src/mesa/drivers/dri/i965/brw_tcs.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_tcs.c
  26  *
  27  * Tessellation control 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_tcs_debug_recompile(struct brw_context *brw,
  39                        struct gl_shader_program *shader_prog,
  40                        const struct brw_tcs_prog_key *key)
  41 {
  42    struct brw_cache_item *c = NULL;
  43    const struct brw_tcs_prog_key *old_key = NULL;
  44    bool found = false;
  45
  46    perf_debug("Recompiling tessellation control 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_TCS_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 |= key_debug(brw, "input vertices", old_key->input_vertices,
  69                       key->input_vertices);
  70    found |= key_debug(brw, "TES primitive mode", old_key->tes_primitive_mode,
  71                       key->tes_primitive_mode);
  72    found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex);
  73
  74    if (!found) {
  75       perf_debug("  Something else\n");
  76    }
  77 }
  78
  79 static bool
  80 brw_codegen_tcs_prog(struct brw_context *brw,
  81                      struct gl_shader_program *shader_prog,
  82                      struct brw_tess_ctrl_program *tcp,
  83                      struct brw_tcs_prog_key *key)
  84 {
  85    const struct brw_compiler *compiler = brw->intelScreen->compiler;
  86    struct brw_stage_state *stage_state = &brw->tcs.base;
  87    nir_shader *nir = tcp->program.Base.nir;
  88    struct brw_tcs_prog_data prog_data;
  89    bool start_busy = false;
  90    double start_time = 0;
  91
  92    memset(&prog_data, 0, sizeof(prog_data));
  93
  94    /* Allocate the references to the uniforms that will end up in the
  95     * prog_data associated with the compiled program, and which will be freed
  96     * by the state cache.
  97     *
  98     * Note: param_count needs to be num_uniform_components * 4, since we add
  99     * padding around uniform values below vec4 size, so the worst case is that
 100     * every uniform is a float which gets padded to the size of a vec4.
 101     */
 102    struct gl_shader *tcs = shader_prog->_LinkedShaders[MESA_SHADER_TESS_CTRL];
 103    int param_count = nir->num_uniforms;
 104    if (!compiler->scalar_stage[MESA_SHADER_TESS_CTRL])
 105       param_count *= 4;
 106
 107    prog_data.base.base.param =
 108       rzalloc_array(NULL, const gl_constant_value *, param_count);
 109    prog_data.base.base.pull_param =
 110       rzalloc_array(NULL, const gl_constant_value *, param_count);
 111    prog_data.base.base.image_param =
 112       rzalloc_array(NULL, struct brw_image_param, tcs->NumImages);
 113    prog_data.base.base.nr_params = param_count;
 114    prog_data.base.base.nr_image_params = tcs->NumImages;
 115
 116    brw_nir_setup_glsl_uniforms(nir, shader_prog, &tcp->program.Base,
 117                                &prog_data.base.base, false);
 118
 119    if (unlikely(INTEL_DEBUG & DEBUG_TCS))
 120       brw_dump_ir("tessellation control", shader_prog, tcs, NULL);
 121
 122    int st_index = -1;
 123    if (unlikely(INTEL_DEBUG & DEBUG_SHADER_TIME))
 124       st_index = brw_get_shader_time_index(brw, shader_prog, NULL, ST_TCS);
 125
 126    if (unlikely(brw->perf_debug)) {
 127       start_busy = brw->batch.last_bo && drm_intel_bo_busy(brw->batch.last_bo);
 128       start_time = get_time();
 129    }
 130
 131    void *mem_ctx = ralloc_context(NULL);
 132    unsigned program_size;
 133    char *error_str;
 134    const unsigned *program =
 135       brw_compile_tcs(compiler, brw, mem_ctx, key, &prog_data, nir, st_index,
 136                       &program_size, &error_str);
 137    if (program == NULL) {
 138       if (shader_prog) {
 139          shader_prog->LinkStatus = false;
 140          ralloc_strcat(&shader_prog->InfoLog, error_str);
 141       }
 142
 143       _mesa_problem(NULL, "Failed to compile tessellation control shader: "
 144                     "%s\n", error_str);
 145
 146       ralloc_free(mem_ctx);
 147       return false;
 148    }
 149
 150    if (unlikely(brw->perf_debug)) {
