1 /* 2 * Copyright 2015 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 DEALINGS 21 * IN THE SOFTWARE. 22 */ 23 24 #include "brw_context.h" 25 #include "brw_state.h" 26 #include "brw_defines.h" 27 #include "brw_program.h" 28 #include "intel_batchbuffer.h" 29 #include "intel_buffer_objects.h" 30 #include "program/prog_parameter.h" 31 #include "main/shaderapi.h" 32 33 static uint32_t 34 f_as_u32(float f) 35 { 36 union fi fi = { .f = f }; 37 return fi.ui; 38 } 39 40 static uint32_t 41 brw_param_value(struct brw_context *brw, 42 const struct gl_program *prog, 43 const struct brw_stage_state *stage_state, 44 uint32_t param) 45 { 46 struct gl_context *ctx = &brw->ctx; 47 48 switch (BRW_PARAM_DOMAIN(param)) { 49 case BRW_PARAM_DOMAIN_BUILTIN: 50 if (param == BRW_PARAM_BUILTIN_ZERO) { 51 return 0; 52 } else if (BRW_PARAM_BUILTIN_IS_CLIP_PLANE(param)) { 53 gl_clip_plane *clip_planes = brw_select_clip_planes(ctx); 54 unsigned idx = BRW_PARAM_BUILTIN_CLIP_PLANE_IDX(param); 55 unsigned comp = BRW_PARAM_BUILTIN_CLIP_PLANE_COMP(param); 56 return ((uint32_t *)clip_planes[idx])[comp]; 57 } else if (param >= BRW_PARAM_BUILTIN_TESS_LEVEL_OUTER_X && 58 param <= BRW_PARAM_BUILTIN_TESS_LEVEL_OUTER_W) { 59 unsigned i = param - BRW_PARAM_BUILTIN_TESS_LEVEL_OUTER_X; 60 return f_as_u32(ctx->TessCtrlProgram.patch_default_outer_level[i]); 61 } else if (param == BRW_PARAM_BUILTIN_TESS_LEVEL_INNER_X) { 62 return f_as_u32(ctx->TessCtrlProgram.patch_default_inner_level[0]); 63 } else if (param == BRW_PARAM_BUILTIN_TESS_LEVEL_INNER_Y) { 64 return f_as_u32(ctx->TessCtrlProgram.patch_default_inner_level[1]); 65 } else { 66 unreachable("Invalid param builtin"); 67 } 68 69 case BRW_PARAM_DOMAIN_PARAMETER: { 70 unsigned idx = BRW_PARAM_PARAMETER_IDX(param); 71 unsigned comp = BRW_PARAM_PARAMETER_COMP(param); 72 assert(idx < prog->Parameters->NumParameters); 73 return prog->Parameters->ParameterValues[idx][comp].u; 74 } 75 76 case BRW_PARAM_DOMAIN_UNIFORM: { 77 unsigned idx = BRW_PARAM_UNIFORM_IDX(param); 78 assert(idx < prog->sh.data->NumUniformDataSlots); 79 return prog->sh.data->UniformDataSlots[idx].u; 80 } 81 82 case BRW_PARAM_DOMAIN_IMAGE: { 83 unsigned idx = BRW_PARAM_IMAGE_IDX(param); 84 unsigned offset = BRW_PARAM_IMAGE_OFFSET(param); 85 assert(offset < ARRAY_SIZE(stage_state->image_param)); 86 return ((uint32_t *)&stage_state->image_param[idx])[offset]; 87 } 88 89 default: 90 unreachable("Invalid param domain"); 91 } 92 } 93 94 95 void 96 brw_populate_constant_data(struct brw_context *brw, 97 const struct gl_program *prog, 98 const struct brw_stage_state *stage_state, 99 void *void_dst, 100 const uint32_t *param, 101 unsigned nr_params) 102 { 103 uint32_t *dst = void_dst; 104 for (unsigned i = 0; i < nr_params; i++) 105 dst[i] = brw_param_value(brw, prog, stage_state, param[i]); 106 } 107 108 109 /** 110 * Creates a streamed BO containing the push constants for the VS or GS on 111 * gen6+. 112 * 113 * Push constants are constant values (such as GLSL uniforms) that are 114 * pre-loaded into a shader stage's register space at thread spawn time. 115 * 116 * Not all GLSL uniforms will be uploaded as push constants: The hardware has 117 * a limitation of 32 or 64 EU registers (256 or 512 floats) per stage to be 118 * uploaded as push constants, while GL 4.4 requires at least 1024 components 119 * to be usable for the VS. Plus, currently we always use pull constants 120 * instead of push constants when doing variable-index array access. 