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 33static uint32_t 34f_as_u32(float f) 35{ 36 union fi fi = { .f = f }; 37 return fi.ui; 38} 39 40static uint32_t 41brw_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 offset = prog->Parameters->ParameterValueOffset[idx]; 72 unsigned comp = BRW_PARAM_PARAMETER_COMP(param); 73 assert(idx < prog->Parameters->NumParameters); 74 return prog->Parameters->ParameterValues[offset + comp].u; 75 } 76 77 case BRW_PARAM_DOMAIN_UNIFORM: { 78 unsigned idx = BRW_PARAM_UNIFORM_IDX(param); 79 assert(idx < prog->sh.data->NumUniformDataSlots); 80 return prog->sh.data->UniformDataSlots[idx].u; 81 } 82 83 case BRW_PARAM_DOMAIN_IMAGE: { 84 unsigned idx = BRW_PARAM_IMAGE_IDX(param); 85 unsigned offset = BRW_PARAM_IMAGE_OFFSET(param); 86 assert(offset < ARRAY_SIZE(stage_state->image_param)); 87 return ((uint32_t *)&stage_state->image_param[idx])[offset]; 88 } 89 90 default: 91 unreachable("Invalid param domain"); 92 } 93} 94 95 96void 97brw_populate_constant_data(struct brw_context *brw, 98 const struct gl_program *prog, 99 const struct brw_stage_state *stage_state, 100 void *void_dst, 101 const uint32_t *param, 102 unsigned nr_params) 103{ 104 uint32_t *dst = void_dst; 105 for (unsigned i = 0; i < nr_params; i++) 106 dst[i] = brw_param_value(brw, prog, stage_state, param[i]); 107} 108 109 110/** 111 * Creates a streamed BO containing the push constants for the VS or GS on 112 * gen6+. 113 * 114 * Push constants are constant values (such as GLSL uniforms) that are 115 * pre-loaded into a shader stage's register space at thread spawn time. 116 * 117 * Not all GLSL uniforms will be uploaded as push constants: The hardware has 118 * a limitation of 32 or 64 EU registers (256 or 512 floats) per stage to be 119 * uploaded as push constants, while GL 4.4 requires at least 1024 components 120 * to be usable for the VS. Plus, currently we always use pull constants 121 * instead of push constants when doing variable-index array access. 122 * 123 * See brw_curbe.c for the equivalent gen4/5 code. 124 */ 125void 126gen6_upload_push_constants(struct brw_context *brw, 127 const struct gl_program *prog, 128 const struct brw_stage_prog_data *prog_data, 129 struct brw_stage_state *stage_state) 130{ 131 const struct gen_device_info *devinfo = &brw->screen->devinfo; 132 struct gl_context *ctx = &brw->ctx; 133 134 bool active = prog_data && 135 (stage_state->stage != MESA_SHADER_TESS_CTRL || 136 brw->programs[MESA_SHADER_TESS_EVAL]); 137 138 if (active) 139 _mesa_shader_write_subroutine_indices(ctx, stage_state->stage); 140 141 if (!active || prog_data->nr_params == 0) { 142 stage_state->push_const_size = 0; 143 } else { 144 /* Updates the ParamaterValues[i] pointers for all parameters of the 145 * basic type of PROGRAM_STATE_VAR. 146 */ 147 /* XXX: Should this happen somewhere before to get our state flag set? */ 148 if (prog) 149 _mesa_load_state_parameters(ctx, prog->Parameters); 150 151 int i; 152 const int size = prog_data->nr_params * sizeof(gl_constant_value); 153 gl_constant_value *param; 154 if (devinfo->gen >= 8 || devinfo->is_haswell) { 155 param = brw_upload_space(&brw->upload, size, 32, 156 &stage_state->push_const_bo, 157 &stage_state->push_const_offset); 158 } else { 159 param = brw_state_batch(brw, size, 32, 160 &stage_state->push_const_offset); 161 } 162 163 STATIC_ASSERT(sizeof(gl_constant_value) == sizeof(float)); 164 165 /* _NEW_PROGRAM_CONSTANTS 166 * 167 * Also _NEW_TRANSFORM -- we may reference clip planes other than as a 168 * side effect of dereferencing uniforms, so _NEW_PROGRAM_CONSTANTS 169 * wouldn't be set for them. 170 */ 171 brw_populate_constant_data(brw, prog, stage_state, param, 172 prog_data->param, 173 prog_data->nr_params); 174 175 if (0) { 176 fprintf(stderr, "%s constants:\n", 177 _mesa_shader_stage_to_string(stage_state->stage)); 178 for (i = 0; i < prog_data->nr_params; i++) { 179 if ((i & 7) == 0) 180 fprintf(stderr, "g%d: ", 181 prog_data->dispatch_grf_start_reg + i / 8); 182 fprintf(stderr, "%8f ", param[i].f); 183 if ((i & 7) == 7) 184 fprintf(stderr, "\n"); 185 } 186 if ((i & 7) != 0) 187 fprintf(stderr, "\n"); 188 fprintf(stderr, "\n"); 189 } 190 191 stage_state->push_const_size = ALIGN(prog_data->nr_params, 8) / 8; 192 /* We can only push 32 registers of constants at a time. */ 193 194 /* From the SNB PRM (vol2, part 1, section 3.2.1.4: 3DSTATE_CONSTANT_VS: 195 * 196 * "The sum of all four read length fields (each incremented to 197 * represent the actual read length) must be less than or equal to 198 * 32" 199 * 200 * From the IVB PRM (vol2, part 1, section 3.2.1.3: 3DSTATE_CONSTANT_VS: 201 * 202 * "The sum of all four read length fields must be less than or 203 * equal to the size of 64" 204 * 205 * The other shader stages all match the VS's limits. 206 */ 207 assert(stage_state->push_const_size <= 32); 208 } 209 210 stage_state->push_constants_dirty = true; 211} 212 213 214/** 215 * Creates a temporary BO containing the pull constant data for the shader 216 * stage, and the SURFACE_STATE struct that points at it. 217 * 218 * Pull constants are GLSL uniforms (and other constant data) beyond what we 219 * could fit as push constants, or that have variable-index array access 220 * (which is easiest to support using pull constants, and avoids filling 221 * register space with mostly-unused data). 222 * 223 * Compare this path to brw_curbe.c for gen4/5 push constants, and 224 * gen6_vs_state.c for gen6+ push constants. 225 */ 226void 227brw_upload_pull_constants(struct brw_context *brw, 228 GLbitfield64 brw_new_constbuf, 229 const struct gl_program *prog, 230 struct brw_stage_state *stage_state, 231 const struct brw_stage_prog_data *prog_data) 232{ 233 unsigned i; 234 uint32_t surf_index = prog_data->binding_table.pull_constants_start; 235 236 if (!prog_data->nr_pull_params) { 237 if (stage_state->surf_offset[surf_index]) { 238 stage_state->surf_offset[surf_index] = 0; 239 brw->ctx.NewDriverState |= brw_new_constbuf; 240 } 241 return; 242 } 243 244 /* Updates the ParamaterValues[i] pointers for all parameters of the 245 * basic type of PROGRAM_STATE_VAR. 246 */ 247 _mesa_load_state_parameters(&brw->ctx, prog->Parameters); 248 249 /* BRW_NEW_*_PROG_DATA | _NEW_PROGRAM_CONSTANTS */ 250 uint32_t size = prog_data->nr_pull_params * 4; 251 struct brw_bo *const_bo = NULL; 252 uint32_t const_offset; 