1 /* 2 Copyright (C) Intel Corp. 2006. All Rights Reserved. 3 Intel funded Tungsten Graphics to 4 develop this 3D driver. 5 6 Permission is hereby granted, free of charge, to any person obtaining 7 a copy of this software and associated documentation files (the 8 "Software"), to deal in the Software without restriction, including 9 without limitation the rights to use, copy, modify, merge, publish, 10 distribute, sublicense, and/or sell copies of the Software, and to 11 permit persons to whom the Software is furnished to do so, subject to 12 the following conditions: 13 14 The above copyright notice and this permission notice (including the 15 next paragraph) shall be included in all copies or substantial 16 portions of the Software. 17 18 THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 19 EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 20 MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. 21 IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE 22 LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION 23 OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION 24 WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. 25 26 **********************************************************************/ 27 /* 28 * Authors: 29 * Keith Whitwell <keithw (at) vmware.com> 30 */ 31 32 33 #include "main/compiler.h" 34 #include "main/context.h" 35 #include "brw_context.h" 36 #include "brw_vs.h" 37 #include "brw_util.h" 38 #include "brw_state.h" 39 #include "program/prog_print.h" 40 #include "program/prog_parameter.h" 41 #include "brw_nir.h" 42 #include "brw_program.h" 43 44 #include "util/ralloc.h" 45 46 GLbitfield64 47 brw_vs_outputs_written(struct brw_context *brw, struct brw_vs_prog_key *key, 48 GLbitfield64 user_varyings) 49 { 50 GLbitfield64 outputs_written = user_varyings; 51 52 if (key->copy_edgeflag) { 53 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_EDGE); 54 } 55 56 if (brw->gen < 6) { 57 /* Put dummy slots into the VUE for the SF to put the replaced 58 * point sprite coords in. We shouldn't need these dummy slots, 59 * which take up precious URB space, but it would mean that the SF 60 * doesn't get nice aligned pairs of input coords into output 61 * coords, which would be a pain to handle. 62 */ 63 for (unsigned i = 0; i < 8; i++) { 64 if (key->point_coord_replace & (1 << i)) 65 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_TEX0 + i); 66 } 67 68 /* if back colors are written, allocate slots for front colors too */ 69 if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC0)) 70 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL0); 71 if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC1)) 72 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL1); 73 } 74 75 /* In order for legacy clipping to work, we need to populate the clip 76 * distance varying slots whenever clipping is enabled, even if the vertex 77 * shader doesn't write to gl_ClipDistance. 78 */ 79 if (key->nr_userclip_plane_consts > 0) { 80 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0); 81 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1); 82 } 83 84 return outputs_written; 85 } 86 87 static void 88 brw_vs_debug_recompile(struct brw_context *brw, struct gl_program *prog, 89 const struct brw_vs_prog_key *key) 90 { 91 perf_debug("Recompiling vertex shader for program %d\n", prog->Id); 92 93 bool found = false; 94 const struct brw_vs_prog_key *old_key = 95 brw_find_previous_compile(&brw->cache, BRW_CACHE_VS_PROG, 96 key->program_string_id); 97 98 if (!old_key) { 99 perf_debug(" Didn't find previous compile in the shader cache for " 100 "debug\n"); 101 return; 102 } 103 104 for (unsigned int i = 0; i < VERT_ATTRIB_MAX; i++) { 105 found |= key_debug(brw, "Vertex attrib w/a flags", 106 old_key->gl_attrib_wa_flags[i], 107 key->gl_attrib_wa_flags[i]); 108 } 109 110 found |= key_debug(brw, "legacy user clipping", 111 old_key->nr_userclip_plane_consts, 112 key->nr_userclip_plane_consts); 113 114 found |= key_debug(brw, "copy edgeflag", 115 