121 * 122 * See brw_curbe.c for the equivalent gen4/5 code. 123 */ 124 void 125 gen6_upload_push_constants(struct brw_context *brw, 126 const struct gl_program *prog, 127 const struct brw_stage_prog_data *prog_data, 128 struct brw_stage_state *stage_state) 129 { 130 const struct gen_device_info *devinfo = &brw->screen->devinfo; 131 struct gl_context *ctx = &brw->ctx; 132 133 bool active = prog_data && 134 (stage_state->stage != MESA_SHADER_TESS_CTRL || 135 brw->programs[MESA_SHADER_TESS_EVAL]); 136 137 if (active) 138 _mesa_shader_write_subroutine_indices(ctx, stage_state->stage); 139 140 if (!active || prog_data->nr_params == 0) { 141 stage_state->push_const_size = 0; 142 } else { 143 /* Updates the ParamaterValues[i] pointers for all parameters of the 144 * basic type of PROGRAM_STATE_VAR. 145 */ 146 /* XXX: Should this happen somewhere before to get our state flag set? */ 147 if (prog) 148 _mesa_load_state_parameters(ctx, prog->Parameters); 149 150 int i; 151 const int size = prog_data->nr_params * sizeof(gl_constant_value); 152 gl_constant_value *param; 153 if (devinfo->gen >= 8 || devinfo->is_haswell) { 154 param = intel_upload_space(brw, size, 32, 155 &stage_state->push_const_bo, 156 &stage_state->push_const_offset); 157 } else { 158 param = brw_state_batch(brw, size, 32, 159 &stage_state->push_const_offset); 160 } 161 162 STATIC_ASSERT(sizeof(gl_constant_value) == sizeof(float)); 163 164 /* _NEW_PROGRAM_CONSTANTS 165 * 166 * Also _NEW_TRANSFORM -- we may reference clip planes other than as a 167 * side effect of dereferencing uniforms, so _NEW_PROGRAM_CONSTANTS 168 * wouldn't be set for them. 169 */ 170 brw_populate_constant_data(brw, prog, stage_state, param, 171 prog_data->param, 172 prog_data->nr_params); 173 174 if (0) { 175 fprintf(stderr, "%s constants:\n", 176 _mesa_shader_stage_to_string(stage_state->stage)); 177 for (i = 0; i < prog_data->nr_params; i++) { 178 if ((i & 7) == 0) 179 fprintf(stderr, "g%d: ", 180 prog_data->dispatch_grf_start_reg + i / 8); 181 fprintf(stderr, "%8f ", param[i].f); 182 if ((i & 7) == 7) 183 fprintf(stderr, "\n"); 184 } 185 if ((i & 7) != 0) 186 fprintf(stderr, "\n"); 187 fprintf(stderr, "\n"); 188 } 189 190 stage_state->push_const_size = ALIGN(prog_data->nr_params, 8) / 8; 191 /* We can only push 32 registers of constants at a time. */ 192 193 /* From the SNB PRM (vol2, part 1, section 3.2.1.4: 3DSTATE_CONSTANT_VS: 194 * 195 * "The sum of all four read length fields (each incremented to 196 * represent the actual read length) must be less than or equal to 197 * 32" 198 * 199 * From the IVB PRM (vol2, part 1, section 3.2.1.3: 3DSTATE_CONSTANT_VS: 200 * 201 * "The sum of all four read length fields must be less than or 202 * equal to the size of 64" 203 * 204 * The other shader stages all match the VS's limits. 205 */ 206 assert(stage_state->push_const_size <= 32); 207 } 208 209 stage_state->push_constants_dirty = true; 210 } 211 212 213 /** 214 * Creates a temporary BO containing the pull constant data for the shader 215 * stage, and the SURFACE_STATE struct that points at it. 216 * 217 * Pull constants are GLSL uniforms (and other constant data) beyond what we 218 * could fit as push constants, or that have variable-index array access 219 * (which is easiest to support using pull constants, and avoids filling 220 * register space with mostly-unused data). 221 * 222 * Compare this path to brw_curbe.c for gen4/5 push constants, and 223 * gen6_vs_state.c for gen6+ push constants. 