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    switch (tep->program.info.tes.spacing) {
  97    case GL_EQUAL:
  98       prog_data.partitioning = BRW_TESS_PARTITIONING_INTEGER;
  99       break;
 100    case GL_FRACTIONAL_ODD:
 101       prog_data.partitioning = BRW_TESS_PARTITIONING_ODD_FRACTIONAL;
 102       break;
 103    case GL_FRACTIONAL_EVEN:
 104       prog_data.partitioning = BRW_TESS_PARTITIONING_EVEN_FRACTIONAL;
 105       break;
 106    default:
 107       unreachable("invalid domain shader spacing");
 108    }
 109
 110    switch (tep->program.info.tes.primitive_mode) {
 111    case GL_QUADS:
 112       prog_data.domain = BRW_TESS_DOMAIN_QUAD;
 113       break;
 114    case GL_TRIANGLES:
 115       prog_data.domain = BRW_TESS_DOMAIN_TRI;
 116       break;
 117    case GL_ISOLINES:
 118       prog_data.domain = BRW_TESS_DOMAIN_ISOLINE;
 119       break;
 120    default:
 121       unreachable("invalid domain shader primitive mode");
 122    }
 123
 124    if (tep->program.info.tes.point_mode) {
 125       prog_data.output_topology = BRW_TESS_OUTPUT_TOPOLOGY_POINT;
 126    } else if (tep->program.info.tes.primitive_mode == GL_ISOLINES) {
 127       prog_data.output_topology = BRW_TESS_OUTPUT_TOPOLOGY_LINE;
 128    } else {
 129       /* Hardware winding order is backwards from OpenGL */
 130       prog_data.output_topology =
 131          tep->program.info.tes.ccw ? BRW_TESS_OUTPUT_TOPOLOGY_TRI_CW
 132                                    : BRW_TESS_OUTPUT_TOPOLOGY_TRI_CCW;
 133    }
 134
 135    /* Allocate the references to the uniforms that will end up in the
 136     * prog_data associated with the compiled program, and which will be freed
 137     * by the state cache.
 138     *
 139     * Note: param_count needs to be num_uniform_components * 4, since we add
 140     * padding around uniform values below vec4 size, so the worst case is that
 141     * every uniform is a float which gets padded to the size of a vec4.
 142     */
 143    int param_count = nir->num_uniforms / 4;
 144
 145    prog_data.base.base.param =
 146       rzalloc_array(NULL, const gl_constant_value *, param_count);
 147    prog_data.base.base.pull_param =
 148       rzalloc_array(NULL, const gl_constant_value *, param_count);
 149    prog_data.base.base.image_param =
 150       rzalloc_array(NULL, struct brw_image_param,
 151                     tep->program.info.num_images);
 152    prog_data.base.base.nr_params = param_count;
 153    prog_data.base.base.nr_image_params = tep->program.info.num_images;
 154
 155    brw_nir_setup_glsl_uniforms(nir, &tep->program, &prog_data.base.base,
 156                                compiler->scalar_stage[MESA_SHADER_TESS_EVAL]);
 157
 158    int st_index = -1;
 159    if (unlikely(INTEL_DEBUG & DEBUG_SHADER_TIME))
 160       st_index = brw_get_shader_time_index(brw, &tep->program, ST_TES, true);
 161
 162    if (unlikely(brw->perf_debug)) {
 163       start_busy = brw->batch.last_bo && drm_intel_bo_busy(brw->batch.last_bo);
 164       start_time = get_time();
 165    }
 166
 167    void *mem_ctx = ralloc_context(NULL);
 168    unsigned program_size;
 169    char *error_str;
 170    const unsigned *program =
 171       brw_compile_tes(compiler, brw, mem_ctx, key, &prog_data, nir,
 172                       &tep->program, st_index, &program_size, &error_str);
 173    if (program == NULL) {
 174       tep->program.sh.data->LinkStatus = false;
 175       ralloc_strcat(&tep->program.sh.data->InfoLog, error_str);
 176
 177       _mesa_problem(NULL, "Failed to compile tessellation evaluation shader: "
 178                     "%s\n", error_str);
 179
 180       ralloc_free(mem_ctx);
 181       return false;
 182    }
 183
 184    if (unlikely(brw->perf_debug)) {
 185       if (tep->compiled_once) {
 186          brw_tes_debug_recompile(brw, &tep->program, key);
 187       }
 188       if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
 189          perf_debug("TES compile took %.03f ms and stalled the GPU\n",
 190                     (get_time() - start_time) * 1000);
 191       }
 192       tep->compiled_once = true;
 193    }
 194
 195    /* Scratch space is used for register spilling */
 196    brw_alloc_stage_scratch(brw, stage_state,
 197                            prog_data.base.base.total_scratch,
 198                            devinfo->max_tes_threads);
 199
 200    brw_upload_cache(&brw->cache, BRW_CACHE_TES_PROG,
 201                     key, sizeof(*key),
 202                     program, program_size,
 203                     &prog_data, sizeof(prog_data),
 204                     &stage_state->prog_offset, &brw->tes.base.prog_data);
 205    ralloc_free(mem_ctx);
 206
 207    return true;
 208 }
 209
 210 void
 211 brw_tes_populate_key(struct brw_context *brw,
 212                      struct brw_tes_prog_key *key)
 213 {
 214    struct brw_program *tcp = (struct brw_program *) brw->tess_ctrl_program;
 215    struct brw_program *tep = (struct brw_program *) brw->tess_eval_program;
 216    struct gl_program *prog = &tep->program;
 217
 218    uint64_t per_vertex_slots = prog->info.inputs_read;
 219    uint32_t per_patch_slots = prog->info.patch_inputs_read;
 220
 221    memset(key, 0, sizeof(*key));
 222
 223    key->program_string_id = tep->id;
 224
 225    /* The TCS may have additional outputs which aren't read by the
 226     * TES (possibly for cross-thread communication).  These need to
 227     * be stored in the Patch URB Entry as well.
