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, struct gl_program *prog,
  39                        const struct brw_tes_prog_key *key)
  40 {
  41    struct brw_cache_item *c = NULL;
  42    const struct brw_tes_prog_key *old_key = NULL;
  43    bool found = false;
  44
  45    perf_debug("Recompiling tessellation evaluation shader for program %d\n",
  46               prog->Id);
  47
  48    for (unsigned int i = 0; i < brw->cache.size; i++) {
  49       for (c = brw->cache.items[i]; c; c = c->next) {
  50          if (c->cache_id == BRW_CACHE_TES_PROG) {
  51             old_key = c->key;
  52
  53             if (old_key->program_string_id == key->program_string_id)
  54                break;
  55          }
  56       }
  57       if (c)
  58          break;
  59    }
  60
  61    if (!c) {
  62       perf_debug("  Didn't find previous compile in the shader cache for "
  63                  "debug\n");
  64       return;
  65    }
  66
  67    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
  68    found |= key_debug(brw, "inputs read", old_key->inputs_read,
  69                       key->inputs_read);
  70    found |= key_debug(brw, "patch inputs read", old_key->patch_inputs_read,
  71                       key->patch_inputs_read);
  72
  73    if (!found) {
  74       perf_debug("  Something else\n");
  75    }
  76 }
  77
  78 static bool
  79 brw_codegen_tes_prog(struct brw_context *brw,
  80                      struct brw_program *tep,
  81                      struct brw_tes_prog_key *key)
  82 {
  83    const struct brw_compiler *compiler = brw->screen->compiler;
  84    const struct gen_device_info *devinfo = &brw->screen->devinfo;
  85    struct brw_stage_state *stage_state = &brw->tes.base;
  86    nir_shader *nir = tep->program.nir;
  87    struct brw_tes_prog_data prog_data;
  88    bool start_busy = false;
  89    double start_time = 0;
  90
  91    memset(&prog_data, 0, sizeof(prog_data));
  92
  93    brw_assign_common_binding_table_offsets(devinfo, &tep->program,
  94                                            &prog_data.base.base, 0);
  95
  96    STATIC_ASSERT(BRW_TESS_PARTITIONING_INTEGER == TESS_SPACING_EQUAL - 1);
  97    STATIC_ASSERT(BRW_TESS_PARTITIONING_ODD_FRACTIONAL ==
  98                  TESS_SPACING_FRACTIONAL_ODD - 1);
  99    STATIC_ASSERT(BRW_TESS_PARTITIONING_EVEN_FRACTIONAL ==
 100                  TESS_SPACING_FRACTIONAL_EVEN - 1);
 101
 102    prog_data.partitioning = nir->info->tes.spacing - 1;
 103
 104    switch (nir->info->tes.primitive_mode) {
 105    case GL_QUADS:
 106       prog_data.domain = BRW_TESS_DOMAIN_QUAD;
 107       break;
 108    case GL_TRIANGLES:
 109       prog_data.domain = BRW_TESS_DOMAIN_TRI;
 110       break;
 111    case GL_ISOLINES:
 112       prog_data.domain = BRW_TESS_DOMAIN_ISOLINE;
 113       break;
 114    default:
 115       unreachable("invalid domain shader primitive mode");
 116    }
 117
 118    if (nir->info->tes.point_mode) {
 119       prog_data.output_topology = BRW_TESS_OUTPUT_TOPOLOGY_POINT;
 120    } else if (nir->info->tes.primitive_mode == GL_ISOLINES) {
 121       prog_data.output_topology = BRW_TESS_OUTPUT_TOPOLOGY_LINE;
 122    } else {
 123       /* Hardware winding order is backwards from OpenGL */
 124       prog_data.output_topology =
 125          nir->info->tes.ccw ? BRW_TESS_OUTPUT_TOPOLOGY_TRI_CW
 126                             : BRW_TESS_OUTPUT_TOPOLOGY_TRI_CCW;
 127    }
 128
 129    /* Allocate the references to the uniforms that will end up in the
 130     * prog_data associated with the compiled program, and which will be freed
 131     * by the state cache.
 132     *
 133     * Note: param_count needs to be num_uniform_components * 4, since we add
 134     * padding around uniform values below vec4 size, so the worst case is that
 135     * every uniform is a float which gets padded to the size of a vec4.