old_key->copy_edgeflag, key->copy_edgeflag); 116 found |= key_debug(brw, "PointCoord replace", 117 old_key->point_coord_replace, key->point_coord_replace); 118 found |= key_debug(brw, "vertex color clamping", 119 old_key->clamp_vertex_color, key->clamp_vertex_color); 120 121 found |= brw_debug_recompile_sampler_key(brw, &old_key->tex, &key->tex); 122 123 if (!found) { 124 perf_debug(" Something else\n"); 125 } 126 } 127 128 static bool 129 brw_codegen_vs_prog(struct brw_context *brw, 130 struct brw_program *vp, 131 struct brw_vs_prog_key *key) 132 { 133 const struct brw_compiler *compiler = brw->screen->compiler; 134 const struct gen_device_info *devinfo = &brw->screen->devinfo; 135 GLuint program_size; 136 const GLuint *program; 137 struct brw_vs_prog_data prog_data; 138 struct brw_stage_prog_data *stage_prog_data = &prog_data.base.base; 139 void *mem_ctx; 140 bool start_busy = false; 141 double start_time = 0; 142 143 memset(&prog_data, 0, sizeof(prog_data)); 144 145 /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */ 146 if (vp->program.is_arb_asm) 147 stage_prog_data->use_alt_mode = true; 148 149 mem_ctx = ralloc_context(NULL); 150 151 brw_assign_common_binding_table_offsets(devinfo, &vp->program, 152 &prog_data.base.base, 0); 153 154 /* Allocate the references to the uniforms that will end up in the 155 * prog_data associated with the compiled program, and which will be freed 156 * by the state cache. 157 */ 158 int param_count = vp->program.nir->num_uniforms / 4; 159 160 prog_data.base.base.nr_image_params = vp->program.info.num_images; 161 162 /* vec4_visitor::setup_uniform_clipplane_values() also uploads user clip 163 * planes as uniforms. 164 */ 165 param_count += key->nr_userclip_plane_consts * 4; 166 167 stage_prog_data->param = 168 rzalloc_array(NULL, const gl_constant_value *, param_count); 169 stage_prog_data->pull_param = 170 rzalloc_array(NULL, const gl_constant_value *, param_count); 171 stage_prog_data->image_param = 172 rzalloc_array(NULL, struct brw_image_param, 173 stage_prog_data->nr_image_params); 174 stage_prog_data->nr_params = param_count; 175 176 if (!vp->program.is_arb_asm) { 177 brw_nir_setup_glsl_uniforms(vp->program.nir, &vp->program, 178 &prog_data.base.base, 179 compiler->scalar_stage[MESA_SHADER_VERTEX]); 180 } else { 181 brw_nir_setup_arb_uniforms(vp->program.nir, &vp->program, 182 &prog_data.base.base); 183 } 184 185 uint64_t outputs_written = 186 brw_vs_outputs_written(brw, key, vp->program.info.outputs_written); 187 prog_data.inputs_read = vp->program.info.inputs_read; 188 prog_data.double_inputs_read = vp->program.info.double_inputs_read; 189 190 if (key->copy_edgeflag) { 191 prog_data.inputs_read |= VERT_BIT_EDGEFLAG; 192 } 193 194 brw_compute_vue_map(devinfo, 195 &prog_data.base.vue_map, outputs_written, 196 vp->program.nir->info->separate_shader); 197 198 if (0) { 199 _mesa_fprint_program_opt(stderr, &vp->program, PROG_PRINT_DEBUG, true); 200 } 201 202 if (unlikely(brw->perf_debug)) { 203 start_busy = (brw->batch.last_bo && 204 drm_intel_bo_busy(brw->batch.last_bo)); 205 start_time = get_time(); 206 } 207 208 if (unlikely(INTEL_DEBUG & DEBUG_VS)) { 209 if (vp->program.is_arb_asm) 210 brw_dump_arb_asm("vertex", &vp->program); 211 } 212 213 int st_index = -1; 214 if (INTEL_DEBUG & DEBUG_SHADER_TIME) { 215 st_index = brw_get_shader_time_index(brw, &vp->program, ST_VS, 216 !vp->program.is_arb_asm); 217 } 218 219 /* Emit GEN4 code. 220 */ 221 char *error_str; 222 program = brw_compile_vs(compiler, brw, mem_ctx, key, &prog_data, 223 vp->program.nir, 224 brw_select_clip_planes(&brw->ctx), 225 !