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 "compiler/brw_nir.h"
  32 #include "brw_program.h"
  33 #include "brw_state.h"
  34 #include "program/prog_parameter.h"
  35
  36 static void
  37 brw_tes_debug_recompile(struct brw_context *brw, struct gl_program *prog,
  38                        const struct brw_tes_prog_key *key)
  39 {
  40    perf_debug("Recompiling tessellation evaluation shader for program %d\n",
  41               prog->Id);
  42
  43    bool found = false;
  44    const struct brw_tes_prog_key *old_key =
  45       brw_find_previous_compile(&brw->cache, BRW_CACHE_TES_PROG,
  46                                 key->program_string_id);
  47
  48    if (!old_key) {
  49       perf_debug("  Didn't find previous compile in the shader cache for "
  50                  "debug\n");
  51       return;
  52    }
  53
  54    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
  55    found |= key_debug(brw, "inputs read", old_key->inputs_read,
  56                       key->inputs_read);
  57    found |= key_debug(brw, "patch inputs read", old_key->patch_inputs_read,
  58                       key->patch_inputs_read);
  59
  60    if (!found) {
  61       perf_debug("  Something else\n");
  62    }
  63 }
  64
  65 static bool
  66 brw_codegen_tes_prog(struct brw_context *brw,
  67                      struct brw_program *tep,
  68                      struct brw_tes_prog_key *key)
  69 {
  70    const struct brw_compiler *compiler = brw->screen->compiler;
  71    const struct gen_device_info *devinfo = &brw->screen->devinfo;
  72    struct brw_stage_state *stage_state = &brw->tes.base;
  73    nir_shader *nir = tep->program.nir;
  74    struct brw_tes_prog_data prog_data;
  75    bool start_busy = false;
  76    double start_time = 0;
  77
  78    memset(&prog_data, 0, sizeof(prog_data));
  79
  80    brw_assign_common_binding_table_offsets(devinfo, &tep->program,
  81                                            &prog_data.base.base, 0);
  82
  83    /* Allocate the references to the uniforms that will end up in the
  84     * prog_data associated with the compiled program, and which will be freed
  85     * by the state cache.
  86     *
  87     * Note: param_count needs to be num_uniform_components * 4, since we add
  88     * padding around uniform values below vec4 size, so the worst case is that
  89     * every uniform is a float which gets padded to the size of a vec4.
  90     */
  91    int param_count = nir->num_uniforms / 4;
  92
  93    prog_data.base.base.param =
  94       rzalloc_array(NULL, const gl_constant_value *, param_count);
  95    prog_data.base.base.pull_param =
  96       rzalloc_array(NULL, const gl_constant_value *, param_count);
  97    prog_data.base.base.image_param =
  98       rzalloc_array(NULL, struct brw_image_param,
  99                     tep->program.info.num_images);
 100    prog_data.base.base.nr_params = param_count;
 101    prog_data.base.base.nr_image_params = tep->program.info.num_images;
 102
 103    brw_nir_setup_glsl_uniforms(nir, &tep->program, &prog_data.base.base,
 104                                compiler->scalar_stage[MESA_SHADER_TESS_EVAL]);
 105
 106    int st_index = -1;
 107    if (unlikely(INTEL_DEBUG & DEBUG_SHADER_TIME))
 108       st_index = brw_get_shader_time_index(brw, &tep->program, ST_TES, true);
 109
 110    if (unlikely(brw->perf_debug)) {
 111       start_busy = brw->batch.last_bo && brw_bo_busy(brw->batch.last_bo);
 112       start_time = get_time();
 113    }
 114
 115    struct brw_vue_map input_vue_map;
 116    brw_compute_tess_vue_map(&input_vue_map, key->inputs_read,
 117                             key->patch_inputs_read);
 118
 119    void *mem_ctx = ralloc_context(NULL);
 120    unsigned program_size;
 121    char *error_str;
 122    const unsigned *program =
 123       brw_compile_tes(compiler, brw, mem_ctx, key, &input_vue_map, &prog_data,
 124                       nir, &tep->program, st_index, &program_size, &error_str);
 125    if (program == NULL) {
 126       tep->program.sh.data->LinkStatus = linking_failure;
 127       ralloc_strcat(&tep->program.sh.data->InfoLog, error_str);
 128
 129       _mesa_problem(NULL, "Failed to compile tessellation evaluation shader: "
 130                     "%s\n", error_str);
 131
 132       ralloc_free(mem_ctx);
 133       return false;
 134    }
 135
 136    if (unlikely(brw->perf_debug)) {
 137       if (tep->compiled_once) {