 136     */
 137    int param_count = nir->num_uniforms / 4;
 138
 139    prog_data.base.base.param =
 140       rzalloc_array(NULL, const gl_constant_value *, param_count);
 141    prog_data.base.base.pull_param =
 142       rzalloc_array(NULL, const gl_constant_value *, param_count);
 143    prog_data.base.base.image_param =
 144       rzalloc_array(NULL, struct brw_image_param,
 145                     tep->program.info.num_images);
 146    prog_data.base.base.nr_params = param_count;
 147    prog_data.base.base.nr_image_params = tep->program.info.num_images;
 148
 149    brw_nir_setup_glsl_uniforms(nir, &tep->program, &prog_data.base.base,
 150                                compiler->scalar_stage[MESA_SHADER_TESS_EVAL]);
 151
 152    int st_index = -1;
 153    if (unlikely(INTEL_DEBUG & DEBUG_SHADER_TIME))
 154       st_index = brw_get_shader_time_index(brw, &tep->program, ST_TES, true);
 155
 156    if (unlikely(brw->perf_debug)) {
 157       start_busy = brw->batch.last_bo && drm_intel_bo_busy(brw->batch.last_bo);
 158       start_time = get_time();
 159    }
 160
 161    void *mem_ctx = ralloc_context(NULL);
 162    unsigned program_size;
 163    char *error_str;
 164    const unsigned *program =
 165       brw_compile_tes(compiler, brw, mem_ctx, key, &prog_data, nir,
 166                       &tep->program, st_index, &program_size, &error_str);
 167    if (program == NULL) {
 168       tep->program.sh.data->LinkStatus = false;
 169       ralloc_strcat(&tep->program.sh.data->InfoLog, error_str);
 170
 171       _mesa_problem(NULL, "Failed to compile tessellation evaluation shader: "
 172                     "%s\n", error_str);
 173
 174       ralloc_free(mem_ctx);
 175       return false;
 176    }
 177
 178    if (unlikely(brw->perf_debug)) {
 179       if (tep->compiled_once) {
 180          brw_tes_debug_recompile(brw, &tep->program, key);
 181       }
 182       if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
 183          perf_debug("TES compile took %.03f ms and stalled the GPU\n",
 184                     (get_time() - start_time) * 1000);
 185       }
 186       tep->compiled_once = true;
 187    }
 188
 189    /* Scratch space is used for register spilling */
 190    brw_alloc_stage_scratch(brw, stage_state,
 191                            prog_data.base.base.total_scratch,
 192                            devinfo->max_tes_threads);
 193
 194    brw_upload_cache(&brw->cache, BRW_CACHE_TES_PROG,
 195                     key, sizeof(*key),
 196                     program, program_size,
 197                     &prog_data, sizeof(prog_data),
 198                     &stage_state->prog_offset, &brw->tes.base.prog_data);
 199    ralloc_free(mem_ctx);
 200
 201    return true;
 202 }
 203
 204 void
 205 brw_tes_populate_key(struct brw_context *brw,
 206                      struct brw_tes_prog_key *key)
 207 {
 208    struct brw_program *tcp = (struct brw_program *) brw->tess_ctrl_program;
 209    struct brw_program *tep = (struct brw_program *) brw->tess_eval_program;
 210    struct gl_program *prog = &tep->program;
 211
 212    uint64_t per_vertex_slots = prog->info.inputs_read;
 213    uint32_t per_patch_slots = prog->info.patch_inputs_read;
 214
 215    memset(key, 0, sizeof(*key));
 216
 217    key->program_string_id = tep->id;
 218
 219    /* The TCS may have additional outputs which aren't read by the
 220     * TES (possibly for cross-thread communication).  These need to
 221     * be stored in the Patch URB Entry as well.
 222     */
 223    if (tcp) {
 224       struct gl_program *tcp_prog = &tcp->program;
 225       per_vertex_slots |= tcp_prog->info.outputs_written &
 226          ~(VARYING_BIT_TESS_LEVEL_INNER | VARYING_BIT_TESS_LEVEL_OUTER);
 227       per_patch_slots |= tcp_prog->info.patch_outputs_written;
 228    }
 229
 230    key->inputs_read = per_vertex_slots;
 231    key->patch_inputs_read = per_patch_slots;
 232
 233    /* _NEW_TEXTURE */
 234    brw_populate_sampler_prog_key_data(&brw->ctx, prog, &key->tex);
 235 }
 236
 237 void
 238 brw_upload_tes_prog(struct brw_context *brw)
 239 {
 240    struct brw_stage_state *stage_state = &brw->tes.base;
 241    struct brw_tes_prog_key key;
 242    /* BRW_NEW_TESS_PROGRAMS */
 243    struct brw_program *tep = (struct brw_program *) brw->tess_eval_program;
 244
 245    if (!brw_state_dirty(brw,
 246                         _NEW_TEXTURE,
 247                         BRW_NEW_TESS_PROGRAMS))
 248       return;
 249
 250    brw_tes_populate_key(brw, &key);
 251
 252    if (!brw_search_cache(&brw->cache, BRW_CACHE_TES_PROG,
 253                          &key, sizeof(key),
 254                          &stage_state->prog_offset,
 255                          &brw->tes.base.prog_data)) {
 256       bool success = brw_codegen_tes_prog(brw, tep, &key);
 257       assert(success);
 258       (void)success;
 259    }
 260 }
 261
 262
 263 bool
 264 brw_tes_precompile(struct gl_context *ctx,
 265                    struct gl_shader_program *shader_prog,
 266                    struct gl_program *prog)
 267 {
 268    struct brw_context *brw = brw_context(ctx);
 269    struct brw_tes_prog_key key;
 270    uint32_t old_prog_offset = brw->tes.base.prog_offset;
 271    struct brw_stage_prog_data *old_prog_data = brw->tes.base.prog_data;
 272    bool success;
 273
 274    struct brw_program *btep = brw_program(prog);
 275
 276    memset(&key, 0, sizeof(key));
 277
 278    key.program_string_id = btep->id;
 279    key.inputs_read = prog->nir->info->inputs_read;
 280    key.patch_inputs_read = prog->nir->info->patch_inputs_read;
 281
 282    if (shader_prog->_LinkedShaders[MESA_SHADER_TESS_CTRL]) {
 283       struct gl_program *tcp =
 284          shader_prog->_LinkedShaders[MESA_SHADER_TESS_CTRL]->Program;
 285       key.inputs_read |= tcp->nir->info->outputs_written &
 286          ~(VARYING_BIT_TESS_LEVEL_INNER | VARYING_BIT_TESS_LEVEL_OUTER);
 287       key.patch_inputs_read |= tcp->nir->info->patch_outputs_written;
 288    }
 289
 290    brw_setup_tex_for_precompile(brw, &key.tex, prog);
 291
 292    success = brw_codegen_tes_prog(brw, btep, &key);
 293
 294    brw->tes.base.prog_offset = old_prog_offset;
 295    brw->tes.base.prog_data = old_prog_data;
 296
 297    return success;
 298 }