224 */ 225 void 226 brw_upload_pull_constants(struct brw_context *brw, 227 GLbitfield64 brw_new_constbuf, 228 const struct gl_program *prog, 229 struct brw_stage_state *stage_state, 230 const struct brw_stage_prog_data *prog_data) 231 { 232 unsigned i; 233 uint32_t surf_index = prog_data->binding_table.pull_constants_start; 234 235 if (!prog_data->nr_pull_params) { 236 if (stage_state->surf_offset[surf_index]) { 237 stage_state->surf_offset[surf_index] = 0; 238 brw->ctx.NewDriverState |= brw_new_constbuf; 239 } 240 return; 241 } 242 243 /* Updates the ParamaterValues[i] pointers for all parameters of the 244 * basic type of PROGRAM_STATE_VAR. 245 */ 246 _mesa_load_state_parameters(&brw->ctx, prog->Parameters); 247 248 /* BRW_NEW_*_PROG_DATA | _NEW_PROGRAM_CONSTANTS */ 249 uint32_t size = prog_data->nr_pull_params * 4; 250 struct brw_bo *const_bo = NULL; 251 uint32_t const_offset; 252 gl_constant_value *constants = intel_upload_space(brw, size, 64, 253 &const_bo, &const_offset); 254 255 STATIC_ASSERT(sizeof(gl_constant_value) == sizeof(float)); 256 257 brw_populate_constant_data(brw, prog, stage_state, constants, 258 prog_data->pull_param, 259 prog_data->nr_pull_params); 260 261 if (0) { 262 for (i = 0; i < ALIGN(prog_data->nr_pull_params, 4) / 4; i++) { 263 const gl_constant_value *row = &constants[i * 4]; 264 fprintf(stderr, "const surface %3d: %4.3f %4.3f %4.3f %4.3f\n", 265 i, row[0].f, row[1].f, row[2].f, row[3].f); 266 } 267 } 268 269 brw_emit_buffer_surface_state(brw, &stage_state->surf_offset[surf_index], 270 const_bo, const_offset, 271 ISL_FORMAT_R32G32B32A32_FLOAT, 272 size, 1, 0); 273 274 brw_bo_unreference(const_bo); 275 276 brw->ctx.NewDriverState |= brw_new_constbuf; 277 } 278 279 /** 280 * Creates a region containing the push constants for the CS on gen7+. 281 * 282 * Push constants are constant values (such as GLSL uniforms) that are 283 * pre-loaded into a shader stage's register space at thread spawn time. 284 * 285 * For other stages, see brw_curbe.c:brw_upload_constant_buffer for the 286 * equivalent gen4/5 code and gen6_vs_state.c:gen6_upload_push_constants for 287 * gen6+. 288 */ 289 void 290 brw_upload_cs_push_constants(struct brw_context *brw, 291 const struct gl_program *prog, 292 const struct brw_cs_prog_data *cs_prog_data, 293 struct brw_stage_state *stage_state) 294 { 295 struct gl_context *ctx = &brw->ctx; 296 const struct brw_stage_prog_data *prog_data = 297 (struct brw_stage_prog_data*) cs_prog_data; 298 299 /* Updates the ParamaterValues[i] pointers for all parameters of the 300 * basic type of PROGRAM_STATE_VAR. 301 */ 302 /* XXX: Should this happen somewhere before to get our state flag set? */ 303 _mesa_load_state_parameters(ctx, prog->Parameters); 304 305 if (cs_prog_data->push.total.size == 0) { 306 stage_state->push_const_size = 0; 307 return; 308 } 309 310 311 uint32_t *param = 312 brw_state_batch(brw, ALIGN(cs_prog_data->push.total.size, 64), 313 64, &stage_state->push_const_offset); 314 assert(param); 315 316 STATIC_ASSERT(sizeof(gl_constant_value) == sizeof(float)); 317 318 if (cs_prog_data->push.cross_thread.size > 0) { 319 uint32_t *param_copy = param; 320 for (unsigned i = 0; 321 i < cs_prog_data->push.cross_thread.dwords; 322 i++) { 323 assert(prog_data->param[i] != BRW_PARAM_BUILTIN_SUBGROUP_ID); 324 param_copy[i] = brw_param_value(brw, prog, stage_state, 325 prog_data->param[i]); 326 } 327 } 328 329 if (cs_prog_data->push.per_thread.size > 0) { 330 for (unsigned t = 0; t < cs_prog_data->threads; t++) { 331 unsigned dst = 332 8 * (cs_prog_data->push.per_thread.regs * t + 333 cs_prog_data->push.cross_thread.regs); 334 unsigned src = cs_prog_data->push.cross_thread.dwords; 335 for ( ; src < prog_data->nr_params; src++, dst++) { 336 if (prog_data->param[src] == BRW_PARAM_BUILTIN_SUBGROUP_ID) { 337 param[dst] = t; 338 } else { 339 param[dst] = brw_param_value(brw, prog, stage_state, 340 prog_data->param[src]); 341 } 342 } 343 } 344 } 345 346 stage_state->push_const_size = 347 cs_prog_data->push.cross_thread.regs + 348 cs_prog_data->push.per_thread.regs; 349 } 350