tep->program,
 171                                &prog_data.base.base,
 172                                compiler->scalar_stage[MESA_SHADER_TESS_EVAL]);
 173
 174    if (unlikely(INTEL_DEBUG & DEBUG_TES))
 175       brw_dump_ir("tessellation evaluation", shader_prog, tes, NULL);
 176
 177    int st_index = -1;
 178    if (unlikely(INTEL_DEBUG & DEBUG_SHADER_TIME))
 179       st_index = brw_get_shader_time_index(brw, shader_prog, NULL, ST_TES);
 180
 181    if (unlikely(brw->perf_debug)) {
 182       start_busy = brw->batch.last_bo && drm_intel_bo_busy(brw->batch.last_bo);
 183       start_time = get_time();
 184    }
 185
 186    void *mem_ctx = ralloc_context(NULL);
 187    unsigned program_size;
 188    char *error_str;
 189    const unsigned *program =
 190       brw_compile_tes(compiler, brw, mem_ctx, key, &prog_data, nir,
 191                       shader_prog, st_index, &program_size, &error_str);
 192    if (program == NULL) {
 193       if (shader_prog) {
 194          shader_prog->LinkStatus = false;
 195          ralloc_strcat(&shader_prog->InfoLog, error_str);
 196       }
 197
 198       _mesa_problem(NULL, "Failed to compile tessellation evaluation shader: "
 199                     "%s\n", error_str);
 200
 201       ralloc_free(mem_ctx);
 202       return false;
 203    }
 204
 205    if (unlikely(brw->perf_debug)) {
 206       struct brw_shader *btes = (struct brw_shader *) tes;
 207       if (btes->compiled_once) {
 208          brw_tes_debug_recompile(brw, shader_prog, key);
 209       }
 210       if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
 211          perf_debug("TES compile took %.03f ms and stalled the GPU\n",
 212                     (get_time() - start_time) * 1000);
 213       }
 214       btes->compiled_once = true;
 215    }
 216
 217    /* Scratch space is used for register spilling */
 218    brw_alloc_stage_scratch(brw, stage_state,
 219                            prog_data.base.base.total_scratch,
 220                            devinfo->max_tes_threads);
 221
 222    brw_upload_cache(&brw->cache, BRW_CACHE_TES_PROG,
 223                     key, sizeof(*key),
 224                     program, program_size,
 225                     &prog_data, sizeof(prog_data),
 226                     &stage_state->prog_offset, &brw->tes.base.prog_data);
 227    ralloc_free(mem_ctx);
 228
 229    return true;
 230 }
 231
 232 void
 233 brw_tes_populate_key(struct brw_context *brw,
 234                      struct brw_tes_prog_key *key)
 235 {
 236    struct brw_tess_ctrl_program *tcp =
 237       (struct brw_tess_ctrl_program *) brw->tess_ctrl_program;
 238    struct brw_tess_eval_program *tep =
 239       (struct brw_tess_eval_program *) brw->tess_eval_program;
 240    struct gl_program *prog = &tep->program;
 241
 242    uint64_t per_vertex_slots = prog->info.inputs_read;
 243    uint32_t per_patch_slots = prog->info.patch_inputs_read;
 244
 245    memset(key, 0, sizeof(*key));
 246
 247    key->program_string_id = tep->id;
 248
 249    /* The TCS may have additional outputs which aren't read by the
 250     * TES (possibly for cross-thread communication).  These need to
 251     * be stored in the Patch URB Entry as well.
 252     */
 253    if (tcp) {
 254       struct gl_program *tcp_prog = &tcp->program;
 255       per_vertex_slots |= tcp_prog->info.outputs_written;
 256       per_patch_slots |= tcp_prog->info.patch_outputs_written;
 257    }
 258
 259    /* Ignore gl_TessLevelInner/Outer - we treat them as system values,
 260     * not inputs, and they're always present in the URB entry regardless
 261     * of whether or not we read them.
 262     */
 263    key->inputs_read = per_vertex_slots &
 264       ~(VARYING_BIT_TESS_LEVEL_INNER | VARYING_BIT_TESS_LEVEL_OUTER);
 265    key->patch_inputs_read = per_patch_slots;
 266
 267    /* _NEW_TEXTURE */
 268    brw_populate_sampler_prog_key_data(&brw->ctx, prog, &key->tex);
 269 }
 270
 271 void
 272 brw_upload_tes_prog(struct brw_context *brw)
 273 {
 274    struct gl_shader_program **current = brw->ctx._Shader->CurrentProgram;
 275    struct brw_stage_state *stage_state = &brw->tes.base;
 276    struct brw_tes_prog_key key;
 277    /* BRW_NEW_TESS_PROGRAMS */
 278    struct brw_tess_eval_program *tep =
 279       (struct brw_tess_eval_program *) brw->tess_eval_program;
 280
 281    if (!brw_state_dirty(brw,
 282                         _NEW_TEXTURE,
 283                         BRW_NEW_TESS_PROGRAMS))
 284       return;
 285
 286    brw_tes_populate_key(brw, &key);
 287
 288    if (!brw_search_cache(&brw->cache, BRW_CACHE_TES_PROG,
 289                          &key, sizeof(key),
 290                          &stage_state->prog_offset,
 291                          &brw->tes.base.prog_data)) {
 292       bool success = brw_codegen_tes_prog(brw, current[MESA_SHADER_TESS_EVAL],
 293                                           tep, &key);
 294       assert(success);
 295       (void)success;
 296    }
 297 }
 298
 299
 300 bool
 301 brw_tes_precompile(struct gl_context *ctx,
 302                    struct gl_shader_program *shader_prog,
 303                    struct gl_program *prog)
 304 {
 305    struct brw_context *brw = brw_context(ctx);
 306    struct brw_tes_prog_key key;
 307    uint32_t old_prog_offset = brw->tes.base.prog_offset;
 308    struct brw_stage_prog_data *old_prog_data = brw->tes.base.prog_data;
 309    bool success;
 310
 311    struct brw_tess_eval_program *btep = brw_tess_eval_program(prog);
 312
 313    memset(&key, 0, sizeof(key));
 314
 315    key.program_string_id = btep->id;
 316    key.inputs_read = prog->nir->info->inputs_read;
 317    key.patch_inputs_read = prog->nir->info->patch_inputs_read;
 318
 319    if (shader_prog->_LinkedShaders[MESA_SHADER_TESS_CTRL]) {
 320       struct gl_program *tcp =
 321          shader_prog->_LinkedShaders[MESA_SHADER_TESS_CTRL]->Program;
 322       key.inputs_read |= tcp->nir->info->outputs_written;
 323       key.patch_inputs_read |= tcp->nir->info->patch_outputs_written;
 324    }
 325
 326    /* Ignore gl_TessLevelInner/Outer - they're system values. */
 327    key.inputs_read &= ~(VARYING_BIT_TESS_LEVEL_INNER |
 328                         VARYING_BIT_TESS_LEVEL_OUTER);
 329
 330    brw_setup_tex_for_precompile(brw, &key.tex, prog);
 331
 332    success = brw_codegen_tes_prog(brw, shader_prog, btep, &key);
 333
 334    brw->tes.base.prog_offset = old_prog_offset;
 335    brw->tes.base.prog_data = old_prog_data;
 336
 337    return success;
 338 }