 151       struct brw_shader *btcs = (struct brw_shader *) tcs;
 152       if (btcs->compiled_once) {
 153          brw_tcs_debug_recompile(brw, shader_prog, key);
 154       }
 155       if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) {
 156          perf_debug("TCS compile took %.03f ms and stalled the GPU\n",
 157                     (get_time() - start_time) * 1000);
 158       }
 159       btcs->compiled_once = true;
 160    }
 161
 162    /* Scratch space is used for register spilling */
 163    if (prog_data.base.base.total_scratch) {
 164       brw_get_scratch_bo(brw, &stage_state->scratch_bo,
 165                          prog_data.base.base.total_scratch *
 166                          brw->max_hs_threads);
 167    }
 168
 169    brw_upload_cache(&brw->cache, BRW_CACHE_TCS_PROG,
 170                     key, sizeof(*key),
 171                     program, program_size,
 172                     &prog_data, sizeof(prog_data),
 173                     &stage_state->prog_offset, &brw->tcs.prog_data);
 174    ralloc_free(mem_ctx);
 175
 176    return true;
 177 }
 178
 179
 180 void
 181 brw_upload_tcs_prog(struct brw_context *brw)
 182 {
 183    struct gl_context *ctx = &brw->ctx;
 184    struct gl_shader_program **current = ctx->_Shader->CurrentProgram;
 185    struct brw_stage_state *stage_state = &brw->tcs.base;
 186    struct brw_tcs_prog_key key;
 187    /* BRW_NEW_TESS_PROGRAMS */
 188    struct brw_tess_ctrl_program *tcp =
 189       (struct brw_tess_ctrl_program *) brw->tess_ctrl_program;
 190    struct brw_tess_eval_program *tep =
 191       (struct brw_tess_eval_program *) brw->tess_eval_program;
 192    assert(tep);
 193
 194    if (!brw_state_dirty(brw,
 195                         _NEW_TEXTURE,
 196                         BRW_NEW_PATCH_PRIMITIVE |
 197                         BRW_NEW_TESS_PROGRAMS))
 198       return;
 199
 200    struct gl_program *prog = &tcp->program.Base;
 201
 202    memset(&key, 0, sizeof(key));
 203
 204    key.program_string_id = tcp->id;
 205
 206    key.input_vertices = ctx->TessCtrlProgram.patch_vertices;
 207
 208    /* _NEW_TEXTURE */
 209    brw_populate_sampler_prog_key_data(ctx, prog, stage_state->sampler_count,
 210                                       &key.tex);
 211
 212    /* We need to specialize our code generation for tessellation levels
 213     * based on the domain the DS is expecting to tessellate.
 214     */
 215    key.tes_primitive_mode = tep->program.PrimitiveMode;
 216
 217    if (!brw_search_cache(&brw->cache, BRW_CACHE_TCS_PROG,
 218                          &key, sizeof(key),
 219                          &stage_state->prog_offset, &brw->tcs.prog_data)) {
 220       bool success = brw_codegen_tcs_prog(brw, current[MESA_SHADER_TESS_CTRL],
 221                                           tcp, &key);
 222       assert(success);
 223       (void)success;
 224    }
 225    brw->tcs.base.prog_data = &brw->tcs.prog_data->base.base;
 226 }
 227
 228
 229 bool
 230 brw_tcs_precompile(struct gl_context *ctx,
 231                    struct gl_shader_program *shader_prog,
 232                    struct gl_program *prog)
 233 {
 234    struct brw_context *brw = brw_context(ctx);
 235    struct brw_tcs_prog_key key;
 236    uint32_t old_prog_offset = brw->tcs.base.prog_offset;
 237    struct brw_tcs_prog_data *old_prog_data = brw->tcs.prog_data;
 238    bool success;
 239
 240    struct gl_tess_ctrl_program *tcp = (struct gl_tess_ctrl_program *)prog;
 241    struct brw_tess_ctrl_program *btcp = brw_tess_ctrl_program(tcp);
 242
 243    memset(&key, 0, sizeof(key));
 244
 245    key.program_string_id = btcp->id;
 246    brw_setup_tex_for_precompile(brw, &key.tex, prog);
 247
 248    /* Guess that the input and output patches have the same dimensionality. */
 249    key.input_vertices = shader_prog->TessCtrl.VerticesOut;
 250
 251    key.tes_primitive_mode = GL_TRIANGLES;
 252
 253    success = brw_codegen_tcs_prog(brw, shader_prog, btcp, &key);
 254
 255    brw->tcs.base.prog_offset = old_prog_offset;
 256    brw->tcs.prog_data = old_prog_data;
 257
 258    return success;
 259 }