253 gl_constant_value *constants = brw_upload_space(&brw->upload, size, 64, 254 &const_bo, &const_offset); 255 256 STATIC_ASSERT(sizeof(gl_constant_value) == sizeof(float)); 257 258 brw_populate_constant_data(brw, prog, stage_state, constants, 259 prog_data->pull_param, 260 prog_data->nr_pull_params); 261 262 if (0) { 263 for (i = 0; i < ALIGN(prog_data->nr_pull_params, 4) / 4; i++) { 264 const gl_constant_value *row = &constants[i * 4]; 265 fprintf(stderr, "const surface %3d: %4.3f %4.3f %4.3f %4.3f\n", 266 i, row[0].f, row[1].f, row[2].f, row[3].f); 267 } 268 } 269 270 brw_emit_buffer_surface_state(brw, &stage_state->surf_offset[surf_index], 271 const_bo, const_offset, 272 ISL_FORMAT_R32G32B32A32_FLOAT, 273 size, 1, 0); 274 275 brw_bo_unreference(const_bo); 276 277 brw->ctx.NewDriverState |= brw_new_constbuf; 278} 279 280/** 281 * Creates a region containing the push constants for the CS on gen7+. 282 * 283 * Push constants are constant values (such as GLSL uniforms) that are 284 * pre-loaded into a shader stage's register space at thread spawn time. 285 * 286 * For other stages, see brw_curbe.c:brw_upload_constant_buffer for the 287 * equivalent gen4/5 code and gen6_vs_state.c:gen6_upload_push_constants for 288 * gen6+. 289 */ 290void 291brw_upload_cs_push_constants(struct brw_context *brw, 292 const struct gl_program *prog, 293 const struct brw_cs_prog_data *cs_prog_data, 294 struct brw_stage_state *stage_state) 295{ 296 struct gl_context *ctx = &brw->ctx; 297 const struct brw_stage_prog_data *prog_data = 298 (struct brw_stage_prog_data*) cs_prog_data; 299 300 /* Updates the ParamaterValues[i] pointers for all parameters of the 301 * basic type of PROGRAM_STATE_VAR. 302 */ 303 /* XXX: Should this happen somewhere before to get our state flag set? */ 304 _mesa_load_state_parameters(ctx, prog->Parameters); 305 306 if (cs_prog_data->push.total.size == 0) { 307 stage_state->push_const_size = 0; 308 return; 309 } 310 311 312 uint32_t *param = 313 brw_state_batch(brw, ALIGN(cs_prog_data->push.total.size, 64), 314 64, &stage_state->push_const_offset); 315 assert(param); 316 317 STATIC_ASSERT(sizeof(gl_constant_value) == sizeof(float)); 318 319 if (cs_prog_data->push.cross_thread.size > 0) { 320 uint32_t *param_copy = param; 321 for (unsigned i = 0; 322 i < cs_prog_data->push.cross_thread.dwords; 323 i++) { 324 assert(prog_data->param[i] != BRW_PARAM_BUILTIN_SUBGROUP_ID); 325 param_copy[i] = brw_param_value(brw, prog, stage_state, 326 prog_data->param[i]); 327 } 328 } 329 330 if (cs_prog_data->push.per_thread.size > 0) { 331 for (unsigned t = 0; t < cs_prog_data->threads; t++) { 332 unsigned dst = 333 8 * (cs_prog_data->push.per_thread.regs * t + 334 cs_prog_data->push.cross_thread.regs); 335 unsigned src = cs_prog_data->push.cross_thread.dwords; 336 for ( ; src < prog_data->nr_params; src++, dst++) { 337 if (prog_data->param[src] == BRW_PARAM_BUILTIN_SUBGROUP_ID) { 338 param[dst] = t; 339 } else { 340 param[dst] = brw_param_value(brw, prog, stage_state, 341 prog_data->param[src]); 342 } 343 } 344 } 345 } 346 347 stage_state->push_const_size = 348 cs_prog_data->push.cross_thread.regs + 349 cs_prog_data->push.per_thread.regs; 350} 351