_mesa_is_gles3(&brw->ctx), 226 st_index, &program_size, &error_str); 227 if (program == NULL) { 228 if (!vp->program.is_arb_asm) { 229 vp->program.sh.data->LinkStatus = false; 230 ralloc_strcat(&vp->program.sh.data->InfoLog, error_str); 231 } 232 233 _mesa_problem(NULL, "Failed to compile vertex shader: %s\n", error_str); 234 235 ralloc_free(mem_ctx); 236 return false; 237 } 238 239 if (unlikely(brw->perf_debug)) { 240 if (vp->compiled_once) { 241 brw_vs_debug_recompile(brw, &vp->program, key); 242 } 243 if (start_busy && !drm_intel_bo_busy(brw->batch.last_bo)) { 244 perf_debug("VS compile took %.03f ms and stalled the GPU\n", 245 (get_time() - start_time) * 1000); 246 } 247 vp->compiled_once = true; 248 } 249 250 /* Scratch space is used for register spilling */ 251 brw_alloc_stage_scratch(brw, &brw->vs.base, 252 prog_data.base.base.total_scratch, 253 devinfo->max_vs_threads); 254 255 brw_upload_cache(&brw->cache, BRW_CACHE_VS_PROG, 256 key, sizeof(struct brw_vs_prog_key), 257 program, program_size, 258 &prog_data, sizeof(prog_data), 259 &brw->vs.base.prog_offset, &brw->vs.base.prog_data); 260 ralloc_free(mem_ctx); 261 262 return true; 263 } 264 265 static bool 266 brw_vs_state_dirty(const struct brw_context *brw) 267 { 268 return brw_state_dirty(brw, 269 _NEW_BUFFERS | 270 _NEW_LIGHT | 271 _NEW_POINT | 272 _NEW_POLYGON | 273 _NEW_TEXTURE | 274 _NEW_TRANSFORM, 275 BRW_NEW_VERTEX_PROGRAM | 276 BRW_NEW_VS_ATTRIB_WORKAROUNDS); 277 } 278 279 void 280 brw_vs_populate_key(struct brw_context *brw, 281 struct brw_vs_prog_key *key) 282 { 283 struct gl_context *ctx = &brw->ctx; 284 /* BRW_NEW_VERTEX_PROGRAM */ 285 struct brw_program *vp = (struct brw_program *)brw->vertex_program; 286 struct gl_program *prog = (struct gl_program *) brw->vertex_program; 287 288 memset(key, 0, sizeof(*key)); 289 290 /* Just upload the program verbatim for now. Always send it all 291 * the inputs it asks for, whether they are varying or not. 292 */ 293 key->program_string_id = vp->id; 294 295 if (ctx->Transform.ClipPlanesEnabled != 0 && 296 (ctx->API == API_OPENGL_COMPAT || ctx->API == API_OPENGLES) && 297 vp->program.ClipDistanceArraySize == 0) { 298 key->nr_userclip_plane_consts = 299 _mesa_logbase2(ctx->Transform.ClipPlanesEnabled) + 1; 300 } 301 302 if (brw->gen < 6) { 303 /* _NEW_POLYGON */ 304 key->copy_edgeflag = (ctx->Polygon.FrontMode != GL_FILL || 305 ctx->Polygon.BackMode != GL_FILL); 306 307 /* _NEW_POINT */ 308 if (ctx->Point.PointSprite) { 309 key->point_coord_replace = ctx->Point.CoordReplace & 0xff; 310 } 311 } 312 313 if (prog->nir->info->outputs_written & 314 (VARYING_BIT_COL0 | VARYING_BIT_COL1 | VARYING_BIT_BFC0 | 315 VARYING_BIT_BFC1)) { 316 /* _NEW_LIGHT | _NEW_BUFFERS */ 317 key->clamp_vertex_color = ctx->Light._ClampVertexColor; 318 } 319 320 /* _NEW_TEXTURE */ 321 brw_populate_sampler_prog_key_data(ctx, prog, &key->tex); 322 323 /* BRW_NEW_VS_ATTRIB_WORKAROUNDS */ 324 if (brw->gen < 8 && !brw->is_haswell) { 325 memcpy(key->gl_attrib_wa_flags, brw->vb.attrib_wa_flags, 326 sizeof(brw->vb.attrib_wa_flags)); 327 } 328 } 329 330 void 331 brw_upload_vs_prog(struct brw_context *brw) 332 { 333 struct brw_vs_prog_key key; 334 /* BRW_NEW_VERTEX_PROGRAM */ 335 struct brw_program *vp = (struct brw_program *)brw->vertex_program; 336 337 if (!brw_vs_state_dirty(brw)) 338 return; 339 340 brw_vs_populate_key(brw, &key); 341 342 if (!brw_search_cache(&brw->cache, BRW_CACHE_VS_PROG, 343 &key, sizeof(key), 344 &brw->vs.base.prog_offset, &brw->vs.base.prog_data)) { 345 bool success = brw_codegen_vs_prog(brw, vp, &key); 346 (void) success; 347 assert(success); 348 } 349 } 350 351 bool 352 brw_vs_precompile(struct gl_context *ctx, struct gl_program *prog) 353 { 354 struct brw_context *brw = brw_context(ctx); 355 struct brw_vs_prog_key key; 356 uint32_t old_prog_offset = brw->vs.base.prog_offset; 357 struct brw_stage_prog_data *old_prog_data = brw->vs.base.prog_data; 358 bool success; 359 360 struct brw_program *bvp = brw_program(prog); 361 362 memset(&key, 0, sizeof(key)); 363 364 brw_setup_tex_for_precompile(brw, &key.tex, prog); 365 key.program_string_id = bvp->id; 366 key.clamp_vertex_color = 367 (prog->nir->info->outputs_written & 368 (VARYING_BIT_COL0 | VARYING_BIT_COL1 | VARYING_BIT_BFC0 | 369 VARYING_BIT_BFC1)); 370 371 success = brw_codegen_vs_prog(brw, bvp, &key); 372 373 brw->vs.base.prog_offset = old_prog_offset; 374 brw->vs.base.prog_data = old_prog_data; 375 376 return success; 377 } 378