 138          brw_tes_debug_recompile(brw, &tep->program, key);
 139       }
 140       if (start_busy && !brw_bo_busy(brw->batch.last_bo)) {
 141          perf_debug("TES compile took %.03f ms and stalled the GPU\n",
 142                     (get_time() - start_time) * 1000);
 143       }
 144       tep->compiled_once = true;
 145    }
 146
 147    /* Scratch space is used for register spilling */
 148    brw_alloc_stage_scratch(brw, stage_state,
 149                            prog_data.base.base.total_scratch,
 150                            devinfo->max_tes_threads);
 151
 152    brw_upload_cache(&brw->cache, BRW_CACHE_TES_PROG,
 153                     key, sizeof(*key),
 154                     program, program_size,
 155                     &prog_data, sizeof(prog_data),
 156                     &stage_state->prog_offset, &brw->tes.base.prog_data);
 157    ralloc_free(mem_ctx);
 158
 159    return true;
 160 }
 161
 162 void
 163 brw_tes_populate_key(struct brw_context *brw,
 164                      struct brw_tes_prog_key *key)
 165 {
 166    struct brw_program *tcp = (struct brw_program *) brw->tess_ctrl_program;
 167    struct brw_program *tep = (struct brw_program *) brw->tess_eval_program;
 168    struct gl_program *prog = &tep->program;
 169
 170    uint64_t per_vertex_slots = prog->info.inputs_read;
 171    uint32_t per_patch_slots = prog->info.patch_inputs_read;
 172
 173    memset(key, 0, sizeof(*key));
 174
 175    key->program_string_id = tep->id;
 176
 177    /* The TCS may have additional outputs which aren't read by the
 178     * TES (possibly for cross-thread communication).  These need to
 179     * be stored in the Patch URB Entry as well.
 180     */
 181    if (tcp) {
 182       struct gl_program *tcp_prog = &tcp->program;
 183       per_vertex_slots |= tcp_prog->info.outputs_written &
 184          ~(VARYING_BIT_TESS_LEVEL_INNER | VARYING_BIT_TESS_LEVEL_OUTER);
 185       per_patch_slots |= tcp_prog->info.patch_outputs_written;
 186    }
 187
 188    key->inputs_read = per_vertex_slots;
 189    key->patch_inputs_read = per_patch_slots;
 190
 191    /* _NEW_TEXTURE */
 192    brw_populate_sampler_prog_key_data(&brw->ctx, prog, &key->tex);
 193 }
 194
 195 void
 196 brw_upload_tes_prog(struct brw_context *brw)
 197 {
 198    struct brw_stage_state *stage_state = &brw->tes.base;
 199    struct brw_tes_prog_key key;
 200    /* BRW_NEW_TESS_PROGRAMS */
 201    struct brw_program *tep = (struct brw_program *) brw->tess_eval_program;
 202
 203    if (!brw_state_dirty(brw,
 204                         _NEW_TEXTURE,
 205                         BRW_NEW_TESS_PROGRAMS))
 206       return;
 207
 208    brw_tes_populate_key(brw, &key);
 209
 210    if (!brw_search_cache(&brw->cache, BRW_CACHE_TES_PROG,
 211                          &key, sizeof(key),
 212                          &stage_state->prog_offset,
 213                          &brw->tes.base.prog_data)) {
 214       bool success = brw_codegen_tes_prog(brw, tep, &key);
 215       assert(success);
 216       (void)success;
 217    }
 218 }
 219
 220
 221 bool
 222 brw_tes_precompile(struct gl_context *ctx,
 223                    struct gl_shader_program *shader_prog,
 224                    struct gl_program *prog)
 225 {
 226    struct brw_context *brw = brw_context(ctx);
 227    struct brw_tes_prog_key key;
 228    uint32_t old_prog_offset = brw->tes.base.prog_offset;
 229    struct brw_stage_prog_data *old_prog_data = brw->tes.base.prog_data;
 230    bool success;
 231
 232    struct brw_program *btep = brw_program(prog);
 233
 234    memset(&key, 0, sizeof(key));
 235
 236    key.program_string_id = btep->id;
 237    key.inputs_read = prog->nir->info->inputs_read;
 238    key.patch_inputs_read = prog->nir->info->patch_inputs_read;
 239
 240    if (shader_prog->_LinkedShaders[MESA_SHADER_TESS_CTRL]) {
 241       struct gl_program *tcp =
 242          shader_prog->_LinkedShaders[MESA_SHADER_TESS_CTRL]->Program;
 243       key.inputs_read |= tcp->nir->info->outputs_written &
 244          ~(VARYING_BIT_TESS_LEVEL_INNER | VARYING_BIT_TESS_LEVEL_OUTER);
 245       key.patch_inputs_read |= tcp->nir->info->patch_outputs_written;
 246    }
 247
 248    brw_setup_tex_for_precompile(brw, &key.tex, prog);
 249
 250    success = brw_codegen_tes_prog(brw, btep, &key);
 251
 252    brw->tes.base.prog_offset = old_prog_offset;
 253    brw->tes.base.prog_data = old_prog_data;
 254
 255    return success;
 256 }