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@vmware.com> 30 */ 31 32 33#include "util/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 "compiler/brw_nir.h" 42#include "brw_program.h" 43 44#include "util/ralloc.h" 45 46/** 47 * Decide which set of clip planes should be used when clipping via 48 * gl_Position or gl_ClipVertex. 49 */ 50gl_clip_plane * 51brw_select_clip_planes(struct gl_context *ctx) 52{ 53 if (ctx->_Shader->CurrentProgram[MESA_SHADER_VERTEX]) { 54 /* There is currently a GLSL vertex shader, so clip according to GLSL 55 * rules, which means compare gl_ClipVertex (or gl_Position, if 56 * gl_ClipVertex wasn't assigned) against the eye-coordinate clip planes 57 * that were stored in EyeUserPlane at the time the clip planes were 58 * specified. 59 */ 60 return ctx->Transform.EyeUserPlane; 61 } else { 62 /* Either we are using fixed function or an ARB vertex program. In 63 * either case the clip planes are going to be compared against 64 * gl_Position (which is in clip coordinates) so we have to clip using 65 * _ClipUserPlane, which was transformed into clip coordinates by Mesa 66 * core. 67 */ 68 return ctx->Transform._ClipUserPlane; 69 } 70} 71 72static GLbitfield64 73brw_vs_outputs_written(struct brw_context *brw, struct brw_vs_prog_key *key, 74 GLbitfield64 user_varyings) 75{ 76 const struct intel_device_info *devinfo = &brw->screen->devinfo; 77 GLbitfield64 outputs_written = user_varyings; 78 79 if (devinfo->ver < 6) { 80 /* Put dummy slots into the VUE for the SF to put the replaced 81 * point sprite coords in. We shouldn't need these dummy slots, 82 * which take up precious URB space, but it would mean that the SF 83 * doesn't get nice aligned pairs of input coords into output 84 * coords, which would be a pain to handle. 85 */ 86 for (unsigned i = 0; i < 8; i++) { 87 if (key->point_coord_replace & (1 << i)) 88 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_TEX0 + i); 89 } 90 91 /* if back colors are written, allocate slots for front colors too */ 92 if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC0)) 93 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL0); 94 if (outputs_written & BITFIELD64_BIT(VARYING_SLOT_BFC1)) 95 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_COL1); 96 } 97 98 /* In order for legacy clipping to work, we need to populate the clip 99 * distance varying slots whenever clipping is enabled, even if the vertex 100 * shader doesn't write to gl_ClipDistance. 101 */ 102 if (key->nr_userclip_plane_consts > 0) { 103 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST0); 104 outputs_written |= BITFIELD64_BIT(VARYING_SLOT_CLIP_DIST1); 105 } 106 107 return outputs_written; 108} 109 110static bool 111brw_codegen_vs_prog(struct brw_context *brw, 112 struct brw_program *vp, 113 struct brw_vs_prog_key *key) 114{ 115 const struct brw_compiler *compiler = brw->screen->compiler; 116 const struct intel_device_info *devinfo = &brw->screen->devinfo; 117 const GLuint *program; 118 struct brw_vs_prog_data prog_data; 119 struct brw_stage_prog_data *stage_prog_data = &prog_data.base.base; 120 void *mem_ctx; 121 bool start_busy = false; 122 double start_time = 0; 123 124 memset(&prog_data, 0, sizeof(prog_data)); 125 126 /* Use ALT floating point mode for ARB programs so that 0^0 == 1. */ 127 if (vp->program.info.is_arb_asm) 128 stage_prog_data->use_alt_mode = true; 129 130 mem_ctx = ralloc_context(NULL); 131 132 nir_shader *nir = nir_shader_clone(mem_ctx, vp->program.nir); 133 134 brw_assign_common_binding_table_offsets(devinfo, &vp->program, 135 &prog_data.base.base, 0); 136 137 if (!vp->program.info.is_arb_asm) { 138 brw_nir_setup_glsl_uniforms(mem_ctx, nir, &vp->program, 139 &prog_data.base.base, 140 compiler->scalar_stage[MESA_SHADER_VERTEX]); 141 if (brw->can_push_ubos) { 142 brw_nir_analyze_ubo_ranges(compiler, nir, key, 143 prog_data.base.base.ubo_ranges); 144 } 145 } else { 146 brw_nir_setup_arb_uniforms(mem_ctx, nir, &vp->program, 147 &prog_data.base.base); 148 } 149 150 if (key->nr_userclip_plane_consts > 0) { 151 brw_nir_lower_legacy_clipping(nir, key->nr_userclip_plane_consts, 152 &prog_data.base.base); 153 } 154 155 if (key->copy_edgeflag) 156 nir_lower_passthrough_edgeflags(nir); 157 158 uint64_t outputs_written = 159 brw_vs_outputs_written(brw, key, nir->info.outputs_written); 160 161 brw_compute_vue_map(devinfo, 162 &prog_data.base.vue_map, outputs_written, 163 nir->info.separate_shader, 1); 164 165 if (0) { 166 _mesa_fprint_program_opt(stderr, &vp->program, PROG_PRINT_DEBUG, true); 167 } 168 169 if (unlikely(brw->perf_debug)) { 170 start_busy = (brw->batch.last_bo && 171 brw_bo_busy(brw->batch.last_bo)); 172 start_time = get_time(); 173 } 174 175 if (INTEL_DEBUG(DEBUG_VS)) { 176 if (vp->program.info.is_arb_asm) 177 brw_dump_arb_asm("vertex", &vp->program); 178 } 179 180 181 /* Emit GFX4 code. 182 */ 183 struct brw_compile_vs_params params = { 184 .nir = nir, 185 .key = key, 186 .prog_data = &prog_data, 187 .log_data = brw, 188 }; 189 190 if (INTEL_DEBUG(DEBUG_SHADER_TIME)) { 191 params.shader_time = true; 192 params.shader_time_index = 193 brw_get_shader_time_index(brw, &vp->program, ST_VS, 194 !vp->program.info.is_arb_asm); 195 } 196 197 program = brw_compile_vs(compiler, mem_ctx, ¶ms); 198 if (program == NULL) { 199 if (!vp->program.info.is_arb_asm) { 200 vp->program.sh.data->LinkStatus = LINKING_FAILURE; 201 ralloc_strcat(&vp->program.sh.data->InfoLog, params.error_str); 202 } 203 204 _mesa_problem(NULL, "Failed to compile vertex shader: %s\n", params.error_str); 205 206 ralloc_free(mem_ctx); 207 return false; 208 } 209 210 if (unlikely(brw->perf_debug)) { 211 if (vp->compiled_once) { 212 brw_debug_recompile(brw, MESA_SHADER_VERTEX, vp->program.Id, 213 &key->base); 214 } 215 if (start_busy && !brw_bo_busy(brw->batch.last_bo)) { 216 perf_debug("VS compile took %.03f ms and stalled the GPU\n", 217 (get_time() - start_time) * 1000); 218 } 219 vp->compiled_once = true; 220 } 221 222 /* Scratch space is used for register spilling */ 223 brw_alloc_stage_scratch(brw, &brw->vs.base, 224 prog_data.base.base.total_scratch); 225 226 /* The param and pull_param arrays will be freed by the shader cache. */ 227 ralloc_steal(NULL, prog_data.base.base.param); 228 ralloc_steal(NULL, prog_data.base.base.pull_param); 229 brw_upload_cache(&brw->cache, BRW_CACHE_VS_PROG, 230 key, sizeof(struct brw_vs_prog_key), 231 program, prog_data.base.base.program_size, 232 &prog_data, sizeof(prog_data), 233 &brw->vs.base.prog_offset, &brw->vs.base.prog_data); 234 ralloc_free(mem_ctx); 235 236 return true; 237} 238 239static bool 240brw_vs_state_dirty(const struct brw_context *brw) 241{ 242 return brw_state_dirty(brw, 243 _NEW_BUFFERS | 244 _NEW_LIGHT | 245 _NEW_POINT | 246 _NEW_POLYGON | 247 _NEW_TEXTURE | 248 _NEW_TRANSFORM, 249 BRW_NEW_VERTEX_PROGRAM | 250 BRW_NEW_VS_ATTRIB_WORKAROUNDS); 251} 252 253void 254brw_vs_populate_key(struct brw_context *brw, 255 struct brw_vs_prog_key *key) 256{ 257 struct gl_context *ctx = &brw->ctx; 258 /* BRW_NEW_VERTEX_PROGRAM */ 259 struct gl_program *prog = brw->programs[MESA_SHADER_VERTEX]; 260 struct brw_program *vp = (struct brw_program *) prog; 261 const struct intel_device_info *devinfo = &brw->screen->devinfo; 262 263 memset(key, 0, sizeof(*key)); 264 265 /* Just upload the program verbatim for now. Always send it all 266 * the inputs it asks for, whether they are varying or not. 267 */ 268 269 /* _NEW_TEXTURE */ 270 brw_populate_base_prog_key(ctx, vp, &key->base); 271 272 if (ctx->Transform.ClipPlanesEnabled != 0 && 273 (ctx->API == API_OPENGL_COMPAT || ctx->API == API_OPENGLES) && 274 vp->program.info.clip_distance_array_size == 0) { 275 key->nr_userclip_plane_consts = 276 util_logbase2(ctx->Transform.ClipPlanesEnabled) + 1; 277 } 278 279 if (devinfo->ver < 6) { 280 /* _NEW_POLYGON */ 281 key->copy_edgeflag = (ctx->Polygon.FrontMode != GL_FILL || 282 ctx->Polygon.BackMode != GL_FILL); 283 284 /* _NEW_POINT */ 285 if (ctx->Point.PointSprite) { 286 key->point_coord_replace = ctx->Point.CoordReplace & 0xff; 287 } 288 } 289 290 if (prog->info.outputs_written & 291 (VARYING_BIT_COL0 | VARYING_BIT_COL1 | VARYING_BIT_BFC0 | 292 VARYING_BIT_BFC1)) { 293 /* _NEW_LIGHT | _NEW_BUFFERS */ 294 key->clamp_vertex_color = ctx->Light._ClampVertexColor; 295 } 296 297 /* BRW_NEW_VS_ATTRIB_WORKAROUNDS */ 298 if (devinfo->verx10 <= 70) { 299 memcpy(key->gl_attrib_wa_flags, brw->vb.attrib_wa_flags, 300 sizeof(brw->vb.attrib_wa_flags)); 301 } 302} 303 304void 305brw_upload_vs_prog(struct brw_context *brw) 306{ 307 struct brw_vs_prog_key key; 308 /* BRW_NEW_VERTEX_PROGRAM */ 309 struct brw_program *vp = 310 (struct brw_program *) brw->programs[MESA_SHADER_VERTEX]; 311 312 if (!brw_vs_state_dirty(brw)) 313 return; 314 315 brw_vs_populate_key(brw, &key); 316 317 if (brw_search_cache(&brw->cache, BRW_CACHE_VS_PROG, &key, sizeof(key), 318 &brw->vs.base.prog_offset, &brw->vs.base.prog_data, 319 true)) 320 return; 321 322 if (brw_disk_cache_upload_program(brw, MESA_SHADER_VERTEX)) 323 return; 324 325 vp = (struct brw_program *) brw->programs[MESA_SHADER_VERTEX]; 326 vp->id = key.base.program_string_id; 327 328 ASSERTED bool success = brw_codegen_vs_prog(brw, vp, &key); 329 assert(success); 330} 331 332void 333brw_vs_populate_default_key(const struct brw_compiler *compiler, 334 struct brw_vs_prog_key *key, 335 struct gl_program *prog) 336{ 337 const struct intel_device_info *devinfo = compiler->devinfo; 338 struct brw_program *bvp = brw_program(prog); 339 340 memset(key, 0, sizeof(*key)); 341 342 brw_populate_default_base_prog_key(devinfo, bvp, &key->base); 343 344 key->clamp_vertex_color = 345 (prog->info.outputs_written & 346 (VARYING_BIT_COL0 | VARYING_BIT_COL1 | VARYING_BIT_BFC0 | 347 VARYING_BIT_BFC1)); 348} 349 350bool 351brw_vs_precompile(struct gl_context *ctx, struct gl_program *prog) 352{ 353 struct brw_context *brw = brw_context(ctx); 354 struct brw_vs_prog_key key; 355 uint32_t old_prog_offset = brw->vs.base.prog_offset; 356 struct brw_stage_prog_data *old_prog_data = brw->vs.base.prog_data; 357 bool success; 358 359 struct brw_program *bvp = brw_program(prog); 360 361 brw_vs_populate_default_key(brw->screen->compiler, &key, prog); 362 363 success = brw_codegen_vs_prog(brw, bvp, &key); 364 365 brw->vs.base.prog_offset = old_prog_offset; 366 brw->vs.base.prog_data = old_prog_data; 367 368 return success; 369} 370