 228     */
 229    if (tcp) {
 230       struct gl_program *tcp_prog = &tcp->program;
 231       per_vertex_slots |= tcp_prog->info.outputs_written &
 232          ~(VARYING_BIT_TESS_LEVEL_INNER | VARYING_BIT_TESS_LEVEL_OUTER);
 233       per_patch_slots |= tcp_prog->info.patch_outputs_written;
 234    }
 235
 236    key->inputs_read = per_vertex_slots;
 237    key->patch_inputs_read = per_patch_slots;
 238
 239    /* _NEW_TEXTURE */
 240    brw_populate_sampler_prog_key_data(&brw->ctx, prog, &key->tex);
 241 }
 242
 243 void
 244 brw_upload_tes_prog(struct brw_context *brw)
 245 {
 246    struct brw_stage_state *stage_state = &brw->tes.base;
 247    struct brw_tes_prog_key key;
 248    /* BRW_NEW_TESS_PROGRAMS */
 249    struct brw_program *tep = (struct brw_program *) brw->tess_eval_program;
 250
 251    if (!brw_state_dirty(brw,
 252                         _NEW_TEXTURE,
 253                         BRW_NEW_TESS_PROGRAMS))
 254       return;
 255
 256    brw_tes_populate_key(brw, &key);
 257
 258    if (!brw_search_cache(&brw->cache, BRW_CACHE_TES_PROG,
 259                          &key, sizeof(key),
 260                          &stage_state->prog_offset,
 261                          &brw->tes.base.prog_data)) {
 262       bool success = brw_codegen_tes_prog(brw, tep, &key);
 263       assert(success);
 264       (void)success;
 265    }
 266 }
 267
 268
 269 bool
 270 brw_tes_precompile(struct gl_context *ctx,
 271                    struct gl_shader_program *shader_prog,
 272                    struct gl_program *prog)
 273 {
 274    struct brw_context *brw = brw_context(ctx);
 275    struct brw_tes_prog_key key;
 276    uint32_t old_prog_offset = brw->tes.base.prog_offset;
 277    struct brw_stage_prog_data *old_prog_data = brw->tes.base.prog_data;
 278    bool success;
 279
 280    struct brw_program *btep = brw_program(prog);
 281
 282    memset(&key, 0, sizeof(key));
 283
 284    key.program_string_id = btep->id;
 285    key.inputs_read = prog->nir->info->inputs_read;
 286    key.patch_inputs_read = prog->nir->info->patch_inputs_read;
 287
 288    if (shader_prog->_LinkedShaders[MESA_SHADER_TESS_CTRL]) {
 289       struct gl_program *tcp =
 290          shader_prog->_LinkedShaders[MESA_SHADER_TESS_CTRL]->Program;
 291       key.inputs_read |= tcp->nir->info->outputs_written &
 292          ~(VARYING_BIT_TESS_LEVEL_INNER | VARYING_BIT_TESS_LEVEL_OUTER);
 293       key.patch_inputs_read |= tcp->nir->info->patch_outputs_written;
 294    }
 295
 296    brw_setup_tex_for_precompile(brw, &key.tex, prog);
 297
 298    success = brw_codegen_tes_prog(brw, btep, &key);
 299
 300    brw->tes.base.prog_offset = old_prog_offset;
 301    brw->tes.base.prog_data = old_prog_data;
 302
 303    return success;
 304 }